Introduction To Theoretical Linguistic Lyons PDF

INTRODUCTION TO
THEORETICAL LINGUISTICS
Downloaded from Cambridge Books Online by IP 152.118.148.226 on Tue Apr 15 07:53:48 BST 2014.
http://ebooks.cambridge.org/ebook.jsf?bid=CBO9781139165570
Cambridge Books Online Cambridge University Press, 2014
INTRODUCTION
TO THEORETICAL
LINGUISTICS
JOHN LYONS
Master, Trinity Hall, Cambridge
CAMBRIDGE
UNIVERSITY PRESS
Published by the Press Syndicate of the University of Cambridge
The Pitt Building, Trumpington Street, Cambridge CB2 IRP
40 West 20th Street, New York, NY 10011-4211, USA
10 Stamford Road, Oakleigh, Melbourne 3166, Australia
Cambridge University Press 1968
First published 1968

Reprinted 1969 (twice) 19711972 1974 1975 1977 1979 1985
1987 1992 1995
ISBN o 521 09510 7

(paperback)
Transferred to digital printing 2001
CONTENTS
i Linguistics: The Scientific Study of Language

I . I Introductory page 1
i. 2 Traditio?ial grammar 4
1.3 Comparative philology 21
1.4 Modern linguistics 38
2 The Structure of Language

2.1 Introductory 53
2.2 Substance and form 54
2.3 Paradigmatic and syntagmatic relations 70
2.4 Statistical structure 81
3 The Sounds of Language

3.1 Introductory 99
3.2 Phonetics IOI
3.3 Phonology 112
4 Grammar: General Principles

4.1 Introductory 133
4.2 Formal grammar 137
4.3 Grammar and lexicon 158
5 Grammatical Units
5.1 Introductory I7O
5.2 77* sentence 172
5.3 TTitf morpheme l8o
5.4 77i ;orrf 194
c;.; 77* notion of 'rank' 2O6
VI CONTENTS
6 Grammatical Structure
6.1 Immediate constituents page 209
6.2 Phrase-structure grammars
6.3 Categorial grammars 227
6.4 Exocentric and endocentric constructions 231
6.5 Context-sensitive grammars 235
6.6 Transformational grammar 247
7 Grammatical Categories
7.2 Deictic categories 275
7.3 Number and gender 281
7.4 CVw 289
7.5 Tense, mood and aspect 34
7.6 The parts of speech 3i7
8 Grammatical Functions
8.1 Subject, predicate and adjunct 334
8.2 Transitivity and ergativity 350
8.3 FOM* 37i
8.4 Existential, locative and possessive constructions 388
9 Semantics: General Principles

9.2 Traditional semantics 43
9.3 ' Meaning fulness* 412
9.4 Reference and sense 424
9.5 'Lexical* and 'grammatical* meaning 435
10 Semantic Structure
10.2 Synonymy 446
CONTENTS vii
10.3 Hyponymy and incompatibility page 453
10.4 Antonymy, complementarity and converseness 460
10.5 Componential analysis and universal semantics 470
Notes and references 482
Addenda 489
Bibliography 490
Table of symbols and notational conventions 506
Index of proper names 507
Index of subjects 510
NOTE
Explanatory notes, bibliographical

references and suggestions for fur-
ther reading for chapters 1 to 10 are
given in Notes and References,
pp. 482-9.
Cambridge Books Online
http://ebooks.cambridge.org/
Introduction to Theoretical Linguistics
John Lyons
Book DOI: http://dx.doi.org/10.1017/CBO9781139165570
Online ISBN: 9781139165570
Hardback ISBN: 9780521056175
Paperback ISBN: 9780521095105
Chapter
PREFACE TO THE 1995 EDITION pp. ix-xiv
Chapter DOI: http://dx.doi.org/10.1017/CBO9781139165570.001
Cambridge University Press

PREFACE TO THE 1995 EDITION
This book was first published in 1968; it has been reprinted several times
and has been translated, so far, into 12 languages; and it is still as widely used
today in colleges and universities throughout the world as it was twenty
years ago. The fact that this is so is of course very gratifying. It is, however,
something that I never anticipated when I wrote the book.
It has been suggested to me that one of the reasons why Introduction to
theoretical linguistics (ITL) has become something of a classic (the term is
not mine) is that in writing it I took a broader view of the subject-matter
than the authors of most other textbooks of the period and emphasized, as
they did not, continuity rather than discontinuity in the development of
what I referred to in the original Preface as contemporary linguistic theory:
continuity between philology and linguistics; between traditional grammar
and modern (i.e., mid-twentieth-century) linguistics; between structur-
alism (European and American) and generativism.
It is arguable that, at certain points and in certain respects, I
overemphasized continuity. I did not appreciate the degree to which my
view of linguistics had been shaped by my own academic background,
which included a training in classical philology and philosophy, on the one
hand, and a certain amount of formal and computational linguistics, on the
other, and by my own research interests. In particular, I did not realize that
the classical philologist's understanding of traditional grammar was not the
same as that of someone whose understanding of it derived primarily from
its application to English and other modern, mainly European, languages.
Also, although I was well aware that traditional grammar is far less
homogeneous than it is commonly represented as being in textbook
accounts of the history of linguistics (and I made this fact explicit in my
own account of the history of linguistics in Chapter 1 and at relevant points
elsewhere in the book), I did not always give alternative traditional analyses
when I might have done. In some cases, this was simply due to ignorance,
on my part, of the work of those whose specialized expertise was greater
than mine. In other cases, it was a consequence of my more or less
deliberate adoption of what historiographers often refer to as Whiggism: a
concentration on those themes or issues which (independently of their
importance in their own historical context) can be seen with hindsight
(from the viewpoint of the winning side, the Whigs) to have prevailed and
to have brought us to where we are now.
http://dx.doi.org/10.1017/CBO9781139165570.001
I make no apologies for the Whiggism or, more generally, for the
historicism that is implicit in my explanation of current attitudes and ideas.
I still think that a historicist account of contemporary linguistic theory - an
inevitably, but I trust responsibly, selective account - is both justifiable and
helpful. It is certainly better than a totally unhistorical account of the kind
that one all too often finds in introductory textbooks. And it is far better
than a revolutionary, not to say messianic, approach which exaggerates the
difference between one orthodoxy and its replacement. If I were rewriting
Chapter i today, there is much that I would change as far as the detail is
concerned. The historiography of linguistics has made enormous strides in
the past quarter-century and both the primary and the secondary sources
are correspondingly richer, more accessible and better indexed for ease of
reference than they were. The empirical database with which theoretical
linguistics operates is also richer and more reliable than it was, descriptively
and typologicaliy. This would obviously influence my choice of what to
include and what to exclude or my judgement of its significance. But I do
think it would be largely a matter of detail. On the whole, I stand by the
historical and interpretative generalizations that I made in 1968 (many of
which continue to be quoted or referred to in the more recent literature).
Students should be encouraged, however, not to take them on trust, but to
compare them with the generalizations made in other textbooks and to
weight them against the evidence that is now more readily available.
The fact that ITL is widely referred to in the literature and quotations
from it are often used as convenient starting-points for the further
discussion of particular theoretical issues means that, for some years now, I
have felt that I was not at liberty to amend or update the text, even in those
places where I was dissatisfied with the position that I had adopted earlier
or with my formulation of it. Over the years I have of course changed my
mind on several issues. I have also introduced and established a number of
terminological distinctions and typographical conventions, partly in
response to what are generally acknowledged to be advances made in
linguistic theory and partly as a consequence of trying to be more precise
about some of the topics that I have been particularly concerned with in my
own work. For example, I have come to see that the distinction between
sentences and utterances needs to be drawn more carefully than it is drawn
in ITL, as do the distinction between language and speech, the distinction
between forms and expressions (to use current terminology) and the
distinction between the use of a language and the product of the use of a
language (which is confused both in Saussure's terminological distinction
between 'langue' and 'parole' and in Chomsky's roughly comparable
distinction between 'competence' and 'performance'). It would be
impossible to list here, still less to discuss, all the changes that would need
to be introduced into the text of ITL if these and other such terminological
distinctions were adopted. I would refer interested readers, and more
especially lecturers and instructors using ITL as a textbook in class, to some
of my more recent works, including Language and linguistics (1981; reissued
1995), Natural language and universal grammar (1991) and Linguistic
semantics (1995). Many of the topics dealt with in ITL are discussed there,
in greater or less detail, with what is intended to be greater precision and
with reference to the subsequent literature.
I should now say something about generative grammar and generativism.
ITL was probably the first relatively comprehensive textbook of general
linguistics to include, and to make central, an introduction to Chomskyan
transformational-generative grammar. I described this, at the time, as
'currently the most familiar and the most highly developed system [of
syntactic analysis based on generative principles]'. I did, however, make
clear that there were alternative approaches to the formalization of the
grammatical structure of languages which were, at least in principle,
generative (as I understood the term), but which did not necessarily respect
the primacy of particular Chomskyan axioms (notably the axiom of
concatenation and those pertaining to the definition of constituent-
structure in terms of dominance and bracketing). Since 1968, several of the
alternative systems of generative grammar which were then in prospect
have been developed, and others have appeared on the scene; and
Chomskyan generative grammar itself has passed through several stages of
modification and development.
In the context of present-day theories of generative grammar, what
might be referred to as the pre-classical version of Chomsky's own system
of transformational-generative grammar that is presented in ITL - the
original system of Syntactic structures (1957) - will strike many of today's
specialists as primitive in the extreme, not to say archaic. However, the very
simplicity of the original system may have its pedagogical virtues: it is
relatively easy to understand, and it can still serve satisfactorily for the
limited, but important, purpose of explaining the basic principles and
providing a solid base upon which to build. Paradoxically, perhaps, it may
serve more satisfactorily for this purpose than either the so-called standard
theory of Chomsky's Aspects of the theory of syntax (1965) or any of its
successors in a direct line of descent. Much of the detailed development of
Chomskyan transformational-generative grammar that took place in the
1970s and early 1980s, though historically important, can now be seen as
irrelevant, in the context, on the one hand, of the most recent, very radical,
revisions of the system favoured by Chomsky himself (a system of
parameters and principles, rather than of fully generative rules) and, on the
other, of the increased number of competing systems of generative (or
partly generative) grammar that are now being discussed and compared in
the specialist literature.
In my exposition of the theory of generative grammar in 1968,1 was at
pains to distinguish the wood from the trees and the essential from the
inessential. In particular, I was careful to distinguish generative grammar as
such from what I have in more recent publications referred to as
generativism (cf. Language and linguistics, 1981, and, more especially, the
third edition of Chomsky, 1991). By 'generativism' I mean, briefly, a whole
cluster of philosophical and psychological doctrines (some, though not all,
of which are both theoretically exciting and empirically well supported)
which motivate many linguists' (including Chomsky's) interest in
generative grammar, as well as that of philosophers, psychologists and
others, but which are logically independent of it. Very few textbook
presentations of generative grammar draw the distinction, as they should,
between generative grammars as formal systems and generativism. Either
may yet survive independently of the other. Time will tell.
Not only generative grammar, but grammatical theory in general, has
moved on from where it was twenty or so years ago. Many of the topics
dealt with in the central chapters of ITL (e.g., grammatical categories,
deixis, transitivity and ergativity, morphological structure) have been
intensively researched during this period. As I said above, my own views
have changed on certain issues, but not to the point that I would wish to
retract or radically modify any of the generalizations that are made in
respect of them in ITL. I trust that what is said about them here will
continue to serve, as it has done for some time now, as an introduction to
more detailed and more specialized treatments and as a bridge between
traditional and more modern treatments of them.
As to semantics, the account that is given in ITL is also somewhat old-
fashioned in certain respects: it is restricted to lexical semantics (roughly,
the study of the meaning of words and phrases) and it does not exploit
properly the distinction between sentence-meaning and utterance-meaning
(to which I referred above). What is actually said about lexical semantics,
however, requires little or no modification in the light of the very
considerable advances that have been made recently in semantics (and
pragmatics). For my own approach to semantics on a broader front and in
the context of what is coming to be a more or less standard theoretical
framework readers are referred to Linguistic semantics (1995).
I would like to reiterate my thanks to all those who helped me directly
or indirectly in writing ITL> especially to my four mentors, W. Sidney Allen
and Robert H. Robins (my PhD supervisors) and the late Charles E. Bazell
and Fred W. Householder. The passage of time has, if anything, made me
even more aware of how much I learned from them and how fortunate I
was to be introduced to modern linguistics by four scholars whose expertise
in their several areas of specialization was unrivalled and who combined an
interest in contemporary linguistic theory with a proper sense of its
diachronic and synchronic continuity. The influence of Noam Chomsky's
work, especially of Syntactic structures, is evident throughout ITLy but
especially of course in Chapter 4; I should none the less like to acknowledge
it formally, and gratefully, here, the more so as I have subsequently taken a
different view from his on certain aspects of what I have elsewhere referred
to as generativism. The other friends and colleagues who influenced my
views on linguistics in the late 1950s and the early 1960s are too numerous
to mention; the references both in ITL and in later works must suffice.
Finally, I must thank the current commissioning editor for linguistics at
Cambridge University Press, Judith Ayling, for encouraging me to reissue
ITL and to write a new Preface for it.
Cambridge, February 1995 J.L.
John Lyons
Online ISBN: 9781139165570
Chapter
1 - Linguistics: The Scientific Study of Language pp. 1-52

1
LINGUISTICS: THE
SCIENTIFIC STUDY OF LANGUAGE
1.1 Introductory
I.I.I Definition of linguistics
Linguistics may be defined as the scientific study of language. This
definition is hardly sufficient to give the reader any positive indication
of the fundamental principles of the subject. It may be made a little
more revealing by drawing in greater detail the implications contained
in the qualification 'scientific'. For the moment, it will be enough to
say that by the scientific study of language is meant its investigation
by means of controlled and empirically verifiable observations and
with reference to some general theory of language-structure.
i. i. 2 Linguistic terminology
It is sometimes suggested that the terminology, or 'jargon', of
modern linguistics is unnecessarily complex. This is a criticism which
need not detain us long. Every science has its own technical vocabu-
lary: it is only because the layman takes on trust the established
sciences, and especially the 'natural* sciences, that he does not
question their right to furnish themselves with special vocabularies.
The technical terms used by linguists arise in the course of their work
and are easily understood by those who approach the subject sympa-
thetically and without prejudice. It should not be forgotten that most
of the terms which the non-linguist employs to talk about language
('word', 'syllable', 'letter', 'phrase', 'sentence', 'noun', 'verb', etc.)
originated as technical terms of traditional grammar and are no less
'abstract' in their reference than the more recent creations of
linguists. If the contemporary linguist requires different terms,
instead of, or in addition to, those familiar to the layman, this is
accounted for partly by the fact that the non-technical employment of
many of the terms of traditional grammar has rendered them in-
sufficiently precise for scientific purposes and partly by the simple
[i]
2 I. LINGUISTICS
fact that modern linguistics has in certain respects advanced beyond
traditional grammar in its attempt to construct a general theory of
language-structure. The technical terms employed in this book will
be introduced gradually, with full explanation and as far as possible
with reference to traditional terms of general currency. As we shall
see, the use of a special vocabulary eliminates a good deal of ambi-
guity and possible misunderstanding in the discussion of language.
1.1.3 Objective approach to language

The chief difficulty facing the person who comes new to the study of
linguistics is that of being prepared to look at language objectively.
For language is something we tend to take for granted; something
with which we are familiar from childhood in a practical, unreflecting
way. And, as has often been observed, it requires a particularly strong
effort to look at familiar things afresh. Nor is it merely our intuitive
or practical familiarity with language that stands in the way of its
objective examination. There are all sorts of social and nationalistic
prejudices associated with language, and many popular misconcep-
tions fostered by the distorted version of traditional grammar that is
frequently taught in the schools. To free one's mind of these pre-
judices and misconceptions is indeed difficult; but it is both a
necessary and a rewarding first step.
i. i. 4 History of linguistics
Nothing is more helpful to the layman or student making his first
acquaintance with the science of linguistics than some knowledge of
the history of the subject. Many of the ideas about language which
the linguist will question, if he does not abandon them entirely, will
seem less obviously self-evident if one knows something of their
historical origin. This is true not only of a good deal that is taught
formally at school, but also of much that at first sight might appear to
be a matter of downright common sense; for, as Bloomfield has
remarked of the common-sense way of dealing with linguistic matters,
' like much else that masquerades as common sense it is in fact highly
sophisticated, and derives, at no great distance, from the speculations
of ancient and medieval philosophers'. As instances of 'common-
sense ' attitudes to language which derive from what Bloomfield refers
to as 'the speculations of ancient and medieval philosophers' one may
cite the commonly-held belief that all languages manifest the same
I.I. INTRODUCTORY 3
* parts of speech' (in the form in which this belief is usually held and
expressed). The traditional theory of 'the parts of speech', and the
standard definitions of classical grammar, reflect, as we shall see in
due course, ancient and medieval attempts to force together the
categories of grammar, logic and metaphysics. Other commonly held
views about language derive not so much from philosophical specu-
lation as from the subordination of grammar to the task of interpreting
written texts, and especially to that of interpreting works written in
Greek and Latin by the classical authors.
But the history of linguistics is of interest today not only in so far as
it enables us to free ourselves of certain commonly held misconcep-
tions about language. Linguistics, like any other science, builds on the
past; and it does so, not only by challenging and refuting traditional
doctrines, but also by developing and reformulating them. As an aid
to the understanding of the principles and assumptions governing
modern linguistics a knowledge of the history of the subject has
therefore a positive, as well as a negative, contribution to make. This
point will be abundantly illustrated in the course of the book. It is
stressed here because many recent works on linguistics, in describing
the great advances made in the scientific investigation of language in
the last few decades, have neglected to emphasize the continuity of
Western linguistic theory from earliest times to the present day.
It may also be pointed out here that what is generally referred to as
' traditional grammar* (and we shall continue to use this term) is
much richer and more diversified than is often suggested in the
cursory references made to it by many modern handbooks of lin-
guistics. Much of the earlier history of Western linguistic thought is
obscure and controversial. This is mainly due to the fact that most of
the original sources have disappeared: from what has survived it is
clear that, although one can trace a continuous line of development
from Plato and the Sophists to the medieval Schoolmen, throughout
this period there were many individual grammarians who were
capable of original thought. A definitive and comprehensive history of
4
traditional grammar' is yet to be written. Although the necessarily
brief outline of the history of linguistics which follows is intended
primarily as an introduction to the present state of the subject, we
shall try, as far as possible, to relate past developments in linguistic
theory to the social conditions and the philosophical ideas current at
the time.
4 I. LINGUISTICS
1.2 Traditional grammar

i. 2. i Philosophical origins of traditional grammar
Traditional grammar, like so many other of our academic traditions,
goes back to Greece of the fifth century before Christ. For the Greeks
'grammar' was from the first a part of * philosophy'. That is to say,
it was a part of their general inquiry into the nature of the world
around them and of their own social institutions.
1.2.2 ' Nature' and ' convention'

The Greek philosophers debated whether language was governed by
'nature* or 'convention*. This opposition of 'nature' and 'con-
vention* was a commonplace of Greek philosophical speculation. To
say that a particular institution was ' natural * was to imply that it had
its origin in eternal and immutable principles outside man himself
(and was therefore inviolable); to say that it was 'conventional*
implied that it was merely the result of custom and tradition (that is,
of some tacit agreement, or 'social contract*, among the members of
the communitya ' contract * which, since it was made by men, could
be broken by men).
In the discussion of language, the distinction of 'nature' and
' convention* was made to turn principally upon the question whether
there was any necessary connexion between the meaning of a word
and its form. Extreme adherents of the 'naturalist* school, like
Cratylus, whose views Plato reports in his dialogue of that name,
maintained that all words were indeed 'naturally* appropriate to the
things they signified. Although this might not always be evident to
the layman, they would say, it could be demonstrated by the philo-
sopher able to discern the * reality * that lay behind the appearance of
things. Thus was born the practice of conscious and deliberate
etymology. The term itself (being formed from the Greek stem etymo-
signifying 'true* or 'real*) betrays its philosophical origin. To lay
bare the origin of a word and thereby its ' true' meaning was to reveal
one of the truths of 'nature*.
Various ways were recognized in which the form of a word might
be 'naturally' appropriate to its meaning. First of all, there was the
relatively small set of words, like neigh, bleat, hoot, crash, tinkle, etc.
(to use examples from English rather than Greek), which to some
degree or other were 'imitative' of the sounds they referred to.
1.2. TRADITIONAL GRAMMAR 5
A different, though related, category comprised words {cuckoo,
peewit, etc.) which were 'imitative' of a particular kind of sound, but
which denoted the source of the sound, rather than the sound itself.
In both cases there is an obvious 'natural* connexion between the
physical form of the word and wThat it signifies. The technical term
employed for words belonging to these two categories, and still used
in this sense, was onomatopoeia. This was simply the Greek word for
* the creation of names \ The fact that it was restricted by grammarians
to words which 'imitate' the sounds they denote reflects the view
maintained by the Greek 'naturalists' (particularly the Stoic philo-
sophers) that such words form the basic set of 'names' from which
language has developed. The fundamental relationship between a
word and its meaning was that of' naming'; and originally words were
' imitative' of the things they named. Onomatopoeic words formed the
nucleus of the vocabulary.
But relatively few words are onomatopoeic. Others were demon-
strated to be of 'natural' origin by reference to one or more of their
constituent sounds. Certain sounds were held to be suggestive, or
' imitative', of particular physical qualities, or activities, being classified
as 'smooth', 'harsh', 'liquid', 'masculine', etc. For instance, one
might maintain, in the spirit of the 'naturalists', that I is a liquid
sound, and that therefore the words liquid,flow,etc., contain a sound
which is 'naturally' appropriate to their meaning. The modern term
for this kind of relationship between the constituent sounds of words
and their nreaning, in so far as it is asserted to be a feature of language,
is sound-symbolism.
After taking full account of onomatopoeia and sound-symbolism,
the Greek etymologists were still left with very many words to
explain. At this point they invoked various principles in terms of
which words could be derived from, or related to, one another; and
these were codified in time as the traditional principles of etymology.
We shall not go into these principles here, except to mention that
they fall into two types. First, the meaning of a word might be
extended by virtue of some 'natural' connexion between the original
and the secondary application: cf. the mouth of a river, the neck of a
bottle, etc. (These are examples of metaphor, one of the many terms
introduced by the Greeks which have passed into traditional grammars
and works on style.) Second, the form of a word might be derived
from that of another by the addition, deletion, substitution and trans-
position of sounds (granted some ' natural' connexion in the meanings
of the two words). It is only by a very free and uncontrolled use of the
6 I. LINGUISTICS
second set of principles, operating upon the form of a word, that the
' naturalists' could maintain their position, claiming to be able to
derive all words from a primary set of words of ' natural * origin.
1.2.3 Analogists and anomalists

The dispute between 'naturalists' and 'conventionalists' was to
endure for centuries, dominating all speculation about the origin of
language and the relationship between words and their meaning. Its
importance for the development of grammatical theory is that it gave
rise to ' etymological' investigations which stimulated and maintained
the interest of scholars in classifying the relationships between words.
For good and ill, it set the study of grammar in the framework of
general philosophical inquiry.
For reasons that need not be discussed here, the controversy
between the 'naturalists' and the 'conventionalists' developed rather
later (from about the second century B.C.) into a dispute as to how far
language was 'regular'. In Greek, as in English, although there are
many obvious instances of 'regular' patterns in the language, there
are also many exceptions. As an example of a 'regular' pattern in
English, take boy: boys, girl: girls, cow: cows, etc. This is an instance
of one kind of ' regularity' in language discussed by the founders of
traditional grammar. Other types will be illustrated below. The
Greek words for regularity and irregularity in this sense of these
terms are 'analogy' and 'anomaly'; those who maintained that
language was essentially systematic and regular are generally called
' analogists', and those who took the contrary view are referred to as
'anomalists'.
It is to be observed that the term 'analogy' is also being used here
in the more particular sense of a mathematical' proportion', according
to which we say, for instance, that the proportion 6:3 equals the
proportion 4:2, 2:1, etc. (The term 'proportion' comes from the
Latin translation of the Greek word analogia.) 'Analogical' reasoning
was widely applied by Plato and Aristotle, and their followers, in the
study of the sciences. On the basis of a proportion like boy: boys, we
can form 'analogically' thousands of other words: cow: cows, girl:
girls, etc.; given either cow or cows, we can 'solve' the equation
boy:boys = cow:x or boy:boys = x:cows.
The analogists devoted their energies to the establishment of the
various models with reference to which the regular words of the
language could be classified (the traditional term ' paradigm' is merely
1.2. TRADITIONAL GRAMMAR J
the Greek word for 'model' or 'example'). The anomalists did not
deny that there were regularities in the formation of words in
language, but pointed to the many instances of irregular words for the
formation of which analogical reasoning is of no avail (child: children,
etc.) and also to the multiplicity of different 'analogies' that had to be
recognized for words of the same class (this is more striking for
Greek or Latin than for English). They also drew attention to the fact
that the relationship between the form of a word and its meaning was
frequently 'anomalous': for instance the names of the cities Thebes
or Athens are plural nouns in Greek, although they denote single
cities; one of the Greek words for ' child' (paidion) is neuter in gender,
although children must be either male or female (cf. the German
word Kindy which is also neuter: examples of this kind of 'anomaly*
could be given from many languages). Another example of 'anomaly*
was afforded by the existence of synonymy (two or more words with
the same meaning) and homonymy (one form with two or more
meanings). If language were really a product of human 'convention'
one would not expect to find 'irregularities' of these various kinds;
and if they existed they should be corrected. The anomalists main-
tained that language, a product of' nature', was only partly susceptible
of description in terms of analogical patterns of formation, and that
due attention had to be given to 'usage', however 'irrational' a
particular fact of 'usage' might be.
That the dispute between the 'analogists' and the 'anomalists' was
not settled once and for all by the Greeks is hardly surprising. In the
first place, the distinction between descriptive and prescriptive (or
normative) grammar was not clearly drawn (that is to say, the distinc-
tion between describing how people actually speak and write and
prescribing how they ought to speak and write: we shall discuss this
distinction in some detail later: cf. 1.4.3). Consequently the 'analogist'
would tend to 'correct' any apparent 'anomalies' with which he
might be confronted rather than change his ideas about the nature of
language. Secondly, and more importantly, since 'irregularities' can
only be determined with reference to the 'regularities' from which
they differ, what is ' irregular' from one point of view, from another
may be regarded as 'regular'. Any general dispute as to whether
language is 'regular' or not involves the further question: what are
in fact the 'regular* patterns? The controversy between the 'ana-
logists' and the 'anomalists' was not therefore, as is sometimes
suggested, a pointless dispute resulting from the perverse refusal of
both sides to recognize the obvious fact that there are both 'analogies'
8 I. LINGUISTICS
and 'anomalies' in language. It was at most a dispute as to what
constitutes 'regularity* in language and how much of the apparent
'irregularity' can be shown, by further analysis, to be describable in
terms of alternative patterns.
The history of the controversy between the 'analogists' and the
' anomalists' is far from clear. Its earlier development is known only
from fragments, and from quotations and comments in the works of
later authors; and it is possible that the later writers (in particular
Varro, a Roman grammarian of the first century B.C.) may have
exaggerated the differences between the two parties to the contro-
versy. Whatever their theoretical pronouncements, both the 'ana-
logists' and the 'anomalists' admitted that there were certain
regularities in language, and both contributed to the systematization of
grammar. Indeed it was the Stoics, usually said to be 'anomalists',
who laid the foundations of traditional grammar in connexion with
their 'etymological' work. And the Alexandrian 'analogists' built
upon these. Such differences as we find between the Stoics and the
Alexandrians can generally be explained in the light of their difference
of purpose. The Stoics were interested primarily in the philosophical
problem of the origin of language, in logic and in rhetoric; the
Alexandrians in literary criticism. Moreover, as we shall see, the
Alexandrian scholars were working upon the literary texts of the past;
where there was no recorded 'usage' to refer to they invoked the
principle of 'analogy' to supply the want of this. Later grammarians,
responsible for the codification of what we now call traditional
grammar, recognized both 'analogy' and 'usage' ('anomaly') as
theoretical principles. However, this did not really solve the problem
since, on the one hand, when one is looking for regularities in language,
one is frequently faced with alternative ways of relating words and
sentences, and, on the other, there still remains the question whose
'usage' is to be taken as correct. Modern linguistics may claim to have
made some progress in the solution of these questions, as we shall see,
but not to have solved them definitively. The controversy between
'analogists' and 'anomalists' is still with us.
i .2.4 Alexandrian period

With the establishment of the great library in the Greek colony of
Alexandria at the beginning of the third century B.C. that city became
the centre of intense literary and linguistic research. The manuscripts
of the authors of the past, and in particular those containing the text
of the Homeric poems, had by now become intolerably corrupt. By
comparing different manuscripts of the same works the Alexandrian
scholars of the third and second centuries B.C. sought to restore the
original text and to decide between genuine and spurious works.
Since the language of the classical texts differed in many respects from
the contemporary Greek of Alexandria, the practice grew up of
publishing commentaries on the texts and grammatical treatises
elucidating the various difficulties that might trouble the reader of the
earlier Greek poets. Admiration for the great literary works of the
past encouraged the belief that the language in which they were
written was itself inherently * purer', more ' correct', than the current
colloquial speech of Alexandria and the other Hellenistic centres. The
grammars produced by Hellenistic scholars came therefore to have
a double purpose: they combined the aim of establishing and
explaining the language of the classical authors with the desire to
preserve Greek from corruption by the ignorant and unlettered. This
approach to the study of language fostered by Alexandrian classicism
involved two fatal misconceptions. The first concerns the relation
between written and spoken language; the second has to do with the
manner in which languages develop. They may both be referred to
what I will call the * classical fallacy' in the study of language.
From the beginning Greek linguistic scholarship had been
concerned primarily with the written language. (The term ' grammar',
which the Greeks applied to the study of language, bears witness to
this: it is derived from the word for 'the art of writing'.) No con-
sistent distinction was drawn between sounds and the letters used to
represent them. In so far as the difference between the spoken and the
written language was perceived at all, the tendency was always to
consider the former as dependent on, and derived from, the latter.
The Alexandrian concern with literature merely reinforced this
tendency.
The second misconception inherent in the Alexandrian approach to
the study of language was the assumption that the language of the
fifth-century Attic writers was more 'correct' than the colloquial
speech of their own time; and in general that the 'purity' of a
language is maintained by the usage of the educated, and ' corrupted'
by the illiterate. For more than two thousand years this prejudice was
to reign unchallenged. It is all the harder to eradicate in that the terms
in which the assumption is usually expressed'purity' and 'correct-
ness 'are taken as absolutes. It should be clear, however, that these
terms have no meaning except in relation to some selected standard.
IO I. LINGUISTICS
The assertion that the language of Plato is a 'purer' form of Greek

than, let us say, that of some illiterate artisan of Alexandria is there-
fore not so much false as either meaningless or tautological. We shall
return to this point and to a more detailed discussion of the relation
between written and spoken language towards the end of the present
chapter (cf. 1.4.2).
1.2.5 Greek grammar

We may now look at some of the more important particular features
of the grammatical analysis of their own language carried out by the
Greeks. To those who have been familiar from their school-days with
the various grammatical categories employed in traditional descrip-
tions of Greek, it might very well appear that the recognition of just
these categories and no others would impose itself immediately upon
anyone who set himself the task of analysing the language. Even the
most superficial knowledge of the history of Greek grammatical
scholarship shows us that this is not true. The particular analysis
reflected in standard school-grammars of Greek was so far from being
self-evident that it took some six centuries to elaborate (from the
fourth century B.C. to the second century A.D.). Moreover it is not the
only analysis possible, and is perhaps not even the best. In any case,
it could not reasonably be said that slightly different ways of
describing the language favoured by some of the Greek grammarians
are necessarily inferior to that which was eventually standardized and
handed to posterity as the grammar of Greek. In the following brief
account of the historical development of the traditional grammatical
framework the various categories recognized by the Greeks and their
successors will not be discussed in any detail. Such discussion will be
postponed until the ground has been prepared in subsequent
chapters.
Protagoras, one of the earliest and most influential of the fifth-
century Sophists, is credited with the distinction of three genders in
Greek. It is Plato (c. 429-347 B.C.) who, as far as we know, first
explicitly distinguished beteeen nouns and verbs. It may be noted,
however, that the two classes of words defined by Plato as ' nouns'
and 'verbs' were not co-extensive with the classes to which these
labels were given in the later systems of analysis upon which our
school-grammars are based. As defined by Plato, * nouns' were terms
that could function in sentences as the subjects of a predication and
'verbs' were terms which could express the action or quality predi-
1.2. TRADITIONAL GRAMMAR II
cated. (Roughly speaking, the subject of a predication names the thing
about which something is said, and the predicate is that part of the
sentence which says something about the thing named by the
subject: cf. 8.1.2.)
Two things may be observed. First, the definition of the major
grammatical classes, 'nouns' and * verbs', was made on logical
grounds: i.e. as constituents of a proposition. Second, what we now
call verbs and adjectives were put together in the same class. Even
when later Greek grammarians abandoned the classification established
by Plato, they did not replace it with the tripartite system into nouns,
verbs and adjectives, with which we are familiar, but substituted
another bipartite system, which brought together what we call nouns
and adjectives. Little attention was given at first to words which were
not members of the major classes.
It was not until medieval times that the division of words into
nouns, verbs and adjectives was made. Aristotle (384-322 B.C.) kept
the Platonic distinction between 'nouns' and 'verbs', but added to
these a further distinct class. These were the 'conjunctions'; by this
term Aristotle meant all those words which were not members of the
major classes, 'nouns' and 'verbs'. He also took over from his pre-
decessors the threefold classification of gender. He observed, however,
that the names of many ' things' (the term employed by Protagoras to
label the third gender) were grammatically either 'masculine' or
'feminine' in Greek, and he introduced the term 'intermediate' to
refer to the third gender. (Later, that which was neither 'masculine'
nor 'feminine' was called, quite simply, 'neither'; and it is the Latin
translation of this which has given us the traditional term 'neuter'.)
A more significant advance made by Aristotle was his recognition of
the category of tense in the Greek verb: that is to say, he noted that
certain systematic variations in the forms of the verb could be cor-
related with such temporal notions as 'present' or 'past'. His teaching
on this point, however (though more explicit than Plato's), is far from
clear.
Of the several different 'schools' of Greek philosophy, it was the
Stoics who gave the most attention to language. The reason for this
lay in their belief that right conduct was a matter of living in harmony
with 'nature' and that knowledge consisted in the conformity of our
ideas with the real things in 'nature', of which these ideas are, or
should be, the image. Language was therefore central to Stoic philo-
sophy, and, in particular, to the part to which they gave the term
'logic', but which included also what we should call epistemology
12 I. LINGUISTICS
and rhetoric, as well as grammar. One of the first and most
fundamental distinctions they made was that between form and
meaning, 'that which signifies* and 'that which is signified*. But the
Stoics did not take language to be a direct reflection of 'nature*. For
the most part they were 'anomalists', insisting on the lack of corre-
spondence between words and things and on the illogicalities of
language. Earlier members of the ' school' distinguished four parts of
speech ('noun', 'verb', 'conjunction', 'article'); later members five
(by separating 'common nouns' and 'proper nouns'). The adjective
was classed with the noun. The classification of what we now call
inflexion (e.g. the relationship between such forms, in English, as
boy, boysy or singf sang, sung) was greatly developed by the Stoics. It
was they too who gave to the term case the sense which it has pre-
served in standard grammatical usage ever since, distinguishing
between the true form of the noun, the 'upright' case (what we now
refer to as the nominative), and the 'oblique' cases, which they
regarded as deviations from the upright. They realized that another
factor in addition to time was involved in determining the form of
Greek verbs; namely, the completion or non-completion of the action
expressed by the form in question. They distinguished between the
active and the passive; and between transitive and intransitive verbs.
The Alexandrian scholars carried further the work of the Stoic
grammarians. And it was in Alexandria that what we now call the
'traditional' grammar of Greek was more or less definitively codified.
Unlike most of the Stoics, the Alexandrian grammarians were
* analogists'; and their search for regularities in language led them to
establish 'canons', or patterns, of inflexion. The grammar of
Dionysius Thrax (late second century B.C.) was, to the best of our
knowledge, the first comprehensive and systematic grammatical
description to be published in the western world. In addition to the
four Stoic parts of speech Dionysius recognized also the adverb, the
participle (so called because of its 'participation' in both nominal and
verbal characteristics), the pronoun, and the preposition. All Greek
words were classified in terms of case, gender, number, tense, voice,
mood, etc. (cf. chapter 7). Dionysius did not deal explicitly with
syntax, the principles according to which words were combined into
sentences. This part of the grammatical description of Greek was
carried out some three centuries later, less systematically, however, by
Apollonius Dyscolus (second century A.D.).
1.2.6 The Roman period

We have now traced briefly the development of grammar among the
Greeks. Less need be said about the work of the Latin grammarians.
It is a matter of common knowledge that in every sphere of Roman
scholarship, art and literature, Greek influence was supreme* From
the second century B.C., and in some cases earlier, the Roman aristo-
cracy enthusiastically adopted Greek culture and Greek methods of
education. Their children were brought up to speak, read and write
Greek as well as Latin, and frequently went to complete their educa-
tion in one of the great Hellenistic centres of philosophy and rhetoric.
It is hardly surprising therefore to find that the Latin grammarians
were almost wholly dependent on their Greek models. The influence
of both the Alexandrians and the Stoics can be seen in Varro's work
on the Latin language (first century B.C.). And at Rome, as in Greece,
grammatical studies remained subservient to philosophy, literary
criticism and rhetoric. The controversy between 'analogists' and
* anomalists' was kept alive and, with other grammatical points, was
the subject of a good deal of dilettante discussion. Caesar himself
wrote a grammatical treatise On Analogy (which he dedicated to
Cicero) in the midst of his military campaigns in Gaul.
The Roman grammarians followed their Greek models not only in
their general assumptions about language, but also in points of detail.
A typical Latin grammar was organized, as was the grammar of
Dionysius Thrax, in three sections. The first section would define the
scope of grammar as the art of correct speech and of the under-
standing of the poets, and would deal also with letters and syllables.
The second section would treat of the * parts of speech* and give, in
greater or less detail, the variations they underwent according to
tense, gender, number, case, etc. Finally there would be a discussion
of good and bad style, warnings against common 'faults' and
'barbarisms', and examples of the recommended 'figures of speech'.
In dealing with the ' parts of speech' the Latin grammarians made
only such minor modifications as the differences between Greek and
Latin forced to their attention. The fact that the two languages are
very similar in their general structure doubtless encouraged the view
that the various grammatical categories elaborated by the Greek
scholarsthe 'parts of speech', case, number, tense, etc.were
universal and necessary categories of language. This view was to be
maintained explicitly by medieval grammarians.
The later period of Latin grammatical scholarship, the period of
14 I. LINGUISTICS
Donatus (c. A.D. 400) and Priscian (c. A.D. 500), like the Alexandrian
period, was an age of classicism. The grammars of Donatus and
Priscian, intended as teaching grammars and used as such through the
Middle Ages and as late as the seventeenth century, set out to
describe not the language of their own day, but that of the 'best
writers', especially Cicero and Virgil, and thus perpetuated what I
have called the 'classical* fallacy in the approach to linguistic
description.
1.2.7. Medieval period

A dominant feature of the medieval period in Europe was the
important place occupied by Latin in the educational system. All
personal advancement, both secular and clerical, depended upon a
sound knowledge of Latin. For Latin was not only the language of the
liturgy and the scriptures, but also the universal language of diplo-
macy, scholarship and culture. Since it was now a foreign language
which had to be learnt at school for important practical, as well as
cultural, purposes, a large number of manuals were written to assist
the schoolboy in the mastering of it. Most of these were based upon
the grammars of Priscian and Donatus. And Latin was not merely a
foreign language; it was primarily a written language. In so far as it
was spoken, each country developed its own pronunciation. And this
fact could not but have reinforced the traditional view of the primacy
of the written language.
Many advances were made in the grammatical analysis of Latin by
medieval scholars and have become part of what we now think of as
traditional grammar. Far more important, however, than the details
of the grammatical analysis of Latin carried out by the scholars of the
later Middle Ages were the philosophical presuppositions which they
brought into the study of language.
The thirteenth century saw a flowering of scholarship in all its
branches: it was the period of the great scholastics, who, under the
influence of the newly-accessible works of Aristotle and other Greek
philosophers, set out to reduce all sciences, including grammar, to a
set of propositions whose truth could be demonstrated conclusively by
deduction from first principles. The scholastic philosophers, like the
Stoics, were interested in language as a tool for analysing the structure
of reality. It was therefore the question of meaning, or 'signification',
to which they attached the greatest importance. Indeed, so many
works were produced with the title ' The Modes of Signifying' (De
modis significandi) that the grammarians of the period are often referred
to collectively as 'modistae'. Inspired with the scholastic ideals of
science as a search for universal and invariant causes, they deliberately
attempted to derive the categories of grammar from the categories of
logic, epistemology and metaphysics; or rather, to derive the cate-
gories of all four sciences from the same general principles. But in
doing this, it should be noted, they took it for granted that the
grammatical categories to be found in the works of Donatus and
Priscian were in general valid. Their objection to Donatus and
Priscian (and this objection was in fact made explicitly) was not that
the Roman grammarians had inaccurately described the facts of Latin
grammar, but that they had not accounted for these facts scientifically,
that is by deducing them from their 'causes'.
It was the task of scientific, or 'speculative', grammar to discover
the principles whereby the word, as a 'sign', was related on the one
hand to the human intellect and on the other to the thing it repre-
sented, or 'signified'. It was assumed that these principles were
constant and universal. For how else could language be the vehicle of
true knowledge? According to the speculative grammarians the word
did not directly represent the nature of the thing it signified; it
represented it as existing in a particular way, or 'mode'as a sub-
stance, an action, a quality, etc.and it did this by having the forms
of the appropriate part of speech. Grammar was therefore a philo-
sophical theory of the parts of speech and their characteristic ' modes
of signifying'. (The term 'speculative' is worth noting. It must not
be taken in its modern sense, but in the more particular sense
deriving from the view that language is like a 'mirror', Latin
speculum, which gives a 'reflection' of the 'reality' underlying the
'phenomena' of the physical world. The Stoics had employed the
same metaphor.)
It is easy enough to say that the method of definition employed by
the scholastic grammarians is circular and their views on grammar
self-evidently false. They were certainly more ready than we are to
assume that the modes of signifying necessarily coincided with the
modes of 'being' and 'understanding'. But before we dismiss their
linguistic speculations as unworthy of our attention (as is often the
tendency) it would be as well to consider whether it is merely the
terminology of the time which we find unacceptable or incompre-
hensible. Statements such as the following are often quoted as if
obviously absurd and needing no further discussion: 'Grammar is
substantially the same in all languages, even though it may vary
l6 I. LINGUISTICS
accidentally'; * Whoever knows grammar in one language also knows
it in another so far as its substance is concerned. If he cannot, how-
ever, speak another language, or understand those who speak it, this
is because of the difference of words and their formations which is
accidental to grammar*. The first quotation is from Roger Bacon
(1214-94), and the second from an anonymous scholar of the same
period. We may be tempted to reject these statements out of hand on
the basis of our experience of foreign languages. Surely, we exclaim,
the grammatical differences between French and English, or Russian
and English, are not just * accidental' and unimportant; and we tend
to attribute the scholastic view of the universality of grammar to the
unique position occupied by Latin throughout the Middle Ages and
the low status of the vernacular languages, many of which were in any
case derived from, or strongly influenced by, Latin. The privileged
position of Latin was doubtless an important factor in the develop-
ment of universal grammar. But the scholastic view of language was
not abandoned at the Renaissance with the new interest in the
vernaculars and their use in literature. What the scholastic statements
which I have quoted mean, when they are stripped of their meta-
physical expression in terms of 'substance' and 'accidents', is merely
this: all languages will have words for the same concepts and all
languages will manifest the same parts of speech and other general
grammatical categories. This view may be true or false; we shall
return to this question in due course (cf. chapters 7 and 8). But it is
a view which, in the form in which it has just been put, would be
accepted by many of those who most vociferously proclaim their
emancipation from the fetters of scholasticism.
1.2.8 The Renaissance and after

The Renaissance scholars certainly thought that they were making a
radical break with the scholastic tradition. Petrarch and his followers
ridiculed the language of the schoolmen for its 'barbarism', and took
Cicero's usage as their model of good Latin style. From Cicero too
they derived their ideal of 'humanism'this term (Cicero's
humanitas) being synonymous with 'civilization' and opposed to
'barbarism'. Holding that the literature of classical antiquity was the
source of all ' civilized' values, they concentrated their energies upon
the collection and publication of the texts of the classical authors;
especially after the invention of printing in the late fifteenth century,
which made possible the wide and rapid distribution of accurate
texts. Once again grammar became an aid to the understanding of
literature and to the writing of 'good* Latin. Erasmus himself (in
1513) published a Latin syntax based on Donatus. Greek also became
the object of intense study, and, somewhat later, Hebrew. Thus it
was that the 'humanists' handed on to succeeding generations of
scholars the languages and literature of three cultures.
The vernacular languages of Europe began to attract the notice of
scholars even before the Renaissance: we have a seventh-century
grammar of Irish, a twelfth-century grammar of Icelandic and a
thirteenth-century grammar of Provencal, not to mention Aelfric's
comparison of Latin and Anglo-Saxon (tenth century) and a number
of observations about Basque which go back to the tenth century.
Several grammars of French were produced for travelling English-
men in the fourteenth and fifteenth centuries. With the Renaissance,
heralded in this respect by Dante's De vulgari eloquentiay interest in
the vernacular languages developed enormously, and grammars were
written in great numbers. In fact, the whole classical conception was
extended to the modern languages of Europe. Language still meant the
language of literature; and literature, when it became the object of
academic study in our schools and universities, continued to mean the
work of the 'best authors' writing in the accepted genresDante in
that of the Virgilian epic, Milton in a more Homeric strain, Racine in
the vein of Sophocles, and so on. It is true that a more satisfactory
academic approach to literature has developed nowadays, and authors
are no longer classified by the normative canons of Alexandria and the
Renaissance. Yet the study of grammar in the language departments
of our schools and universities still tends to be classical in spirit.
The ideals of 'speculative' grammar were revived in France in the
seventeenth century by the teachers of Port Royal. In 1660 they
published their Gratnmaire genirale et raisonneey the aim of which was
to demonstrate that the structure of language is a product of reason,
and that the different languages of men are but varieties of a more
general logical and rational system. The Port Royal grammar had an
immense influence both in France and abroad, and the 'Age of
Enlightenment' was to see the publication of many such works. All
these 'rational' grammars worked within the confines of the classical
tradition and produced no new linguistic theories.
No more striking evidence of the endurance of the classical tradi-
tion in the study of language could be given than the definitions to be
found in the most recent editions (1932) of the dictionary and
grammar of the French Academy, which, since its foundation by
l8 I. LINGUISTICS
Richelieu (in 1637), has been charged with the task of establishing
authoritatively the vocabulary and grammar of French. Grammar is
defined as ' the art of speaking and writing correctly'; its object is to
discover the relations holding between the elements of language,
whether these relations be 'natural' or 'conventional'; the gram-
marian's task is to describe 'good usage', that is the language of
those educated persons and writers who write 'pure' French, and to
defend this ' good usage' from ' all causes of corruption, such as the
invasion of the vocabulary by foreign words, technical terms, slang,
and those barbarous expressions which are constantly being created
to satisfy the dubious needs of trade, industry, sport, advertising,
etc.'; as for the rules of grammar, these are not arbitrary, but 'derive
from the natural tendencies of the human mind'.
There exists no official body whose function it is to legislate
authoritatively about English usage. Nevertheless, the literary and
philosophical prejudices embodied in the French Academy's defini-
tions quoted above, the origin of which can be traced in Greece and
Alexandria, are hardly less prevalent in the English-speaking countries
than they are in France.
The true followers of the classical and scholastic grammarians are
not those who seek to preserve intact the whole framework of classical
grammar, but rather those who carry out free and critical inquiry into
the role and nature of language writhin the context of present-day
scientific thinking, and with the more extensive knowledge of
languages and cultures that is now available. As we shall see in the
course of this book, many of the insights into the structure of
language obtained by the classical grammarians were valuable and
revealing, but demand reformulation in more general and more
empirical terms.
1.2.9 Wider influence of the Greco-Roman tradition

Throughout this section the term 'traditional grammar' has been
employed, as is customary, to refer to that tradition of linguistic
analysis and linguistic theory which originated in Greece, was
further developed in Rome and in medieval Europe, and was extended
to the study of the vernacular languages at the Renaissance and after-
wards. This, the Greco-Roman tradition of linguistic analysis (as it
may be called for convenience), influenced the descriptions of certain
non-European languages even before the Renaissance. The grammar
of Dionysius Thrax was translated into Armenian in the fifth century
A.D., and somewhat later into Syrian. Subsequently, the Arab
grammarians drew upon the Syrians, and they also came more directly
into contact with the Greco-Roman tradition in Spain. And the
Hebrew grammarians were influenced by the Arabs. So it was that
the native grammatical descriptions of Armenian, Syrian, Arabic and
Hebrew were already strongly influenced by the Greco-Roman
tradition even before these languages attracted the attention of
European scholars at the Renaissance.
1.2.1 o The Indian tradition

Before turning our attention to the period of * Comparative Philology*
in the next section, we must look briefly at one other tradition of
grammatical analysis which has exercised considerable influence upon
the development of modern linguistics: this is the Indian, or Hindu,
tradition.
The Indian grammatical tradition is not only independent of the
Greco-Roman but also earlier, more diverse in its manifestations and
in some respects superior in its achievements. Panini (? fourth
century B.C.), acknowledged as the greatest of the Indian grammarians,
mentions a large number of predecessors, and it may be assumed that
he is working in a tradition which started some centuries before him.
As for the diversity and extent of Indian grammatical work: about
twelve different schools of grammatical theory have been recognized
in the Indian tradition (most, if not all, to some degree dependent on
Panini), and there are about a thousand separate grammatical works
preserved.
Although Indian grammar and Greco-Roman grammar were, as
far as we know, independent of one another, both in their origins and
in their development, there are certain points of similarity. In India,
as in Greece, there was a controversy about the 'natural* or 'con-
ventional' status of language; and just as the Alexandrian scholars,
dealing with the classical texts of the past, produced glossaries and
commentaries to explain words or constructions which were no
longer current in contemporary Hellenistic Greek, so the Indian
grammarians compiled glossaries and commentaries on the sacred
Hindu texts, the earliest of which, the Vedic hymns, were composed
some centuries earlier; and the distinction between what we may call
'nouns' and 'verbs' in Sanskrit was drawn in much the same way as
it was drawn by Plato for Greek, that is to say in terms of a distinction
between 'subject' and 'predicate'. (The Sanskrit grammarians also
20 I. LINGUISTICS
recognized two other parts of speech, which we may translate as
'preposition' and 'particle'.)
There are two respects in which Indian linguistic work may be held
to be superior to Western traditional grammar: first in phonetics, and
second in the study of the internal structure of words. Indian gramma-
tical studies seem to have had their origin in the necessity of preserving
intact, not only the text, but also the pronunciation of the Vedic
hymns, the precise and accurate recitation of which is held to be
essential to their efficacy in Hindu ritual. The Indian classification of
speech sounds was more detailed, more accurate and more soundly
based upon observation and experiment than anything achieved in
Europe (or elsewhere as far as we know) before the late nineteenth
century, when the science of phonetics in Europe was in fact strongly
influenced by the discovery and translation of the Indian linguistic
treatises by Western scholars. In their analysis of words the Indian
grammarians went well beyond what might be thought necessary for
the original purpose of preserving the language of the sacred texts.
And Panini's grammar is not in fact specifically devoted to the
language of the Vedic hymns, but to the language of his own day.
Panini's grammar of Sanskrit has frequently been described, from
the point of view of its exhaustiveness (within the limits which it sets
itself: i.e. mainly with regard to the structure of words), its internal
consistency and its economy of statement, as far superior to any
grammar of any language yet written. The main part of the grammar,
which is a highly technical work and can be interpreted only with the
aid of the commentaries of his successors, consists of about 4,000
rules (some of them extremely short) and lists of basic forms (' roots'),
to which reference is made in the rules. The rules are ordered in
sequence in such a way that the scope of a particular rule is defined or
restricted by the preceding rules. Further economy is achieved by the
use of abbreviations and symbols.
As we shall see in the following section, the discovery of Sanskrit by
Western scholars was one of the principal factors in the development
of comparative philology in the nineteenth century. This was not
simply a matter of being brought into contact with the Sanskrit
language, but also of becoming acquainted with the Indian gram-
matical tradition. There are many aspects of nineteenth-century
linguistics which are clearly derived from the practice or theory of the
Indian grammarians. But the influence of Panini's principles
(exhaustiveness, consistency and economy) is to be seen even more
clearly in some of the most recent work in linguistics.
I. LINGUISTICS 21
1.3 Comparative philology

i. 3. i * Language-familiesy
It is a well-known fact that different languages resemble one another
in different degrees. Speaking very generally, we may say that
resemblances between languages are of two kinds: resemblances of
vocabulary and resemblances of grammatical structure. If we compare
English and German, for example, we shall find many words, in all
parts of the everyday vocabulary of the two languages, which are
similar in form and meaning (son: Sohn, mother: Mutter, brother:
Bruder, six: sechs} seven: siebenf have: habe, must: muss, can: kann,
etc.). There are far fewer instances of vocabulary resemblances
between English and Russian; and very few indeed between English
and Turkish, or English and Chinese (if we discount ' international'
scientific terminology). Moreover, in the case of the languages I have
mentioned, the degree of vocabulary resemblance is supported by
the degree of grammatical resemblance. German and English are
more alike in their grammatical structure than are Russian and
English, and still more alike than are Turkish and English, or Chinese
and English. These facts are a matter of experience and observation,
immediately obvious to anyone who sets out to learn, or simply
examines, the languages in question. The facts are explained by saying
that English and German are closely related; that English and
Russian are more distantly related; and that English and Turkish, or
English and Chinese, are (as far as we can tell) totally unrelated.
In this context the term ' relationship' is being used, as is customary
in linguistics, to refer to a historical, or 'genetic', relationship. To
say that two languages are related is equivalent to saying that they
have developed from some earlier single language. This is otherwise
expressed by saying that they belong to the same family of languages.
Most of the languages of Europe and many languages of Asia belong
to what is called the Indo-European family. Within this wider family,
however, there are many different branches, or sub-families:
Germanic, comprising German, English, Dutch, Swedish, etc.;
Slavonic, comprising Russian, Polish, Czech, etc.; Romance (the term
used for languages deriving from Latin), comprising French, Italian,
Spanish, etc.; Greek; Indo-Iranian, comprising Sanskrit (with its
medieval and modern descendants), Persian, etc.; Celtic, comprising
Gaelic (Irish and Scottish), Welsh, Breton, etc. This is far from being
a complete list of the distinct branches, let alone of the individual
22 I. LINGUISTICS
languages, but it will suffice to give the reader an idea of the extent
of the Indo-European family. Other major language-families (to
mention but a few) include Semitic (Hebrew, Arabic, etc.), Finno-
Ugrian (Finnish, Hungarian, etc.), Bantu (Swahili, Kikuyu, Zulu,
etc.), Altaic (Turkish, etc.), Sino-Tibetan (Chinese, Tibetan, etc.)
and Algonquian (including a number of American Indian languages).
To have established the principles and methods used in setting up
these, and other, language-families and, what is more important, to
have developed a general theory of language change and linguistic
relationship was the most significant achievement of nineteenth-
century linguistic scholarship. The term 'comparative philology',
which I shall use to refer to this period of linguistics and to its
characteristic aims and methods, is one which came to replace the
earlier and less satisfactory * comparative grammar' in the course of
the nineteenth century and which, though less commonly used these
days by linguists themselves (who tend to prefer * comparative and
historical linguistics'), is not infrequently met in general books on
language and, like many other unsuitable terms, has been perpetuated
in the titles of university chairs and departments and of prescribed
courses of study. The important thing to notice is that ' philology' in
this context has no connexion wTith textual criticism and literary
scholarship, quite separate branches of study for which the term
'philology* is also used, though less commonly in Great Britain and
America than in Europe.
1.3.2 'Scientificy linguistics

It is usually said that the nineteenth century saw the birth of the
scientific study of language in the western world. And this statement is
true, if we give to the term 'scientific* the sense it generally bears
today; it was in the course of the nineteenth century that facts of
language came to be carefully and objectively investigated and then
explained in terms of inductive hypotheses. It should not be forgotten,
however, that this conception of science is of quite recent development.
The speculative grammar of the scholastics and of their philosophical
successors at Port Royal was scientific according to their under-
standing of what constituted sure knowledge. Their causal demon-
strations of why languages were as they are were based on principles
assumed to be universally valid. The difference between this way of
looking at linguistic questions and that which resulted in the im-
mensely fruitful period of comparative philology was not so much
1.3. COMPARATIVE PHILOLOGY 23
that the latter was more respectful of the ' facts' and more careful in
its observation and collection of them (this is effect, rather than
cause), but by the end of the eighteenth century there had developed
a general dissatisfaction with a priori and so-called ' logical' explana-
tions and a preference for historical reasoning.
1.3.3 Evolutionary point of view

The change of outlook that led to the adoption of the historical point
of view was general and was not confined to the study of language. The
abandonment of a priori reasoning had first taken place in the so-
called ' naturaly sciences. Later this attitude was extended to the study
of human institutions as well. It was observed that all human institu-
tionslaws, customs, religious practices, economic and social groups,
and languageswere continually changing, and it was no longer felt
to be satisfactory to explain their state at a particular time in terms of
abstract principles, but rather in terms of their development from
some previous different state by adaptation to changing external
conditions. The 'providential' theory of history of the general
Christian tradition had been increasingly challenged, and was
replaced by evolutionary and secular theories of human development.
1.3.4 A wider range of languages

As we have seen, the Renaissance had already brought about a far
greater interest in the contemporary languages of Europe, in addition
to promoting the more intensive study of Greek, Latin and Hebrew.
From the sixteenth century onwards, dictionaries and collections of
sample texts came to be published exemplifying more and more
languages, including (in a limited way) some of the languages of the
Middle and Far East and even of America. Various attempts were
made to group languages into families; but most of these attempts
were vitiated by the assumption that, since Hebrew was the language
of the Old Testament, it was the source from which all other languages
were to be derived.
1.3.5 Romanticism
Of particular importance in the study of language was the new spirit
of romanticism which developed at the end of the eighteenth century,
particularly in Germany, as a reaction against the classicism and
24 I. LINGUISTICS
avowed rationalism of the previous age. The leaders of the romantic
movement rejected the view that the canons of literary excellence had
been fixed for all time by the classical tradition. Their interest in
German antiquities led to the publication and study of texts and
glossaries of the older Germanic languages (Gothic, Old High
German, and Old Norse). Herder (i744-1803) maintained that there
was an intimate connexion between language and national character.
Following him, the statesman and polymath Wilhelm von Humboldt
(1767-1835) gave more definite form to this thesis, saying that each
language had its own distinctive structure, which reflected and
conditioned the ways of thought and expression of the people using
it. This belief in the connexion between national language and
national character, which took firm root in Germany, later gave rise
to a good deal of extravagant and mischievous speculation in which
the notions of 'race* and 'language' were inextricably confused
(notably with respect to the term 'Aryan'). At the period we are
discussing it promoted not only an interest in the earlier stages of the
German language, but a more general enthusiasm for linguistic
variety itself and a readiness to consider all languages, however
'barbarous', on their own terms. It is no accident that German
scholars were pre-eminent among the founders of comparative
philology.
1.3.6 Discovery of Sanskrit

At the end of the eighteenth century it was discovered that Sanskrit,
the ancient and sacred language of India, was related to Latin and
Greek and to other languages of Europe. This discovery was made
independently by several scholars. Of these the most influential was
the British orientalist, Sir William Jones, who declared (in 1786), in
words that have since become famous, that Sanskrit bore to Greek and
Latin ' a stronger affinity both in the roots of verbs and in the forms of
grammar than could possibly have been produced by accident: so
strong indeed that no philologer could examine them at all without
believing them to have sprung from some common source, which
perhaps no longer exists'. It is not difficult to recapture the feeling of
excitement and wonder which this discovery kindled in the minds of
classically-trained Western scholars. They had long been aware of the
similarities that existed between Greek and Latin. But they also knew
of the strong and lasting cultural and political ties which had united
Greece and Rome; and this might seem to account for the resem-
blances between the two languages. By the end of the eighteenth
century enough information was available about languages of different
structures for scholars to realize immediately that the resemblances
observed between the classical languages of Europe and Sanskrit were
so striking as to demand an explanation.
1.3.7 Importance of Indo-European languages

Once the imagination of scholars had been fired by discoveries in the
field of the Indo-European languages and their permanent interest
secured by the development of what were felt to be sound principles
of comparison, it was natural that they should direct their attention to
the determination of other language families, some of which I have
mentioned above. However, the Indo-European family has, and
perhaps always will have, pride of place in the historical and compara-
tive study of languages. This is not because of any intrinsic qualities
of the Indo-European languages themselves. The reason is simply
that many of the Indo-European languages have very ancient written
records, going back hundreds and even thousands of years. Since
related languages are for the most part divergent forms of some earlier
single language, the further back we go in time the less difference will
we find between the languages being compared. Although some of the
relationships within the Indo-European family could be demonstrated
from the evidence of the modern spoken languages, the details of
these relationships could certainly not have been worked out without
the help of the older texts.
1.3.8 'Loan-words*
So far I have talked as though a certain degree of resemblance in
vocabulary and grammara degree of resemblance greater than can
reasonably be attributed to chancewere a sufficient proof of
relationship. This is an over-simplification; it takes no account of what
is referred to technically, though perhaps misleadingly, as borrowing.
It is a well-known fact that languages in geographical or cultural
contact borrow' words from one another quite freely; for words tend
to travel across geographical and linguistic boundaries together with
the object or custom to which they refer. Much of the resemblance
in the vocabularies of different languages may therefore be due to
their having borrowed words either from one another or from some
third language. We have only to think of the vast numbers of words
26 I. LINGUISTICS
of Greek and Latin origin in the vocabularies of modern European
languages to see the force of this point. (It may be noted that, if we
use the term ' borrowing' here to refer, not only to words which have
been taken directly from the classical languages, but also to words
which have been created in recent times by the deliberate combination
of parts of Greek and Latin words, then we must say that most
modern scientific terms, as well as the names of such modern inven-
tions as the telephone, television, automobile, cinema, etc., have been
indirectly borrowed from Greek and Latin.) Borrowing from the
classical languages explains many of the more obvious similarities in
the words of different modern European languages. In addition, these
languages have all borrowed from one another in different degrees and
at different periods of history, and continue to do so. With our know-
ledge of the languages involved and of the general development of
European culture, we have little difficulty in recognizing most of the
loan-words in the vocabularies of modern European languages. It is
clear, however, that if we were unable to identify the loan-words, and
so discount them in our calculations, we should very probably over-
estimate the degree of relationship holding between the languages in
question. The founders of comparative philology were well aware
that words pass easily from one language to another, but had no sure
way of distinguishing loan-words from the rest of the vocabulary. For
this reason, they tended to rely particularly upon grammatical
resemblances as evidence of linguistic relationship and to be cautious
in invoking resemblances between words except where these were
found in the ' basic' vocabulary of the languages concernedthe most
essential words, those which are learned earliest and used constantly.
We now know that the grammar of a language can also be influenced
by that of another language with which it is in contact; and, although
it is undeniable that what may be called 'cultural' words are more
prone to borrowing than others, it is doubtful whether there is any
particular set of words so basic in a language that they are immune
from replacement by borrowing. However, this question is less
important than it might seem to be. For, in the case of those languages
which have been most successfully investigated from a comparative
point of view, the vague notion of resemblance has been abandoned
in favour of the more precisely definable concept of systematic
correspondence.
1.3.9 * Grimm's law '

Instances of partly systematic correspondences between the sounds
of equivalent words in different languages had been noted by the
earliest comparative philologists. In 1822, Jacob Grimm, following
the Danish scholar, Rasmus Rask, pointed out, for example, that the
Germanic languages frequently had: (i) an f where other Indo-
European languages (e.g. Latin or Greek) had a p; (ii) a p where other
languages had a b; (iii) a 'th-sound' where other languages had a t;
(iv) a t where other languages had a d; and so on according to the
following (partial and somewhat simplified) table for Gothic (the
earliest Germanic language for which we have substantial records),
Latin, Greek and Sanskrit:
Table 1
Gothic f P b e t d h k g
Latin P b f t d t c g h
Greek P b ph t d th k g kh
Sanskrit P b bh t d dh j h
(Notice that 8 in the transcription of Gothic denotes the voiceless

fricative 'th-sound', found in such English words as thick, thin, etc.,
as distinct from the corresponding voiced fricative of e.g. then, there,
etc.; and that th, ph, kh, dh, bh used in the transcription of Greek and
Sanskrit denote aspirated stops: for the distinction between fricatives
and aspirated stops, cf. 3.2.6.) Instances of words which illustrate
some of these correspondences are: Go. fotus, L. pedis, Gk. podos,
Skt. padas ('foot'); Go. taihun, L. decem (n.b. the Latin c represents
a 'k-sound'), Gk. deka, Skt. dasa ('ten'). Grimm explained such
correspondences by postulating a 'sound-shift' in a prehistoric period
of Germanic whereby the original Indo-European 'aspirate' con-
sonants (bh, dh and gh) became unaspirated (b, d and g), the original
voiced consonants (b, d and g) became voiceless (p, t and k) and the
original voiceless consonants (p, t and k) became 'aspirates' (f, 8 and
h). It should be observed that the formulation of the correspondences
given here is not quite accurate as a restatement of what Grimm said
about the Germanic 'sound-shift': it substitutes the modern terms
'voiced' and 'voiceless' (cf. 3.2.6) for the traditional terms of classical
grammar employed by Grimm and, more important, it distinguishes
28 I. LINGUISTICS
phonetically between the original Indo-European and the later
Germanic * aspirates'; moreover, it leaves out of consideration the
later 'High German Sound-Shift', which Grimm regarded as part of
the same general process working itself out over many centuries.
Also, our formulation is imprecise in that it gives only a partial
account of the correspondences between Gothic and the other
languages. However, it conveys the essence of what later came to be
known as * Grimm's law'the most famous of all the * sound-laws'
established by comparative philologistsand is sufficient for our
present purpose.
Grimm and his contemporaries saw that there were many excep-
tions to such generalizations as those we have summarized in the
table of correspondences given above. They observed, for instance,
that, although the word for ' brother' was completely * regular' in the
development of the consonants (cf. Go. brodar: L. frater; b = f,
9 = t), the word for 'father* was only partially so (cf. Go. fadar:
L. pater; f = p, but d = t). They were not troubled by exceptions of
this kind, for they had no reason to believe that sound-change was
'regular'. Grimm himself remarked: 'The sound-shift succeeds in
the majority of cases, but never works itself out completely in every
individual case; some words remain in the form they had in the older
period; the current of innovation has passed them by.'
i.3.10 The 'Junggrammatiker'

Some fifty years later, a strikingly different principle was proclaimed
by a group of scholars who saw themselves as revolutionaries and
rejoiced in the label which their opponents contemptuously applied
to themthe 'Junggrammatiker' (the 'Young Grammarians' or
'Neogrammarians'). As expressed by Wilhelm Scherer (1875), tê
principle is this: 'The sound changes which we can observe in docu-
mented linguistic history proceed according to fixed laws which
suffer no disturbance save in accordance with other laws.' The claim
that all sound-change could be accounted for by laws which operate
without exceptions was highly controversial, but now appeared more
reasonable than it would have done in Grimm's day. Between 1820
and 1870 an enormous amount of research had been devoted to the
investigation of sound-change in the different branches of the Indo-
European languages; and at the end of this period various brilliant
discoveries were made which had the effect of explaining many of the
apparently 'irregular' correspondences noticed by earlier scholars.
1.3.11 ' Verner }s law' and other ' sound-laws '

In 1875, the Danish scholar, Karl Verner, published a particularly
influential article in which he was able to demonstrate that corre-
spondences such as Gothic d = Latin t (in, e.g. Go. fadar = L. pater),
though exceptional in terms of * Grimm's law', were perfectly regular
if this * law' was modified to take account of the place of the accent in
corresponding Sanskrit words (marked in the following table by ' ' ' ) :
Table 2
Sanskrit bhratar- pitdr-

Gothic brodar fadar
Latin frdter pater
Verner assumed that Sanskrit had preserved the place of the earlier
Indo-European word-accent, and that the Germanic 'sound-shift'
had taken place before the accent was shifted to word-initial position
in some prehistoric period of Germanic. On this assumption, occur-
rences of d (which, between vowels, may have been a voiced fricative;
i.e. pronounced like the th of English, father) could be explained by the
following Maw': the voiceless 'aspirates' resulting from 'Grimm's
law' (f, th, h) were preserved if the preceding syllable bore the accent
(hence Go. brodar, etc.), but otherwise were voiced (as b, d, g,
respectively: hence Go. fadar, etc.). It may be added that the identifi-
cation of the medial consonants in the English words father and
brother is due to developments particular to English; they were
distinguished in Old English, and they are distinguished in modern
German (cf. Voter \ Bruder), although they have changed phonetically
as a result of the 'High German Sound-Shift', which was compara-
tively recent and applied only to German.
In addition to ' Verner's law', a number of other important 'sound-
laws' were established at about the same time which accounted
successfully for some of the more troublesome exceptions to the
general correspondences holding between the various Indo-European
languages. The general effect of these 'sound-laws' was to give
scholars a much clearer idea of the relative chronology of develop-
ments wTithin the different branches of Indo-European and to increase
their confidence in the principle of regularity in sound-change.
Although this principle met with violent and widespread opposition
30 I. LINGUISTICS
when it was first proclaimed by the ' Junggrammatiker' it came to be
accepted by the majority of comparative philologists towards the end
of the nineteenth century.
The methodological significance of the principle of regularity in
sound-change was tremendous. By concentrating their attention on
the exceptions to the * laws' which they established it forced scholars
either to formulate these * laws' more precisely (in the way that
Grimm's law* was made more precise by 'Verner's law') or to
provide a satisfactory explanation for those words which had not
developed in accordance with the * laws' whose conditions they
appeared to fulfil.
i. 3.12 Exceptions explained by ' borrowing'

Many apparent exceptions to the * sound-laws' could be explained as
loan-words, borrowed from some neighbouring related language or
dialect after the operation of the Maw' which they appeared to violate.
An example is one of the Latin words for 'red', rufus, which has an f
where comparison with words in other Indo-European languages and
the relevant * sound-law' set up for Latin (we shall not go into the
details here) would lead us to expect a b. And there is another Latin
word for 'red', ruber, which has the expected b (it has a different
ending; but this does not concern us here). The existence of the
* doublets', and the difference between them, is plausibly explained
on the assumption that Latin borrowed rufus from one of the
neighbouring closely related dialects in which, as we know from other
evidence, the f was a regular development.
i. 3.13 Role of analogy

The second main factor which the ' Junggrammatiker' invoked to
account for exceptions to their sound-laws was what they referred to
as analogy. It had long been recognized that the development of
language had frequently been influenced by the tendency to create
new forms 'by analogy with' the more common or more regular
patterns of formation in the languageby the same tendency that
underlies, for instance, the English-speaking child's production of
forms like fixed, goed, tooths (for flew, went and teeth). Since this
tendency was thought to have the effect of introducing into the
language 'incorrect' forms, it was regarded as one of the factors
responsible for the ' corruption' of the language in a degenerate and
illiterate age. And it was thought that, just as parents and teachers
should correct the false analogical formations of children, so should
grammarians correct the 'false analogies' of adult speakers which
threatened to gain wider currency. With the increased attention that
was devoted to the historical and prehistorical development of the
classical and vernacular languages of Europe in the course of the nine-
teenth century, it came to be realized that 'analogy* was a major
factor in the development of languages at all periods and could not be
attributed merely to periods of decline and corruption.
The influence of analogy and its explanatory force in the case of
exceptions to the 'sound-laws' can be illustrated by reference to a
certain class of nouns in Latin. Consider, for example, the nominative
and genitive singular of the Greek word genos, the Latin genus and
the Sanskrit janas (words which are related to the English kin and
which can be translated as 'family', 'race', 'species', etc.): Gk. genos,
genous, L. genus, generis, Skt. janas, janasas. On the basis of the corre-
spondences exemplified in these (and many other) words the following
sound-laws can be set up for the development of the reconstructed,
'original' Indo-European *s:
(i) in Sanskrit, the 'original' *s is preserved (in all cases
relevant to the present example);
(ii) in Greek, the 'original' *s
(a) is preserved before and after consonants and in word-
final position,
(b) becomes h in initial position (cf. Gk. heptd, L. septem,
Skt. sapta, Eng. seven), and
(c) disappears between vowels (hence *genesos - geneos,
-> genous);
(iii) in Latin, the 'original' # s
(a) is preserved before and after consonants, and also in
word-initial and word-final position, and
(b) becomes r between vowels (hence Agenesis -> generis).
The forms of Gk. genos and L. genus are therefore ' regular' in their
development. The asterisk is used, in accordance with the customary
practice of comparative philologists, to distinguish 'reconstructed'
sounds or forms from the sounds or forms actually attested in the
recorded texts. Although we are concerned here only with the
development of Indo-European # s, the examples are 'regular' also in
terms of the sound-laws set up for the other consonants, as well as for
the vowels, of Indo-European. But in both Greek and Latin there are
32 I. LINGUISTICS
4
exceptions' to the 'sound-laws'. To take a Latin example: in the
classical authors we find a difference in the form employed for the
nominative singular of the word for 'honour'in general, the earlier
authors have honos and the later authors honorwhereas they all agree
in the form of the other 'cases', honoris, honorem, etc. The forms
honoris, honor em, etc., are 'regular' (from *honosis, *honosem, etc.) in
terms of the 'laws' given above; so also is honos (*s is preserved in
word-final position). Now there are many nouns in Latin in which the
r has developed 'regularly' from an Indo-European *r: cultor,
cultoris, cultorem, etc. The form honor may be explained therefore as
an ' analogical' replacement of honos, which had the effect of bringing
honor: honoris, etc. into line with cultor: cultoris. In fact, the word
honor is only one example from a large class of Latin nouns of the
same type (cf. amor, 'love'; labor, 'work'; timor, 'fear'; etc.); to be
contrasted with them are such monosyllabic nouns as fids: floris
('flower'), mos: moris ('custom'), etc., in which the 'regular' s was
preserved in the nominative singular of the classical Latin forms. (For
those readers who know Latin it may be pointed out that the account
of the development of these words given here has been simplified
somewhat in that it has made no reference to the length of the vowel
o in the endings; 'analogy' is relevant here also.)
Some of the 'laws' set up by the 'Junggrammatiker' and their
followers were of extraordinarily limited scope; many implausible
'analogies' were suggested; and forms were frequently assumed to
have been borrowed without any indication of the dialect from which
they might have been borrowed. In the light of the research carried
out in the field of historical and comparative linguistics over the last
eighty years, we can now see that the distinction made by the
' Junggrammatiker' between' sound-laws',' borrowing' and' analogy'
was drawn too sharply: certain sound-changes may have their origin
in one word or a small set of words ' borrowed' from a neighbouring
dialect and may then spread 'by analogy' into a wider class of words.
After this has happened, it may be possible to describe the change
('macroscopically' as it were) by means of a general 'law'. But this
does not mean that the change in question has taken place as a result
of some 'law' operating upon language from without.
1.3.14 Positivist outlook of nineteenth-century linguistics

The notion of' evolution', although it was by no means a new concept,
was one of the dominant ideas of nineteenth-century thought. It was
an idea which was enthusiastically taken up by the romantic move-
ment in its reaction to the classical tradition. With the publication of
Darwin's Origin of Species (1859) and the substitution of the principle
of natural selection for the notion of purpose or design, not only was
evolutionary biology offered the possibility of adopting the prevailing
mechanistic, or positivist, outlook of the ' natural' sciences, but the
whole idea of ' evolution' was thought to have been put on a sounder
' scientific' footing. There are many particular features of nineteenth-
century linguistic thought which can be attributed to the influence of
evolutionary biology which need not be dealt with here. What must
be stressed is the fact that the apparent success of the positivist out-
look in biology tended to promote the search for ' laws' of' evolution'
in all the social sciences. In their attempt to construct a theory of
language-change on the basis of what were conceived as the sound
positivist principles of the * exact' sciences the ' Junggrammatiker'
were merely falling into line with the social scientists of the time.
Contemporary linguistics is no longer committed to a positivist con-
ception of science; and, as we shall see, it is no longer predominantly
concerned with the 'evolution' of languages.
1.3.15 Comparative philology and general linguistics

Comparative linguistics (as a branch of general linguistics) is an
explanatory science. It sets out to explain the evident fact that
languages change and that different languages are related to one
another in different degrees. The changes that languages undergo and
the different degrees of relationship between languages are accounted
for in terms of hypotheses which, like any other scientific hypotheses,
are subject to revision as a result of the discovery of new evidence or
of the adoption of a new way of looking at, and systematizing, the
evidence. The 'Indo-European' hypothesis has been continually
modified for both of these reasons. We now give a rather different
interpretation to the term ' evolution' from that given to it by certain
nineteenth-century scholars; we may understand the terms 'sound-
law', 'reconstruction' and 'analogy' somewhat differently; we may
recognize more clearly than our predecessors that language-change is
not simply a function of time, but also of social and geographical con-
ditions ; and we may admit that languages can, under certain conditions,
' converge' as well as ' diverge' in the course of time. However, none
of these modifications is sufficient to invalidate completely either the
methods or the earlier conclusions of comparative linguistics.
34 I. LINGUISTICS
Since we are not concerned in this book with the history of
linguistics for its own sake, we shall not go further into the principles
of the comparative method as these were elaborated in the course of
the nineteenth and twentieth centuries. We will conclude this section
by making explicit the main ways in which comparative philology has
contributed to the formation of the attitudes and assumptions
characteristic of contemporary theoretical linguistics.
One of the most immediate and most important effects of the nine-
teenth-century concern with the 'evolution* of languages was the
realization that developments in the forms of words and phrases in the
written texts and inscriptions of the past could generally be explained
on the basis of attested or postulated changes in the corresponding
spoken language (in terms of * sound laws'). The earliest comparative
philologists inherited the classical view, that the written language was
in some sense prior to the spoken, and continued to describe sound-
change in terms of changes in the constituent 'letters' of words.
However, it was soon appreciated that any systematic account of
language-development must give theoretical and practical recognition
to the principle that letters (in an alphabetic writing-system) are
merely symbols for the sounds in the corresponding spoken language.
As we shall see in the following section, it is one of the fundamental
assumptions of modern linguistics that sound, not writing, is the
primary medium of language. Comparative philology gave a powerful
impetus to the development of phonetics (which was also influenced
by the theories of the Indian grammarians: cf. 1.2.10), which
contributed, in its turn, to the formulation of more general and more
satisfactory 'sound-laws'.
No less important was the gradual development, from the middle
of the nineteenth century, of a more correct understanding of the
relationship between' languages' and' dialects'. Intensive study of the
history of the classical and modern languages of Europe made it
quite clear that the various regional' dialects', far from being imperfect
and distorted versions of the standard literary languages (as they were
frequently thought to be), had developed more or less independently.
They were no less systematicthey had their own regularities of
grammatical structure, pronunciation and vocabularyand they were
no less suitable as tools for communication in the contexts in which
they were used. It became clear, in fact, that the differences between
'languages' and closely-related 'dialects' are for the most part
political and cultural, rather than linguistic. From a strictly linguistic
point of view, what are customarily regarded as ' languages' (standard
Latin, English, French, etc.) are merely ' dialects' which, by historical
'accident', have become politically or culturally important. From
this point of view, for example, it was linguistically ' accidental' that
the ' dialect' of Rome and the surrounding area should have spread
with the growth of the Roman empire and become the ' language' we
call Latin: there is nothing in the structure of Latin itself to account
for this development. Of course, the use of a particular * dialect* for
literature, administration, philosophy and a wide range of other
purposes and activities may have the result that this 'dialect* will
develop a commensurately wide vocabulary, incorporating all the
distinctions necessary for it to operate satisfactorily in this way. This
is a different matter; and it is the effect, not the cause, of the impor-
tance achieved by the speech of a particular community. In general,
the standard languages of various countries have originated in the
'dialects' spoken by the socially-dominant or governing classes in
those countries.
A further point that has become clear as a result of the investigation
of regional 'dialects' (in the branch of linguistics referred to as
dialectology, or dialect-geography) is the impossibility of drawing a
sharp line of demarcation between 'dialects' of the same or neigh-
bouring languages. In those areas of the world where there have been
frequent changes of political boundaries or where the principal lines
of trade and communication cross political boundaries, what is
generally regarded as a dialect of one language may shade more or less
imperceptibly into a dialect of another. For example, there are
dialects spoken on both sides of the Dutch-German border which are
equally close to (or equally remote from) both standard Dutch and
standard German. If we feel that they must be dialects of either the
one or the other language, we are victims of the traditional view of the
relationship between 'language' and 'dialect'. It may be added that
judgements on questions of this kind are only too frequently influenced
by political or nationalist prejudice.
The assumption that all languages have the same grammatical
structure (cf. 1.2.7)ls n o longer generally accepted by linguists. One
reason for its abandonment derives from the demonstration by
nineteenth-century comparative philologists that all languages are
subject to continuous change; in particular, that classical Greek and
Latin were, from a linguistic point of view, merely stages in a process
of continuous development, and that much of their grammatical
structure could be accounted for in terms of reduction or expansion
of an earlier system of grammatical distinctions. It was realized that
36 I. LINGUISTICS
different languages, and different chronological stages of the same
language, might vary considerably in grammatical structure; and it
was no longer possible to assert that the traditional framework of
grammatical categories was essential to the functioning of human
language. This conclusion was reinforced by the investigation of a far
wider range of languages than those that were accessible to earlier
scholars who had maintained the universal validity of traditional
grammatical theory. As we shall see, current linguistic theory rests on
far more general assumptions than those of traditional grammar.
1.3.16 * Analogy' and ' structure'

The reader cannot fail to have been struck by the fact that, whereas the
traditional grammarian regarded 'analogy* as the principle of
regularity in language, the comparative philologist of the late nine-
teenth century tended to look upon it as one of the main factors which
inhibited the 'regular* development of language (cf. 1.2.3). This
paradox can now be resolved. If we are primarily concerned with the
establishment of a theory of language-change, as were most of the
nineteenth-century scholars, we may overlook the fact that, not only
the differences, but also the identities in the speech of successive
generations need to be accounted for. If the child learning English
uses the form corned, rather than came, or tooths, rather than teeth, it is
presumably because he has acquired an understanding of the prin-
ciples whereby the past tense of most verbs and the plural of most
nouns in English are formed: he has learned some of the regularities,
or some of the rules, of the language. If the child did not succeed in
acquiring an understanding of these principles of regular formation,
he would never be able to do more than repeat, parrot-wise, utterances
of the language which he has previously heard around him. We do not
say that he 'knows* the language, until he is capable of constructing
new utterances which are recognized as normal and can be under-
stood by other speakers of the language.
The important point is that language is patterned, or structured (we
shall come back to this term in the following chapter), on a number of
different levels. In every language there are regular principles
according to which sounds combine with one another to form words,
and regular principles according to which the sounds may be pro-
nounced somewhat differently in different positions of the word or
sentence. At the same time there are regularities in the formation of
words and sentences from the point of view of their grammatical
function. There is nothing anomalous about the form came, con-
sidered solely from the point of view of its pronunciation: cf. lame,
maim, game, same, etc. What is irregular about came is its relationship
to come, by contrast with love: loved, jump: jumped, etc. Because forms
like came stand outside the regular patterns of grammatical formation,
they will tend to be regularized as the language is transmitted from
one generation to the next. So much was clear to the 'Junggram-
matiker'. But they drew a sharp distinction between the 'physical'
medium in which language was manifestits soundsand the
' psychological' aspects of grammatical structure; and they assumed
that only the 'physical' was subject to 'laws' of regular development.
We now know that the way in which sounds combine with one
another to form words and sentences cannot be explained solely, or
even primarily, in terms of the ' physical' nature of these sounds. The
principle of analogy is no less relevant to the formation of regular
groups of sounds in particular languages than it is to the formation of
grammatical 'paradigms' (cf. 1.2.3). The child must learn the one,
just as he must learn the other.
The process of change in language can be looked upon as the re-
placement of one system of'analogies' and 'anomalies' with another.
Changes may be brought about either by such 'external' causes as
borrowing or by the 'internal' factor of 'structural pressure'. By
structural pressure is meant the tendency to regularize the 'ano-
malies' in accordance with the general patterns of the language.
Since changes may be taking place simultaneously at different levels
of the language-system, the result is not necessarily the gradual
elimination of all irregularity. For example, the change of s to r
between vowels in prehistoric Latin was a matter of regular develop-
ment (cf. 1.3.13). It may have started in just one or two words, and
then spread ' by analogy'; whether the change took place in all words
simultaneously or not, the outcome was a somewhat different system
at the level of the permissible combinations of sounds (s no longer
appearing between vowels). But this new regularity at the level of
sounds, as we have seen, created such grammatical 'anomalies' as
honos: honoris, etc., by contrast with cultor: cultoris, etc. And these
'anomalies' were subsequently regularized by structural pressure at
the grammatical level (honor: honoris). It is because the changes at
different levels of the system are not always in phase, because con-
flicting principles of analogy may be operative at the same time, and
because particularly frequent forms (e.g. came, went) are resistant to
regularization, that languages never achieve a state of anomaly-free
38 I. LINGUISTICS
equilibrium. Not all these points have been illustrated here. But
enough may have been said to make it clear that even the irregularities
in language may originate from what were once regularities, however
paradoxical this may appear; and also that analogy, or pattern, or
structure, is the dominating principle, without which languages
could not be learned or used to say things which had not been said
before. Recognition of the different levels of structure, or patterning,
in language carries us into the third of our chronological periods in
the history of linguisticsthe period of twentieth-century ' structural
linguistics*.
1.4 Modern linguistics
1.4.1 Ferdinand de Saussure

If any one person is to be called the founder of modern linguistics it is
the great Swiss scholar, Ferdinand de Saussure, whose lectures (recon-
structed from the notes of his students after his death) were published
in 1915 as Cours de Hnguistique generate. Many different schools of
linguistics can be distinguished at the present time, but they have all
been directly or indirectly influenced (in various degrees) by de
Saussure's Cours. Although we shall have occasion to refer to some of
these different * schools' in later chapters of this work, we shall make
no general attempt to assess their contribution to the development of
the subject over the last fifty years. In this section, we will in fact
abandon the principle of chronological sequence entirely. Instead, we
will list the most important features which distinguish modern
linguistics as a whole from the linguistics of previous periods.
1.4.2 Priority of the spoken language

As we have already seen (cf. 1.2.4 a n d 1.2.8), the traditional gram-
marian tended to assume that the spoken language is inferior to and
in some sense dependent upon the standard written language. In
conscious opposition to this view, the contemporary linguist maintains
(though with certain important qualifications which we shall intro-
duce presently) that the spoken language is primary and that writing
is essentially a means of representing speech in another medium.
The principle of the priority of the spoken language over the
written implies, first of all, that speech is older and more widespread
than writing. It is sometimes asserted that speech cannot be * proved*
to be older than writing. But this is true only if the term * proof* is
1.4- MODERN LINGUISTICS 39
made to bear a far greater load than we generally require it to bear in
questions of historical fact. We know of no system of writing with a
history of more than some six or seven thousand years. On the other
hand, there is no group of people known to exist or to have existed
without the capacity of speech; and many hundreds of languages
have never been associated with a writing-system until they were
committed to writing by missionaries or linguists in our own day.
It seems reasonable to suppose therefore that speech goes back to the
origins of human society.
The relative antiquity of speech and writing is, however, of
secondary importance. Far more relevant to understanding the
relation between speech and writing is the fact that all systems of
writing are demonstrably based upon units of spoken language. (In
some cases it is necessary to go back to an earlier form of a particular
language or to some other language from which the writing-system
was borrowed; but this does not invalidate the principle.) In the
description of spoken language, the linguist generally finds that he
must recognize units of three different kinds (as well as many others,
of course): 'sounds', 'syllables' and 'words'. Now, all commonly-
used systems of writing take one or other of these units as basic;
alphabetic systems being based on 'sounds', syllabic on 'syllables',
and ideographic on 'words'. Granted that all three 'layers' are present
beforehand in spoken language, it is easy enough to explain the
derivation of each of the main systems of writing from a different
' layer' of the spoken language. Although a particular alphabet or a
particular syllabary may be more suitable for certain languages than
for others, there is no correlation between the general structure of
different spoken languages and the type of writing-system used to
represent them. Spoken Turkish did not change as a result of the
replacement of the Arabic script by the Roman in 1926; and no
modification of spoken Chinese is implied by the present government's
proposal to introduce an alphabetic script in place of the traditional
ideographic system.
For historical reasons (usually as a result of sound-change or of
borrowing from other languages with different orthographic con-
ventions) certain words may be distinguished in their written, but not
their spoken, form: instances of such words (traditionally called
homophones) are great and gratey meat and meety seen and scene, in
English. Conversely, words that are written in the same way may be
pronounced differently: English examples of such forms (which we
may refer to as homographs) are the words written lead and read
40 I. LINGUISTICS
(which may rhyme with either bleed or bled). The longer the period
during which a language has been committed to writing (and used for
literature and administration) the greater will be the discrepancy
between spelling and pronunciation, unless, of course, this discrepancy
has been corrected periodically by spelling reforms. Throughout
history, scribal traditions have tended to be conservative: this
conservatism is partly explained by the prevalence of the ' classical'
view of language-change as 'corruption' (cf. 1.2.4); a n d it has been
reinforced, in modern times, by the standardization of spelling for
printing.
There are more important differences between spoken and written
language than those brought about by the development of homo-
phony and homography. No writing-system represents all the
significant variations of pitch and stress which are present in spoken
utterances; and the conventions of punctuation to distinguish
different kinds of sentences (e.g. the use of an exclamation-mark or
question-mark, rather than a full-stop) and the practice of italicizing
words for emphasis constitute, at best, an indirect and imperfect
means of supplying this deficiency. Moreover, in the typical situations
in which the written language is used there is no direct, face-to-face
confrontation of writer and reader; information which might be
carried by the gestures and facial expressions accompanying speech
must therefore be conveyed verbally. The fact that there are invari-
ably such differences as these between speech and writing means that
written language cannot be regarded as merely the transference of
spoken language to another medium.
In the case of particular languages the principle of the priority of
speech over writing must be qualified even further. In a literate
society, especially in one that is educated to a knowledge and an
appreciation of the writings of the past, the written and the spoken
language may develop at different rates and may come to diverge
from one another considerably in vocabulary and grammar. Modern
French affords a particularly striking example of this. Not only is there
a good deal of homophony between grammatically unrelated forms
(cf. cou 'neck', coup 'blow', etc., coUt 'cost', which are all pronounced
alike), but many of the grammatically related forms (e.g. the singular
and plural of the same noun or verb), though spelled differently, are
not distinguished at all in the spoken language (cf. il pense and ils
pensent, 'he thinks' v. 'they think'). As a result there are many
French sentences which are ambiguous when spoken (and taken out
of context), although they are perfectly clear in the written form
(cf. il vient toujours a sept heures: il vient toujours d cette heurey ' he
always comes at seven o'clock*: 'he always comes at this time'). More
important than homophony is the fact that certain verbal forms,
notably the 'simple past* tense forms (e.g. donna 'gave', ripondit
'answered', etc.) and the past subjunctive qu'il donndt 'that he should
give', etc.), have disappeared from spoken French and are learned by
French children, to be used thereafter only in writing, when they go
to school. In addition to these differences of grammatical structure,
there are many differences of vocabulary: as in English, there are
many words and expressions which would be regarded as ' bookish' if
they were used in conversation, and many others which do not appear
in written French (except, of course, in the written representation of
conversation in a novel or a play). In other words, written and spoken
French (to a greater degree than written and spoken English) are
learned and used by educated Frenchmen as partially independent
languages. Written and spoken Chinese are even more independent
of one another, since what are conventionally referred to as different
'dialects' of Chinese (Mandarin, Cantonese, etc.) are written in
essentially the same way. Educated speakers of Mandarin and Can-
tonese are thus able to communicate with one another in writing,
although they may not be able to do so in speech.
Although facts such as these lead us to modify the principle of the
priority of the spoken language, they do not oblige us to abandon it
entirely. For it is only exceptionally that a written language becomes
completely independent of the spoken language from which it
originally derived. This happened notably in the case of Latin, which
was used for centuries in Europe as the language of religion, adminis-
tration and scholarship (and is still used in this way by the Catholic
Church). The Latin of scholars, priests and diplomats in medieval
and Renaissance Europe was a 'dead' language; it was not their
normal medium of everyday discourse, learned 'naturally' in child-
hood, but a language which they learned and used for restricted
purposes. Moreover, it was essentially an unchanging, written
language (based on the 'living', spoken Latin of many centuries
before); and, when spoken, drew constantly and deliberately upon
the written works of the past. As we have seen, the peculiar status of
Latin in medieval and Renaissance Europe tended to confirm the
scholars of the time in their acceptance of the classical principle of the
priority of the written language (cf. 1.2.7). Other well-known examples
of 'dead' languages which continued to be used for religion or
scholarship are Sanskrit, Byzantine Greek and Old Church Slavonic.
42 I. LINGUISTICS
In discussing the relationship between written and spoken language
(which, as we have seen, is by no means a simple relationship and
varies considerably in the case of different languages), we have said
nothing about the different 'styles' that must be distinguished in
dealing both with spoken and with written language. When the
traditional grammarian maintained the principle of the priority of
the written language, he was, of course, thinking primarily of the
language of literature (rather than, for instance, the language of
telegrams, newspaper-headlines or public notices); and he would
tend to say that the literary language was the 'noblest* or most
' correct * form of the language. We may now take up the discussion of
this question. We will return to the principle of the priority of the
spoken language in the following chapter (cf. 2.2.6).
1.4.3 Linguistics is a descriptive, not a prescriptive, science

The traditional grammarian tended to assume, not only that the
written language was more fundamental than the spoken, but also
that a particular form of the written language, namely the literary
language, was inherently 'purer* and more 'correct* than all other
forms of the language, written and spoken; and that it was his task,
as a grammarian, to ' preserve * this form of the language from ' cor-
ruption*. There are several different points here, but they may all be
treated conveniently under the distinction to be drawn between
prescriptive and descriptive linguistics.
The first question to be discussed is that of' purity * or' correctness *.
It should be evident that there are no absolute standards of ' purity *
and ' correctness * in language and that such terms can only be inter-
preted in relation to some standard selected in advance. We can say that
a foreigner has made a mistake because he has said something which
would not be said by a native speaker. We can also say, if we wish,
that some speaker of a regional dialect of English has produced an
'incorrect' or 'ungrammatical* form because this form is not in con-
formity with the patterns of standard English: but we are now
assuming that he ought to have been speaking standard English in
that particular situation. To assert that any linguistic form is ' correct *
or 'incorrect* because it is at variance with some other form taken
(explicitly or implicitly) as the standard is therefore tautological. Each
socially or regionally differentiated form of the language has its own
standard of 'purity* and 'correctness* immanent in it. Once this is
realized and accepted, the way is clear to a more satisfactory descrip-
tion of languages. Whether the speech of one region or of one social
group should be taken as the standard for wider use (e.g. as the basis
for a literary language), is a question of a different order. The linguist's
first task is to describe the way people actually speak (and write) their
language, not to prescribe how they ought to speak and write. In other
words linguistics (in the first instance at least) is descriptive, not
prescriptive (or normative).
The second point to be made has to do with the notion that
linguistic change necessarily involves 'corruption'. All languages are
subject to constant change. This is an empirical fact, explicable in
terms of a number of factors, some of which, as we saw in the previous
section, are now more or less well understood. Nor is it in the least a
matter for regret that languages should change in the several ways in
which they do. All living languages, it may be assumed, are of their
nature efficient and viable systems of communication serving the
different and multifarious social needs of the communities that use
them. As these needs change, languages will tend to change to meet
the new conditions. If new terms are required they will be incor-
porated in the vocabulary, whether by 'borrowing* them from other
languages or by forming them from existing elements in the vocabu-
lary by the productive resources of the language; fresh distinctions
may be drawn and old distinctions lost; the same distinctions may
come to be expressed by different means. In denying that all change
in language is for the worse, we are not of course implying that it
must be for the better. What we are saying is merely that any standard
of evaluation applied to language-change must be based upon a
recognition of the various functions a language 'is called upon' to
fulfil in the society which uses it.
It should be stressed that in distinguishing between description and
prescription, the linguist is not saying that there is no place for pre-
scriptive studies of language. It is not being denied that there might
be valid cultural, social or political reasons for promoting the wider
acceptance of some particular language or dialect at the expense of
others. In particular, there are obvious administrative and educational
advantages in having a relatively unified literary standard. It is
important, however, to realize two things: first, that the literary
standard is itself subject to change; and second, that from the point
of view of its origin the literary standard is based generally upon the
speech of one socially or regionally determined class of people and,
as such, is no more 'correct', no 'purer' (in any sense that the
linguist can attach to these terms) than the speech of any other class
44 I. LINGUISTICS
or region. If the literary standard has a richer vocabulary (that is to
say, if those who do a good deal of reading and writing have a larger
vocabulary) this is because, through literature, we may enter
vicariously into the lives of many societies, including those of the
past, and share in their diverse experiences.
In condemning the literary bias of traditional grammar, the
linguist is merely asserting that language is used for many purposes
and that its use in relation to these functions should not be judged by
criteria which are applicable, only or primarily, to the literary language.
The linguist is not denying that there is a place in our schools and
universities for the study of the literary purposes to which language is
put. Still less is he claiming to enter the field of literary criticism.
This point has often been misunderstood by critics of linguistics.
1.4.4 The linguist is interested in all languages

This principle is merely a generalization of the preceding one. It is
still fairly common to hear laymen talking about * primitive* languages,
and even repeating the myth that there are some peoples whose
language consists of a couple of hundred words supplemented by
gestures. The truth is that every language so far studied, no matter
how * backward' or 'uncivilized' the people speaking it, has proved on
investigation to be a complex and highly developed system of com-
munication. Moreover, there is absolutely no correlation between the
different stages of cultural development through which societies have
' evolved' and the ' type' of language spoken in these stages of cultural
development. If there is any truth at all in the nineteenth-century
speculations about the development of languages from structural
complexity to simplicity, or from simplicity to complexity, this is not
recoverable from the study of any of the thousands of different
languages spoken throughout the world today. Most linguists these
days refrain from speculating about the origin and development of
language in general terms. They have found that the study of all
languages on equal terms is rewarding. The results of such study
have so far thrown no light on the more general question of the origin
and development of language in the remote past of mankind.
It may be necessary to learn another language, or at least a
specialized vocabulary, in order to study a particular subject or talk
satisfactorily about it. If anyone wants to understand Greek philo-
sophy or scholastic philosophy, for example, he must learn Greek or
medieval Latin (or at least, by diligent and intelligent study of com-
mentaries and expositions, come to understand the significance of the
key-termsand this is a way of learning, indirectly, a certain small
part of the languages). He may then find that it is impossible to discuss
the questions that exercised the Greek and scholastic philosophers at
all adequately except by using their terminology. It may thus be said
that his own language, say modern English, is 'poorer* than Greek or
Latin, since it does not provide him with the necessary distinctions in
this particular field. This might seem to run counter to the assump-
tion that all languages are efficient and viable systems of communica-
tion serving the needs of the community in which they operate. But
the contradiction is only apparent. The number of distinctions one
can draw in classifying the features of 'the world* is in principle
infinite. Only those that are of importance in the life of a particular
community will be given recognition in the vocabulary of that com-
munity. The fact that we do not have in normal colloquial English
terms to refer to some of the concepts of ancient philosophy is
merely a reflection of the fact that most of us do not discuss problems
of ancient philosophy: if we did, the terms would have been created
for this. Whether we choose to say that the extended specialized
vocabularies that are used by different sub-groups for different
specialized purposes (e.g. discussing nuclear physics or heraldry) are
part of English depends upon the way in which we define 'English'
(we shall have more to say about this later). The point is that no
language can be said to be intrinsically ' richer' than another: each is
adapted to the characteristic pursuits of its users.
The linguist's concern (in principle) with all languages derives from
the proclaimed aims of his subject: the construction of a scientific
theory of the structure of human language. All recorded and observ-
able instances of language serve as data to be systematized and
' explained' by the general theory.
1.4.5 Priority of synchronic description

One of the most important of the many conceptual and terminological
distinctions introduced into linguistics by de Saussure was his
distinction between the diachronic and the synchronic study of
language. (The distinction is sometimes drawn by opposing 'histori-
cal ' to ' descriptive \ This is a different sense of the term ' descriptive'
from that which is intended when 'descriptive' is opposed to 'pre-
scriptive ': cf. 1.4.3. For that reason, it is preferable to use the technical
terms coined by de Saussure.) By the diachronic study of a particular
46 I. LINGUISTICS
language is meant the description of its historical development

('through time'); for example, a diachronic study of the English
language might treat of its development from the time of our earliest
records to the present day, or might cover some more limited period
of time. By the synchronic study of a language is meant the descrip-
tion of a particular * state' of that language (at some * point' in time).
It is important to realize that synchronic description is not restricted
in principle to the analysis of modern spoken language. One can carry
out a synchronic analysis of ' dead' languages provided that there is
sufficient evidence preserved in the written records that have come
down to us. The description of a ' dead' language will necessarily be
less complete than the description of a modern spoken language,
simply because it is impossible to check the validity of certain state-
ments made about the language by appealing to native speakers as a
source of further evidence. But there are many 'dead' languages for
which there exists enough material for a reasonably comprehensive
synchronic description.
As we have seen, nineteenth-century linguistics ('comparative
philology') was primarily concerned with the diachronic (cf. 1.3.1).
The principle of the priority of synchronic description, which is
characteristic of most twentieth-century linguistic theory, implies
that historical considerations are irrelevant to the investigation of
particular temporal 'states' of a language. The application of this
principle may be illustrated by means of a famous analogy used by
de Saussure. In this analogy (which he drew upon to make a number of
different theoretical points) de Saussure compared languages to
games of chess.
In the course of a game of chess the state of the board is constantly
changing, but at any one time the state of the game can be fully
described in terms of the positions occupied by the several pieces.
It does not matter by what route (the number, nature or order of the
moves) the players have arrived at the particular state of the game:
this state is describable synchronically without reference to the
previous moves. So it is with language, said de Saussure. All languages
are constantly changing; and just as the state of the chess-board at
some particular time can be described without reference to the
particular combination of moves that has brought the game to that
point, so can the successive, or socially and geographically delimited,
states of a language be described independently of one another.
An example will make clear the several points involved. In Latin
of the classical period up to six ' cases' of the noun were distinguished
by a difference of ending (cf. Table 9, in 7.4.2). The different cases
marked the different kinds of relation holding between words in a
sentence. In later Latin, as a result of an increasing use of prepositions
and certain ' sound-changes' which reduced the number of distinct
endings, a system gradually developed in which no more than two
cases of the noun were distinguished: a ' nominativey and an * oblique'
case, the former being used for the subject of the sentence and the
latter for the object of a transitive verb or of a preposition. This is the
situation that is found in Old French. However, the difference between
the two cases was not marked in all types of nouns (and adjectives).
In fact the following three main noun-classes can be set up for the
oldest period in French:
I II in
Sing. Nom. murs porte chantre
Obi. mur porte chanteur
Plur. Nom. mur portes chanteurs
Obi. murs portes chanteurs
(The reason for the development in words of class in of a difference

of stem in the nominative singular is the difference in the historical
development of stressed and unstressed vowels of Latin: chantre is
from Latin cantor, chanteur from cantorem.) If wTe compare the
different classes, we shall see that in words of Class 1 the presence or
absence of the ending s cannot be taken as marking the form for
either case or number independently. It is only in construction with
other words that murs can be recognized as either singular (and
nominative) or plural (and oblique). By contrast in words of class 11
the s can be regarded as marking portes as plural and its absence as
marking porte as singular, there being no indication of case in the
internal structure of words of this class. Between the eleventh and
fourteenth centuries the whole of the French nominal (and adjec-
tival) system was regularized on the pattern of the class 11 words,
with the result that the distinction between the nominative and the
oblique cases disappeared and the final s came to be the mark of the
plural. (The distinction between the two cases in the singular endured
a little longer in words of class illchantre: chanteur, mairei majeur,
pastre (pdtre): pasteur, etc.but finally this disappeared also as a
grammatical feature, although in some instances the two forms have
been preserved in modern French as * doublets', i.e. as different
words each with its own singular and plural.) When nouns such as
4 I. LINGUISTICS
fils ('son'), Georges, Louis, etc., preserved what was, from the
historical point of view, the s of the old nominative singular, they lost
the contrasting form without s (Old French has the contrast fils: fil
in the singular). The result of the change that took place between the
two periods was the development of the system still reflected in
traditional French orthography in which, by and large, number is
marked in the noun (and adjective) by the presence or absence of s,
and case plays no part.
We may now interpret the example in the light of de Saussure's
analogy. There are two states of the language in question. The
historical development of the later system cannot be understood
except by reference to the earlier, but the facts of the historical
development are clearly not relevant to an understanding of how the
later system worked. It would be absurd to suggest, for instance, that
the relation of portes to porte in the later period was different from
that of tnurs to mur. Each state of the language can, and should, be
described on its own terms without reference to what it has developed
from or what it is likely to develop into. Although the words in the
two different states of the language are identical in form (in order to
simplify the exposition we may assume this to be so), the grammatical
relationship between them is quite different. The pieces, as it were,
are the same, but the positions they occupy on the board have
changed. (In modern spoken French the state of the board is different
again: the distinction between singular and plural in the noun is
generally made, if at all, not in the form of the word itself, but in a
number of different ways, including the form of the accompanying
definite article, agreement of the verb, 'liaison* before a word be-
ginning with a vowel, etc. We have already seen that there is consider-
able difference between the structure of modern spoken and written
French: cf. 1.4.2.)
Relatively few speakers of a language know very much about its
historical development; and yet, learning the language 'naturally* as
children, they come to speak it according to certain systematic
principles, or 'rules', 'immanent' in the utterances they hear about
them. It is the task of synchronic linguistic description to formulate
these systematic 'rules' as they operate in the language at a particular
time. (It may be that the way in which the rules are integrated in the
system of description will reflect particular historical processes in the
development of the language. If so this is an important fact about
the structure of language. But it does not affect the general principle of
the priority of the synchronic, since the native speakers of a particular
language are able to learn and apply the * rules' of their language
without drawing upon any historical knowledge.) With regard to those
few members of a speech-community who do have some knowledge of
previous states of the language, the following argument would seem
to apply. Either their specialized knowledge has some effect upon the
way they speak the language, or it does not. If it affects their usage, so
that their speech is in some respect different from (presumably more
' archaic' than) that of other members of the community, it is to that
extent a different language; and will not therefore fall within the
scope of a description of the more typical usage of the speech-
community. And if it has no noticeable effect upon their normal
speech, it is even more clearly irrelevant to the synchronic description.
In either case, therefore, synchronic analysis is independent of such
knowledge of the history of the language as may exist in the speech-
community.
The principle of the priority of synchronic description is generally
taken to carry the further implication that diachronic description
presupposes the previous synchronic analysis of the various 'states'
through which languages have passed in their historical development.
Since we are not primarily concerned with historical and comparative
linguistics, we will not go further into this question. There is, how-
ever, one important point that must be made here.
The convenient terminological distinction between synchronic and
diachronic description must not be understood to imply that time is
itself the determining factor in language-change. Strictly speaking,
change in language is never ' a function of time' in this sense. There
aife many different factors, both within a language and external to it,
which may determine its development from one synchronic 'state* to
another; and the passage of time merely allows for the complex
interaction of these various factors. Furthermore, one must bear in
mind that the notion of diachronic development (language-change) is
most usefully applied microscopicallythat is to say, in the com-
parison of 'states' of a language relatively far removed from one
another in time. One would be quite mistaken if one were to assume
that the language of a particular 'speech-community* at a particular
time is completely uniform, and that language-change is a matter of
the replacement of one homogeneous system of communication by
another equally homogeneous system at some definite 'point' in time.
The 'speech-community' is always made up of many different
groups, and the speech of the members of these groups will re-
flect in various ways (in pronunciation, grammar and vocabulary)
50 I. LINGUISTICS
differences of age, place of origin or prolonged residence, professional
interest, educational background, and so on. Any particular member
of the speech-community belongs of course simultaneously to many
such linguistically-relevant groups. In addition to the differences in
the language which derive from the existence within the community of
various social groups, there are also important differences of * style'
which relate to the various functions of language and the different
social situations in which the language is used: such differences as can
be referred to the distinction of 'formal' and 'colloquial', etc. It is
customary (except in works devoted specifically to this question) to
abstract from synchronic variation in language, either by restricting
the description of a language to the speech of a particular group using
a particular * style', or by describing the language in terms of such
generality that the description is valid (in intention at least) for all
'varieties'. Some degree of' idealization' is involved in either of these
two procedures, and this may be necessary at the present stage of
linguistic theory. It is important to realize, however, that much of the
difference between two diachronically-determined 'states' of a lan-
guage may be present in two ' varieties' of the language existing at the
same time. From the microscopic, as distinct from the macroscopic,
point of view it is impossible to draw a sharp distinction between
diachronic 'change' and synchronic 'variation'.
1.4.6 The structural approach

The most characteristic feature of modern linguisticsone which it
shares with a number of other sciencesis ' structuralism' (to use the
label which is commonly applied, often pejoratively). Briefly, this
means that each language is regarded as a system of relations (more
precisely, a set of interrelated systems), the elements of which
sounds, words, etc.have no validity independently of the relations
of equivalence and contrast which hold between them. (The reader
may well have observed that the key terms 'system' and 'relation'
have already been used in the discussion of de Saussure's distinction
of the synchronic and diachronic. In fact, de Saussure drew this
distinction as a consequence of his conviction that every language, at
a given time, constitutes an integrated system of relationships.)
The more particular implications of structuralism may be left for
the following chapter. Here it is sufficient to remark that there is no
conflict between the peculiarly abstract approach to the study of
language, which is characteristic of modern, ' structural' linguistics,
1.4. MODERN LINGUISTICS 51
and more 'practical' approaches. However abstract, or 'formal',

modern linguistic theory might be, it has been developed to account
for the way people actually use language. It derives from, and it is
validated or refuted by, empirical evidence. In this respect linguistics
is no different from any other science; and the point would not be
worth stressing, if it were not the case that some linguists, out of
sympathy with current developments, have seen a necessary opposi-
tion between what have been called ' formalism' and ' realism' in the
study of language.
1.4.7 ' Langue' and 'parole'

At this point, it will be convenient to introduce, for future reference,
de Saussure's distinction between langue and parole. (English equi-
valents have occasionally been proposed, but most scholars have
continued to use the French terms employed by de Saussure.
Chomsky has recently drawn roughly the same distinction in terms of
linguistic 'competence' and 'performance' with respect to particular
languages.)
The distinction is intended to eliminate an ambiguity in the use of
the word 'language'. Suppose we were to propose as a provisional
definition of 'English' the following: the English language may be
defined as the set of utterances produced by speakers of English when
they are speaking English. We see the ambiguity immediately. When
we say of someone that he 'speaks English* (or is a 'speaker of
English') we do not imply that he is actually' speaking English' on any
one occasion. It would be quite reasonable to say of a parrot, in the
appropriate circumstances, that it 'is speaking English', but not that
it 'speaks English'. Let us follow de Saussure, and say that all those
who 'speak English' (or are 'speakers of English') share a particular
langue and that the set of utterances which they produce when they
are 'speaking English' constitute instances of parole.
A number of questions now arise. What is the relationship between
langue and parole! Which, if either, does the linguist claim to be
describing when he writes the grammar of 'English' (or of any other
' language') ? De Saussure's own answer to both of these questions was
very largely determined by his adherence to the psychological and
sociological theories of Durkheim; and we need not go into these
details. It suffices, for the present, to have posed the problem and to
have introduced de Saussure's terms.
The relationship between langue and parole is very complex, and
52 I. LINGUISTICS
somewhat controversial. For the moment, we may be content with

the statement that all members of a particular language-community
(all those who speak a particular language, e.g. English) produce
utterances, when they are speaking that language, which, despite their
individual variations, are describable in terms of a particular system
of rules and relations: in some sense, they have the same structural
characteristics. The utterances are instances of parole, which the
linguist takes as evidence for the construction of the underlying
common structure: the langue. It is therefore the langue, the language-
system, which the linguist describes. We shall see later that a distinc-
tion must be made between ' utterances' and * sentences'; and that the
description of a 'language' is, in principle, a two-stage operation. The
utterances of a particular 'language* (what speakers actually produce,
when we say that they 'are speaking the language') can be described
only indirectly, and at the present time very inadequately, on the
basis of a prior description of the sentences of the 'language*. This
distinction between utterances and sentences is fundamental in most
modern linguistic theory. But we can develop certain preliminary
notions without invoking it. Although we shall use the terms
'sentence' and 'utterance' more or less synonymously throughout the
next three chapters, the reader should bear in mind that the relation-
ships we are describing hold within what will later be established as
sentences (units of langue), not utterances (instances of parole).
John Lyons
Online ISBN: 9781139165570
Chapter
2 - The Structure of Language pp. 53-98

2
THE STRUCTURE OF LANGUAGE
2.1 Introductory
2. i. i ' Sounds' and 'words'
If we were to ask a non-linguist what are the ultimate units of
language, the building-blocks, so to speak, out of which utterances
are constructed, he might well reply that the ultimate units of
language are * sounds' and * words'. He might add that words are
made up of sequences of sounds, each sound being represented,
ideally, by a particular letter of the alphabet (in the case of languages
customarily represented by a system of alphabetic writing); and that,
whereas the words of a language have a meaning, the sounds do not
(their sole function being to form words). These several propositions
underlie the traditional view of language reflected in most grammars
and dictionaries: the grammar gives rules for the construction of
sentences out of words, and the dictionary tells us what the individual
words mean. In the following chapters, we shall have occasion to
examine critically the terms * sound', 'word', 'meaning' and 'sen-
tence' which figure in these general statements about language.
However, for the purpose of the present preliminary discussion of the
structure of language, we may leave these terms undefined. Certain
distinctions will be indicated in the course of this chapter and will
be made explicit later.
2.1.2 Phonology, grammar and semantics

The traditional view of language incorporates the notions of composi-
tion (a more complex unit is composed of simpler, or smaller, units:
a word is composed of sounds, a phrase of words, a clause of phrases,
a sentence of clauses, and so on) and of correlation (each word is
correlated with one or more meanings). If we use the term 'level' for
the former and the term 'plane' for the latter we may say that,
according to the model of language-structure which we have pro-
visionally adopted, every language can be described in terms of two
planes: 'form' and 'meaning', or preferably (for the traditional terms
3 [53] LIT
54 2. THE STRUCTURE OF LANGUAGE
have various conflicting interpretations in linguistics) expression and
content. And the expression-plane of language can be described in
terms of (at least) two levels: that of sounds and that of words. To
introduce now the terms generally used by linguists: the sounds of a
given language are described by phonology; the form of its words and
the manner of their combination in phrases, clauses and sentences by
grammar \ and the meaning, or content, of the words (and of the
units composed of them) by semantics.
2.1.3 The 'double articulation* of language

Linguists sometimes talk of the 'double articulation' (or 'double
structure') of language; and this phrase is frequently understood,
mistakenly, to refer to the correlation of the two planes of expression
and content. What is meant is that the units on the 'lower' level of
phonology (the sounds of a language) have no function other than
that of combining with one another to form the 'higher' units of
grammar (words). It is by virtue of the double structure of the
expression-plane that languages are able to represent economically
many thousands of different words. For each word may be repre-
sented by a different combination of a relatively small set of sounds,
just as each of the infinitely large set of natural numbers is distin-
guished in the normal decimal notation by a different combination of
the ten basic digits.
2.1.4 'Level' versus 'plane', 'expression' versus 'content'

Having recognized the two planes of expression and content and the
two levels of phonology and grammar we have not, of course,
advanced at all beyond the traditional view of the structure of
language. We must now see what it is that characterizes the more
modern approach to the study of language to which it is customary to
attach the label 'structural linguistics'.
2.2 Substance and form
2.2.1 Structure of the vocabulary

Few people these days would maintain that the correlation of a
particular word and a particular meaning is other than conventional.
The long controversy between the 'naturalists' and the 'conven-
2.2. SUBSTANCE AND FORM 55
tionalists' may be considered closed (cf. 1.2.2). But the very means by
which the conventionality of the relationship between 'form' and
'meaning' (between expression and content) is demonstrated, namely
the citation from different languages of quite different words referring
to the same thing or having the same meaning (e.g. tree in English,
Baum in German, arbre in French), tends to encourage the view that
the vocabulary of any given language is essentially a list of names associ-
ated by convention with independently existing things or meanings.
And yet one soon comes to realize, in learning a foreign language,
that there are distinctions of meaning made in one language that are
not made in another; that learning the vocabulary of another language
is not simply a matter of acquiring a fresh set of labels to attach to
familiar meanings. To take an example: the English word brother-in-
law can be translated into Russian as zjatj, shurin, svojak, or deverj;
and one of these four Russian words, zjatj, must sometimes be trans-
lated as son-in-law. From this it should not be concluded, however,
that the word zjatj has two meanings, and that in one of its meanings
it is equivalent to the other three. All four words in Russian have a
different meaning. It so happens that Russian brings together (under
zjatj) both sister's husband and daughter's husband, but distinguishes
wife's brother (shurin), wife's sister's husband (svojak) and husband's
brother (deverj). So there is really no word which means 'brother-in-
law ' in Russian, just as there is no word which means' zjatj' in English.
Every language has its own semantic structure. To the degree that
the meanings of one language can be brought into one-to-one cor-
respondence with those of another we will say that the two languages
are semantically isomorphic (have the same semantic structure). The
degree of semantic isomorphism between different languages varies
considerably. In general (this point will be discussed and exemplified
more fully in the chapter on semantics: cf. 9.4.6), the structure of the
vocabulary of a particular language will reflect the distinctions and
equivalences which are of importance in the culture of the society in
which the language operates. The degree of semantic isomorphism
between any two languages will therefore depend very largely upon
the amount of overlap there is in the culture of the two societies using
those languages. Whether there are, or could be, two languages whose
vocabularies are to no degree whatsoever isomorphic with one
another is a question with which we need not be concerned. We can
at least allow for the possibility that all the meanings recognized by a
given language are unique to that language and have no validity or
relevance outside it.
3.2
2.2.2 Substance and form

De Saussure and his followers account for the differences in the
semantic structure of different languages in terms of a distinction
between substance and form. By the form of the vocabulary (or the
form of the content-plane: cf. 2.1.4) is meant the abstract structure of
relationships which a particular language imposes, as it were, on the
same underlying substance. Just as the same lump of children's clay
can be fashioned into objects of different shapes and sizes, so the
substance (or medium) within which distinctions and equivalences of
meaning are drawn can be organized into a different form in dif-
ferent languages. De Saussure himself conceived of the substance of
meaning (the substance of the content-plane) as the whole mass of
thoughts and emotions common to mankind independently of the
language they speaka kind of nebulous and undifferentiated
conceptual medium out of which meanings are formed in particular
languages by the conventional association of a certain complex of
sounds with a certain part of the conceptual medium. (The reader
should note that in this section the terms ' substance' and ' form' are
being employed in the sense in which they were introduced into
linguistics by de Saussure: cf. 4.1.5.)
2.2.3 Semantic structure exemplified with colour-terms

There is much in de Saussure's account of semantic structure that
may be attributed to outdated psychological theories and rejected.
The whole notion of a conceptual substance independent of language
and culture is of doubtful validity. Indeed, many philosophers,
linguists and psychologists today would be reluctant to admit that
meanings can be described satisfactorily as ideas or concepts in the
mind. The notion of substance can be illustrated, however, for certain
words at least, without the postulation of an underlying conceptual
medium. It is an established fact that the colour-terms of particular
languages cannot always be brought into one-to-one correspondence
with one another: for example, the English word brown has no
equivalent in French (it would be translated as brun, marron, or even
jaune, according to the particular shade and the kind of noun it
qualifies); the Hindi word pild is translated into English as yellowy
orange or even brown (although there are different words for other
shades of * brown'); there is no equivalent to blue in Russianthe
words goluboj and sinij (usually translated as 'light blue' and 'dark
blue', respectively), refer to what are in Russian distinct colours, not
different shades of the same colour, as their translation into English
might suggest. To make the point as generally as possible, let us
compare part of the English vocabulary with part of the vocabulary of
three hypothetical languages: A, B and C. For simplicity, we will
restrict our attention to the area of the spectrum covered by the five
English red orange yellow green blue
A a b c d e
B f g h i j
C 1> 9 r s
1 2 3 4 5 6 78 9 10
Fig. i.
terms: red, orange, yellow, green, blue. We will assume that the same
area is covered by thefivewords a, b, c, d and e in A, by thefivewords
/, g, h, i and j in B and by the four words p, q, r and s in C (see
Fig. i). From the diagram, it is clear that language A is semantically
isomorphic with English (in this part of its vocabulary): it has the
same number of colour-terms, and the boundaries between the area of
the spectrum covered by each of them coincide with the boundaries
of the English words. But neither B nor C is isomorphic with English.
Although B has the same number of terms as English, the boundaries
come at different places in the spectrum; and C has a different number
of terms (with the boundaries in different places). In order to
appreciate the practical implications of this, let us imagine that we
have ten objects (numbered i to io in Fig. i), each of which reflects
light at a different wavelength, and that we wish to group them
according to their colour. In English, object i would be described as
'red' and object 2 as 'orange'they would differ in colour; in
language A they would also differ in colour, being described as a and
b, respectively. But in B and C they would be described by the same
colour-term, / or p. On the other hand, objects 2 and 3 would be
distinguished by B (as/and g), but brought together by English and
by A and C (as 'orange', b and^>). From the diagram it is clear that
there are many cases of non-equivalence of this kind. It is not being
maintained, of course, that the speakers of B cannot see any difference
in the colour of objects i and 2. They will presumably be able to
distinguish them in much the same way as the speakers of English can
distinguish objects 2 and 3, by referring to them as reddish-orange and
yellow-orange. The point is that the primary classification is different;
and the secondary classification rests upon and presupposes the
primary (in terms of the semantic structure of English, for instance,
crimson and scarlet denote different * shades * of the same colour redy
whereas the Russian words goluboj and sinijy as we saw above, refer to
what are different colours under the primary classification). The
substance of the vocabulary of colour may therefore be thought of as
a physical continuum within which languages may draw either the
same or a different number of boundaries and within which they may
draw the boundaries at the same or different places.
It would be unreasonable to maintain that there are no perceptually
discrete objects and features of the world external to language and
independent of it; that everything is amorphous until it is given form
by language. At the same time it is clear that the manner in which
objects, flora and fauna, etc., are grouped together under particular
words may vary from language to language: the Latin word mus refers
to both mice and rats (as well as to certain other rodents); the French
singe refers to both apes and monkeys; and so on. A rather more
abstract notion of substance is required to bring facts of this kind
within the scope of the Saussurean explanation of semantic structure.
And there is quite clearly no possibility of accounting for the vocabu-
lary of kinship in terms of the imposition of form upon an underlying
physical substance. Only a limited number of words can be described
in terms of their reference to contiguous areas within a physical
continuum. And we shall see later that even the vocabulary of colour
(which is frequently cited as one of the clearest examples of what is
meant by the imposition of form on the substance of the content-
plane) is rather more complex than is commonly supposed (cf. 9.4.5).
The additional complexities do not affect the points that have been
made in this section. It is sufficient that for at least some parts of the
vocabulary one may postulate an underlying substance of content.
But the notion of semantic structure is independent of the assump-
tion of substance. As the most general statement of what is meant by
semantic structurea statement which applies to all words, whether
they refer to objects and features of the physical world or notwe may
adopt the following formulation: the semantic structure of any
system of words in the vocabulary is the network of semantic relations
that held between the words in the system in question. The nature of
these relations we must leave for the chapter on semantics. The
important point to notice in the definition of semantic structure that
has just been given is that it makes use of the key-terms system and
relation. Colour-words (like kinship-words, and many other sets of
words in the vocabularies of languages) constitute an organized system
of words which are related to one another in a certain way. Such
systems are isomorphic if they have the same number of terms in them
and if the terms are related in the same way.
2.2.4 'Language is form, not substance*

Before discussing the distinction of substance and form with respect
to the expression-plane of language (where it is in fact of more general
validity), it might be helpful to return briefly to de Saussure's analogy
of the chess-game. First of all, it may be pointed out that the actual
material out of which the chess-pieces are made is irrelevant to the
operation of the game. They can be made of any material at all (wood,
ivory, plastic, etc.), provided that the physical nature of the material
is such as to maintain the significant differences of shape between the
pieces in the conditions under which the game is normally played.
(This last point, the physical stability of the material, is obviously
important; it was taken for granted, rather than stressed, by de
Saussure. Chess-pieces carved out of ice would not do, if the game
was to be played in a heated room.) Not only is the material out of
which the pieces are made irrelevant; so also are their particular
shapes. All that is necessary is that each piece should be identifiable as
one which by the rules of the game moves in a particular way. If one
of the pieces is lost or broken, we can replace it with some other
object (a coin or a piece of chalk, for example) and establish the con-
vention that the substituted object will be treated for the purposes of
the game as the piece it represents. The relationship between the
shape of a piece and its function in the game is a matter of arbitrary
convention. Provided that the conventions of interpretation are
accepted by the participants, the game can be played equally well with
pieces of any shape. If we draw the implications of the analogy in
respect of the expression-plane of language, we shall come closer to
understanding one of the fundamental principles of modern lin-
guistics: to put it in de Saussure's own terms, language is a form, not
a substance.
6o 2. THE STRUCTURE OF LANGUAGE
2.2.5 ' Realization' in substance

As we saw in the previous chapter, the spoken language is prior to the
written (cf. 1.4.2). In other words, the primary substance of the
expression-plane of language is sound (in particular, that range of
sound which can be produced by the human speech organs); and
writing is essentially a technique for transferring the words and
sentences of a language from the substance in which they are normally
realized to the secondary substance of shape (visible marks on paper
or stone, etc.). Further transference from the secondary to a tertiary
substance is also possible, as, for instance, in the transmission of
messages by teleprinter. The fact that this kind of transference can
be carried out (one is tempted to call it 'transubstantiation') means
that the structure of the expression-plane of language is very largely
independent of the substance in which it is realized.
For simplicity, let us first of all consider the case of languages
which make use of an alphabetic system of writing. We will assume
that the sounds of the language are in one-to-one correspondence
with the letters of the alphabet used to represent them (in other
words, that each sound is represented by a different letter and each
letter represents always the same sound). If this condition is met, there
will be neither homography nor homophonythere will be one-to-
one correspondence between the words of the written language and
the words of the spoken language; and (on the simplifying assumption
that sentences are composed of words, and nothing more) all the
sentences of the written and spoken language will also be in one-to-
one correspondence. The written and spoken language will therefore
be isomorphic. (That the written and spoken language are never
completely isomorphic, as we have seen already, is irrelevant to the
argument at this point. To the degree that they are not isomorphic
they are different languages. This will be one of the implications of
the principle that language is form, not substance.)
To prevent confusion, we will use square brackets to distinguish
sounds from letters (this is a standard notational convention:
cf. 3.1.3). Thus [t], [e], etc., will denote sounds, and t, e> etc., will
denote letters. We may now introduce a distinction between formal
units and their substantial realization as sounds and letters. When we
say that [t] is in correspondence with , [e] with e, and in general that
a particular sound is in correspondence with a particular letter, and
vice versa, we can interpret this to mean that neither the sounds nor
the letters are primary, but that they are both alternative realizations
2.2. SUBSTANCE AND FORM 6l
of the same formal units, which of themselves are quite abstract
elements, independent of the substance in which they are realized.
For the purpose of the present section let us call these formal units
' expression-elements'. Using numbers to refer to them and putting
the numbers between slants, we may say that /i/ denotes a certain
expression-element, which might be realized in phonic substance by
the sound [t] and in graphic substance by the letter t; that jzj denotes
another expression-element, which might be realized by [e] and e\
and so on.
Now it is clear that, just as the chess-pieces can be made out of
various kinds of material, so the same set of expression-elements can
be realized, not only in sound and shape, but in many other kinds of
substance. For instance, each element could be realized by a light of
a different colour, by a particular gesture, by the emission of a
characteristic odour, by exerting pressure of a certain degree of
intensity on the hand, and so on. It might be possible in fact to
construct a system of communication within which each expression-
element is realized by a different kind of substancea system within
which, for example, /i/ might be realized by a sound (of any kind
whatsoever), \z\ by a light (of any colour), fa/ by a gesture of the
hand, etc. However, we may neglect this possibility and concentrate
rather upon the realization of expression-elements by means of
differences in one homogeneous substance. This is far more typical
of human language. Although spoken language may be associated
with various conventional gestures and facial expressions, these
gestures and expressions do not realize formal units of the same level
as those realized by the constituent sounds of the accompanying
words: that is to say, a particular gesture does not combine with
sounds to make a word in the way that two or more sounds combine
to make a word.
In principle, the expression-elements of language may be realized
in any kind of substance, provided that the following conditions are
satisfied, (a) The sender of a 'message' must have available the
necessary apparatus for the production of the significant differences
in the substance (differences of sound, shape, etc.), and the receiver of
the message must have the apparatus necessary for the perception of
these differences: in other words, the sender (speaker, writer, etc.),
must have the requisite ' encoding' apparatus and the receiver (hearer,
reader, etc.) must have the appropriate 'decoding' apparatus, (b) The
substance itself, as the medium within which the differences are
drawn, must be sufficiently stable to maintain the differences between
the realizations of the expression-elements under the normal condi-
tions of communication long enough for the transmission of messages
between the sender and the receiver.
2.2.6 The substance of spoken and written language

Neither of these conditions requires much comment. Nevertheless,
a brief comparison of speech and writing (more precisely, of phonic
and graphic substance) might be helpful from the point of view of
(a) their availability and convenience, and (b) their physical stability
or durability.
In their speculations about the origins of human language many
linguists have suggested that sound was a more convenient medium
for the development of language than any of the available alternatives.
By contrast with gestures or any other substance within which
differences are perceived by the sense of sight (a very highly-developed
sense in human beings), sound does not depend upon the presence of
a source of light and is not so frequently obstructed by objects which
lie in its path: it is therefore equally well suited for communication
by day and by night. By contrast with various kinds of substance
dependent upon the sense of touch for the production and perception
of distinctions within them, sound does not require that the sender
and receiver should be in very close proximity; and it leaves the hands
free for other tasks. Whatever other factors may have influenced the
development of human speech, it is clear that phonic substance (that
range of sound which can be produced by the human speech organs
and falls within the normal range of human hearing) satisfies the
conditions of availability and convenience fairly well. Relatively few
human beings are physically unable either to produce or to perceive
differences of sound. Phonic substance also satisfies the condition of
physical stability well enough for what one must assume to have been
the most normal and most necessary form of communication in
primitive societies.
Graphic substance differs somewhat from phonic substance in
respect of convenience and availability: it requires the use of some
implement or other, and does not leave the hands free for the
performance of various co-operative tasks.
Far more important, however, are the differences between the two
kinds of substance with regard to durability. Until recent times (with
the invention of the telephone and sound-recording equipment), it
was a characteristic of phonic substance that it did not serve as a very
reliable medium for communication unless the sender and the
receiver were present in the same place at the same time. (Oral
tradition and the employment of a third person as a messenger were
dependent upon memory.) Of themselves, as it were, sound-sequences
died away and, if they were not * decoded* on the spot, were lost
for ever. But with the invention of writing an alternative and more
durable medium was made available for * encoding* language.
Although writing was less convenient (and was therefore not used)
for communication of a more ephemeral kind, it made possible the
transmission of messages over considerable distances and also their
preservation for future reference. These differences in the conditions
under which speech and writing were, and still are, most typically
employedthe one in unreflecting, face-to-face communication, the
other in more carefully composed texts intended to be read and
understood without the aid of all the clues in the immediate situa-
tiongo a long way towards explaining both the origin of writing in
the first place and many of the subsequent divergences between
written and spoken languages. As we have already seen, these
differences are such that it would be inaccurate to say of languages
with a long history of literacy behind them that writing is merely the
transference of speech to an alternative substance (cf. 1.4.2). Granted
the differences in the physical stability of phonic and graphic
substance and their importance in the historical development of
written and spoken languages, it remains true that both kinds of
substance are sufficiently stable to maintain the perceptual differences
between the sounds or shapes which realize the expression-elements
in the conditions under which speech and writing are customarily
employed.
2.2.7 Arbitrariness of substantial realization

We may turn now to the second of the points which de Saussure made
about the substance in which language is realized: just as the shape
of the chess-pieces is irrelevant to the operation of the game, so also
are the particular distinctions of shape or sound whereby the
expression-elements of language are identified. In other words, the
association of a particular sound or letter with a particular expression-
element is a matter of arbitrary convention. The point may be
illustrated from English. Table 3 gives, in column (i), six of the
expression-elements of English, numbered arbitrarily from 1 to 6;
column (ii) gives their normal orthographic realizations, and column
Table 3
Expression-elements
(i) (ii) (iii) (iv) (v) (vi)

M t [t] p [p] e
M e [e] i [i] b
b
hi [b] d
td] d
d
hi W b [b] P
ISI i [i] [e] t
M P [Pj t [t]
Words
(vii) (viii) (ix) (x) (xi)

A 'bet' dip [dip] dbe
B 'pet' tip [tip] ibe
C 'bit' dep [dep] dte
D 'pit' tep [tep] ite
E 'bid' deb [deb] dtp
F 'bed' dib [dib] dbp
(iii) their realizations as sounds. (For simplicity, we will assume that

the sounds, [t], [e], etc., are not further analysable and realize the
minimal expression-elements of the language, as they are found, for
example, in the words written bet, pet, bid, etc. Although this assump-
tion will be questioned in the following chapter, the argument is
unaffected by any subsequent modifications we may feel obliged to
make.) Let us now establish a different arbitrary convention according
to which /i/ is realized orthographically by p, jzj by i, and so on;
cf. column (iv). As a result, the word A (which means 'bet', and was
formerly written bet) will now be written dip, the word B will be
written tip, and so on: cf. columns (vii), (viii) and (ix). Quite clearly,
every word and sentence of written English that is distinguished
under the normal conventions of the orthography is distinguished
under the new conventions. The language itself is totally unaffected
by the change in its substantial realization.
The same argument applies with respect to the spoken language (but
with certain limitations which we will introduce presently). Suppose
that the expression-element /i/ were realized in phonic substance as
[p], \z\ as [i], and so on: cf. column (v). Then the word that is now
written bet (and might still be written bet, since there is obviously no
intrinsic connexion between the letters and the sounds) would be
pronounced like the word that is now written dip (although it would
still mean 'bet'); and so on for all other words: cf. column (x).
Once again, the language itself is unchanged by this alteration in its
substantial realization.
2.2.8 Priority of the phonic substance

There is, however, an important difference between the graphic and
the phonic realization of language; and it is this difference which
compels us to modify the strict Saussurian principle that the expression-
elements are completely independent of the substance in which they
are realized. Whereas there is nothing in the shape of the letters d, b,
e> etc., which prohibits their combination with one another in any
way we choose to select, certain combinations of sounds are un-
pronounceable. For example, we might decide to adopt for the written
language the set of realizations listed in column (vi) of the table, so
that word A would be written dbe, word B ibe, and so on: cf. column
(xi). The letter-sequences of column (xi) can be written or printed
no less easily than those of column (ix). By contrast, such sound-
complexes as would result from the substitution of [d] for [b], of [t]
for [i] and of [p] for [d] in the word for 'bid* (the word E) would be
unpronounceable. The fact that there are limits upon the pronounce-
ability (and audibility) of certain groups or complexes of sounds
means that the expression-elements of language, or rather their
combinations, are partly determined by the nature of their primary
substance and the 'mechanisms' of speech and hearing. Within the
range of possibilities left open by the limits imposed by the condition
of pronounceability (and audibility) each language has its own
combinatorial restrictions, which may be referred to the phonological
structure of the language in question.
Since we have not yet drawn a distinction between phonetics and
phonology (cf. chapter 3), we must be content here with a rather
imprecise statement of the point at issue. We will take for granted the
classification of sounds into consonants and vowels; and we will
assume that this classification is valid both in general phonetic theory
and also in the description of the combinatorial possibilities for
particular languages, including English. Now, the substitution of [p]
for [t], of [e] for [i], etc. (cf. column (iv)), would not affect the pro-
nunciation of the language, because (amongst other things) it holds
constant under the substitution the consonantal and vocalic nature of
the sounds. This not only guarantees the pronounceability of the
resultant spoken words, but also respects their phonological structure
(as words of English) in terms of the ratio of consonants to vowels and
the way in which these two classes of sounds combine with one
another. It should be clear, however, that other substitutions might
be made which, though they might maintain pronounceability, would
alter the balance of consonants and vowels and their patterns of
combination in words. Nevertheless, provided that all the words of
spoken English are kept distinct under the new system for realizing
the expression-elements, the grammatical structure of the language
would be unchanged. It is therefore in principle possible for two (or
more) languages to be grammatically, but not phonologically, iso-
morphic. Languages are phonologically isomorphic if, and only if, the
sounds of the one are in one-to-one correspondence with the sounds
of the other and corresponding classes of sounds (e.g. consonants and
vowels) conform to the same principles of combination. One-to-one
correspondence between sounds does not imply identity. On the
other hand, as we have seen, the principles of combination are not
wholly independent of the physical nature of the sounds.
The conclusion to be drawn from the argument of the preceding
two paragraphs supports the principle that the spoken language is to
be given priority over the written language in general linguistic theory
(cf. 1.4.2). The patterns of combination to which the letters con-
form in the written language are totally inexplicable in terms of the
shapes of the letters, whereas they are at least partly explicable in
terms of the physical nature of the sounds in the corresponding spoken
words. For example, u and n are related to one another in shape in
precisely the same way as d and p. But this fact is completely irrele-
vant to the combination of these letters with one another in the
written words of English. Far more relevant is the fact that the letters
in question are in partial correspondence with the sounds of the
spoken language. The study of the substance of sound is therefore of
more central concern to the linguist than is the investigation of
graphic substance and writing-systems.
2.2.9 Combination and contrast

The only properties that expression-elements have, considered in
abstraction from their substantial realization, are (i) their combinatorial
functiontheir capacity to combine with one another in groups or
complexes which serve to identify and distinguish words and sentences
(and we have just seen that the combinatorial properties of the
expression-elements of language are in fact partly determined by the
nature of their primary substance, sound) and (ii) their contrastive
functiontheir difference from one another. It was the second of
these properties that de Saussure had in mind when he said that the
expression-elements (and, more generally, all linguistic units) are
essentially negative in nature: the principle of contrast (or opposition)
is fundamental in modern linguistic theory. The point may be
illustrated with reference to Table 3 on p. 64. Each of the expression-
elements (numbered from 1 to 6 in the table) contrasts with, or is in
opposition with, each other element that could occur in the same
position in English words, in the sense that the substitution of one
element for another (more precisely, the substitution of the sub-
stantial realization of one element for the substantial realization of
another) will have the effect of changing one word into another. For
instance, the word A {bet) is distinguished from the word B {pet) in
that it has fa/ rather than /6/ initially, A is distinguished from C {bit) in
that it has J2J rather than l$j medially, and A is distinguished from F
{bed) in that it has / i / rather than \\\ in final position. On the basis of
these six words, we can say that / i / is in contrast with \\\y \z\ with
/5/, and fa/ with /6/. (By bringing other words forward for comparison
we should of course be able to establish other contrasts and other
expression-elements.) As a formal unit, and within the limits of the
sample of the language we are considering, /i/ can be defined as * that
element which is not /4/, and which combines with either jzj or j$l
and either j$j or /6/'; and all the other elements in the table can be
defined similarly. In general, any formal unit can be defined (i) as
being distinct from all other elements which contrast with it, and
(ii) as having certain combinatorial properties.
2.2.10 Discreteness of expression-elements

Certain important principles may now be introduced on the basis of
the distinction between form and substance. Let us take for the
purpose of illustration the contrast between foj and /6/, which is
maintained in the spoken language by the difference between the
sounds [b] and [p]. As we have seen, the fact that it is this particular
difference in sound, rather than some other, is irrelevant to the
structure of English. It should also be noticed that the difference
between [b] and [p] is not absolute, but relative. That is to say, what
we have been referring to as 'the sound [b]' or 'the sound [p]' are
ranges of sound; and there is in fact no determinate point at which the
*[b]-range* begins and the '[p]-range* ends (or vice versa). From the
phonetic point of view, the difference between [b] and [p] is a matter
of continuous variation. But the difference between the expression-
elements I2I and /6/ is absolutein the following sense. The words
A and B (bet and pet), and all other words of English kept apart by the
occurrence of either /3I or /6/, do not gradually change into one
another in the spoken language as [b] is gradually varied towards [p].
There might be some point at which it is impossible to tell whether
A or B was intended, but there is no word of English which is in some
sense half-way between the two with respect to its grammatical
function or meaning and is identified by a sound midway between
[b] and [p]. From this it follows that the expression-plane of language
is constructed of discrete units. But these discrete units are realized
in physical substance by ranges of sound within which there is the
possibility of considerable variation. Since the expression-units must
not be confused with one another in their substantial realization as
sounds, there must be some 'safety-margin' between the range of
sounds which realizes one and the range of sounds which realizes
another. Certain contrasts may be lost in the course of time or may
not be maintained in all words by all speakers of a language. This fact
may be accounted for by assuming that the contrasts in question fall
below the ' threshold' of importance in the number of utterances of
the language they keep distinct. It does not mean that the difference
between certain pairs of expression-elements is relative, rather than
absolute.
2.2.11 Grammatical and phonological words

We are now in a position to eliminate an ambiguity in the term
'composition', as it was used in the previous section. It was said that
words were composed of sounds (or letters) and that sentences, clauses
and phrases were composed of words (cf. 2.1.1). It will now be clear
that the term ' word' is ambiguous. In fact, it is customarily used in a
number of different senses, but here we need distinguish only two.
As formal, grammatical units, words may be regarded as totally
abstract entities, whose only properties are that they have a particular
contrastive and combinatorial function (what the principles of
contrast and combination are for grammatical units we shall discuss
later), But these grammatical words are realized by groups or com-
plexes of expression-elements, each of which (in the spoken language)
is realized by a particular sound. We may refer to the complexes of
expression-elements as phonological words. That such a distinction is
necessary (and we will return to it later: cf. 54.3) is evident from the
following considerations. First of all, the internal structure of a
phonological word is in general irrelevant to the fact that it realizes
a particular grammatical word. For example, the grammatical word A
(which means 'bet': cf. the table on p. 64) happens to be realized by
the complex of expression-elements /321/; but it might equally well
be realized by a complex of different expression-elements, not neces-
sarily three in number. (Notice that this is not the same point as that
made earlier about the realization of the expression-elements. A
phonological word is not composed of sounds, but of expression-
elements.) Furthermore, the grammatical and phonological words of
a language are not necessarily in one-to-one correspondence. For
instance, the phonological word which we may identify by its normal
orthographic representation, down, realizes at least two grammatical
words (cf. down the hill: the soft down on his cheek): they are different
grammatical words because they have different contrastive and
combinatorial functions in sentences. An instance of the converse of
this phenomenon is afforded by the alternative realizations, which may
be written dreamed and dreamt9 of one and the same grammatical word
(the past tense of a particular verb). It may be pointed out in passing
that the two phenomena that have just been exemplified are commonly
treated as kinds of homonymy and synonymy (cf. 1.2.3). We have
made no appeal to the meaning of the words in the argument pre-
sented above, but only to their grammatical function and to their
phonological realization. To summarize the points that have been
made: grammatical words are realized by phonological words (and
there is no presupposition of one-to-one correspondence); and phono-
logical words are composed of expression-elements. It is clearly
possible to distinguish yet a third sense of the term * word', according
to which we would say that the English word written cap and the
French word cap are identical: they are the same in (graphic)
substance. But we are not generally concerned with the substantial
identity of words in linguistics. The relationship between the
7O 2. THE STRUCTURE OF LANGUAGE
grammatical word and its substantial realization in sound or shape is
indirect, in the sense that it is handled through the intermediate
level of phonology.
2.2.12 * Abstractness * of linguistic theory

The considerations of this section might seem to be remote from all
practical concerns. This is not so. It is the rather abstract approach
to the study of language made possible by the distinction of substance
and form that has given us a greater understanding of the historical
development of languages than was possible in the nineteenth century
and has led, more recently, to the construction of more comprehensive
theories of the structure, the acquisition and the use of human lan-
guage. And such theories have been put to eminently practical
purposes in the development of more efficient means of teaching
languages, in the construction of better systems of telecommuni-
cation, in cryptography and in the design of systems for the analysis
of languages by computer. In linguistics, as elsewhere, abstract theory
and practical application go hand in hand; but the theory is prior to
its application and is justified independently by its contribution to our
greater understanding of the subject-matter it deals with.
2.3 Paradigmatic and syntagmatic relations

2.3.1 The notion of distribution
Every linguistic unit (with the exception of the sentence: cf. 5.2.1) is
to a greater or less degree restricted with respect to the contexts in
which it can occur. This fact is expressed by saying that every
linguistic unit (below the level of the sentence) has a characteristic
distribution. If two (or more) units occur in the same range of contexts
they are said to be distributionally equivalent (or to have the same
distribution); if they have no contexts in common they are in
complementary distribution. Between the two extremes of total
equivalence and complementary distribution there are two kinds of
partial equivalence that we need to recognize: (a) The distribution of
one unit may include (without being totally equivalent to) the distri-
bution of another: if x occurs in all the contexts in which y occurs,
but there are some contexts in which y but not x occurs, then the
distribution oiy includes the distribution of x. (b) The distribution of
two (or more) units may overlap (or intersect): if there are some
2.3. PARADIGMATIC AND SYNTAGMATIC 71
contexts in which both x and y occur, but neither x nor y occurs in all
the contexts in which the other occurs, then x and y are said to have
an overlapping distribution. (To those readers who are familiar with
certain elementary notions of formal logic and mathematics, it will be
clear that the various kinds of distributional relationship that hold
between linguistic units can be brought within the scope of class-
(i)
Distributional Complementary distribution
equivalence
( \
* A - A 1
J
\ V '
J \ \
^ ^x
(iii) (iv)
Distributional inclusion Overlapping distribution
Fig. 2. Distributional relations (x occurs in the set of contexts A,
and B is the set of contexts in which y occurs).
logic and set-theory. The fact that this is so is highly relevant to the
study of the logical foundations of linguistic theory. What one might
refer to loosely as * mathematical' linguistics is now a very important
part of the subject. Although we cannot go into the details of the
various branches of 'mathematical linguistics' in the present ele-
mentary treatment of linguistic theory, we shall make reference to
some of the more important points of contact as the occasion arises.)
It should be emphasized that the term 'distribution' is applied to
the range of contexts in which a linguistic unit occurs only in so far as
that range of contexts can be brought within the scope of a systematic
statement of the restrictions determining the occurrence of the unit
in question. What is here meant by 'systematic* may be explained by
way of an example. The elements /I/ and /r/ of English are at least
partially equivalent in distribution (for the convention of oblique
strokes, cf. 2.2.5): both may occur in a number of otherwise phono-
logically-identical words (cf. light: right, lamb: ram, blaze: braise,
climb: crime, etc.). But many of the words in which one of the
elements occurs cannot be matched with otherwise phonologically-
identical words in which the other occurs: there is no word srip to
match slip, no tlip to match trip, no brend to match blend, no blick to
match brick, and so on. However, there is an important difference
between the non-occurrence of words like srip and tlip, on the one
hand, and of words like brend and blick, on the other. The first pair
(and others like them) are excluded by certain general principles which
govern the phonological structure of English words: there are no
words which begin with /tl/ or /sr/ (the statement could be made in
more general terms, but this formulation of the principle will suffice
for our present purpose). By contrast, no such systematic statement
can be made about the distribution of /I/ and /r/ which would account
for the non-occurrence of blick and brend. Both elements are found
elsewhere in the contexts /b-i... / and /b-e... /: cf. blink: brink, blessed:
breast, etc. From the point of view of their phonological structure
brend and blick (unlike tlip and srip) are acceptable words of English.
It is a matter of * chance', as it were, that they have not been given a
grammatical function and a meaning, and put to use in the language.
The point that has just been illustrated by means of a phonological
example applies also at the grammatical level. Not all combinations of
words are acceptable. Of the unacceptable combinations, some can be
accounted for in terms of a general distributional classification of the
words of the language, whereas others must be explained by reference
to the meaning of the particular words or to some other fact specific
to them as individual words. We shall return to this question in a later
chapter (cf. 4.2.9). For the purpose of the present argument, it is
sufficient to observe that distributional equivalence, total or partial,
does not imply absolute identity in the range of contexts in which the
units in question occur: what it implies is identity in so far as the
contexts are specified by the phonological and grammatical regu-
larities of the language.
2.3.2 Free variation

As we saw in the previous section, every linguistic unit has both a
contrastive and a combinatorial function. Now it is clear that two
units cannot be in contrast unless they are at least partially equivalent
in distribution (for units that are in complementary distribution the
question of contrast does not arise). Units which occur, but are not in
contrast with one another, in a given context are in free variation. For
example, the vowels of the two words leap and get contrast in most
contexts in which they both occur (cf. bet: beat, etc.), but are in free
variation in the alternative pronunciations of the word economics.
Both in phonology and in semantics, one must be careful not to
confuse free variation (equivalence of function in context) with
distributional equivalence (occurrence in the same range of contexts).
What is meant by free variation and contrast will depend upon the
nature of the units to which the terms are applied and the point of
view from which they are regarded. As we have seen, two expression-
elements are in contrast if the substitution of one for the other in the
same context produces a different word or sentence; otherwise they
are in free variation. But words (and other grammatical units) may
be looked at from two different viewpoints. It is only when one is
concerned with their grammatical function (roughly speaking, whether
they are nouns, verbs or adjectives, etc.) that the notions of contrast
and free variation are interpreted in terms of distributional equi-
valence; and this is because there is a direct connexion between
grammatical function and distribution (cf. 4.2.6). Although there is
also some connexion between the meaning of a word and its distri-
bution, neither is completely determined by the other; and for that
reason the two notions are theoretically distinct. In semantics free
variation and contrast are to be interpreted as * sameness and difference
of meaning'. (It is more usual, however, to employ the traditional
term 'synonymy', rather than 'free variation', in semantics.)
2.3.3 * ParadigmaticJ and ' syntagmatic'

By virtue of its potentiality of occurrence in a certain context a
linguistic unit enters into relations of two different kinds. It enters
into paradigmatic relations with all the units which can also occur in
the same context (whether they contrast or are in free variation with
the unit in question); and it enters into syntagmatic relations with the
other units of the same level with which it occurs and which constitute
its context. To return to the example used in the previous section:
by virtue of its potentiality of occurrence in the context /-et/ the
expression-element /b/ stands in paradigmatic relationship with /p/,
/s/, etc.; and in syntagmatic relationship with /e/ and /t/. Likewise,
/e/ is in paradigmatic relationship with /!/, /a/, etc., and in syntagmatic
relationship with /b/ and /t/. And /t/ is related paradigmatically with
/d/, /n/, etc., and syntagmatically with /b/ and /e/.
Paradigmatic and syntagmatic relationships are also relevant at the
word-level, and indeed at every level of linguistic description. For
example, by virtue of its potentiality of occurrence in such contexts as
a.. .of milky the word pint contracts paradigmatic relations with such
other words as bottle, cup, gallon, etc., and syntagmatic relations with
0, of and milk. In fact, words (and other grammatical units) enter into
paradigmatic and syntagmatic relations of various kinds. ' Potentiality
of occurrence* can be interpreted with or without regard to the
question whether the resultant phrase or sentence is meaningful; with
or without regard to the situations in which actual utterances are
produced; with or without regard to the dependencies that hold
between different sentences in connected discourse; and so on. We
shall have more to say later about the different conditions that can be
imposed upon the interpretation of the term * potentiality of occur-
rence* (cf. 4.2.1, on the notion of 'acceptability'). Here it must be
emphasized that all linguistic units contract syntagmatic and para-
digmatic relations with other units of the same level (expression-
elements with expression-elements, words with words, etc.); that the
context of a linguistic unit is specifiable in terms of its syntagmatic
relations; and that the range of contexts in which it is said to occur,
as well as the extent of the class of units with which it is said to be
paradigmatically related, will depend upon the interpretation
explicitly or implicitly attached to 'potentiality of occurrence' (or
'acceptability').
This last point has been formulated in what might appear to be
an unnecessarily devious manner. It will become clear later that one
of the advantages of this formulation is that it enables us to draw the
distinction between grammatical sentences and meaningful sentences,
not in terms of the combination of grammatical units in the one case
and of semantic units ('meanings') in the other, but in terms of the
degree or kind of ' acceptability' that is preserved by various combi-
nations of the same units.
2.3.4 The interdependence of paradigmatic and syntagmatic relations

Two important points must now be made about paradigmatic and
syntagmatic relations. The first of these, which (together with
distinction of substance and form) might be taken as the defining
characteristic of modern, 'structural' linguistics, is as follows:
linguistic units have no validity independently of their paradigmatic
and syntagmatic relations with other units. (This is a more specific
formulation of the general 'structural* principle that every linguistic
unit has a certain place in a system of relationships: cf. 1.4.6.) To
illustrate from the level of expression-elements: In our earlier
discussion of such words as bet, pet, etc., in English, it was assumed
that all these words were realized as sequences of three expression-
elements (just as they are all written as sequences of three letters by
the orthographic conventions of the language). We may now verify
this assumption. Let us suppose, contrary to the facts, that there were
words whose realization in sound was identical with that of put, tit,
cat, pup, tip, cap, puck, and tick, but no words realized (' pronounced')
like but, pet, pit, bit, cut, gut, kit, duck, cab, cad, kid, cud, etc. What
is being assumed (to put it in rather imprecise phonetic terms) is that
all the phonological words realized as complexes of three sounds can
be described, from the point of view of their substantial realization
(i.e. as phonetic words), as sequences of consonant + vowel + con-
sonant (where the consonants are [p], [t] or [k] and the vowels [u], [i]
and [a]for simplicity, we have assumed no other consonants or
vowels), but that the only combinations of consonant and vowel
possible in first and second position of the sequence are [pu], [ti] and
[ka]. Given this situation, it is clear that [u], [i] and [a] do not realize
three distinct expression-elements of the language since they are not
in paradigmatic relationship (and, a fortiori, not in contrast). Just how
many expression-elements would be recognized in a situation of this
kind (and it is not at all untypical of what is found in language)
depends upon certain more particular phonological principles which
we will discuss later. We might say that in each word there are only
two positions of contrast, of which the first is 'filled* by one of three
consonant-vowel complexes and the second by one of three con-
sonants: we might therefore recognize six expression-elements
(realized as / i / : [ka], \z\: [pu], / 3 /: [ti], / 4 /: [p], / 5 /: [t] and /6/: [k]).
Alternatively, we might recognize four expression-elements, of which
three are realized by the consonants [p], [t] and [k], occurring in
initial and final position, and the fourth, which occurs in medial
position, is realized by a vowel, the phonetic quality of which is
determined by the preceding consonant. The point is therefore that
one cannot first establish the elements and then state their permissible
combinations. The elements are determined by taking account simul-
taneously of their paradigmatic and syntagmatic relations. The reason
why we recognize three positions of contrast in the English words
bet, pet, bit, pit, bid, tip, tap, etc., is that paradigmatic and syntag-
matic relations can be established at three points. We shall see that
the interdependence of the paradigmatic and syntagmatic dimensions
is a principle that holds at all levels of language-structure.
2.3.5 'Syntagmatic9 does not imply 'sequential*

The second important point is the following: syntagmatic relations
do not necessarily presuppose an ordering of units in linear sequence,
such that the substantial realization of one element must precede the
substantial realization of another in time. Compare, for instance, the
two Chinese words hao ('day') and hdo ('good'), which differ from
one another phonologically in that the first has what is conventionally
referred to as 'the fourth tone' (/x/, realized as a fall in the pitch of
the syllable from high to low) and the second has 'the third tone' (/v/,
realized as a variation in the pitch of the syllable from medium to low
and back to medium). The two elements, /x/ and /v/> are in paradig-
matic contrast in the context /hao/; that is to say, in this context (and
many others) they enter into the same syntagmatic relations. If we
say that one word is to be analysed phonologically as /hao/ -f /x/ and
the other as /hao/ + /v/, it is obviously not being implied that the sub-
stantial realization of the tone follows the substantial realization of
the rest of the word. Language-utterances are spoken in time and can
therefore be segmented into a sequence of successive sounds or
complexes of sounds. But whether or not this time-sequence is
relevant to the structure of the language depends once again upon the
paradigmatic and syntagmatic relations of the linguistic units and not,
in principle, on the successivity of their substantial realizations.
Relative sequence is one of the potentialities of the phonic sub-
stance (as it is also one of the derivative potentialities of graphic
substance under the conventions of left-to-right, right-to-left or up-to
down writing) which may or may not be actualized. The application
of the principle may be illustrated conveniently from the grammatical
level. English is a language with what is commonly called 'fixed word
order', and Latin a language with 'free word order'. (In fact, the
word-order of English is not completely 'fixed' and the word-order
of Latin is not completely 'free', but the difference between the two
languages is sufficiently clear for the purpose of the present illustra-
tion.) In particular, an English sentence consisting of a subject, a
verb and an object (e.g. Brutus killed Caesar) will normally be spoken
(and written) with the substantial realizations of the three units in
question sequentially-ordered as subject + verb + object, and the
permutation of the two nouns or noun-phrases will have the effect of
rendering the sentence ungrammatical or converting it into a different
sentence: Brutus killed Caesar and Caesar killed Brutus are different
sentences; although The chimpanzee ate some bananas is a sentence,
Some bananas ate the chimpanzee (we will assume) is not. By contrast,
Brutus necavit Caesarem and Caesarem necavit Brutus are alternative
substantial realizations of the same sentence (' Brutus killed Caesar'),
as are Caesar necavit Brutum and Brutum necavit Caesar ('Caesar
killed Brutus'). The relative order in which words occur in the Latin
sentence is therefore grammatically irrelevant, although of course the
words cannot be spoken other than in one order or another.
2.3.6 Sequential and non-sequential syntagmatic relations

Let us now make the point in a more general way. For simplicity, we
will assume that we are dealing with two classes of units (or tenta-
tively-established units), the members of each class being in para-
digmatic relationship with one another. The classes are X, with a and
b as members, and Y, with p and q as members: using a standard
notation for expressing class-membership,
(These equations may be read as follows:' X is a class whose members

are a and b', 'Y is a class whose members are p and q\) The sub-
stantial realization of each unit is represented by the corresponding
italic letter (a realizes a, etc., and X and Y are variables standing for
the realizations of the units), and we will assume that these substantial
realizations cannot occur simultaneously (they might be consonants
and vowels, or words), but are sequentially-ordered relative to one
another. There are three relevant possibilities: (i) the sequence may
be 'fixed' in the sense that, say, X necessarily precedes Y (i.e. ap, aq,
bp, bq occur, but not pa, qa, pb, qb); (ii) the sequence may be 'free'
in the sense that both XY and YX occur, but XY = YX (where ' = '
means 'is equivalent to'equivalence being defined for the parti-
cular level of description); (iii) the sequence may be 'fixed' (or 'free')
in the different sense that both XY and YX occur, but XY + YX
(' =(=' means ' is not equivalent to'). It may be observed in passing that
the three possibilities are not always distinguished in discussing such
questions as word-order. Of the three possibilities, the last two are the
simplest to handle from a theoretical point of view. In the case of (ii),
since XY and YX are not in contrast, the units a, b, p and q, as they
are realized in such sequences as op or pa, are in a non-sequential
syntagmatic relationship (this is the situation with respect to the words
in a language with free word order). In the case of (iii), since XY
contrasts with YX, the units are in a sequential syntagmatic relation-
ship (this is the situation with respect to the adjective and noun, for
certain adjectives, in French). It is in the case of (i), which is extremely
common, that confusion is likely to arise. Since YX does not occur,
the members of the classes X and Y cannot be sequentially-related at
that level. On the other hand, at some point in the description of the
language, we must specify the obligatory sequence of their realiza-
tions in substance; it may therefore serve the purpose of economy in
the integration of one level with another if instances of (iii) are
assimilated with instances of (ii). It was by tacit appeal to this
principle that we said earlier that English words like bet, pet, etc., had
the phonological structure of consonant + vowel + consonant (using
the terms 'consonant' and * vowel' for classes of expression-elements).
That some of the syntagmatic relationships between the expression-
elements in English are sequential is clear from the comparison of
such words as pat, apt, cat, act, etc. Although the sequence CCV
(consonant + consonant + vowel: the consonants in question being
realized as [p], [t], [k], [b], [d] and [g]) does not occur, both CVC and
at least some instances of VCC do, as we have just seen. At the same
time there are systematic restrictions upon the co-occurrence of
consonants in VCC: for example, a word that would be realized in
substance as [atp] or [atk] is systematically excluded; so, too, are
[akk], [app], [att], English therefore exemplifies both (i) and (iii) in
the phonological structure of the words we have been considering.
By assimilating them to the same sequentially-ordered formulae we
simplify the statement of their substantial realization. It should be
stressed, however, that this does not mean we must not also bring out
the difference between suchc accidental' gaps in the English vocabulary
as [git] or [ped] and such systematically-excluded * words' as [pti]
or [atp] (cf. 2.3.1).
Further discussion of the question of sequence would be out of
place at this point. We shall return to it in later chapters. But before
we continue with the exposition, it should be emphasized that the
whole of the present discussion has been deliberately restricted by the
assumption that all units in syntagmatic relationship co-occur on
equal terms and that there are no groupings within complexes of such
units. The discussion may also appear to have been based on the
additional assumption that units are necessarily realized each by one
and only one isolable segment or feature of the phonic substance.
This is not the case, as we shall see later. The two general points that
have been made are these: (i) the paradigmatic and the syntagmatic
dimensions are interdependent, and (2) the syntagmatic dimension is
not necessarily time-ordered.
2.3.7 * Marked 'and' unmarkedy

So far we have recognized only two possibilities for units in para-
digmatic relationship: that they should be in contrast or in free
variation. It is frequently the case that of two units in contrast (and
for simplicity we may restrict ourselves to two-term contrasts) one
will be positive, or marked, the other being neutral, or unmarked.
What is meant by these terms may be made clear by way of an
example. Most English nouns have a plural and a singular form which
are related, as are boys: boy, days: day, birds: bird, etc. The plural is
positively marked by the final s, whereas the singular is unmarked.
Another way of expressing what has just been said would be that in
a given context the presence of a particular unit is in contrast with its
absence. When this situation holds it is usually the case that the un-
marked form is more general in sense or has a wider distribution than
the marked form. And it has become customary to employ the terms
'marked* and 'unmarked* in a rather more abstract sense, such that
the marked and unmarked members of a contrasting pair are not
necessarily distinguished by the presence and absence of a particular
overt unit. For instance, from the semantic point of view the words
dog and bitch are unmarked and marked for the contrast of sex. The
word dog is semantically unmarked (or neutral), since it can be applied
to either males or females {That's a lovely dog you ve got there: is it a
he or a she?). But bitch is marked (or positive), since it is restricted to
females and may be used in contrast with the unmarked term to
determine the sense of the latter as negative, rather than neutral (Is it
a dog or a bitch?). That is to say, the unmarked term has a more
general sense, neutral with respect to a certain contrast; its more
specific negative sense is derivative and secondary, being a consequence
of its contextual opposition with the positive (non-neutral) term. It
follows from the particular relationship that holds between the words
dog and bitch that, whereas both female dog and male dog are per-
fectly acceptable, both female bitch and male bitch are semantically
anomalous, the one being tautological and the other contradictory. This
notion of 'marking' within paradigmatic oppositions is extremely
important at all levels of language-structure.
2.3.8 Syntagmatic length

One final general point may be made here with respect to the relation-
ship between the paradigmatic and the syntagmatic dimensions. If
there is a given set of units to be distinguished in terms of their
composition out of 'lower-level' elements, then (independently of
certain statistical considerations, which we shall discuss in their
following section) the ' length' of each ' higher-level' unit, measured
in terms of the number of syntagmatically-related elements in the
complex which identifies it, will be inversely proportionate to the
number of elements in paradigmatic contrast within the complex. Let
us suppose, for example, that in one system there are just two
expression-elements (which we will refer to as o and 1) and that in
another system there are eight expression-elements (which we will
number from o to 7); and for simplicity, since this assumption does
not affect the general principle, let us assume that all combinations
of the expression-elements are permissible by the 'phonological'
rules governing the two systems. If there are eight 'phonological'
words to be distinguished within the first (binary) system, each of the
words will be at least three elements long (000, 001, 010, o n , 100,
101, n o , i n ) , whereas each of the eight words can be distinguished
by a single occurrence of a different element of the second (octal)
system (o, 1, 2, 3, 4, 5, 6, 7). If there are sixty-four words to be
distinguished, then complexes at least six elements long will be
required by the binary system, and complexes at least two elements
long by the octal system. In general, the minimal number of ' higher-
level ' units that can be distinguished by a set of' lower-level' elements
syntagmatically related in complexes is determined by the formula:
N = px xp2 xp3".pm (where N is the number of * higher-level' units;
where m is the number of positions of paradigmatic contrast for the
' lower-level' elements; and where p1 denotes the number of elements
in paradigmatic contrast in the first position, p2 denotes the number
of elements in paradigmatic contrast in the second position, and so
on to the mth position). It will be observed that this formula does not
presuppose either that the same elements can occur in all positions
or that the number of elements in paradigmatic contrast is the same
in all positions. The simple illustration given above for a binary and an
2.3. PARADIGMATIC AND SYNTAGMATIC 8l
octal system, within which all elements occur in all positions and all
syntagmatic combinations are possible, is therefore merely a special
case which falls within the scope of the more general formula:
2 x 2 x 2 = 8, 2 x 2 x 2 x 2 = 1 6 , etc.,
and
8 = 8, 8 x 8 = 64, 8 x 8 x 8 = 512, etc.
The reason why we chose to compare a binary system (with two
elements) and an octal system (with eight elements) is that 8 is an
integral power of 2: it is 2 to the power 3, rather than 2 to the power
3-5 or 4*27, etc. And this brings out clearly the relationship between
paradigmatic contrast and syntagmatic 'length'. Other things being
equal, the minimal length of our binary words will be three times
that of our octal words. We shall make use of this particular numerical
relationship in the following section. And we shall make appeal to the
more general principle that distinctions can be made either syntag-
matically or paradigmatically in later chapters, especially in the
chapter dealing with semantics.
It should be observed that the notion of 'length' that has just been
discussed is defined in terms of the number of positions of paradig-
matic contrast within a syntagmatic complex. It is not necessarily
bound to temporal sequence. This point (which follows from what
was said earlier in this section: cf. 2.3.6) will be of considerable
importance in the subsequent discussion of phonological, grammatical
and semantic structure.
2.4 Statistical structure

2.4.1 Functional load
Not all paradigmatic contrasts are of equal importance in the
functioning of language. They may vary considerably in respect of
their functional load. To illustrate what is meant by this term we may
consider some contrasts within the phonological system of English.
There are very many words of spoken English that are distinguished
in their substantial realization by the occurrence of [p] rather than [b]
in the same context (cf. pet: bet, pin: bin, pack: back, cap: cab, etc.);
and, on the basis of this contrast, we may establish an opposition
between /p/ and /b/, which, for the present at least, we may regard as
two minimal expression-elements of the language (by a iminimal'
unit is meant one that is not further analysable). Since there are many
words that are distinguished by the opposition of /p/ and /b/, the
contrast between these two elements has a high functional load. Other
contrasts have a lower functional load. For example, there are
relatively few words that are kept apart from one another in their
substantial realization by the occurrence of one rather than the other
of the two consonants that occur in the final position of wreath and
wreathe (the symbols for these two sounds in the International
Phonetic Alphabet are [9] and [5] respectively: cf. 3.2.8); and there are
very few, if any, that are kept apart by the occurrence of the initial
sound of ship rather than the second consonantal sound of measure or
leisure (these two sounds are symbolized as [f] and [3] respectively in
the International Phonetic Alphabet). The functional load of the
contrasts between /8/ and /fl/ and between /J/ and j^j is therefore
much lower than that of the contrast /p/: /b/.
The importance of functional load is obvious. Misunderstanding
will tend to occur if the speakers of a language do not consistently
maintain those contrasts which serve to distinguish utterances which
differ in meaning. Other things being equal (and we will return to this
condition in a moment), the higher the functional load the more
important it is that the speakers should learn the particular contrast
as part of their ' speech-habits * and should subsequently maintain it
in their use of the language. It is to be expected therefore that
children will tend to learn first those contrasts which have the highest
functional load in the language which they hear about them; and, as
a consequence of this fact, that contrasts with a high functional load
will be correspondingly resistant to disappearance in the transmission
of the language from one generation to the next. Investigation of the
ease with which children master the contrasts in their native language
and the study of the historical development of particular languages
lend some empirical support to these expectations. In each case,
however, there are additional factors which interact with, and are
difficult to isolate from, the operation of the principle of functional
load. We shall not go into these other factors here.
The precise quantification of functional load is complicated, if not
made absolutely impossible, by considerations excluded above under
the proviso 'other things being equal'. First of all, the functional load
of a particular contrast between expression-elements will vary
according to the structural position they occupy in the word. For
example, two elements might contrast frequently at the beginning of
a word, but only very rarely at the end of a word. Do we simply take
an average over all the positions of contrast? The answer to this
question is not clear.
2.4. STATISTICAL STRUCTURE 83
Secondly, the importance of a particular contrast between expres-
sion-elements is not just a function of the number of words they
distinguish: it also depends on whether the words themselves can
occur and contrast in the same context. To take the limiting case: if
A and B are two classes of words in complementary distribution, and
if each member of A differs in its substantial realization from a
member of B solely in that it has the element /a/ where the corre-
sponding word of B has /b/, it is clear that the functional load of the
contrast between /a/ and /b/ is nil. So the functional load of a parti-
cular contrast must be calculated for words which have the same or
an overlapping distribution. It is also clear that any 'realistic*
measure of the importance of a particular contrast should take
account, not simply of the distribution of the words as specified by
the rules of the grammar, but of the actual utterances in which the
words might be confused if the contrast were not maintained. How
often, for instance, or in what circumstances, would an utterance
such as You'd better get a cab be confused with You'd better get a cap,
if the speaker failed to make the distinction in the final consonants
of cab and cap} The answer to this question is evidently relevant to
any precise quantification of the functional load of the contrast in
question.
Finally, the importance of a particular contrast would seem to be
related to its frequency of occurrence (which is not necessarily deter-
mined by the number of words it distinguishes). Let us assume that
the three expression-elements, /x/, /y/ and /z/, occur in the same
structural position of words of the same distributional class. But let
us further suppose that, whereas the words in which /x/ and /y/ occur
are in frequent contrast in the language (they are high-frequency
words), the words in which /z/ occurs have a low frequency of
occurrence (although they might be equally numerous in the vocabu-
lary). If a speaker of the language failed to acquire the contrast
between /x/ and /z/, he would be less severely handicapped in
communication than he would be if he failed to acquire the contrast
between /x/ and /y/. The functional load of the latter contrast is,
ex hypothesis higher than that of the former.
The considerations of the previous paragraphs demonstrate the
difficulty of arriving at any precise measure of functional load. So far,
the various measures that have been proposed by linguists cannot lay
claim to the precision that their mathematical sophistication might
appear to invest them with. Nevertheless, we must allow a place in
oiir theory of language-structure for the undoubted importance, both
synchronic and diachronic, of the concept of functional load. For it
is still possible to say of certain contrasts that they have a higher load
than others, even if we cannot say by how much.
2.4.2 Information-content and probability of occurrence

Another important statistical notion has to do with the amount of
information carried by a linguistic unit in a given context; and this
also is determined by (or is generally held to be determined by) its
frequency of occurrence in that context. The term * information* is
here being employed in the specialized sense it has acquired in
communication-theory, which we will now explain. The information-
content of a particular unit is defined as a function of its probability.
To take the simplest case first: if two or more units are equiprobable
in a given context, they each have the same information-content in
that context. Probability is related to frequency in the following way.
If two, and only two, equiprobable units may occur, x and y, each of
them will occur (on average) in just half the instances of the context
in question: each has a probability, apriorif of J. The probability of a
particular unit x is denoted by px. Thus, in the present instance,
px = \ and py = J. More generally, each of n equiprobable units
(xly x2, #3, ..., xn) has a probability of-. (It will be observed that the
n
sum of all the probabilities in the set of units is 1. This holds inde-
pendently of the more particular assumption of equiprobability.
A special case of probability is 'certainty*. Units which cannot but
occur in a given context have a probability of 1.) If the units are equi-
probable, each of them carries the same amount of information.
More interesting, since they are more typical of language, are
unequal probabilities. Suppose, for example, that two, and only two,
units occur, x and y; and that x occurs twice as often, on average, as
y: then, px = f and py = . The information-content of x is half that
of y. That is to say, information-content is inversely proportionate (and,
as we shall see, logarithmically related) to probability: this is the
fundamental principle of information-theory.
At first sight, this principle may appear somewhat surprising. But
consider first the limiting case of complete predictability. In written
English, the letter u approximates very closely to complete predict-
ability when it follows q\ if we disregard certain borrowed words and
proper names, we may say that it is completely predictable (it has a
probability of 1). Similarly, the word to has a probability of 1 in such
sentences as / want.. .go home, I asked him.. .help me (assuming that
only one word is permitted in the contexts in question). If it were
decided to omit the u (in queeny queer, inquest, etc.) or the word to,
in the contexts referred to, no information would be lost (here we see
the connexion between the general and the more specialized sense of
the word 'information'). Since the letter u and the word to are not in
paradigmatic contrast with any other units of the same level which
might have occurred in the same context, they have a probability of
1 and their information-content is o: they are totally redundant.
Consider now the case of the two-term contrast where px = and
py = \. Neither is totally redundant. But it is clear that the omission
of x is of less consequence than the omission of y. Since x is twice as
probable as yy the receiver of a message (knowing the a priori prob-
abilities) would stand twice as good a chance on average of ' pre-
dicting ' where x had been omitted as he would of ' predicting' where
y had been omitted. Redundancy is therefore a matter of degree. The
redundancy of x is twice that of y. In general, the more probable a
unit is, the greater its degree of redundancy (and the lower its informa-
tion-content).
2.4.3 Binary systems

Information-content is usually measured in bits (the term is derived
from * binary digits'). Any unit with a probability of \ conveys one
bit of information; any unit with a probability of carries 2 bits of
information; and so on. The convenience of this measure of informa-
tion-content will be clear if we consider the practical problem of
* encoding* a set of units (which we will first of all assume to be equi-
probable) as groups of binary digits. In the previous section, we saw
that each of a set of eight units could be realized as a distinct group of
three binary digits (cf. 2.3.8). This fact rests on the relationship
between the number 2 (the base of the binary system of calculation)
and 8 (the number of units to be distinguished): 8 = 23. More
generally, if N is the number of units to be distinguished, and m is
the number of positions of contrast in the groups of binary digits
required to distinguish them, then N = 2m. The relationship
between the number of paradigmatic contrasts at the * higher* level
(N) and the syntagmatic length of groups of * lower '-level elements
(m) is therefore logarithmic: m = log2iV. (The logarithm of a
number is the power to which the base of the numerical system must
be raised in order to arrive at the number in question. If N = xm,
then m = log^. N: 'if iV equals x to the power m, then m equals the
logarithm to the base x of N\ It will be remembered that in decimal
arithmetic, the logarithm of 10 is i, of ioo is 2, of 1000 is 3, etc.:
that is, log10 10 = 1, log10 100 = 2, log10 1000 = 3, etc. If informa-
tion-theory had been based on the decimal, rather than the binary,
system of measurement, it would have been more convenient to
define the unit of information in terms of a probability of y^.) It
should be clear to the reader that the equation, N = xm, given here,
is a special case of the equation N = pi x p 2 x p 3 ...p m , introduced in
2.3.8. The equation N = xm holds when there is the same number
(x) of elements in paradigmatic contrast at each of the positions in the
syntagmatic group.
The reason why information-content is generally measured in bits
is simply that many physical systems for the storage and transmission
of information operate on a binary principle: they are two-state
systems. For instance, information may be coded on magnetic tape
(for processing by a digital computer) as a sequence of magnetized or
unmagnetized positions (or groups of positions): each position is in
one of two possible states, and may therefore carry one bit of informa-
tion. Again, information might be transmitted (as, for instance, in
Morse code) as a sequence of 'impulses', each of which can take one
of two values: short or long in duration, positive or negative in
electrical charge, etc. Any system which makes use of an 'alphabet*
of more than two elements can be encoded into a binary system at the
source of transmission and decoded into the original 'alphabet' when
the message is received at its destination. This is what happens when
messages are transmitted by teleprinter. That information-content
should be measured logarithmically to the base 2, rather than
logarithmically to some other numerical base, is a consequence of the
fact that communications-engineers customarily work with two-state
systems. Whether the principle of binary 'coding' is directly relevant
to the study of languages under their more normal conditions of
' transmission' between speaker and hearer, is a matter of consider-
able dispute among linguists. It is certainly the case that many of the
most important phonological, grammatical and semantic distinctions
in language are binary, as we shall see in later chapters; and we have
already seen that one of the two terms of a binary contrast can
frequently be regarded as positive, or marked, and the other as
neutral, or unmarked (cf. 2.3.7). We shall not enter here into the
question whether all linguistic units can be reduced to complexes of
hierarchically-ordered binary 'choices'. The fact that many units (at
all levels of language-structure) are reducible in this way means that
the linguist should be accustomed to thinking in terms of binary
systems. At the same time, it should be realized that the fundamental
concepts of information-theory are quite independent of the parti-
cular assumptions of binarism.
2.4.4 Unequal probabilities

Since each binary digit can carry no more than one bit of information,
a group of m binary digits can carry a maximum of m bits. So far we
have assumed that the higher-level units to be distinguished are equi-
probable. Let us now consider the more interesting and more typical
case, where the probabilities are unequal. For simplicity, we will take
a set of three units, a> b and c, with the following probabilities:
Pa = h Pb = h Pc = h The unit a carries i bit, and b and c each
carry 2 bits of information. They might be encoded in a binary
system of realization as a: 00, b: 01 and c: 10 (with 11 left unassigned).
But if the digits were transmitted in sequence along some channel of
communication and if each digit took the same length of time to send
and receive, this would be an inefficient convention of encoding to
adopt. For a would take up as much of the capacity of the channel as
b and c, although it carries only half as much information. It would be
more economical to encode a as one digit, say 1, and to distinguish b
and c from a by encoding them with the contrasting digit, o, in the
first position: b and c would then be distinguished from one another
in the second position of contrast (which is of course empty for a).
Thus a: 1, b: 00 and c: 01. This second convention makes the most
economical use of the capacity of the channel, because it maximizes
the amount of information carried by each group of one or more
digits. Since a, which occurs twice as often as b and c, is transmitted
in half the time, the maximum number of messages can be trans-
mitted in the shortest time (on the assumption that the messages are
long enough or numerous enough to reflect the average frequencies of
occurrence). In fact, this simple system represents the theoretical ideal:
each of the three units, a, b and c, carries an integral number of bits of
information and is realized by that number of distinctions in substance.
2.4.5 Redundancy and noise

The theoretical ideal is never achieved in practice. First of all, the
probabilities of the units will generally lie between, rather than
4-2
correspond with, the terms in the series i, J, J, -J-, yV> ..., . For
example, a particular unit might have a probability of : it will
therefore convey log2 5approximately 2-3bits of information. But
there is no such thing as o- 3 of a distinction in substance: distinctions in
substance are absolute, in the sense explained above (cf. 2.2.10). And
if we use three digits to identify a unit with a probability of i, we
thereby introduce redundancy into the substantial realization. (The
average redundancy of the system can be minimized to whatever point
is considered desirable; and the mathematical theory of communi-
cation is primarily concerned with this question. But we need not go
into the more technical details here.) The important point is that some
degree of redundancy is in fact desirable in any system of communi-
cation. The reason is that, whatever medium is used for the purpose
of transmitting information, it will be subject to various unpredictable
physical disturbances, which will obliterate or distort part of the
message and thus lead to the loss of information. If the system were
free of redundancy, the information lost would be irrecoverable. The
term used by communication-engineers for random disturbances in
the medium, or channel of communication, is noise. The optimum
system, for a particular channel, is one in which there is just enough
redundancy to enable the receiver to recover the information lost as
a result of noise. It should be observed that the terms ' channel * and
' noise' are to be interpreted in the most general sense. They are not
restricted to acoustically-based systems, still less to the systems
constructed by engineers (telephone, television, teleprinter, etc.). The
distortions produced in one's handwriting by writing in a moving
train can be attributed to * noise'; so too can the distorting effects
on speech of a cold in the head, drunkenness, distraction and lapse of
memory, etc. (Misprints are one of the effects of noise in the 'en-
coding ' of the written language: they are frequently unnoticed by the
reader, because the redundancy of most written sentences is high
enough to counteract the distorting influence of random errors.
Misprints are more serious in a string of figures, any combination of
which is a priori possible. This fact is recognized by the practice of
accountants, who deliberately introduce redundancy into their
ledgers by requiring that the totals of different columns should
balance. And the convention that cheques should specify the amount
to be paid in both words and figures makes it possible for banks to
detect, if not to correct, many of the errors caused by noise of one
kind or another.) As far as the spoken language is concerned, the term
2-4- STATISTICAL STRUCTURE 89
* noise* can be taken to include any source of distortion or misunder-
standing, whether this be attributable to the imperfect performance of
the speaker and hearer or to the acoustic conditions of the physical
environment in which the utterances are produced.
2.4.6 Summary of general principles of information-theory

From the early 1950s communication-theory (or information-theory)
has exerted a powerful influence on a number of different sciences,
including linguistics. The fundamental principles may be summarized
as follows:
(i) All communication rests upon the possibility of choice, or
selection from a set of alternatives. In the chapter on semantics we
shall see that this principle provides us with an interpretation of the
term 'meaningful' (in one of its senses): a linguistic unit, of whatever
level, has no meaning in a given context if it is completely predictable
in that context.
(ii) Information-content varies inversely with probability. The
more predictable a unit is, the less meaning it has. This principle is in
accord with the commonly-expressed view of writers on style, that
cliches (or 'hackneyed expressions' and 'dead metaphors') are less
effective than more 'original* turns of phrase.
(iii) The redundancy in the substantial realization of a linguistic
unit (its 'encoding') is the difference between the number of distinc-
tions in substance required to identify it and its information-content.
A certain degree of redundancy is essential in order to counteract
noise. Our previous discussion of the stability of the substance in
which language is realized, and of the necessity of some 'safety
margin* between the realizations of contrasting elements, can also be
brought within the scope of the more general principle of redun-
dancy (cf. 2.2.10).
(iv) The language will be more efficient (in terms of information-
theory) if the syntagmatic length of units is inversely related to their
probability. That some such principle is indeed operative in language
is suggested by the fact that the most frequently-used words and
expressions tend to be shorter. This was first observed empirically,
rather than deduced as a testable consequence of theoretical prin-
ciples; and a particular formula known as 'Zipf's law' (after its
originator) has been worked out to express the correlation between
length and frequency. (We shall not give 'Zipf's law' here or discuss
its mathematical and linguistic basis; it has in any case been modified
9O 2. THE STRUCTURE OF LANGUAGE
by more recent work.) At the same time, it must be recognized that

the length of a word in letters or sounds (in the sense in which we
have been employing the term * sound* so far) is not necessarily a
direct measure of syntagmatic length. We will return to this point. It is
extremely important, and it has not always been stressed in statistical
treatments of language.
2.4.7 Diachronic implications

Languages, as they develop through time and 'evolve* to meet the
changing needs of the societies that use them for communication, can
be regarded as komeostatic (or 'self-regulating') systems, the state of a
language at any one time being' regulated' by two opposing principles.
The first of these (which is sometimes referred to as the principle of
'least effort') is the tendency to maximize the efficiency of the system
(in the particular sense of 'efficiency' explained above): its effect is to
bring the syntagmatic length of words and utterances closer to the
theoretical ideal. The other principle is 'the desire to be understood':
this inhibits the shortening effect of the principle of 'least effort' by
introducing redundancy at various levels. As a consequence, it is to
be expected that the changing conditions of communication will tend
to keep the two tendencies in balance. If the average amount of noise
is constant for different languages and different stages of the same
language, it would follow that the degree of redundancy in language
is constant. Unfortunately, it is impossible (at present at least) to
verify the hypothesis that languages keep these two opposing
principles in 'homeostatic equilibrium'. The reasons for this will
occupy us presently. Nevertheless, the hypothesis is suggestive. Its
general plausibility is supported by 'Zipf's law'; and also by the fact
(pointed out long before the era of information-theory) that words
tend to be replaced by longer (and more 'colourful') synonyms,
especially in colloquial usage, when frequent use has robbed them of
their 'force' (by reducing their information-content). The extreme
rapidity with which 'slang' expressions are replaced can be accounted
for in these terms.
So too can the phenomenon of 'homonymic conflict' and its dia-
chronic resolution (which has been abundantly illustrated by Gillieron
and his followers). 'Homonymic conflict' may arise when the opera-
tion of the principle of 'least effort' may concur with other factors
determining sound-change to reduce, or eliminate, the 'safety
margin' between the substantial realizations of two words and so
2.4- STATISTICAL STRUCTURE 91
produce homonymy. (The term * homonymy' is commonly used to
refer to either homophony or homography these days: cf. 1.4.2. Here
it is of course homophony.) If the homonyms are more or less
equally probable in a large number of contexts, the 'conflict* will
tend to be resolved (or inhibited) by the replacement of one of the
words. A well-known example is the disappearance from standard
Modern English of the word quean (originally meaning 'woman', and
later 'hussy' or 'prostitute') which was brought into 'conflict' with
queen by the loss of the earlier distinction between the vowels realized
orthographically as ea and ee. The most famous example of 'homo-
nymic conflict' in the literature is probably the case of the words for
'cat' and 'rooster' in the dialects of south-west France. Distinguished
as cattus and gallus in Latin, they had both developed, as a result of
sound-change, into [gat]. The 'conflict' was resolved by the replace-
ment of [gat] = ' rooster' with various other words, including the local
variants offaisan ('pheasant') or vicaire ('curate'). The substitution
of the second of these presumably rested upon some previous con-
nexion in 'slang' usage between 'rooster' and 'curate'. There is a
very rich literature on the subject of 'homonymics'. (References are
given in the notes.)
2.4.8 Conditional probabilities of occurrence

As we have seen, the occurrence of a particular unit (a sound or
letter; an expression-unit; a word; etc.) may be totally or partially
determined by its context. We must now clarify this notion of
contextual determination (or conditioning), and draw out its impli-
cations for linguistic theory. For simplicity, we will first of all restrict
our attention to the consideration of contextual determination as it
operates within syntagmatically-related complexes of units on the
same level of language-structure; that is to say, we will for the present
neglect the all-important point that the complexes of lower-level units
realize higher-level units which themselves have contextually-deter-
mined probabilities.
We will use the symbols x and y as variables, each standing for a
particular unit or syntagmatically-related group of units; and we will
assume that x and y are themselves in syntagmatic relationship. (For
example, at the level of expression-units x might stand for /b/ or
/b/ + /i/, and y for jtj or for /i/ + /t/; at the level of words x might
stand for men or old + men, and y for sing or sing +beautifully.) Both
x and y will have an average, a priori, probability of occurrence: px
andpy, respectively. So too will the combination x+y which we will
symbolize as pxy.
In the limiting case of statistical independence between x and y,
the probability of the combination x +y will be equal to the product of
the probabilities of x andjy ip^ = px xpy. This fundamental principle
of probability theory may be illustrated by means of a simple
numerical example. We will consider the numbers from 10 to 39
(inclusive), and let x and y stand for 2 and 7, in the first and second
position of their decimal representation: the combination of x and y
will therefore stand for the number 27. Within the range of numbers
we are considering (on the assumption that each of the thirty numbers
is equiprobable) px = and py = T^. If we were to * think of a
number between 10 and 39' and to ask someone to guess which
number we had in mind, his chance of guessing correctly (without
further information) would be one in thirty: p^ = ^ . But suppose
that we told him that the number was a multiple of 3. Clearly, his
chance of guessing correctly would now improve to one in ten. More
important from our point of view (since we are considering the prob-
ability of one digit in the context of the other), the choice of one of
the two digits is no longer statistically independent of the choice of
the other. The probability of y, given that x = 2, is ^, since there are
only three multiples of 3 within the range (21, 24, 27); and the
probability of x, given that y = 7, is 1, since there is only one
multiple of 3 that ends in 7 within the range. We may symbolize these
equations as py(x) = J and px(y) = 1. The conditional probability of
occurrence of y in the context of x is J, and the conditional probability
of x given y is 1. (The two expressions 'in the context* and 'given'
are to be understood as equivalent; both are used in works on
statistical linguistics.) To generalize from the example: if px(y) = px
(that is, if the probability of x in the context of y is equal to its
a priori, unconditioned, probability), x is statistically independent of
y; if however, the occurrence of x is rendered more or less probable by
the occurrence of yif either px(y) > px or px(y) < pxthen x is
'positively' or 'negatively' conditioned by y. The extreme case of
'positive'conditioning is, of course, total redundancy, with px(y) = 1
(y presupposes x); and the extreme case of 'negative' conditioning
is, 'impossibility', withpx(y) = o (y precludes x). It is important
to realize that contextual conditioning may be either 'positive' or
'negative' (in the sense in which these terms have been used here);
and also that the probability of x given y is not always, indeed rarely,
equal to the probability of y given x.
2-4- STATISTICAL STRUCTURE 93
Various kinds of conditioning must be distinguished, if the results
of any statistical analyses are to be of significance for linguistics. As
we saw above, syntagmatic relations may be sequential or non-
sequential ; conditioning may therefore be sequential or nonsequential
When x and y are sequentially-related, px(y) is progressive if y
precedes x, and regressive if y follows x. Independently of whether
the conditioning is progressive or regressive, x and y may be con-
tiguous (next to one another in a sequentially-ordered syntagmatic
complex); in this case, if x is conditioned by y, px(y) is transitional.
Many popular accounts of the statistical structure of language tend
to give the impression that the conditional probabilities operative at
all levels of language-structure are necessarily sequential, transitional
and progressive. This is clearly not so. For example, the conditional
probability of a particular noun as subject or object of a particular
verb in Latin is unaffected by the relative order with which the words
occur in temporal sequence (cf. 2.3.5); t n e occurrence of the prefixes
un- and in- in English (in such wTords as unchanging or invariable) is
regressively conditioned; the occurrence of a particular expression-
unit at the beginning of a word may be ' positivelyy or * negatively'
conditioned by the occurrence of a particular expression-unit at the
end of a word (or conversely); and so on.
Of course, it is in principle possible to calculate the conditional
probability for any unit relative to any context. The important thing is
to choose the context and the direction of conditioningthat is, to
calculate px(y) rather than py(x)in the light of what is already
known of the general syntagmatic structure of the language. (A certain
class of units, X, may presuppose or allow the occurrence of another
syntagmatically-related class of units, Y, at a particular position
relative to it (it may also preclude the occurrence of a third class, Z.)
Given that this is so, one may then calculate the conditional prob-
ability of a particular member of the class, Y.) The results will be of
statistical interest if, and only if, px(y) or py(x) is significantly
different from px and py.
2.4.9 Positional probabilities for English consonants

Probabilities can also be calculated for particular structural positions.
For example, three sets of probabilities are given in Table 4 for each of
twelve consonants in spoken English: (i) their a priori probability
averaged over all positions; (ii) their probability in word-initial
position before vowels; and (iii) their probability in word-final
Table 4. Probabilities of selected English consonants

in different positions of the word
(i) (iii)
'Absolute' Initial Final
w 0-070
0-063
0-072
0-042
0-105
0-127
M
[1] 0-052 0-034 0-034
[d] 0-030 0-037 0-039
[h] 0-026 0-065
[m] 0-026 0-058 0-036
M 0-025 0-046 0-014
M 0-019 o-oio 0-048
m 0-017
o-oi6
0-044
O'o6i
o-oio
P>] 0-0005
[P] 0-016 O-O2O 0-008
[g] 0-015 0-027 O-OO2
position after vowels. It will be observed that there are some striking
differences between the frequency with which particular consonants
occur in different positions of the word. For example, of the units
listed [v] is the least frequent in word-initial position, but the third
most frequent in word-final position; on the other hand, [b] is the
third most frequent in word-initial position, but the least frequent in
word-final position (apart from [h], which does not occur at all
finally: n.b. we are talking of sounds, not letters). Others (like [t])
have a high probability, or (like [g] and [p]) a low probability for both
positions. It will also be observed that the range of variation between
the highest and the lowest probability is greater for the end of the word
than it is for the beginning. Facts of this kind find their place in
a description of the statistical structure of phonological words in
English.
It was said earlier (with reference to 'Zipf's law': cf. 2.4.6.) that
the number of sounds or letters in a word is not a direct measure of its
syntagmatic length in terms of information-theory. The reason is of
course that not all sounds or letters are equiprobable in the same
context. If the probability of a phonological or orthographic word
were directly related to the probabilities of its constituent expression-
elements, it would be possible to derive the probability of the word by
multiplying together the probabilities of the expression-elements for
each of the structural positions in the word. For example, given that
x is twice as probable as y in initial position and a is twice as probable
as b in final position, one would expect xpa to be twice as frequent as
either ypa or xpb and four times as frequent zsypb. That this expecta-
tion is not fulfilled in particular instances is evident from the con-
sideration of a few words in English. The expression-elements
realized by [k] and [f] are more or less equally probable initially, but
the word call is far more frequent than fall (according to various
published frequency-lists for English words); although the element
realized by [t] has about fifty times the probability of that realized
by [g] in word-final position, the word big occurs with about four
times the frequency of bit; and so on.
The probabilities for initial and final position used in these calcu-
lations (cf. Table 4) are probabilities based on the analysis of con-
tinuous text. This means that the occurrence of a particular consonant
in relatively few high-frequency words may outweigh the occurrence
of another in very many low-frequency words (cf. the remarks made
in 2.4.1, in connexion with the concept of * functional load'). The
consonant [8], which occurs initially in such words as the, then, their,
them, etc., in English, illustrates the effect of this weighting. In initial
position, it is the most frequent of all consonants, with a probability
of about o-io (compared with 0-072 for [t], 0-046 for [k], etc.). But it
occurs in only a handful of different words (less than thirty in modern
usage). By contrast initial [k] is found in many hundreds of different
words, although its probability in continuous text is less than half
that of [8]. Comparison of all the words in English realized as
consonant + vowel + consonant (which in itself is a very common
structure for English phonological words) shows that in general
there are more words with a high-frequency initial and final con-
sonant than there are words with a low-frequency initial and final
consonant, and also that the former tend to be of more frequent
occurrence. At the same time, it must be stressed that certain words
are far more frequent or far less frequent than one might predict from
the probabilities of their constituent expression-elements.
2.4.10 * Layers' of conditioning

Although we have so far discussed the question of contextual determi-
nation with reference to the conditional probabilities holding between
units of the same level, it is clear that the occurrence of a particular
expression-element is very largely determined by the contextual
probability of the phonological word of which it forms a part. For
example, the three words written book, look and took are all of frequent
occurrence: they differ from one another phonologically (and ortho-
graphically) solely in the consonant which appears in the initial
position. From the point of view of the grammatical structure of
English, the possibility of contrast between the three words in actual
utterances is relatively small (and it is totally unrelated to the prob-
abilities of the initial consonants). The word took stands apart from
the other two in a number of ways, the most important of these being
that it realizes the past tense of a verb. It therefore occurs more freely
with such words and phrases as yesterday or last year than do look
and book (where the phonological words corresponding to took are the
words written looked and booked); it can have he, she or it or a singular
noun as subject (he took, etc., but not he look or he book, etc.); and it
cannot occur with to preceding it (e.g. / am going to took is unaccept-
able). But book and look also differ from one another grammatically.
Each of them may be used as a noun or a verb in the appropriate
context (it must not be forgotten that a phonological word may
realize more than one grammatical word: cf. 2.2.11). Although look is
far more frequent as a verb and book as a noun, this difference is of
small account compared with such non-statistical grammatical facts
as the following: as a verb book (i.e. 'make reservations', etc.), but
not look, may take a noun or noun-phrase as its direct object (/ will
book my seat, He is going to book my friend for speeding: the word look
is impossible here); and look normally takes a * prepositional phrase'
(/ will look into the matter, They never look at me: here book is im-
possible). It would seem to be the case that in the majority of English
utterances produced by speakers in their everyday use of the language
the confusion of book and look is precluded by grammatical constraints
of one kind or another. And this is quite typical of minimally-
contrasting phonological words in English.
But let us now consider the relatively small set of utterances in
which both look and book are grammatically acceptable. It is quite easy
for a speaker of English to devise, and on occasion he will produce or
hear, such utterances. An example might be / looked for the theatre:
I booked for the theatre. If we assume, for the sake of the argument,
that all but the initial consonant of the words booking or looking is
'transmitted' to the hearer without significant distortion by 'noise'
in the 'channel', he is faced with the problem of 'predicting', on the
basis of the, redundancies in the language and in the situation of
utterance, which of the two words was intended by the speaker. (For
simplicity, we will suppose that cooked, etc., are impossible, or highly
improbable, in the particular situation.) Although looked may be
assumed to occur far more frequently than booked in any representa-
tive sample of (British) English, it is quite clear that the occurrence
of theatre raises the probability of booked considerably. Whether
booked or looked is more probable with for the theatre, it would be
very hard to say. But in any particular situation either one may be
more highly determined than the other. This is evident from a
comparison of the following two longer utterances:
(i) / looked for the theatre, but I couldn't find it.
(ii) I booked for the theatre, but I have lost the tickets.
The word booked would appear to be contextually precluded in (i)
and looked in (ii). But the situation itself, including any previous con-
versation, might also have introduced various 'presuppositions', the
determining force of which is as strong as that of but and couldritfind
in (i) and but and tickets in (ii). If this is so, the hearer will already
be 'conditioned' by these presuppositions to 'predict* (that is, in
fact, to hear) looked rather than booked (or the converse) in the
shorter ' frame' / -ookedfor the theatre. For the present, we may refer
to these probabilities, which derive from the co-occurrence of one
word with another and the 'presuppositions' of the particular
situation of utterance, as 'semantic'. (In later chapters we will distin-
guish other levels of acceptability within what we are here calling
'semantics'.)
Our example has been drastically over-simplified: we have recog-
nized only three levels of conditioning (phonological, grammatical
and semantic), and we have assumed that only one expression-unit
has been lost, or distorted, through 'noise'. These simplifications do
not, however, affect the general conclusion of the argument. When it
comes to the discussion of particular utterances, it must be recognized
that the semantic probabilities outweigh the grammatical and the
grammatical outweigh the phonological. Since it is impossible (in the
present state of linguistic research at least) to identify all the seman-
tically-relevant factors in the external situations in which particular
utterances occur, it is also impossible to calculate the probability, and
therefore the information-content, of any part of them. This is one
of the points that was stressed in our earlier discussion of functional
load and information-theory (cf. 2.4.1).
2.4.11 A dilemma and its methodological resolution

Two apparently contradictory principles have been maintained in this
section: first, that statistical considerations are essential to an under-
standing of the operation and development of languages; second, that
it is in practice (and perhaps also in principle) impossible to calculate
precisely the information carried by linguistic units in actual utter-
ances. This apparent contradiction is resolved by recognizing that
linguistic theory, at the present time at least, is not, and cannot, be
concerned with the production and understanding of utterances in
their actual situations of use (except for a relatively small class of
language-utterances which can be handled directly in this way:
cf. 5.2.5), but with the structure of sentences considered in abstraction
from the situations in which actual utterances occur.
John Lyons
Online ISBN: 9781139165570
Chapter
3 - The Sounds of Language pp. 99-132

3
THE SOUNDS OF LANGUAGE
3.1 Introductory
3. i. i Phonetics and phonology
So far we have been taking for granted the traditional view of
language, according to which sentences are composed of words and
words of 'sounds' (cf. 2.1.1). In the course of our discussion of
substance and form in the previous chapter we saw that the term
* sounds' is potentially ambiguous. We must now clear up this
ambiguity.
If the linguist is asked whether two 'sounds' are the same or
different, or how many 'sounds' there are in a given language, he
must know whether the question is one of substance or form;
whether these 'sounds' are to be regarded as physical entities which
can be described without knowing to what language they belong or
whether they are to be described in terms of such differences and
similarities of sound as are functional in the language (by 'functional'
is to be understood 'relevant for the purpose of communication'). In
the first case he will give a phonetic description of what he hears or
analyses instrumentally; in the second he will give a phonological
description.
3.1.2 Speech-sounds
Let us now distinguish, provisionally, between the terms 'speech-
sound' and 'phoneme'. A speech-sound is any phonetically distinct
unit of sound; that is to say, any unit of sound produced by the
speech-organs that can be distinguished by the phonetician from all
other units of sound produced by the speech-organs. Since there is
practically no limit to the number of different speech-sounds that can
be produced by the human speech-organs and distinguished by the
phonetician, it follows that the speech-sound is somewhat indetermi-
nate in nature. Individual phoneticians will differ in expertise; and
even with the various instrumental techniques now available for
measuring the acoustic properties of sound, there is in the last resort
[99]
IOO 3. THE SOUNDS OF LANGUAGE
room for minute differences and discrepancies. The fact is that there
is no natural limit to the divisibility of the range of sounds used in
speech. The point at which the phonetician stops distinguishing
different speech-sounds is dictated either by the limits of his own
capacities and those of his instruments or (more usually) by the
particular purposes of the analysis. For some purposes he will wish
to draw more refined distinctions than for others. According to the
degree of refinement in representation which he seeks to achieve he
will use what is called a 'broader* or a 'narrower* transcription.
(There are, of course, any number of intermediate stages between
'broadest* and 'narrowest*.) Let me give some examples.
In English, p, t and k in certain positions of the word are slightly
aspirated (that is, pronounced wTith an accompanying slight puff of
breath); in other positions, after s for example, they are unaspirated
(cf. top: stop,pot: spot, etc.). In a broad transcription, therefore, the
phonetician might well use the same letter (or other symbol) to
represent both the English speech-sounds (as indeed the alphabet
used for English generally does), although they are quite easily
distinguishable, phonetically. Again, in English there are two main
'I-sounds*, impressionistically referred to as 'clear' and 'dark*: the
former occurs before vowels, the latter before consonants and at the
end of words (cf. leaf, late: feel, field). A narrow transcription would
mark the difference between the 'clear' and the 'dark' I; a broad
transcription might not.
3.1.3 Phonemes and allophones

A discussion of broad transcriptions brings us naturally to the notion
of the phoneme. To return to the example of the English p, t and k:
the reason why we can represent both the aspirated and the un-
aspirated consonants with the same symbol in broad transcription is
that the distinction between the aspirated and the unaspirated variety
never has the function of keeping apart different words in English
(this is a very crude, and partly inaccurate, statement of a principle
that will be treated more fully later); it is not a functional difference:
it is a phonetic difference, but not a phonological, or phonemic,
difference of English (for the present, we shall not distinguish between
the terms 'phonological* and 'phonemic'). In other languages (for
example, in some Northern Indian languages and in most dialects of
Chinese) the distinction between aspirated and unaspirated conso-
nants is functional, or phonemic. And there are phonemic differences
3.1. INTRODUCTORY IOI
of English that play no role (though they may occur, as speech-
sounds) in other languages. For example, in certain languages the
distinction between the speech-sounds d and t, b and p, g and k, etc.
(between what the phonetician describes as voiced and voiceless
consonants) is not a phonemic differencefor instance, p, t and k
might occur at the beginning and end of words and b, d and g only
in the middle of words. In cases of this kind we say that the phonetic-
ally distinguishable pairs of speech-sounds are positional variants, or
allophoneSy of the same phoneme. They are called positional variants
because the occurrence of one rather than another of the phonetic
variants of a particular phoneme is determined by the position of the
phoneme in the word.
It will now be clear why the word 'sounds' is ambiguous, and why
a distinction must be made between the units of phonetic description
(speech-sounds), on the one hand, and the units of phonological
description (phonemes), on the other. In order to distinguish speech-
sounds from phonemes in the transcription of utterances, it has
become customary to put symbols standing for the latter between
obliques and the the former between square brackets; and we shall
follow this convention throughout. We might say, for instance, with
reference to the example used above, that /p/, /t/ and /k/ have as
allophones [p], [t] and [k] in initial and final position and [b], [d] and
[g] in medial position. A fuller treatment of the concept of the
phoneme in modern linguistic theory will be given presently. We
must first discuss some of the general principles of phonetic analysis.
3.2 Phonetics
3.2.1 Different branches of phonetics
Phonetics is not to be regarded as just an ancillary tool for the linguist
and the student of foreign languages. As practised today, it is a highly
developed science, incorporating parts of physiology and physics, but
with its own conditions of relevance, its own methods of investigation
and experiment, and its own technical vocabulary.
Speech-sounds are often described non-tectmically in subjective,
and largely meaningless, terms, as 'harsh', 'guttural', 'soft', 'flat',
etc. They can be described objectively from three main points of
view: (1) in terms of the manner of their production by the human
speech-organs; (2) in terms of the acoustic properties of the sound-
waves travelling between speaker and hearer; and (3) in terms of
THE
102 3- SOUNDS OF LANGUAGE
their physical effects upon the human ear and its associated mechan-
isms. This yields a threefold division of the subject into articulatory,
acoustic and auditory phonetics.
3.2.2 Auditory phonetics

The auditory classification of speech-sounds has not yet been carried
very far; and for practical purposes, at the present time, phonetics
can be regarded as being made up of two main branches: articulatory
and acoustic phonetics. It would be impossible in the space available
here to give anything like a satisfactory treatment of either of these
branches. All one can hope to do is to provide a brief account of the
general principles of classification and theory.
3.2.3 Articulatory phonetics

The oldest, and still the most common, method of phonetic descrip-
tion is that made in terms of 'articulation' by the speech-organs.
What the phonetician, for convenience and from his own particular
point of view, calls 'the organs of speech* do not, it is true, form a
primary physiological system. Most of the speech-organs (the lips, the
teeth, the tongue, the vocal cords, the nose, the lungs, etc.) serve some
other purpose also, which, in man as in the other animals, can be
considered as biologically primary. However, the speech-organs are
legitimately considered by the phoneticians as a unitary, inter-
connected system. Individual differences in the speech-organs of
different persons (apart from obvious physical defects) have no
significant effect on the kind of sounds produced. The difference
between a boy's voice and a man's, or between a man's and a woman's,
is not generally relevant in speech. Nor is it true, although it is often
said, that the speech-organs of certain 'races' are physically adapted
to the production of certain sounds rather than others. 'Inability' to
produce certain sounds is generally a result of environmental factors
in childhood, the main factor being that of learning one's native
language as one hears it pronounced.
3.2.4 Voice and pitch

Most phoneticians divide speech-sounds into consonants and vowels.
This classification depends upon a prior distinction of voiced and
voiceless sounds. Voice is produced by the regular vibration of the
3-2. PHONETICS IO3
vocal cords in the larynx, or * Adam's apple*. The rate of vibration
depends on the degree of tension, and this governs the pitch of the
resulting sound. (Pitch is usually treated as a concomitant feature of
the sounds produced, rather than as an integral part of them.) Vowels
are defined (in articulatory terms) as voiced sounds in the formation
of which the air passes through the pharynx and the mouth without
obstruction (by the tongue, lips, teeth, etc.). All speech-sounds other
than vowels are defined to be consonants. Consonants are therefore
rather heterogeneous. Indeed, a number of phoneticians would deny
that the distinction between vowels and consonants is as fundamental
as it is usually said to be. Not only are there such apparently para-
doxical sounds as * voiceless vowels', but voicing may be complete or
partial and the term 'obstruction' is far from being as precise as it
might appear at first sight. Supplementary criteria of different kinds
must be brought to bear in the case of partially obstructed sounds to
decide whether they are to be classed as vowels or consonants. These
are points we need not dwell upon. They are mentioned here in
order to support the view, which will be stressed throughout this
section, that sound as produced by the human speech-organs is a
continuum within which there are perhaps no 'natural' absolute
categories.
3.2.5 Vowels
Vowels are generally classified in terms of three main articulatory
dimensions: the degree to which the mouth is opened (close v. open);
the position of the highest part of the tongue (front v. back); and the
position of the lips (rounded v. spread, or unrounded). Thus, a certain
sound may be described as a close, front, rounded vowel (e.g. the
vowel in the French word lune); another as close, front, unrounded
(e.g. that of the French si or of the English word sea). The terms
'close', 'front', etc., are not of course used absolutely. Each of the
three dimensions represents a continuum and, in principle, is
infinitely divisible. The vowel of the French word si, for instance, is
closer than that of the English word sea (as well as being different in
certain other respects). What has been done for the purpose of
classification is to establish a set of so-called ' cardinal' vowels; that is
(to quote Daniel Jones, the originator of this system of classification)
'a set of fixed vowel-sounds having known acoustic qualities and
known tongue and lip positions'. Although Jones talks of the 'known
acoustic qualities' of the cardinal vowels, the system is primarily, and
104 3- T H E SOUNDS OF LANGUAGE
to the extent that it is objective, one of articulatory classification, The
principle followed is that of first defining a set of articulatory extreme*
(the * closest possible front vowel', 'the openest of the back vowels',
etc.) and then selecting intermediate positions in such a way that 'the
degrees of acoustic separation are approximately equal'. The judge-
ment of the intervals of 'acoustic separation' is subjective (though of
course that of a brilliant and highly skilled phonetician). Now that the
value of each of the cardinal vowels has been fixed and Daniel Jones's
own recording of them has been analysed instrumentally their
acoustic properties are, within certain limits, known objectively. That
the cardinal vowel-system is in the last resort arbitrary does not make
it any the less useful or scientific: the values of the cardinal vowels are
fixed in the International Phonetic Alphabet (IPA), and by reference
to them as a standard of measurement trained phoneticians can
satisfactorily describe the vowels of any language, to a greater or less
degree of precision according to need.
In addition to the three articulatory variables which the phonetician
selects as basic in the classification of vowels, there are several others
which he customarily treats as secondary modifications of what is
taken to be the 'normal* pronunciation. For instance, vowels are
assumed to be pronounced 'normally* without the passage of air
through the nose. If the nasal passage is kept open during the pro-
duction of the vowel so that air passes out from both the mouth and
the nose, the vowel is said to be nasalized; it is regarded, and repre-
sented by the IPA, as a modification of the corresponding non-
nasalized vowel (cf. [a]: [3]).
3.2.6 Consonants
Consonants fall into several different, interesting categories. They
may be voiced or voiceless; and oral or nasal. If the obstruction in the
air passage is complete, the resulting sounds are described as stops
(or plosives); if the obstruction is only partial, but produces friction,
they are called fricatives (or spirants). The place at which the obstruc-
tion occurs is referred to as the point of articulation: the lips, the
teeth, the alveolae (or teeth ridge), the palate (i.e. the hard palate),
the velum (or soft palate), the uvula, the pharynx, the glottisthese
are the principal points of articulation to which recognition is given
in the IPA. The organ (or that part of it) which is brought into contact
with (or close to) the point of articulation is called the articulator: in
most cases this is some part of the tongue, which for the purpose is
3-2. PHONETICS IO5
regarded as being made up of four partsthe apex (or tip), the blade,
the middle, the dorsum (or back). Apical consonants are those in
which the tip of the tongue acts as articulator; dorsal consonants
those in which the articulator is the back of the tongue. Most of the
traditional terms used by phoneticians are interpreted under the tacit
assumption that certain articulators are normally associated with
particular points of articulation: for instance, the term Mental* is
used for sounds (e.g. [t], [d]) which are produced by bringing the tip
of the tongue into contact with, or close to, the (upper) teeth; the
term ' velar', on the other hand, is used for sounds (e.g. [k], [g]) which
are produced by contact, or friction, between the back of the tongue
and the soft palate. In other cases the term employed is a compound
label specifying the articulator and the place of articulation: thus
* labiodental' denotes sounds (e.g. [f], [v]) which result from contact,
or friction, between the (upper) teeth and the (lower) lip. And, if
necessary, further terms can be freely created on the same pattern.
Consonants may be classified then in terms of a number of different
articulatory variables (we have mentioned only some of them). For
example, according to the conventions of the IPA, [p] is a bilabial,
voiceless, oral, stop; [b] is a bilabial, voiced, oral, stop; [f] is a labio-
dental, voiceless, (oral) fricative; [m] is a bilabial, (voiced) nasal, stop;
[t] is a dental (or alveolar), voiceless, oral stop; [n] is a dental (or
alveolar), (voiced) nasal stop; and so on. Perceptible deviations from
what are taken to be the ' normal' points of articulation are described
as being 'advanced' or 'retracted'. For instance, the initial sounds of
the English words key and car are phonetically different: they would
commonly be described as an 'advanced' and 'retracted' (or 'front'
and 'back') variety of [k]. This does not mean that they are different
'species' of some natural 'genus' of sounds, but only that the roof
of the mouth and the surface of the tongue are conventionally divided
by phoneticians into a smaller, rather than a larger, number of regions.
3.2.7 Articulatory variables: 'long' and 'short' components

So far we have talked as if speech were made up of sequences of
physically discrete units. This is not the case. Speech consists of
continuous bursts of sound of varying length, within which the only
physical breaks are at points where the speaker pauses to take breath.
What the phonetician calls speech-sounds are abstracted by him from
the continuous streams of sound according to some explicit or implicit
criteria of demarcation. Simplifying the question somewhat, we may
106 3. THE SOUNDS OF LANGUAGE
say that the number of speech-sounds recognizable in a continuous
stretch of speech is determined by the number of successive distinct
states of the speech-organs: a change in the state of the speech-organs
may be defined as a perceptible change in at least one of the recognized
articulatory variables. It follows that certain of the variables may keep
the same value throughout a number of successive states of the speech-
organs. Consider the two English words cats and cadsy which may be
written, in a broad transcription, as [kats] and [kadz]. It will be
observed that the phonetic notation used here represents the dif-
ference between the two words as the sum of the differences between
two distinct speech-sounds (i.e. [t]: [d] and [s]: [z]). In fact the
difference lies in only one of the articulatory variables (we may
neglect the difference in the vowels): in [kats] the portion of the word
occurring after the vowel is voiceless; in [kadz] it is voiced (this is
something of a simplification; but it does not affect the point being
made). We can distinguish therefore between two kinds of phonetic
components, which may be conveniently called 'short* and 'long*
according to whether they occupy one or more than one position. It is
important to observe that the same articulatory component may be
long or short in different phonetic sequences. For example, in the
word under the component of voicing extends over the whole word,
and that of dental contact over two positions [nd], whereas nasality
is confined to only one position [n]; in the word omnipotent, how-
ever, nasality is a long component, in [mn], and it is the shift from
labial to dental (more precisely, alveolar) contact which leads us to
distinguish between the positions occupied by the speech-sounds
[m] and [n].
Certain articulatory components, notably voice and nasality, being
unaffected by the presence or absence of obstruction in the mouth,
may enter into the formation of both consonants and vowels. For
example, the relations that hold between the vowels written as [a],
[a] and [a] in the International Phonetic Alphabet (IPA) are identical
with those that hold between the consonants [p], [b] and [m]; [t], [d]
and [n]; [k], [g] and [Q]: the first member of each set of three is
neither voiced nor nasalized, the second is voiced but not nasalized,
and the third is both voiced and nasalized. This point may be
illustrated by means of a hypothetical example. Let us suppose that in
a given language there are three words which may be transcribed
phonetically as (a) [patak], (b) [badag] and (c) [manarj]. There are five
articulatory variables involved: three points of articulation
(i) labial, (ii) dental and (iii) velar contactand two components
32. PHONETICS IO7
which may enter into the formation of both consonants and vowels
the presence or absence of (iv) voice and (v) nasality. The phonetic
relationships between the three words can be represented by means
of the two-dimensional matrices given in Fig. 3, in which the
horizontal dimension is used for temporal succession and the vertical
dimension for the specification of the 'positive* or 'negative' value
for each of the five relevant articulatory variables: the 'positive'
value (voiced, rather than voiceless; nasal, rather than oral; dental,
rather than either labial or velar, contact; etc.) is marked with a
horizontal broken line; and the 'negative' value is left unmarked.
These matrices are absolutely equivalent to the more usual alphabetic
transcriptions: (a) [pgtak], (b) [badag], (c) [manarj].
(c)
Labial
Dental -
Velar
Nasal
Voiced
patak badag m a n a ij
Fig. 3. Note that, since 'labial*, 'dental' and 'velar* imply obstruction at
these points of articulation (the articulatory components of the vowels are
not giventhey are assumed to be constant in value), the distinction of
vowels and consonants, and hence the recognition of five states of the speech
organs, is clear from the absence of obstruction in two portions, and the
presence of obstruction in three portions, of each word.
In the case of [patak], [badag] and [manag] it is clear that the

presence or absence of voice and nasality are long components
running from the beginning to the end of each word (it will be
recalled that the small circle put under a vowel marks it as voiceless
and the 'tilde' (~) placed over a vowel marks it as nasalized, and that,
unless they are specified as voiceless, vowels are assumed to be
voiced, as 'normally'); it is also clear that the sequence of obstructed
and unobstructed segments distinguishes consonants and vowels, and
in this case the number of speech-sounds. Careful consideration of
this example and generalization of the facts implicit in it should lead
us to the conclusion that such expressions as 'a [p]-sound', 'an
108 3. THE SOUNDS OF LANGUAGE
[mj-sound', etc. are convenient, but potentially misleading, ways of
saying ' a stretch of sound which is characterized at a particular place
by labiality, occlusion, voicelessness and lack of nasality', 'a stretch
of sound characterized at a particular place by labiality, occlusion,
voice and nasality', etc. (to mention only a few of the relevant
articulatory variables). The alphabetic system of transcription
customarily used by phoneticians, and used here in citing the
examples, is therefore a far from perfect system for the transcription
of speech. A more satisfactory system might be one which gave a
direct representation of each of the overlapping articulatory com-
ponents of any given stretch of speech and explicitly marked the
relative lengths of the components. As it is, anyone who wishes to
read phonetic transcriptions with understanding must first learn to
* de-alphabetize' them, mentally substituting for each of the symbols
the simultaneous articulatory components they imply and then
combining these components with one another in sequence ac-
cording to the principles illustrated above.
This question has been discussed at some length because it is not
always stressed sufficiently, and is sometimes not even mentioned,
in elementary and popular treatments of language. Only too often
the impression is given that speech is composed of sequences of
discrete and independent speech-sounds, and that these speech-
sounds are the ultimate units of articulatory phonetic analysis. We
shall see presently that the distinction that has been made here on
phonetic grounds between short and long components is one which can
be made also on phonological grounds.
3.2.8 Phonetic alphabets

The reader will now readily appreciate the following point. The
International Phonetic Alphabet (IPA) is based upon the Roman
alphabet. Moreover, the phonetic values assigned to the letters were
established before the necessity of distinguishing between phonetics
and phonology had been clearly recognized. The IPA betrays its
origins in the inconsistency with which it represents certain articu-
latory features and in its assumptions of 'normality' and 'abnor-
mality'. In general, the letters were defined to have the values of the
'sounds' they denoted in one or more of the major European
languages using the Roman alphabet; some letters which were
redundant, e.g. x, as used in English, etc. (= [ks]) and c, as used in
English and French, etc. (= [k] or [s]), were given a different
3-2. PHONETICS IO9
phonetic interpretation; and a certain number of new symbols
(borrowed from the Greek alphabet or specially designed for the
purpose) were introduced, e.g. [9] and [6] to represent the initial
' sounds' of the English words thick and there. Although the inventors
of the IPA and those responsible for its further development have
shown great ingenuity, as well as a commendable attention to the
requirements of typographical clarity, in the design of new symbols
for the alphabet, there was clearly a practical limit to the number of
totally distinct symbols that could be constructed. So the IPA has
from the outset made provision for the use of diacritics in addition to
letters. The diacritics are employed, as we saw earlier, either to give a
' narrower' transcription than is possible by using the letters alone or
to show that a certain ' sound' is pronounced in a manner which the
conventions of the IPA regard as 'abnormal'.
The distinction that is made in the IPA system between letters and
diacritics can be misleading. Almost inevitably it suggests that the
differences between the sounds represented by distinct letters are
more fundamental than those marked by diacritics. For example,
voiced and voiceless oral stops are represented by distinct letters
(M- [*]> [b]: [p]> [g] [k], etc.); but as we have seen, vowels and nasal
consonants are assumed to be voiced unless marked as voiceless by
means of a special diacritic. From the point of view of articulatory
description, as we have seen above, the relationship between [d] and
[t] is the same as that between [a] and [a] or between [n] and [n]. The
conventions of the IPA might tend to suggest, or support, the notion
that, whereas [d] and [t] are two quite different 'sounds', the voiceless
[] and [n] are merely less common variants of the more 'normal'
voiced [a] and [n]. There are many inconsistencies. Nevertheless, the
IPA remains a very useful tool for the phonetician; for its conventions
have been carefully defined and have been accepted by phoneticians
of many different countries.
There are other alphabetic systems of phonetic transcription besides
the IPA. Some of these have the advantage that they can be typed on
a standard typewriter, and are for this reason preferred by many
authors, especially in America. It should not be thought, however,
that any important difference of principle distinguishes the various
phonetic alphabets in common use. All the remarks which have been
made above about the IPA apply to phonetic alphabets in general.
IIO 3. THE SOUNDS OF LANGUAGE
3.2.9 Acoustic phonetics

It is in the field of acoustic phonetics that the most striking develop-
ments have taken place since the Second World War. With the aid
of various kinds of electronic equipment, the most widely used of
which is the sound spectrograph, it is now possible to analyse the
complex sound waves produced in speech into their component
frequencies and relative amplitudes as these vary continuously
through time. Furthermore, considerable progress has been made in
what is called speech-synthesis: the artificial reconstruction of recog-
nizable utterances in particular languages by producing sound waves
at the frequency-bands that have been found to be of particular
importance in human speech and combining these in the appropriate
way. We shall not go into the principles of acoustic phonetics. All that
is required in a book of this nature is some brief indication of the
results that have been achieved and their implications for linguistic
theory.
First of all, acoustic analysis has confirmed (if confirmation was
needed) that speech is not made up of a sequence of discrete * sounds \
The determinable components of the sound-waves produced when we
talk are of varying lengths and overlap one another in the time-
sequence. Rather more surprisingly perhaps, the factors identified by
acoustic analysis are not always relatable in a very straightforward
manner to the factors traditionally regarded as criterial in the distinc-
tion of different speech-sounds by the articulatory phonetician. The
acoustic analysis of vowels in terms of their constituent formants
(i.e. the two frequency-bands at which there is the maximum concen-
tration of energy) correlates quite well with their description in terms
of the articulatory dimensions of front v. back and close v. open; the
articulatory feature of rounding can also be identified acoustically.
So too can voice and nasality; as well as obstruction and friction. It is
the acoustic identification of the place of articulation that presents the
greatest difficulty. From the acoustic point of view, it would appear
that the distinction between consonants produced at different places
of articulation is carried mainly by contextual and transitional features
in the sound-wave, rather than by some features inherent in the
consonants themselves.
Although it has long been known that what is perceptually the
same sound can, in certain instances, be produced by a different
combination of 'articulations', it was probably assumed by most
phoneticians until recently that the articulatory distinctions were
3.2. PHONETICS III
for the most part maintained by the sound-waves as these were
transmitted for analysis by the auditory system of the hearer.
It is now realized that the transmission of speech is rather more
complex.
Allowance must be made for feedback. In the case of speech, we are
not dealing with a system of sound-production and sound-reception
in which the 'transmitter' (the speaker) and the 'receiver' (the
hearer) are completely separate mechanisms. Every normal speaker of
a language is alternately a producer and a receiver. When he is
speaking, he is not only producing sound; he is also 'monitoring'
what he is saying and modulating his speech, unconsciously cor-
relating his various articulatory movements with what he hears and
making continual adjustments (like a thermostat, which controls the
source of heat as a result of' feedback' from the temperature readings).
And when he is listening to someone else speaking, he is not merely a
passive receiver of sounds emitted by the speaker: he is registering
the sounds he hears (interpreting the acoustic 'signal') in the light
of his own experience as a speaker, with a 'built-in' set of contextual
cues and expectancies. For this reason, therefore, the primary medium
of language (the substance of the expression-plane: cf. 2.2.8) is
perhaps not rightly conceived in wholly physical terms (in the
narrowest sense of the term 'physical'). The phonic data is not just
sound as it might be treated by the physicist, but sound as 'filtered'
and categorized by human beings in their use of language. In other
words, the phonic medium of language has a psychological, as well as
a purely physical, aspect. It is possible that this psychological aspect
of the medium may yet be reducible to a rather more complex
physical description of the properties of sound. But this should not
be taken for granted.
It may be pointed out here that the principle of 'feedback' is not
restricted to the production and reception of physical distinctions in
the substance, or medium, in which language is manifest. It operates
also in the determination of phonological and grammatical structure.
Intrinsically ambiguous utterances will be interpreted in one way,
rather than another, because certain expectancies have been estab-
lished by the general context in which the utterance is made or by
the previous discourse (cf. 2.4.5, o n redundancy).
THE
3.3 Phonology
3.3.1 The phoneme
In the previous section a distinction was made between the phonetic
and the phonological analysis of language. The principles underlying
phonetic analysis have now been discussed in sufficient detail, and we
may look a little more closely at the concept of the phoneme, which
has been introduced as the unit of phonological description.
3.3.2 Complementary distribution of allophones

We may begin our further discussion of the theory of the phoneme
from the point we reached earlier at which we distinguished between
phonemes and * speech-sounds' (cf. 3.1.2). Two phonetically different
'sounds' in the same environment which have the effect of dis-
tinguishing different words are recognized as different phonemes.
Thus, for example, [I] and [r] are different phonemes in English
because they distinguish numerous pairs of words, such as lamb: ram,
lot: rot, light: right, etc. We may refer to these phonemes as /I/ and /r/,
respectively. (It will be recalled that square brackets are being used
for phonetic symbols and oblique strokes for phonemic symbols,
according to the general practice of linguists.) There are many
languages in which [I] and [r] either do not both occur or do not both
occur in the same environment with the effect of distinguishing
different words. In such languages (e.g. in Chinese and Japanese) the
difference between [I] and [r] is not phonemic. Phonetic units that
never occur in the same environment (and therefore cannot distin-
guish different words) are said to be in complementary distribution
(cf. 2.3.1). The example of the 'clear' and 'dark' [I] in English has
already been given: the 'clear' [I] occurs only before vowels and the
phoneme /y/, and the 'dark' [I] before all other consonants and at the
end of words. The phonetically distinct 'clear' and 'dark' [I] are
recognized therefore as positional variants, or allophones, of the same
phoneme. There is no reason to consider the distinction between [I]
and [r] as more 'naturally' phonemic than that between the 'clear'
and ' dark' [I]. This is a contingent fact of English. Not only are there
languages in which [I] and [r] are in complementary distribution and
recognized by the phonologist as allophones of the same phoneme,
but there are also languages (e.g. Russian and certain dialects of
Polish) in which the distinction between a 'clear' and 'dark' [I] is
phonemic. It may be observed in passing that the application of the
3.3. PHONOLOGY 113
principle of complementary distribution will generally correlate pretty
well with the phonetically-untrained native speaker's judgement about
what is and what is not the same 'sound'. This may be attributed to
the fact that the native speaker has learned to respond to certain
phonetic differences as functional in his language and to neglect others
as irrelevant for the purpose of communication.
3.3.3 Phonetic similarity of allophones

Implicit in what has been said so far is a further condition for
grouping different speech-sounds into a smaller set of phonemes. It is
clearly not a sufficient condition for their allocation as variants of the
same phoneme that speech-sounds should be in complementary
distribution. To return to yet another earlier example, the voiceless
stop consonants of English are pronounced differently in different
environments, with slight aspiration in some cases and without any
aspiration in others. This means that each of the set of aspirated stops
(let us write them in a narrow transcription as [ph], [th], [kh]) is in
complementary distribution with each of the set of unaspirated stops
(let us write these as [p], [t], [k]). Why not group [ph] with [t] or [k],
rather than with [p]? Notice that, provided the conventions of
phonetic interpretation are made clear, /top/: /spop/ and /pot/: /stot/
are no less adequate representations of the contrasts between the
words normally written top: stop and pot: spot9 than are /top/: /stop/
and /pot/: /spot/. All we have to do, if we choose to make the identifi-
cations implied in the first way of transcribing the words, is to
associate with the phonemes /t/ and /p/ the following conventions of
phonetic interpretation: /t/ and /p/ are realized phonetically as [th]
and [ph] in initial and final position, and as [p] and [t], respectively,
when they occur after /s/. However, no phonologist describing
English would make this particular set of identifications. And there are
a number of reasons why he would not. The first, and most important,
supplementary criterion (to which most linguists would give as much
weight as they do to the condition of complementary distribution) is that
of phonetic similarity. And it is this criterion that has been implicit in
the discussion of the various examples brought forward to illustrate
the difference between speech-sounds and phonemes, [ph] and [p],
rather than [ph] and [t], are considered as allophones of the same
phoneme /p/ because phonetically they are more similar to one
another than [ph] and [t] (which are also in complementary distri-
bution).
THE
It will be clear from our discussion of the principles of phonetic
analysis that whether or not two speech-sounds are similar is not a
question that admits of a simple answer. It follows from the multi-
dimensional nature of the sounds used in human speech that they
may be alike in some respects and unlike in others. This means that
the phonologist analysing a particular language may be faced with
alternative possibilities. Is a voiceless, unaspirated stop more like a
voiced, unaspirated stop or a voiceless, aspirated stop? This is a
question that confronts the linguist analysing English. For [p], [t]
and [k] are in complementary distribution not only with [ph], [th]
and [kh], but also with [b], [d] and [g], which never occur after /s/
in the same word. As long as we confine our attention to just this
isolated problem, then neither of the two possible solutions forces
itself upon us as more reasonable than the other. (We must be careful
not to assume that [ph] and [p] are phonetically more similar simply
because the International Phonetic Alphabet represents them as two
kinds of [p]. This point has been stressed in the preceding section.)
In fact the linguist never considers such problems in isolation from
the rest of the analysis. If we take a general view of the phonology of
English, we shall see that the opposition of voiced and voiceless
consonants is one of greater importance in the language than that of
the opposition of unaspirated and aspirated consonants. Not only are
stops distinguished in certain positions by the presence or absence of
voice (bet: pet, etc.), where they are also distinguished by the absence
or presence, respectively, of aspiration; but other consonants (frica-
tives and affricates: leave: leaf, jeep: cheap, etc.) are distinguished by
the presence or absence of voice, and here there is no question of an
opposition between aspirated and unaspirated variants of a pair of
phonemes. This being so, it is doubtless preferable to consider the
lack of aspiration in [p] vis-d-vis [ph] as a matter of allophonic
variation, rather than its lack of voice vis-a-vis [b]. It should be
realized, however, that this decision has not, and could not have,
been made on the grounds simply of phonetic similarity.
3.3.4 Free variation in phonology

It may happen that two phonetically different units occur, but do not
contrast, in the same environment: that is to say, the substitution of
one for the other does not produce a different word, but merely a
different * pronunciation' of the same word. In this case the phoneti-
cally different units are said to be in free variation (cf. 2.3.2). For
3-3- PHONOLOGY 115
example, the 'glottal stop' (a sound produced by first bringing the
vocal cords together and then releasing them so that there is a sudden
escape of air: the IPA symbol for this sound is [?]) is a free variant of
/t/ at the end of a syllable before a consonant in what is called the
(standard) * Received Pronunciation* of English. Most speakers who
use this kind of pronunciation are probably not conscious of the fact
that they have two alternative pronunciations of such words as fort-
nighty etc., according to whether they are speaking more or less
formally and deliberately, whereas they are well aware of the occur-
rence of the same speech-sound as a 'substandard' variant of /t/
before vowels in, for example, the 'Cockney* pronunciation of words
like city. It may also happen that alternative pronunciations of a word
are current involving a difference that is generally phonemic in the
language. For example, the word economics is sometimes pronounced
with the vowel found in bet and sometimes with that of the word
beat; the word either is sometimes pronounced with the vowel of beat
and sometimes with that of bite. And there are many such alter-
natives, often in the speech of the same person, in English. However,
this would not normally be described as allophonic free variation.
The fact that the difference of the vowels serves to distinguish at least
some words in English means that it is always recognized as phonemic.
This is by virtue of the principle which is sometimes expressed in the
words 'once a phoneme, always a phoneme'. Moreover, it is not
possible to give an account of the sporadic fluctuation of phonemes
of the kind exemplified here other than by listing the words in which
it is to be found. From this point of view, the difference between the
two pronunciations of economics, either, etc., is, as it were, 'accidental':
it is not part of the regular phonological structure of the language. In
this respect it differs from the kind of free variation mentioned above
(between [t] and the glottal stop) where we were able to state con-
cisely, and in phonological terms (that is to say, without reference to
the particular words, but in terms of the categories of phonemes
preceding and following the free variants) the conditions under which
it takes place.
3.3.5 Neutralization in phonology

A more common phenomenon than free variation between phonemes
in phonologically definable conditions is the so-called neutralization
of the distinction between them in certain positions. This may be
explained by means of a commonly-used example. In many languages,
Il6 3. THE SOUNDS OF LANGUAGE
including German, Russian, Turkish, there is a phonemic distinc-
tion between voiced and voiceless consonants in most positions of the
word, but at the end of words voiced consonants do not occur. Thus,
both the German words Rad (' wheel', * bicycle') and Rat (' council',
* advice') are pronounced alike, namely as [rait]. (Many speakers of
German distinguish this pair of words, Rad and Rat, in terms of the
length or quality of the vowel. But the point being made is unaffected
by this fact.) The normal orthography takes account of the fact that
in the forms of words in which the same consonants occur in non-final
position the distinction between the voiced and voiceless member of
the pair is consistently made: cf. zum Rade [raids] verdammen ('to
condemn to torture on the wheel') and meinem Rate [raita] folgen
('follow my advice'). The distinction between voiced and voiceless
consonants, i.e. between /d/ and /t/, /b/ and /p/, etc., is said to be
'neutralized' in final position; and this is a phonological statement
about the language.
However, there are several different ways of treating this kind of
neutralization. Some linguists would say that it is the phoneme /t/
that occurs finally in both Rad and Rat and account for the change of
/d/ to /t/ manifest in the relation between Rade and Rad in a section
of linguistic description intermediate between grammar and phono-
logy (to which the name morphophonemics is given: this is the approach
followed in many standard American textbooks and published
analyses of languages). Another school would say that the fact that
certain phonological oppositions are 'neutralizable', whereas others
are not, in a given language constitutes such an important and basic
feature of the phonology of the language as to justify the recognition
of two different kinds of phonological unit. Linguists who take this
line of approach (principally the so-called ' Prague school') recognize
in addition to the phonemes, which preserve their distinctive function
in all positions, also what are termed archiphonemes restricted to the
positions of neutralization. A common and convenient way of
symbolizing archiphonemes is by the use of capital letters: thus /T/ is
an archiphoneme in German, as distinct from /d/ and /t/, which are
phonemes and do not occur in final position. A word like Tod
('death') would be transcribed /toD/; that is to say, in a manner
which makes it clear that the unit which occurs finally is of a different
order from that which occurs initially. Phonetically speaking, the
speech-sounds that occur at the beginning and end of this word are
the same (up to a certain degree of narrowness in transcription).
Whether or not they are phonologically the same is a question that
3.3. PHONOLOGY 117
can only be answered with reference to a particular theory of phono-
logy. The linguist who accepts the principle ' once a phoneme, always
a phoneme' will say that they are. The linguist who draws a distinc-
tion between phonemes and archiphonemes will say that they are not.
We shall look further into some of the theoretical differences under-
lying these different answers later in this chapter. Enough has been
said here to arouse in the reader the suspicion (which will become
stronger as we advance further in the subject) that the 'facts' which
a linguist discovers about the structure of a particular language are not
entirely independent of the theories with which he approaches it in
the first place.
3.3.6 Syntagmatic relations between phonemes

In our discussion of phonological theory so far, we have been
concerned solely with the paradigmatic dimensions of speech (cf.
2.3.3). We have said that /p/, /b/, /I/, etc., are different phonemes in
English because they are in paradigmatic contrast in various contexts.
(This notion of 'contrast' has already been treated, from a more
general point of view, in the previous chapter: cf. 2.2.9). I*1 t n e word
pet, for example, we may say that the phoneme /p/ occurs in the
initial position, where it is in contrast with /b/, /I/, etc. (cf. bet, let,
etc.); that the phoneme /e/ occurs in the second position, where it is in
contrast with /i/, /o/, etc. (cf. pit,pot, etc.); and that the phoneme /t/
occurs in final position, where it is in contrast with /k/, /n/, etc.
(cf. peck, pen, etc.). The word pet may, therefore, be represented as a
sequence of three phonemes: /p/ + /e/ + /t/, abbreviated to /pet/. (The
reasons why the phonemes are said to be sequentially-ordered have
been discussed above: cf. 2.3.6.)
For simplicity, let us assume that three, and only three, phonemes
are in contrast in the context /-et/, namely, /p/; /b/ and /I/: that
three, and only three, are in contrast in the context /p-t/: namely,
/e/, /i/ and /o/: and that three, and only three, are in contrast in
the context /pe-/; jtj, /k/ and /n/. On the basis of this assump-
tion, we can construct the following two-dimensional matrix, in
which the vertical columns represent the sets of phonemes in contrast
in each of the three positions:
l?l N N
N N N
N H M
Il8 3. THE SOUNDS OF LANGUAGE
The horizontal dimension represents the possibility of syntagmatic
combination (cf. 2.3.3). The matrix may be interpreted in the
following way: at any of the positions of paradigmatic contrast in the
word /pet/, which is given in the top row of the matrix, one may
substitute another of the phonemes from the second or third row, and
the result will be a different English word. The matrix summarizes,
therefore, the facts described in the previous paragraph. It tells us
that the following seven syntagmatic combinations of phonemes
constitute English wTords: /pet/, /bet/, /let/, /pit/, /pot/, /pen/ and
/pek/
-
Under this interpretation of the matrix, we have taken the word pet
as ' focal'; and we have held constant two of its constituent phonemes
as the context for paradigmatic substitution. But we may also
interpret the matrix in such a way that no particular word is taken as
' focal' in this sense. Many other English words can be accounted for,
if we allow any phoneme from the first column to combine syntag-
matically with any phoneme from the second and third columns:
cf. bin, bit, lick, lock, etc. (to put the words in their standard ortho-
graphic form). But this extension will also admit certain combinations
which do not constitute English words: e.g. /bik/ or /Ion/. At this
point, we must decide whether such combinations are systematically
excluded by virtue of some general restrictions upon the combination
of English phonemes with one another. If no such restrictions can be
established, we will say that forms like /bik/ and /Ion/ are phonologic-
ally acceptable ' words' of English, which have not been * actualized',
as it were, by the language and invested with a particular meaning and
a particular grammatical function. 'Non-actualization', in this sense
of the term, contributes to the redundancy of utterances (cf. 2.4.5).
The determination of the phonological structure of a language may
be thought of, then, in the following way. Every language has a
vocabulary of ' actualJ words, each of which in the first instance may
be taken as phonologically ' regular' (conforming to certain systematic
principles of combination operating upon the constituent phonemes).
The linguist's task is to account for their phonological 'regularity'.
(For simplicity of exposition, we are deliberately restricting the
discussion of phonological structure at this point by assuming that
phonology is concerned solely with the formation of words, inde-
pendently of their occurrence in sentences.) Given a set of 'actual'
words, the linguist will describe their phonological structure in terms
of rules which specify the permissible combinations of classes of
phonemes, each member in the class being in contrast with every
3.3. PHONOLOGY II9
other. For example, let us say that each column of phonemes in the
matrix given above constitutes a class, as follows: X = {/p/, /b/, /I/},
Y = {/e/> //> //}> ^ = {/t/, /n/, /k/}. We may now formulate a rule
which says that any member of X may combine with any member of
Y and Z (in that order). The rule, as we have just seen, will not only
account for the phonological regularity of such * actual' words as pet,
bet, lit, lick, peck, etc.; but it will also admit as regular a number of
non-occurrent * words'.
Let us now assume that we have established a whole set of rules of
this kind, each of which accounts for the ' regularity' of a class of
English words of different phonological structure, but each of which
also defines as phonologically regular many non-occurrent * words'.
By contrast with the 'actual* words of English, the non-occurrent
'words' also defined as regular by the same phonological rules may
be called 'potential'. In addition there will be many combinations of
phonemes (e.g. /pta / or /atp/: cf. 2.3.6) which are explicitly or
implicitly excluded by the phonological rules, and are defined as
irregular.
3.3.7 'Actual' and 'potential' phonological words

It might be thought that the ideal system of phonological rules would
be one which successfully defined as regular all the words actually
used by the speakers of the language, and no others. But quite apart
from the practical impossibility of formulating rules of this kind, it is
theoretically undesirable that one should attempt to eliminate the
distinction between the 'actual' and the 'potential'. We shall return
to this question in connexion with the principles of generative
grammar (cf. 4.2.2). Here it is sufficient to point out that any attempt
to identify the 'regular' with the 'actually occurrent' in phonology
would fail in two equally important respects.
The vocabulary of most languages will contain a number of words
which are phonologically 'irregular* in the sense that they do not
conform to the patterns of formation characteristic of the majority of
words in the language: they may have been borrowed from other
languages and may not have been ' assimilated' fully, or they might be
'onomatopoeic' (cf. 1.2.2). Although the linguist, as a matter of
principle, tends to assume at the outset that all the words used by the
speakers of a language are phonologically 'regular', he will be
prepared to revise this assumption in the case of particular words, if
he finds that he is unable to bring them within the scope of the rules
5-2
THE
he sets up for the majority of the words in the vocabulary. And he will
the more readily treat them as ' irregular*, when native speakers of the
language agree that there is something 'unusual* or 'alien* about
them.
But the identification of the 'actual* and the 'regular* also fails
in another respect. Many of the non-occurrent combinations of
phonemes would be accepted by native speakers as more 'normal*
than others; they are, not only easily pronounceable, but in some way
similar in form to other words of the language. The phonological
description of the language should reflect (although it should not
necessarily be determined by) feelings of this kind. It is noticeable,
for instance, and it has often been pointed out, that writers of
nonsense verse (like Lewis Carroll or Edward Lear) will create
'words* which almost invariably conform to the phonological
structure of actual words in the language; and the same is true of
brand-names invented for manufactured products. The ideal system
of phonological rules for a language will therefore be one which
correctly characterizes as regular, not only those combinations of
phonemes which constitute the majority of words in the vocabulary,
but also many others, wrhich native speakers of the language might
accept as 'possible*, or 'potential*, words. Needless to say, there
might be considerable dispute as to the ' correctness * of the charac-
terization. The point is that 'regularity* cannot be defined except in
terms of the rules which specify the permissible combinations of the
phonological units. And this point is valid at all levels of linguistic
description (cf. 4.2.13).
3.3.8 Distinctive features

So far we have assumed that the phonemes of a language are not
susceptible of further analysis: that they are minimal expression-
elements (to use the terminology of the previous chapter, cf. 2.2.5).
This assumption is still made in most of the standard textbooks of
linguistics, although it was challenged over thirty years ago by
Trubetzkoy, Jakobson and other members of the Prague school.
Their approach to phonology has recently been gaining ground, not
only in Europe, where it has always had many adherents, but also in
America.
According to Trubetzkoy and his followers, the phoneme is further
analysable into distinctive features. Consider, for example, the
following nine phonemes of English: /k/, /g/, /Q/, /t/, /d/, /n/, /p/,
3.3. PHONOLOGY 121
/b/, /m/. Typical instances of words in which they occur, in initial
and final position, are:
Table 5
M kill, lack M pet, cap N top, cat

III gull, peg M bid, hub /<"/ doll, kid
hi song H man, ham N net, pan
The gap in the row for /g/ is accounted for by the fact that this
phoneme does not occur in word-initial position. Under the inter-
pretation of phonemic theory outlined above each of these nine
phonemes is totally different from all the others. Phonetic considera-
tions are relevant to the question of allophonic variation (i.e. to the
grouping together of speech-sounds in complementary distribution as
contextually-determined variants of the same phoneme: cf. 3.3.3),
but not to the relations which hold between one phoneme and
another. To quote Hockett: ' . . .it must constantly be remembered
that a phoneme in a given language is defined only in terms of its
differences from the other phonemes in the same language'. The Prague
school phonologists would amend this statement by defining each
phoneme in terms of both its similarities and its differences with
respect to other phonemes in the same language.
In the case of the nine consonantal phonemes of English illustrated
above we can recognize seven features', or components, to which we
can give the following names (derived from a partial description of
their allophones in terms of articulatory phonetics): velar, labial,
dental (or alveolar); voiceless, voiced; oral, nasal. Of these, the
distinction between voiced and voiceless can be regarded as a distinc-
tion between the presence or absence of the feature of voice; and the
distinction between nasal and oral as a distinction between the
presence or absence of nasality. If we treat the velar, labial and
dental-alveolar components as independent features, each of which is
either present or absent, we can analyse the nine consonants in terms
of five two-valued variables. The variables are La (labial), Ve (velar),
De (dental or alveolar), Vo (voice), Na (nasality), and the two values
are positive (presence) and negative (absence). We will use the binary
digits 1 and o to indicate these positive and negative values (cf. 2.4.3).
Given these notational conventions, we can represent each of the
nine phonemes as a column of values in a two-dimensional matrix
THE
Table 6. Articulatory features of English stop consonants
N iii hi IPI N H N hi
La o 0 0 I 1 I 0 0 0
Ve I 1 I 0 0 O 0 0 O
Be o 0 O 0 0 O 1 1 I
Vo 0 1 I 0 1 I 0 1 I
Na o 0 I 0 0 I 0 0 I
Table 7. Distinctive feature*i of English stop consonants
M Izl hi IPI M H N / hi
La + + +
Ve + +
Be + 4. +
Vo + + +
Na + +
(cf. Table 6). For example, the column for /k/ reads (from top to
bottom) as 01000, the column for /g/ as 01010, and so on. It will be
observed that every phoneme is distinguished from every other by at
least one value of the five variable features.
We must now recognize a distinction between functional and non-
functional values. The phonological contrast between /k/ and /g/ is
maintained solely by the negative or positive value of the variable Vo;
so too, is the contrast between /p/ and /b/, and between /t/ and /d/.
The opposition of voiceless v. voiced is therefore a minimal functional
contrast within the English oral stop consonants: it is a distinctive
feature. On the other hand, /Q/ is opposed to both /k/ and /g/, /m/ to
both /pi and /b/, and /n/ to both /t/ and /d/, by the positive value of
Na (nasality); but the fact that the nasal consonants (/Q/, /m/ and /n/)
are also realized with voice can be regarded as irrelevant to the phono-
logical structure of English. Nasality presupposes, or determines, the
occurrence of voice in the phonetic realization of English words: there
are no words which are distinguished by the occurrence of a voiceless
nasal, rather than a voiced nasal, in the same position of paradigmatic
contrast. In this sense, voice is non-functional when it combines with
3.3. PHONOLOGY 123
nasality. Similarly, since there are no consonants which are both labial
and velar, both labial and dental, both dental and velar, etc., the
positive specification of La> Ve or De in any column of the matrix
determines the negative value of the other two. We may therefore
construct a new matrix in which * + ' will be used for * positive',
' ' for negative, and the absence of either * + * or ' ' for ' non-
functional* (cf. Table 7). Each of the nine phonemes we have been
discussing is now distinguished from the others in terms of either
two or three distinctive features.
The advantage of this approach is that it enables us to state more
systematically and more economically restrictions upon the distribu-
tion of particular classes of phonemes. For example, although there
are many words of English which have /sp/, /sk/ or /st/ in the first
two positions (e.g. spoty skip, step), there are none which begin with
/sb/, /sg/, or /sd/. Quite clearly, this is not an 'accidental* coincidence
in the combinatorial properties of /p/, /k/ and /t/, on the one hand,
and of /b/, /g/ and /d/, on the other. There are not six independent
'facts' to be described at this point, but simply one: 'in the context
/s-/ the distinction between the voiced and voiceless consonants is
non-functional.' And there are many other contexts in which we
would wish to say that a particular opposition between one set of
phonemes and another is non-functional.
3.3.9 'Grimm's law' reformulated in terms of distinctive features

Furthermore, in the diachronic description of languages, develop-
ments in the phonological system, which would otherwise be
accounted for in terms of' sound-laws' operating independently upon
particular phonemes, can often be more satisfactorily stated in terms
of distinctive features. In fact, this kind of formulation was implicit
in what was said about 'Grimm's law' in the first chapter (cf. 1.3.9).
It is made explicit in Fig. 4 below, in which + As stands for ' aspira-
tion' (as against As9 absence of aspiration) and +Fr stands for
'friction (partial obstruction, characteristic of fricatives, as against
total obstruction, characteristic of stop consonants: cf. 3.2.6). This is an
over-simplified representation of 'Grimm's law'; and it incorporates
certain assumptions about the reconstructed Indo-European con-
sonant system which we need not discuss here. But it will serve for
the purpose of illustration.
First of all, it will be noted that what may be regarded as the same
speech-sounds (e.g. [p], [g]) are described in terms of different
THE
distinctive features in the two systems labelled 'Indo-European* and
'Germanic'. For example, in the former system, [p] opposes [b] as
voiceless v. voiced, and it opposes [bh] as unaspirated v. aspirated. In
the 'Germanic' system, [p] opposes [f] as non-fricative v. fricative,
and it opposes [b] as voiceless v. voiced. The presence or absence
of aspiration is taken as distinctive for 'Indo-European' (but not for
Indo-European Germanic
/P. t. k/
Vo
- As + - Vo +
/b. d, g[ /bh, dh, gh/ /p, t, k/ /b, d, g/
Fig. 4. 'Grimm's law' in terms of distinctive features.
'Germanic'), and the presence or absence of friction is regarded as

distinctive in ' Germanic' (but not in 'Indo-European'). The features
labial, dental and velar are not indicated, since they remain constant
in value between the two stages (i.e. from the phonological point of
view, /f/ is regarded as a labial and [h] as a velar). In terms of this
analysis, 'Grimm's law' may be stated by means of three 'rules'
operating upon the distinctive features: (i) Vo-> +Fr, (ii) + Vo->
Vo, (iii) +As-> As. This does not mean that there were three
independent and chronologically-separated developments. For sim-
plicity, however, we may describe it as if the phonetic changes in the
realization of the phonemes took place in the following order, (i) In
certain positions, notably at the beginning of the word, /p/, /t/, /k/
come to be realized as [fj, [6] and [h] (or [x]). This means that they
are now distinguished phonetically from /b/, /d/ and /g/ both as
voiceless v. voiced and as fricative v. non-fricative. At this hypo-
thetical stage of the language (on the basis of the simplifying assump-
tions we are making here) 'voiceless and fricative' may be regarded as
the complex phonetic realization of the value of one phonologically-
distinctive feature, (ii) But, subsequently or concomitantly, /b/, /d/
and /g/ come to be realized, phonetically, without voice. As a con-
sequence the opposition between /p/ and /b/, etc., is realized phone-
3.3. PHONOLOGY 125
tically as [f] v. [p], etc. (iii) And, subsequently or concomitantly,
aspiration, which was formerly distinctive in /b/ v. /bh/, etc., becomes
non-functional (and is lost), when these pairs of consonants are
distinguished by the opposition of voice.
This summary of the diachronic development also illustrates the
important fact that the same speech-sounds may realize different
combinations of distinctive features in different languages. For
example, [p] has been assigned the values Vo and As in ' Indo-
European', and the values Fr and Vo in * Germanic*. (It may be
that, in each case, one of the two values is non-functional. We will
not discuss this question, since it would require a rather extensive
treatment of the * Indo-European* and 'Germanic* phonological
systems; and the evidence is not always easy to interpret.)
A further point can now be made in relation to this example. The
reader may have been puzzled by the apparently unmotivated shift
from a phonetic to a phonological description of 'Grimm's law* in
the course of the previous paragraph. As we saw in the first chapter
'Grimm's law* did not in fact operate independently of the phono-
logical context in which the consonants occurred. It was 'inhibited',
first of all by the accentual conditions described by Verner (cf. 1.3.11),
and also by the occurrence of a preceding /s/ in word-initial position.
Consequently, a certain number of the voiceless fricatives which
resulted from the phonetic change [p] -> [f], etc., were subsequently
voiced (and realized either as voiced stops or voiced fricatives
in either case, they were identical with the consonants which
resulted from [bh] -> [b], etc.); and the word-initial groups [sp], etc.,
did not develop into [sf], etc., but were phonetically unchanged.
It is for this reason that we say that the phonological system changed.
This point may be illustrated by means of the following schematic
representation of what is assumed to have taken place. The
'words' have been deliberately constructed in order to demonstrate
the principles involved without the introduction of extraneous
complications:
Stage 1 Stage 2
(1) [beda] -> [peta]
(2) [p*ba] -> [fepa]
(3) [speda] -> [speta]
(4) [dpar] -> [tefar]
(5) [depar]-> [tebar]
(6) [p<*bhar]-> [febar]
126 3-T H E
SOUNDS OF LANGUAGE
It will be observed that the place of the accent is given for stage i.
This is to take note of 'Verner's law', according to which in (4) the
[p] develops to [f], but in (5) it becomes [b]. As a result, whereas at
stage 1 (4) and (5) had the same consonant in medial position and
thus differed from (6), at stage 2 (5) and (6) share the same medial
consonant and together differ in this respect from (4). And, whereas
at stage 1 the consonant that occurs after [s] in (3) is phonetically
identical with that which occurs initially in (2), at stage 2 it is identical
with the initial consonant of (1). If therefore we describe each system
independently of the other, we will say that the distribution of the
phonemes is quite different. However, if the phonetic changes had
taken place without affecting the syntagmatic and paradigmatic
relations of any phoneme relative to any other, we would say that the
two systems, despite their phonetic realization, were phonologically
isomorphic (cf. 2.2.1).
3.3.10 'Neutralization* in relation to 'marked* and' unmarked* terms

In the discussion of the stop consonants in English, we said that the
distinction between + Vo and Vo was non-functional in the context
/s-/(cf. 3.3.8). It will be clear that this is a somewhat different sense of the
term 'non-functional* from the sense in which it was employed a little
earlier, when we said that the distinction between + Vo and Vo was
non-functional in combination with -f Na (cf. 3.3.8). In the one case,
the presence or absence of voice is rendered non-functional by the
simultaneous occurrence of a particular distinctive feature as a
component of the same phoneme; in the other, its redundancy is
determined by the context in which the phoneme in question occurs.
Contextually-determined redundancy of this kind is described as
neutralization: the opposition of voice is neutralized in the context /s-/.
As we have seen, the Prague school phonologists would say (in earlier
statements of the theory at least) that it is not the phoneme /p/ which
occurs in the word spot, but the archiphoneme /P/ (cf. 3.3.5).
The notion of neutralization is important (not only in phonology,
but also in grammar and semantics), because it correlates with the
distinction of unmarked and marked terms of an opposition (cf. 2.3.7).
In general, it is the unmarked term which occurs in a position of
neutralization. We tacitly took account of this fact in the assignment
of the values 'positive* and 'negative* in Table 7. (The assignment of
values in the discussion of the 'Indo-European* and 'Germanic*
systems is more problematical, and may be treated with some reserve.)
3.3. PHONOLOGY 127
The presence of voice, rather than the * presence' of voicelessness, was
taken as the ' positive' value of the feature in question because it is
the voiceless stop consonants which realize the archiphonemes in
positions of neutralization: e.g. /s-/.
3.3.11 Recent developments in distinctive-feature theory

Recent developments in distinctive-feature theory (associated mainly
with the names of Jakobson and Halle) have been characterized by
three main tendencies. The first is the attempt to establish a relatively
small set of distinctive features for the analysis of the phonological
structure of all languages. The second is the tendency to interpret the
phonetic correlates of the distinctive features (their realization in
substance: cf. 2.2.5), n o t m articulatory terms, as we have done here,
but in terms of acoustically-based categories (more precisely, perhaps,
in terms of categories which can be given an acoustic as well as an
articulatory interpretation). The third is the tendency to analyse all
contrasts in terms of one or more binary (or two-valued) features.
This might involve, for example, the recognition that /p/ and /k/
share some positive or negative value which distinguishes them from
/t/, instead of saying, as we have said above, that each of the three
differs from the other two by having a positive feature (La> Ve or De)
which the other two not only lack, but could not possibly have, under
the standard interpretation of such articulatory-phonetic terms as
'labial', Mental'and'velar'.
The three tendencies listed in the previous paragraph might be
summarized as: (i) universalist, (ii) acoustically-based, and (iii)
binarist. All three are highly controversial at the present time; and it
would be inappropriate to do more than mention them in an ele-
mentary book of this nature. For further discussion, reference may be
made to the more technical works cited in the notes.
3.3.12 Prosodic analysis

We must now refer briefly to yet a third approach to phonology. This
is generally referred to as prosodic analysis: it has been practised
mainly by Firth and his followers in London over the last twenty
years. It differs from other theories of phonology in that it gives no
place at all to the phoneme, whether as a minimal unit of analysis or as
a complex of distinctive features. Otherwise, as we shall see, it has a
certain affinity with Prague school phonology. We will use Turkish
THE
for the purpose of exemplification. This has the advantage that we
can simultaneously illustrate the phenomenon of vowel harmony,
which is found (in one form or another) in very many languages; and
it will make the Turkish examples used in the grammatical sections
of the book easier to follow.
3.3.13 Vowel-harmony in Turkish

The current orthography for standard literary Turkish recognizes
eight vowels: i,'i,u\ u, e, a, 0 and 0. (Here and elsewhere in this book,
we shall use i in place of the undotted form which is normally used in
Turkish orthographic practice.) These eight vowels can be described
in terms of the following three articulatory oppositions: high v. low,
back v. front, rounded v. unrounded (cf. 3.2.5). For simplicity we will
regard low as 'negative' with respect to high (Hi), front as the 'nega-
tive* of back (Ba), and unrounded as the 'negative* of rounded (Ro);
but this attribution of 'positive* and 'negative* values should not be
taken here as implying anything about the status of these terms as
marked or unmarked. Since each of the eight vowels is in contrast
with all the others in monosyllabic words, they may be regarded as
eight distinct phonemes, analysable into distinctive features as
indicated in Table 8.
Table 8. The vowel-phonemes of Turkish
/a/ M ll /o/
Hi + + + +
Ba - + - + 4.
Ro + + + +
However, it is a characteristic of Turkish that the principle of

'vowel-harmony* operates throughout the word, in the following
sense: (i) all the syllables of the word take the same 'value* for the
feature Ba\ and (ii) the feature Ro can take the value 'positive* only
in the first syllable of the word and in suffixes which have a high
vowel ( + Hi). There are a number of exceptions to this formulation
of the general principles. They fall into two main classes: (a) borrowed
words (e.g. salon, 'drawing-room*, which has +Ro in the second
syllable and Ro in the first, kitap, 'book*, which has Bain the
3.3. PHONOLOGY 129
first syllable and + Ba in the second), and (b) words containing * non-
harmonic' suffixes (e.g. sev-iyor> 'he loves', where the first vowel of
-iyor- is 'harmonic', in that it has Ba and Ro like sev-, but the
second is not: it has +Ba and +R6). Both classes of words are
phonologically 'irregular' in terms of the general structure of
Turkish (cf. 3.3.7). It is customarily said that the values of Ba and
Ro are determined by the value they have in the preceding syllable;
and there is certainly good reason to assert that vowel-harmony is
'progressive', rather than 'regressive', in Turkish (cf. 2.4.8, for this
sense of 'progressive' conditioning). Alternatively, however, we
might say that it is neither one nor the other. Instead we can define
Ba and Ro not as distinctive features of phonemes, but as prosodies
of the word. We can think of them as being 'switched' to a particular
value throughout the domain of their operation (in the case of Ba
and Ro this is the whole word, in the case of +Ro it is the first
syllable: and the occurrence of an irregular 'non-harmonic' vowel
'resets' the 'switch', as it were). By contrast, Hi can take the value
'positive' or 'negative' independently in all syllables. We will now
distinguish +Hi and Hi as phonematic units (n.b. not 'phonemes')
and symbolize them as ' I ' and 'A', respectively.
The effect of this distinction between prosodies and phonematic
units is to attribute to the Turkish word a two-dimensional phono-
logical structure. The sequence of phonematic units constitutes the
segmental 'infrastructure' of the word, whereas the prosodies form
its 'superstructure'. This can be represented notationally in various
ways. The system we will use here, for typographical simplicity, is
somewhat unconventional. We will put the phonematic units within
parentheses and prosodies outside the parentheses: thus +Ba, Ro
(klzlArlmlzdAn). The orthographic form of this word, hyphenated
to show the grammatical units out of which it is composed, is kiz-lar-
imiz-dan ('from our daughters'). It will be observed that the phonetic
realization of the vowel in each syllable of the word is determined in
part by the phonematic unit A or I which occurs there and in part by
the prosodies + Ba and Ro which extend over the whole word. The
main advantage of this two-dimensional phonological analysis of
Turkish is that it gives a more correct impression of the number of
phonological contrasts that are possible within the word.
If each of the eight vowels is regarded as being in contrast with all
the others, independently in all syllables of the word, this implies that
the number of possible vowel sequences in a Turkish word of n
syllables is 8 n . But the number of regular sequences is in fact 4 x zn;
I3O 3- T H E SOUNDS OF LANGUAGE
i.e. 2 x 2(2). This will be clear from the statement of the principles of
vowel-harmony given above; and it is reflected in the two-dimensional
prosodic analysis. There are four classes of words determined by the
two values of each of the prosodic * variables': (1) + Ba, + Ro;
(2) -Ba, + Ro; (3) + Ba, -Ro; and (4) -Ba, -Ro. And each of
these classes contains words of one or more syllables, which may be
extended, as the word kiz = -f Ba, Ro (klz) was extended above,
by the addition of further syllables. In principle, since there is the
possibility of either A or I in each syllable, for every value of n (where
n is the number of syllables in the resultant word), there are 2n
phonologically-regular sequences of phonematic units. There are
therefore 4X2 1 ( = 8 ) different 'sequences' of vowels in mono-
syllabic words; 4X2 2 (= 16) two-vowel sequences; 32 three-vowel
sequences; and so on. Once we take note of the difference in the
* domains' of contrast for phonematic units and prosodies the formula
2 x 2(2n) can be interpreted as an instance of the more general
formula N = px xp2xp3-..pm given in 2.3.8.
For the benefit of those readers who are not familiar with Turkish,
the following list of six words may be helpful for the interpretation of
the statements made above about vowel-harmony:
gozleriniz Ba, +Ro (gAz-lAr-In-Iz) 'eye-plural-your-
pluraV = 'your
eyes'
kollarimiz + Ba, + Ro (kAl-lAr-Im-Iz) ' arm-plural-my
plural9 = 'our
arms'
adamlarin + Ba, Ro (AdAm-lAr-In) ' man-plural-oV
= 'of (the)
men'
evlerinde Ba, Ro (Av-lAr-I(n)-dA) 'house-p/wra/-
his-in' = 'in
their house'
kolum +Ba9 +Ro (kAl-Im) 'arm-my' = 'my
arm'
goziimuz Ba, + Ro (gAz-Im-Iz) 'zyz-my-pluraV
= ' our eyes'
A prosodic representation of the words is given in the second column,
together with an indication of the way in which the words are com-
posed by the 'agglutination' (cf. 5.3.7) of smaller grammatical units.
It will be observed that the phonematic structure of these smaller
3-3- PHONOLOGY 131
grammatical units remains constant (lAr, Im, etc.). By virtue of their
occurrence in words of one prosodic class rather than another, they
are realized phonetically in different ways: e.g. [ler] v. [lar]; [im] v.
[im] v. [um] v. [iim] (the orthographic representation in the first
column is * broadly* phonetic: cf. 3.1.2).
3.3.14 *Multidimensional^ nature of prosodic analysis

One of the main differences, then, between phonemic analysis and
prosodic analysis is that the former represents words (and utterances
composed of words) as a unidimensional sequence of elements
(phonemes), the latter as a multidimensional structure composed of
prosodies and phonematic units. It is multidimensional, rather than
simply two-dimensional, because there may be several 'layers' of
prosodies operating over 'domains' of different lengths (pairs of
consonants, syllables, words, and even groups of words or whole
utterances): there is no space to illustrate all these possibilities here.
The phonematic units are like phonemes in the sense that they are
regarded as discrete units ordered serially, or sequentially, in the
unidimensional sequence of phonemes (and archiphonemes).
3.3.15 Difference between various 'schools' of phonology

There are many other important similarities and differences between
the various approaches to phonology that have been developed in
recent years. It should be emphasized, however, that they all rest
upon the prior acceptance of the notion of contrast, or opposition, and
the recognition that the paradigmatic and syntagmatic 'dimensions'
of language are interdependent. The differences between the various
'schools' of phonology can frequently be accounted for in terms of
the more particular assumptions which they accept, often for purely
methodological reasons.
3.3.16 Phonology and grammar

Of these assumptions only one need be mentioned here. For a number
of years, many linguists (including Bloch, Harris and Hockett, who
did so much to establish the principles and procedures characteristic
of the 'orthodox' American approach to phonemics) held that
phonological analysis must precede and be independent of grammati-
cal analysis. This assumption, or methodological principle, was never
THE
universally accepted (even in America). Nowadays, most, if not all,
linguists would agree that it imposes an unnecessarily strong and
undesirable restriction upon the theory of language. If anything, it
reverses the order of precedence, since the notion of contrast can
only apply to words which are capable of occurring in the same
context (cf. 2.4.10). The occurrence of words is determined partly by
their grammatical function (as nouns, verbs, etc.) and partly by their
meaning. It follows, therefore, that two words whose grammatical
function prohibits their occurrence in the same context cannot be in
phonological contrast. And it is frequently the case that a quite
different set of phonological contrasts is relevant for different
grammatical classes.
This is true of Turkish. It has already been mentioned that in
Turkish (as in Russian and German: cf. 3.3.5) the contrast of voiced v.
voiceless is neutralized for stop consonants in word-final position.
However, it is also neutralized for the majority of suffixes that are
added to words in the process of 'agglutination* illustrated above. For
example, the suffix meaning 'in* or 'at* (which was written as -dA-
in the prosodic representation of the word evlerinde) does not contain
a voiced dental stop, but a dental stop which is neutral with respect to
voice. Whether it is realized as [d] or [t] is determined by its occur-
rence in a voiced or voiceless environment. (This is a somewhat
inaccurate statement of the conditions, but it will suffice.) The point
is that the phonological structure of words may be partly determined
by their grammatical structure. Although it has been merely men-
tioned, rather than discussed or fully illustrated, in this section, it
will be taken for granted throughout the rest of this book.
3.3.17 Limited coverage of phonology in present treatment

Whole areas of phonology have been left untouched in the very brief
treatment of the subject that has been presented here. Nothing has
been said about stress and intonation in phrases and utterances;
nothing about the various functions that tone may serve in different
languages; and nothing about the status of the syllable as a phono-
logical unit. For an account of these and other topics, reference
should be made to the works cited in the bibliographical notes. In
the following chapters we shall be concerned with grammar and
semantics.
John Lyons
Online ISBN: 9781139165570
Chapter
4 - Grammar: General Principles pp. 133-169

4
GRAMMAR: GENERAL PRINCIPLES
4.1 Introductory
4.1.1 'Grammar'
The term 'grammar* goes back (through French and Latin) to a
Greek word which may be translated as ' the art of writing \ But quite
early in the history of Greek scholarship this word acquired a much
wider sense and came to embrace the whole study of language, so far
as this was undertaken by the Greeks and their successors. The
history of western linguistic theory until recent times is very largely
the history of what scholars at different times held to fall within the
scope of ' grammar* taken in this wider sense.
4.1.2 Inflexion and syntax

More recently the term 'grammar' has developed a narrower inter-
pretation. As used nowadays, it tends to be restricted to that part of
the analysis of language which was handled in classical grammar
under the headings of inflexion and syntax. The traditional distinc-
tion between inflexion and syntax, which rests on the acceptance of
the individual word as the fundamental unit of language may be
formulated as follows: inflexion treats of the internal structure of
words, and syntax accounts for the way in which words combine to
form sentences. For the present, we may continue to take the notions
of 'word* and 'sentence* for granted. Grammar, we will say, gives
rules for combining words to form sentences. It thus excludes, on
the one hand, the phonological description of words and sentences,
and, on the other, an account of the meaning that particular words
and sentences bear. This, it may be observed, is also the sense in
which the non-linguist usually intends the word 'grammatical*
to be understood when he says that such-and-such a combination
of words or the form of a particular word is 'grammatical* or 'un-
grammatical*.
[133]
134 4- GRAMMAR: GENERAL P R I N C I P L E S
4.1.3 * Notional' grammar

Modern grammatical theory is frequently said to be 'formal', in
contrast with traditional grammar, which was 'notional*. According
to Jespersen, a distinguished representative of an older school of
grammarians standing between the traditional and the modern
approach to grammatical analysis, 'notional* grammar starts from
the assumption that there exist ' extralingual categories which are
independent of the more or less accidental facts of existing languages'
and are 'universal in so far as they are applicable to all languages,
though rarely expressed in them in a clear and unmistakable way*.
Whether there are any universal 'categories' of grammar which
hold for all languages (the 'parts of speech', 'tense', 'mood', etc.)
in the sense in which Jespersen and traditional grammarians under-
stood the term is a question that we will discuss later (cf. 7.iff.). For
the present we shall assume that there are not (or, at least, that there
may not be); and in making this assumption we shall be adopting the
' formal' approach to grammatical analysis.
It should be noticed that this is a rather special use of the term
'formal' and has nothing to do with the distinction of 'form' and
'substance' which we made in an earlier chapter (cf. 2.2.2). It differs
too from the more general scientific use according to which a formal,
or formalized, theory is one in which theorems may be derived from
a set of basic terms and axioms by the application of explicit rules of
inference. As we shall see later, recent developments in linguistics
have gone a considerable way towards formalizing grammatical theory
in this more general sense of the term 'formal'. For the moment, the
term 'formal' may be taken to imply the rejection of 'notional'
assumptions: in other words, formal grammar makes no assumptions
about the universality of such categories as the 'parts of speech' (as
they were traditionally defined) and claims to describe the structure
of every language on its own terms. A more positive indication of what
linguists have had in mind when they have used the term ' formal' in
opposition to ' notional' will be given in the following section.
4.1.4 Semantic considerations in grammar

The 'formal' approach to grammatical description has frequently
been understood to carry the further implication that semantic
considerations are irrelevant both in the determination of the units of
grammatical analysis and in the establishment of rules for their
4-1. INTRODUCTORY I35
permissible combination in sentences of the language. The proposal
to make the theory and practice of grammar independent of questions
of meaning has been responsible for the most interesting and most
fertile developments in modern grammatical theory. But it has often
been mistakenly assumed that linguists who refuse to admit considera-
tions of meaning in grammar take up this position because they have
no interest in semantics. This is not so. Nor is it because they believe
that semantic analysis is necessarily more subjective than phono-
logical or grammatical analysis. The reason is simply that the gram-
matical structure of a language and its semantic structure tend to be
highly, but not totally, congruent with one another. As soon as the
linguist becomes seriously interested in semantics, he must see that
nothing but advantage can come from the methodological separation
of semantics and grammar. As long as it is maintained that every
identity or difference of grammatical structure must be matched with
some corresponding identity or difference of meaning (however subtle
and difficult to determine) there is a danger that either the gram-
matical description or the semantic, or both, will be distorted.
Furthermore, it should be noted at this point that the methodo-
logical separation of grammar and semantics refers only to the way
in which the description of the language is presented. It does not
mean that the linguist will deliberately refuse to take advantage of his
knowledge of the meaning of sentences when he is investigating their
grammatical structure. We shall see later that the fact that a particular
sentence is ambiguous (has two or more different meanings) is
often an indication that it should be assigned two or more different
grammatical analyses (cf. 6.1.3). But we shall also see that ambi-
guity of itself is insufficient to justify a difference in the grammatical
analysis.
4.1.5 The term 'formal*

It is unfortunate that the term * formal* (like many other terms
employed by linguists) is used in so many different senses in the
literature of the subject. It would be easy enough to avoid such terms.
However, anyone who becomes interested in linguistics must sooner
or later learn to adjust his interpretation of words like 'formal'
according to the context in which he meets them. This adjustment
becomes easier when one knows something about the historical
development of the subject. For that reason, we shall frequently give
a brief account of the etymology or historical development of the
more important technical terms when the notions they are associated
with come up for discussion. As far as the word 'formal' is concerned,
it should be remembered that it is commonly used in the following
senses: (i) with reference to the phonological and grammatical
structure of language, in contrast with the semantic (by virtue of the
traditional distinction between the 'form' of a word and its
'meaning': cf. 2.1.2); (ii) with reference to the phonological, gram-
matical and semantic structure of language as distinct from the
' medium' in which language is realized or the conceptual or physical
continuum ' structured' by the lexical elements of language (in this
sense it is opposed to 'substantial': by virtue of the Saussurian
distinction of 'substance' and 'form', cf. 2.2.2); (iii) as equivalent to
'formalized' or 'explicit', in contrast with 'informal' or 'intuitive';
(iv) in opposition to 'notional', in the sense in which this latter term
has been illustrated above with the quotation from Jespersen. Of these
four senses, (i) and (iv) are not always distinguished in the literature:
when linguists talk about 'notional' grammar, they tend to assume
that any universal categories postulated by the grammarian will
necessarily be based on meaning. (Whether this is a valid assumption
or not we will not discuss at this point: we merely draw attention to
the fact that it is often taken for granted that 'notional' implies
'semantically-based'.) It is sense (ii), which is in direct conflict with
sense (i), that tends to cause the most confusion.
For the sake of completeness, one should perhaps mention that in
the very recent literature a somewhat different opposition has been
introduced by Chomsky, in his discussion of the universal properties
of language. This is a distinction between 'formal' and 'substantive'
universals (n.b. 'substantive', not 'substantial'). Roughly, this rests
upon a distinction between the nature of the rules used by the linguist
in the description of languages and the elements (linguistic units or
classes of units) to which reference is made in the rules. If one holds,
for example, that there is a fixed set of distinctive features, a particular
selection of which is combined in various ways in the phonological
systems of different languages (cf. 3.3.11), one may say that these
distinctive features constitute the substantive universals of phono-
logical theory. By contrast, any condition imposed upon the manner
in which the phonological rules operate or the way in which the
phonological units combine with one another according to the
specification of the rules is a formal universal of phonological theory.
The postulate of unidimensionality, for instance, could be regarded
as a formal universal of 'orthodox' phonemic theory (cf. 3.3.14).
4-1. INTRODUCTORY 137
Chomsky's use of * formal' is related to sense (iii) of the previous
paragraph. The universal categories of traditional grammar (in
particular the ' parts of speech') cannot be described as either * formal'
or 'substantive', in any very strict application of this distinction,
since the rules of traditional grammar were not explicitly formalized.
However, it is probably true to say, with Chomsky, that they were
defined primarily in 'substantive' terms. It should be noticed, how-
ever, that the traditional definitions of the 'parts of speech' frequently
included some specification of their combinatorial properties. We
shall return to this question in a later chapter (cf. 7.6.1 ff.).
These terminological remarks are intended to assist the reader who
may have read, or who may go on to read, other works on linguistic
theory. In this book, the term 'formal' will be employed only in
senses (iv) and (iii), that is to say, in opposition to 'notional', on the
one hand, and to 'informal', on the other: the transition from one
sense to the other will be made in the course of the present chapter.
When the noun 'form' is used in a technical sense, it will be clearly
distinguished by its juxtaposition with either 'meaning' or 'sub-
stance ': cf. (i) and (ii) above.
4.2 Formal grammar

4.2.1 ' Acceptability'
We must now try to characterize modern grammatical theory a little
more positively than we have done so far.
We may begin by invoking the notion of 'acceptability'. 'Accept-
able' is a primitive, or pre-scientific, term, which is neutral with
respect to a number of different distinctions that will be made later,
including the distinction that is drawn traditionally between 'gram-
matical' and 'meaningful' (or 'significant'). It is a more primitive
term than either 'grammatical' or 'meaningful' in the sense that,
unlike these terms, it does not depend upon any technical definitions
or theoretical concepts of linguistics. An acceptable utterance is one
that has been, or might be, produced by a native speaker in some
appropriate context and is, or would be, accepted by other native
speakers as belonging to the language in question. It is part of the
linguist's task, though not the whole of it, to specify as simply as
possible for the language he is describing what sentences are accept-
able, and to do this in terms of some general theory of language-
structure.
In the description of a modern spoken language the linguist will

usually have available a collection of recorded utterances (his 'data',
or * corpus') and he will also be able to consult native speakers of the
language (his * informants'). He may of course be his own informant
if he is describing his own language; but in this case he must be on
his guard against the danger of producing for description a corpus of
material which includes only such sentences as satisfy his precon-
ceived ideas about the structure of the language. As the description
proceeds the linguist can obtain further utterances of various kinds
from his informants, and so extend the corpus; and he can check with
them the acceptability of sentences which he himself constructs in
order to test the generality of his tentative rules. If he finds that his
informants will not accept as a natural or normal sentence some
utterance which satisfies the rules of acceptability which he has so far
established, then he must, if possible, revise the rules so that they
exclude the 'sentence* in question, whilst still allowing all the
acceptable sentences for which they were set up in the first place.
In the case of the so-called ' dead* languages, like Latin, it is of course
impossible to verify one's rules by checking with native speakers the
acceptability of all the sentences accounted for by the rules. For this
reason a description of one of the classical languages will inevitably
be incomplete in certain respects. However, the adequacy of the
description will be proportionate to the amount and variety of the
material upon which it is based.
At first sight it might appear that the term 'acceptable', as it has
been employed here, is redundant and introduces unnecessary compli-
cations. It might be thought that to say that a given utterance is
acceptable is to say no more than that it has at some time been
produced by a native speaker and that it would be possible, in
principle, for a linguist, or team of linguists, to collect all the sentences
of a language and put them in the corpus. But this view is erroneous.
The term 'acceptable' not only has the advantage of stressing the
operational connexion between the linguist's 'raw material' and its
ultimate source of control in the reactions of native speakers. It also
emphasizes the fact that the linguist must account not only for the
utterances which have actually occurred in the past, but also for very
many others which might equally well have occurred and might occur
in the future. This was generally taken for granted and occasionally
made explicit by traditional grammarians. More recently, in their
desire to avoid the pitfalls of prescriptive grammar (cf. 1.4.3), many
linguists have declared that their descriptions of a particular corpus
4*2. FORMAL GRAMMAR 139
of material are valid only for the sentences actually occurring in the
corpus and carry no implications as to what other sentences might be
produced by native speakers of the language in question. But this
attitude, however well-intentioned, is, as we shall see, both theoreti-
cally and practically untenable.
4.2.2 The sentences of a language may be unlimited in number

Every native speaker of a language is able to produce and understand,
not merely those sentences which he has at some time heard before,
but also an indefinitely large number of new sentences which he has
never heard from other speakers of the language. It seems probable,
in fact, that most of the sentences produced by native speakers, apart
from a limited set of * ritual' utterances (like How do you do?y Thank
you!, etc.), are 'new* sentences in this sense. And the 'new' sentences
will satisfy the same operational test of acceptability to other native
speakers as 'old' sentences, which might have been produced simply
from memory. They will exhibit the same regularities and can be
accounted for by the same rules. In other words, it is the class of
potential utterances which we must identify as the sentences of the
language. And the number of potential utterances in any natural
language is unlimited. Any given collection of utterances, however
large, is but a 'sample' of this unlimited set of potential utterances.
If the sample is not only large, but representative of the totality of
potential utterances, it will, ex hypothesi, manifest all the regularities
of formation characteristic of the language as a whole. (The distinction
made here between the 'sample' and the language as a whole is
essentially the distinction drawn by de Saussure between 'parole' and
'langue': cf. 1.4.7.) It is the linguist's task therefore in describing a
language to establish rules capable of accounting for the indefinitely
large set of potential utterances which constitute the language. Any
linguistic description which has this capacity of describing actual
utterances as members of a larger class of potential utterances, is said
to be generative (for further discussion of this term, cf. 4.2.13). We
shall see that if rules are established to account for the acceptability
of any representative sample of utterances, the same rules will
necessarily account for a much larger set of utterances not in the
original corpus, unless the application of the rules is very severely,
and 'unnaturally', restricted. Moreover, if certain rules with parti-
cular properties are incorporated into the description, it will be
capable of accounting for an infinite, but specified, set of acceptable
utterances. In this respect therefore a generative description will
reflect, and indeed can be thought of as * explaining', the native
speaker's ability to produce and understand an indefinitely large set
of potential utterances.
4.2.3 * Layers' of acceptability

The question which now confronts us is this: How much of accept-
ability, or what kind of acceptability, falls within the scope of the
grammar, and how much is to be accounted for by other parts of the
linguistic description or by disciplines outside linguistics? It seems
clear that utterances can be acceptable or unacceptable in various
ways or in various degrees. We might say of a foreigner's English, for
example, that it is * grammatically' acceptable (or correct), but that his
'accent1 is faulty and marks him immediately as a non-native speaker
of the language. We might say of certain sentences (as Russell did,
for instance, of the sentence Quadruplicity drinks procrastination) that
they are 'grammatical', but 'meaningless'; we might wish to say the
same of the nonsense verse of Lewis Carroll, but for somewhat
different reasons. Then there are other kinds of acceptability and
unacceptability, which have nothing to do with whether an utterance
is meaningful or not. Fairy-tales and science fiction provide many
instances of sentences which would be unacceptable in 'everyday*
English. Again, some utterances, though meaningful, would be
regarded by certain people or in certain circumstances as 'blasphe-
mous' or 'obscene': this we may refer to, loosely, as 'social accept-
ability'. In what follows we shall neglect 'social acceptability'.
4.2.4 ' Idealization' of the data

When we say that two people speak the same language we are of
necessity abstracting from all sorts of differences in their speech.
These differences, reflecting differences of age, sex, membership of
different social groups, educational background, cultural interests, and
so on, are important and, in principle at least, are to be accounted for
by the linguist. However, in the speech of any persons who are said
to 'speak the same language' there will be what may be described as
a 'common core'a considerable overlap in the words they use, the
manner in which they combine them in sentences and the meaning
which they attach to the words and sentences. The possibility of
communication depends upon the existence of this 'common core'.
4-2. FORMAL GRAMMAR 141
For simplicity of exposition, we shall assume that the language we are
describing is uniform (by 'uniform' is meant 'dialectally and
stylistically' undifferentiated): this is, of course, an lidealization' of
the facts (cf. 1.4.5) a n c * t n a t all native speakers will agree whether an
utterance is acceptable or not. We shall also assume that the 'odd'
sentences which one might find in children's stories or science fiction
are describable in terms of an extension of the description which
accounts primarily for more 'normal' utterances. We shall concen-
trate therefore upon the distinction of phonologically acceptable and
grammatically acceptable, on the one hand, and of grammatically
acceptable (grammatical) and semantically acceptable (meaningful)
on the other.
4.2.5 Phonological and grammatical acceptability

In the preceding section we saw that every language has its own
phonological structure, describable in terms of a set of units (pho-
nemes or phonological units of diverse kinds according to the nature
of the language and the model of analysis adopted by the linguist) and
a statement of their possibilities5 of combination. A certain amount of
unacceptability (including much of what is popularly referred to as
'accent') can be accounted for at the phonological, or even the
phonetic, level. If we were to produce a large set of sequences of
phonological units, each sequence being constructed in accordance
with the rules of combination established by phonological analysis of
the language in question, we should find that only an infinitesimal
fraction of the resulting 'utterances' were acceptable to a native
speaker. This point is readily illustrated by means of an English
example. For simplicity we may assume that the orthography of
English exactly reflects the phonological structure of the language.
This assumption, though of course false, does not affect the validity
of the point we are illustrating.
Let us take as our example the following English sentence:
iwantapintofmilk. (It is printed without spaces to indicate that for the
moment we are considering it simply as a sequence of letters, each of
which is assumed to represent one phonological unit; there is of
course no pause between the words in the normal pronunciation of the
corresponding spoken utterance.) Making use of our knowledge of the
permissible sequences of English letters, we might try the effect of
substituting single letters and groups of letters at different places in
our model sentence. We know, for instance, that s may follow i and
142 4- GRAMMAR: GENERAL PRINCIPLES
precede a (cf. isangasongofsixpence); but *isantapintofmilk is not an
acceptable sentence. (Throughout this and the following sections we
shall use an asterisk before a word or sentence to denote unaccept-
ability. This is now standard practice in linguistics. In historical and
comparative studies the asterisk generally denotes an unattested, or
'reconstructed', word or phonological unit (cf. 1.3.13).) Nor is
*iwantapinkofmilt, or *ipindawantopfilk, although none of these
'utterances' can be excluded by virtue of the phonological structure
of English. Even if we were to work out over a large body of text, not
only the permissible pairs, triples, quadruples, etc., of letters, but also
the probabilities of occurrence of particular letters relative to their
neighbours (such calculations have been carried out for English and
for certain other languages, and the results have been used in crypto-
graphy and in the design of communication channels for the trans-
mission of written messages), this would not help us very much in our
attempt to construct other acceptable sentences of English by substi-
tuting single letters or groups of letters in the model sentence. The
fact is that certain' blocks' of letters within the language form' higher-
level* units at the boundaries of which the probabilities of occurrence
of single letters relative to one another are of little significance in
determining which 'blocks' can combine with one another to form
acceptable utterances.
More relevant than the general probability of occurrence of w
after i and before a is the fact that w forms part of the 'block* want.
From one point of view this can be considered as a permissible sequence
of letters in English; from another point of view it must be considered
as a unit, for which other 'blocks' may be substituted to produce
further acceptable utterances: idrinkapintofmilk, itakeapintofmilk, etc.
And i is also a 'block': the fact that it is composed of only one
letter is irrelevant. No other single letter may be substituted for it in
the present environment to produce a different acceptable utterance,
but only such other 'blocks' as we, they, thejoneses, etc. What we have
been calling 'blocks' of letters we may now recognize, provisionally at
least, as words of the language. (To simplify the argument, we will
here neglect the fact that some 'blocks' which we might substitute
for / in the present example are not single words, but combinations
of words, like The Joneses, They all, etc. At this point the reader
may wish to refer again to the section dealing with the * double
articulation' of the expression-plane: cf. 2.1.3.)
4.2.6 Distributional approach to grammatical description

It will be observed that we have been brought to the traditional view
that English sentences are * structured * on two levels, that of letters
(or phonological units) and that of words, without explicitly invoking
the notion of meaning. What we have done is to distinguish a phono-
logical (or orthographic) component of acceptability from the rest of
acceptability and to leave the rest (or Residue') of acceptability to be
analysed into other 'higher-level' components. It is of course true
that the utterances / want a pint of milk, I drink a pint of milk, etc.
(we may now insert the spaces between the words) are meaningful
when used in an appropriate context. Moreover, they are different in
meaning; and their difference of meaning is properly described as a
function of the meanings of the constituent words, want, drink, etc.
But so far we have not taken account of these facts; and we shall not
do so within the theory of grammar that is being presented here. They
will be handled by the theory of semantics. All that we have done in
principle is to construct a set of acceptable sentences by placing
different words in the same 'frame', or context. The whole set of
contexts in which a linguistic unit may occur is its distribution (cf.
2.3.1). What is being developed here, therefore, is the distributional
approach to grammatical analysis.
We have now advanced to the point at which we can say that
phonology accounts for the acceptability and unacceptability of
utterances, so far as this can be done, by means of rules or formulae
which specify permissible combinations of the phonological units of
the language under description and that grammar 'takes over', as it
were, at the 'higher* level, accounting for acceptability in terms of
permissible combinations of words.
4.2.7 Interdependence of phonology and grammar

At this stage in the argument, it should be pointed out that we are
still working with the assumption that sentences are composed of
words and that words are composed of phonological units (or letters,
if we are dealing with the written language). Both of these assump-
tions will be modified later. (The argument would be made somewhat
more complicated, but it would not be invalidated, if we were to
introduce the modifications here.) It should also be mentioned that,
although we have talked as though grammatical description will
necessarily come after (but be independent of) phonological analysis,
144 4* GRAMMAR: GENERAL PRINCIPLES
thisneednot be the case(cf. 3.3.16). Grammar and phonology are estab-
lished as theoretically distinct levels of linguistic structure. But the
phonological and the grammatical structures of particular languages
are usually interdependent (in different ways and in different degrees).
It will be the linguist's task in describing a given language to account
for this interdependence, where it exists (as it will also be his task to
account for such interdependence as may hold between the gram-
matical and the semantic structure- of the language).
4.2.8 A simple example of distributional analysis

We are still far from having arrived at a satisfactory account of
grammatical description. For we have not yet said kow the acceptable
combinations of words are to be specified. Listing all the acceptable
sequences of particular words is out of the question, since, as we have
seen, no natural language can be regarded as a limited set of sentences.
In the case of a closed corpus of material it would be possible of course
to compile a list of all the phonologically (or orthographically)
different sentences and then to decide the question of acceptability by
reference to the list. But this would not only be unrevealing, in the
sense that it would not contribute to our understanding of the native
speaker's ability to produce 'new' sentences. It would not even be the
most economical way of describing the given text. If we pursue the
matter a little further, we shall see why.
In a reasonably large and representative sample of the sentences of
a language there will be a considerable overlap in the distributions of
different words. For example, not only could beer, water, gin, etc., be
substituted for milk in I drink a pint of milk, but the same set of words
could occur in many of the other environments in which milk can
occur. Likewise they, we and you can be substituted for /, and buy
take, order, etc., for drink in many other environments than the one
in the model sentence. Words that can generally be substituted for
one another in many different sentences can therefore be grouped into
distributional classes by virtue of this fact.
Let us assume that we have a corpus of material for analysis
consisting of the following seventeen i sentences': ab, ar, pr, qab, dpb9
aca, pea, pep, qar, daca, qaca, dacp, dacqa, dacdp, qpeda, acqp, acdp.
Each different letter represents a different word: the symbolic
notation is used for reasons of generality, and also to make clear that
at this point we are not directly invoking semantic considerations.
(Since acceptability is assumed to be preserved by the operation of
4-2 FORMAL GRAMMAR 145
substitution and since 'acceptability* includes 'meaningfulness', it
may appear that we are guilty of equivocation here. We are working
towards, but still have not reached, the distinction of ' grammatical *
and 'meaningful'.) It will be seen that a and^> have certain environ-
ments in common (cf. -r, pc-9 dac-)9 and so do b and r (cf. a-, qa-)y
and d and q (cf. dac-a> -aca9 ac-p)y but that c has a unique distribution
(a-a9p-a9p-p9 qa-a9 da-a, da-p, etc.) in the sense that no other 'word*
occurs in any of the environments in which c occurs. Let us now put
a and p into a class X, and substitute the class-label X at each place
where either a or p occurs (the sentences which differ only in that
where one has a the other has p being reduced formulaically to one
class of sentences): Xb9 Xr (ar9 pr)9 qXb9 dXb9 XcX (aca9 pca9 pcp)9
qXry qXcX9 dXcX (daca9 dacp)9 dxcqX, dXcdX, qXcdX, XcqX9
XcdX. Let us now group b and r into a distributional class Y, and
d and q into a class Z. Substituting Y for b and r, and Z for d and q9
we get: (i) XY9 (Xb9 Xr); (2) ZXY (qXb9 qXr9 dXb); (3) XcX;
(4) ZXcX (qXcX9 dXcX); (5) ZXcZX (dXcqX9 dXcdX9 qXcdX);
(6) XcZX (XcqX9 XcdX). We can thus account for the sentences of
our corpus in terms of six structural formulae specifying the accept-
able sequences of classes of words (c being a one-member class).
These formulae are linear (in a sense to be explained later: cf. 6.1.1).
For the present we may be content with the description of the
sentences of our corpus in terms of their linear structurethat is to
say, in terms of the following formulae, or rules:
(i)XY
(2) ZXY
(3) XcX
(4) ZXcX
(5) ZXcZX
(6) XcZX
Each of these rules may be thought of as describing a distinct
sentence-type. (The fact that it is possible to reduce these sentence-
types to subtypes by invoking the principles of constituent-structure
is irrelevant at the present stage of the exposition: cf. 6.1.2 ff.) It will
be observed that this system of rules satisfactorily accounts for the
acceptability of the seventeen sentences of the corpus (it defines them
as grammatical). But it does so only by including the occurrent
sentences as members of a total set of forty-eight sentences. (The
number 48 is obtained by applying the formula given in 2.3.8 to
each of the six sentence-types and summing the totals. There are
2 x 2 = 4 sentences of type (1), 2 x 2 x 2 = 8 sentences of type (2),

2 x 1 x 2 = 4 sentences of type (3), 16 sentences of type (4), 8 of type
(5) and 8 of type (6): 4 + 8 + 4+16 + 8 + 8 = 48.) The language
described by this grammar therefore contains exactly 48 sentences.
The thirty-one non-occurrent sentences must either be allowed as
acceptable or excluded, if for any reason it is decided that they must
be excluded, by means of additional rules prohibiting certain combi-
nations of particular words. The additional rules would obviously
much complicate the * grammar'. In this sense, therefore, the most
economical way of describing the given text is one which represents
it as a random sample of seventeen sentences from the total set of
forty-eight sentences which constitute the language. And the
' grammar* we have set up to describe the text is generative in the
sense explained above (cf. 4.2.2). We will say that it generates, or
characterizes, the language of the text, assigning to each of the
sentences that occur in the 'sample' (as well as to those which do not
occur in the * sample') a particular structural description: pr is a
sentence of structure XYy pcda of structure XcZX, etc.
In the case of this restricted artificial 'language', used for the
purpose of exemplification, only seven distinct words occur, and there
are in the corpus only seventeen simple sentences out of a total of
forty-eight which the grammar generates. With a natural language the
situation is of course far more complex. The number of individual
words will run into many thousands; their distributional classification
will not be so straightforward and certainly cannot be carried out in
the manner illustrated. Moreover, there will be a number of different
types of sentences to be accounted for, including sentences of con-
siderable complexity. However, these facts do not affect the principle.
The words of natural languages can be grouped into distributional
classes (and, in practice, as we shall see, always have been by gram-
marians); and the distributional classes established for particular
positions in one type of sentence are generally valid for particular
positions in other, and more complex, types of sentences. Grammar,
as it is understood here, is nothing other than the description of the
sentences of a language in terms of the combination of words (and
phrases, etc.) by virtue of their membership of distributional classes.
It is a kind of algebra in which the 'variables' are word-classes and
the 'constants', or the 'values' taken by the variables in particular
sentences, are individual words.
In order to see that this yields, in principle, the kind of description
which would normally be called 'grammatical', we need only inter-
pret the example with reference to English. Let a = men, p = women,
b = live, r = die, c = love, d = old, q young. In other words, let
the class X include all those words which are usually referred to as
'plural nouns'; let Y be the class of 'intransitive verbs'; c, the class
of' transitive verbs'; and Z, the class of' adjectives \ Our statement of
the permissible combinations of word-classes implies that sentences
such as Men die, Old men love young women, etc., which would be
described in traditional grammar as instances of a simple * subject ' -
'predicate* construction, are grammatically acceptable, while *Die
men or *Old love young men women, etc., are not.
4.2.9 Grammatical classes

In traditional grammatical theory the ' parts of speech' (nouns, verbs,
adjectives, etc.) were defined, ostensibly, in 'notional' terms ('a noun
is the name of any person, place or thing', etc.). But, as we shall see
in a later chapter (cf. 7.6.1 ff.), traditional grammarians have tended
to confuse two different questions in their discussion of the ' parts of
speech'. The first, which is the question which concerns us here, is
that of establishing the conditions under which a certain word may be
said to belong to a particular grammatical class: 'Is the word men a
member of class X or class Y?' In practice, this was always deter-
mined in terms of the distribution of the wordits potentiality of
occurrence in sentences relative to the occurrence of other words in
the same sentences. In this respect, modern linguistics has merely
given recognition, within the theory of grammar, to the distributional
principle by which traditional grammarians were always guided in
practice. The theoretical acceptance of this principle distinguishes
'formal' from 'notional* grammar. The second question has to do
with the naming of the grammatical classes (once their membership
has been established on ' formal' grounds): ' Is X appropriately called
the class of "nouns"?' From the point of view of 'formal* grammar,
any label is as good as any other; and the traditional terms, 'noun',
'verb', 'adjective', etc., are neither more nor less satisfactory for the
purpose than any other terms would be.
We shall see later that it may be possible to define such terms
as 'noun* or 'verb* in a way that both respects their traditional
'notional' interpretation and also makes them applicable in the
'formal' analysis of different languages. But this question will not be
pursued at this point. For the present we will assume that the terms
' noun',' verb',' adjective', etc., and other traditional terms of grammar,
have no ' notional' (or universal) application, but denote distribution-

ally justifiable grammatical classes to which any arbitrary labels
might have been given.
We are now in a position to discuss, in a preliminary way, the
distinction between 'grammatical' and 'meaningful'. (We shall,
however, return to this question at various places later in the book.)
For simplicity, we shall restrict our attention at this point to one class
of English sentences (all of which will be described here in a deliber-
ately simplified manner). The class in question may be illustrated by
the following examples, assumed to be acceptable:
(1) The dog bites the man
(2) The chimpanzee eats the banana
(3) The wind opens the door
(4) The linguist recognizes the fact
(5) The meaning determines the structure
(6) The woman undresses the child
(7) The wind frightens the child
(8) The child drinks the milk
(9) The dog sees the meat
In a traditional grammatical description of English, all these sentences
would be defined as simple sentences of subject-predicate structure.
Furthermore, it would be asserted that (in all the instances of the
class of sentences we are at present concerned with) the subject is a
phrase (i.e. a unit made up of mpre than one word) composed of an
Article (here the, 'the definite article') and a Noun (e.g. dog)y and (as
is normally the case in English) it precedes the predicate; and that the
predicate is a phrase composed of a (transitive) Verb (e.g. bites)
followed by the object of the verb, the object being a phrase composed
of an Article and a Noun (e.g. man). Various other statements would
also be made about the particular class of sentences illustrated above:
that the subject and the verb agree in 'number' (i.e. if the subject-
noun is singular, the verb is also singular, but if the subject-noun is
plural, then the verb is plural: thus The dog bites v. The dogs bite);
that the verb is 'inflected' for 'tense' (i.e. bites is 'present', bit is the
corresponding 'past tense' of the same verb, and so on). All these
various points can be more or less satisfactorily interpreted within the
framework of modern 'formal' grammar (cf. especially chapters 7
and 8). For the moment, however, we shall work solely with the three
word-classes articles, nouns and verbs. The article may be regarded
as a one-member class, containing the word the; it will be abbreviated
here as T. The class of nouns and the class of verbs, abbreviated as
N and V, contain several thousands of words, including those which
occur in the ' actual' sentences listed above:
N = {dog, man, chimpanzee, banana, wind, door, linguist, fact,
meaning, structure, child, milk, meat, . . . }
V = {bites, eats, opens, recognizes, determines, undresses,
frightens, drinks, sees, . . . }
Given this tentative distributional classification of particular words of
English (and it will be observed that we are, by implication, restricting
ourselves to the consideration of sentences in which only isingular*
nouns and 'present-tense* verbs occur), we may propose the following
grammatical rule (which assumes a purely linear structure: cf. 6.1.1):
S i: T+N+V+T+N
The symbol ' 2 ' ('sigma') stands for 'sentence': the Greek letter is
employed rather than the Roman ' S' (practice varies in this respect in
current work in linguistics) in order to emphasize the difference of
4
status' in grammatical theory between sentences and the units out of
which sentences are constructed, e.g. words (cf. chapter 5). The sub-
script numeral attached to S indicates that the rule accounts for only
one class of sentences. The choice of the numeral is purely arbitrary.
The rule may be read as follows: * Any combination of words which
results from the substitution of one member of the appropriate word-
class, chosen at random from the word-lists in the lexicon of the
language, in place of the symbols Ty N and V at each position of
the linear formula T+N+ V+ T+ N is a sentence of type 1.' The
grammatical rule presupposes, therefore, not only a lexicon (or
dictionary) in which all the words of the language are given the
appropriate grammatical classification as N, V or T, but also one or
more rules of lexical substitution for the replacement of the word-class
symbols with words. The existence of such rules we may take for
granted at this point: they will be discussed in the following section.
Given the grammatical rule proposed above and the associated lists
specifying the members of the grammatical classes, all the sentences
in our 'sample* are defined to be grammatical, with the structural
description T+N+V+T+N.
We have said that N and V might have several thousand members.
To simplify the arithmetical calculation and yet obtain some reason-
able indication of the number of English sentences that would be
generated by the single rule given above, let us assume that each of the
6 LIT
two classes, N and V, contains exactly one thousand (io3) members.
On this assumption, the proposed rule generates no more and no less
than I X I O 3 X I O 3 X I X I O 3 = io 9 (one thousand million) sentences,
each with the same structural description. This is but one rule, and it
accounts for a very simple class of very short English sentences.
It is therefore a very comprehensive rule. It is comprehensive in the
sense that it undoubtedly generates an enormous number of accept-
able sentences. But it is perhaps too comprehensive, since it also
generates (and defines to be grammatical) very many sentences which
would fail to pass the test of acceptability in normal circumstances of
use. (The condition 'in normal circumstances', however indetermi-
nate it might be in application, cannot be omitted. For example, many
'normally* unacceptable 'sentences' are deliberately introduced in
the context of linguistic discussion, and in similar 'abnormal*
circumstances.) Since all the sentences generated by the proposed
rule are thereby defined as grammatical, we must either amend the rule
to exclude some of the sentences which we consider to be unaccept-
able or account for their unacceptability, if it can be accounted for in
the total description of the language, in terms of the incompatibility
of the meanings of particular subclasses of words (or in some other
way). These two alternatives are not, in fact, mutually exclusive, as
we shall see. But let us first consider the implications of the first
alternative within the framework of ' formal' grammar.
4.2.10 Subclassification
One obvious way of amending the proposed grammatical analysis is to
subdivide the classes N and V and formulate, not one new rule, but a
whole set of different rules. Let us therefore reclassify the vocabulary
as follows:
Na = {dog, man> chimpanzee, linguisty child, wind, . . . }
Nb = {banana, door, milk, meat, . . . }
Nc = {fact, meaning, structure, . . . }
Vd = {eats, bites, frightens, undresses, sees, . . . }
Ve = {recognizes, determines, sees, eats, . . . }
Vf = {determines, . . . }
Before we proceed, a number of points should be stressed in con-
nexion with this reclassification. First, the way in which we have
arrived at the particular decisions incorporated in it is, in principle,
irrelevant. It is not being suggested here that linguistic theory can, or
should, yield a set of procedures for the determination, or * discovery',
of the distributional classes referred to in the grammatical rules. What
matters is whether one classification rather than another enables the
grammarian to formulate a set of rules which will include the
maximum number of acceptable sentences and the minimum number
of unacceptable sentences among the total set of sentences which the
grammar generates. (There are additional considerations, which we
will discuss later with reference to the distinction of 'strong' and
'weak' adequacy: cf. 6.5.7. They may be disregarded at this point.)
The second point to notice is that the new subclasses, despite the use
of subscripts, are now assumed to be totally unrelated to one another.
In other words, Na, Nb and iVc are subclasses of N only in the ' acci-
dental' sense that we have gone from an earlier tentative grammatical
classification to a later tentative classification. In principle, what we
have done is to construct a completely new classification of the vocabu-
lary and a completely new grammar for the language we are describing.
As we shall see in the following section, it is possible to revise the
notion of distributional classification with which we are at present
operating in such a way that Naf Nb> Nc can be regarded as subclasses
of the wider class N, and Vd> Ve and Vf as subclasses of V. Finally, it
should be observed that we have introduced a certain amount of
multiple membership: determines occurs in both Ve and Vf, and sees
in both Vd and Ve. Apart from anything that might be said against the
double classification of these words, this has the undesirable con-
sequence that within the present framework sees in a sentence like
The child sees the banana is, from the grammatical point of view, a quite
different element from sees in such sentences as The child sees the
meaning. We shall return to this problem also in the next section.
Given the new classification of the words we are concerned with, we
will substitute for the previous rule a whole set of rules (each one, it
should be noticed, defining a completely different sentence-type):
(a) S x : T+Na + Vd+ T+Na (cf. The dog bites the man)
(b) 2 2 : T+Na + Vd+T+Nb (cf. The chimpanzee eats the
banana)
(c) S 3 : T+Na + Ve + T+Nc (cf. The linguist recognizes the fact)
(d) S 4 : T+Nc + Vf+T+Ne (cf. The meaning determines the
structure)
Although these four rules suffice to generate the nine sentences of the
'sample' (and very many others), it will be evident that further rules
6-2
are now required to account for other sentences composed of the
words used in the * sample' which we might wish to regard as
acceptable (The banana frightens the linguist, etc.). The reader is
invited to construct some additional rules and also to extend the lists
of words given above.
The most important point that arises in connexion with the revision
of the grammatical rules is this: the distinction between the gram-
matical and the ungrammatical sentences of English has now been
redefined. Such combinations of words as *The banana bites the
meanings *The structure drinks the chimpanzee, etc., which we will
assume to be unacceptable as utterances of English, are now defined to
be ungrammatical. On the other hand, there are many other unaccept-
able utterances which would be allowed by the rules as grammatical:
*The chimpanzee drinks the door, *The dog undresses the wind, etc.
In principle, we might hope to make the classification of words and
the system of rules progressively more detailed, making continual
adjustments until it becomes capable of generating the maximum
number of acceptable sentences and the minimum number of un-
acceptable sentences. With each successive modificationand this is
the theoretical import of the illustrationthe limits of grammaticality
are redefined for the language being described. From the 'formal'
(v. 'notional') point of view, grammaticality is nothing more than
acceptability to the extent that this can be brought within the scope
of a particular set of rules and a particular classification of the lexical
and grammatical elements in the language. (For the present, the
distinction between' lexical' and' grammatical elements' that has been
slipped in at this point may be disregarded: ' lexical and grammatical
elements' may be interpreted as 'words'.)
4.2.11 Indeterminacy of grammar

In describing a given language, the linguist will draw the limits of
grammaticality at a particular point. His decision to draw these limits
at one place rather than another, if the decision is made consciously
after weighing the various alternatives that present themselves, will
tend to be determined by two main factors. The first may be referred
to as the principle of 'diminishing returns'. It is possible to go a lot
further with the distributional subclassification of words than would
have been thought feasible, or even desirable, by traditional gram-
marians. But sooner or later, in his attempt to exclude the definitely
unacceptable sentences by means of the distributional subclassifi-
cation of their component words, the linguist will be faced with a
situation in which he is establishing more and more rules, each
covering very few sentences; and he will be setting up so many over-
lapping word-classes that all semblance of generality is lost. This is
what is meant by the principle of 'diminishing returns': there comes
a point (and where this point is might be legitimate matter for dispute)
at which the increase in the complexity of the rules is too * costly' in
proportion to its * yield', a relatively small increase in the coverage of
acceptable and unacceptable sentences. But the second factor is no less
important. Since the sentences of the language being described are so
numerous (and, as we shall see later, for both practical and theoretical
reasons we may wish to say that they are infinite in number), one
cannot hope to decide for every sentence generated by the grammar
that it is definitely acceptable or unacceptable. In fact, one does not
have to go very far with the grammatical description of any language
before one finds disagreement among native speakers about the
acceptability of sentences generated by the rules tentatively estab-
lished by the grammarian. There is therefore a real, and perhaps
ineradicable, problem of indeterminacy with respect to acceptability
and unacceptability.
It would seem to follow from these considerations that the gram-
matical structure of any language is in the last resort indeterminate.
It is not only that linguists will differ in their interpretation of what
constitutes the optimum degree of generality in the scope of the rules
and in their evaluation of the acceptability of particular sets of
sentences. There is the additional problem that the generation of one
set of sentences of a particular type may make the generation of other
sentences of a different type extremely difficult to handle within the
theoretical framework established for the first set. Some instances of
the problems which arise will be given in a later chapter (cf. 8.3.6 ff.).
It may be that they will be lessened, and even eliminated, by advances
in grammatical theory that are now being made: but at the present
time they seem to be rather intractable. We may therefore restate as a
general principle which governs all grammatical description (and it
will be unaffected by anything that is said in subsequent chapters
about the nature and manner of operation of grammatical rules) the
following fact: whether a certain combination of words is or is not
grammatical is a question that can only be answered by reference to a
particular system of rules which either generates it (and thus defines
it to be grammatical) or fails to generate it (and thereby defines it to
be ungrammatical).
Most writers on grammatical theory, including those who have
made major contributions to the development of transformational
grammar (in particular, Chomsky), would seem to reject this principle.
They suggest that the grammatical structure of any language is
determinate and is known 'intuitively* (or 'tacitly') by native
speakers. This appears to be an unnecessarily strong assumption. It
is undoubtedly the case that native speakers will agree that certain
sets of utterances 'belong together', or are 'similar* or 'different* in
some way. These' intuitions', in so far as they are ascertainable, are an
important part of the linguist's data; and he will try to account for
them by distinguishing various kinds of acceptability (or well-formed-
ness) and various kinds of relatedness between sentences. But he need
not assume that there will be any very direct correspondence between
the 'intuitions' of the speakers and the statements made by the
linguist.
One should not exaggerate the difference of opinion between
linguists on this question. To assert that the grammatical structure of
language is in the last resort indeterminate is not the same as to assert
that no part of the grammatical structure is determinate. There are
many combinations of words (e.g. *They likes shey *The dog bite the
man, etc.) which all linguists will characterize immediately, not only
as unacceptable, but also as 'ungrammatical' (without necessarily
producing a set of grammatical rules). One can say that their immedi-
ate reaction is based on an 'intuitive* awareness of the grammatical
structure of standard English; one can equally well say that the
combinations in question infringe principles of such generality in
utterances of standard English that any grammar would necessarily
have to take account of them. It is with respect to the less general
principles that alternative grammars might differ in their charac-
terization of sentences as grammatical and ungrammatical. And the
'intuitions' of linguists and speakers of the language tend to be un-
reliable and inconsistent at this point anyway.
4.2.12 ' Grammatical' and ' meaningful9

As we saw earlier in this section, it seems reasonable to distinguish
different kinds of acceptability. The traditional distinction between
'grammatical' and 'meaningful* (not to mention the other kinds of
acceptability referred to earlier) is not affected by the conclusion we
have just reached: that the limits of grammar are ultimately indeter-
minate; or rather, that each grammar defines its own limits and, in
that sense, makes them determinate. When we say that a certain
utterance is ungrammatical (by reference to a given grammar), we are
not implying that it is not also unacceptable for other reasons. Of some
unacceptable combinations of words we will say that they are gram-
matical, but meaningless; of others that, although they are both
grammatical and meaningful, they would not normally occur, because
the occasion for saying what they 'express* could hardly arise. Of yet
another class of combinations of words we might be inclined to say
that they are both ungrammatical and meaningless. At one end of the
' continuum' of acceptability there are to be found certain combina-
tions of words whose acceptability or unacceptability is accounted for
only by the grammar (cf. He gives, They give v. *He give, *They
gives). At the other end of the continuum the grammatical description
is of no avail. But there will be many utterances in the description of
whose acceptability or unacceptability both a grammatical account and
an explanation in terms of the meaning of the component words will
be valid: and it is this fact which gives to the 'notional' definitions of
certain grammatical classes the considerable degree of validity that
they have in certain languages ('verbs of motion', 'masculine noun',
'locative phrase', etc.). We will return to this question of the cor-
relation between the grammatical and the semantic classification of
words in the next section.
4.2.13 The term 'generative*

But first we must say something more about the terms 'generate' and
'generative', since they have often been misunderstood. The first
point to be stressed is the negative one: a generative grammar is not
necessarily a transformational grammar (cf. 6.6.1). The terms
'generative' and 'transformational' are frequently confused, because
they were introduced into linguistics at the same time by Chomsky.
(The term 'transformation' was also used by Harris in roughly the
same sense as it was used by Chomsky.) By Chomsky and his followers
the term 'generative' is usually understood to combine two distin-
guishable senses: (i) 'projective' (or 'predictive'); and (ii) 'explicit'
('formal' v. 'informal'). It has been employed in both of these senses
in the course of this section.
It was first introduced in the sense of' projective' (or ' predictive'):
to refer to any set of grammatical rules which, explicitly or implicitly,
described a given corpus of sentences by 'projecting' them upon, or
treating them as a ' sample' of, a larger set of sentences. A grammar of
this kind is ' predictive' in that it establishes as grammatical, not only

' actual' sentences, but also 'potential* sentences. It is important to
realize that most of the grammars that have ever been written through-
out the history of linguistics are generative in this first sense of the
term. There would be no reason to stress this point, were it not for
the fact that the distinction between prediction and the prescription
of normative standards of ' correct usage' is not always appreciated
(cf. 1.4.3).
But the term 'generative* was subsequently used in this section in
a rather particular sense of 'explicit* (cf. 'generate', 'characterize',
4.2.8). This approximates to, and indeed derives from, one of the
senses in which the term ' generate' is employed in mathematics. Let
us consider, in a non-technical way, a statement such as the following:
'The number 2 generates the set, or series, of numbers 2, 4, 8, 16,
32, ....' This set of numbers, which is infinite, can be ordered into
a series, as illustrated in the statement, of 'ascending powers of 2',
the number 2 being the base of the series. Every number generated by
the base satisfies the function 2 n (where 2 is the base and n is a
variable ranging over the natural numbers, 1 , 2 , 3 , 4 , . . . ) . ^ Y
number of whatever magnitude either belongs to the set generated
by 2 in the sense defined or it does not; whether it belongs to the set
('the powers of 2') is decidable. The decision-procedure, for any given
number, might take the following form: generate the series in question
up to or beyond the magnitude of the particular number, testing for
identity with each of the numbers generated by the base. In this way
we might decide, for instance, that 9 is not in the set, since it does not
equal 2, 4, 8 or 16 (the first number in the series greater than 9).
Furthermore, in order to emphasize the grammatical analogy, one
might say that each number in the set is generated with a particular
'structural description': the 'structural description' of 64, for
instance, is '2 n , where n = 6 \
When we say that a grammar generates the sentences of a language
we imply that it constitutes a system of rules (with an associated
lexicon) which are formulated in such a way that they yield, in
principle, a decision-procedure for any combination of the elements
of the language (let us call them 'words' at this juncture) in more or
less the above sense. Furthermore, the grammar not only 'decides'
whether a given combination is grammatical or not (by generating or
failing to generate a combination of symbols which can be tested for
'identity' with the utterance in question); but it provides for each
grammatical combination at least one structural description. (We shall
4.2. FORMAL GRAMMAR 157
see later that sentences with more than one structural description are
defined to be grammatically ambiguous: cf. 6.1.3.)
This second, more or less mathematical, sense of the term' generate'
presupposes, for its applicability to grammar, a rigorous and precise
specification of the nature of the grammatical rules and their manner
of operation: it presupposes the formalization of grammatical theory.
In the course of this section, we have therefore moved from
'formal* v. 'notional' to 'formal* v. 'informal' in our interpretation
of the term 'formal grammar'. This transition reflects the historical
development of grammatical theory over the last ten or fifteen
years.
4.2.14 'Distribution* and 'discovery procedures'

The principles of distributional analysis (which we have been taking
to be definitive for 'formal' v. 'notional' grammar) were most
thoroughly and extensively discussed by Harris, especially in Methods
in Structural Linguistics (published in 1951, but written a few years
earlier). Harris and other American linguists at the time developed
these principles within the framework of 'procedural' linguistics: a
set of assumptions about the nature of linguistic theory and methodo-
logy which was not universally accepted then and is still less generally
accepted now. In particular, it was assumed that the proper task of
'structural linguistics' was to formulate a technique, or procedure,
which could be applied to a corpus of attested utterances and, with
the minimum use of the informant's judgements of 'sameness' and
'difference', could be guaranteed to derive the rules of the grammar
from the corpus itself. For this reason, the term 'distribution', and
the term 'structuralist', has, for purely historical reasons, come to be
associated with the view that it is possible to formulate 'discovery
procedures' for the establishment of the rules of particular grammars
on the basis of attested utterances. It should be clear that the term
'distribution' is used in this book without any of these adventitious
implications. It should also be realized that the work of Harris and
his colleagues, with its strong tendency towards rigorous formulation
of distributional principles, served as the foundation upon which
generative grammar has been built, since the publication of Chomsky's
Syntactic Structures in 1957.
158 4* GRAMMAR: GENERAL P R I N C I P L E S
4.3 Grammar and lexicon
4.3.1 Analysis and synthesis

As we saw in the previous section, every grammar presupposes a
lexicon (or dictionary) in which the words of the language are classified
according to their membership of the distributional classes referred to
in the grammatical rules.
Both the grammar and the lexicon can be looked upon from two
different points of view, according to whether the linguist is concerned
with the analysis ('recognition') of a corpus of utterances or the
synthesis ('production') of grammatical sentences. Although practical
convenience may dictate that the lexicon and the grammar be
organized in a somewhat different way, according to whether they are
being employed for 'recognition* or 'production', it is important to
realize that they are themselves neutral with respect to this distinction.
Any corpus of attested utterances can only be described satisfactorily
as a' sample' of the sentences the grammar generates (cf. 4.2.8). There is
therefore no opposition between generative and' descriptive' grammar.
However, the fact that grammar is neutral, in principle, with
respect to analysis and synthesis does not mean that the adoption
of one point of view rather than the other will not have any practical
consequences. If the grammar is to be used for synthesis, it will be
convenient to have the lexicon organized in such a way that, given a
particular word-class symbol (e.g. N9 'noun'), one can readily find
the members of the class in question for the operation of the lexical-
substitution rules (cf. 4.3.2). One obvious way of doing this is to
organize the lexicon as a set of lists, each of the following form
N = {man, boy, chimpanzee, . . . }
On the other hand, if we are engaged upon the analysis of a given
text, it will be easier to work with a master list in which the words are
ordered according to some principle (e,g. alphabetically) which
enables us to find quickly any individual words occurring in the
sentences being analysed and discover their grammatical properties,
e.g.
beauty: Noun
die: Nouny Verb
warm: Adjective, Verb
Our conventional dictionaries are of this second type. Traditional
grammarians did not usually compile lists of the first type except for
4-3 GRAMMAR AND LEXICON 159
'irregular' forms (which they might put in the grammar, as well as
listing them alphabetically in the dictionary). They assumed the
possibility of constructing lists of words for the purpose of 'synthesis'
on the basis of the meaning of particular words and the definitions of
the 'parts of speech*. Some recent generative grammars, or gram-
matical sketches, provide partial lists for each word-class (and we shall
discuss the nature of these lists below). The difference between the
two kinds of lexicon is not one of principle, but simply one of
convenience of reference. Modern generative grammars of the kind
referred to have been more concerned to establish the grammatical
classes required in the description of the language they are dealing
with than exhaustively to classify all the words in these languages. If
all the words of the language are not classified appropriately in the
lexicon, the grammar will not be generative in the sense referred to
as 'explicit' (cf. 4.2.13).
One consequence of the adoption of the point of view of analysis
rather than synthesis may be mentioned here. If the linguist knows
that his description of a particular language is going to be used only
for the analysis of recorded material (this is the case, for example, in
some of the projects which have as their aim the automatic analysis of
written texts for the purpose of machine-translation or library-
classification and the retrieval of information), he can afford to make
a less exhaustive classification of the lexicon and a less complete
grammatical description of the language.
For example, there are very many English nouns which end in ness
(e.g. goodness, correctness, etc.). Most of these, like the two instances
just cited, can be 'derived' from adjectives (e.g. good, correct, etc.).
Without going into the nature of 'derivation' at this point (cf. 5.4.2),
we can say one word, a noun, may be derived from another, an
adjective, by means of the following formula: Ax + ness = Ny. (This
may be read as follows: 'Any word composed of a member of word-
class Ax and ness is a member of word-class Ny.') Since this is a very
productive derivational rule of English, we may assume that it is
included in the grammar; and all the words ending in ness which can
be derived by means of the formula may be removed from the lexicon.
Now, if we are concerned with synthesis, we must decide which
adjectives belong to the class Ax: whether, for instance, this class
includes true and strong, so that trueness and strongness (in addition to,
or rather than, truth and strength) would be generated as grammatical
or excluded as ungrammatical. But a 'recognition' grammar need not
wait upon this decision. Then it would be quite reasonable to work
i6o 4. GRAMMAR: GENERAL P R I N C I P L E S
with the more general rule A + ness = Nv. ('Any word which occurs
in a sentence at a position in which Ny is permissible, and which can
be analysed into A and ness is to be accepted by the recognition pro-
gramme.') If trueness and strongness were to turn up in the texts being
analysed they would be analysed and accepted as grammatical; if not,
the question is irrelevant. (The words truth and strength would be listed
in the lexicon or analysed in terms of other rules. The example that
has been used here comes from an actual computer programme which
successfully analysed very many English derivational formations.)
There is no difference of principle associated with the difference of
viewpoint. Whether it is used for analysis or synthesis, the formula
A + ness = Ny generates the same set of words (assuming that it
makes reference to the same list of adjectives). But, if one is concerned
solely with analysis, one can afford to set one's sights lower. One can
deliberately generate (in the abstract, mathematical sense of ' gener-
ate'and this is the sense in which the term must always be inter-
preted) a set of sentences which includes many that one would
normally wish to exclude, on the assumption that they will not occur
anyway. To exclude sentences which one assumes will not occur
would add considerably to the 'cost' (cf. 4.2.11). This principle of
'cost-effectiveness' has frequently been applied in automatic
language-analysis by computer, since the principle of 'diminishing
returns' has a very direct economic interpretation (in terms of the
additional time required for programming, extra running time for
the computer, etc.).
But one must forestall a possible misunderstanding in connexion
with analysis and synthesis. The fact that the grammar is neutral
between the two does not imply that analysis is simply the converse of
synthesis (or vice versa). It should not be supposed that a computer pro-
gramme, for instance, might work 'downwards' through a set of rules
(and from the grammar to the lexicon) in the' production' of sentences
and 'upwards' through the same set of rules (and from the lexicon to
the grammar) in the 'recognition' of a given corpus of material.
Both 'production' and 'recognition', whether by speakers and
hearers of a language or by a computer programme designed to simu-
late their 'behaviour' in the use of language, would seem to involve
'feedback' from one process to the other (cf. 3.2.9). Little progress
has yet been made in the investigation of this problem from a psycho-
logical point of view; and a certain amount of 'psycholinguistic'
research has been vitiated by a failure to realize that 'generative' does
not mean 'productive'. Hence this cautionary paragraph.
4.3. GRAMMAR AND LEXICON l6l
4.3.2 Lexical substitution rules

We may now return to our discussion of the structure of the lexicon.
For simplicity, we shall continue to operate with our very simple
notion of grammatical rules. Although this will be revised in sub-
sequent chapters, the revision will have no effect upon the general
points made here. Let us therefore return to the first rule used in the
previous section:
S i: T+N+V+T+N
and the word-classes it presupposes
T = {the}
N = {mariy dog, chimpanzee, . . . }
V = {bites, eats, opens, . . . }
The process of lexical substitution (the insertion of particular words
at the positions established by the grammatical rule: cf. 4.2.9) can
be described as follows: For each occurrence of a grammatical-class
symbol in the structural description of a sentence put any member of
the class referred to, taking this member from the class listed in the
lexicon. (When all the grammatical-class symbols have been replaced,
or * rewritten', by means of the repeated application of this principle
of lexical substitution, the * output* of the generative system is a
sentence with a specified structural description.) The operation of
lexical substitution can be formalized by means of the following rule
X->x|xeX
' Rewrite X, where X is a variable ranging over (" taking as its values ")
all the grammatical classes referred to in the generative system
(e.g. T, N or V), as x, where x is any member of the class X.' The
repeated application of this lexical-substitution rule would convert
T+N+ V+ T+ N into a sentence like The dog bites the man, which is
represented in the tree-diagram in Fig. 5. (It will be observed that the
distinction between the part of the * output' which comes from the
grammatical rule and the part which comes from the lexicon is
represented in the distinction between solid and broken lines in the
tree-diagram. This is a useful convention, which we will adhere to
throughout the present work.)
Since the lexical-substitution rule operates in the same way
independently of the 'value' taken by X, we can look upon the lists
of words as a set of rules appended to the grammar and thus dispense
162 4. GRAMMAR: GENERAL P R I N C I P L E S
with the generalized lexical-substitution rule. In taking this view

(which we shall presently revise) we shall be following the practice
of Chomsky and the earliest writers on generative grammar. Let us
therefore organize the lexicon in the following form:
T->{the}
N -> {many dog, chimpanzee, . . . }
V-> {bites, eats, opens, . . . }
The arrow may be interpreted as an instruction to replace, or
* rewrite', the element occurring to the left of the arrow with one of
the elements listed to the right of the arrow: e.g. fy -> z* would mean
'rewrite^ as z (under the conditions governing the system of rules)'.
Any system of rules, each of which is cast in this form (the manner of
their operation will be discussed in greater detail in chapter 6) will
be referred to as a rewrite-system (or system of rewriting rules).
We will now use the rewriting arrow for the grammatical rule also,
and integrate the grammatical rule and the lexical-substitution rules
in the one system:
(1) X-+T+N+V+T+N
(2) T^{the)
(3) N-> {man, dog, chimpanzee, . . . }
(4) V -* {bites, eats, opens, . . . }
This is a very simple generative grammar, which we will now extend to
accommodate a certain amount of subclassification of the word-classes.
It was pointed out in the previous section that, within the frame-
work of assumptions with which we were working in our discussion
of formal grammar, any reclassification of the words in the lexicon
yielded, not subclasses of the original wider classes, but entirely
unrelated new classes. From the grammatical point of view, this
inadequacy may be remedied by the inclusion of additional rules, as
follows
N-*{Na,Nb,Nc}
V-+{Vd,Ve,V,}
We must now amend the lexical-substitution rules (increasing their
number). Thus, the new grammar-and-lexicon takes the following
form:
(1) X-+T+N+V+T+N
(2) N->{Na,NbyNc}
4-3- GRAMMAR AND LEXICON 163
(3) V->{Vd,Ve,Vf}
(4) Na -> {man, dog, chimpanzee, . . . }
(5) Nh -> {banana, door, milk, . . . }
(6) iVc -> {fact, meaning, structure, . . . }
(7) Va -> {eats, bites, frightens, . . . }
(8) Ve -> {recognizes, . . . }
(9) P/ "^ {determines, . . . }
S
This system of rules formalizes within the grammar the fact that Na,
Nb and iVc are subclasses of AT (the members of the subclasses are
'nouns') and Vd, Ve and Vf are subclasses of V (the members are
'verbs'). There is consequently an additional 'layer' of grammatical
structure introduced by the system for the purpose of formalizing
this fact (cf. Fig. 6, which represents the additional 'layer' by means
of the branches of the 'tree' connecting N and Na and V and Vd).
However, it does so by allowing as grammatical the very combinations
of subclasses that the process of subclassification was designed to
prohibit. (There is no restriction in the above set of rules to prevent
the selection of, say, Nb in the second position, of Vd in the third
position, and of Nc in the fifth position.)
What is involved here is the principle of syntagmatic conditioning,
or compatibility, between one subclass of words and another: what is
generally referred to as lexical selection. At this point, we are con-
cerned with the structure of the lexicon, and we will not go into the
grammatical aspects of this. Let us simply assume that an appropriate
formalization is available (making use of 'context-sensitive* rules:
cf. 6.5.1) which will enable us to preserve the notion of subclassifi-
cation, expressed in rules (2) and (3), and yet to generate the desired
combinations of subclasses.
4.3.3 Grammatical features

We may now envisage a more and more detailed subclassification of
the vocabulary of the language (to the point of * diminishing returns').
The grammar might be extended, for example, to include such
further rules as
Na^{NaVNJ
Nb->{Nbl9Nb2}
etc.
Each successive subclassification of this kind implies an increase in
the number of lexical-substitution rules at the end of the grammar.
Moreover, it will be evident that this formalization is based upon a
very particular (and, as we shall see, false) assumption about the
grammatical structure of language. The rules divide the vocabulary
into hierarchically-ordered classes and subclasses (cf. Fig. 7), such
that iV all and Nal2 are totally included in Nal, Nal is totally included
in Na, and Na is totally included in N; and so on. This assumption
was made in the earliest generative grammars which adopted the
rewrite-system of formalization (introduced into linguistics by
Chomsky).
It is unsatisfactory in two respects. First, it leads to a large number
of separate lists of words in the lexicon, with a commensurately high
degree of multiple-membership (cf. 4.2.10). Second, and more im-
portant, it makes the formulation of the grammatical rules more
complicated than the 'facts' would suggest is necessary. To quote
4.3. GRAMMAR AND LEXICON 165
Chomsky: 'The difficulty is that this subcategorization [i.e. the sub-
classification of the vocabulary] is typically not strictly hierarchic, but
involves cross classification. Thus, for example, Nouns in English are
either Proper (John, Egypt) or Common (boy, book) and either
Human (John, boy) or non-Human (Egypt, book)... But if the sub-
categorization is given by rewriting rules, then one or the other of
these distinctions will have to dominate, and the other will be
unstatable in the natural way.' For instance, if the class of Nouns is
Fig. 7.
first divided into Proper Nouns and Common Nouns, then each of
these in turn divided into a Human and non-Human subclass, the
only way in which a rule can be formulated to refer to all Human
nouns is by making it refer to both of the completely unrelated classes
Proper-Human and Common-Human (since there is no list of
Human Nouns in the lexicon). Chomsky goes on to point out: 'As
the depth of the analysis increases [i.e. with successive subclassifi-
cation], problems of this sort mount to the point where they indicate
a serious inadequacy in a grammar that consists entirely of rewriting
rules.'
We shall not discuss the revision of the grammatical rules which
Chomsky has proposed in order to solve these problems: in this
chapter we are working with a very simple system. As far as the
lexicon is concerned the effect of 'cross-classification* is clear. It
means that each word must be indexed in a way that makes it
possible to select, for instance, any 'human' noun (regardless of
whether it is 'proper' or 'common'), any 'concrete' noun (regardless
of whether it is 'animate' or 'inanimate'), and so on. The technical
term which has come to be associated with this kind of classification,
i66 4. GRAMMAR: GENERAL PRINCIPLES
or * indexing', is feature (cf. the use of the term * feature' in phonology,
with which there is a certain parallelism). Every word, we assume,
must be listed in the lexicon (which can now no longer take the form
of a set of rewrite-rules included in the grammar) with a set of features,
as follows:
boy: [common], [human], [masculine], . . .
door: [common], [inanimate], . . .
The rule, or rules, of lexical substitution will then be formulated in
a manner which makes it possible to select a particular word according
to one or more specified features. At what point in the generative
system the lexical-substitution rules will apply is a disputed question.
It should be observed, however, that, although we must now abandon
the view of the lexicon which represented it as a set of rules of the
form Na -> {boy, . . . } , our more general rule still holds (cf. 4.3.2):
X->X|XX
('rewrite X as x, where x is a member of the word-class X'). The
difference is that now X is that class of words which satisfy a parti-
cular feature-specification. For example, if the sentence generated by
the grammar calls for a 'common, human, masculine' noun, then X
is the class composed of all the words in the lexicon which include
among their grammatical features [common], [human] and [mascu-
line] : e.g. boy. But there is no list of this composite class.
4.3.4 Implications of congruence in grammatical and semantic

classification
We may now take up a further point. Although it is hardly relevant
to the understanding of generative grammar as this is represented so
far in the published literature, it is beginning to assume some
importance in connexion with certain proposals that have been made
recently for the formalization of semantics. As we shall see in later
chapters, there is good reason to believe that the integration of
grammar and semantics will bring about some rapprochement between
'formal' and 'notional* grammar. For that reason, it is especially
important not to lose sight of the general principles.
The terms used for the features mentioned in the previous para-
graphs ('proper', 'common', 'animate', 'masculine', etc.) were
described as grammatical We have not abandoned the principle that
such terms, when they denote classes of words referred to by the
4-3- GRAMMAR AND LEXICON 167
grammatical rules, are labels for what are assumed to be distribution-
ally-based classes. However, the labels that have been employed at
this point (which derive from 'notional* grammar) obviously carry
with them certain semantic implications. We have already referred
to the congruence which holds, in various degrees, between the
grammatical and the semantic structure of language; and we shall
come back to this question later. It may be assumed that most of the
grammatically 'animate' nouns will denote human beings or animals,
that most 'masculine* nouns will denote males, and so on. But the
classification of words in terms of such features as 'animate* or
'masculine* will often conflict with a classification based on the
meaning of the words (cf. chapter 7). This is well-recognized, and it is
the main reason why most linguists have turned away from ' notional *
grammar.
At the same time, it should be clearly understood that in a compre-
hensive description of a language the lexicon will include both
grammatical and semantic information for every word that is listed
there. It is not inconceivable that the semantic information should be
organized in such a way that it then becomes possible to derive part
of the grammatical information (required for the operation of the
grammatical rules) from the statement of the meaning of the word
whenever there is congruence between the grammatical and the semantic
classification. Let us suppose, for example, that the word man is given
in the lexicon with a statement of its meaning (in whatever form this
might take) and let us further suppose that from this information it is
possible to derive by a statable rule the fact that man (in at least one
of its meanings) denotes a male, adult, human being (n.b. this does
not imply that the meaning of man is 'male, adult, human being*:
this proviso is important in view of what will be said in the chapters
on semantics). Given this fact, we can derive from it the grammatical
classification of the word man as [masculine], [human], which implies
[animate], etc. This is a grammatical classification, because it is
designed to account for such distributional facts as the use of the
forms who v. which; he, him, his v. she, her v. it, its, etc. For example,
*The man which came..., or *The man washed her own shirt must be
excluded as unacceptable, and we will assume as ungrammatical.
At first sight, it might appear that this proposal to derive the
grammatical classification of a word from a statement of its meaning
contradicts the principles of formal grammar discussed in the previous
section. This is not so, provided that we respect the following
important condition: any general rule that is established for the
l68 4. GRAMMAR: GENERAL PRINCIPLES
derivation of grammatical features from the statement of the meaning
of a word given in the lexicon is inhibited by the overt specification of
the word with a contradictory feature. Let us assume, for example, that
all words denoting human beings are grammatically [human], and
that this also implies that they are [animate]; further, that they are
[masculine] if they denote males, and [feminine] if they denote
females. This is the general principle, which will apply if there is no
contradictory grammatical classification associated in the lexicon with
a given word. We will now state (and assume the grammatical
apparatus for the formalization of this statement) that [inanimate]
contradicts [animate]. Since the word man does not have any contra-
dictory grammatical features associated with it in the lexicon, the
general rules for the conversion of semantic information into gram-
matical will apply correctly.
But the word child might be entered in the lexicon with the feature
[inanimate], in order to allow such sentences as The child ate its
dinner. One realizes immediately, however, that the facts are far more
complicated than this simple technique can handle. Such sentences
as The child ate his dinner and The child ate her dinner are also
acceptable. We might therefore introduce a distinction between
[inanimate] and [neuter]. Normally, [inanimate] (whether it is
specified in the grammatical information or derived for a particular
word from an analysis of its meaning) will imply [neuter]; and
[animate] will imply either [masculine] or [feminine], the choice
being left free if the noun is not determined as one or the other. But
we might consider the possibility of entering child as both [animate]
and [neuter]. In this case, we must set up the grammatical rules in
such a way that the selection of either he or she on the basis of
[animate] is not inhibited by the presence of [neuter]; and, con-
versely, that the selection of it on the basis of [neuter] is not inhibited
by the presence of [animate]. The situation is in fact rather more
complicated than we have indicated here. But the general principles
are not affected by the additional complexity.
The proposal that has just been outlined has not yet been imple-
mented in any of the published literature (as far as I am aware). But
it has much to recommend it; and its implementation might well result
from developments that have been taking place recently in 'com-
ponentiar semantics (cf. 10.5.iff.). At the same time, it must be
stressed that the proposal is not only highly speculative at the present
time, but any attempt to implement it would entail a rather radical
revision of the formalization of generative grammar as this has been
4.3. GRAMMAR AND LEXICON 169
developed by Chomsky and his followers. Various suggestions have
been made recently by other authors which, for other reasons, would
tend to imply that this revision is necessary; and reference to these
will be made in the notes appended to later sections of this book.
However, in the body of the book we shall follow (in general, if not
always on points of detail) the main line of development in the theory
of generative grammar, as this has been traced by Chomsky and those
working most closely with him.
4.3.5 Summary
In this chapter, we have introduced the notion of 'generative
grammar'; and we have deliberately simplified the treatment of
grammatical rules, in order to concentrate upon the more general
principles. We have also maintained the view that sentences are
constructed by the simple operation of taking words from the lexicon
(according to their grammatical classification) and combining them in
sequence. Furthermore, we have not distinguished consistently
between 'sentence* and 'utterance', since we have tacitly assumed
that the grammar (supplemented by the rules of phonology and their
phonetic interpretation in the substance of sound) generates sentences
which are 'identical* with the potential utterances of the language.
Modifications will be introduced on all these points in the course of
the next two chapters. Before we return to the further development of
the generative framework, we must discuss the nature of sentences
and other grammatical units.
John Lyons
Online ISBN: 9781139165570
Chapter
5 - Grammatical Units pp. 170-208

5
GRAMMATICAL UNITS
5.1 Introductory
5.1.1 Words, sentences, morphemes, phrases and clauses
Traditional linguistic theory operates with two fundamental units of
grammatical description: the word and the sentence. Both of these
units are given practical recognition in the conventions of different
writing systems. For instance, in the various alphabetic systems
employed for European languages, as well as for many other languages
throughout the world, sentences are separated from one another by
using special marks of punctuation (full-stop, question-mark, ex-
clamation-mark) and by capitalizing the first letter of the first word in
each sentence; and, within sentences, words are separated from one
another by spaces. For this reason, the educated layman is familiar
with the terms 'word' and 'sentence', and uses them freely in talking
about language.
So far we have been employing the terms 'word* and 'sentence*
without definition or explanation. We must now examine these terms
in the light of the general principles discussed in the previous chapter,
taking account of the implications that 'word' and 'sentence' carry in
everyday usage and in traditional grammatical theory.
For reasons which will be explained later in this chapter, the
classical grammarians were little concerned with the analysis of words
into smaller elements. However, it is clear that, in many languages at
least, such elements exist. For instance, the English word unaccept-
able is made up of three smaller units, each of which has a charac-
teristic distribution: uny accept, and able. Moreover, these are minimal
units in the sense that they cannot be analysed further into distri-
butionally-classifiable units of English. Such minimal units of
grammatical-analysis, of which words may be composed, are
customarily referred to as morphemes.
We have therefore three different units of grammatical description
to consider in this chapter: sentences, words and morphemes.
Intermediate between the word and the sentence, two other units are
commonly recognized by grammarians: phrases and clauses. Tradi-
[170]
tionally, the distinction between the two was formulated somewhat as
follows: any group of words which is grammatically equivalent to a
single word and which does not have its own subject and predicate is
a phrase; on the other hand, a group of words with its own subject
and predicate, if it is included in a larger sentence, is a clause. The
distinction between phrases and clauses was not always drawn clearly
or consistently in the analysis of particular sentences. Theoretically,
the traditional distinction between phrases and clauses amounts to a
distinction between word-like and sentence-like groups of words
within sentences; for the sentence itself, as we shall see, was tradi-
tionally defined in terms of 'subject* and 'predicate*. The phrases
and clauses of traditional grammar are therefore secondary units
defined in terms of their grammatical equivalence to the primary
units, words and sentences. We shall have little to say about phrases
and clauses from the point of view of modern grammatical theory in
this book (but cf. 5.5.1, 6.2.10).
The relation between the five units of grammatical description (in
languages for which allfiveare established) is one of composition. If we
call the sentence the' highest' unit and the morpheme the' lowest', we
can arrange all five units on a scale of rank (sentence, clause, phrase,
word, morpheme), saying that units of higher rank are composed of
units of lower rank. Alternatively, we can say that units of higher rank
can be analysed (or 'decomposed') into units of lower rank.
Many older books on language devoted a good deal of space to
discussing which of the two traditional primary units of grammatical
description, the word or the sentence, it to be regarded as 'basic':
does the grammarian first of all identify words and then account for
the structure of sentences in terms of the permissible combinations
of words, or does he start by recognizing the sentences in his material
and then analyse these sentences into their constituent words? We
shall not go into this question here. For the present, we will assume
that the grammar of a language is neutral with respect to the question
whether one works 'up* or 'down* the scale of rank, just as it is
neutral to the distinction of analysis and synthesis (cf. 4.3.1).
5.1.2 The utterance

One reason why linguists no longer argue whether the sentence or the
word is more 'basic' is that they now realize, more clearly than did
their predecessors, that neither words nor sentences, nor indeed
any of the other units of linguistic description, are 'given' in the
172 5- GRAMMATICAL UNITS
unanalysed material. When the linguist sets out to describe the grammar
of a language on the basis of a recorded corpus of material, he starts
with a more primitive notion than that of either the word or the
sentence (by 'primitive* is meant 'undefined within the theory',
'pre-theoretical'). This more primitive notion is that of the utterance.
Just as 'acceptability* is a more primitive notion than 'grammati-
cally' or'significance* (cf. 4.2.1), so'utterance* is more primitive than
'word*, 'sentence*, 'morpheme*, etc., in that its application does not
rest upon any technical definitions or postulates of the science of
language. The utterance has been defined by Harris as ' any stretch of
talk, by one person, before and after which there is silence on the
part of that person*. It must be remembered that we are not dealing
here with the formal definition of some linguistic unit, but with a pre-
scientific description of the linguist's data. 'Silence* and the other
terms used to characterize and delimit utterances are to be under-
stood with the tolerance customarily granted to everyday non-
scientific discourse. Furthermore, we must not assume that there will
necessarily be a high degree of correspondence between utterances
and sentences, or indeed between utterances and grammatical units of
any one particular type. As Harris goes on to say, ' The utterance is,
in general, not identical with the "sentence" (as that word is
commonly used), since a great many utterances in English, for
example, consist of single words, phrases, "incomplete sentences",
etc. Many utterances are composed of parts which are linguistically
equivalent to whole utterances occurring elsewhere.'
5.2 The sentence

5.2.1 Bloomfield's definition of the sentence
Bloomfield's definition of the sentence will serve as a starting-point for
our discussion. According to Bloomfield a sentence is' an independent
linguistic form, not included by virtue of any grammatical construc-
tion in any larger linguistic form*. He exemplifies his definition with
the following utterance: How are you? It's a fine day. Are you going to
play tennis this afternoon? \ and goes on to say, 'whatever practical
connexion there may be between these three forms there is no
grammatical arrangement uniting them into one larger form: the
utterance consists of three sentences*. The point of Bloomfield's
definition can be stated more concisely as follows: the sentence is the
largest unit of grammatical description. A sentence is a grammatical
5-2. THE SENTENCE 173
unit between the constituent parts of which distributional limitations
and dependencies can be established, but which can itself be put into
no distributional class. This is equivalent to saying that the notion of
distribution, which is based on substitutability, is simply not applic-
able to sentences. Consider Bloomfield's example. Unless we take the
wider context into account, it is pointless to talk of substituting other
forms for, say, It's a fine day. And even if we do take into account the
wider context, what Bloomfield calls the practical connexions'
between the three sentences could not be brought within the scope of
general rules of distributional selection. Not only are the constituent
elements of the sentence It's a fine day not predictable from the
wider context; it is not even possible to say that a statement, rather
than a question, is bound to occur. How are you?, It's a fine day., Are
you going to play tennis this afternoon?, are all distributionally indepen-
dent of one another; and for that reason they are recognized as three
distinct sentences.
5.2.2 *:Derived* sentences

A few examples may now be given of various kinds of utterances or
parts of utterances which are traditionally regarded as sentences,
although they are not distributionally independent in the sense in
which we have been using the term ' distribution'. First of all, we may
consider utterances which contain pronouns of 'personal' reference
(he, she, it, they, in English). An utterance like He'll be here in a
moment presupposes the previous occurrence of some masculine noun
or noun-phrase (e.g. John, the milkman, etc.) to which the pronoun he
refers. The distributional restrictions on the occurrence of 'personal'
pronouns which are traditionally formulated by saying that he (him,
his) 'stands for* masculine singular nouns, she (her) for feminine
singular nouns, it (its) for neuter singular nouns, and they (them,
their) for plural nouns, may operate over very long utterances (and
indeed over several different utterances by several speakers). How-
ever, if, as is generally the case, the utterances can be segmented into
stretches which are distributionally independent of one another in all
respects apart from the selection of the 'personal' pronouns, these
stretches may be regarded as 'derived' sentences within which nouns
have been replaced with pronouns (masculine, feminine, etc., as
appropriate) by secondary grammatical rules. This is what is implied
by the traditional term 'pronoun' (a grammatical element of variable
reference which 'stands for' some noun previously given, explicitly
or implicitly, in the context). And it is clear that in order to under-
stand a 'derived* sentence, like He'll be here in a moment, the hearer
must be able to substitute for the pronoun he the correct noun or
noun-phrase which it 'replaces*.
As another example of distributional constraints running over a
sequence of what would be normally regarded as separate sentences
we may consider the 'indirect discourse' construction in Latin. This
construction is seen in its simplest form in a sentence such as Dico te
venisse in Marci Laecae domum (from Cicero: ' I assert that you came
into Marcus Laeca's house'), where the segment te venisse in.,,
domum is the 'indirect' form (in the 'accusative and infinitive'
construction) of the 'direct' form (tu) venisti in.. .domum. It is not
at all uncommon in Latin authors for a whole speech to be put in
'indirect discourse' with each segment in it that is traditionally
regarded as a separate sentence (and punctuated as such) marked as
dependent on some previous 'verb of saying' by the occurrence of
the'accusative and infinitive' (and other features). This phenomenon,
which may be called 'extended indirect discourse', is also found in
English (and many other languages). Take for instance a passage such
as the following: The prime minister said that he deeply regretted the
incident. He would do everything he could to ensure that it did not happen
again. On the following day he would confer with his colleagues. He was
confident that... Once again, passages of this kind are best accounted
for in two stages: first of all, by describing a set of independent
sentences in their 'direct' form (/ deeply regret...; I will do every-
thing I can...; Tomorrow I will confer.. .;Iam confident that...) and
then, by specifying the secondary grammatical rules which will
transpose each of these sentences into the corresponding 'indirect'
form when they occur in sequence after a 'verb of saying'.
5.2.3 'Incomplete9 sentences

Finally, we may consider the case of what are traditionally called
'incomplete' or 'elliptical' sentences.This category has recently come
in for a lot of criticism from linguists, who have pointed out (correctly,
but irrelevantly) that, when such utterances occur in a particular
context, they are perfectly comprehensible, and from this point of
view can hardly be called 'incomplete'. One must distinguish between
contextual completeness and grammatical completeness.
There are many utterances of normal, everyday conversation which
are dependent for their internal form on the preceding utterances of
5.2. THE SENTENCE 175
the same speaker or the person with whom he is conversing. An
example might be John's, if he gets here in time, which could hardly
occur except immediately after a question which 'supplied* the words
required to make it into what would traditionally be regarded as a
Complete' sentence. For instance, it might occur after Whose car are
you going in?, but not after When are you going there? An utterance
like John's, if he gets here in time is therefore grammatically 'in-
complete', since it is not itself a sentence (it is not distributionally
independent) and yet it can be derived from a sentence which is
constructed by adding to the utterance various elements 'given* in
the context: We are going in John's car, if he gets here in time.
Somewhat different is the case of an utterance such as Got the
tickets? At first sight it might appear that there is no reason to call this
an 'elliptical' sentence at all, a shortened form of the sentence Have
you got the tickets?, since it can be interpreted without reference to any
previously occurring utterance. It is probable that the simplest way
of accounting for the utterance in question is by deriving it by means
of a rule which deletes Have you from sentences beginning Have you
got... For in this way we can account for the fact that it will be
understood to 'contain* the pronoun you and also for the form of the
reply Yes, I have {got them). The difference between this example ^&d
the one considered in the previous paragraph is, however, quite clear.
Although they are both 'elliptical', shorter forms of some longer
version of the same sentence, they are ' elliptical' in a different sense.
John's, if he gets here in time is grammatically incomplete; that is to
say, it is not a sentence, and therefore is not to be described directly by
the grammar, but by supplementary rules (if such rule* can be
established) which account for the deletion of contextually-determined
elements in the sentences from which the utterances of connected
discourse are derived. On the other hand, Got the tickets? (in British
English at least) is a sentence (and in that sense it is 'complete'); the
ellipsis that is involved in its derivation from the alternative version
of the same sentence Have you got the tickets? is purely a matter of
grammar and is independent of the wider context. Similarly, the
grammatical rules of English must allow for the generation of such
grammatically equivalent (but perhaps stylistically distinct) forms as
don't and do not, can't and cannot, etc., occurring in otherwise
identical sentences. Traditional grammarians often failed to distin-
guish between grammatical and contextual completeness.
176 5 GRAMMATICAL UNITS
5.2.4 Two senses of the term 'sentence9

The sentence is the maximum unit of grammatical analysis: that is, it
is the largest unit that the linguist recognizes in order to account for
the distributional relations of selection and exclusion that are found
to hold in the language he is describing. It will be clear from the
discussion of the examples given above (which could be multiplied)
that distributional relations frequently hold across the boundaries of
segments of utterances which would normally be regarded as
separate sentences. This would appear to be a contradiction. It may
be resolved, however, by distinguishing two senses of the term
'sentence'. As a grammatical unit, the sentence is an abstract entity
in terms of which the linguist accounts for the distributional relations
holding within utterances. In this sense of the term, utterances never
consist of sentences, but of one or more segments of speech (or
written text) which can be put into correspondence with the sentences
generated by the grammar. On the other hand, the segments them-
selves are often referred to as sentences, as for instance by Bloomfield,
when he says that the utterance How are you? It's a fine day. Are you
going to play tennis this afternoon? * consists of three sentences'. In the
case of segments which are distributionally independent (such as the
three segments into which Bloomfield divides his utterance) it is
tempting to say that the grammar generates them directly without
recourse to the more abstract grammatical units. But, as we have seen,
the correspondence between sentences and the segments referred to
them is often far less straightforward. It would seem to be preferable
therefore to restrict the term 'sentence' (as traditional grammarians
did, though perhaps not explicitly or consistently) to the more
abstract sense suggested here. To invoke de Saussure's distinction
explained above (cf. 1.4.7), utterances are stretches of parole produced
by native speakers out of sentences generated by the system of
elements and rules which constitute the langue. The linguist
describes instances of parole by establishing the langue and relating
them to it in the simplest way. Having made this point, we shall
continue to follow the normal practice of linguists by saying that
utterances are composed of sentences. This form of expression is to
be understood in the way explained here.
5.2.5 'Ready-made' utterances

At this point we should perhaps mention a further category of
utterances or parts of utterances which resemble * incomplete'
sentences in that they do not correspond directly to sentences
generated by the grammar, but differ from them in that their
description does not involve the application of the rules established to
account for the vast mass of more * normal' utterances. These are
what de Saussure has called * ready-made utterances' ('locutions
toutes fakes'): expressions which are learned as unanalysable wholes
and employed on particular occasions by native speakers. An example
from English is How do you do?, which, though it is conventionally
punctuated as a question, is not normally interpreted as such; and
unlike genuine questions beginning How do you..., constructed by
means of the productive rules of English grammar, it cannot be
matched with corresponding sentences of the form / very well,
Hozv does he ?, He beautifully, etc. Another * ready-made'
English expression is Rest in peace (as a tombstone inscription) which,
unlike for example Rest here quietly for a moment, is not to be
regarded as an instruction or suggestion made to the person one is
addressing, but a situationally-bound expression which is unanalysable
(and which does not require any analysis) with reference to the
grammatical structure of contemporary English. The stock of
proverbs passed on from one generation to the next provides many
instances of 'ready-made utterances' (cf. Easy come easy go; All that
glisters is not gold; etc.). From a strictly grammatical point of view
such utterances are not profitably regarded as sentences, even though
they are distributionally independent and thus satisfy the definition
of the sentence given above. Their internal structure, unlike that of
genuine sentences, is not accounted for by means of rules which
specify the permissible combinations of words. However, in a total
description of the language, which brings together the phonological and
the grammatical analysis, they might be classified as (grammatically
unstructured) sentences, since they bear the same intonation contour
as sentences generated by the grammar. Apart from this fact, they are
to be accounted for simply by listing them in the dictionary with an
indication of the situations in which they are used and their meaning.
In addition to those 'ready-made' expressions which may serve as
complete utterances and which permit no extension or variation,
there are others which are grammatically unstructured, or only
partially structured, but which can yet be combined in sentences
according to productive rules. Examples are: What's the use of
-ing... ?; Down with / \for 's sake. There is no generally accepted
term for such elements. We will refer to them as schemata. Schemata,
it will be observed, may be of different ranks. What's the use of
-ing... ? and Down with / are sentence-schemata (and so belong to
the only class of grammatical units which concerns us in this section).
An indefinitely large number of sentences can be generated from
them by ' filling' the vacant * slot * in the schema with a member of the
appropriate grammatical class: thus, What's the use of worrying?,
What's the use of getting everything ready the night before ?, Down with
the King!, Down with the Sixth Republic! On the other hand, for 's
sake is a phrase-schema; and the grammar must not only account for
the class of elements which can *fillthe slot' in the schema {for his sake,
for my mother's sake, etc.), but must also classify the resultant phrase
according to its distribution in sentences (/ did it for 's sake, etc.).
5.2.6 Different types of sentences

In traditional grammar, sentences are classified into different types in
two ways: first of all by function, as statements, questions, exclama-
tions and commands; and secondly according to their structural
complexity, as simple or compound. Complex sentences are made
up of a number of simple sentences (which when incorporated as
constituents of larger sentences are, by virtue of this fact, called
clauses). Thus: / saw him yesterday and I shall be seeing him again
tomorrow is a complex sentence. Complex sentences are divided
into: (a) those in which the constituent clauses are grammatically
co-ordinate, no one being dependent on the others, but all being, as it
were, added together in sequence, with or without the so-called
co-ordinating conjunctions {and, but, etc.); and {b) those in which one
of the clauses (the 'main clause') is 'modified* by one or more
subordinate clauses grammatically dependent upon it and generally
introduced (in English) by a subordinating conjunction {if, when,
etc.). Subordinate clauses are subdivided by function as nominal,
adjectival, adverbial, etc.; and further as temporal, conditional,
relative, etc. We shall not go into the details of this classification here.
The notion of grammatical dependence has been given a distribu-
tional interpretation in this section; and the other notions, of
'function* and 'modification', whereby sentences may be classified
into various types will be treated from a distributional point of view
in a later chapter (cf. 6.4.1).
There is one general point which must be made before we tem-
porarily leave the question of sentence-types. The * formal' approach
to grammatical description adopted in this book and accepted by most
linguists today implies a rejection of any attempt to categorize
sentences 'notionally' in advance of their classification in terms of
internal structure (that is, in terms of the distributional relations
holding between their parts). It must not be assumed in advance that
every language will have formally differentiated patterns of sentence-
structure for each of the four major sentence-types (statements,
questions, exclamations, commands) recognized in traditional
grammar. What must be done, as we have insisted in the previous
chapter, is first to establish for each language independently the
grammatical units and patterns of combination valid for that language,
and only then, if at all, to give them such labels as Statement',
1
question', etc., in terms of their semantic or contextual correlations.
Traditional grammar recognized four main sentence-types because
Greek and Latin had four formally distinct patterns of sentence-
construction which could be classified, roughly, into the four semantic
categories of statements, questions, exclamations, and commands.
The doctrine that these four types are universal grammatical cate-
gories, like the doctrine that the 'parts of speech' are universal
features of language, was part of the wholesale transference of the
particular details of Greek and Latin to the plane of the theoretically
necessary and a priori categories assumed by the medieval * specu-
lative' grammarians and their successors.
5.2.7 Phonological criteria

The criterion of distributional independence is not always sufficient
of itself to segment an utterance uniquely into a determinate number
of sentences. (It will be remembered that utterances are said to be
composed of sentences only with the reservations set forth earlier in
this section.) Take for example the utterance I shouldn't botheriflwere
you Td leave it till tomorrow. The segment if I were you is not distri-
butionally independent (and so is not a sentence), since it presupposes,
generally at least, another segment (either preceding it or following it)
in the same utterance containing would, should, 'd> etc. Both the
segment preceding if I were you and the segment following it satisfy
these conditions of distributional presupposition, so that the utterance
can be divided into two sentences in two different ways; / shouldn't
bother if I were you.jFd leave it till tomorrow, and / shouldn't bother, jIf I
l8o 5. GRAMMATICAL UNITS
were you Fd leave it till tomorrow. In deciding between these two
possibilities in any particular case we should have recourse to other
considerations, principally to the criteria of potential pause and
intonation. Distributionally defined sentences in English which
present no problems of demarcation have a characteristic intonation
pattern and may be set off from one another when they occur in
sequence in the same utterance by pauses of greater or less duration.
Hence, in the total description of English the sentence, which is
defined primarily in grammatical terms, is also to be recognized as
the domain of the phonological features summed up in the term
4
intonation*. Since this is so, it makes for a simpler total description
of the language if the phonological features in question are allowed
to determine the cases, such as the one exemplified above, left un-
decided by the strictly grammatical criteria. It is, of course, theoreti-
cally conceivable that in a particular language the domain of intonation
should be quite differently established (as a given number of syllables,
for example) and correspond to no grammatically delimited unit.
However, it is probably the case that in all languages the sentence is
the unit at which there is the greatest 'congruence of levels', particu-
larly between the phonological and grammatical levels of description.
It may be mentioned here that what are traditionally referred to as
complex sentences made up of co-ordinate clauses would not be
recognized as single sentences, but as sequences of separate sentences,
on purely grammatical criteria. An utterance such as / saw him
yesterday and I shall be seeing him again tomorrow would be segmented
by the test of distributional independence into two sentences (the
break coming between yesterday and and). However, the supple-
mentary criteria of potential pause and intonation will distinguish
utterances in which two or more consecutive sentences are to be taken
as clauses in a single sentence or as independent sentences. Ortho-
graphical practice reflects the distinction in such cases: cf. / saw him
yesterday. And I shall be seeing him again tomorrow, on the one hand,
and / saw him yesterday and I shall be seeing him again tomorrow, on the
other.
5.3 The morpheme

5.3.1 Word and morpheme
In the discussion of the other two 'primary* units of grammatical
analysis, the word and the morpheme, we are faced with the difficulty
that, whichever one we take first, we must presuppose some knowledge
5.3. THE MORPHEME l8l
of the other. Most modern treatments of grammatical theory have
glossed over this difficulty by defining the morpheme as the minimal
unit of grammatical analysis (which is the definition we will pro-
visionally adopt) and then failing to point out that the general practice
of linguists is not always consistent with this definition, but is
conditioned equally by some explicit or implicit reference to the
word as a grammatical unit. The reasons for this ambivalence or
equivocation are historically explicable and will become clear in the
course of our discussion. We shall then see that neither words nor
morphemes (as these terms as generally applied by linguists) are
universal features of language, although it would be possible to make
them so by definition. However, in order to make either one or the
other of the two units universal, we should have to make a more
radical break with the past than most linguists have been prepared to
do so far. Although we will deal mainly with the morpheme in this
section and with the word in the next, there will necessarily be a
certain measure of overlap between the two sections.
5.3.2 Segmentability of words

We have described morphemes as minimal units of grammatical
analysisthe units of * lowest' rank out of which words, the units
of next 'highest' rank, are composed (cf. 5.1.1). By way of example,
we said that the English word unacceptable is composed of three
morphemes, uny accept, able, each one of which has a particular
distribution and also a particular phonological (and orthographical)
form, or 'shape'. We must now draw a distinction between the
morphemes themselves, as distributional units, and their phonological
(or orthographical) 'shape'.
Whether a word can be divided into smaller grammatical segments
is a matter of degree. The words boys; jumps, jump-ed, jump-ing;
tall-er, tall-est; etc. can be segmented into their constituent parts no
less readily than un-accept-able; and so can the majority of English
nouns, verbs and adjectives. Such words, we will say, are determinate
with respect to segmentation. But there are many other English nouns,
verbs and adjectives which either cannot be segmented at all or are
only partially determinate with respect to segmentation. Examples
are: the irregular plurals men, children, mice, sheep, etc.; the ' strong'
verbs went, took, came, run, cut, etc.; the irregular comparatives and
superlatives better, best, worse, worst. These words all present
problems of segmentation, different in degree or in kind. For instance,
l82 5. GRAMMATICAL UNITS
men stands in the same grammatical relationship to man as boys does
to boy (as boys is the plural of the singular form boy, so men is the
plural of the singular man); and there is at least some phonological
(and orthographical) resemblance between men and man which could
be made the basis for segmenting men into two parts. The same is true
for mice and mouse (notice that here the orthographical difference is
greater than the phonological). On the other hand, although worse
and went stand in the same grammatical relationship to bad and go
as do taller and jumped to tall and jump, there is no phonological
resemblance at all between worse and bad or between went and go.
Words such as worse or went cannot be segmented into parts. Linguists
have exercised considerable ingenuity in arguing for one Solution'
rather than another to the problem of words that are indeterminate
with respect to segmentation and even for the * segmentation* of
such words as worse and went. We shall not discuss these * solutions'
here since they are motivated by certain methodological assumptions
which are less general than the assumptions we are making in the
present work. It suffices for our purpose to have drawn attention to
the fact that, in some languages at least, there are words which cannot
be segmented into parts, except arbitrarily, although these words
belong to the same grammatical class as other words which are
segmentable.
5.3.3 The morpheme as a distributional unit

Now there is nothing in the definition of the morpheme to imply that
it must always be an identifiable segment of the word of which it is
a constituent. To say that worse is composed of two morphemes, one
of which it shares with bad (and worst) and the other of which it shares
with taller, bigger, nicer, etc., is equivalent to saying that worse differs
from taller, bigger, nicer, etc. in grammatical function (that is, in its
distribution throughout the sentences of English) in the way that bad
differs from tall, big, nice, etc. (and worst from tallest, etc.). This is
commonly expressed as a proportion of grammatical, or distributional,
equivalence (cf. the original sense of 'analogy*: 1.2.3).
bad : worse : worst = tall: taller : tallest
This proportion expresses the fact that, for example, worse and taller
(as well as bigger, nicer, etc.) are grammatically alike in that they are
comparative forms of the adjectivethey can occur in such sentences
as John is worse {taller, etc,) than Michael, It is getting worse [taller,
5-3- THE MORPHEME 183
etc.) all the time. Worse and taller (as well as bigger, nicery etc.) differ
from one another, however, in that they cannot occur in exactly the
same set of sentencesfor instance, as traditional grammarians would
say, they cannot 'qualify* exactly the same set of nouns. In so far as
the class of nouns which can be qualified by a particular adjective is
grammatically determined (and here we touch upon a point to which
we shall return presently), this feature of their distribution is accounted
for by postulating a particular morpheme as a component of one
adjective and another morpheme as a component of another adjective
which * qualifiesy a different class of nouns.
In order to make this point clear, let us first put in symbolic form
the distributional proportion we have just set up, representing each
different word with a different letter, and factorize this as we would
any other algebraic proportion:
A:B:C = D:E:F
Factorizing (and employing arbitrary symbols) we obtain:
ax : bx : ex = ay : by : cy
That is to say, each word is factorized into two components; all the
words on the left-hand side of the equation have the component x,
and all the words on the right-hand side the component y; with
regard to their other component (a, b or c) the first word on the left-
hand side agrees with the first word on the right-hand side, the
second word on the left-hand side agrees with the second word on the
right-hand side, and so on. The components, or distributional factors,
of the words are morphemes.
By factorizing the distribution of words in this way we can account
for their occurrence in sentences in terms of the distribution of their
component morphemes: the distribution of a word is the product of
the distribution of the morphemes of which it is composed.
5.3.4 Morpheme and morph

It is clear that from this point of view the question whether words can
be segmented into parts or not is quite irrelevant. The morpheme is
not a segment of the word at all; it has no position in the word (for
example, in the analysis of A into its component morphemes we could
just as well put xa as ax), but mere "r its 'factorial' function. When the
word can be segmented into parts, these segments are referred to as
morphs. Thus the word bigger is analysable into two morphs, which
7-2
184 5' GRAMMATICAL UNITS
can be written orthographically as big and er (with the orthographical
conventions of English accounting for the additional ' linking' g) and
in a phonological transcription as /big/ and /a/. Each morph represents
(or is the exponent of) a particular morpheme.
The distinction that we have drawn here between morphs and
morphemes can be expressed in terms of de Saussure's distinction of
substance and form (cf. 2.2.2). Like all grammatical units, the mor-
pheme is an element of 'form', 'arbitrarily' (cf. 2.2.7) related to its
' substantial' realization on the phonological (or orthographical) level
of the language. As we have seen, morphemes may be represented
directly by phonological (or orthographical) segments with a particular
'shape' (that is, by morphs), but they may also be represented in the
substance of the language in other ways. In order to refer to mor-
phemes, it is customary to use one of the morphs which represents the
morpheme in question and to put it between braces. Thus {big} is the
morpheme which is represented in phonological substances by /big/
and in orthographic substance by big; and the word went (phono-
logically /went/), which cannot be segmented into morphs, represents
the combination of the two morphemes {go} and {ed}. Although we
shall follow this convention, it must be realized that the particular
notation chosen to refer to morphemes is a matter of arbitrary
decision. We might just as well number the morphemes and say, for
instance, that {207} is represented by /big/ (or big); and that {1039} +
{76} is represented by the substantially unitary form /went/ (or
went).
5.3.5 Allomorphs
A further point may now be made with regard to the relationship
between morphemes and morphs. It frequently happens that a par-
ticular morpheme is not represented everywhere by the same morph,
but by different morphs in different environments. These alternative
representations of a morpheme are called allomorphs. For example,
the plural morpheme in English, which we may refer to as {s}, is
regularly represented by the allomorphs /s/, /z/ and /iz/. These are
phonologically conditioned, in the sense that the selection of any one is
determined by the phonological form of the morph with which it is
combined. The rule is as follows: (i) if the morph representing the
noun morpheme with which {s} is combined to form the plural ends
with a 'sibilant' (/s/, /z/, /J/, jij, /tj/, /d3/), {s} is represented by /iz/
(cf. /bAsiz/, buses; /saiziz/, sizes; /fijiz/, fishes; /gara:3iz/, /gara:d3iz/,
5-3 THE MORPHEME 185
garages (n.b. the variation in the phonological representation of this
word by speakers of standard British English); /batjiz/, batches; etc.);
(ii) otherwise, if (a) the morph ends in one of the voiced phonemes
(including the vowels), {s} is represented by /z/ (cf. /dogz/, dogs;
/bedz/, beds; /laiz/, lies; etc.) and if (b) the morph ends in a voiceless
(consonant) phoneme, {s} is represented by /s/ (cf. /kats/, cats; /bets/,
bets; etc. (It will be observed that the orthographical conventions of
English distinguish only two of these three allomorphs, with -s
standing for both /s/ and /z/ and -es standing for /iz/. The present
tense singular morpheme, which we can refer to as {z} (in order to
distinguish it from the morpheme {s} which forms the plurals of
English nouns) is regularly represented by the same three allomorphs
as {s}. And the statement of the phonological conditioning of their
occurrence is identical: cf. the verbs (i) /fijiz/, fishes; /kaetjiz/,
catches; etc.; (ii) /digz/, digs; /ebz/, ebbs, etc.; and (iii) /kiks/, kicks;
/sips/, sips; etc. The past tense morpheme of English, {ed}, is also
regularly represented by three phonologically-conditioned allo-
morphs: /t/, /d/ and /id/. The rule governing their distribution is as
follows: (i) /id/ occurs after morphs ending in alveolar stops (i.e. after
/t/ and /d/) (cf. /wetid/, wetted; /wedid/, wedded, etc.); elsewhere,
(ii) /d/ occurs after voiced phonemes (including the vowels and the
nasals) and (iii) /t/ after voiceless phonemes (cf. (ii) /saegd/, sagged;
/Uvd/, loved; /moud/, mowed; /maind/, mined; etc. and (iii) /saekt/,
sacked; /pAft/, puffed, etc.). In fact, the allomorphic variation associ-
ated with the regular representations of all the three morphemes
discussed here, {z}, {s} and {ed}, can be subsumed under a more
general rule whereby the appropriate morph is generated from an
underlying invariant morph, neutral with respect to voicing, which is
syllabified (as /iz/ or /id/) when combined with a morph ending in
4
the same sound* (i.e. for the purpose of this rule, all the sibilants
are considered to 'contain the same sound' as the underlying 's-
sound' representing both {s} and {z} and both /t/ and /d/ to ' contain
the same sound' as the underlying alveolar sound representing {ed}).
It will be obvious that a rule of this kind can be stated in terms of
prosodic or distinctive feature analysis more readily than with
reference to a phonemic analysis of English: cf. 3.3.8 ff.
The limiting case of allomorphic variation is found where no
generalization can be made, in terms of phonological structure or in
any other terms, about the selection of a particular allomorph. This
situation can be illustrated from English. In addition to the three
regular allomorphs of the English plural morpheme {s} one might also
establish the form /an/ which is to be found in the word oxen,
/oksan/. Since all other morphs ending in /ks/ which represent noun-
morphemes in English have the regular /iz/ in the plural (cf. /boksiz/,
boxes; /foksiz/, foxes, etc.), the occurrence of /an/ in /oksan/ is not
phonologically conditioned. In fact, it is not determined by any
feature of the morpheme {ox} or the morph /oks/ which can be
brought within the scope of any general statement about the structure
of English. It is true that the plural nouns children and brethren also
end in /an/. But whereas oxen presents no problems of segmentation,
since it can be analysed into two morphs, /oks/ and /an/, the former of
which is identical with the morph representing the singular ox (and
in this respect oxen is like the regular plurals in English), the recogni-
tion of /an/ in children and brethren would leave us with two morphs,
/tjildr/ and /bredr/, neither of which is identical with the morph
representing the singular of these nouns (even granting that brethren
has a singular in modern English) and neither of which occurs else-
where in the language. Since the formation of the word oxen is an
irregular fact of English, which, despite the segmentability of the
word into two constituent morphs, can only be handled by an ad hoc
'rule' applying to this one instance, there is little point in recog-
nizing /an/ as an allomorph of {s} in the description of contemporary
English.
The reader may be tempted to think that the elaboration of such
subtle distinctions as those we have drawn in this section between
morpheme, morph, and allomorph is something of an idle, scholastic
pastime which serves no useful purpose. But such distinctions are
essential if we wish to construct a general theory of language-
structure. As we shall see, in certain languages words can generally be
segmented into parts (morphs), in others they cannot; in some
languages the morphs each tend to represent a single minimal
grammatical unit (a morpheme), in others they do not; and in some
languages each morpheme is usually represented by a segment of
constant phonological form, whereas in others certain morphemes
are represented by a set of alternant morphs (allomorphs) the selection
of which in particular environments may be conditioned by phono-
logical or grammatical factors.
It is true that a good deal of what is often regarded as phono-
logically-conditioned allomorphic variation may be eliminated from
the description by adopting a prosodic or distinctive-feature analysis
for the phonology. But grammatically-conditioned variation of allo-
morphs cannot be eliminated in this way, and only a certain amount
5.3. THE MORPHEME 187
of phonologically-conditioned variation. The concept of the allo-
morph is therefore useful. It is, however, the distinction between the
morpheme and the morph, between the grammatical unit and its
' substantial' representation, which is particularly important. For it is
by making this distinction that we can bring out clearly both the
grammatical similarity and the formational difference between such
words as went and killed, or worse and bigger. In the purely gram-
matical part of the description both the 'regular' and 'irregular'
forms can be handled alike: {go} + {ed}, {kill+ed}; {bad} + {er},
{big} + {er}; etc. The difference between the ' regular' and ' irregular'
forms is seen at that point in the description where words as purely
grammatical units are 'embodied', as it were, in phonological (or
orthographic) substance. In the case of the regular forms, like killed,
rules can be set up to combine morphs (and one now sees the
advantage, from the point of view of total description of the language,
of using the morph to stand for the morphemea convention which
was described above as purely arbitrary). These rules are of very
general applicability, and, in many cases, their scope can be left open
by employing a formulation which says, in effect, but more formally:
' any forms which are not accounted for by one of the special rules are
to be handled by the following general rule(s) according to the
following conditions'. The irregular words are handled by special
rules of restricted scope, applying in the limiting case to one and only
one word: e.g. c{go} + {ed} is realized by went9. One means of ensuring
that both the ' regular' and the ' irregular' forms are handled appro-
priately is by ordering the rules in such a way that the rule of limited
scope is applied first, where applicable, and then the rules of general
applicability, whose scope may now be left totally unrestricted.
5.3.6 Isolating, agglutinating and inflecting languages

Now that we have distinguished between morpheme, morph and
allomorph, we can draw upon these distinctions in the exemplification
of some of the differences between languages referred to in the previous
paragraph. Languages are frequently classified into structural types
(in terms of a system of classification which originated in the nine-
teenth century) as isolating, agglutinating and inflecting (or 'fusional').
An isolating (or 'analytic') language is defined as one in which all
words are invariable. (Since we are for the present taking the concept
of the word for granted, we shall postpone to the following section the
question whether there is any need to distinguish between the word
and the morpheme in the description of isolating languages.) Chinese
is often cited as a well-known example of the isolating type of
language; but nowadays scholars appear to agree that many Chinese
words are composed of more than one morpheme, and Vietnamese is
said to be a more ' typical' isolating language than Chinese. Whether
a language is isolating or not is obviously a matter of degree. Granted
the recognition of words and morphemes for the language in question,
the average degree of 'isolation' can be expressed as a ratio of the
number of morphemes over the number of words: the lower the
ratio, the more highly isolating is the language (a ratio of i-oo being
characteristic of the 'ideal' isolating language). Average ratios which
have been calculated over a body of continuous text for a number of
languages show that, for instance, English (with a ratio of i-68) is
more 'analytic' than Sanskrit (2-59) or the highly 'synthetic'
Eskimo (3*72). These figures, it must be remembered, give average
ratios over running text. Since a language might be, and frequently is,
relatively isolating with respect to certain classes of words and
relatively synthetic with respect to other classes of words, the ratios
might be quite different if they were calculated over all the words in
the language with each word counted once.
5.3.7 Turkish: an 'agglutinating' language

It is the distinction between agglutinating and inflecting languages
(both types being 'synthetic') which is more interesting for our
present purpose. An agglutinating language is one in which words are
typically composed of a sequence of morphs with each morph
representing one morpheme. Turkish may be taken as an example
which approximates very closely to the 'ideal' of the agglutinating
type. In Turkish the plural morph is {ler}, the possessive morph (' his',
'her', 'its') is {*}, and the 'ablative' morph is {den}. There are many
other morphs which are added to nouns in Turkish; but these will
serve to illustrate the nature of 'agglutination'. Ev ('house') is the
'nominative', singular form; evler ('houses') is 'nominative', plural;
evi is singular, possessive ('his/her house'); evleri ('his/her houses',
'their house(s)') is plural, possessive; evden ('from the house') is
ablative, singular; evler den ('from the houses'); evinden ('from his/
her house') is singular, possessive, ablative; and evlerinden ('from
his/her houses', 'from their house(s)') is plural, possessive, ablative.
(The insertion of n between i and den is automatic and regular.) The
first thing to notice is that each of the three morphs representing,
respectively, the plural, the possessive and the ablative morphemes,
{let}, {} and {den}, preserve their phonological identity and are
immediately recognizable; they are, as it were, simply * stuck on'
('agglutinated') in sequence. Consequently, Turkish words are, in
general, readily segmented into their constituent morphs: ev-ler-
i(n)-den, etc.
A second and no less important feature of Turkish is that in a
particular word each morph represents just one morpheme. These two
features, (i) determinacy with respect to segmentation into morphs
(cf. 5.3.2), and (ii) the one-to-one correspondence between morph
and morpheme, are characteristic of * agglutinating* languages. It
should be noticed, however, that the two features are independent of
one another: as we shall see below, a language may manifest either
one without the other. It should also be observed that the one-to-one
correspondence between morph and morpheme referred to here is to
be understood as holding within a given word: certain Turkish
morphs (including {Y})may represent different morphemes in different
classes of words, in the same way as, for instance, the English morphs
/s/, /z/ and /iz/ represent the present singular morpheme in verbs and
the plural morpheme in nouns. The degree to which languages employ
this kind of multiple representation of morphemes by single morphs
(which is the converse of the representation of one morpheme by
many allomorphs) varies considerably.
5.3.8 Latin: an inflecting language

Let us now take Latin as an example of a language of the 'inflecting'
type. Generally speaking, Latin words cannot be segmented into
morphs; rather, they can only be segmented into morphs at the price
of arbitrariness, inconsistency and the proliferation of allomorphs.
Take, for instance, the words domus ('house'; nominative singular),
dotnl('of the house'; genitive, singular), domum (accusative, singular),
domo ('from the house'; ablative, singular), dotnl ('houses'; nomi-
native, plural), domorum (genitive, plural), domos (accusative, plural),
domls (ablative, plural). We might be tempted to analyse these forms
into dom, on the one hand, and us, 1, urn, 6, drum, os, is, on the other;
and this is what some grammarians do. But let us now bring forward
for comparison another set of words of a very common type: puella
('girl'; nominative, singular), puellae (genitive, singular), puellam
(accusative, singular), puella (ablative, singular), puellae (nominative,
plural), puelldrum (genitive, plural), puellds (accusative, plural),
190 5* GRAMMATICAL UNITS
puellis (ablative, plural). What principle of segmentation should we
follow here? If we try to match the * endings' of the two types as far
as possible, we should doubtless segment the forms into puell and a,
ae, am, a, drum, as, is (with is being the only morph common to the
two types of nouns). But if we segment puella, puellae, etc., in this
way, we should certainly be left with the uncomfortable feeling that,
since a or a is found in all forms of the second type except puellis and
is never to be found in the first set of forms, domus, etc., we ought
perhaps to recognize two allomorphs: puell (which combines only
with is) and puella (which combines with 'zero', in the nominative
singular, and with e, m, a, arum, as: n.b. the long vowel a has now
been resolved into a sequence of two instances of the short a). This is
attractive enough as far as it goes. Similarly, we could recognize two
allomorphs for the first type, dom and domo (with a third 'pseudo-
allomorph', domu, accounted for as a variant of domo which occurs
before consonants), and thus identify the accusative singular morph
(m) and the accusative and genitive plural morphs of both types
(vowel lengthening + s, and vowel lengthening -\-rum). But we should
still be left with a number of allomorphs for the endings of the two
types of nouns: 'zero' and s (or us, or indeed os) for the nominative
singular, etc. And, as anyone who has any acquaintance with Latin
knows, we have yet to take account of the other three regular types of
formation (traditionally called 'declensions'), not to mention the
numerous irregular nouns. What may seem a reasonable procedure
when just two types are compared would no longer seem so when these
other types are brought into the picture. It is no doubt because Latin
(and Greek) words are not readily segmentable into morphs that their
formation was handled in a quite different way by the classical
grammarians. The traditional manner of dealing with the formation
of words in Latin (and Greek) was to classify them into types
('declensions' for nouns and adjectives, 'conjugations' for verbs) and
to set up for each type a table, or 'paradigm', giving all the forms for
one chosen member of the type. It was then left to the person using
the grammar to construct the forms for other members of the type by
reference to the appropriate 'paradigm' (the term 'paradigm' derives
from the Greek word for 'pattern' or 'example'). That is to say, the
classical grammarians did not establish rules, but merely 'patterns',
of formation. Some recent grammars of Latin (and also of Greek)
have kept the traditional 'paradigms', but have superimposed upon
the traditional method of handling 'inflexion' some attempt to seg-
ment words into 'stems' and 'endings'. In doing so, they have
frequently been influenced by historical considerations. According to
the hypotheses of the comparative philologists of the nineteenth
century and their successors a good deal of Latin (and Greek) * in-
flexion* can be plausibly explained as being due to the coalescence
of once distinct morphs. This is largely irrelevant in the synchronic
analysis of the language. We cannot get away from the fact that Latin
(and Greek) words do not lend themselves to segmentation into
morphs.
The impossibility of segmenting Latin words neatly or consistently
into morphs illustrates one feature of the language which makes it
inflecting' (or 'fusional'), rather than * agglutinating*. (It should be
observed that the term employed here is inflecting, not inflexional
inflexion* is used for both 'fusional' and 'agglutinating' languages:
cf. 5.4.2.) The other, more important, feature is the lack of any
correspondence between such segments of the word as we might
recognize and morphemes (assuming that the morpheme is still
defined as 'minimal grammatical unit'). Even if we were to segment
domus, domty etc., into the morph dom (or the allomorphs dom, domo)
and a set of 'endings', us (or s), f, drum (or vowel lengthening +rum)y
etc., we could not say that one part of us (or s) represents {singular}
and another part {nominative}; that part of 1 represents {singular}
and another part {genitive}; and so on. We should have to say that us
(or s) represents simultaneously {singular} and {nominative}; that 1
represents simultaneously either {singular} and {genitive} or {plural}
and {nominative}; etc. The difference between Latin and Turkish in
this respect is striking.
5.3.9 No 'pure' types

It must be realized that the recognition of three 'types' of languages,
by the opposition of 'analytic' to 'synthetic' (which is in any case a
question of degree, as we have seen), and then, within 'synthetic', of
'agglutinating' to 'inflecting' does not imply that any one language
will fall neatly into one 'type'. Still less does it imply (as certain
nineteenth-century linguists maintained) that there is necessarily any
'evolutionary' law which governs the historical development of
languages from one 'type' to another. As Sapir has put it: 'A linguist
who insists on talking about the Latin type... as though it were
necessarily the high-water mark of linguistic development is like the
zoologist that sees in the organic world a huge conspiracy to evolve
the race-horse or the Jersey cow.' The typological classification of
languages into 'inflecting' or 'agglutinating' is only one of many
ways in which languages can be classified according to their structure.
Turkish, though largely' agglutinating', is to some degree' inflectingJ;
and Latin provides instances of' agglutination'. And both languages are
in part analytic. English, we as have seen, is fairly 'analytic', having a
large number of single-morpheme words: man, book, go, tall, good,
etc. As for the 'synthetic' words in English some are 'agglutinative'
(books, taller, etc.); others are 'semi-agglutinative' (or 'semi-in-
flecting') in that they are partly or wholly indeterminate with respect
to segmentation (cf. men, mice, worse, etc.) or contain segments which
represent simultaneously more than one minimal grammatical unit.
As an instance of the latter kind of 'semi-agglutination' consider the
forms /z/, /s/ and /iz/ (orthographically, s and es), which we earlier
regarded as allomorphs of the present singular 'morpheme' {z}. These
endings, which may be reduced to a single underlying 'sibilant'
ending, represent simultaneously both singular and 'third person'
(cf. the verbs jumps, love-s,fish-es,etc.). There are few instances of
fully 'inflecting' words in English (satisfying both the conditions
mentioned above). One such instance is the 'third person singular of
the verb to be\ This form, is, might be segmented into i (an allo-
morph of {be}) and the regular s, but there is little point in doing so,
since i occurs nowhere else as an allomorph of {be}.
There is yet another kind of 'semi-agglutination'. The same morph
may represent different grammatical units either in different positions
of the same word or in different words of the same class. This happens
on a small scale in many languages, including Turkish and English.
For example, in English the same morph (or allomorphs: /s/, /z/,
/iz/) represents 'third person, singular (present)' in verbs and {plural}
in nouns, as we have already seen. And the morph er is found both in
the comparative form of adjectives (tall-er, etc.) and in 'agent' nouns
formed from verbs (run(n)-er, read-er, etc.). This kind of 'semi-
agglutination' (if it may be so called) is what was described above as
the multiple representation of morphemes by single morphs. It is
particularly characteristic of various so-called ' Austronesian' langua-
ges (Sundanese, Tagalog, Malay, etc.).
5.3.10 Inconsistency between theory and practice

It should now be clear that the relationship between the morpheme
and the morph is not a purely grammatical relationship. The words of
Latin can be analysed into their distributional 'factors' as easily as
can the words of Turkish. The difference between the 'inflecting'
and the ' agglutinating' (and the various kinds of * semi-agglutinating')
languages is not therefore one of grammatical structure; it is a
difference in the way in which the minimal grammatical units are
represented in the phonological (or orthographical) form of the word.
It should also be clear, from our discussion of the various ways in
which minimal grammatical units are represented in different
languages, that the distinction of morpheme and morph is forced
upon us, in the first instance, by the fact that not all languages are
'agglutinating' or 'inflecting' and, more particularly, by the fact that
there are languages which do not conform to the 'ideal' of either
'type' (indeed there is probably no purely 'agglutinating' or purely
'inflecting' language). If all 'synthetic' languages were purely
'agglutinating' the morpheme would be a minimal grammatical
unit of constant phonological shape, and we could account directly
for the combination of morphemes into words. On the other hand, if
all ' synthetic' languages were fully inflecting, we should have no use
for the morph (which is by definition a phonological segment of a word
representing a morpheme). It is because many languages are partly
'agglutinating' and partly 'inflecting' that the distinction between
morpheme and morph must be made.
In the case of languages that are mainly ' agglutinating' (and most
languages of the world are said to approximate to the 'agglutinating'
type) the distinction is undoubtedly useful. As we have seen, it
enables us to account for the distribution of grammatically equivalent
words in the same way in the grammar (boys: {boy} + {s}> oxen:
{ox} + {s}, mice: {mouse}+ {s}} sheep: {sheep}+ {s}> etc.) and then, in the
'conversion' of these words into their phonological form, to set up
general rules of 'direct transcription' for the regular, 'agglutinating'
forms and special, more complex rules (of the kind required for
'inflecting' languages) for the conversion of the irregular, non-
agglutinating forms.
However, with regard to languages that are mainly 'inflecting', we
must now face a problem which has so far only been implicit in our
discussion of this 'type'. We have observed that such segments
(granted their recognition as segments) as a (or 'zero') inpuella, and
us (or s) in domus do not represent minimal grammatical units, such as
{singular} or {nominative}, but the combination {singular + nomi-
native}. On our definition of the morpheme as the minimal gram-
matical unit (which is the definition to which most linguists, in
theory, adhere) it is such elements as {singular} and {nominative}
which are morphemes. Yet most linguists who have dealt with the
morphemic analysis of Latin have regarded a and us as allomorphs
of 'the nominative singular morpheme*. This is the inconsistency
between theory and practice to which we referred at the beginning of
the present section. The reason for the inconsistency is doubtless the
historical one that the morpheme was first established for those
languages in which the formation of words could be described in
terms of the combination of phonologically constant segments (this is
the sense in which the concept was employed by the Sanskrit
grammarians); later the morpheme was defined as a more 'abstract',
distributional unit, and this led to the distinction of morpheme and
morph. In Latin there is a conflict between these two conceptions of
the morpheme. If we define the morpheme as the minimal gram-
matical unit, then it cannot simultaneously be a unit functioning in
the formation of words in Latin; if on the other hand we define it as a
formational segment of the word, then it is not a minimal grammatical
unit in Latin (or a universal concept of linguistic theory).
5.4 The word
5.4.1 Morphology and syntax

The word is the unit par excellence of traditional grammatical theory.
It is the basis of the distinction which is frequently drawn between
morphology and syntax and it is the principal unit of lexicography (or
' dictionary-making').
According to a common formulation of the distinction between
morphology and syntax, morphology deals with the internal structure
of words and syntax with the rules governing their combination in
sentences. The very terms 'morphology* and 'syntax*, and the way in
which they are applied, imply the primacy of the word. Etymologically
speaking, * morphology * is simply ' the study of forms * and ' syntax *
the theory of ' putting together *: it was taken for granted by tradi-
tional grammarians that the ' forms * treated in grammar are the forms
of words and that words are the units which are 'put together*, or
combined in sentences. In older books on language, the distinction
between morphology and syntax is sometimes represented in terms
of a distinction between 'form* and 'function*. This also rests upon
the implied primacy of the word: according to their 'function* in the
sentence, which is accounted for by the rules of syntax (with reference
T H E
54- WORD 195
to such notions as * subject', * object', * complement', etc.), words are
said to assume a different 'form', and the different 'forms' are
handled by morphology.
5.4.2 Inflexion and derivation

Although the term 'morphology' is now sufficiently well-established
to be called 'traditional', it is not in fact a term employed in classical
grammar. The term which opposes 'syntax' in classical grammar is
'inflexion'. The standard reference grammars of Greek and Latin,
and the grammars of modern languages which are based on classical
principles, are generally divided into three sections, not two: namely,
into inflexion (or 'accidence'), derivation (or 'word-formation'), and
syntax. But these three sections are not regarded as being of equal
importance to the grammarian. Whereas some hundreds of pages may
be devoted to inflexion and syntax, there will usually be not more than
half-a-dozen or so pages on derivation. The reason for this dis-
proportionate treatment is that there is really no place in classical
grammar for derivation. Inflexion is defined in classical grammatical
theory somewhat as follows: inflexion is a change made in the form
of a word to express its relation to other words in the sentence. And
in the grammars of particular languages the section on inflexion will
describe the ' declensions' of nouns, adjectives, and pronouns, and the
' conjugations' of verbs, according to selected models of formation, or
'paradigms'. The section on derivation will list various processes
whereby new words are formed from existing words (or * roots'):
adjectives from nouns {seasonal from season), nouns from verbs
{singer from sing), adjectives from verbs {acceptable from accept), and
so on. That any space at all is allotted to derivation is an awkward,
and theoretically inconsistent, gesture in recognition of the fact that
some words can be further analysed into components even if the rules
in the main body of the grammar have nothing to say about the
function or distribution of these components.
It was in the nineteenth century that the term 'morphology* was
introduced into linguistics to cover both inflexion and derivation.
(The term itself appears to have been invented by Goethe and to have
been first applied, in biology, to the study of the 'forms' of living
organisms; as we saw in the first chapter, from the middle of the
nineteenth century linguistics was very much influenced by evolu-
tionary biology.) The reason why a section on derivation wTas grafted
on to grammars of the western classical languages at this period was
that the comparative philologists, to a considerable extent influenced
by the Sanskrit grammatical treatises now available to them, had
become interested in the systematic study of the formation of words
from a historical point of view. And it was realized that inflexional
and derivational processes had a good deal in common.
But in classical grammar the distinction between inflexion and
derivation is absolutely fundamental. Whereas singing is but a form
of the word sing, syntactically determined, singer is a different word
with its own set of forms, or * paradigm' (cf. 1.2.3). The fact that from
the point of view of their formation both singing and singer can be
regarded as being composed of a 'root' (or 'stem'), sing, and a suffix,
ing or er (the one process of formation being hardly less productive
than the other in English) is obscured by the difference of treatment
imposed by the presuppositions of classical grammatical theory.
Standard dictionaries of English (and of most other languages), which
are based upon the assumptions of classical grammar, list derivational
forms as distinct words, but not the regular inflexional forms, which
can be constructed by reference to the ' paradigms' set out in a con-
ventional grammar of the language. In later chapters it will become
clear that much of what is traditionally referred to as derivation can
be, and ought to be, integrated with the syntactical rules of English
in a generative grammar of the language. For the present, however, we
may leave the question of derivation and look more closely at the
concept of the 'word*.
5.4.3 Ambiguity of the term 'word'

The term 'word' has been used in the preceding paragraphs in three
quite different senses. The first two senses are readily distinguished
in terms of the notion of 'realization' (cf. 2.2.11). Just as we must
distinguish between the morph as the phonological (or orthographical)
representation of the morpheme, so we must distinguish between
phonological (or orthographical) words and the grammatical words
which they represent. For example, the phonological word /saeg/ and
the corresponding orthographic word sang represent a particular
grammatical word, which is traditionally referred to as 'the past
tense of sing9; whereas the phonological word /kAt/ and the corres-
ponding orthographic word cut represent three different grammatical
words: ' the present tense of cut', ' the past tense of cut', and the ' past
participle of cut*. It has already been mentioned that phonological
and orthographic words in English are generally in one-to-one
T H E
54- WORD 197
correspondence with one another in the sense that they represent
the same set of (one or more) grammatical words (cf. the examples just
given). But there are some instances of (a) one-many or (b) many-one
correspondence between phonological and grammatical words: cf.
(a) /poustman/: postman, postmen; /miit/: meaty meet, etc. (b) /ri:d/,
/red/: read ('the present tense of read\ 'the past tense of read*\
n.b. /red/ is also in correspondence with the orthographic word red,
and /ri:d/ with the orthographic word reed). Many other languages
besides English, whose spelling conventions are popularly said to be
only partly * phonetic' (in a non-technical usage of the term * phonetic'),
provide similar examples of one-many or many-one correspondence
between phonological and orthographic words (cf. 1.4.2).
5.4.4 Word and 'lexeme*

There is a third, more 'abstract', usage of the term 'word'. It was
this usage that we employed above when we said, for instance, that in
traditional grammar * whereas singing is but a form of the word sing...
singer is a different word, with its own set of forms, or " paradigm " ' ;
and the same, more 'abstract' sense was implicit in our reference to
sang as 'the past tense of sing9. Modern linguists have tended to
neglect, or even to condemn, this more 'abstract' usage. Bloomfield,
for example, says that the school tradition is ' inaccurate' in referring
to units like book, books, or do, does, did, done, as ' different forms of
the same word'. But it is Bloomfield himself who is here guilty of
inaccuracy. It is up to us to decide which way we wish to define the
term 'word'. The important thing is to keep the three senses apart.
Modern linguists have not always done this consistently and as a
result they have frequently misinterpreted traditional grammatical
theory. It is, of course, the more ' abstract' sense that the term 'word'
bears in classical grammar. However, since most linguists now employ
the term 'word' to refer to such phonological or orthographic units
as /saerj/ or sang, on the one hand, or to the grammatical units they
represent, on the other (and indeed do not always distinguish even
between these two senses), we shall introduce another term, lexeme, to
denote the more 'abstract' units which occur in different inflexional
' forms' according to the syntactic rules involved in the generation of
sentences. Notationally, lexemes will be distinguished from words by
the use of capitals. Thus the orthographic word cut represents three
different inflexional 'forms' (i.e. three different grammatical words)
of the lexeme CUT.
In later sections of this book, we shall use the term * lexeme' only
when it might not be clear from the context which sense of * word' is
intended. It is unfortunate that modern linguistics has not followed
the traditional practice of defining the word to be the more 'abstract*
unit.
5.4.5 'Accidence'
It is worth while dwelling for a moment on the implications of the
term 'form* as it is employed in traditional grammar. We have already
seen that the opposition drawn by de Saussure between ' substance'
and ' form' is to be distinguished from the Aristotelian and scholastic
opposition of these terms. We shall not go into the details of Aristo-
telian metaphysics with its rich terminology of distinctions ('sub-
stance', 'matter', 'form', 'essence', 'existence', etc.), but it must be
realized that classical grammar rests upon metaphysical assumptions
of a more or less Aristotelian kind. In particular, it presupposes the
distinction between the ' essential' and the ' accidental' properties of
an object. For instance, it is part of the 'essence' of man (let us say)
to be intelligent and to have two legs; whereas it is ' accidental' that
particular men should have red hair or blue eyes. Similarly, the words
that occur in sentences have particular 'accidental' properties
(e.g. nouns are either singular or plural, verbs are in the present, past
or future tense, and so on); and the grammarian's ' paradigms' (and
lists of irregularities) describe the ' forms' of words of different classes.
Lexemes (the 'words' of traditional grammar) are the underlying
invariant units considered in abstraction from their 'accidental'
properties: lexemes are 'substances' which occur in various 'acci-
dental' 'forms'. It is this conception, it will be observed, which
underlies Roger Bacon's view quoted above (1.2.7): 'Grammar is
substantially the same in all languages, even though it may vary
accidentally.' And it is this conception which explains the traditional
term 'accidence' for what we are calling 'inflexion': strictly speaking,
words are 'inflected' (i.e. they vary their 'form') according to their
'accidence' (from the Latin accidentia, 'accidental properties'). And
the classical grammarians assumed that the ' accidental' properties of
lexemes, like the 'accidental' properties of everything else in the
universe, could be classified under a restricted set of 'categories'
(cf. 7.1.1). A more extended treatment of the basis of classical
grammar than can be given in an introductory work of this kind
would show that it was not always possible to apply the Aristotelian
T H E
54* WORD 199
system consistently to the description of language. The present brief

account will be sufficient for the elucidation of the traditional concept
of the * word'. And, as we shall see, the traditional view of the 'word*
(or lexeme), when stripped of its metaphysical implications, merits a
more sympathetic consideration than it has received from most
modern linguists.
5.4.6 Orthographic words

We may now take up the question of the word and its status in
general linguistic theory. We have already seen that it is the word,
rather than the lexeme, which has engaged the interest of most modern
linguists, although they have not generally distinguished between these
concepts. With the habits of reading and writing drilled into us at
school and sustained by continuous practice thereafter, it is perhaps
difficult for us to think of utterances except as being composed of
words. But the ability to break utterances up into words is not only
characteristic of educated and literate speakers of a language. Sapir
tells us that uneducated American-Indian speakers, with no experience
of writing any language at all, when asked to do so, were perfectly
capable of dictating to him texts in their own language 'word by
word', and had little difficulty in isolating words from utterances and
repeating them to him as units. Whatever else we may say about the
word as a linguistic unit, we must reject the view which has sometimes
been advanced that 'primitive languages' do not have words. The
habit of reading and writing, especially in a complex, industrialized
society like our own which is founded on literacy, may well reinforce
the native speaker's consciousness of the word as an element of his
language (and may also maintain certain inconsistencies: cf. all right,
altogether); but it certainly does not create his ability to break
utterances up into words in the first place.
5.4.7 ' Potential pause'

It has been suggested that the word should be defined as ' any segment
of a sentence bounded by successive points at which pausing is
possible9. Even if we grant that this criterion of potential pause will in
fact segment utterances into units which we should wish to regard as
words, it should be considered as a procedural help to the linguist
working with informants, and not as a theoretical definition. The fact
is that speakers do not normally pause between words. Since the
native speaker is able to actualize the * potential pauses' in his
utterances when he wishes to, even though he does not do this
normally, it follows that the words must be identifiable as units in
his language under the normal conditions in which he uses it. It is
this functional unity of the word in the language as it normally
operates that we must try to capture in our definition; and it is
presumably the native speaker's consciousness of the word as a
functional unit which also lies behind the recognition of the word in
most systems of orthography.
5.4.8 Semantic definition of the word

One well-known definition of the word runs as follows: ' A word may
be defined as the union of a particular meaning with a particular
complex of sounds capable of a particular grammatical employment/
This definition, it will be observed, makes it a necessary condition
that the word should be simultaneously a semantic, a phonological
and a grammatical unit. Now it may well be true that all the units
which we wish to regard as words in the description of a given
language satisfy these three conditions. But they are certainly not the
only units to satisfy them. Entire phrases, like the new book, have
a definite meaning, a definite phonological shape and a definite
grammatical employment. And so do distributionally limited segments
of even higher rank. Some linguists have suggested that the definition
can be made satisfactory by saying that words are the smallest
segments of utterances which fulfil the three conditions. But this will
not do either. The un and the able of unacceptable satisfy the three
criteria. Yet they would not generally be regarded as words. Moreover,
the word unacceptable is more or less synonymous with the phrase
not acceptable] both can be analysed into three meaningful units (and
three grammatical units: three morphemes each represented by one
morph). We must conclude that semantic considerations are irrelevant
in the definition of the word, as in the definition of other grammatical
units. As for the phonological criterion: although, as we shall see later,
words are phonologically delimited in various ways in many languages,
such phonological features are never more than secondary correla-
tions. We shall therefore concentrate upon defining the word in
purely grammatical terms.
To simplify the discussion, for the present we will continue to
assume, with the majority of linguists, that in all languages the
morpheme is the minimum unit of grammatical analysis. The
THE
54' WORD 201
question we have set ourselves therefore is this: how shall we define a
unit intermediate in rank between the morpheme and the sentence
and one which will correspond fairly closely with our intuitive ideas of
what is a 'word', these intuitive ideas being supported, in general, by
the conventions of the orthographic tradition?
5.4.9 * Minimal free form'

The first answer to be considered is Bloomfield's, which is probably
the most famous of all modern attempts to define the word in general
terms applicable to all languages. According to Bloomfield the word
is ' a minimum free form'. This depends upon the prior distinction of
'free' and 'bound' forms in the following sense: forms which never
occur alone as whole utterances (in some normal situation) are bound
forms; forms which may occur alone as utterances are free forms. Any
free form no part of which is itself a free form is, by Bloomfield's
definition, a word. It will be evident to the reader that this definition
applies, in so far as it does apply, to phonological words rather than
grammatical words. Bloomfield did not distinguish clearly between
these two concepts.
There is no doubt that Bloomfield's definition will cover a good
number of the forms in different languages that we should wish to
recognize as words. And it would account satisfactorily for the native
speaker's ability to actualize the 'potential pauses' between such
forms as are covered by the definition when they occur in the stream
of speech. But, as Bloomfield himself pointed out, the definition will
not capture certain forms that have traditionally been considered as
independent words: e.g. the or a in English. Such forms are not likely
to occur as whole utterances in any normal situation of language use.
(Clearly the word the might be said as a whole utterance in reply to
someone doing a crossword-puzzle and asking for ' a three-letter word
beginning with th'; or in reply to the inquiry 'Did you say " a " or
"the"}' In all such contexts linguistic forms are being 'mentioned'
rather than 'used': and in contexts of 'mention' linguistic units of
every rank and level can occur as whole utterances.) Bloomfield is, of
course, aware of this difficulty; but, instead of making a stand on his
primary criterion and saying that, despite their traditional classifica-
tion, such forms as the and a are not in fact words, he introduces a
supplementary criterion of 'parallelism' with forms which are
classified as words by the first criterion of freedom of occurrence.
The and a occur in the same environments in sentences as this and that
202 5' GRAMMATICAL UNITS
(cf. the many a man, this man, that man, etc.); this and that may occur
as minimal free forms, and so are classified as words; therefore the
and a are words.
Although Bloomfield's definition of the word in terms of 'freedom*
and 'bondage' has been accepted by many eminent linguists, it can
hardly be regarded as satisfactory. We must not lose sight of the
primary object of grammatical description: to generate sentences from
which can be derived the utterances and potential utterances of the
language being described. All questions of classification must be sub-
ordinated to this purpose.We may assume that, in general, sentences
are composed of many morphemes; and that the word is a unit,
'below* the rank of sentence, composed, typically, of a number of
morphemes. But to call a particular 'complex* of morphemes a 'unit*
implies that these morphemes are in greater 'cohesion* than other
groupings of morphemes in the sentence which are not recognized as
words; that sentences can be generated more satisfactorily by taking
into account (at least) two different principles of composition, one of
which determines the combination of morphemes into 'complexes'
(which we will call grammatical words) and the other the combination
of words into sentences. Now, as a matter of empirical fact it may be
true that the set of 'minimal free forms* will generally correspond in
all languages to the set of phonological units representing gram-
matical words; but, if so, this fact presumably depends upon and
reflects the structural 'cohesion' of the word in sentences, and is of
only indirect concern to the grammarian. Like the criterion of
'potential pause*, that of occurrence as a 'minimum free form* is at
best a procedural aid to the linguist working with informants. Let us
now look a little more closely therefore at some of the features
involved in what has been referred to as the 'cohesion' of the word
as a grammatical unit.
5.4.10 ' Internal cohesion' of the word

The grammatical 'cohesion' of the word (regarded as a combination
of morphemes) is commonly discussed in terms of two criteria:
'positional mobility' and 'uninterruptability'. The first of these may
be illustrated with reference to the following sentence, which we have
segmented into morphemes (or morphs):
the-boy-s-walk-ed-slow-ly-up-the-hill
1 2 3 4 5 6 7 8 9 10
T H E
5-4- WORD 203
The sentence may be regarded as a combination of ten morphemes,
which occur in a particular order relative to one another. However,
various permutations are possible, if we start from the order in which
the morphemes occur in the sentence cited above: slow-ly-the-boy-s-
walk-ed-up-the-hill, up-the-hill-slowl-y-walk-ed-the-boy-s> etc. Sub-
stituting the numbers for these two other possibilities, we get (in
place of 1 2 3 4 5 6 7 8 9 10):
6 7 1 2 3 4 5 8 9 10
8 9 10 6 7 4 5 1 2 3
There are other possible permutations which will yield an acceptable
English sentence. The point is, however, that under all the permuta-
tions certain pairs or triples of morphemes will behave as * blocks',
not only occurring always together, but also in the same order
relative to one another: there is no possibility of the sequence 3 2 1
(*s-boy-the), or 5 4 (*ed-walk). One of the characteristics of the word
is that it tends to be internally stable (in terms of the order of the
component morphemes), but positionally mobile (permutable with
other words in the same sentence). Clearly, this characteristic is far
more striking in languages with a 'free word-order* (cf. 2.3.5).
It is worth pointing out that positional mobility and internal
stability are independent of one another. Suppose, for example, we
found a language in which the order of words was fixed, but the order
of morphemes within words freely subject to permutation. This may
be indicated symbolically as follows (taking A> B and C to be words):
A C
I 2 3 4 5 6 8
7 9 10
2 3 i 6 4 5 IO 9 8 7
3 2 i 5 6 4 9 7 8 10
2 I 3 4 5 6 9 io 7 8
Here, it is the order of the intermediate units relative to one another
which is 'fixed', by contrast with the 'freedom' of order within the
intermediate units (one might say that the sentence is 'internally
stable' at the word-rank, whereas the morphemes are 'positionally
mobile' within words). To illustrate from English: this situation
would hold if, for example, not only Tke-girl-s have-be-en-eat-ing
apples were acceptable, but also *Girl-the-s en-have-be-eat-ing
s-apple> etc. (We have here taken the girls to be one word, and also
have been eating: in terms of the present criterion, this is correct.)
So far (to the best of my knowledge) no language has been found
which manifests this particular feature. As we have already seen,
'free* order tends to be found, if at all, at the higher ranks (cf. 2.3.6).
This is an empirical fact about the structure of language. The con-
verse situation, which we have just envisaged, is not only logically
conceivable, but it would define the word as a structural unit of
language no less clearly than the common phenomenon of relatively
'free' word-order.
But we have said that positional mobility and internal stability are
independent of one another. Once again we may illustrate from
English. The criterion of positional mobility would fail to define the
'definite article', the, as a word: it cannot be moved from one place
in the sentence to another independently of the noun it 'modifies'. In
this respect, it is like the so-called 'postpositive' articles of Swedish,
Norwegian, Rumanian, Bulgarian, Macedonian, etc. ('postpositive'
simply means 'following', rather than 'preceding': cf. 'postposition'
v. 'preposition', 7.4.7): e.g. Rumanian lup 'wolf: lupul 'the wolf,
Macedonian grad 'city': gradot 'the city'. It is the criterion of
'interruptability' (or 'insertability') which distinguishes the English
article as more 'word-like' than the Rumanian or Macedonian article.
It is possible to 'interrupt' (or 'insert' within) the sequence the-boy,
whereas it is not possible to 'interrupt' the sequence grad-ot or
hip-ul: by 'interruptability' is meant the possibility of inserting other
elements, more or less freely, between the morphemes or 'blocks' of
morphemes. For example, between the and boy one may insert a
whole sequence of other elements: the big strong strapping boy, etc.
The fact that one criterion, but not the other, applies to the
English article implies that, even if it is taken to be a word, it is not
so 'fully' a word as other elements to which all the relevant criteria
apply. There are many marginal cases in various languages which
have been much discussed in the literature.
5.4.11 Phonological correlations

In many languages the word is phonologically marked in some way.
For instance, a great number of languages have what is called a word-
accent: in such languages every word is 'accented' (this may be a
matter of stress or pitch, or both) on one and only one syllable. The
accent may be 'free' (as in English or Russian), in the sense that the
syllable on which it falls is not generally determined by the phono-
logical structure of the word or its grammatical classification; or
' restricted' (the accent is restricted to one of the last three syllables
T H E
54- WORD 205
in both Classical and Modern Greek); or 'fixed', with reference to the
beginning or end of the word or to some other feature. Examples of
well-known languages with a fixed accent are: Latin, where the place
of the accent is generally determined by the length of the penultimate
syllable; Polish, where the accent (generally) occurs on the next to the
last syllable; Turkish, where it (generally) occurs on the last syllable;
and Czech, where it falls on the initial syllable of the word. To mention
just two other ways in which words may be phonologically delimited:
' vowel-harmony' in Turkish and Hungarian operates throughout
words, but not beyond (cf. 3.3.13); and in many languages a more
restricted set of phonological units occur at the beginning or end of
words than in other positions. For all languages with a word-accent
it is true (in general) that there will be the same number of words in
an utterance as there are accents; and in the case of languages with a
fixed accent the boundaries between words may be determined by
reference to the syllables on which the accent falls. However, the
very fact that we can say that there are exceptions to the general rules
governing the place of the accent (that, for instance, the Russian word
ne, 'not', is never stressed; that the reflexive forms of the verb in
Polish have the accent on the same syllable as in the corresponding
unreflexive forms; that such Turkish forms as gitmiyorduy l he was not
going', or askerken 'when (he was) a soldier', are words despite the
place of the accent and their partial violation of the rules of ' vowel-
harmony') shows that the accent is not the primary defining feature of
the word in these languages. We could hardly determine the place of
the Polish or Turkish accent with respect to the beginning or end of
the word, if the word-boundaries themselves were determined solely
by reference to the place of the accent! The partial' congruence' of the
phonological and grammatical levels by virtue of the double status of
the word as both a grammatical and phonological unit is therefore a
common, though not a universal, feature of languages. There are
languages (a notable instance is French) where whatever congruence
there is between phonological and grammatical structure seems to
hold over units of higher rank than the word.
5.4.12 Independence of criteria

For a fuller account of the various features by means of which the
grammatical ' cohesion' of the word may be established, and also of
the different ways in which words may be phonologically marked in
different languages, reference may be made to the works cited for this
2O6 5. GRAMMATICAL UNITS
section in the bibliographical notes. Enough has been said in the
preceding paragraphs to show that the grammatical criteria are inde-
pendent of one another, and that phonological criteria are not only
independent of one another, but necessarily subordinate to the gram-
matical. It follows from these facts that what we call ' words' in one
language may be units of a different kind from the ' words' of another
language; and yet the application of the term 'word* is not entirely
arbitrary, since the relevant features whereby words are established
for different languages all tend to support their identification as
structural units.
Since the criteria for the establishment of words apply not only
independently of one another, but also independently of the criteria
whereby morphemes are defined as minimal grammatical units, in
certain languages the same units may be simultaneously both words
and morphemes. For instance, in English the morphs /nais/, /boi/,
/wont/ (orthographically, nice, boy, want) represent simultaneously
the morphemes 'nice', 'boy', and 'want* and (on the assumption
that this is in fact the correct analysis of these forms) grammatical
words composed each of one morpheme. As we have already seen
(5.3.6), a one-to-one ratio of morphemes to words is the defining
characteristic of 'isolating* languages.
5.5 The notion of 'rank'

5.5.1 'Rank* is a surface-structure notion
In this chapter we have given a very brief account of the primary
units of grammatical analysis and the way in which the terms applying
to them have been defined in modern linguistics. As we have seen, the
sentence and the word (in the sense of 'lexeme': cf. 5.4.4) were the
main units with which traditional grammar operated, the phrase and
the clause being defined, not always consistently, in terms of them.
The morpheme has replaced the word as the minimal unit of
grammatical analysis in many modern treatments of linguistic
theory; but few linguists have adhered in practice to a theoretically
consistent definition of the morpheme.
At the beginning of the chapter, we said that the relationship
between sentences, clauses, phrases, words and morphemes was one
which could be expressed by saying that a unit of ' higher' rank was
composed out of units of' lower' rank. The term' rank' has been taken
from Halliday; other linguists use 'level* (e.g. Pike) or 'stratum'
5.5. THE NOTION OF 'RANK* 207
(e.g. Lamb) in roughly the same sense. Various attempts have been
made to formulate a consistent theory of grammatical structure based
on the notion of 'composition' (notably by the three authors men-
tioned in the previous sentence). We shall not discuss these theories
here, but simply refer the reader to the works cited in the notes.
In the next chapter, we shall adopt a transformational approach to
grammatical description. One consequence of this is that we shall be
led to the distinction of what will be referred to as the ' deep' and
the 'surface* structure of sentences. It will then appear that such
distinctions as can be drawn consistently between units of different
rank (and it is undeniable that sentences have a grammatical structure
that can be described appropriately in such terms) are drawn with
respect to the surface structure.
5.5.2 An illustration
There is no reason to believe, or to make it a theoretical requirement,
that the rules of the grammar must be organized in such a way that
one set of rules (say) generates words out of morphemes, that another
set of rules then combines these words into phrases, that a third set
of rules generates clauses out of phrases, andfinallythat a fourth set of
rules generates sentences out of clauses. Within a transformational
framework, much of the importance that has been attached in the past
to the definition of units of different ranks disappears. We are no
longer committed to the necessity of saying that every sentence must
be analysable without residue into units of one rank or another.
To give just one example at this point. There is a large class of
adjectives in English, exemplified by red-haired, blue-eyed, one-legged,
etc., which quite clearly 'contain' three morphemes (as the term
'morpheme' is generally applied): e.g. {red} + {hair} + {ed}. In each
case, two of the constituent morphemes may be regarded as words,
e.g. red and hair. They are free forms, and they satisfy the various
other criteria that have been proposed for the definition of the word
(cf. 5.4.6 ff.). From one point of view, it might be desirable to regard
red-haired as one word: in particular, for the purpose of integrating
the grammar and phonology in the determination of the distribution
of stress. (Most English words have one primary stress; and the
'complex' adjectives we are discussing would seem to satisfy this
criterion.) On the other hand, red-haired 'contains' two words and a
bound morpheme. Furthermore, there would seem to be some sense
in which the bound morpheme {ed} is ' added' to the phrase red hair:
208 5. GRAMMATICAL UNITS
there is no form ^haired, which we can combine at a higher rank with
red. Problems of this kind cease to be worrying if we accept that
adjectives like red-haired can be generated by the grammar in a
sequence of operations which are not constrained by the requirements
of linguistic theory to follow the principle that units of ' higher' rank
are composed of an integral number of units of ' lowerJ rank.
With these remarks, we can embark upon our further discussion of
grammatical structure. We shall continue from the point we reached
in the previous chapter. But we shall no longer maintain the view that
the vocabulary of a language is a set of words, listed in the lexicon,
which are combined into sentences by means of rules operating
directly upon word-classes.
John Lyons
Online ISBN: 9781139165570
Chapter
6 - Grammatical Structure pp. 209-269

6
GRAMMATICAL STRUCTURE
6.1 Immediate constituents

6.1.1 Concatenation and linearity
In our treatment of the general principles of formal grammar* in
chapter 4, we deliberately adopted the view that all sentences had a
simple linear structure: i.e. that every sentence of the language could
be satisfactorily described, from the grammatical point of view, as a
string (or sequence) of constituents (which we assumed to be words).
As an abstract illustration of what is meant by the term * string'
(which is the technical term used in mathematical treatments of the
grammatical structure of language) we may consider the following
instances:
a+b+c+d
The plus-sign is employed here (other conventions of notation are
also to be found in the literature) to indicate concatenation (' chaining
together*). The string results from the combination of the consti-
tuents, or elements, in a particular order. What the order denotes
depends upon the interpretation given to the system in its application
to particular phenomena. In the case of natural languages, the left-to-
right ordering of the constituents in the string may be thought of as
reflecting the time-sequence (from earlier to later) in spoken utterances
or the left-to-right ordering of written sentences in the conventions
used for English and most languages of the world today. At the same
time, it should be realized that the same abstract principle of linear
ordering might also be used for other purposes in the description of
language. Indeed, there is no reason why linear ordering should not
be interpretable in different ways in different parts of a grammar. We
tacitly assumed, in chapter 4, that the combination of words resulting
from the application of a grammatical rule constituted a string, with
the order of the concatenated words being determined by the order
in which the words occur in sentences of the language.
It was pointed out in chapter 2 (cf. 2.3.5) t n a t t n e notion of syntag-
matic relationship did not necessarily presuppose an ordering of the
[209]
210 6. GRAMMATICAL STRUCTURE
elements between which the relationship held: and instances were
given of both sequential and non-sequential combinations of elements.
It is important to realize that a string is a particular kind of syntagm,
as concatenation is a particular kind of combination. If the data
suggested that we should, we might formalize the theory of gram-
matical structure in terms of a non-concatenating system of rules
which generated not strings of elements, but unordered sets (between
the members of which there held certain relations of syntagmatic pre-
supposition, dependency, etc.). For example, a rewrite-rule (cf. 4.3.2)
of the form
might be interpreted (the plus-sign now no longer indicating con-

catenation) to mean solely that X is a syntagm composed of the un-
ordered combination of elements a, 6, c and d. Although most of the
discussion in this chapter will take for granted the restriction to
concatenating systems of rewrite-rules, the point should be borne in
mind. Some of the rules suggested in a later chapter will be non-
concatenating. But before we continue with the more formal dis-
cussion, we must say something about the historical background.
6.1.2 Immediate constituents

Most modern textbooks of linguistics attach great importance to what
is called immediate constituent analysis. The term 'immediate con-
stituent' was introduced by Bloomfield (1933), as follows: 'Any
English-speaking person who concerns himself with this matter, is
sure to tell us that the immediate constituents of Poor John ran away
are the two forms poor John and ran away; that each of these is, in
turn, a complex form; that the immediate constituents of ran away
are ran... and away...; and that the constituents of poor John are
.. .poor and John.'
There is an obvious parallelism between immediate constituent
analysis and the traditional procedure of 'parsing' sentences into
* subject' and 'predicate', and each of these, where appropriate, into
words, phrases and clauses of various types. Bloomfield's sentence
might be described as a simple sentence whose subject is a noun-
phrase, made up of the noun John 'modified' by the adjective poor,
and whose predicate is a verb-phrase, consisting of the verb ran
'modified' by the adverb away. Underlying both approaches to
grammatical analysis is the view that sentences are not just linear
6.1. IMMEDIATE CONSTITUENTS 211
sequences of elements, but are made up of 'layers' of immediate
constituents, each lower-level constituent being part of a higher-level
constituent. The analysis of a sentence into its several * layers' of
constituents can be represented graphically in a number of ways.
We may use brackets: [{Poor John) {ran away)]. Or we may construct
a tree-diagram (cf. Fig. 8).
Poor John ran away

Fig. 8. Bloomfield's immediate constituents.
These two methods of representation are equivalent. (The symbols

xy y and z are employed here merely for convenience of reference to
the diagram.) The tree-diagram given above is to be interpreted as
follows: the ultimate constituents of the sentence (the elements out of
which the sentence is constructed) are poor, John, ran and away; the
words poor and John are the immediate constituents of one con-
struction, poor John, so the branches leading to them derive directly
from one 'node* {y); the words ran and away are the immediate
constituents of another construction, being related through the next-
highest ' node' common to them both {z); and the two constructions
poor John and ran away are the immediate constituents of the highest-
level construction, the sentence itself, so they both derive directly
from the ' node' x. It will be observed that neither in the representa-
tion of the constituent structure of the sentence by means of brackets
nor in the equivalent tree-diagram have we incorporated the informa-
tion that poor is an adjective, that poor John is a noun-phrase, etc.
Nor have we taken account of the fact that poor John is 'subject* and
ran azvay is 'predicate' or of the notion of 'modification'. In these
respects our analysis of the sentence into its constituents differs from,
and so far is' poorer' than, the analysis that would be given in terms of
the categories of traditional grammar.
One can distinguish three periods of development in the theory of
constituent structure. Bloomfield himself did little more than intro-
duce the notion and explain it by means of examples. He spoke of a
' proper analysis' of the sentence into constituents as ' one which takes
account of the meanings'. His followers, notably Wells and Harris,
formulated the principles of constituent analysis in greater detail and
replaced Bloomfield's somewhat vague reference to 'taking account
of the meanings' with explicitly distributional criteria. Finally, in the
last few years, the theory of constituent-structure has been formalized
and subjected to mathematical study by Chomsky and other scholars,
who have given considerable attention to the nature of the rules
required to generate sentences with the appropriate constituent-
structure.
6.1.3 Grammatical ambiguity

One of the most cogent reasons for recognizing that sentences, and
parts of sentences, have a non-linear constituent-structure is that it
enables us to handle a large class of ambiguities at the grammatical
level of description. If we compare the following three examples, we
shall see that ambiguity may be a function either of the distributional
classification of the elements or of the constituent structure, or of both
together:
(a) They can fish
(b) Beautiful girVs dress
(c) Some more convincing evidence
In the case of the first example, (a) They can fish (cf. They could fish,
They mayfish,on the one hand, and They cannedfish,They eatfish,on
the other), the ambiguity is accounted for by the double classification
of both can (as a modal auxiliary or a transitive verb) and fish (as an
intransitive verb or a noun). Since the verb-phrase in an English
sentence may consist of a transitive verb and a noun or of an intransi-
tive verb with a preceding auxiliary (this is of course an over-simplified
account), it follows that They canfishis analysable in two ways. There
is, however, no difference of bracketing; under both interpretations
the sentence is to be bracketed as They (can fish). By contrast, the
ambiguity of (b) beautiful girl's dress is to be accounted for, not in terms
of the distributional classification of the elements (beautiful, girl, dress),
but in terms of a difference of constituent-structure; under one inter-
pretation, 'girPs dress which is beautiful', the words girVs and dress
form a constituent; under the other, * dress of (a) beautiful girl', it is
beautiful and girl that are bracketed together to form a constituent.
The third example combines both factors: the phrase (c) some more
convincing evidence may be paraphrased, unambiguously, as either
(i) some evidence which is more convincing, or (ii) some more evidence
which is convincing; in other words, it can be bracketed into consti-
tuents as either (i) some [(more convincing) evidence], or (ii) [(Some more)
(convincing evidence)]. But correlated with this difference of con-
stituent-structure there is also a difference in the distributional
classification of the elements. This becomes clear, if we substitute less
for more and good (or bad, etc.) for convincing: some less convincing
evidence is not ambiguous, nor is some more good evidence, nor is some
better evidence, nor is some more evidence. The word more belongs to
(at least) two distributional classes: (i) like less, it combines with
adjectives to form adjectival-phrases (however, its distribution is more
restricted than that of less, since more is here in complementary
distribution with the suffix -er: cf. nicer v. *more nice, etc.), and
(ii) unlike less, it combines with a preceding some to form a * modifier'
of nouns and noun-phrases (cf. some more evidence v. msome less
evidence).
Ambiguity may be a function then either of constituent-structure or
of the distributional classification of the ultimate (and intermediate)
constituents; and this is the case not only for English, but for many
other languages as well. We must therefore take account of both
factors in the analysis of sentences. This can be done easily enough
by labelling the bracketed structures or the 'nodes' of the tree-
diagram. Thus:
"L{NP(A [poor] + N [John]) + VP( V [ran] + Adv [away])}
There is no distinction associated with the different kinds of brackets
used here. The employment of different brackets makes it easier to
locate, visually, the corresponding left and right member of a pair of
brackets. An equivalent tree-diagram (which is much easier to 'read')
is given in Fig. 9.
We have now brought into the representation of the grammatical
structure of sentences such as Poor John ran away the fact that poor
John is a noun-phrase (NP) consisting of the adjective (A) poor and the
noun (N)jfohn and that ran away is a verb-phrase (VP) composed of
the verb (V) ran and the adverb (Adv) away. And from now on we
shall consider such information to be an integral part of the analysis of
sentences in terms of their constituent-structure. We have still not
introduced the traditional notions of 'subject* and 'predicate'. Nor
have we yet said anything about 'modification'. In traditional
grammatical theory the phrase poor John would be classed as a noun-
phrase because it 'functions as', or 'does the work of, a noun in
8 LIT
sentences. This can be interpreted as meaning that phrases of the
form adjective + noun have the same distribution in the sentences
generated by the grammar as nouns; that poor John, new car, etc.,
can be freely substituted tor John, car, etc., in any given sentence, and
the result will be another grammatically acceptable sentence. This
fact is accounted for in our analysis of such sentences as Poor John
ran away by labelling the 'node' through which the adjective and the
noun are related with the term 'noun-phrase'.
Fig. 9.
Before we develop further the notions of constituent-structure

from a generative point of view, we must explain what is meant by the
term 'grammatical ambiguity', as it has been used above. Consider all
the noun-phrases in English of the form A + Nx + N2 (A = adjective,
N = noun). Many of these are ambiguous; and their ambiguity can
be resolved by specifying their constituent-structure as either
A + (N + N2) or (A + N^ + N2. Now, many of these phrases are not
subject to misinterpretation when they are actually used in sentences,
because either the rest of the sentence or the general context in which
the language operates makes it certain, or at least very probable, that
one interpretation rather than the other is the correct one. The phrase
fresh fruit market, for instance, is hardly likely to be understood in the
sense indicated by the bracketing fresh (fruit market); conversely,
the phrase new fruit market is not likely to be used in the sense of
(new fruit) market. Let us assume, for the sake of the argument, that
neither fresh fruit market nor new fruit market has more than one
interpretation: from the semantic point of view, we will say that they
are not ambiguous. Are they grammatically ambiguous? Is the
constituent-structure fresh (fruit market), in the one case, and (new
fruit) market, in the other, grammatically acceptable ? To answer these
questions we must of course refer to some explicit grammar of
English. It is clear that, in general, the bracketing A + (A^ + N2) is
acceptable if the first noun can combine wTith the second (fruit
market) and if the adjective can combine with the second noun (new
market, }fresh market); and the bracketing (A + N^ + N2 is accept-
able if the first noun can combine with the second noun and if the
adjective can combine with the first noun (fresh fruit, Inew fruit).
The question therefore is essentially the same as that discussed above
in connexion with the limits of grammar: the question of sub-
classification with the two limiting factors of indeterminacy and of
1
diminishing returns' (cf. 4.2.11). Any phrase of the form A + N + N2
will be given two grammatical analyses, unless the grammar and
lexicon to which we refer prohibits explicitly the combination of the
adjective in question with one or other of the nouns.
The notion of constituent-structure does not rest solely upon its
capacity to handle ambiguities of the kind we have discussed. The
main reason for regarding sentences as being composed of * layers* of
constituents is that in this way we can achieve a more economical and
intuitively more satisfying description. This will be clear from the
discussion which follows. It may be mentioned in passing that there
are other kinds of ambiguity which can be accounted for by an
essentially different kind of grammatical structure from that which
we are at present considering. Some of these will be discussed later in
this chapter (cf. 6.6.2).
6.2 Phrase-structure grammars

6.2.1 * Phrase-structure'
It is one thing to know where to put the brackets in a given sentence:
it is quite a different matter to construct a system of generative rules
of such a kind that they will explicitly assign the correct constituent-
structure to sentences. Although many different systems of
grammatical rules can be designed which will impose a constituent-
structure analysis upon the sentences they generate, we shall restrict
ourselves initially to the consideration of concatenating rewrite-
systems of the type which Chomsky has made familiar. Such systems
we will call phrase-structure grammars. By implication, * constituent-
structure grammars' will be a wider term.
2l6 6. GRAMMATICAL STRUCTURE
6.2.2 Rewrite-systems
We may begin by providing a set of rewrite-rules (they are assumed to
be concatenating: cf. 6.6.1) of the following form:
(1) I<->NP+VP
(2) VP->V+Adv
(3) NPÂ + N
If these rules are applied in sequence in such a way that each rule
(apart from the first) is used to replace, or * rewrite', in the 'output'
of the previous rules, whatever symbol occurs in the left-hand side of
the rule with whatever symbols occur 'bracketed' after the arrow,
we shall get by rule (1) the output
2 {NP+ VP)
and by the application to this of rule (2)
I<{NP+VP(V + Adv)}
and, finally, by rule (3)
2 {NP(A + N)+VP(V+ Adv)}
That is, each rule brackets together the constituents which form the
construction it defines, and labels the construction; and the structural
layers are determined by the order in which the rules are applied.
In the present instance the result would be the same if rule (3) were
applied before rule (2). But rule (1) must be applied first, in order to
produce the elements required for rules (2) and (3). The term occurring
in the left-hand part of rule (1) of the grammar denotes the highest-
level construction, of which all other constructions generated by
the grammar are constituents. We shall call this term ('sentence') the
initial symbol. After the application of all the rules relevant in the
generation of a particular type of sentence, the grammar will have
'produced' a bracketed 'string' of symbols (adjective, noun, verb,
etc.), each of which denotes a class of elements in the lexicon. The
symbols denoting lexical classes we shall refer to as terminal symbols,
and bracketed strings of terminal symbols as terminal strings. If we
now substitute for each of the terminal symbols in a terminal string
a member of the lexical class it denotes, we shall get a sentence whose
constituent structure is completely determined by the rules which
generate the terminal string. Thus, if the lexicon contains the fol-
lowing information (cf. 4.3.1 ff.)
6.2. PHRASE-STRUCTURE GRAMMARS 217
TV = {John, etc.}
V = {ran, etc.}
A = {poor, etc.}
Adv = {amry, etc.}
the three-rule grammar given above will generate such sentences as
Poor John ran away with their correct constituent-structure. Con-
versely, presented with this sentence and using a lexicon which is
organized more conveniently for analysis, i.e. in some such form as
away: Adv
John: N
poor: A
ran: V
we shall be able to replace each word of the sentence with its class-
symbol and then work through the rules of the grammar upwards and
from right to left until, if we arrive successfully at the initial symbol
in rule (1), we thereby recognize the sentence as being grammatically
acceptable and as having a certain constituent structure. It should be
stressed that the reason why it is possible to work through the rules
in either direction in the case of the present grammar is that it
conforms to certain general conditions (which we will not discuss here:
cf. 6.2.11). Independently of this particular property it is neutral with
respect to analysis and synthesis (cf. 4.3.1); and it formalizes the
notion of * labelled bracketing' (cf. 6.1.3).
6.2.3 Alternative rules

Let us now extend the grammar, in order to make it capable of
'producing' and 'recognizing' sentences like Old men love young
women, in addition to such sentences as Poor John ran away. We can
do this by introducing rules which allow for alternative ways of
* rewriting' the element VP (verb-phrase). Thus:
(1) !>->NP+VP
(za) VP->Vintr
(2b) VP-+Vir
(3) NP-+A + N
It is to be understood that one of the sub rules (2 a) and (2b) must be
applied but that the choice between them is free. When we introduce
2l8 6. GRAMMATICAL STRUCTURE
the possibility of choosing between (2 a) and (26) in the grammar, we

must also change the classification of words in the lexicon:
Vintr = {'My etc.}
Vtr = {love, kill, etc.}
Our grammar is still somewhat unsatisfactory in a number of ways.
First of all, it will be observed that, although it correctly generates the
sentences we started with, it also generates such unacceptable
' sentences' as *Poor John kill old women. At some point we shall have
to take account of the 'agreement* that holds in English between the
'subject* of the sentence and the verb (if the verb is in the 'present
tense'). This question we shall leave for the moment (cf. 6.5.4).
A second deficiency of the grammar is that, if each of the rules is
applied where it is applicable, as we have so far assumed to be the
case, the grammar will necessarily generate either five-word sentences,
like Old men love young women, or four-word sentences, like Poor John
ran away. It will not generate John ran away, John ran, Men love
young women, Old men love women, etc. And it will not generate, for
example, Old men love young women passionately.
6.2.4 Optional and obligatory rules

Suppose we now make rule (3) into two rules, viz.:
(3) NP->N
(4) N-+A + N
specifying that, whereas rule (3) is obligatory, rule (4) is optional. The
grammar will now generate John ran away, Men love women, Old men
love women, Men love young women, etc.
Consider now the tree-diagrams (given in Fig. 10) representing the
sentences generated by the rules set up so far, as exemplified by
(i) John ran away, (ii) Poor John ran away, (iii) Men love women,
(iv) Old men love women, (v) Men love young women, (vi) Old men love
young women.
By reference to these tree-diagrams or to the rules by which they
are constructed, we can see that all the sentences in question are
subtypes of the sentence-type H(NP+ VP). In other words, at a
certain level of analysis they are structurally identical. At a lower
level, (i) and (ii) are identical with one another, but different in
structure from (iii)-(vi); and so on. The introduction of optional rules
and alternative rules therefore increases the power of the grammar, and
also groups sets of constructions into subtypes of a common type.
JVP FP NP VP
Adv N I
AN,
John ran away j 1 ran away
1 1
1
Poor John
0)
2
/ \
\
NP VP NP VP
/ \
\
N
i
Vtr X N NP
NP
i
i \ A
N
T
1
Men 1 A
love i love
women Old men women
(iv)
NP VP NP VP
N /. NP N Vtr NP
/ \ \ 1 / N.
\ i / \
Men love Aii Ni A N \ A N
i i love I
i I i
young women Old men young women
(v) (vi)
Fig. 10.
(Note that the branch for NP->N has been omitted from (v) and (vi).)
6.2.5 Ordering of rules

In all that has been said so far, it has been assumed that the rules of
the grammar are to be applied (where applicable) in the order in which
they are numbered. We must look at this question more closely, for
the imposition of one order rather than another upon the rules may
make a difference to the output of the grammar. Consider the
following system of rules, which is the same as that given earlier,
except that a new rule (4) has been added: (rules (4) and (5) are
optional rules):
(1) "L-+NP+VP
{za) VP->Vintr
{zb) VP->Vtr
(3) NP->N
(4) N-+T+N
(5) N->Adj+N
It is assumed that the information T = {the} is given in the associated
lexicon. Since rules (4) and (5) are optional, the grammar will
generate men, the meny good men, and the good men. But if rule (5) had
been allowed to apply before rule (4), such unacceptable sequences as
*good the men would have been permitted as grammatical. Notice also
that the place of rule (3) is important. If it were put before (26), it
would have to be repeated again after {2b), in order to provide for the
expansion of the NP which results from VP -> Vtr -f NP.
We have seen that the order of rules in a 'rewrite' constituent-
structure grammar may be critical for two reasons: to prevent the
generation of unacceptable sequences of elements and to achieve a
reduction in the number of rules required. In certain instances, it
might be desirable to allow a particular group of rules to apply in
various orders in order to account for grammatical ambiguity of the
kind referred to above (cf. 6.1.3). Take, for instance, an optional rule
for the co-ordination of nouns of the form
(6) N^N+and+N
If it is applied before rule (5) in the present system of rules, the
output will be phrases like {old men) and women and men and {old
women). But if it is applied after rule (5), the result will be phrases like
old {men and women). The ambiguity of the sequence old men and
women, that is as either {old men) and women or old {men and women),
can be satisfactorily accounted for, therefore, in terms of a difference
in the relative order of application of rules (5) and (6). It may be
observed, however, that the bracketed phrase generated by the appli-
cation of rule (6) after rule (5), old (men and women), is semantically
equivalent to (old men) and (old women), which could be generated
by a double application of rule (5) to the output of rule (6).
6.2.6 Recursive rules

The rule for the co-ordination of nouns is one that we should wish
to be applicable not just once, but many times; indeed innumerable
times. For it is possible to have phrases formed of indefinitely many
nouns linked by and. The simplest way of handling these might seem
to be by introducing into the grammar a set of alternative rules:
(6a) N->N+and+N
(6b) N->N+and+N+and+N
(6c) N^N+and+N+and+N+and+N
etc.
However , this is clearly unsatisfactory since there is no upper limit
to the number of constituent nouns that can be co-ordinated in this
way in English. It is true that in a given text, or in the total set of
utterances that have ever been produced, there will be some noun-
phrase (or a number of noun-phrases) which contains more co-
ordinated nouns than any other noun-phrase. It may also be true that
there is very little likelihood of a noun-phrase being produced by a
native speaker which includes more than this highest number of
constituent nouns. The fact remains that, whatever number we
decide upon as the upper limit, it will always be possible to produce a
sentence containing a noun-phrase the number of whose constituent,
co-ordinated nouns exceeds any alleged 'upper limit'. Therefore,
since it is impossible in principle to list an infinite set of alternative
rules, we may specify that rule (6) may be applied indefinitely many
times. Rules that may be applied an indefinite number of times are
called recursive. It is very probable that in all languages there are
many constructions, the generation of which requires recursive rules.
Other examples from English are the co-ordination of clauses in
sentences (e.g. He came in and he sat down and he said that.. .and
he...); and the modification of nouns with adjectives (e.g. great big
black.. .dog).
If we wish to make our grammar a more 'realistic* model for the
production of sentences and to construct it in such a way that it will
generate more instances of recursive structures with two constituents
than with three, more instances of structures with three constituents
than with four, and so on, we can do this by assigning a particular
probability to each number of times that the rules in question are to
be applied (the probabilities being generated as an infinite series
whose limit is zero). Probabilities could be assigned in the same way to
each member of a set of alternative rules and to each optional rule in
the grammar. But we must be careful not to confuse probability of
occurrence with grammaticality. Although native speakers may
hesitate about the acceptability of certain particular constructions of
low frequency, it does not appear to be the case that they will always
consider constructions with a high probability of occurrence to be
more acceptable than constructions with a low probability of
occurrence.
6.2.7 Recursive co-ordinate structures

The treatment of co-ordination by means of a recursive rule brings
us to another problem. All rules of the type we are discussing in
this section assign a bracketed structure to the constructions they
generate; as we have seen, the different structural layers' are deter-
mined by the order of application of the' rewrite * rules. Consider now
a phrase like Tom and Dick and Harry. This sequence of words will
be produced by a double application of rule (6) given above. But it will
necessarily be generated as grammatically ambiguous: as either {Tom
and Dick) and Harry or Tom and {Dick and Harry). (Noun-phrases
containing more than three co-ordinated nouns can be generated in
several different ways, and will therefore be described by the grammar
as multiply ambiguous.) Although the bracketing {Tom and Dick)
and Harry or Tom and {Dick and Harry) might well be intuitively
satisfying in certain instances, we should normally wish to regard
noun-phrases composed of three (or more) co-ordinated nouns as
internally unbracketed. Constituent-structure grammars therefore
present us with a dilemma with regard to the treatment of co-
ordination. Recursive rules will fail to yield what is intuitively the
correct structural description. Yet recursive rules are necessary in
order to generate the infinity of such constructions. Here we merely
point to the dilemma. (It has been much discussed in works dealing
with transformational grammar.)
6.2.8 Discontinuous constituents

In all that has been said so far we have tacitly assumed that the
constituents of a given construction will always occur next to one
another in sentences. Consideration of a sentence such as John called
Bill up shows that this is not so; in this sentence, called.. .up is a
discontinuous constituent, enclosing the other constituent, Bill, with
which it is in construction. Let us now compare the following three
sentences, to see how they might be handled in a generative grammar:
(a) John called up Bill
(b) John called Bill up
(c) John called him up
The important thing to notice is that called up is optionally dis-
continuous when the object is a noun, but obligatorily discontinuous
when the object is a pronoun (*John called up him, with normal stress
and intonation, is unacceptable). We will say, not only that called up
is a constituent of all three sentences, but that, in some sense or at
some more abstract level, all three sentences have the same con-
stituent-structure; and we will introduce a supplementary rule (of a
different kind from the constituent-structure rule) which will apply
optionally to the terminal string 'underlying* (a) and (b) and
obligatorily to the string * underlying' (c), the effect of which will be
to insert the 'object* between the two parts of the verb called up.
In this way (provided that we can specify the conditions under which
it applieswith what class of verbs, how the' object' is identified, etc.)
we shall be able to handle the problem of' discontinuous constituents*
satisfactorily.
A similar treatment suggests itself, and is much more generally
applicable, in the case of languages with 'free word-order'. An
example from Latin will make the point clear. The English sentence
Catullus loved Clodia can be translated into Latin by any one of the
following six sequences of words: (i) Catullus Clodiam amabat,
(ii) Catullus amabat Clodiam, (iii) Clodiam Catullus amabat, (iv) Clo-
diam amabat Catullus, (v) Amabat Clodiam Catullus, (vi) Amabat
Catullus Clodiam. (There are differences of emphasis or contextual
presupposition associated with each of the six versions, but all six
are acceptable.) The form of the word Catullus (rather than Catullum,
Catulli, etc.) marks it as the subject, in the 'nominative* case; and the
form of the word Clodiam (rather than Clodia, Clodiae, etc.) indicates
that it is the object, in the 'accusative* case. The simplest way of
generating the six alternative versions of the sentence would seem to
be by means of a set of supplementary rules of permutation operating
upon the same terminal string. If all word-order were completely
' free* in Latin, the supplementary rules would be quite simple. The
fact that there are constraints upon the order of certain words in
Latin sentences makes the question more complicated, but does not
affect the principle being illustrated.
6.2.9 * Supplementary* rules

This brief discussion of discontinuous constituents and 'free' word-
order has brought us to the view that the grammar of a language might
include rules of two different kinds: on the one hand, phrase-structure
rules and, on the other, what we have been calling * supplementary*
rules whose function it is to * transform* the terminal strings generated
by the constituent-structure rules into actual sentences (or, more
precisely, into sequences of elements which are more directly relatable
to sentences). We will return to this more * abstract* view of con-
stituent-structure in a later section (cf. 6.6).
6.2.10 Complex sentences

Within certain limits constituent-structure grammar can handle the
formation of complex, as well as simple sentences (for the traditional
distinction of simple and complex sentences, cf. 5.2.6). For example,
a sentence such as Bill was reading the newspaper when John arrived
might be generated by a system of 'rewrite* rules which included the
following (and others which, for simplicity, we will omit):
(a) "L-+NP+VP (Time Adverb)
/i_\ rn- AJ L (Preposition + Time Noun
(b) Time Adverb -> L , r , . . ,
11 emporal nConjunction + v2
It will be observed that Time Adverb is introduced as an optional
constituent of the sentence by rule (a): the parentheses signify that it
is optional. We will assume that rule (b) is optional: so that, if rule (b)
is 'bypassed*, Time Adverb will appear in the terminal string and be
replaced in sentences by yesterday, etc. On the other hand, if rule (b)
is applied and the option Preposition + Time Noun is taken, the final
result will be on Tuesday, in March (further rules would be required
to account for the occurrence of different prepositions with different
subclasses of 'time nouns* and also for the generation of last week,
etc.). It is, however, the option Temporal Conjunction + S which
concerns us at this point. The selection of this option, with the
appropriate development of the constituent sentence, will account for
when John came in (and will classify it as a Time Adverbial, like
yesterday or on Tuesday). The constituent-structure assigned to a
sentence like Bill was reading the newspaper when John came in would
be as illustrated in Fig. 11. The effect of permitting the inclusion of
NP Temporal
Conpmctiwi
was reading the token

newspaper
Adv
came
Fig. 11.
the initial symbol in the right-hand side of a 'rewrite' rule is to

introduce the possibility of * recycling* into the grammar. This
feature is certainly required, in order to account for the 'embedding'
of sentences (as clauses, or constituents of clauses) within other
sentences: cf. He said that..., or The man who.. .has just arrived.
At the same time, it is clear that sentence-embedding rules of the
kind illustrated here are not capable of handling such features of
complex sentences as what is traditionally called the 'sequence of
tenses' in the constituent clauses (and so preventing the generation
of such ' sentences' as *Bill was reading the newspaper when John has
come in). We shall have more to say about this and related questions in
the following sections. It is sufficient for our present purpose to have
shown how a 'rewrite' constituent-structure grammar might formal-
ize the distinction that is traditionally drawn between simple and
complex sentences: any sentence that is generated by a group of rules,
at least one of which contains the initial symbol in the right-hand side
is a complex sentence; all other sentences are simple.
6.2.11 Formalization of context-free phrase-structure grammars

All the constituent-structure rules given so far in this section have
been of the general form A -> B ('rewrite A as B', in the synthesis of
sentences). Nothing has been said about the 'values' taken by A and
B, except that the first rule of the system must have the initial symbol,
S (Sentence), as A (i.e. on the left-hand side) and that the terminal
symbols (N, Vtr) Vintr> A, etc.), which occur in the strings generated
by the grammar, denote lexical classes: N = {boy, girl, . . . } ,
Vlr = {eat, kill, . . . } , Vintr = {die, go, . . . } , Adj = {good, old, ...}.
Rules of the form A -> B, which rewrite A as B without reference to
the context of A, are referred to as context-free rules. (We return to
this question in 6.5.1.)
The formal properties of context-free constituent-structure gram-
mars have been intensively studied in the last few years. Chomsky and
others have discussed the effect of ordering the rules in a system, of
allowing both optional and obligatory rules, of introducing alternative
subrules and of having recursive rules in the system. Furthermore
they have pointed out that two important restrictions must be
imposed upon the values of A and B in rules of the form A -> B, if
a grammar containing such rules is to be capable of assigning a unique
structure to the sentence it generates: (i) A and B must not be
identical (i.e. A must not be rewritten as itself); and (ii) A must be a
single symbol, although B may be, and generally is, a string con-
taining more than one symbol. Some scholars have worked with
constituent-structure grammars which impose more particular
limitations upon B, notably that it must consist of two and only two
elements (e.g. NP+ VP> or V^ + N), For a fuller explanation of the
implications of such restrictions the reader is referred to the more
technical works cited in the notes.
6.2.12 Weak and strong equivalence

The most important result of this theoretical study of grammar
(which draws upon research into the formalized systems of logic and
mathematics) has been the demonstration that constituent-structure
grammars with different formal properties may generate exactly the
same set of sentences. Grammars which generate the same set of
sentences are said to be weakly equivalent \ grammars which not only
generate the same sentences, but assign to them the same structural
description are said to be strongly equivalent. To illustrate the
difference between weak and strong equivalence we will consider
briefly constituent-structure grammars of a somewhat different kind
from the 'rewrite' systems discussed above: namely, ' categorial'
grammars.
6.3 Categorial grammars
6.3.1 Fundamental and derived categories

Categorial grammars have their origin in the work of the Polish
logician Ajdukiewicz (following Lesniewski); they have been further
developed by Bar-Hillel, Lambek, and other contemporary logicians
and linguists. (The choice of the term * categorial' in this connexion
is explained by particular developments in the history of logic and
philosophy, which we will not go into here.)
In a categorial system there are just two fundamental grammatical
categories, sentence and noun: we shall represent these as 2 and n,
respectively. All lexical items other than nouns are given a derived
categorial classification in the lexicon according to their potentiality
for combination with one another or with one of the fundamental
categories in the constituent-structure of sentences. The derived
categories are complex, in the sense that they specify simultaneously
(i) what other category the element in question can combine with in
order to form a sentence-constituent, and (ii) the categorial classifi-
cation of the constituent that results from this operation. To take a
simple example: an element like run or exist (an 'intransitive verb')
can combine with a noun (as * subject') to form a sentence: e.g. John
ran (we shall continue to disregard the question of tense and features
of * agreement').
6.3.2 ' Cancellation'

In the 'quasi-arithmetical' notation of Bar-Hillel the categorial
classification of elements like run can be expressed as a 'fraction'
whose denominator denotes what other category run, etc. can combine
with and whose numerator denotes the category of the resultant
2.
construction. Thus, the lexical classification of run, etc., as indicates
n
that such elements combine with nouns to form sentences. Given this
classification of run and given that John is a noun (n), we know that
John ran is a grammatically well-formed sentence. We can determine
this fact automatically by a simple rule of * cancellation' similar to the
arithmetical rules of cancellation: just as
x 2 _
y x - = x, so n. = It
y n
(i.e. we * cancel out' the numerator and denominator when they are
identical and, in this case, we are left with 2, which indicates that
the expression is a sentence: the dot is being used here to represent
linear concatenation). But the system must also be capable of
excluding as ungrammatical such sequences as .n (Ran John). We
must therefore specify the direction in which an element combines, to
the left or to the right. We will indicate this by means of an arrowhead
attached to the horizontal * fractional* line: thus *- denotes an
n
element which combines with a noun to the left to form a sentence;
whereas, for example, -* denotes an element (e.g. an 'adjective' like
poor, oldy etc.) which combines with a noun to its right to form a noun
(or noun-phrase).
6.3.3 More complex categories

Derived categories may have, not only fundamental categories, but
also derived categories as their numerator or denominator. For
example, an * adverb' combines with a' verb' to its left to form a * verb *
(or verb-phrase); that is to say it combines to its left with an element
which combines to its left with a noun to form a sentence (for
simplicity we will consider only 'intransitive verbs', and we will
disregard the fact that 'adverbs' may be given other categorial
classifications as well). This is indicated by the 'fraction':
6.3.4 Possible extensions

(i
The categorial system we have just described is a relatively simple one,
but it is satisfactory enough for our purpose. It is bidirectional, in the
sense that it allows for combination to either the left or the right: it
6.3. CATEGORIAL GRAMMARS 229
could be extended in various ways (by permitting the occurrence of
two or more concatenated categories in the denominator of a derived
category; by permitting an element to combine both to the left and
to the right simultaneously; and so onwe shall not go into the effect
of such extensions here); and it could be restricted (made uni-
directional) by allowing for combination in only one specified
direction. It will also be observed that this categorial system, like the
'rewrite* system discussed above, cannot handle discontinuous
constituents (cf. 6.2.8).
6.3.5 Notational conventions

At this point we will introduce a space-saving notational convention,
which has no effect upon the formal capacity of the system. The
convention is to write the * fractions' as an ordered pair of bracketed
symbols with the numerator coming first and the denominator second
and an arrow over the denominator to indicate direction. By this
convention run will be classified as (2 n), poor as (nn) and an ' adverb'
such as away as {(Sn)(Sn)}.
6.3.6 * Categorial9 analysis of constituent-structure

We may now illustrate the operation of this simple bidirectional
categorial grammar with reference to Bloomfield's sentence Poor John
ran away. The categorial classification of all the lexical elements has
been given in the previous paragraphs. We will write them under the
words in the sentence:
Poor John ran away
(nn) . n . (Sn). {(Sn)(Sn)}
Now for the 'cancellations'. It is clear that there are three possibilities
at the first stage: (1 a) the ' cancellation' of poor with John; (1 b) of John
with ran; and (1 c) of ran with away. But if (ife) is selected, the result
will be
(1 ft) (nn) . S . {(Sn)(Zn)},
which establishes John ran as a sentence and leaves poor and away
unaccounted for outside the sentence. If (1a) is selected, at the next
stage we will have
(1 a) n . (Sn) . {(Sn)(Sn)}.
There are now two possibilities: and it is clear that, in order to
analyse the sentences without residue, we must now cancel ran with
away to yield
(a)n.(Sn);
and, finally
(3)2.
Alternatively, we might have first combined ran with away (ic), and
then poor with John: in other words we might have cancelled the same
elements, but in a different order.
6.3.7 Comparison of phrase-structure and categorial analysis

The operation of 'cancellation' reflects the notion of bracketing with
which we are concerned in this section; it will be clear that the
analysis of Poor John ran away which has just been given agrees with
what Bloomfield said about the structure of this sentence: that it is
composed of two immediate constituents, poor John and ran away, and
that each of these consists of two immediate constituents, poor and
John, on the one hand, and ran and away, on the other. Let us now
compare the categorial analysis with an analysis in terms of the
following 'rewrite* rules (cf. 6.2.2):
(1) X-+NP+VP
(2) VP + Vinir + Adv
(3) NP-+A + N
From the composite tree-diagram given in Fig. 12, it will be clear that
the two grammars are not only weakly equivalent (in that they
generate the same set of sentencesthe fact that we have put the
categorial system in the form of a 'recognition* grammar and the
'rewrite* system in the form of a 'production* grammar is irrelevant
to this issue: cf. 4.3.1); they are also equivalent with respect to the
bracketing they assign to sentences like Poor John ran away. But
they are far from being equivalent in all respects. First of all, it may
be noted that the 'rewrite* system has two auxiliary symbols (NP
and VP) in addition to the four terminal symbols which denote lexical
classes (N, Vintr> A and Adv); so that, whereas the categorial analysis
asserts that poor John is a phrase of the same category as John and
ran away a phrase of the same category as ran, the 'rewrite* analysis
does not. However, we could easily make the two systems equivalent
6.3. CATEGORIAL GRAMMARS 231
with respect to this feature by substituting N for NP and Vinir for
VP in the ' rewrite' rules (we will not go into the further implications
of this revision). The main difference between the two systems lies
in the fact that the categorial grammar, unlike the 'rewrite' grammar,
regards one constituent in each construction as dependent upon the
other: the categorial notation makes it clear which is the dependent
constituent (the one with the more complex classification) and this is
fundamental to the operation of 'cancelling*. The 'rewrite* system
Ado
(25) (25)
Fig. 12.
represents the notion of dependency only partially and indirectly:

from a rule of the form iV -> A + N it can be inferred that A is
dependent upon N, but in a rule like S -> N+ Vintr (or 2 -+ NP+ VP)
no conclusions can be drawn about any relationship of dependency
that might hold between the two constituents. This does not mean
that the categorial system is necessarily better than the 'rewrite*
system, but only that the two systems are not strongly equivalent.
There may be reasons for preferring one analysis to the other; if so,
we shall say that the grammar which assigns the preferred analysis is
more strongly adequate than the grammar which does not. This is a
separate question, to which we will return briefly later (cf. 6.5.7).
6.4 Exocentric and endocentric constructions

6.4.1 Distributional interpretation
We may now look a little more closely at the distributional basis for
immediate constituent analysis. Constructions may be classified,
according to their distribution and that of their constituents, into
what are commonly called (in the terminology introduced by Bloom-
field) endocentric and exocentric constructions. An endocentric
construction is one whose distribution is identical with that of one
or more of its constituents; and any construction which is not endo-
centric is exocentric. (In other words, exocentricity is defined nega-
tively with reference to a prior definition of endocentricity, and all
constructions fall into one class or the other.) For example, the now
famous phrase poor John is endocentric, since it has the same distri-
bution as its constituent John; any English sentence in which John
occurs can generally be matched with another sentence in which
poor John occurs in the same position; and, since John is a noun,
poor John is described as a noun-phrase. On the other hand, in
Vancouver is exocentric, since its distribution is different from that of
either the preposition in or the noun Vancouver, The phrase in Van-
couver has much the same distribution in English sentences as there and
other adverbs (of place); therefore it is classified as an adverbial
phrase (of place). These examples show that, although the distinction
between endocentric and exocentric constructions is one that was not
made as such in traditional grammar, the traditional notion of
'function' (according to which poor John is said to 'function as' a
noun and in Vancouver as an adverb) can be given a natural inter-
pretation in terms of the distribution of the elements in question.
It is often said that no construction has exactly the same distri-
bution as any of its constituents; that the substitution of the con-
struction for one of the constituents will result in a certain number of
unacceptable utterances. This is doubtless true. But, as we saw in a
previous chapter, by the distribution of an element we do not mean
the totality of acceptable utterances in which the element occurs, but
that set of utterances whose acceptability is accounted for by the
grammatical description of the language, the limits to which this can
be pushed being determined by what we have called the 'law of
diminishing returns' (cf. 4.2.11). In this sense, therefore, any two
elements either have or have not the same distribution; and con-
structions are either endocentric or exocentric. Any grammar of
English will recognize many different (and overlapping) subclasses of
nouns, verbs, adjectives, etc. For instance, a distinction will be drawn
between 'countable' nouns and 'uncountable' (or 'mass') nouns (to
handle the acceptability of / like wine, etc., and the unacceptability
(except in cannibalistic circumstances) of */ like boy, etc.); between
'human' nouns and 'non-human' nouns (to handle, inter alia, the
distribution of who: which: cf. the boy who... and the book which...);
between singular nouns and plural nouns (to handle 'agreement';
cf. the boy is... and the boys are...); between masculine, feminine
6.4. EXOCENTRIC AND ENDOCENTRIC 233
and neuter nouns (to handle the 'reference* of he, she, it); and so on.
When we say that two nouns have exactly the same distribution, we
mean that they are given the same classification at the 'lowest* level
of distributional subclassification that the grammar achieves. But we
can say that two elements have the same distribution at various
specified 'depths'. Thus, all nouns (and all noun-phrases and noun-
clauses) have the same distribution at the relatively high level of classi-
fication for which the term' noun' is used without further qualification.
At a lower level two nouns might have a different distribution, one
being 'animate' and the other 'inanimate', etc. The concepts of endo-
centricity and exocentricity are therefore to be used with respect to
some specified 'depth' of subclassification.
6.4.2 'Nesting' of endocentric constructions

Endocentric constructions fall into main types: co-ordinating and
subordinating. Co-ordinating constructions have the same distribution
as each of their constituents taken separately. Thus bread and cheese
and coffee or tea are co-ordinating noun-phrases. However, the two
phrases belong to different subtypes, the first taking a plural verb
and the second a singular verb (if, as here, each of the constituents is
singular). Subordinating constructions have the same distribution as
one of their constituents. Examples from English include: A + N
(poor John); Adv + A (awfully clever); N (or NP) + Adv (or adverbial
phrase) (the girl upstairs, the man on the bus). The constituent whose
distribution is the same as that of the resultant construction is called
the head; the other constituent is the modifier. In subordinative
constructions one modifier may be recursively 'nested' within
another. For example, in the man on the top of the bus there are two
constituents, the man (head) and on the top of the bus (modifier); on the
top of the bus is an exocentric adverbial phrase consisting of the pre-
position on and the noun-phrase the top of the bus; the top of the bus
is endocentric, its constituents being the top (head) and of the bus
(modifier). Noun-phrases of the kind illustrated here can be extended
more or less indefinitely: cf. the man on the top of the bus in the park
in the centre of the town on the top of the hill in the centre of the plain....
The classification of the different types of construction that has
just been given is based upon the work of Bloomfield and his
followers; whereas the account of constituent-structure in terms of
generative ' rewrite' rules which has taken up the greater part of this
section derives mainly from the more recent treatment of the subject
by Chomsky. It is to be noticed that, while Bloomfield talked always
of the classification of constructions, Chomsky talks of the generation
of constructions. As we have already seen (cf. 4.2.8), any analytic
classification of the constructions of a language must also be
synthetic (implicitly, if not explicitly). However, the change of out-
look represented by the conscious and deliberate adoption of the
generative approach has the practical effect of making the task of
classification secondary to the main purpose of the grammar, the
generation of sentences: the generative approach imposes limits on
classification. For example, Bloomfield describes the English words
longlegs and butterfingers as exocentric, 'because they occur both as
singulars.. .and plurals {that longlegs)9. But this fact is sufficient to
show that the forms in question are not constructions at all; they are
to be entered in the lexicon as unanalysable wholes, as members of
the appropriate distributional class, and are not to be generated by the
grammar. The distribution of longlegs9 unlike the distribution of long
legs (in He has long legs) cannot be accounted for by any productive
formational rules of modern English.
6.4.3 Endocentricity in a phrase-structure grammar

The distributional basis for the rules of a 'rewrite* constituent-
structure grammar is clear enough. Every rule of the form A -> B + C
rests upon the distributional identity of A and B + C; more precisely,
upon the inclusion of the distribution of B + C in the distribution of
A (identity being a special case of inclusion). The difference between
identity and inclusion (or, in logical terms, between 'improper* and
'proper* inclusion) comes out in the distinction between rules which
' rewrite * all instances of A as B + C and those which consist of a set of
alternative subrules. For example, a rule such as N-> A + N rests
upon the distributional identity of John and poor John, of man and
old many etc. On the other hand, a rule such as
implies that the distribution of phrases like eat meat, on the one hand,
and of words like die, on the other, is included in the distribution of
the class of verb-phrases.
The difference between endocentric and exocentric constructions,
on the one hand, and between the two main subtypes of endocentric
constructions (co-ordinating and subordinating), on the other hand,
6.4. EXOCENTRIC AND ENDOCENTRIC 235
is manifest in the form of the 'rewrite' rules which account for the
formation of these constructions. It should be noticed, however, that
the notions of endocentricity, subordination, etc., have no systematic
significance in a 'rewrite* grammar of the kind outlined above: that is
to say, these notions are not invoked as such in the grammar. More-
over, since the effect of an obligatory rule (or set of subrules) is to
'rewrite* all instances of A as B + C it does not matter what A is
called: after the operation of the rule, A disappears, except as the
name of the higher 'node*. For example, the choice of the terms noun-
phrase (NP) and verb-phrase (VP) in the first rule of the grammar
X-+NP+VP
is a perfectly arbitrary choice in a 'rewrite* constituent-structure
grammar: for both NP and VP are obligatorily 'rewritten* as various
other elements or constructions. (They might equally well be called
Y and Z.) The relationship between noun-phrases, nouns and pro-
nouns (of various subclasses) is not expressed by the nomenclature
applied to them, as in 'classificatory* treatments, but by the fact of
their deriving from some common ' node*. In respect of the notions of
endocentricity and exocentricity, and of subordination and co-
ordination, a categorial grammar might claim to be a truer formaliza-
tion of the principles outlined by Bloomfield and developed by his
successors, with a good deal of exemplification from many different
languages. As we have seen, a categorial grammar takes as fundamental
the notion of 'dependency* (which is generally to be equated with
subordination); and the difference between endocentric and exo-
centric constructions is evident from the structure of the dependent
category in a 'cancellation'if the numerator and denominator are
identical, the construction is endocentric; it not, it is exocentric
(cf. above 6.3.2).
6.5 Context-sensitive grammars

6.5.1 The term 'context-sensitive*
We must now introduce a distinction between context-sensitive and
context-free rules. (The terms 'context-restricted* and 'context-
dependent' are used synonymously with 'context-sensitive' in the
literature.)
All the rules given so far have been context-free, in the sense that
the symbol appearing to the left of the rewrite-arrow was replaced in
the ' output' of the rule as the symbol or string of symbols appearing
to the right of the arrow in the statement of the rule. To take an
example, the following rule (cf. 6.2.6)
N-+N+and+N
was interpretable as an instruction to 'rewrite' AT as N+and+N
(optionally or obligatorily) in any string of symbols to which the rule
was applied. In other words, the sole condition for the operation of
the rule was the occurrence of N in the 'input* to the rule. For
instance, given as alternative 'input' strings
(i) X + 7V+Y
(ii) W + AT+Z
the 'output', after the operation of the rule in question, would be,
in the one case,
X + N+and+N+Y
and in the other,
W + N+and+N+Z
The point is that there were no contextual restrictions imposed upon
the operation of the rule.
Suppose, however, that we had formulated the rule in the following
manner
N - N+ and+Nfm the context X +... + Y
This might be defined to mean that N was to be' rewritten' (optionally
or obligatorily) only if it occurred in the 'input* string with X
immediately to its left and Y immediately to its right. In which case,
the rule would apply to (i), but not to (ii). The specification of the
contextual condition to the right of the oblique stroke in the above rule
makes it a context-sensitive rule.
6.5.2 Various kinds of context-sensitive grammars

Very many different types and subtypes of context-sensitive rules can
be formulated. We shall confine our attention to those which fall
within the scope of phrase-structure grammars (formalized as a set of
ordered or unordered, optional or obligatory, recursive or non-
recursive, concatenating rules). Any such grammar that includes one
or more context-sensitive rules is defined to be a context-sensitive
phrase-structure grammar.
6.5. CONTEXT-SENSITIVE GRAMMARS 237
Within this class of grammars, one may distinguish those in which
X and Y, in a rule of the kind that has just been given by way of
exemplification, are restricted to cover only one symbol each; those
in which either X or Y, or both, may refer to a string of more than
one symbol; and so on.
We will tacitly assume that the class of context-sensitive grammars
with which we are concerned here is defined by the condition that
X and Y, in a rule of the following kind
A -> B/in the context X + . . . + Y
may refer, each independently, to any finite number of concatenated
symbols, but that A must be a single symbol. Furthermore, we will
assume that B cannot be identical with A, and cannot be 'zero'
(cf. 6.2.11). These conditions would allow the following rules as
* well-formed' within the system:
(a) P->)/in the context E+F+... + G
(b) P-> Q + R/'m the context E+ ... + G + H+K+L
(c) P-^R + S+T/in the context G+...+H
etc.
They would prohibit the inclusion in the grammar of rules such as
the following:
(d) P-P/in the context E+...+F
(e) P-> 0/in the context E+...+F
Rule (d) is 'ill-formed* because it 'rewrites' P as itself (i.e. it breaks
the condition that A and B must not be identical). Rule (e) contains
the 'zero* symbol (0) immediately to the right of the arrow: this
defines rule (e) as a deletion-rule (' rewrite P as zero' means ' delete P ' :
the 'output* of rule (e), if it were allowed in the system, would be
E+F, derived by the rule from the 'input' E+P+F).
It should be noticed that in the above rules some symbols are
italicized, whereas others are not. The symbols printed in italics are
constants, the others are variables. We shall return later to the
distinction of constants and variables (cf. 6.6.6). For the purposes of
this section, the difference is as follows: if a symbol occurs in any
part of a rule as a constant, this means that the rule applies to
occurrences of that particular symbol; if the symbol occurs as a
variable, this means that the rule applies to any of the constants that
are defined to fall within the class referred to by the variable.
In this section, the difference is relevant only to the extent that it
distinguishes the actual rules of the grammar (which, we will assume,
do not contain any variables) from an abstract specification of the
format of the rules. In other words,
A -> B/in the context X +... + Y
is not a rule, but summarizes a whole class of rules, according to the
* values' defined as permissible for the variables A, B, X and Y. The
conditions that restrict the scope of the variables have been given
above.
6.5.3 Context-sensitive grammars include context-free grammars

If we add a further condition, we can define context-free phrase-
structure grammars to be a subclass of context-sensitive grammars.
The condition is that the * value' of X and Y may be left unrestricted
in rules of the form
A - B/in the context X +... + Y
We will first distinguish, for purely terminological convenience,
between the null-value (0) 2nd positive values (any permissible * value',
other than 0, that is assigned to a variable). Now, if it is specified, for a
particular rule, that the contextual variables, X and Y, are unrestricted
as to ' value* (each, independently, may be either positive in * value'
or null), the rule in question is context-free. If the 'value' of either
X or Y is restricted, i.e. specified to be either null or positive, then the
rule is context-sensitive.
To take a few examples. The following rules are all context-
sensitive
(f) P -> Qfrn the context 0 +... + 0
(g) P-><3/in the context 0 + ...+R + S
(h) P-> Ql'm the context T+... + 0
Rule (f) says, in effect, that P is to be * rewritten' as Q only if it has
no other symbol to its left and to its right in the * input' string.
(Normally, the only symbol that would satisfy this contextual condi-
tion is the initial symbol, 2. But one might have a system of rules in
which the initial symbol had been 'rewritten' as P.) Rule (g) says that
P is to be 'rewritten' as ), only when the 'input' string is P+R + S;
and rule (h), that P is to be'rewritten', only when it occurs in the final
position of the 'input' string in questionT+P. These rules may
also be put in another form (and frequently are in published
grammars)
(P) 0 + P+ 0-+0
(g') 0 + P+R +
(h') T+P+0-+T+Q + 0
Generalizing this format (by using A, B, X and Y are variables) we can
say that all the rules of a context-sensitive grammar are of the form
And a context-free rule of the form
may be taken as a special case of a context-sensitive rule in which the

* values' of X and Y are unrestricted.
It is along these lines that the theory of phrase-structure grammars
has been generalized to bring both context-free and context-sensitive
rules within the same formal framework. We may now move on to a
less abstract discussion of the implications of context-sensitivity in
language.
6.5.4 Concord and government

One of the most obvious of the phenomena in language falling within the
scope of the notion of * context-sensitivity' is concord (or * agreement').
We shall first of all give a fairly traditional account of the principles.
In many languages, the constituents of a particular syntactic
construction are said to * agree', or be 'in concord', with respect to
such features as ' gender', * number', 'case', 'person', etc. (The terms
'gender', 'number', etc., as well as the terms 'subject' and 'object',
will be fully discussed from a theoretical point of view in chapters 7
and 8.) For example, in noun-phrases in French the article and the
adjective must be in agreement with the noun in number ('singular'
v. 'plural') and gender ('masculine' v. 'feminine'): cf. un Hvre
interessanty 'an interesting book', v. une piice interessante, 'an
interesting play'; and des livres inter essants, '(some) interesting
books' v. des pieces inter essantes, ('some) interesting plays'. Here un
and interessant are 'masculine, singular' in concord with Hvre; une
and interessante are 'feminine, singular' in agreement with piece;
interessants is 'masculine, plural' in agreement with livres, and
interessantes 'feminine, plural' in concord with pieces; des is
'plural', but unmarked for the distinction of 'masculine' and
'feminine*. (For simplicity, we are referring to the concord-system
of written French. We have already mentioned that written and
spoken French are to a certain extent different languages: cf. 1.4.2.)
As in French, so also in Italian, Spanish, German, Russian, Latin,
and in many other languages of the world, the adjective must agree
with the noun in constructions similar to the one illustrated. Again, in
many languages, the verb must agree with either the subject or the
object, or both, in number, gender or person. Part of the relatively
complex gender-and-number concord of Swahili will be illustrated
in the following chapter (cf. 7.3.5); and the 'ergative' concord-
system of Eskimo will be referred to in the sections on 'transitivity*
and 'ergativity' (cf. 8.1.6, 8.2.2).
By comparison with the languages mentioned in the previous para-
graph, English has relatively little concord, Within noun-phrases it is
only the demonstrative pronouns which agree with the noun they
'modify': cf. this book, that book v. these books, those books. However,
concord is operative with respect to the verb in certain 'tenses' (more
precisely, it is not simply 'tense' that is involved, but also 'mood'
and 'aspect': cf. 7.5.1 ff.). The following simple sentences illustrate
the principle, as it applies to English, that the verb is in agreement
with the subject:
(1) He goes v. I/You) We/They go
(2) He has gone v. I/You/We/They have gone
(3) He is going v. You)We /They are going v. / am going
(4) He 11 was going v. You j'We I They were going
In the first two sentences, the person-and-number concord operates
in the following manner: if the subject is 'third person, singular' {he,
she, it, someone, etc., or a noun-phrase with a singular noun as its
head, e.g. John, the boy), the verb is 'third person, singular'; whereas,
if the subject is not' third person, singular', the verb is in what we will
refer to as the 'unmarked' form. In (3) and (4), which both contain
what is traditionally called 'the verb to bey (here employed as an
'auxiliary verb'), concord operates somewhat differently: in (3),
there is once again 'third person, singular' concord, but also 'first
person, singular' concord (/ am); and in (4), 'first person, singular'
and 'third person, singular' fall together under subject-verb concord.
With the 'simple past tense' of the verb (e.g. went, loved), there is no
subject-verb concord: cf. He\I\You\We\They went, etc.
Concord, as illustrated above, is usually distinguished from govern-
ment (or * rection', in the usage of some authors). For example, in many
languages the verb is said to 'govern* its object in a particular case:
e.g. Latin (Ego) amo te, 'I love you*, (Ego) suadeo tibi, 'I advise
you* (te v. tibi, 'accusative' v. 'dative', are governed by, or dependent
on, the verbs amo v. suadeo). In Latin (Russian, German, etc.), not
only verbs, but also prepositions, may govern the noun, pronoun or
noun-phrase dependent upon them in a particular case: e.g. ad urbem,
'to the city' (ad 'takes the accusative': urbem) v. ab urbey 'from the
city' (ab 'takes the ablative': urbe). For a fuller discussion of what is
meant by such expressions as 'takes the ablative', cf. 7.4.1 ff.
It will appear from the above examples that the difference between
concord and government lies in the fact that under concord two or
more words or phrases are 'inflected' for the same category (e.g.
number or person), whereas under government the principal and the
dependent member of a syntactic construction do not both exhibit the
same category: instead the dependent member is determined with
respect to the relevant category (e.g. case) by the principal member.
We have just stated the difference between concord and government
in a traditional manner; and it is a traditional distinction. In more
recent grammatical theory, the distinction is frequently drawn with
respect to the prior recognition of a distinction between endocentric
and exocentric constructions (cf. 6.4.1). For instance, Hockett says
that government is found 'only in exocentric constructions' (of a
particular subtype): the Latin ad urbem is exocentricit differs distri-
butionally from its constituents, ad and urbem. Concord, on the
other hand, 'is found in endocentric constructions, and in a tie that
cuts across hierarchical [i.e. immediate-constituent] structure to link
certain predicate attributes to subjects'. In other words, concord is
found in both endocentric and exocentric constructions: the French
un livre inter essant is endocentric, since its distribution is identical with
that of un livre; but Le livre est interessant (which also manifests
concord between livre and interessant), 'The book is interesting', is
exocentric, because its distribution (in so far as the notion of distri-
bution is relevant to sentences: cf. 5.2.1) differs from the distribution
of le livre, on the one hand, and est interessant, on the other.
Although what Hockett and other authors have to say in this
connexion is correct as far as it goes, it should be pointed out that
there is both a principal and a dependent member in constructions
which manifest concord: it would be incorrect, for example, to
maintain (as some linguists have maintained) that the person and
number of the subject is determined by the person and number of
the verb. It would be equally incorrect to say (and this view has been
expressed even more frequently) that neither the subject nor the
verb determines the other, but that both the subject and the verb
manifest a category which pertains to the construction of which they
are members. As we shall see later, number and person are nominal
categories, which, if they are manifest in a given language, may be
marked, inflexionally or otherwise, in the 'surface-structure' of the
verb-phrase (cf. 7.2.6, 7.3.1). This is implicit in the traditional
formulation of the facts: 'the verb agrees with the subject in number
and person*. We shall assume that this view is correct in our discus-
sion of context-sensitive rules in English in the following paragraphs.
The distinction between 'surface-structure* and 'deep-structure*
will be explained in a later section (cf. 6.6.1). Once we draw this
distinction, it will become clear that the difference between concord
and government (especially if it is made with reference to the notions
of 'endocentric5 and 'exocentric*) is essentially a 'surface-structure*
distinction.
6.5.5 A context-free interpretation of subject-verb concord

Let us now consider how we might generate some simple English
sentences that manifest subject-verb concord. For simplicity, we will
restrict our attention to transitive sentences in the 'present tense*:
examples are
(id) The dog bites the man
(ib) The dog bites the men
(ic) The dogs bite the man
(id) The dogs bite the men
(2 a) The chimpanzee eats the banana
etc.
We will assume a lexical classification of the occurrent words: for
simplicity, the lexicon may be regarded as a list of the members of
the terminal classes of the grammar (T> N, V) similar to that given in
chapter 4 (cf. 4.3.2).
Consider, now, the following context-free grammar:
K)
\NPplur+VPplur
(2) VPHng->
(3) VPplur^Vplur
(5) NPsing->T+Nsing
(6) NPplur^T+Nplur
(7)Nsing->N+0
(9)
(10)
There are other ways of generating the sentences in question by

means of context-free rules, but this set of rules will serve for the
purpose of illustration. It is assumed that it would be supplemented
by further rules for the phonological realization of the words occurring
in sentences, such that man+0 is 'rewritten* as /man/, man + s is
'rewritten* as /men/, hit+0 as /hit/, hit + s as /hits/, and so on. It may
be observed, in passing, that phonological realization rules of this kind
(e.g. hit + 0 -> /hit/) break one of the conditions imposed upon simple
phrase-structure grammars abovethey rewrite more than one
symbol at once (cf. 6.2.11). The phonological realization rules must
operate, of course, after the lexical-substitution rules (cf. 4.3.2).
Throughout this book we have assumed that the lexical-substitution
rules are outside the grammatical system properly so called.
Let us now interpret the grammatical rules that have just been
given: a sample sentence with its structural description, in terms of
this grammar, is represented in Fig. 13. The implication of the choice
introduced by rule (1) is that there are two unrelated sentence-types,
which we might call 'singular sentences' and 'plural sentences'. The
dog bites the men is a 'singular sentence', and The children drink the
milk a 'plural sentence', according to this interpretation of the
grammar. In other words, number is defined to be a category of the
sentence for subject-verb concord. But number is also introduced
into the object noun-phrase by rule (4). The grammar, therefore,
treats the choice between 'singular' and 'plural' in the subject-
position of the sentence as something quite different from, and totally
unrelated to, the choice between 'singular' and 'plural' in the object-
position. It fails to represent the fact that number is a category of the
noun-phrase (strictly speaking of the noun, or pronoun, rather than
the noun-phrase: but we may neglect this difference here); that the
choice between 'singular' and 'plural* is independently made in the
subject-position and object-position; and that the prior determination
of the subject as either 'singular' or 'plural' subsequently determines
the verb as either ' singular' or ' plural' according to the principles of
concord discussed above (cf. 6.5.4).
chimpanzee
banana
Fig. 13. Concord within a context-free grammar (the numerals attached to
the branches refer to the rules of the grammar in 6.5.5).
6.5.6 A context-sensitive interpretation of subject-verb concord

Let us now suppose that we wish to formalize the fact that the
number of the subject determines the number of the verb (within
the framework of a phrase-structure grammar). One way of doing so
would be as follows (and this is essentially the way adopted by
Chomsky in Syntactic Structures: for reasons which we will not discuss
here he puts the equivalent of rule (4) of the following system among
the transformational rules of his grammarbut we are still operating
without transformational rules):
(1) X-+NP+VP
(2) VP-+Verb + NP
( \ V h iV+Sl[n tthe
h e Ccontext
O n t e X t NP NPinn+...
sing
W
~* \V+ 0/in the context NPplur + ...
(5)NP9lno->T+N+0
(6) NPplur->T+N+s
There are a number of differences between this set of rules and the set
given in 6.5.5. I* *s shorter, with six rules rather than ten. Rules (5)
and (6) analyse 'singular' and 'plural' noun-phrases into three
constituents, of which one is an article, one a noun and the third a
member of the category of number (either * singular* or ' plural'). At
the same time, it should be noticed that there is no formalization of
the fact that the noun is the head of the noun-phrase. Although many
other shortcomings of the analysis assigned to such sentences by a
phrase-structure grammar can be remedied, as we shall see, by the
introduction of transformational rules, this particular type of
deficiency (and we will assume it is a deficiency) requires a rather
more fundamental revision of the formalization (cf. 7.6.8).
For the moment, however, we are concerned solely with the
question of context-sensitivity. The important point is that rule (1)
defines all sentences generated by these rules to be sentences of the
same type (NP+VP); rule (3) makes number a category of the
noun-phrase, independently of the occurrence of the noun-phrase as
subject or object; and rule (4) says that number in the verb (more
precisely, in V: we will not go into the reasons why Verb and V are
distinguished in this set of rules) is determined by the number of the
preceding noun-phrase. Rule (4) is a context-sensitive rule of the form
A-> B/in the context X + . . . + Y , where X is (in this case) assigned
a positive 'value* and Y is left unrestricted (cf. 6.5.3). (^ wu*l be
observed that the rule can be formulated in this way only within a
system of concatenating rules: cf. 6.1.1. It is the left-hand noun-
phrase, not the right-hand noun-phrase, which controls concord.
Since the left-hand noun-phrase derives from the NP generated by
rule (1) it may be interpreted, as we shall see later, as the subject
rather than the object: cf. 7.6.2.) As far as it goes, therefore, the
above system of six rules correctly formalizes the facts of concord
in English (in this deliberately restricted class of sentences). The
analysis of a sample sentence is given in Fig. 14.
It requires but little consideration to see that, as soon as we start
9 LIT
increasing the class of sentences generated by the grammar to
include all the other kinds of concord referred to in 6.5.4, tne number
of rules required in a context-free grammar would increase consider-
ably. There would be little difference in the number of rules added to
the context-sensitive grammar. Independently of any evaluation of
the two grammars in terms of economy of statement, however, we
wTish to say that the context-sensitive rules correctly formalize the
facts of concord. This brings us to a discussion of the distinction
between 'strong' and 'weak' adequacy.
NP VP
(3) \
NPsing Verb AT
/
A/
4/ \ i
/(4)\
s
(3)
N Pplur
T / K
eat /
T N
the
chimj.ânzee
the banana
Fig. 14. Concord within a context-sensitive grammar (the numerals attached
to the branches refer to the rules of the grammar in 6.5.6).
6.5.7 Strong and weak adequacy

We have already explained what is meant by strong and weak
equivalence (cf. 6.2.12). Two or more grammars are weakly equiva-
lent, if they generate the same set of sentences; they are strongly
equivalent, if they generate the same set of sentences with the same
structural description assigned to each sentence. The distinction
between weak and strong adequacy is somewhat similar.
A grammar is said to be weakly adequate if it generates the set of
sentences that one wishes to generate; a grammar is strongly adequate
if it not only generates the desired set of sentences but also assigns
to each sentence the correct structural description.
The term 'correct' is controversial, not to say tendentious. But it
must be remembered that this definition of the difference between
weak and strong adequacy carries with it absolutely no implications
about the interpretation of * correct': it does not even prejudge the
question whether there are any standards of ' correctness '. However,
we shall assume that, in certain cases at least, it is possible to say of
two alternative descriptions of the same set of sentences that one
description is ' more correct' than the other, or ' correct' in certain re-
spects, even though we might not wish to say that any description
is absolutely * correct'.
It will be evident that the two sets of rules introduced in 6.5.5 an<^
6.5.6 are weakly, though not strongly, equivalent. There can be no
doubt that the context-sensitive system is more strongly adequate (more
* correct') than the context-free system wTith regard to its characteriza-
tion of the principles of subject-verb concord in English. It was for
this reason that we deliberately introduced the notions of * strong'
and 'weak' adequacy at this point. The two grammars are readily
comparable, because they are weakly equivalent and they both
conform to the same general conditions imposed by the formalization
of phrase-structure grammars (cf. 6.5.3).
6.6 Transformational grammar

6.6.1 Deep and surface structure
We may begin our discussion of transformational grammar with a
quotation from Hockett: ' . . .consider English Atoms are too small to
see by any possible technique and They are too much in love to see clearly.
The surface grammar of the two sentences is much the same, though
not identical. But cutting across the surface grammar is a difference
in the deeper connexions of to see. In the first sentence, atoms and see
are related as they are in You can't see atoms \ in the second, they and
see are related as they are in They see you' In the same chapter,
Hockett says of the ' various layers of deep grammar' that lie ' beneath'
the ' surface grammar' (the 'one layer' which 'is immediately apparent
to the analyst') that, although 'they have much to do with how we
speak and understand', they 'are still largely unexplored, in any
systematic way, by grammarians'.
By the time that Hockett's textbook appeared (1958), Chomsky and
Harris had published works (1956 and 1957) in which proposals were
made for the systematic analysis of both the surface structure and the
9-2
deep structure of sentences by means of what they both called
* transformational' rules. Although Harris and Chomsky were
working in close association (and in their publications each has
acknowledged his debt to the other), they have from the outset
developed the theory of 'transformational' grammar along different
lines. More recently, the Soviet linguist, Shaumjan, has published a
book (1965) in which yet a third system of grammatical analysis has
been proposed making use of ' transformational' rules.
We shall not go into the differences between these theories here.
It is important to realize, however, that there is a serious ter-
minological problem in deciding what 'legitimately' counts as a
transformational grammar. If we tried to formulate a definition of
' transformational' in such a way that it would apply to the theories
of Chomsky, Harris and Shaumjan, then we should probably find that
it also applied to many other theories of grammatical structure which
are described by their authors as 'non-transformational'.
The purpose of this assertion (which is not intended to be pole-
mical) is merely to draw the reader's attention to the fact that the
term ' transformational' is strictly defined, but differently defined, in
the formalizations proposed by Chomsky, Harris and Shaumjan.
The term ' transformational' has unfortunately engendered a good
deal of unnecessary controversy and confusion in the recent literature
of linguistics. If we use the term in a general and rather informal
sense, rather than in the particular sense in which it is defined in tny
one theory, we can say, quite reasonably, that the ' deeper connexions'
between sentences which 'cut across the surface grammar' (cf. the
quotation from Hockett given above) are transformational relation-
ships: this is a perfectly legitimate use of the term 'transformational'.
Many of these transformational relationships between sentences are
well-recognized in traditional grammar; but it is only recently that
linguists have made any progress in accounting for them in an
explicitly generative framework. Any grammar that claims to assign
to each sentence that it generates both a deep-structure and a surface-
structure analysis and systematically to relate the two analyses is a
transformational grammar (whether it uses the label or not).
Of the many theories that are transformational in this wide, but
quite reasonable, sense of the term, the best-known and so far the
most highly-developed is that of Chomsky. In what follows, we shall
therefore restrict our attention to the generative system which he and
his associates have developed, and applied to the description of parts
of the grammatical structure of many languages, over the last few
6.6. TRANSFORMATIONAL GRAMMAR 249
years. This system, like all others, has a number of known inade-
quacies. But so far it has proved itself capable of handling an im-
pressively large number of the * deeper connexions' between sentences
that we have described as 'transformational*.
6.6.2 Transformational ambiguity

One reason for introducing the notion of 'bracketing* (or consti-
tuent-structure) into the theory of grammar is that it enables us to
account systematically for various kinds of grammatical ambiguity
(cf. 6.1.3). But there are many other ambiguous constructions in
different languages which depend upon the ' deeper connexions* we
have decided to call transformational rather than upon a difference of
' bracketing*.
Let us begin by taking a well-known example from traditional
grammar. The Latin phrase amor Dei, like its English translation the
love of God, is ambiguous (out of context). Traditional grammars of
Latin would say that the word Dei ('of God') is either a subjective or
objective 'genitive*. This is a transformational explanation of the
ambiguity: it implies that the phrase amor Dei is related to, and
indeed in some sense derivable from, two sentences: (i) a sentence in
which Deus (cited now in the 'nominative* case: cf. 6.2.9) *s t n e
subject of the verb amare ('to love*); (ii) a sentence in which Deum
(cited now in the 'accusative') is the object of the verb amare, (From
our discussion of the traditional concept of the 'word' as a lexeme,
cf. 5.4.4, it will be clear that these statements can, and should, be
reformulated in terms of the lexemes DEUS and AMO.) Similarly, the
love of God is related to two sentences in English: (i) a sentence in
which God is the subject of the verb love (cf. God loves mankind)]
(ii) a sentence in which God is the object of the verb love (cf. Mankind
loves God), The phrase might still be ambiguous in particular sentences:
It is the love of God which inspires men to work for their fellows.
One of Chomsky's most famous examples, the phraseflyingplanes
(in a sentence such as Flying planes can be dangerous), is ambiguous for
much the same reasons as the love of God is ambiguous: under one
interpretation flying planes is related to a sentence in which planes is
the subject of fly or areflying,under the other to a sentence in which
planes is the object oifly (cf. Planes fly v. Johnfliesplanes).
In traditional grammar, there is a distinction drawn between the
'participle' and the 'gerund*. In so far as this distinction applies in
English (and there are situations in which it is unclear) it might be
25O 6. GRAMMATICAL STRUCTURE
formulated as follows: (a) A participle is a word which is derived from
a verb and used as an adjective, (b) A gerund is a word which is
derived from a verb and used as a noun. This distinction is clearly
relevant to the analysis of an ambiguous phrase like flying planes. If
we consider the following two sentences (in which the principles of
subject-verb concord * disambiguate' the phrase in question)
(1) Flying planes are dangerous
(2) Flying planes is dangerous
the difference between the 'participle' and the 'gerund' comes out
quite clearly. The verb are in (1) is 'plural' because its subject is
planesy a plural noun, which is the head of the endocentric phrase
flying planes: moreover, in (1) flying is distributionally equivalent to
an adjective (e.g. supersonic). The recognition of a head and a modifier
in flying planes in (2) is more problematical: but flying is nominal and
the whole phrase is the subject (cf. Flying is dangerous). Traditional
statements about the 'participle' and the 'gerund' are transforma-
tional in nature. We can interpret them to mean that a particular
word (in the sense of 'lexeme': cf. 5.4.4) may be 'verbal' in one
sentence and 'adjectival' in a transformationally-related phrase, or
'verbal' in one sentence and 'nominal' in a transformationally-
related phrase. Without, for the moment, considering the nature of the
rules which might account for these relationships, let us merely say
that in (1) the phrase flying planes is to be derived by a rule which
'transforms' the structure underlying a sentence like Planes are flying
and assigns to the resultant noun-phrase the derived structure of
adjective + noun; and that in (2) the phrase flying planes is to be
derived by transformational rule from the structure underlying a
sentence like Johnfliesplanes and assigns to the resultant noun-phrase
the derived structural description noun + noun (the first of the two
nouns, if any, being the one that controls concord). If we now assume
that the rules of the grammar generate sentences like (1) and (2) with
both an underlying ('deep') and a derived ('surface') structural
description, we have in principle explicated the 'subjective' and
'objective' interpretations of noun-phrases like flying planes.
Consider now a phrase like eating apples: this is also ambiguous.
Under one interpretation (cf. to eat apples and to fly planes) it is
structurally comparable with flying planes in (2). But the other
interpretation, which is illustrated by
(3) Eating apples cost more than cooking apples
cannot be accounted for by saying that apples is in a 'subjective*
relationship with eating in the deep structure of (3). The subjective
interpretation of eating apples might be possible in somewhat unusual,
or bizarre, situations in which apples are * personified * (to use the
traditional term). In such situations a sentence like the following
(4) Apples eat with a hearty appetite
would, presumably, be equally acceptable. Let us grant, however,
that (4) is 'abnormal'; and that, whatever account we give of its
' abnormality ', this account simultaneously explains the' abnormality'
of the 'subjective' interpretation of eating apples.
There are many phrases of the form V+ing + N which are
multiply-ambiguous: indeed, one might maintain that flying planes
can be interpreted in the sense suggested by the paraphrase 'version'
planes for flying. In the case of flying planes, this third interpretation
is perhaps tautologous. It is quite likely, however, that any grammar
which defines eating apples, etc., to have at least two deep-structure
analyses will also assign at least three analyses toflyingplanes.
The reason why these phrases are said to be grammatically, and
not just semantically, ambiguous is essentially the same as the reason
given in 6.1.3. To illustrate this point with reference to another of
Chomsky's examples (which is very similar to the love of God
discussed above): a phrase like the shooting of the hunters is ambiguous
(if it occurs in a context which does not ' disambiguate' it) because
(a) shoot may be used both 'transitively' and 'intransitively' (more
precisely, both 'transitively' and 'pseudo-intransitively': cf. 8.2.11),
and (b) the hunters may occur in sentences containing the verb shoot
as either the subject of the 'intransitive' (e.g. The hunters shoot) or
the object of the 'transitive' (e.g. John shot the hunters). It is worth
pointing out that the objective interpretation of the phrase the
shooting of the hunters is closely related to passive constructions:
cf. The hunters were shot (by John). With a 'fully transitive' verb
(i.e. with a verb which has an overt and specific object) phrases of the
form the V+ing of NP do not normally admit of a subjective inter-
pretation: they cannot be extended with an objective of NP (*the
shooting of the hunters of the deer). Instead, the subjective NP takes
the 'possessive' suffix and the objective NP the preposition of: cf. the
hunters' shooting of the deer. In a later chapter (cf. 8.2.3 R-)> w e shall
see that this is but one of many facts which suggest that in English
(as in many languages) there is a particular relationship between
the object of a transitive verb and the subject of a corresponding
intransitive verb. For the present, however, we may be content with
the informal account given above of the conditions which determine
the ambiguity of phrases like the shooting of the hunters.
Let us now introduce the purely ad hoc convention (which is
frequently used for this purpose in the literature) of employing
numerical subscripts to identify the words and phrases which are
said to be in correspondence in transformationally-related construc-
tions. For example, we will say that a sentence like John shoots the deer
has the form NP1 Vir NP2(NP1 = John; Vir stands for a particular
member of the class of transitive verbs, shoot; and NP2 = the
hunters); and a sentence like The hunters shoot has the form NPX Vinlr
(NP1 = the hunters, Vintr stands for a particular member of the
class of intransitive verbs). Given this convention, we can say that a
phrase of the form the V-{-ing of NP is grammatically ambiguous (and
may or may not be semantically ambiguous) if, and only if, the
grammar generates sentences of the form
(5)NP1VtrNP2
and
(6) NP, Vinlr
and if (a) the V of the V+ ing of NP is identical with a member of Vtr
in (5) and a member of Vintr in (6), and (b) the NP of the V+ing of
NP can occur both as NP2 in (5) and NPX in (6). These conditions
are satisfied in the case of the shooting of the hunters. But are they
satisfied in the case of the eating of the apples} The verb eat (for
simplicity, we will assume that the 'transitive' and the 'intransitive',
or ' pseudo-intransitive', eat are instances of the ' same' verb, although
this begs certain theoretical questions) occurs in sentences of the form
represented in (5) and (6): cf. John eats the apples and John is eating.
The phrase the eating of the apples is therefore interpretable 'objec-
tively' (the apples is NP2 in a transitive sentence with the verb eat).
Whether it is defined as being 'subjectively' interpretable, from the
syntactic point of view, will depend upon the generation or exclusion
of a sentence like The apples are eating. The point is that a phrase like
the eating of the apples manifests the same ' deep' relationship between
the apples and eat as does the sentence (or non-sentence) The apples
are eating. In other words, either the 'subjective' phrase the eating of
the apples and sentence The apples are eating should both be generated
as grammatical (and systematically related to one another in terms of
their 'deep' structure) or they should both be excluded as ungram-
matical. And their grammaticality or ungrammaticality will depend
upon whether the noun apple and the verb eat are subclassified in the
lexicon (by means of grammatical features': cf. 4.3.3) in such a way
that the grammatical rules will admit or prohibit the combination of
a noun with a given * feature* (e.g. [inanimate]) as the subject of the
verb-class of which eat is a member. (It is the same principle that is
involved as was discussed in connexion with the constituent-structure
analysis of fresh fruit market and new fruit market in 6.1.3).
6.6.3 Neutralization and diversification in syntax

Reference was made in an earlier chapter to the * indirect discourse'
construction in Latin (5.2.2). It was suggested (and this is a very
traditional view) that a sentence like Dico te venisse (or its English
translational equivalent / say that you came) should be generated in
two 'stages': by first generating a sentence in its * direct' form, (Tu)
venisti, and then converting it into the * indirect' form, te venisse,
'after' a 'verb of saying* (e.g. dico). This was a rather imprecise way
of formulating the principle. We shall see later that, in Chomsky's
system of transformational grammar, it is the ' deep' structure under-
lying (Tu) venistiy rather than this sentence itself, that is 'embedded'
as the object of dico (for the term 'embedding', cf. 6.6.8).
It is well-known that the embedding of a transitive sentence in the
' indirect discourse' construction may lead to ambiguity in Latin. For
example, (a) Clodia amat Catullumy 'Clodia loves Catullus', and
(b) Clodiam amat Catullus, 'Catullus loves Clodia', are distinguished
by the opposition of the 'nominative' and the 'accusative' case
(cf. 7.4.2). But Dico Clodiam amare Catullum results from the em-
bedding of either (a) or (b) as the object of dico; and, unless it is
' disambiguated' by the wider context, it is translatable into English as
either ' I say that Clodia loves Catullus' or ' I say that Catullus loves
Clodia'. (Word-order in Latin is irrelevant to the resolution of
ambiguities of this kind.) The sentence is ambiguous, because both
the subject and the object of, the embedded sentence are in the
' accusative' case and, in this instance, each of the two nouns could be
either the subject or object of amare. By contrast, a sentence like Dico
Horatium edere oleas, 'I say that Horace is eating olives', is presum-
ably not ambiguous; rather, it is ambiguous only if the underlying
structure of Horatium edere oleas is relatable to that of Horatium edunt
oleae ('The olives are eating Horace') as well as to that of Horatius
edit oleas ('Horace is eating olives').
Let us now consider the ' indirect discourse' construction in Greek.
There are a number of alternative constructions. Of these we will
mention only two: one that is very similar to the Latin construction
and one that is similar to the English construction with that. Certain
verbs require, or 'prefer', one construction rather than the other.
But some verbs of 'saying' occur freely with both. For example:
.((a) Kuron poreuesthai.
agg
1 (b) hoti Kiiros poredetai.
These are two alternative 'versions' of the same underlying structure;
both are translatable as 'He announces that Cyrus is on the march'.
In (a) Kuron is 'accusative' and poreuesthai is the 'infinitive'; in
(b) hoti is an introductory particle, or conjunction, comparable with
that in English, and Kuros poreuetai is identical in form with the
'direct discourse' simple sentence.
As we have seen, the distinction between dico + Clodia amat
Catullum and dico + Clodiam amat Catullus (the plus-sign is here used
as an ad hoc device to refer to what was described above as 'em-
bedding') is neutralized in the surface-structure of Dico Clodiam
amare Catullum (for the term 'neutralization', cf. 3.3.5). The Greek
example illustrates the converse phenomenon, which (following
Lamb, who might, however, describe the facts in a somewhat
different framework) we may call diversification: two sentences may
differ in surface-structure, but be identical in their deep-structure.
(The English sentences They denied the existence of God and They
denied that God exists also illustrate the phenomenon of diversifi-
cation.)
Although we have so far discussed transformational relations with
particular reference to neutralization and diversification (and we have
deliberately adopted a traditional viewpoint), it must be emphasized
that the distinction between surface structure and deep structure is
independently justifiable. Traditional grammars of English would say
that active and passive sentences (e.g. John opened the door and The
door was opened by John) are syntactically-relatable, the object of the
former 'becoming' the subject of the latter: this is a transformational
statement, which implies a deep-structure identity or similarity. The
relationship between active and passive sentences must be accounted
for, whether this is a case of diversification or not (and opinions differ
on this question). Furthermore, let us suppose, for the sake of the
argument, that the shooting of the hunters (and all phrases of the form
the V+ing of NP) are not ambiguous, but necessarily 'objective',
and that the hunters' shooting (and all phrases of the form NP's
V+ing) are necessarily * subjective \ We should still want the grammar
to relate the shooting of the hunters to the transitive sentence NP1 shoot
the hunters and the hunters' shooting to the intransitive The hunters
shoot.
The phenomena of syntactic neutralization and diversification,
which are widespread in language, bring the distinction between deep
and surface structure into sharper focus.
6.6.4 PS-rules and T-rules

Our discussion of transformational grammar has so far been very
informal; and it has been more or less neutral with respect to various
possible systems of rules which seek to formalize the relationship
between deep structure and surface structure. At this point we will
move on to consider the theory of grammatical structure which
Chomsky developed in Syntactic Structures. Later in the section we
will mention some of the more important modifications that have been
introduced in the last few years.
For a number of reasons (not all of which have been mentioned
above) Chomsky came to the view that ' notions of phrase structure
are quite adequate for a small part of the language and that the rest
of the language can be derived by repeated application of a rather
simple set of transformations to the strings given by the phrase
structure grammar. If we were to attempt to extend phrase structure
grammar to cover the entire language directly, we would lose the
simplicity of the limited phrase structure grammar and of the trans-
formational development.'
Chomsky's proposal, therefore, was to split the syntax of the
language into two parts: (i) a phrase-structure component (or base-
component), containing rules of the form X + A + Y - > X + B + Y
(cf. 6.5.2), and (ii) a transformational component, containing 'sup-
plementary' rules not necessarily constrained to the form of the
phrase-structure rules. The transformational rules were to b e ' supple-
mentary ' in the sense that they were to operate after, and upon the
output of, the phrase-structure rules. Henceforth, we will use the
abbreviations 'PS-rule' and 'T-rule' for rules of the base-component
and transformational component, respectively. The PS-rules given in
Syntactic Structures are as follows:
(1) X-+NP+VP
(2) VP->Verb + NP
p
1
sing
(4)NP8ing->T+N+0
(5)NPplur->T+N+s
(6) Verb-> Aux+V
(7) Aux -> C(M) (have + en) (be + ing)
We have omitted the rules for lexical substitution from the PS-rules
(cf. 4.3.2). Instead, we give Chomsky's sample lexicon in the following
form (the difference is not relevant to any of the points being
illustrated):
T = {the}
N = {man, ball, . . . }
V = {hitf take, walk, ready . . . }
M = {willy can, mayy shall, must}
The only rule that requires any comment at this stage is (7): all the
others conform to conventions used in previous sections. Rule (7)
develops Aux (which has a mnemonic connexion with 'auxiliary
verb') into a string of up to four elements: of these C is obligatory
(and will be interpreted by a transformational rule to handle concord);
the other elements on the right-hand side of rule (7) are optional, and
each is independent of the other two. The output of rule (7) will
therefore contain (in place of Aux in the output of rule (6)) one of the
following eight strings:
(i) C
(ii) C+M
(iii) C + have + en
(iv) C+be + ing
(v) C + M+have + en
(vi) C + M+be + ing
(vii) C+have + en + be + ing
(viii) C+M+have + en + be + ing
Other features of rule (7) will be explained later.
The output of the PS-rules is a kernel string. The reader must be
careful not to confuse this term with kernel sentence. A kernel sentence
(according to the system proposed in Syntactic Structures) is any
sentence which is generated from a single kernel string without the
application of any optional transformations. No sentences are generated
without the application of at least a limited number of obligatory
T-rules. This point is important, since it has frequently been
assumed that kernel sentences (e.g. the simple, active, declarative
sentences of English) were to be generated solely by the phrase-
structure rules of a transformational grammar.
6.6.5 The passive transformation

We will illustrate the nature of the T-rules of Syntactic Structures
with reference, first of all, to the generation of passive sentences in
English. (The fact that Chomsky has modified his view of the relation-
ship between active and passive sentences in his more recent wTork is
not relevant to the points being made here.) The relevant T-rule, as
given in Syntactic Structures, is as follows:
Passive (optional)
Structural Analysis: NP-Aux- V-NP
Structural Change: X 1 - X 2 - X 3 - X 4 - ^
(An alternative, less formal but equivalent, statement of the rule is as

follows:
NP1-Aux-V-NP2->NP2-Aux + be + en-V-by + NP1
There is some significance in the distinction of dashes and plus-signs
as concatenation symbols, but we need not go into this question.)
The first point to notice is that the above T-rule contains two parts:
* Structural Analysis' and * Structural Change' (henceforth SA and
SC, respectively). Unlike PS-rules, which apply to the output of the
rule previously applied (whether optionally or obligatorily, recur-
sively or non-recursively, with,contextual restrictions or without, etc.),
T-rules are defined to apply only to strings that are analysable in
terms of the elements referred to in their SA. In the present instance,
the transformation (the SC part of the rule) is restricted to strings
which can be analysed into A7P> Aux, V and NP. Each of these four
elements is identified in the left-hand part of SC by means of the
subscript attached to the symbol X: Xx denotes the first element
referred to in SA (i.e. NP); X2 denotes Aux\ etc. The transformation
(defined in SC) has the following effects: (a) the first and the fourth
elements are permuted; (b) by is attached to what is now the last NP
in the string; and (c) be + en is attached to Aux. In other words, the
effect of the transformation is precisely that suggested in the alter-
native, less formal, statement of the rule
NP1-Aux-V-NP2->NP2-Aux + be + en-V-by + NP1
We must now see what is meant by saying that a string is analysable
in terms of the four elements NP, Aux, V and NP. Let usfirstrun
through the PS-rules in sequence, listing the strings that result from
the application of each rule to the output of the previous rule
(cf. 6.6.4):
By rule (1): NP+VP
(2): NP+Verb + NP
(3): NPsing+Verb + NPsing
(4): T+N+0+Verb+T+N+0
(6): T+N+0 + A11X+V+T + N+0
(7): T+N+0 + C + M+have + en+V+T+N+0
It will be observed that rule (3) was applied twice. Since NPsing was
selected for both positions in the output of rule (2), rule (4) was also
applied twice, but rule (5) was inapplicable. In rule (7) Aux was
rewritten as C + M+ have + en (i.e. as option (v) of the eight listed in
6.6.5 a s possible under the notational conventions). The output of
rule (7) is a kernel string of the type that underlies such corresponding
active and passive sentences as The man will have read the book and
The book will have been read by the man.
We now wish to apply the passive transformation to this string,
provided that it is analysable in terms of the conditions specified in
the SA part of the rule. Notice, first of all, that none of the elements
referred to in the SA for the passive transformation occurs in the
kernel-string. Furthermore, at no stage in the derivation of the
kernel-string by the PS-rules did we meet the string NP + Aux+ V
+ NP. Let us therefore run through the rules again; but this time we
will generate the phrase-structure of the kernel string in question.
By rule (1) Z(NP+VP)
(2) I,(NP+VP(Verb + NP))
(3) X(NP(NPsing)+VP(Verb + NP(NPsing)))
(A)?:(NP(NPsing(T+N+0))
+ VP(Verb + NP(NPsing(T+N+ 0))))
(6) Z(NP(NPHno(T+N+0))
+ VP(Verb(Aux+ V) + NP(NPsing(T+N+ 0))))
(7) H(NP(NPsing(T+N+0))+VP(Verb(Aux(C +M
+ have + en) + V) + NP(NPsing(T+N+ 0))))
This is the phrase-structure of the kernel string underlying such
sentences as The man will have read the book and The book will have
been read by the man.
A labelled-bracketing of a string is referred to, technically, as a
phrase-marker. Phrase-markers may also be represented by means of a
tree-diagram with labelled nodes: cf. Fig. 15. We must now introduce
two further notions: * substring' and ' domination'.
A substring is any part of a string which is itself a string (of one or
more elements). For example, the string a + b + c is analysable in
various ways into substrings: into the two substrings a and b + cf or
a + b and c, or into the three substrings a and b and c. The kernel
string we are at present concerned with
V+T+N+0
can be analysed into very many different sets of substrings. What we
want to know is whether any one set of substrings satisfies the condi-
tions of analysability specified in the SA of the T-rule for the passive.
We now need the notion of 'domination'.
A symbol dominates everything enclosed within the brackets
opened immediately after the symbol in question in the phrase
marker (or, equivalently, a symbol dominates everything which is
traceable back to the node labelled with the symbol in question in the
corresponding tree-diagram for the phrase-marker). Thus, in the
phrase-marker we are considering: 2 dominates everything between
the left-most and the right-most bracket; VP dominates Verb
(Aux(C + M+have + en)+V) + NP; and so on.
We can now define the notion of structural analysability as a
condition for the application of T-rules. If a string (not necessarily
a kernel string) is analysable (without remainder) into substrings, each
of which is dominated in the phrase-marker for the string by a symbol
referred to in the SA of the T-rule, then the string satisfies the con-
ditions defined by the SA. If the T-rule in question is obligatory, it
must be applied; if it is optional, it may be applied. The passive trans-
formation is defined to be optional in the rule we are considering.
The (kernel) string in question satisfies the conditions for the
application of the passive transformation. This will be clear from the
following diagram (which may be interpreted with reference to the
phrase-marker in Fig. 15):
Aux - V - NP9
By the operation of the T-rule proper (in SC) the output will be
another string (no longer a kernel string: cf. 6.6.4), which will then
serve as the input, together with its associated phrase-marker, for
further T-rules. If all the relevant obligatory T-rules are applied, but
no more optional T-rules, the output of the generative system will be
a passive sentence like The book will have been read by the man.
C M have en
T N'
Fig. 15. Phrase-marker for a kernel sentence.
6.6.6 Heterogeneity of T-rules

We have gone into a certain amount of detail in our discussion of the
notion of analysability; and this is essential to an understanding of
Chomsky's formalization of transformational grammar. Many
popular accounts of the theory (even when they are not factually
incorrect or misleading) fail to give any indication of what is meant by
the term * transformational rule'.
One of the most striking features of the T-rules in Syntactic
Structures is their heterogeneity. It is important to realize that the
construction of even a partial transformational grammar for any
language is a task of immense complexity; and there are many different
factors which may influence the linguist in his decision to handle a
particular feature by means of a T-rule rather than a PS-rule. As we
shall see, not all the T-rules are ' inherently' transformational.
On the basis of our discussion of structural analysability, we may
distinguish two different criteria which define a rule as * inherently'
transformational. The first may be summarized as follows: any rule
which does not conform to the conditions imposed upon phrase-
structure rules is transformational. Some of these conditions have
been mentioned in an earlier section (cf. 6.2.11). If various formal
restrictions of this nature are imposed upon the rules of the base-
component of a transformational grammar, one may guarantee that
the kernel-strings at least are uniquely analysable into substrings in
terms of the conditions specified in the SA of a transformational rule.
The second criterion is somewhat different: a transformational rule
is one which contains in the string of symbols to the left of the
rewrite-arrow (or in the SA associated with the rule) at least one
symbol which functions as a variable, which takes as its * value' any
one of the whole class of substrings dominated by that symbol in the
phrase-marker associated with the string serving as input to the rule.
For example, all the symbols (except V) in the SA of the passive
transformation discussed above
SA: NP-Aux-V-NP
are variables in this sense. By contrast, V is a constant: it is a terminal
symbol of the system of PS-rules given in 6.6.4, dominating no
substring other than itself ('self-domination', in this sense, is a
formal requirement of the system). This distinction between 'con-
stants' and ' variables' (which may be regarded as a special case of the
more general distinction drawn in 6.5.2) is fundamental to the
definition of 'transformation' in Chomsky's system.
It is by virtue of their capacity to refer to classes of substrings, not
simply to a given substring, that the transformational rules are more
powerful than phrase-structure rules. The term 'powerful' is rather
difficult to define, at least in a non-technical way. Let us be content
with the statement that one rule is more powerful than another if it
accounts for more 'facts' or accounts for them more 'correctly'.
(Lurking behind this statement are various considerations of weak
and strong adequacy: cf. 6.5.7.) In particular instances, it is easy
enough to see what is implied by the term 'powerful'. We will give
just one example.
The system of PS-rules given in 6.6.4 allows for the generation
of only two substrings dominated by NP: T+N+0 and T+N+s
(e.g. the boy and the boys). But it is clear that in a more complete
grammar of English the passive transformation will be required to
permute noun-phrases of all kinds: cf. The man who was here yesterday
read a thousand books last year and the corresponding passive * version'
A thousand books were read last year by the man who was here yesterday.
Provided that all the rules of the grammar are correctly integrated in
the generative system, both the man who was here yesterday and
a thousand books will be analysed as * constants' falling within the
scope of the ' variables' NP1 and NP2; and the passive transformation
will operate correctly. Other rules will account for the various possible
positions of last year, the 'embedding' of who was here yesterday
within the first NP of the string that serves as input to the trans-
formational rule for the passive, etc. Apart from any other considera-
tions, a rule containing variables (in this sense of the term) can handle
whole classes of sentences which would otherwise require a lot of
separate (and unrelated) phrase-structure rules.
If the grammar is transformational, in the sense that it includes
some 'inherently' transformational rules operating upon the output
of phrase-structure rules, then a number of other important factors
come into play. Only two will be mentioned here. It might be
decided, for instance, to transfer to the transformational component
such 'complexities' as recursion and context-sensitivity (although
these properties, of themselves, do not violate the conditions of
analysability): for reasons which we need not discuss here, Chomsky
made this decision, in principle, for the system proposed in Syntactic
Structures. Another consideration has to do with the relative order in
which the rules of the grammar are applied. A perfectly 'straight-
forward' phrase-structure rule may be required to operate upon the
output of an 'inherently' transformational rule. If the system is
formalized in such a way that the T-rules come after the PS-rules
(and this is so in Syntactic Structures), then the 'straightforward'
phrase-structure rule must be ' reclassified' as a T-rule, by virtue of
the place it occupies in the grammar. (This is possible, because a
constant can always be regarded as a variable with only one 'value'.)
Considerations of this kind (and there are others) account for the
heterogeneity of the transformational rules in grammars organized in
terms of Chomsky's earlier formalization of the theory.
As we shall see, a number of modifications have recently been
introduced into the system by Chomsky and others working upon the
foundations of transformational theory. Analysability is guaranteed
by somewhat different formal devices in the system recently proposed
by Chomsky: but (to quote his own statement of the point) it remains
'the basic predicate in terms of which transformational grammar is
developed*.
6.6.7 Subject-verb concord in a transformational grammar

At some point in the description of English we must account for the
principles of subject-verb concord. We have already referred to these
in connexion with the notion of context-sensitivity (cf. 6.5.6). We will
now look briefly at the transformational rule which is introduced in
Syntactic Structures to handle the concord of number ('singular* v.
'plural') that holds between the subject of the sentence and the verb.
This will serve to illustrate some of the general points that have been
made above about the nature of T-rules.
The rule (modified in one respect) is as follows:
Number transformation (obligatory)
SA:X-C-Y
Q<- n fSI t H e C O n t e X t NP
*ing+
(o/otherwise
We now meet a different kind of SA from that which appeared in the
passive transformation. X and Y are variables denoting any substrings
of the input string without reference to their analysability; and C is a
terminal symbol of the PS-rules given in 6.6.4. ft *s t r u e t n a t t n e
contextual restriction for C -> s makes reference to the non-terminal
symbol NPsing. But we might consider the possibility of including the
present 'transformational* rule among the PS-rules. Notice that, if
the PS-rules of 6.6.4 a r e applied in the following order (1), (2), (3),
(6), (7), (4), (5), and if the rule for number-concord is applied after
(7) and before (4), the effect is exactly the same as it would be if we
applied the rule to a terminal string generated by the rules of 6.6.4
in the order in which they were listed there. The reader may verify
this fact for himself.
There is, however, a very g6od reason why Chomsky puts the rule
for number-concord at a particular place among the T-rules of
Syntactic Structures. The principles of subject-verb concord in
English apply both to active and to passive sentences: cf. The men see
the cows, The man sees the cows, The cow is seen by the men, The cow is
seen by the man, The cows are seen by the men, etc. If the passive trans-
formation (which is an 'inherently* transformational rule) precedes
the rule for number-concord, the latter can be formulated in such a
way that it applies with respect to the number of the NP occurring
to the left of C (the first element of Aux) in the input string.
Thus the kernel string
(1) T+N+0 + C+M+have + en+V+T+N+s
(which is an instance of NP1 AuxVNP2) will be transformed
into
{id) T+N+0 + s + M+have + en+V+T+N+s
and (2) T+N+s+C + M+have + en + be + en
+ V+by+T+N+0
(which is the corresponding NP2 Aux + be + enVby + NPV
resulting from the application of the optional passive transformation)
will be transformed by the rule for number-concord into
(20) T+N+s+0 + M+have + en + be + en
+ V+by+T+N+0
These are the strings underlying such sentences as The man will have
read the books and the corresponding passive The books will have been
read by the man.
Among the further transformational rules that must be applied to
(la) and (2a) are the rules for the attachment of the verbal suffixes
and the marking of word-boundaries. We shall not give the rules (the
reader should consult Syntactic Structures, pp. 39, 113: he will see
that the rule for the * auxiliary transformation* makes use of ad hoc
variables). The effect of these rules is to convert (la) and (za) into
(ib) ...M+s++have+ + V+en+ + ...
(2b) ...M+0+ +have++be + en+ + V+en+ + . . .
(We have given only the Aux V portions of the strings. The double
plus-sign is used in place of Chomsky's special boundary-symbol.)
Let us now apply the lexical substitution rules (cf. 4.3.2: there is
some divergence here from the Syntactic Structures system, according
to which the lexical items are the terminal symbols of the PS-rules),
using a space to indicate the word-boundaries
(1 c) the man + 0 will+$ have read + en the book + s
(2 c) the book + s will+ 0 have be + en read+ en by the man+0
Further rules (in Syntactic Structures ' morphophonemic' rules) will
now convert these strings into the correct phonological form:
> /man/, ;///+$->/wi I/, read + en -> /red/, etc. Alternatively,
for the written language,' morphographemic' rules would convert the
strings into
(id) The man will have read the books
(zd) The books will have been read by the man
6.6.8 Generalized transformations

So far, in our discussion of the system of rules introduced by Chomsky
in Syntactic Structures, we have assumed that only one terminal
string will serve as input to the transformational component. How-
ever, the system also provides for the combination of two or more
terminal strings (by the successive concatenation of pairs of strings)
by means of optional generalized (or * double-base') transformations.
By contrast, T-rules operating solely upon one terminal string (and
its subsequent transforms) are referred to as singulary (or 'single-
base'). As we have seen, the singulary transformations of Syntactic
Structures are specified as optional (e.g. the passive transformation)
or obligatory (e.g. the number-concord transformation).
Generalized transformations fall into two classes: embedding and
conjoining rules. There is a rough, but not exact correspondence, with
the distinction sometimes drawn in traditional grammar (and rather
difficult to apply in many instances) between complex and compound
sentences (cf. p. 266). Among the many constructions that would be
handled by means of an embedding rule, there are various * nominal-
izations'. We have already discussed (in the earlier, more informal,
parts of this section) sentences like Flying planes are dangerous. We
said that flying planes was a noun-phrase in the surface structure of
this sentence (and had the internal surface structure of adjective +
noun, like supersonic planes), but that it was * transformationally'
related to the deep structure of Planes fly and Planes are flying. We
can now say that the grammar generates a matrix-string (of the form
NP be A) and a constituent-string (of the form NPVintr); and
that the constituent-string is transformed ('nominalized') into an NP
of the form A + N> and then embedded as the subject of the matrix
NP-be-A,
We shall not go into the details of the formalization of embedding
transformations. It was little more than adumbrated in Syntactic
Structures; and the work that has been done since then has revealed
a number of problems. The statement made in the previous paragraph
was intended solely as a general indication of what is meant by the
term 'embedding'. The main point is that an embedded structure
(whether it be a phrase or a clause, according to the traditional
distinction between these two kinds of 'secondary' units: cf. 5.1.1) is
the transform of a string which could also serve as the underlying
structure for a whole sentence, but which ' functions' as a constituent
of another sentence. In a certain sense, it is a sentence within a
sentence. The phrase-marker of the matrix-sentence, dominated by 2,
will therefore contain another instance of S dominated by the appro-
priate symbol for the function* performed by the constituent-
sentence in the structure of the whole.
But conjoining transformations also combine sentences (more
precisely, the structures underlying sentences) within a 'larger'
sentence. The difference is (and this is the difference between
'complex', embedding, sentences and 'compound', conjoining,
sentences in traditional grammar) that conjoining transformations do
not subordinate one sentence, as a 'constituent', to the other or
some part of the other, the 'matrix'. Both of the conjoined structures
(assuming that only two underlying structures are involved in the
operation of conjunction) preserve their ' sentential' status within the
'larger' sentence. The phrase-marker for the 'larger' sentence will
therefore contain two (or more) instances of 2, co-ordinate with one
another, below the topmost 2 which dominates the whole phrase-
marker. The most obvious (and prima facie the simplest) example of a
sentence which results from the application of a conjoining trans-
formation is a sentence with two co-ordinate clauses linked by a co-
ordinating conjunction (e.g. and): cf. Caesar advanced and Pompey
retreated,
6.6.9 The present state of transformational grammar

We began this section by saying that any grammar which claims to
assign both a deep-structure analysis and a surface-structure analysis
to the sentences it generates is a transformational grammar; and we
saw that traditional grammars were transformational in this respect.
They were also generative in the first of the two senses currently
associated with this term (i.e. they were in principle 'projective', but
they were not 'explicit': cf. 4.2.13).
Later in the section we restricted our attention to the best-known
and most highly-formalized system of generative-transformational
grammar that has so far been developed. This was the system intro-
duced by Chomsky in Syntactic Structures (a relatively non-technical
abridgement of previous work, for the most part unpublished,
although it has circulated quite widely among linguists who have
concerned themselves with formalized generative grammar). Many
important theoretical questions have not been mentioned in this
necessarily brief and fragmentary account of Chomsky's system. It has
seemed preferable, in an introduction to linguistic theory, to concen-
trate upon such fundamental notions as 'structural analysability* and
the nature of transformational rules, and to illustrate the operation of
one or two simple rules in sufficient detail for the reader to appreciate
the significance of the fundamental principles of the system.
It has also seemed preferable to illustrate these principles with
reference to the earlier system of transformational grammar (as out-
lined in Syntactic Structures). One reason for this decision is that
most of the detailed exemplification of transformational grammar so
far published conforms, in general, to the earlier system of Chomsky.
Another more important reason is that the major modifications which
have recently been proposed by Chomsky and others derive from the
serious attempts they have made over the last ten years to apply the
earlier system to an ever wider range of syntactic relationships in
language (some of which were not even handled 'informally* in
traditional grammar). To understand the import of these modifi-
cations, one must first of all understand the earlier system.
Furthermore, there is a sense in which Syntactic Structures has not
been superseded by later works. It was a book of revolutionary
importance in the history of modern linguistics. In it, Chomsky did
not amass a wealth of detail from many languages, but selected a few
evident, and in a sense well-understood, facts about the structure of
a very familiar language (English) and about the nature of language
in general, and demonstrated, conclusively, that these facts could not
be 'explained1 within the framework of current linguistic theory. But
in doing so, he set more rigorous standards of 'explanation' than
linguists had hitherto been accustomed to. These standards are still
somewhat controversial (although a good deal of the controversy is
based on misunderstanding). But they have been accepted by many
linguists, including some who have rejected the details of the formal-
ization of grammatical theory developed by Chomsky. It is this fact
which gives to Syntactic Structures its 'revolutionary' importance.
The principles of transformational grammar outlined and discussed
in Syntactic Structures are simple enough (although it may require a
certain effort to see the implications of the particular system of
formalization). The application of these principles to the description
of more than a fragment of the grammatical structure of any one
language is far from simple. The use of T-rules in a generative
grammar with a simple phrase-structure base-component leads to
very serious problems in the specification of the order in which the
T-rules (and in particular the generalized transformations of
Syntactic Structures) operate relatively to one another. We have seen
that the T-rules of the earlier system were very heterogeneous. The
problem of analysability was solved for the kernel-strings by imposing
certain restrictions upon the nature of the PS-rules in the base-
component. But the problem was still there in the grammar. Since a
T-rule could operate upon a string which resulted from the applica-
tion of previous T-rules (involving permutations and deletions), and
since (in theory) each T-rule transformed one phrase-marker (and not
simply an unbracketed string) into another phrase-marker, it was
necessary to establish the principles which govern the assignment of
derived phrase-markers by particular types of T-rules. Although
Chomsky and others concerned themselves with this problem from
the outset, it is hardly discussed at all in Syntactic Structures; and it
was never satisfactorily solved.
Over the last few years, many scholars have illustrated the problems
of ordering T-rules in a grammar organized according to the system
proposed in Syntactic Structures, Recently, Chomsky himself has
modified the theory in a way which is designed to reduce, if not
eliminate, these problems. Briefly (and this is not intended as a self-
sufficient summary of the modifications), Chomsky's proposal
amounts to the removal of generalized transformations from the
grammar by introducing into the base-component optional recursive
rules with 2 to the right of the rewrite-arrow. All transformations are
then made singulary and obligatory (' triggered off', in certain cases,
by optional elements in the rules of the base-component) and ordered
in such a way that they apply to constituent 'sentences' before
embedding and to matrix 'sentences' after embedding (embedding,
and conjoining, being controlled by the base-component rules).
Apart from reasons connected with the problem of ordering the
rules, one of the principal motives for the revision of the system has
been to integrate syntax with phonology and semantics. It has been
tentatively suggested that the deep-structure analysis (generated by
the base-component) is irrelevant for the operation of the phono-
logical rules and that the surface-structure (resulting from the
application to the deep-structure phrase-marker of the singulary
transformations) is irrelevant for the semantic interpretation of
sentences. This is undeniably an attractive hypothesis; but its
empirical validation, or refutation, will require a considerable amount
of further detailed work. Until such work has been carried out and
evaluated, both parts of the hypothesis must be treated with some
degree of scepticism.
Although the irrelevance of surface structure to semantic analysis
remains to be proved or disproved, most linguists who accept the
validity of the distinction between deep grammatical structure and
surface grammatical structure (including those who have described
the relationship between the two in very different terms from Chomsky)
assume that there is some particularly intimate connexion between
deep syntax and semantics. We shall make the same assumption
throughout this book.
In conclusion, one must make the following point. The evaluation
of grammars that do not conform to the same system of formalization
(or are unformalized) poses problems of a very different order from
those posed by the evaluation of grammars that are weakly equivalent
and formally comparable (cf. 6.2.12). The complexity of transforma-
tional grammars (and the term ' transformational' is now being used
in its widest sense) is such that it becomes very difficult to compare
transformational grammars of different types. It is also difficult to
strike a balance between the conflicting claims of weak and strong
adequacy with respect to many sets of sentences that we might wish
to generate by means of a grammar of a particular type. In a later
chapter of this book we shall mention some 'deep connexions'
between sentences which Chomsky's theory of syntactic structure, in
its present form, does not handle (as Chomsky himself has pointed
out). It should be realized, however, that it is no difficult matter for
a linguist (drawing upon the knowledge about language accumulated
by his predecessors) to produce sets of sentences which he will
correctly describe ('informally') as 'transformationally-related'. The
problem is to give a systematic account of these relationships within
the theoretical framework already established for other sets of
sentences. If this cannot be done, one may be justified in modifying
the framework. But the modification can then have the effect of
making it impossible to handle the relationships which were satis-
factorily accounted for by the previous system. This point should be
borne constantly in mind during the more informal discussion of
grammatical theory which follows.
John Lyons
Online ISBN: 9781139165570
Chapter
7 - Grammatical Categories pp. 270-333

7
GRAMMATICAL CATEGORIES
7.1 Introductory
7.1.1 The term 'category* in traditional grammar
So far we have adopted a purely 'formal* approach to grammatical
analysis. In the present chapter we will discuss the traditional
assumption that there exist certain * notional' categories which are
universal in the sense that they are common to all languages (cf. 4.1.3,
4.2.9). In particular, we shall be concerned with the traditional theory
of the * parts of speech' and such 'grammatical categories' tradition-
ally associated with the parts of speech as person, tense, mood,
gender, number and case. First of all, something must be said in this
introductory section about the term 'grammatical category' and its
traditional implications.
There is very little consistency or uniformity in the use of the term
'category' in modern treatments of grammatical theory. It is fre-
quently employed, like 'class' or 'set', to refer to any group of
elements recognized in the description of particular languages. Some
authors refer to the ' parts of speech' as ' categories'; others, following
the more traditional usage, restrict the application of the term to such
features associated with the ' parts of speech' in the classical languages
as have been mentioned above (person, tense, mood, etc.). And there
are otherwider, narrower or quite differenttechnical senses in
which the term has been employed. No attempt will be made here to
give to the term 'grammatical category' a precise, technical inter-
pretation. It is merely being used as a convenient heading for the
chapter in which we will discuss certain important questions of
traditional grammar in the light of more recent developments in
general syntactic theory.
The term ' category' is but one of the many traditional terms used
by linguists which owe their origin to the fact that western gram-
matical theory was developed on the basis of that particular philo-
sophical system which, for the present purpose, we can refer to,
somewhat loosely, as 'Aristotelian'. The term 'category' derives
from a Greek word which is otherwise translated as 'predication'
[270]
(in the logical, or philosophical, sense of 'attributing properties' to
things). In Aristotelian (and scholastic) philosophy, the 'categories'
were the different ways, or modes, in which predications could be
made of things; and it was assumed that the different modes of
predication represented differences in the objective world, the different
modes of 'being'. Underlying the classification of the modes of
predication and of 'being' was the assumption that the physical world
consists of things ('substances') which have certain properties
('accidents'), initiate or undergo certain processes, stand in a certain
relationship to one another, or have a certain extension or location
in space or time. We shall not go into the details of the Aristotelian
theory of the 'categories'. But two general points may be made. First,
it should be noted that a fundamental distinction was drawn between
the category of 'substance', on the one hand, and the other 'acci-
dental' categories, on the other: the 'substance' was the individual
thing in abstraction from its 'accidental' properties. Second, it was
assumed by Aristotle and his followers, and stated explicitly by the
medieval 'speculative' grammarians, that the structure of language
reflected the structure of the world: that words signified things
according to their mode of 'being', as 'substances' or 'accidents'. It
was because of this correspondence between the modes of 'being'
and 'signifying' that knowledge of the world was possible. The
categories of 'being', 'signifying' and 'understanding' were con-
gruent with one another; and the congruence of the three sets of
categories was held to justify the intimate and indissoluble association
of philosophy, grammar and logic.
7.1.2 'Matter' v. 'form': 'substance' v. 'accidents'

A further distinction of Aristotelian philosophy is relevant to the
development of the traditional theory of the ' parts of speech' and the
' grammatical categories': this is the distinction that was drawn between
'matter' and 'form'. Every individual thing (that is, every 'sub-
stance') was said to consist of 'matter' (which we may think of as
some physical material) and of the particular 'form' (the individu-
ating principle) imposed upon this physical' matter' which gave to the
resulting thing its identity and internal stability. (The Aristotelian
distinction of 'matter' and 'form', it will be observed, is very similar
to the distinction drawn by de Saussure between 'substance' and
'form': cf. 2.2.2. As we have just seen, the traditional sense of the
term 'substance' was quite different.)
272 7- GRAMMATICAL CATEGORIES
Since language was both an object of analysis and also the instru-
ment with which all philosophical analysis was carried out, the theory
of the 'categories' has had a double effect upon traditional grammar.
Not only were the elements of language themselves analysed in terms
of 'matter' and 'form* and, as 'substances', classified with respect to
their 'accidental' properties, but they were also grouped into classes
('parts of speech') according to their 'mode of signifying' the things,
properties and relations to which they referred.
As we have already seen (cf. 5.4.1), the basic unit of grammatical
analysis in traditional theory was the word (dictio, to use the Latin
term). Each word, as a 'sign', was composed of a certain combination
of sounds (vox) and meaning (significatio). The physical realization of
the word (vox) wTas not an essential part of it, but merely the way in
which that word happened to be pronounced, 'accidentally' and by
'convention' and 'usage', in a particular language. As a physical
object the vox could be further analysed into its 'elements', sounds
or letters. But the grammatical analysis proper began with the analysis
of the ways in which words operated as ' signs', as instruments for the
description and understanding of ' reality'; with their classification as
'parts of speech' and with the establishment of patterns (or 'para-
digms') of 'declension' and 'conjugation'.
Words, like everything else that came within the scope of scientific
inquiry, had to be described in terms of the traditional, definitive list
of Aristotelian 'categories'. The 'substance' of the word had to be
distinguished from its 'accidents'the different forms it assumed
according to its syntactic function and its particular 'mode of
signifying'. Certain 'accidental' categories were typical and definitive
for particular ' parts of speech': nouns were inflected for case (nomi-
native, accusative, etc.) and number (singular, plural), and belonged
to a particular gender (masculine, feminine, neuter); verbs were
inflected for tense (present, past, future), person, number, etc.; and so
on. What are traditionally referred to as * the grammatical categories'
are therefore the 'accidental' categories of grammatical theory; and
this explains the older term 'accidence' (from the Latin accidentia)
for what we now refer to as 'inflexional variation'.
Some of the 'accidental' grammatical categories, such as tense
(derived from the Latin word for 'time'), could be referred without
difficulty to the Aristotelian categories of predication. Others, such
as case, were peculiar to language. What is important for our purpose
is not a detailed understanding of the relation between the Aristotelian
categories and the traditional 'grammatical categories', but merely an
appreciation of the general relationship between them. The assumed
universality of the Aristotelian categories of predication reinforced, or
promoted, the further assumption that the ' grammatical categories'
were universal features of human language: that every language
necessarily manifested such categories as tense, number, case, etc.;
and that these categories were typical of particular * parts of speech*.
7.1.3 * Major' and ' minor' parts of speech

The * parts of speech' themselves were also defined by reference to
the Aristotelian 'categories' (independently of their inflexion
according to such 'accidental* categories as case, tense, etc.). Nouns
were defined, according to their 'mode of signifying', as words which
referred to 'substances' (hence the term 'substantive'); adjectives as
words which denoted ' qualities' and so on. The reader will doubtless
be familiar with definitions of the 'parts of speech' cast in this
traditional, 'notional' terminology. Later in this chapter we shall
investigate the validity of such definitions.
The Aristotelian opposition of 'matter' and 'form' was also
invoked by certain grammarians to distinguish between what we may
refer to as major and minor 'parts of speech'. Only the major 'parts
of speech' (nouns, verbs, adjectives and adverbs) were meaningful in
the proper sense of the term: they 'signified' the objects of thought
which constituted the 'matter' of discourse. The other 'parts of
speech' (prepositions, conjunctions, etc.) did not 'signify' anything
of themselves, but merely contributed to the total meaning of
sentences by imposing upon them a certain 'form', or organization.
This distinction recalls, on the one hand, that made by Aristotle
between 'words' properly so called and 'conjunctions' (cf. 1.2.5) a n ^ ,
on the other, the distinction traditionally drawn between 'full' and
' empty' words in Chinese grammatical theory. It was taken up and
given further currency by the grammarians of the Port Royal (cf. 1.2.8),
and was used by Leibniz and many philosophers who have been
influenced by him. The distinction is frequently drawn in modern
treatments of grammatical theory in terms of 'lexical' and 'gram-
matical' meaning (cf. 9.5.1).
7.1.4 Logic and grammar

To conclude the present section, something must be said about the
general influence of ' Aristotelian' logic upon traditional grammatical
theory. It has frequently been asserted by philosophers that Aristotle's
theory of 'categories', especially his distinction between 'substance'
and 'accidence', is simply a reflection of the grammatical structure of
Greek, and that if he had spoken a language with a very different
grammatical structure he would have established a quite different set
of ' categories' and possibly a different system of logic. On the other
hand, grammarians are apt to say that some of the distinctions drawn
in traditional grammar are purely ' logical' distinctions (for example,
the distinction between proper nouns and common nouns) and
cannot be justified for Greek and Latin. The fact that both these
positions have been maintained suggests that the relationship
between traditional grammar and 'Aristotelian' logic is far more
complex than is commonly supposed. And this is, in fact, the case.
As we shall see later in this chapter, recent developments in the theory
of syntax (in particular, the distinction of 'deep' and 'surface'
structure: cf. 6.6.1) may enable us to determine to what degree the
' categories' of logic and grammar are congruent with one another.
7.1.5 Primary, secondary and functional categories

For terminological convenience in the sections which follow, we shall
refer to the 'parts of speech' as primary grammatical categories and
such notions as tense, mood, case, etc., as secondary grammatical
categories. The traditional syntactic notions of 'subject', 'predicate',
'object', etc., will be referred to as functional categories. For reasons
which will become clear presently we will deal first with the secondary
grammatical categories.
Most treatments of what we are calling, for convenience, the
secondary grammatical categories give pride of place to the tradi-
tionally-recognized inflexional categories of Latin and Greek:
numbery gender and case for the noun; person, tense, mood and voice
for the verb. It will be recalled that case was regarded as criterial for
the noun and tense for the verb in traditional inflexional definitions
of the parts of speech (cf. 1.2.5). The present treatment will follow
the general practice by concentrating upon the categories of tradi-
tional grammar. We shall see that, although these categories are of
fairly wide applicability, they require reformulation in the framework
of modern generative grammar, which takes syntax as primary and
regards inflexion as but one way of specifying syntactic relationships
within sentences (cf. 5.4.2).
7- GRAMMATICAL CATEGORIES 275
7.2 Deictic categories

7.2.1 Deixis and the situation of utterance
At this point we will introduce the notion of deixis to which we shall
make frequent appeal in the discussion of grammatical categories.
Every language-utterance is made in a particular place and at a
particular time: it occurs in a certain spatio-temporal situation. It is
made by a particular person (the speaker) and is usually addressed to
some other person (the hearer); the speaker and the hearer, we will
say, are typically distinct from one another (there may of course be
more than one nearer) and moreover are typically in the same spatio-
temporal situation. (There are many common situations of utterance
which are 'untypical* in these respects: it is possible to 'talk to
oneself'; and, if one is speaking on the telephone, the hearer will not
be in the same spatio-temporal situation.) We will further assume
that the typical utterance includes a reference to some object or
person (which may or may not be distinct from the speaker and hearer,
cf. Have youfinishedyet?: Has hefinishedyet?, etc.); for the present,
we will call this object or person to which reference is made in the
utterance the 'subject of discourse*. The utterance will therefore
contain as many 'subjects of discourse' as there are lexical items in
the utterance which refer to objects and persons.
The notion of deixis (which is merely the Greek word for ' pointing'
or 'indicating'it has become a technical term of grammatical
theory)is introduced to handle the'orientational' features of language
which are relative to the time and place of utterance. The so-called
'personal pronouns' (/, you, he, etc.) constitute only one class of the
elements in language whose meaning is to be stated with reference to
the 'deictic co-ordinates' of the typical situation of utterance. Other
elements which include a component of deixis are such adverbials of
place and time as here and there ('in the vicinity of the speaker: 'not
in the vicinity of the speaker') and now and then ('at the time of
speaking': 'not at the time of speaking'). These are just the most
obvious instances of the way in which the grammatical structure of
language may reflect the spatio-temporal co-ordinates of the typical
situation of utterance.
The typical situation of utterance is egocentric: as the role of speaker
is transferred from one participant to another in a conversation, so the
' centre' of the deictic system switches (/ being used by each speaker
to refer to himself, you being used to refer to the hearer). The speaker
is always at the centre, as it were, of the situation of utterance. It
should also be observed that the participants in a situation of
utterance not only assume the roles of speaker and hearer (we shall
refer to these as roles). They may also stand in a certain linguistically-
relevant relationship of status vis-a-vis one another (parent: child,
master: servant, teacher: pupil, etc.). Status-relations interact with,
and, in certain languages, may override the participant roles of speaker
and hearer.
7.2.2 Person
The category of person is clearly definable with reference to the notion
of participant-roles: the 'first* person is used by the speaker to refer
to himself as a subject of discourse; the * second' person is used to
refer to the hearer; and the * third' person is used to refer to persons
or things other than the speaker and hearer. So much is straight-
forward enough. There are, however, a number of points in the
traditional treatment of the category of person which require
clarification.
The * third* person is to be distinguished from the 'first' and
* second' persons in several respects. The speaker and hearer are
necessarily present in the situation, whereas other persons and things
to which reference is made may not only be absent from the situation
of utterance, they may be left unidentified. This means that the
category of third person may combine with such other categories
as * definite' or 'indefinite' and 'proximate' or 'remote' (the category
of 'proximity' being determined, as we shall see, by reference to the
participants). The English pronouns he> she and it are definite, as
against someone, somebody and something, which are indefinite. In
Turkish the category of definiteness is obligatorily marked (by means
of a suffix) in the pronouns of the first and second person (in the
accusative) in the same way as it is marked, optionally, for nouns:
cf. ben-i, 'me', sen-i, 'you\-kitab-i, 'the book' v. kitab, 'a book'. But
the English pronouns of first and second person are no less definite
than the corresponding pronouns in Turkish. From the gramma-
tical point of view, independently of their phonological realization,
pronouns of first and second person are necessarily 'definite'
( + def), whereas third person pronouns may be either 'definite'
or 'indefinite' ( def), if this distinction is actualized in the language
in question. Furthermore, pronouns of the first and second per-
son necessarily refer to human beings. (In fables and fairy-stories,
7-2. DEICTIC CATEGORIES 277
animals and things are linguistically ' personified' and, when they are
made to speak of themselves in languages whose gender-system makes
this clear, it seems to be the case that they are automatically 're-
categorized' as animate for the purpose.) Pronouns of the third
person may refer to human beings, to animals and to things. This
does not mean, of course, that these distinctions are syntactically
relevant in the third person for all languages; merely that, if such
distinctions are drawn in a particular language they will be neutral-
ized in combination with first and second person. Finally, it seems
reasonable to say that, whereas first and second person are the
positive members of the category of person, third person is essen-
tially a negative notion; unlike first and second person, it does not
necessarily refer to participants in the situation of utterance. In many
languages there is no overt recognition of what is traditionally called
third person, merely the absence of the formal markers offirstand
second person.
Traditional terminology is rather misleading in the way in which it
represents the combination of the categories of person and number. It
is clear, for instance, that we ('first person plural') does not normally
stand in the same relationship to / ('first person singular') as boys,
cozvs, etc., do to boy, cow, etc. The pronoun we is to be interpreted as
'I, in addition to one or more other persons'; and the other persons
may or may not include the hearer. In other words, we is not 'the
plural of / ' : rather, it includes a reference to ' I ' and is plural.
According to whether the *firstperson plural' pronoun includes a
reference to the hearer or not, it is customary to distinguish between
an ' inclusive' and an ' exclusive' use of the pronoun. Although this
distinction is generally not relevant in English, there are many
languages in which the distinction between the 'inclusive' and the
' exclusive' use of the ' first person plural' is drawn systematically in
sentences of all types and realized as a distinction between two
phonologically-unrelated pronouns. A distinction might also be made
between an * inclusive* and 'exclusive' use of the 'second person
plural' (in a slightly different sense of 'inclusive' v. 'exclusive'). The
English pronoun you may of course be either singular or plural
(cf. wash yourself and wash yourselves). As a plural form, it may be
either 'inclusive' (referring only to the hearers presentin which case
it is the plural of the singular you, in the same sense as cows is the
plural of cow) or 'exclusive* (referring to some other person, or
persons, in addition to the hearer, or hearers). There may be good
reason to suggest, therefore, that' first' (ego) and ' second' (tu) are not
10 LIT
of equal status in the category of person: that the primary distinction
is between 'first' ( + ego: 'plus ego') and 'not-first' ( ego: 'minus
ego') and that the distinction of 'second' and 'third' is secondary.,
If this is so, the correct syntactic analysis of the traditional singular
personal pronouns of English in terms of their component deictic
features would be as follows: / = [ + ego]; you (singular) = [ ego,
+ tu]; he, she, it = [ ego, tu], (He, she and it are further distin-
guished by the non-deictic features of gender. These features of
gender are neutralized in English when person is combined with
plural, they being neutral with respect to gender. This kind of
neutralization is quite common: it is to be found, for instance, in
Russian and German. By contrast, other languages, including French
and Italian, maintain the distinction of gender in the plural personal
pronouns. In French, for instance il ' he' and elle ' she' are matched
in the plural by Us 'they' (masculine) and elles 'they' (feminine).
But here there are neutralizations of a different sort which are
operative. Elles is feminine in the sense that all the persons, or things,
it refers to are feminine; Us is masculine in the sense that at least one
of the persons, or things, referred to is masculine. In other words,
masculine is here the dominant term of the gender system, as ' first'
is the dominant term of the category of person.)
7.2.3 Demonstrative pronouns and adverbs

In this connexion one should mention the ' demonstrative pronouns'
(e.g. this, that: these, those) and the 'situationally-bound adverbs of
place' (e.g. here, there). It is evident that these forms 'include' an
element of deixis; furthermore, that they are 'definite', rather than
'indefinite' (cf. a, some (.. .or other)] somewhere), and that they are to
be distinguished in terms of a category of 'proximity'. What is
perhaps not so clear is that the category of proximity is typically
determined in relation to the category of person. (Strictly speaking,
both person and proximity are determined in relation to the elements
of the typical situation of utterance.) Both this and here are to be
interpreted as 'proximate' with respect to the speaker. It is to be
noticed that, whereas English neutralizes the distinction between the
'second' and 'third' person in this category of 'proximity', there are
languages for which this is not so. In both Latin and Turkish there is
a three-term system of 'demonstrative' pronouns: Latin, hie, iste,
Me; Turkish, bu, su, 0. The Latin hie and the Turkish bu correspond
to the English this: they are 'first person demonstratives' (indicating
7-2. DEICTIC CATEGORIES 279
proximity to the speaker). But Latin iste and Me and Turkish u and 0
(which can generally be translated by the English that) are, respec-
tively, ' second' and' third person demonstratives *. Iste and u indicate
proximity to the hearer, whereas tile and 0 indicate remoteness from
both speaker and hearer.
In many languages no distinction can be drawn between the
' demonstratives* and the 'third person pronouns'. This is the case,
for example, in Turkish where he, she and it (if translated by a
pronoun) would normally be translated with 0 (there is no gender in
Turkish), and they with the plural form of the same ' demonstrative',
onlar. In classical Latin (as in Greek), there was no 'third person
pronoun* at all: where 'pronominal* reference was made to some
'subject of discourse* (other than the speaker or hearer), the appro-
priate 'demonstrative* was usedhie, iste or Me (in addition to these
three and distinguished as to 'proximity* in relation to person, there
was also the word is, which was not normally used for the grammatical
subject, but in other syntactic positions was the most 'neutral* of the
'demonstrative pronouns', standing outside the category of 'proxi-
mity*). The 'third person pronouns' of the Romance languages
(French, Italian, Spanish, etc.) have in fact developed from ' demon-
strative pronouns'; and the same is true of the' third person pronouns *
of English and German. Moreover, in all these languages the ' definite
article* (English, the; French, le, la; etc.) has also developed from
what was originally a 'demonstrative pronoun*. This gives us the
clue to the relationship between these three different' parts of speech'.
They all 'include' the feature 'definite': from this point of view the
man, this man, that man contrast with a man, and he contrasts with
someone. But the man and he, being undetermined with respect to
proximity, are both in contrast with this man ('proximate') and that
man ('remote'). The traditional separation of the 'articles', the
'personal pronouns' and the 'demonstrative pronouns' obscures
these relationships.
7.2.4. 'Attraction' of person and number

In this brief account of the category of person, it is impossible to
pursue all the interesting additional distinctions that are to be found
in particular languages. Reference should be made, however, to the
phenomenon of 'attraction' and also to the 'honorific dimension' of
the system of 'personal pronouns'. What is meant by 'attraction' in
this connexion maybe illustrated from Russian and French. A sentence
28o 7. GRAMMATICAL CATEGORIES
such as John and I arrived might be translated into Russian and
French respectively as My s Ivanom priechali (literally 'We with
John arrived*) and Nous sommes arrives avec Jean (literally 'Wre
arrived with John*). In both instances, it will be observed, the phrase
'with John* specifies the other person already included in the
reference of the pronoun 'we*. * Attractions' of this kind illustrate
further the necessity of analysing the 'personal pronouns* which
appear in the surface structure of sentences into their component
features of deixis and number.
7.2.5 'Honorific* distinctions

The * honorific' dimension is introduced to account for the dif-
ferentiation of the personal pronouns in certain languages, not in terms
of their reference to the roles of the participants in the situation of
utterance, but in terms of their relative status or degree of intimacy.
Well-known and obvious instances of this are to be found in a number
of European languages (e.g. French, German, Russian, Italian, etc.)
where the 'second person singular pronoun* is used in addressing
children and friends (there are considerable differences in the con-
ventions governing the application of the 'intimate* form in the
various languages and they must be stated with reference to social
class and other factors) and the ' second person plural * or some other
pronominal form (e.g. Spanish usted) is used for more 'formal*
reference to the hearer. It is important to realize that such uses of a
' plural * or ' third person * form to refer to the hearer do not invalidate
the distinction of number in the ' second person * or the distinction of
the 'second* and 'third* person in the singular. Not only is the
recognition of these distinctions still required for the proper semantic
interpretation of sentences containing the 'honorific* form of
reference to the hearer, it is frequently required also in the statement
of the syntactic rules for gender or number concord.
7.2.6 Person and the verb

Traditionally, person is regarded as a category of the verb; and it is
certainly marked in the inflexional form of the verb or verb phrase in
many languages. This fact may be illustrated with reference to the
following Latin words: amo ' I love', amas 'you love*, amat 'he, she
loves'. In Latin, and in many other languages, there is normally no
pronominal specification of the subject: ego amo, tu amas, etc., would
7-2. DEICTIC CATEGORIES 28l
be employed only when ego ('I'), tu ('you'), etc., are emphasized or
contrasted with some other form, expressed or implied. On the basis
of evidence such as this, it is frequently suggested that the selection
of the pronoun, when it occurs, is controlled by the form of the verb:
that amOy under conditions of emphasis, selects ego; that amas selects
tu; and so on. This view is, however, mistaken, since it takes no
account of the distinction between the ' surface' structure of sentences
and their underlying ' deep' structure. In terms of the underlying,
semantically-interpretable, structure of sentences, there is little
difference between Latin and English. In both cases we must postu-
late an abstract 'pronominal' element (determined with respect to
person and number) which is the subject of the verb and controls the
rules governing the phonological realization of the verb in surface
structure. This 'pronominal' element is not normally realized in
Latin: in the generation of Latin sentences it is deleted, after the
attachment of the features of person and number to the verb, whereas
in English it is ' rewritten' by the rules of phonological realization as a
'personal pronoun' (/, you, etc.). As we have seen, the 'pronominal'
element is also given a phonological realization in Latin under
certain conditions of emphasis or contrast. And it is phonologically
realized (as me> te> etc.) in other syntactic environments: cf. (dtxit)
me amare..., '(He said) that I was in love with...'. In general
syntactic theory, person (like number) is only secondarily, and deriva-
tively, a category of the verb; and that only in certain languages.
7.3 Number and gender

7.3.1 Number and countability
The most common manifestation of the category of number is the
distinction between singular and plural (cf. boy: boys, etc.), which is
found in many languages all over the world. This distinction clearly
rests upon the recognition of persons, animals and objects which can
be enumerated (as 'one' or 'more than one') and referred to, indi-
vidually or collectively, by means of nouns. Number is therefore a
category of the noun; we have already seen that it combines in a
special way with the category of person to form 'personal pronouns'
(which may be regarded as syntactically equivalent to nouns).
At first sight, it might appear that the application of the distinction
between 'one' and 'more than one' is straightforward enough. It is
important to realize, however, that what counts as 'one object', and
what as ' more than one object',' a group of objects' or an unindividu-
ated 'mass of materiar, in the inanimate world at least, is to a
considerable degree determined by the lexical structure of particular
languages. For example, the English word grape is a * countable' noun,
in the sense that it can be pluralized {Willyou have some grapes?, He
ate six grapes), whereas the German Traube and the Russian vinograd
are ' mass' nouns (like the English word fruit). The French word raisin
may be used in the singular either as a 'mass* noun (Vous voulez du
raisin?) or as a 'collective' noun (Prenez un raisinwhich is normally
to be translated, not as 'Have a grape', but 'Have some grapes'). By
contrast, French fruit and the Russian frukt are 'countable' (cf. Have
some fruit: Prenez un fruit). And in all the languages referred to here
certain words may be used either as 'mass' or 'countable' nouns
(cf. Have some grapefruit: Have a grapefruit). Examples like this
could be multiplied indefinitely, to demonstrate the point that the
lexical categorization of the world in terms of 'countable', 'collec-
tive ' and ' mass' nouns varies considerably from language to language,
even when the languages in question do have such syntactically-
distinct subclasses.
7.3.2 ' Secondary recategorization'

Over and above the primary categorization of nouns in such terms as
'countable', 'mass' and 'collective', there may be the possibility of
'secondary recategorization'. In English, for example, most 'mass'
nouns may be 'recategorized' as 'countable' in certain contexts (cf.
They drink three or four different wines at every meal). There may be
individual 'anomalies': it has often been pointed out that oats,
unlike wheat, barley, etc., is plural (cf. these oats: this wheat), but it is
nonetheless a 'mass', rather than a 'countable', noun. (Not only is it
impossible to singularize oatsto talk of *one oat as distinct from
*two oatsbut its anomalous status creates a certain tension in respect
of 'secondary recategorization'. One might wonder whether One oats
is very like another is possible or not.) To be distinguished from a
word like oats are such 'plural only' forms as scissors, trousers, etc.;
and also certain 'collectives' like committee, Government (cf. This
Government have/has decided..., but not * These Government have
decided...). Finally, there may be stylistic differences between the
usage of different groups in the speech-community. It is noticeable,
for instance, that hunters tend to use such words as lion, elephant,
buffalo, etc., as 'semi-mass' nouns: cf. an elephant, but also a herd of
7-3- NUMBER AND GENDER 283
elephant. Compare, for all styles of English, a fish: a shoal offish; and
contrast, on the one hand, a cow: a herd of cows, and, on the other,
*a cattle: a herd of cattle. The analysis of the category of number in
particular languages may be a very complex matter.
The notion of ' countability' is inherent in the lexical structure of
all human languages, since (as we shall see in our discussion of the
'parts of speech': cf. 7.6.1 ff.) the recognition of what is probably a
universal primary category of language, the noun, presupposes the
' individuation' and enumeration of persons, animals and at least a
certain number of perceptually-discrete objects. However, not all
languages have a grammatical category of number. In Chinese and
Vietnamese (and many other languages), the distinction between, for
example, 'I wrote a letter' and 'I wrote some letters' can be made, if
necessary, by means of a numeral or a word meaning 'several', but it
may be equally well left unexpressed. Other languages (e.g. Classical
Greek, Sanskrit and certain Slavonic languages) have a dual, in addi-
tion to a singular and plural (the dual being used to refer to two objects:
cf. Greek dmpho kheire 'both hands'but in Greek there was
considerable fluctuation in the use of the dual and the plural). Fijian
and a few other languages are reported to have a trial (for reference
to three objects) as well as a singular, dual and plural.
7.3.3 Gender
The traditional names for the three genders found in the classical
Indo-European languages'masculine', 'feminine' and 'neuter'
clearly reflect the association which traditional grammar established
between sex and gender (cf. 1.2.5). But the term 'gender' itself
derives from an extremely general word meaning 'class' or 'kind'
(Latin genus): the three genders of Greek and Latin were the three
main noun-classes recognized in the grammar. From the grammatical
point of view, the nouns of Greek and Latin were classified into three
genders in order to account for two distinct phenomena: (i) pro-
nominal reference, and (ii) adjectival concord (or 'agreement'). For
the same reasons, the nouns of French, Italian and Spanish are
classified into two genders, the nouns of Russian and German into
three genders, the nouns of Swahili into at least six genders, and so
on. Gender plays a relatively minor part in the grammar of English by
comparison with its role in many other languages. There is no
gender-concord; and the reference of the pronouns he, she and it is
very largely determined by what is sometimes referred to as ' natural'
genderfor English, this depends upon the classification of persons
and objects as male, female or inanimate. The operation of gender-
concord will be illustrated below from Swahili.
7.3.4 'Natural' basis for gender

The first general point that must be made is that the recognition of
gender as a grammatical category is logically independent of any
particular semantic association that might be established between the
gender of a noun and the physical or other properties of the persons
or objects denoted by that noun. On the other hand, it is a matter of
empirical fact that in most languages that have gender (defined as a
classification of nouns for pronominal reference or concord) there is
some 'natural', semantic basis for the classification. This is not
necessarily sex. It may be shape, texture, colour, edibilityin short,
any set of 'natural' properties. The degree of correspondence be-
tween the classification of nouns by grammatical gender and a classi-
fication of the persons and objects denoted by them according to the
relevant 'natural' properties will vary considerably from language
to language. It is well-known that in the Indo-European languages
many words which denote inanimate objects are 'masculine' or
'feminine' in gender; to this degree 'natural' and grammatical
gender fail to correspond.
7.3.5 Gender in Swahili

In Swahili (and other Bantu languages) the nouns are classified into
genders according to the singular and plural prefixes attached to them.
There are at least six genders in Swahili (and many more in some
other Bantu languages). They may be illustrated by the following
words:
I. mtUy watu ('person', 'people')
II. kisuy visu ('knife', 'knives')
III. mtiy /mfr'('tree', 'trees')
IV. nchi, nchi ('country', 'countries')
V. jiwe, tnawe ('stone', 'stones')
VI. udevuy ndevu ('single hair of beard', 'beard')
It will be observed that, although there are only five singular prefixes
and five plural prefixes (m-y ki-y n-, ji-> w-, and wa-, vi-> m-, n-y ma-,
respectively), the possibilities of combination yield six classes of
7-3- NUMBER AND GENDER 285
nouns (cf. classes I and III, on the one hand, and IV and VI, on the
other). Gender is relevant, not only for the selection of the correct
singular or plural form of any given noun, but also for the determina-
tion of the verbs, adjectives and other modifiers in construction with it:
mtu amefika, watu wamefika ('The man has arrived', 'The
people have arrived')
kisu kimeanguka, visu vimeanguka ('The knife fell', 'The
knives fell')
mti umekauka, mitt imekauka ('The tree withered', 'The trees
withered')
mtu mzuri yule, watu wazuri wale ('That beautiful person',
'Those beautiful people')
tniti mzuri ule, mitt viizuri He ('That beautiful tree', 'Those
beautiful trees')
From these examples it will be clear that there is gender-concord
between the subject of the sentence and the predicate (cf. the first
three pairs of sentences with the prefix-patterns m- a-y wa- wa-;
hi- ki-, vi- vi-; m- u-, mi- i-: notice that in the case of class I in the
singular and class III in both the singular and the plural the prefix
on the verb differs from the prefix of the noun) and also between any
noun and the adjectives and other modifiers dependent upon it
within the same phrase (cf. the fourth and fifth pairs with the prefix-
patterns m- m- yu-, wa- wa- wa-; m- m- u-, mi- mi- i-: the word-order
in these phrases is noun-adjective-demonstrative).
The subject prefix is an obligatory part of the verb in Swahili,
whether there is a noun present as the subject of the sentence or not:
cf. Amefika, Wamefika ('He/She has arrived', 'They have arrived').
The subject prefixes therefore operate like pronouns; and they are
determined by the gender of the nouns they refer to. For example,
kimeanguka would be used in reference to a word of class II, but
umeanguka in reference to a class III word. The verb may also have an
object prefix: cf. Amewaona, Ameviona, Amemiona, etc. ('He saw
them', where the prefixes wa-, vi-, mi-, etc. refer to nouns of classes
I, II, III, etc.). Pronouns of the first and second person also occur in
the same positions to mark the subject or object of the verb: cf.
Nimewaona, Umewaona, Tumewaona, etc. ('I, you, we, etc., saw
them'). The subject and object prefixes in the verb thus have a
pronominal function (cf. 7.2.2) in addition to marking concord
between the verb and the nouns actually present in the sentence.
Although there is no distinction of masculine and feminine in the
Swahili classification of nouns, there is some 'natural' basis to the
gender system. Most nouns denoting human beings fall into class I,
words denoting inanimate objects into class II, names of trees,
plants, etc., into class III, abstract nouns into class VI, and so on.
There are many words whose classification appears arbitrary or
anomalous, but this does not invalidate the statement that there is a
considerable degree of correspondence between gender and a classifi-
cation of nouns from a semantic point of view.
7.3.6 Clash between 'natural' and 'grammatical* gender

But the distinction between 'animate* and * inanimate* nouns cuts
across their classification by gender in Swahili, and in certain circum-
stances may be in conflict with it. There are a number of nouns
denoting human beings which belong to classes other than class I.
Being 'animate', they tend to determine the verbal prefixes according
to the principle of selection for class I nouns (although they select the
prefixes corresponding to their own gender-class in the adjectives and
modifiers dependent upon them within a noun-phrase). This also
holds for most nouns which refer to animals (though the majority of
these words belong to class IV). In other words, there is a certain
clash between 'natural' and 'grammatical' gender which is reflected
in the syntax of the language. It may be worth pointing out that this is
also true in the more familiar European languages with anomalous
gender classifications. For instance, in French the word professeur is
masculine, although it may refer to a man or a woman. The gender of
the noun determines concord within the noun-phrase as masculine,
regardless of whether the person referred to is male or female: le
nouveau professeur ('the new teacher'). But when the noun professeur
refers to a woman and an adjective which would normally be in
concord with it occurs in the predicate, there is a clash between the
masculine gender of the noun and the female sex of the person
denoted by the noun. Neither the masculine form of the adjective
(e.g. beau) nor its feminine form {belle) can be used appropriately in
these circumstances without resolving, as it were, the 'conflict'
between 'grammatical' and 'natural' gender. Neither Le nouveau
professeur est beau (which necessarily refers to a man) nor *Le
nouveau professeur est belle (which is ungrammatical) is possible. The
'conflict' is resolved with a sentence like Elle est belle, le nouveau
professeur ('She is beautiful, the new teacher'). The resulting con-
7.3. NUMBER AND GENDER 287
struction, it will be observed, is very similar syntactically to the
normal pattern in Swahili (and a number of other languages); the
subject of the sentence is referred to both pronominally and by
means of a noun. In both languages ' grammatical' gender is dominant
within the noun-phrase; but 'natural' gender may prevail in pro-
nominal reference and for concord with the predicate. This pheno-
menon, which is found in a number of languages, would suggest that
grammatical cohesion is stronger in the noun-phrase than it is between
subject and predicate. It also shows that there is indeed some ' natural'
basis for the gender systems of the languages in question.
7.3.7 Redundancy of gender

From the semantic point of view, gender distinctions in the noun are
usually redundant. For instance, the forms la and ronde (rather than
le and rond) are determined by the feminine gender of the noun table
with which they are in concord in the French phrase la table ronde
(' the round table'). It is true that there are a few homonyms in French
which are distinguished by their gender: e.g. le mousse: la mousse ('the
cabin boy': 'the moss'). Although the occurrence of le rather than la
with mousse would be sufficient to resolve any potential ambiguity
between 'cabin boy' and 'moss', it seems clear that the two words
would rarely be used in the same context. The functional load of the
difference in gender is therefore negligible (cf. 2.4.1). This would seem
to be true of most pairs of homonyms that have a different gender in
French. It is probably true of the Indo-European languages generally.
Gender is semantically relevant, however, in the case of words
which are traditionally described as having 'common gender'. For
example, the French word enfant (' child') is inherently indeterminate
with regard to gender, but is determined as either masculine or
feminine according to the sex of the child referred to. Gender is also
semantically relevant in the case of sets of words, such as the French
pairs le chien: la chienne ('dog': 'bitch'), le chat: la chatte ('cat':
'female cat'), etc., where the formally-marked difference of gender
is used systematically to make a particular difference of meaning.
Compare, for example, the Spanish and Italian words for 'boy' and
'girl': muchachoy muchacha; ragazzoy ragazza. In Swahili, too, the
use of one prefix rather than another may have the effect of deter-
mining the sense of a noun in a particular way. For example, if -tu
occurs with class I prefixes (m-, wa-) it means 'person', but if it
occurs with class II prefixes (ki-, vi-) it means ' thing': in other words,
-tu itself is neutral as between the senses * person' and 'thing', but is
determined as one or the other by giving it ' human' or * inanimate'
gender in a particular context. (This is no different in principle from
the determination of the Italian word-stem ragazz- as either 'boy'
or 'girl' according to the particular gender suffix that is attached to
it.) But such instances are untypical. For most nouns, both in Indo-
European and Bantu languages, gender is inherent. Where there is
some correspondence between 'natural' and grammatical gender, it is
usually the meaning of the noun which may be regarded as deter-
mining the gender, rather than the converse. It is in this sense that
gender distinctions in nouns are semantically irrelevant (except for
such instances of' common gender' and formally-related sets of words
as have been exemplified in this paragraph). And it is worth noting
that the most common of the words that are semantically related in a
manner that could be marked by gender tend to be completely
different in phonological form: the absence of phonological corres-
pondence that is found in the English or French pairs boy: girl,
garfon: fille is more typical of the way the vocabulary is organized
with respect to common words than is the parallelism of the Italian
ragazzo: ragazza. Gender-concord is a 'surface-structure' pheno-
menon of certain languages (cf. 6.6.1). It is clearly the pronominal
function of gender which is of primary importance in communication.
7.3.8 'Classifiers'
Rather similar to the category of gender is the system of noun-
classification for the purpose of enumeration and individuation that is
found in many languages of south-east Asia. In Chinese, for example,
unless the noun itself denotes a unit of measurement (e.g. nidny 'year',
tian, 'day'), it must be preceded by a 'classifier' when it occurs with
a numeral or a demonstrative (zhe, 'this'). Some of the classifiers are
very general and may be regarded as semantically empty. Others are
specific to certain classes of nouns, and they may even be used
themselves elsewhere as nouns. It is as if the words thing, person,
animal, tree, fruit, etc., in English were used in this way: so that, for
instance, one would say that person policeman, three tree banana (for
'that policeman', 'three banana trees'); and the difference between
' three bananas' and ' three banana trees' would be made solely in the
classifierthree fruit banana: three tree banana.
7. GRAMMATICAL CATEGORIES 289
7.4 Case
7.4.1 The term 'case*
The nouns (as well as the pronouns and adjectives) of Greek and
Latin were classified by traditional grammarians according to
particular paradigms of declension for the inflexional categories of case
and number. The terminology that has just been employed
' inflexion',' case', and' declension ' (the word * paradigm 'simply meant
' pattern* or * model': cf. 1.2.3)testifies to the importance that
traditional grammar has always attached to the category of case in the
definition and classification of nouns. The Latin word casus (and the
Greek word which it translates) means 'falling' or 'deviation'.
Whatever might be the precise metaphorical origin of the technical
sense of the term in grammatical theory (and there is some dispute
about this), it is clear that variation in the forms of a lexeme according
to the syntax of the language was regarded as deviation from its
normal 'upright' form (cf. 1.2.5; f r 'lexeme', see 5.4.4). The terms
'inflexion' and 'declension' are explained by the same metaphor. It
was the Stoics who gave to the word ' case' the more particular sense
that it has since borne in grammatical terminology (cf. 1.2.5). Case was
the most important of the inflexional categories of the noun, as tense
was the most important inflexional category of the verb. It is signifi-
cant that the term 'case' (originally more or less synonymous with
what was later called 'inflexion') was restricted to one particular
inflexional category. The reason for this would seem to be that most
of the other categoriesgender, number, tense, person, etc.could
be related to a principle of semantic classification. Although each case
of the noun was given a label suggestive of at least one of its principal
semantic functions (e.g. the 'dative' was the case associated with the
notion of 'giving', the 'ablative' was the case associated with
'removal', etc.), it was impossible to give a satisfactory general
definition of the category of case itself in semantic terms. In effect,
traditional definitions of the category of case for Latin and Greek
(and other Indo-European languages) say little more than the
following: of the two inflexional categories of the noun, one is
number (definable in semantic terms and relatable to the Aristotelian
category of quantity), the other is case. The category of case was, as
it were, the most peculiarly grammatical of all the traditional categories
of inflexion, for it had no counterpart in the sister sciences of logic,
epistemology and metaphysics.
7.4.2 Case in Latin and Turkish

At this point it will be helpful to refer to three Latin paradigms
illustrating some of the most common patterns of declension. The
paradigms are set out in the traditional way in Table 9. We shall not
discuss the origin and significance of the traditional names for the
cases in Latin. It is sufficient to say that the most common function
of the 'nominative* is to mark the subject of the sentence; the
' vocative* is the case of address; the 'accusative* is used to mark the
object of a transitive verb; the 'genitive* is the case of 'possession*;
the 'dative* marks the indirect object (e.g. the equivalent of boy
would be in the dative and the equivalent of book in the accusative
in the Latin translation of I gave the boy a book); and the 'ablative*,
has a variety of functions, including that of marking the 'instrument*
with which something is done (e.g. the equivalent of sword in the
translation of / killed him with a sword). It should also be noted: that
particular classes of verbs may determine their objects in cases other
than the accusative; that different prepositions select particular cases
(most commonly the accusative or the ablative); and that the accusa-
tive and the ablative (with and without prepositions) have various
Table 9. Three Latin declensions

(The lexeme PUELLA 'girl* is feminine, LUPUS 'wolf* is
masculine, BELLUM 'war' is neuter.)
Singular
Nominative puella lupus bellum
Vocative puella lupe bellum
Accusative puellam lupum bellum
Genitive puellae lupi belli
Dative puellae lupo bello
Ablative puella lupo bello
Plural
Nominative puellae lupi bella
Vocative puellae lupi bella
Accusative puellas lupos bella
Genitive puellarum luporum bellorum
Dative puellis lupis bellis
Ablative puellis lupis bellis
74- CASE 291
* adverbial' functions with respect to the distinctions of place and
time. All these points will be of importance in the general discussion.
We have started with Latin, since it exemplifies a number of more
general and more particular features of the category of case in the
Indo-European languages. As we shall see, the main problem in
defining case as a grammatical category is that of deciding just how
many of the more particular 'Indo-European' features we wish to
preserve in the definition.
It has already been pointed out in an earlier chapter that case and
number are ' fused' in Latin, in the sense that the inflexional suffixes
(in those instances where there is determinacy of segmentation into
stem and suffix) mark the noun simultaneously for a particular case
and a particular number; furthermore, that the combination of a
particular case and a particular number may be marked by a quite
different suffix in different' declensions' and the same suffix may mark
different combinations of case and number. In short, Latin is' fusional'
with respect to inflexion in the noun (cf. 5.3.8). There is no reason to
make it a defining condition of the category of case that, either alone
or 'fused' with some other nominal category, it should subclassify
nouns into various declensions. The syntactic functions of the cases
in Turkish are very similar to the functions of the cases in Latin, but,
as we have already seen, they are marked by a suffix which (under
the general principles of vowel harmony) is segmentable and constant
for all nouns. There is thus only one regular pattern of ' declension' in
Turkish (cf. Table 10).
Six cases are traditionally recognized in the description of Latin.
But there is no Latin noun in which all six cases are distinguished
morphologically. For most declensions the nominative and the
vocative forms are identical in both the singular and the plural: the
principal exception to this generalization is illustrated by the declen-
Table 10. Case and number in Turkish
Singular Plural
Nominative ev evler
Accusative evi evleri
Genitive evin evlerin
Dative eve evlere
Locative evde evlerde
Ablative evden evlerden
sion of LUPUSa very common paradigm. In some declensions there
is no morphological distinction between the genitive and dative
singular (cf. PUELLA), in others the dative and ablative singular are not
distinguished (cf. LUPUS, BELLUM); in no Latin noun is the dative
plural morphologically distinct from the ablative plural. The
principle upon which the traditional paradigms are based seems to be
this: six cases are recognized, because this is the minimum number
of syntactically-relevant distinctions with which it is possible to state
rules of selection valid for all declensions in both the singular and the
plural. If all nouns were declined like LUPUS, it would be unnecessary
to draw a distinction between the dative and the ablative; if all nouns
were declined like BELLUM, there would no be need to distinguish the
nominative, vocative and accusative; and so on. Within the traditional
system of classification, it becomes possible to make such general
statements as 'the case of "possession" is the genitive* (puellae et
lupi, 'of the girl and the wolf'), 'the indirect object is in the dative*
{puellae et lupo> 'to the girl and the wolf), 'the preposition a selects
the ablative* (a puella et lupo, 'by the girl and the wolf*). But this
general principle is not in fact applied consistently in most grammars
of Latin. For there is at least one other case which must be recognized
in the rules which relate the inflexion of nouns to the syntax of the
sentence. This is the locative of 'place where*. Locative phrases of
' place where * in Latin are usually of the form preposition + ablative
of the noun (e.g. in ripd, 'on the bank*; in oppido, 'in the town*).
With a certain set of nouns, however, no preposition is used; and the
form of the noun that occurs is identical with the genitive singular for
most nouns that are declined like PUELLA, LUPUS and BELLUM (cf.
Romae,' in Rome *; domi,' at home *), but for certain other declensions
(which have not been illustrated) it is like the ablative. The rule for
the selection of the locative case is usually stated in the form in which
it has just been given. This is evidently inconsistent with the principle
implicit in the determination of the six traditional cases. But this
particular inconsistency is merely one of many. Its importance for
general grammatical theory is that it shows quite clearly that syntac-
tically-equivalent constructions may be realized by words or phrases
which might be classified quite differently in terms of their ' surface
structure*. Even if we were to say that there is a locative case in
Latin which happens to be identical with the genitive or the ablative
according to the declension and number of the noun, we should still
have to identify, syntactically, such expressions as Romae ('locative*)
and in oppido (preposition + ablative). Any general theory of case must
74- CASE 293
recognize two facts: (i) that the same case may realize more than one
syntactic function; and (ii) that a particular syntactic function may be
realized by a variety of means in the same languagein particular,
that there is a ' deeper' relationship between cases and prepositional
phrases in Latin than the traditional analysis of inflexion would
suggest. Both of these facts are relevant to the description of many
other languages, both within and outside the Indo-European family.
7.4.3 Interdependence of case and gender in Indo-European languages

The traditional method of grammatical description, which did not
proceed to the statement of the rules of syntax until the various
inflexional forms had been classified and displayed in paradigms,
might encourage us to believe that each case has a typical syntactic
function, and that all the different cases of the noun occupy the same
place, as it were, in the grammatical structure of a language. The
case-systems we find, in the Indo-European languages and elsewhere,
are far less symmetrical than they are made to appear in their
traditional presentation. We have already seen that in Latin, case-
distinctions are made in one declension that are not made in another,
and that the traditional system of six cases rests upon the somewhat
inconsistent application of a principle which might be described as
postulating for all ' declensions' whatever distinctions of case are
drawn in any one. But there is at least one point at which this principle
tends to obscure a very important feature of the category of case, not
only in Latin, but in the Indo-European languages generally. The
distinction between a nominative and an accusative form is never
made for neuter nouns in any of the Indo-European languages: the
declension of BELLUM is typical in this respect. It is also typical in that,
apart from its nominative/accusative plural (we may neglect the
vocative for the moment), it is declined like the masculine LUPUS. This
would suggest that the category of case is interdependent with the
category of gender in the Indo-European languages. To the extent
that gender is * naturally* based, on animacy and sex, there are two
distinctions involved, animate v. inanimate and feminine v. non-
feminine. The difference between masculine and neuter in the
nominative may be regarded as a consequence of the greater impor-
tance of the distinction between subject and object with respect to
animate nouns. Since many inanimate nouns are either masculine or
feminine, the distinction is drawn more or less redundantly for them.
But the important point is that relatively few animate nouns are
neuter. Even in languages without case-inflexion there is a strong
connexion between animacy and subject-position: the principle is
therefore one of very general application, and we shall return to it
later (cf. 8.2.5). The same principle is even more clearly operative in
Russian (and other Slavonic languages), where masculine and neuter
nouns are related in essentially the same way as they are in Latin, but
where the distinction between animate and inanimate is 'redrawn',
as it were, for masculine nouns on the basis of the ' natural' criterion
of animacy. The distinction between the subject-case and the object-
case is made for masculine animate nouns by what is frequently
described, rather misleadingly, as using the 'genitive* rather than
'accusative* (the latter being described as identical with the nomi-
native for masculine and neuter nouns in conventional descriptions of
Russian). The effect of this distinction between animate and inanimate
in masculine nouns is to mark the object of a transitive verb as
inflexionally different from the subject in that particularly important
set of sentences in which both the subject and the object are animate:
cf. Ivan videl Borisa, Borisa videl Ivan ('John saw Boris', where the
-a marks Borisa as the object). By contrast, in Boris videl stol, Stol
videl Boris ('Boris saw a/the table') the inanimate stol is unmarked
inflexionally as the object of the verb videl. It is relatively uncommon
for two inanimate nouns to be equiprobable as subject or object of
the same transitive verb; and when this happens, the distinction can
be made in a number of ways. The relative order of the two nouns is
one of the ways in which, residually as it were, the syntactic distinc-
tion between subject and object may be drawn in Russian (and in a
number of other languages, including Latin and Greek). We must
therefore consider the relationship between word-order and case, as
well as the relationship between prepositional phrases and case, in
any general discussion of the category of case.
7.4.4 Case and definiteness

A further category which is interdependent with case in a number of
languages is the deictic category of definiteness (marked in English by
the article the). The Turkish' accusative' is distinguished inflexionally
from the 'nominative' only when the noun is definite (proper nouns,
like personal pronouns, being inherently definite)cf. ev ald'tm,
' I bdught a house': evi aldim, 'I bought the house'. In Finnish,
the converse situation holds: if definite, a singular noun may be in the
'nominative'; if it is indefinite, it will be in the 'partitive' (or
74- CASE 295
* genitive'). In making this point we are not only indicating once again
that the traditionally-recognized case-inflexions of the noun do not
necessarily have just one, or even a primary, syntactic function but
also that definiteness, like animacy, tends to be interdependent with
the syntactic distinction of subject and object even in languages
without distinctions of case, like English.
7.4.5 * GrammaticaV functions

In many treatments of the case-inflexions found in various languages
a distinction is drawn between their 'grammatical* and their 'local*
functions: the distinction is sometimes formulated in terms of an
opposition between the more 'abstract' ('grammatical') and the more
' concrete' (' local') functions of particular cases. Of the ' grammatical'
functions that have been recognized in the description of many
languages, the following are of particular importance in the context of
the present discussion:
(i) subjective ('nominative')
(ii) objective ('accusative')
(iii) indirect objective ('dative')
(iv) adnominal 'possessive' ('genitive')
(v) instrumental
(vi) agentive
(vii) comitative
In this section we will take for granted the general validity of most of
these 'grammatical' functions, although we shall have occasion to
examine them more closely later. It is important to realize that the
functions themselves are not necessarily realized by case-distinctions
in all languages. The labels in brackets after the first four terms in the
list give the traditional names for the four cases which have these
functions in the Indo-European languagesLatin, Greek, Sanskrit,
German, Russian, etc. The reason why no labels are attached to the
instrumental, agentive and comitative functions is that they are
realized differently in different Indo-European languages. The
following English sentences illustrate the application of the seven
functions distinguished above:
(1) Bill died
(2) John killed Bill
(3 a) John gave the book to Tom
(3 ^) John gave Tom the book
(4) It is Harry's pencil
(5) John killed Bill with a knife
(6) Bill was killed by John with a knife
(7) John went to town with Mary
In (1) Bill is the subject, in (2) and (5) Bill is the (direct) object; in
(2), (3), (5) and (7) John is the subject; in both (3a) and (36) Tom is
the indirect object (n.b. the preposition to is normal with nouns,
though not necessarily with pronouns, in the version exemplified by
(3a): there are many factors involved in the determination of the
relative order of animate and inanimate nouns and of animate and
inanimate pronouns, and there are certain dialectal differences); in
(4) Harry is in the adnominal 'possessive' relationship with respect
to pencil; in (5) and (6) a knife has the instrumental function; in
(6) John is agentive; and in (7) Mary has comitative function ('in
company with'). There are certain transformational relations between
particular pairs of sentences: the object of the active (5) 'becomes*
the subject of the corresponding passive (6), and the subject of (5)
' becomes' the agent in (6); the subject of the intransitive (1) * becomes'
the object of the transitive (2), given that the verbs die and kill are in
a 'causative* relationship. Sentence (4) is also transformationally-
related to The pencil is Harry's and Harry has a pencil (notice that
definiteness and indefiniteness, as well as other factors, are involved
in the distinction of these last two sentences). We will return to these
several transformational relationships later, in the discussion of
subject and object, active and passive, intransitive, transitive and
causative. The term 'adnominal' should, however, be explained at
this point.
It has long been recognized by grammarians that the term ' posses-
sive* is semantically far top specific for the most typical function of
the 'genitive*. Not only in Indo-European, but in many genetically-
unrelated languages throughout the world, there is a striking
parallelism (a) between the adjective and the 'genitive*, on the one
hand, and (b) between the 'genitive* and the subject and/or the object
of a verb, on the other. The most typical function of what is called the
'possessive*, or 'genitive', is to modify a noun, or noun-phrase, in an
endocentric construction (for endocentric v. exocentric constructions,
cf. 6.4.1), and this is also the most typical function of the adjective:
74- CASE 297
cf. Harry's pencil, the red pencil. It is this function to which the
traditional term 'adnominal' is given (as the term 'adverbial' is given
to the function of modifying the verb in an endocentric construction:
but the term 'adverb* was used far more widely in traditional
grammar). And the adnominal 'attributive' function of nouns and
adjectives is related to their 'predicative' function with 'the verb to
be': cf. The pencil is Harry'*s: Harry's pencil, The pencil is red: the red
pencil. So too the endocentric construction adnominal + nominalized
verb is transformationally-related to either the subject-predicate
construction or the object-verb construction. These are traditionally
distinguished as the subjective and objective 'genitive': cf. Bill's
death, John's killing of Bill: Bill died, John killed Bill: Bill's murder
{by John): Bill was murdered {by John). Here again one observes the
relationship between the object of the active and the subject of the
passive. In English, unlike Latin for instance, the adnominal sub-
jective and objective functions have alternative realizations (which
are partly in free variation and partly in complementary distribution):
the ' genitive' Bill's and the ' prepositional phrase' of Bill Only the
first of these is a case of the noun in the traditional sense of the term
case. In fact, even the English 'genitive' is at some remove from the
traditional conception of a case, since the inflexional suffix -'s is not
necessarily attached to the head-noun of the noun-phrase: cf. the
queen of Sheba's beauty. In the following chapter, we will endeavour
to relate adnominal function to subjective and objective function
within a more general framework.
With the exception of the 'genitive' {Bill's) there are no case-
inflexions in the English noun. The distinction between subject and
object is marked in some of the personal pronouns (/: me, he: him,
etc.)the object case also being used with prepositions {to me, with
me, etc.); but the distinction is not made for the inanimate, or neuter,
pronoun (it). Otherwise the 'grammatical' functions listed above are
marked in English by relative word-order or prepositions. The
sentences (i)-(7) illustrate this. In Latin, the instrumental case is the
'ablative', in Greek it is the 'dative', in Russian and Sanskrit it is a
case (distinct from the ' dative' in Russian and distinct from both the
'dative' and the 'ablative' in Sanskrit) to which the label 'instru-
mental' is given; in German a preposition is used. In Latin, the
agentive function is realized by a prepositional phrase (d+ ablative), so
also in Greek and German; in Russian and Sanskrit the 'instru-
mental' case also marks the agent. In Latin, the comitative is
realized by a prepositional phrase (cum + ablative), so also in Greek,
Russian, German and Sanskrit. This very concise (and somewhat
simplified) statement of the way the 'grammatical' functions are
realized in a few Indo-European languages shows, first of all, that
there is a similarity of function between the cases and prepositional
phrases (with a particular preposition selecting a particular case-
inflexion) ; and also that there is some overlap between the realizations
of instrumental and agent on the one hand (cf. Russian and Sanskrit:
and note that in English by the knife may be instrumental if there is no
agent or active subject in the sentence), and between the instrumental
and the comitative on the other (cf. with the knife: with Bill; the
typically comitative construction in Latin is also found with instru-
mental function). There are two obvious ways of reacting to this
'overlap* of function: one is to say that the distinction between
agentive and instrumental, or between instrumental and comitative,
cannot be drawn sharply in general grammatical theory; the other is
to say that the 'merging* that is found in particular languages (the
traditional term for this is syncretism) rests upon the neutralization
of the distinction at a more superficial level of the grammar or upon
'recategorization' in terms of animacy or some other syntactically-
relevant notion. To illustrate this notion of recategorization: in a
pair of sentences like This is the man that killed Bill and This is the
knife that killed Bill, the phrases the man and the knife would appear
to have the same syntactic function. One might maintain, however,
that the knife is an instrumental which has been syntactically 're-
categorized* (the traditional notion of 'personification* is here
relevant) as an 'actor' rather than a 'thing*, and that this is possible
in English when there is no 'actor' overtly referred to as subject or
agent. 'Notional' explanations of this kind, although they have been
abused in the past, should not be condemned out of hand. In the
present instance, the analysis finds some support in the fact that in the
corresponding passive sentences the man and the knife are overtly
distinguished as agent (by) and instrument (with): This is the man that
Bill was killed by, This is the knife that Bill was killed with. If This is
the knife that Bill was killed by is possible, it certainly suggests ' re-
categorization ' of the knife as an agent.
7.4.6 'Local* functions

We may turn now to the so-called 'local' functions of the category of
case. The term ' local' must be understood to include temporal as well
as spatial distinctions, since these are commonly brought together in
74- CASE 299
the ' orientational' systems associated with different languages (cf.

7.2.1): it is for this reason that many linguists prefer to talk of
'concrete', rather than * local', case-distinctions. As we shall see, it is
in connexion with the category of case that we find one of the strongest
reasons for insisting upon the necessity of relating syntax to the
spatiotemporal, 'orientational', framework within which language
operates.
Table 11. The so-called 'local* cases of Finnish
' Interior' Inessive talossa 'in the house'

Elative talosta 'from (inside) the house'
Illative taloon 'into the house*
' Exterior' Adessive talolla 'at (or near) the house*
Ablative talolta 'from (outside) the house'
Allative talolle 'to (or towards) the house'
'General' Essive talona
Partitive taloa
Translative taloksi
A relatively simple system of 'local' oppositions is found in

Turkish: eve> 'to the house'; evde, 'in the house'; evden, 'from the
house' (cf. Table 10, 7.4.2). The same three-way opposition of 'to':
'in/at': 'from' is combined in Finnish with what is traditionally
referred to as 'exterior' v. 'interior' (cf. Table 11: except for the
'illative' and 'allative', it will be noted that the distinction between
the 'exterior' and 'interior' cases is quite evidently marked by the
'agglutinating' suffixes -/- and -s-: we will refer to the 'general'
cases presently). Other languages form more complex combinations
of 'local' case-distinctions, involving either an 'absolute' or a
'relative' point of reference. By an 'absolute' point of reference is
meant some feature of the environment (a particular river or mountain,
for example) to which reference is made in indicating situation or
direction ('upstream of, 'far from the mountain', etc.): case-
distinctions of this kind are found in some Australian and American-
Indian languages. Also to be included under the heading 'absolute'
are such 'local' distinctions as 'north', 'south', 'east' and 'west':
cf. Eskimo anna, 'the one in the north'; qanna, 'the one in the
south', etc. By a 'relative' point of reference is meant some com-
ponent of the typical situation of utterance which serves for the
3OO 7- GRAMMATICAL CATEGORIES
indication of situation or direction. The primary ' local* distinctions

(' to': * in/at': ' from') may be combined with a reference to either the
speaker or the hearer by adding the appropriate case-distinctions to
stems formed from the deictic * demonstratives * (cf. the older English
three-way opposition of hither, here and hence; thither, there and
thence): this feature is found in very many languages. Such notions as
'to the left of* v. 'to the right of, or 'in front of v. 'behind', may
refer implicitly to the speaker, as the 'centre* of the typical situation
of utterance, or explicitly, when they are marked by the inflexion of a
particular noun ('behind the house') to some other 'relative* point
of reference. It is worth noting that in many languages which express
these 'local* distinctions by means of a set of nouns (comparable with
back or top in the English ' complex prepositions' at the back of, at the
top of) the nouns in question are those used otherwise to denote parts
of the human body'head*, 'foot*, 'back*, 'face*, etc. But even in
those languages where the same, or similar, 'local* distinctions are
made by means of case-inflexions (comparable with the Finnish
marking of 'exterior* v. 'interior*) it is clear that the application of
these distinctions rests upon the transference to the explicit point of
reference of a categorization primarily suited for orientation with
respect to the participants in a typical situation of utterance.
The most general distinction to be recognized within the 'local'
functions of the cases is locative v. directional ('in/at* v. 'to* or
'from*). Whether the resultant form has spatial or temporal reference
will depend primarily upon the subclassification of the inflected
noun: cf. 'in the house*, 'from the house*: 'in childhood*, 'from
childhood*, which would be translated in a number of languages by
single inflected words. The terms 'locative* and 'directional* them-
selves are to be interpreted (like 'local*) as neutral with respect to
the distinction of space and time; and the distinction between ' to * and
'from' is a secondary distinction within 'directional*.
The opposition of 'locative* and 'directional* may be regarded as a
particular manifestation of a more general distinction between static
and dynamic. We shall return to this point in our discussion of the
category of aspect in the following section. Here it is sufficient to say
that as location is to motion, so being in a certain state or condition is
to change into (or from) that state or condition: in other words, there
is a notional parallelism between such static expressions as '(be) in
London', '(happen) on Tuesday*, '(be) a teacher* and their dynamic
counterparts' (go/come) to London',' (last) until Tuesday','(become)
a teacher*. Similarly, as location is to motion, so possession is to
CASE
74- 301
1
acquisition (and loss): as '(be) in London is to '(have) a book', so
* (go/come) to London' is to '(get) a book' (cf. 8.4.7).
If we use the general term * state' to refer to location, quality, condi-
tion, possession, etc., we can draw a distinction between those states
which are seen as permanently (or necessarily) associated with
particular persons and objects and those states which are regarded as
only temporarily (or contingently) associated with them. Now the use
of the dynamic form (expressing motion, change of condition,
acquisition or loss) presupposes that the state in question is contingent
(i.e. ' non-essential' or * accidental'), rather than necessary.
Furthermore, this distinction between the contingent and the neces-
sary is marked in the case-inflexion of predicative nouns and adjec-
tives in certain languages. In Russian, for instance, the ' instrumental'
case is used (except in the 'present tense') for contingent states of
quality or condition: ja byljstal soldatom ('I was/became a soldier'),
whereas the 'nominative' is used for more permanent, or necessary,
states. In Finnish, the 'essive' is employed in the static form for
contingent, periodic or temporary states of quality or condition
('while he was a teacher', 'in his capacity as teacher', etc.), and the
'translative' in the dynamic counterpart of these ('he became a
teacher'; also in such expressions as 'it turned blue', etc.). The
distinction between the contingent and the necessary may also be
relevant within the class of relationships to which we have given the
name 'possession': if the 'possessed' item is contingently associated
with the 'possessor', it is marked in some languages (notably in
Chinese, and also in the Siouan family) as alienable (' capable of being
given away'), whereas it is unmarked, or marked as inalienable, if it
is necessarily associated with the 'possessor'. Typical instances of
alienable and inalienable relationships would be 'John's book' and
'John's father' respectively.
From antiquity, grammarians have argued about the relationship
between the 'local' and the 'grammatical' functions of the category
of case (in this context 'local' means 'relating to place and time'). In
the classical languages (and in many other languages), the 'local'
and the ' grammatical' functions of a particular case are often hard to
distinguish; so that it is tempting to say that one is derivable from
the other, or that both are derivable from some more general principle
which is neutral with respect to the spatiotemporal and the syntactic.
This also holds for the ' local' and ' grammatical' functions of the pre-
positions in English (which, as we shall see, may be regarded as cases
of the nouns they govern, if the term 'case' is not restricted to
302 7 GRAMMATICAL CATEGORIES
inflexional variation). One might reasonably wonder whether the
* sense' of from in / am from London (i.e. 'a native of London') or
/ came from school is synchronically unrelated to the ' sense' of from
in I got the book from John; and the directional * sense' of to in I went
to London unrelated to the to of the indirect object in I gave the book
to John. No language has yet been studied in sufficient detail from a
generative point of view for it to be possible to say just how much of
the coincidence between the more clearly * local' and the more clearly
'grammatical' functions of cases and prepositions is synchronically
relevant in a particular language. At the same time, certain points of
coincidence across genetically-unrelated languages are so striking that
they demand an explanation in general syntactic theory. One such
point of coincidence, to which we will give some attention later, is
the similarity between the directional and the indirect object in many
languages (cf. I went to London and I gave the book to John, and note
that in / sent the book to John one might hesitate as to whether to John
is directional or indirect object).
7.4.7 Prepositions
Although the category of case is traditionally restricted to inflexional
variation, it is clear that both the 'grammatical' and the 'local'
functions discussed in the preceding paragraphs are logically inde-
pendent of the way in which they are realized in particular languages.
Furthermore, these 'grammatical' and 'local' functions may be
realized in the same language partly by case-inflexions and partly by
other meansmost commonly by prepositions or postpositions, or by
word-order. This means that the category of case cannot be discussed
solely from a morphological point of view.
The difference between prepositions and postpositions is trivial; and
many linguists would say that it is mere pedantry to maintain the
terminological distinction. In the traditional theory of the 'parts of
speech', as it was developed for the description of the classical
languages of Europe, the term 'preposition' was employed to refer
to that class of invariable words, or particles, which had a 'gram-
matical' or 'local' function and which, as it happens in Latin and
Greek, tend to occur immediately before the noun or noun-phrase
they modify. In many other languages (Turkish, Japanese, Hindi,
etc.), particles with similar 'grammatical' or 'local' functions to those
of the Latin, Greek or English prepositions occur after the noun they
modify (cf. Turkish Ahmet ifin, ' for Ahmet'; Japanese Tokyo e} ' to
74- CASE 303
Tokyo'; Hindi Ram ko, ' to Rama'; etc.); and for this reason they are
usually called 'postpositions'. Whether a particle occurs before or
after its noun is a matter of small consequence in general grammatical
theory: for convenience, we will therefore use the more familiar term
' preposition' to cover both classes of particles.
Whether the term ' case * should be extended beyond its traditional
application, to include prepositions as well as inflexional variation, is
also a question of little importance. The difference between inflexional
variation and the use of prepositions is a difference in the ' surface'
structure of languages. What is of importance, from the point of view
of general linguistic theory, is the fact that the 'grammatical' and
'local' functions traditionally held to be inherent in the category of
case can be no more sharply distinguished in those languages which
realize them by means of prepositions than they can in languages in
which they are realized inflexionally. It is the ' interpenetration' of the
syntactic and the ' orientational' framework of language which has
been especially emphasized in this section.
The fact that the ' grammatical' and the ' local' functions of case-
inflexions and prepositions cannot be separated in many, or even in
most, instances does not, however, imply that there is no validity at
all in the traditional distinction of 'grammatical' and 'local' (or
'abstract' and 'concrete'). First of all, it should be observed that it is
not generally possible to substitute one 'abstract' case-inflexion or
preposition for another without making consequent changes elsewhere
in the sentence. One cannot substitute a 'nominative' for an 'accusa-
tive' or 'genitive' in Latin, for example; just as one cannot substitute
he for him or his in an English sentence. This fact can be taken as
typical of what is implied by the term 'abstract' (or 'grammatical')
in this connexion. By contrast, it is frequently possible to substitute
an 'exterior' case for the corresponding 'interior' case in Finnish,
just as it is possible to substitute at for in> or to for into, in English
(/'II meet you atjin the church, Let's go to /into the church). The distinc-
tion between 'exterior' and 'interior' is therefore more 'concrete': it
is not controlled by other syntactic variables in the sentence, and may
convey, of itself, a difference of meaning. As an example of a ' local'
distinction which is intermediate between the fully 'abstract' and the
fully' concrete' we may consider the opposition between the' locative'
and the 'directional': this is less 'abstract' than the distinction
between 'subjective' and 'objective', but less 'concrete' than the
distinction between 'exterior' and 'interior'. The preposition at
cannot be substituted for to in a sentence like He went to church
without also substituting a 'locative' verb (e.g. was) for the * direct-
ional ' went; in that respect the difference is' abstract'. At the same time,
it would be reasonable to maintain that the choice of both went and to
or was and at is determined by the prior selection of a ' directional'
rather than a 'locative' predicate at a 'higher' (or 'deeper') level.
This is the analysis that will be adopted below (cf. 8.4.7). If ft *s
correct, it implies that the distinction between 'locative' and 'direc-
tional' is a semantically-relevant, or 'concrete', distinction of 'deep'
structure, but that the choice of to rather than at, or into rather than
in, is grammatically-determined; and, of itself, it cannot carry a
difference of meaning.
Although the difference between inflexional variation and the use
of prepositions has been described above as a rather unimportant
difference in the 'surface' structure of languages, there is perhaps
some empirical reason to suggest that, if a language has both in-
flexional distinctions and prepositions, the former will tend to have a
more 'abstract' and the latter a more 'concrete' function by reference
to the criterion suggested above. This is certainly true of the
Indo-European languages; and it seems to be true of many other
languages with case-inflexions. Furthermore, it may also be true that,
although word-order is commonly used as a means of distinguishing
between the 'subject' and the 'object', or between the 'modified'
element and the 'modifier', in a sentence, it rarely, or never,
has the more 'concrete' function typical of prepositions. In this
respect, word-order may be a more typically 'grammatical' device
than inflexion, and inflexion more typically 'grammatical' than
the use of prepositions.
7.5 Tense, mood and aspect

7.5.1 Tense
The term 'tense' derives (via Old French) from the Latin translation
of the Greek word for 'time' (Greek khronos, Latin tempus). The
category of tense has to do with time-relations in so far as these are
expressed by systematic grammatical contrasts. Three such contrasts
were recognized by traditional grammarians in the analysis of Greek
and Latin: 'past', 'present' and 'future'. And it has often been
supposed that the same three-way opposition of tense is a universal
feature of language. This is not so. In fact tense itself is not found in
all languages; and, as we shall see, the opposition of 'past', 'present'
7-5- TENSE, MOOD AND ASPECT 305
and * future' is not simply a matter of tense even in Greek and Latin.
The essential characteristic of the category of tense is that it relates
the time of the action, event or state of affairs referred to in the
sentence to the time of utterance (the time of utterance being 'now').
Tense is therefore a deictic category, which (like all syntactic features
partly or wholly dependent upon deixis: cf. 7.2.1) is simultaneously
a property of the sentence and the utterance (cf. 5.2.4). Many
treatments of tense have been vitiated by the assumption that the
'natural' division of time into 'past', 'present' and 'future' is
necessarily reflected in language. Even Jespersen falls victim to this
assumption in his discussion of tense in The Philosophy of Grammar,
before after
o
'now'
Fig. 16. Time and tense.
Making reference to a diagram similar to that which is given in Fig. 16,

he first of all establishes the 'present' as contemporaneous with the
theoretical zero-point (the 'now' of the time of utterance), the
'past' as 'before-now' and the 'future' as 'after-now'. The 'primary'
distinctions of 'past' and 'future' are then subdivided by Jespersen
by means of a 'secondary' application of the notions 'before' and
'after': 'before-past', 'after-past', 'before-future' and 'after-future'.
(As the 'theoretical zero-point', the 'present' is not subdivided.)
The result is a seven-term ' notional' tense-system, partly or wholly
realized in various languages.
But tense admits of categorization in many different ways. One
might grant (as has often been suggested) that the directionality of time
is given in 'nature' (as expressed by the arrowhead in the diagram in
Fig. 16), but this may or may not be relevant to the analysis of tense
in particular languages. Various categorizations are possible. The
'theoretical zero point' (the 'now' of utterance) might be included
with either 'past' or 'future' to yield, on the one hand, a dichotomy
between 'future' and 'non-future', or,"on the other, a dichotomy
between 'past' and 'non-past'. A different dichotomy (based on the
distinction of' now' and ' not-now' without reference to the direction-
ality of time) could be 'present' v. 'non-present'. Other possible
categorizations might depend upon the notion of'proximity' (with or
without reference to directionality): e.g. a dichotomy of 'proximate'
306 7. GRAMMATICAL CATEGORIES
v. 'non-proximate' (with respect to time of utterance), a trichotomy
of 'now' v. proximate' v. 'remote*. And these distinctions might be
combined in various ways, and not merely as suggested in Jespersen's
scheme.
There is no space here to discuss, or even to illustrate, the wide
variety of tense-systems that are found in different languages. Indeed,
it would be difficult to do this satisfactorily quite apart from the
limitations of space, since the analysis of tense, even in English, is a
matter of considerable controversy. The major tense-distinction in
English is undoubtedly that which is traditionally described as an
opposition of ' past* v. ' present*: e.g. They jump: They jumped. But this
is best regarded as a contrast of 'past* v. 'non-past*. The reason is
that, whereas the past tense does typically refer to 'before-now*, the
non-past is not restricted to what is contemporaneous with the time
of utterance: it is used also for 'timeless* or 'eternal* statements {The
sun rises in the east, etc.) and in many statements that refer to the
future ('after-now*). In other words, a form like jumped is positively
'marked* as past, whereas jump (or jumps) is 'unmarked*. This
analysis is supported, if not confirmed, by the fact that the opposition
of past and non-past is realized systematically by sufRxation of the
first element of the verb-phrase: jump: jumped; will jump: would
jump; has jumped: had jumped; is jumping: was jumping; will have been
jumping: would have been jumping; etc.; and it is the one obligatory
opposition of tense in the simple sentence.
What is traditionally described as the 'future' tense in English is
realized by means of the ' auxiliary verbs' will and shall (the rules that
are given for the choice between the two auxiliaries by normative
grammarians being based, for the most part, upon certain pre-
conceived ideas as to what ought to be the difference between them,
rather than upon the usage of any group of English speakers).
Although it is undeniable that will and shall occur in many sentences
that refer to the future, they also occur in sentences that do not. And
they do not necessarily occur in sentences with a future time reference.
They are most appropriately described as modal (like cany may, must,
etc.); and, in our discussion of the category of mood, we shall see
that 'futurity' is as much a matter of mood as it is of tense. Even in
the analysis of Greek and Latin (where the 'future', like the 'present'
and the 'past', is realized inflexionally), there is some reason to
describe the 'future tense' as partly modal.
7.5.2 Mood
Moody like tense, is frequently realized by inflecting the verb or by
modifying it by means of 'auxiliaries'. It is best defined in relation
to an ' unmarked' class of sentences which express simple statements
of fact, unqualified with respect to the attitude of the speaker towards
what he is saying. Simple declarative sentences of this kind are,
strictly speaking, non-modal ('unmarked* for mood). If, however, a
particular language has a set of one or more grammatical devices for
'marking* sentences according to the speaker's commitment with
respect to the factual status of what he is saying (his emphatic
certainty, his uncertainty or doubt, etc.), it is customary to refer to
the ' unmarked' sentences also (by courtesy as it were) as being * in a
certain mood'; and the traditional term for this ' unmarked' mood is
indicative (or declarative).
Two classes of sentences tend to stand apart from all others by
virtue of their modality. The first class comprises imperative sentences,
which do not make statements at all, but express commands or
instructions {Come here!, Put your coat on!, etc.). Since commands or
instructions are generally issued directly to the hearer, what one might
call the ' central' class of imperative sentences are associated with the
'second person'; and it is a rather striking fact that in very many
languages which inflect the verb for person, number, tense, mood, etc.
(including the Indo-European languages) the form of the verb which
occurs in 'second person singular' imperative sentences is unin-
flected for all these categories (i.e. it is identical with the stem). Many
linguists have taken this fact as evidence that giving commands, rather
than making statements, is the more 'basic' function of language.
Since it is not clear what is implied by 'basic' in this context, we will
not go into the question. In any case, the distinction between giving
commands and making statements cannot be sharply drawn. I want you
to come here would normally be classed as a declarative sentence; and
yet the corresponding utterance, in the right context, might be
understood to express a command no less peremptory or authorita-
tive than Come here!
Interrogative sentences also stand in contrast to declarative
sentences by virtue of their modality. They are not traditionally re-
garded as modal, because in most languages (including Latin, Greek,
English) the syntactic distinction between declarative and interroga-
tive sentences is not associated with a difference of verbal inflexion
or the selection of a particular auxiliary, but with the employment
3O8 7. GRAMMATICAL CATEGORIES
of various interrogative particles or pronouns, with a difference of
word-order, or with intonation, together with the 'indicative mood'.
Linguists do not usually speak of an 'interrogative mood', except in
relation to those languages where questions are distinguished from
modally * unmarked' sentences by the same kind of syntactic devices
as those which characterize other modally 'marked' sentences. But
from a more general point of view, interrogative sentences are quite
clearly modal; and they may be characterized by additional modalities
which indicate the expectations of the speaker. For instance, three
types of 'yes'-'no' questions are commonly recognized in the
description of Latin: (i) 'open' questions (which do not indicate
whether the speaker expects either 'yes' or 'no') with the suffix -ne;
(ii) those expecting the answer 'yes', introduced by the particle
nonne; and (iii) those expecting the answer 'no', introduced by num.
There is a somewhat similar contrast in English between an 'open'
question like Is he here? on the one hand, and He's here, isn't he?, or
He isn't here, is he?, on the other. (By a 'yes'-'no' question is not
meant one that must be answered with ' yes' or ' no', but one that may
be so answered. For instance, Did John come?, but not Who came?, is
a 'yes'-'no' question.)
It we turn now to other modalities, apart from command and
interrogation, we find a large variety of ways in which the 'attitude'
of the speaker is grammatically marked in different languages. At
least three ' scales' of modality may be relevant. The first is the scale
of 'wish' and 'intention'. This may be illustrated by the epitaph
Requiescat in pace, 'May he rest in peace': the Latin 'subjunctive',
requiescat, is in modal contrast with the 'indicative', requiescit, 'he
rests'. (The Greek ' optative', which is distinct from the ' subjunctive',
owes its name to the fact that one of its principal functions was
conceived to be that of expressing wishes.) The second scale is that of
'necessity' and 'obligation': / must go to London next week, etc. The
third is that of ' certainty' and ' possibility': He may be here, He must
be here, etc. I have used the term 'scale' for these different modalities,
because they may be categorized into a larger or smaller number
of subdistinctions (e.g. 'certainty', 'probability', 'possibility', or
'stronger' and 'weaker', or different kinds of, 'obligation' and
'necessity'; and so on). Furthermore, particular languages may merge
any two, or all three, of these scales of modality; or give no gram-
matical recognition to them at all. It should also be noted that there is
an affinity between imperative sentences and the modalities of ' wish'
and 'necessity', on the one hand, and between interrogative sentences
7-5 TENSE, MOOD AND ASPECT 309
and the modality of * possibility', on the other. Indeed, an overtly
interrogative sentence like Will you come here? (we will neglect the
possibility of intonational differences) may be semantically equiva-
lent to the imperative Come here, will you? (or simply Come here!) and
can hardly be classified as being in one modality rather than another.
7.5.3 Intersection of tense and mood

The categories of mood and tense may 'intersect' in various ways.
First of all, a particular modal distinction may be drawn in combina-
tion with one tense, but neutralized with another. For example, the
distinction between the 'obligative' and the * inferential' sense
associated with the auxiliary verb must in English is neutralized in a
non-past sentence like He must come regularly (which might be
equivalent to either 'He has to come regularly' or 'I infer that he
comes regularly'), but it is overtly drawn in the past: He had to come
regularly v. He must have come regularly. (It is also worth observing
that the neutralization of the 'obligative' and the 'inferential' must
is not solely dependent upon tense. The ambiguity of He must come
regularly depends upon the fact that it is determined as ' habitual' in
aspect by the occurrence of the adverb regularly, by contrast, He
must come now is determined as 'non-habitual', and it hardly admits
of the ' inferential' interpretation. The category of aspect is discussed
below.) There is a further distinction within the 'obligative' in
English, which has to do with the acceptance or fulfilment of the
obligation; and this is associated with the choice between must or
have to (with certain differences between these two in the non-past)
and ought. The distinction becomes clear if we compare the perfectly
acceptable sentences / ought to go to New York tomorrow but Tm not
going to or He ought to have gone to New York yesterday but he didn't
with the unacceptable sentences *I must go to New York tomorrow but
Vm not going to or *He had to go to New York yesterday but he didn't.
There is, in other words, a difference in the modal presuppositions
attached to the selection of must/have to and ought. Although these
presuppositions are not dependent upon tense, they have the effect
that from the past-tense sentence He ought to have gone the hearer
will correctly infer that (as far as the speaker is aware) the person
referred to did not in fact go, and from He had to go that he did in
fact go. This distinction within the 'obligative' is sometimes rather
misleadingly described in terms of a difference in the ' force' of the
obligation, or in terms of a difference between 'duty' and 'necessity'.
3IO 7- GRAMMATICAL CATEGORIES
Earlier in this section, it was suggested that the expression of
' futurity' in English (and in other languages) was as much a matter of
mood as of tense. In the first place, it should be observed that
sentences containing will and shall do not necessarily refer to the
future. Among the definitely modal uses of will one may note the
' putative' (e.g. He will be quite a big boy now: to be distinguished from
the * inferential* He must be quite a big boy now) and the ' inductive',
which (to quote Palmer) * is used for *' general'' timeless truths, that may
be proved inductively, of the kind illustrated by Oil will float on
water \ In addition to the purely future sense of shall, there is also a
more obviously modal use, the * promissive': here the speaker puts
himself forward as guarantor, as it were, of the truth or occurrence of
the event he refers to (e.g. You shall have your money by the end of the
week). There are other modalities associated with shall and will.
Furthermore, not only have both 'verbs' developed historically from
what were definitely modals in earlier stages of the language, but
they are paradigmatically opposed to such other modals as may, can
and must in present-day English syntax.
It is true that will and shall are commonly used in sentences referring
to the future. But this may be regarded as a ' natural' consequence of
the fact that statements made about future occurrences are necessarily
based upon the speaker's beliefs, predictions or intentions, rather
than upon his knowledge of 'fact'. It is noteworthy that in very many
languages in which it is customary to recognize a future tense, this is
also employed in sentences with modal implications similar to those
mentioned above as characteristic of English sentences containing will
and shall We will give just a few examples. The French f a sera le
facteur (with the future sera) is used in exactly the same circum-
stances as the English That will be the postman. The Russian ' im-
perfective future* (e.g. Ja ne budu rabotatj, 'I won't work') may be
used with implications of 'intention' or 'determination'. Even the
Greek and Latin future tenses were modal in many of their uses; and
their modality (by contrast with the 'present', i.e. non-past, and past
tenses) is confirmed by two facts which relate the future ' tense' to the
subjunctive mood: (i) in certain contexts the future may replace the
subjunctive, and (ii) there is no 'future subjunctive' form in either
language. Thus the traditional scheme of three tenses, with two moods
for Latin and three moods for Greek (excluding the imperative in
both instances), is not wholly satisfactory. For general syntactic
theory, it may be taken as axiomatic that 'futurity' is a notion that
cuts across the distinction of mood and tense.
Whether a language recognizes categories both of tense and mood
is of course a matter for empirical investigation. According to
Hockett: 'Hopi has three tenses: one used in statements of general
timeless truth ("Mountains are high"), a second used in reports of
known or presumably known happenings ("I saw him yesterday",
"I'm on my way right now"), and a third used of events still in the
realm of uncertainty, hence often where we would think of the event
in the speaker's future ("He's coming tomorrow").' In view of what
is said about the implications of these three 'tenses', one might well
think that they would be more appropriately described as 'moods'.
But the general point being made here is that mood and tense may
' intersect' in such a way that either one or the other label is equally
appropriate. The same point may be illustrated by what Hockett says
of Menomini (an American Indian language described by Bloom-
field) : ' Menomini has a five-way contrast, largely of the mode type
[Hockett follows Bloomfield in preferring 'mode' to 'mood'],
though semantically there are traces of tense-like meanings also:
/pi?w/ "he comes, is coming, came": /pi?wen/ "he is said to be
coming, it is said that he came": /pi?/ " Is he coming? Did he come?":
/piasah/ "so he is coming after all (despite our expectation to the
contrary)": /piapah/ "but he was going to come! (and now it turns
out that he is not!)".' One may note in passing that the form /pi?/
could be described as being in 'the interrogative mood' (cf. 7.5.2).
We may note yet another way in which tense and mood may
'intersect'. There are many places in English where would, should,
could and might are rightly described as past-tense forms corres-
ponding to the non-past forms will, shall, can and may (with had to
corresponding to both must and have to): cf. / am going, I will go:
He said that he was going, He said that he would go, etc. But there are
other sentences with would, should, etc., which have no reference to
past time: e.g. That would be a good place for a picnic, You should see
a doctor, It could be true, etc. In such sentences, the ' tense' distinction
of non-past v. past would seem to subcategorize the modality in
question in such a way that 'past' combines with mood to introduce a
more 'tentative', 'remote' or 'polite' sense. In other words, 'tense'
is here 'converted' into a secondary modality.
7.5.4 The modality of subordinate clauses

Our whole discussion of mood has been based so far upon the
assumption of contrast between simple sentences. Something must
now be said about the modality of subordinate clauses in complex
sentences. The traditional term * subjunctive' is revealing in this
connexion: it comes from the Latin translation of the Greek word for
4
subordinating', and shows that for the traditional grammarian the
subjunctive was the mood of subordination par excellence. This point
may be illustrated with reference to French, where (as in Greek and
Latin) what is traditionally referred to as the subjunctive mood is
obligatory in many constructions, and the indicative in others:
e.g. Je crois qu'il vient v. Je ne crois pas qu'il vienne (' I think he is
coming' v. ' I don't think he is coming'; vient is in the indicative and
vienne in the subjunctive). In fact, there are very few contexts in
which the indicative and the subjunctive are interchangeable in
French. The subjunctive rarely occurs except in subordinate clauses,
where its occurrence is very largely determined by the type of
sentence of which the clause is a constituent, by the selection of a
particular main verb, by negation, and by other factors. In other
words, the indicative and the subjunctive forms of the verb are in
almost complementary distribution.
The question now arises whether the French subjunctive is
correctly described as a mood. More generally (since French is by
no means unique in this respect), what is the connexion between
mood and subordination? First of all, it should be noted that, in those
contexts in which the selection of one form of the verb rather than
another (e.g. vienne v. vient) is determined by other syntactic features
of the sentence, the occurrence of one form rather than the other
cannot of itself carry any distinction of modality. For instance, the
occurrence of vienne rather than vient in Je ne crois pas qu'il vienne
does not indicate any particular semantic distinction associated with
the choice of the subjunctive v. the indicative: there is no choice
open to the speaker in this context. Why, then, do we say that the
verb vienne is in the subjunctive 'mood'? Why do we not say, using
a more neutral term, that it is in the subjunctive 'form'?
Three relevant possibilities may be distinguished in the case of
languages in which one set of verbal forms occurs mainly in sub-
ordinate clauses, (i) The distribution of a set of forms, A, and a set
of forms, B, may be partly complementary and partly overlapping;
and the distinction between them, in the contexts in which they
contrast, may be modal. This situation obtains, though minimally,
in French. One might contrast, for example, Dieu vous benit (indica-
tive, 'God blesses you') and Dieu vous benisse (subjunctive, 'May
God bless you'). In Latin and Greek there was a greater degree of
7.5. TENSE, MOOD AND ASPECT 313
contrast between the subjunctive and the indicative, but, as in
French, the selection of one or the other was determined by other
factors in the majority of subordinate clauses. (2) The occurrence of
either A or B in subordinate clauses may correlate with a difference
of modality which is also indicated elsewhere in the sentence. For
instance, the negative sentence Je ne crois pas qvHil vienne expresses
the speaker's doubt, in contrast to the relative assurance of the
positive Je crois qvHil vient. (Actually, this difference in the modality
of the two sentences is one which, in certain contexts, may be
expressed solely by the selection of the indicative rather than the
subjunctive. But this fact may be neglected for the purpose of
illustration.) We may say, therefore, that in the case of Je ne crois pas
qu'il vienne v. Je crois qu'il vient, the form of the verb is compatible
with the modal context in which it occurs; in the same way, the
form of the verb in He came yesterday is compatible with the time-
reference of yesterday. (3) There may be no correlation at all
between the occurrence of one set of forms rather than another and
the modality of the context by which they are determined. In this
situation one would not describe the difference between A and B
as modal, even though one might quite reasonably use the term
* subjunctive' to refer to that set of forms which was more or less
restricted to subordinate clauses. The subjunctive in Greek and
Latin (as in French) is traditionally regarded as a mood because it
satisfies both the first and second of the conditions described above;
but the term 'subjunctive', of itself, carries no implications of
modality.
7.5.5 Aspect
The term aspect (which is a translation of the Russian word vid) was
first used to refer to the distinction of 'perfective1 and 'imperfective'
in the inflexion of verbs in Russian and other Slavonic languages. The
term 'perfective* (or 'perfect') is reminiscent of that used by the
Stoic grammarians for the somewhat similar notion of ' completion'
found in Greek. As we have seen, the Stoics realized that something
other than, and additional to, temporal reference of the kind indicated
by tense was involved in the analysis of the Greek verbal forms
(cf. 1.2.5). But this insight into what we now call 'aspect' was not
taken over by the Alexandrians and their successors.
7.5.6 'Perfective* v. limperfective'

The distinction between the perfective and the imperfective in
Russian may be illustrated by means of the two sentences (i) Ja
procital roman, and (ii) Ja cital rotnan. Although each of these can be
translated as ' I read a novel' (in answer to a question like * What did
you do last night?'), they are not equivalent in Russian. In (i) the
perfective is used (pro-cital), and in (ii) the imperfective (cital). It is
important to realize that the choice between one aspect and the other
is not determined solely by the' facts * of the situation being described.
The perfective is the ' marked * term in the Russian aspectual system,
and the imperfective is ' unmarked' by contrast with it. (This relation
between the two aspects holds independently of the morphological
derivation of the forms: many perfectives are derived by prefixation
of the corresponding imp erfectives, e.g. pro-cital: cital; in other
instances, it is the imperfective which is derived from the perfective
by suffixation, e.g. za-pis-al (perfective): za-pis-yv-al (imperfec-
tive), 'jotted down'.) The use of the perfective in (i) carries with it
the positive implication that the * action' of reading was completed
(i.e. that the book was finished). But (ii) does not say whether the
' action' was completed or not; it merely tells us that the speaker spent
some time reading a novel.
In Greek, as in Russian, the perfective is ' marked' by contrast with
the imperfective. But there is a third term in the Greek aspectual
system, the 'aorist', which, in certain syntactic positions, is in
opposition with both the imperfective and the perfective; and this
three-term system may well have been a feature of the Indo-European
'parent-language'. The Greek perfective is the most 'marked* of the
three aspects: it is perhaps correct to say that, whereas the imper-
fective is 'unmarked' with respect to the perfective, the aorist is
'unmarked' with respect to the imperfective. As we have seen, the
perfective v. imperfective opposition in Greek is one of 'completion'.
It may be illustrated by means of the following sentence (taken from
Plato's Crito): 'Now is no time to be deciding [imperfective, bouleues-
thai] but to have already decided [perfective, bebouleilsthai].' The
Greek perfective refers to the state which results from the completion
of the action or process, so that the verb bebouleusthai in the sentence
just quoted might be translated, somewhat unnaturally but more
revealingly for our present purpose, as 'to be in a state of decision'.
The opposition between the imperfective and the aorist has to do with
the duration of the action or process described. In contrast with the
aorist bouleuseasthai ('to decide', with no indication as to whether the
decision is momentary or not), the imperfective bouleiiesthaimeans 'to
be in the process of deciding*. The three-term opposition of per-
fective, imperfective and aorist in Greek is therefore the resultant of
two binary distinctions: perfective (or completive) v. non-perfective,
and durative v. non-durative. The aspectual notions of 'completion'
and ' duration' are found, either together or singly, in many languages.
We shall not discuss any of the other notions that are customarily
brought together under the term 'aspect': iterative (or frequentative),
punctual (or momentary), habitual, inchoative (or inceptive), etc. To
list just these few examples is sufficient to show that the category of
aspect includes a wide variety of possible distinctions. Like tense-
distinctions, these all have to do with time; but (as Hockett puts it)
with the 'temporal distribution or contour' of an action, event or
state of affairs, rather than with its 'location in time'. Aspect, unlike
tense, is not a deictic category; it is not relative to the time of
utterance.
7.5.7 Aspect in English

English has two aspects which combine fairly freely with tense and
mood: the 'perfect' (e.g. / have /had read the book, I will/would have
read the book) and the 'progressive' (/ am/was reading the book, I
will/would be reading the book). They also combine freely with one
another (7 have/had been reading the book)unlike the two aspectual
distinctions of Greek discussed above (which are in any case only
roughly comparable). There are a number of other aspectual distinc-
tions in English of more limited distribution, including the ' habitual'
(which occurs only with the past tense: I used to read) and the
'imitative' (which is restricted to the passive: I got killed). For a
detailed treatment of the category of aspect (and also of mood and
tense) in English, reference may be made to a number of recent works
listed in the bibliography. One or two points of more general
theoretical interest may be mentioned here.
There are some verbs in English which do not normally occur with
progressive aspect, even in those contexts in which the majority of
verbs necessarily take the progressive form. Among the so-called
'non-progressive' verbs are think, know, understand, hate, love, see,
taste, feel, possess, own, etc. The most striking characteristic that they
have in common is the fact that they are 'stative'they refer to a
state of affairs, rather than to an action, event or process. Since the
3l6 7. GRAMMATICAL CATEGORIES
most common function of the progressive is to indicate duration, one

might say that it is only 'natural* for stative verbs not to combine
with the progressive: the implication of duration is already contained
in the general meaning of these verbs. This is also true, it should be
observed, of most predicative adjectives in English; a sentence like
*The book is being very heavy is abnormal, if not impossible. But there
are some adjectives which occur with the progressive (of 'the verb
to be'), when they refer to an activity rather than to a state or quality:
John is being very good now. We shall return to this distinction between
'states' and 'activities' in our general discussion of the difference
between verbs and adjectives below (cf. 7.6.4). Here it should be
observed that all the 'non-progressive' verbs can in fact take the
progressive aspect under particular circumstances. One of the
occasions for the use of the progressive aspect with stative verbs in
English is in fact to ' recategorize' them, ad hoc, as 'verbs of activity'.
Even such a sentence as / am having a headache, unusual though it is,
can be given an interpretation as 'activity' in the appropriate circum-
stances. It must also be noted that the 'non-progressive' verbs in
English are not necessarily stative wThen they combine with either the
past tense or one of the modals: e.g. As soon as I saw him, I knew that
there was something zvrong; You will feel a slight pain when I insert the
needle (in these sentences the verbs knew and feel refer to an ' event'
to the beginning of a state, rather than to the state itself).
7.5.8 Intersection of tense and aspect

We have already seen that the distinction between tense and mood
cannot be sharply drawn in English (or in other languages) with
respect to future time. Aspect also 'merges' with both tense and
mood. For example, the perfect is used with just to refer to the very
recent past (/ have just seen him); it forms a ' past-in-the-past' in
reported speech (He said he'd seen him the day before may correspond
to I saw him yesterday in the same way as He said he was reading
corresponds to / am reading); and (to quote Palmer's formulation of
this point) the past and the non-past perfective forms 'indicate
periods of time that began before and continued up to a point of
time, the present moment in the case of the present tense, and a point
of time in the past in the case of the past tense'. All these features of
the English perfect give support to the traditional view that, in
certain circumstances at least, it is a secondary, or relative, tense,
rather than an aspect. The progressive is not exclusively aspectual
either, but may have modal implications: in particular, it may express
intention (e.g. / am going to London tomorrow).
Tense, mood and aspect 'merge' into one another in many other
languages besides English. This is partly because certain notions, as
we have seen, might be classed equally well as modal, aspectual, or
temporal; and partly because more distinctions have to be recog-
nized in the semantic analysis of these languages than are overtly
distinguished by the systematic morphological and syntactic contrasts
which we label as 'tense', 'mood' or 'aspect'. No one of the three
categories is essential to human language; and different languages vary
considerably in the wray in which they group together or distinguish
temporal, modal and aspectual notions. It is only wThen these notions
are expressed by means of some such device as inflexion or the use of
particles that linguists tend to refer to them as grammatical notions.
There are many languages, for instance, in which temporal reference
is optionally, if at all, by means of lexical items like 'yesterday', 'to-
morrow', etc., or 'now* and 'then'. This is not normally regarded as
tense. In much the same way, various modalities of the sentence can
be expressed in English by means of such adverbials as certainlyp,
probably, perhaps, etc., instead of, or in addition to, one of the modal
auxiliaries.
To conclude this section, it should be pointed out that the account
of tense, mood and aspect that has been given here has necessarily
been rather sketchy and over-simplified in places. In particular, we
have said nothing about the important fact that in any given context
the meaning of one of these grammatical categories may be more
precisely specified, or determined, by the adverbs or adverbial phrases
to which it is syntagmatically related in the sentence.
7.6 The parts of speech

7.6.1 Alleged circularity of traditional definitions
One of the criticisms most commonly made of the traditional defini-
tions of the parts of speech is that they are circular. It is pointed out,
for example, that, if the class of nouns is defined, in 'notional' terms,
as that class of lexical items whose members denote persons, places
and things (and this is one way in which the noun was defined in
traditional grammar), the definition cannot be applied without
circularity to determine the status of such English words as truth,
beauty, electricity, etc. The circularity lies in the fact that the only
3l8 7. GRAMMATICAL CATEGORIES
reason we have for saying that truth, beauty and electricity are
'things' is that the words which refer to them in English are nouns.
The criticism of circularity loses its force as soon as we take into
account the distinction between 'formal' and 'notional' definitions,
and the possibility that the c notional' definitions of the parts of speech
may be used to determine the names, though not the membership, of
the major syntactic classes of English and other languages. Let us
assume that we have established for English a set of syntactic classes,
X, Y and Z (as well as a number of other classes) on 'formal',
distributional grounds (cf. 4.2.9); and that the members of each of
these classes are listed in the lexicon, or dictionary, associated with the
grammar:
X = {boy, woman, grass, atom, tree, cow, truth, beauty,
electricity, . . . }
Y = {come, go, die, eat, love, exist, . . . }
Z = {good, beautiful, red, hard, tall, . . . }
By reference to the lexicon, we can decide for each word in the
language to what syntactic class or classes it belongs. It is true that
not all the members of class X denote persons, places and things (if
* thing' is interpreted as 'discrete, physical object'). However, it may
still be true that all (or the vast majority) of the lexical items which
refer to persons, places and things fall within the class X; and, if this
is so, we may call X the class of nouns. In other words, we have the
'formal' class X and a 'notional' class A; they are not co-extensive,
but, if A is wholly or mainly included in X, then X may be given the
label suggested by the 'notional' definition of A. It is for this reason
that the lexical class which has as its members, not only boy, woman,
etc., but also truth, beauty, etc., is appropriately called the class of
nouns in English. Whether there exists any language in which the
noun cannot be defined in this way is an empirical question. Most of
the statements made by linguists to the effect that the noun is not a
universal category of human language are vitiated by the failure to
take note of the distinction between the criteria for membership and
the criteria for naming the classes. In practice, linguists seem to have
had little difficulty in deciding that one class, rather than another, in a
particular language, is correctly identified as the class of nouns. As we
shall see, the situation is somewhat different with respect to the other
parts of speech recognized in traditional grammar. But in principle the
distinction between 'formal' and 'notional' definition is applicable
there too.
7.6. THE PARTS OF SPEECH 319
It is a more serious criticism that the criteria incorporated in the
definitions are obviously language-dependent (or 'glossocentric'), in
the sense that they do not apply outside a very narrow range of
languages (including Latin and Greek, for which they were primarily
established). This point may be illustrated with reference to the
definitions of the noun, verb and preposition given by Dionysius
Thrax (and taken over by most of his successors in the mainstream
of the Western grammatical tradition: cf. 1.2.5): 'The noun is a part
of speech having case-inflexions, signifying a person or a thing*;' The
verb is a part of speech without case-inflexion, admitting inflexions of
tense, person and number, signifying an activity or a being acted
upon'; 'The preposition is a word placed before all other parts of
speech in word-formation and syntactic constructions'. The noun and
the verb are defined, not only 'notionally' in terms of what they
'signify', but also in terms of their inflexional characteristics; whereas
the definition of the preposition invokes the quite different property
of relative position (in both morphological and syntactic construc-
tions). First of all, inflexion is far from being a universal feature of
language; secondly, languages that have inflexion do not necessarily
manifest the categories of case, number and tense; and the sharp
distinction that is drawn between cases and prepositions in tradi-
tional grammar cannot be sustained in general syntactic theory
(cf. 1.2.5, 74-I> 74-7)- The question that the definitions of Dionysius
Thrax were intended to answer may be put in the following terms:
Given that the sentences of the language have been segmented into
words, to what class would each word be assigned? And the gram-
matical criteria for classification were mainly based upon the surface-
structure properties of words.
It may be taken for granted that any general theory of the parts of
speech which is intended to apply to more than a narrow selection of
the world's languages must give explicit recognition to the distinction
between deep and surface structure and must define the parts of
speech, not as classes of words in surface structure, but as deep-
structure constituents of sentences (cf. 6.6.1). For the remainder of
this section, we shall be concerned with the question whether any, or
all, of the traditional parts of speech can be defined in this way.
7.6.2 Syntactic function of major parts of speech

The distinction between deep and surface syntax was not made
explicitly in traditional grammar. But it was implied, in part at least,
32O 7- GRAMMATICAL CATEGORIES
by the assumption that all complex sentences, as well as various kinds
of non-declarative sentences, clauses and phrases, were derived from
simple, modally * unmarked' sentences. Furthermore, the major parts
of speech were associated with certain typical syntactic functions in
simple sentences; and this was assumed by some grammarians to be
a more important property of the parts of speech than their 'acci-
dental', inflexional, characteristics in particular languages (cf. 7.1.2).
It was asserted that every simple sentence is made up of two parts:
a subject and a predicate. The subject was necessarily a noun (or a
pronoun 'standing for* a noun). But the predicate fell into one of
three types, according to the part of speech or parts of speech which
occurred in it: (i) intransitive verb, (ii) transitive verb with its object,
(iii) the 'verb to be9 (or some other 'copula') with its complement.
The object, like the subject, must be a noun. The complement must
be either (a) an adjective, or (b) a noun.
The notions of 'subject', 'predicate', 'object' and 'complement'
will be discussed in the following chapter. For the present, we will
take them for granted; and we will assume that the statements made
in the previous paragraph about their association with particular parts
of speech are correct, for English at least. They may be illustrated by
the following sentences:
(1) Mary dances
(2) Mary cooks fish
(3 a) Mary is beautiful
(3 b) Mary is a child
Omitting distinctions of tense, mood, aspect, countability and
definiteness, we can generate these sentences with the appropriate
structural descriptions by means of the following grammatical rules
and associated lexicon:
Grammar Lexicon
(i) S ->A + X A = {Mary, fish, child, .
B = {dance, . . . }
(") X ^lc+A
[D+Y C = {cook, ...}
IE D = {be}
Y
E = {beautiful, .. }
A diagrammatic representation of the underlying constituent-
structure of these sentences is given in Fig. 17. It will be observed that
X and Y are auxiliary symbols, the sole function of which is to label
the nodes (cf. 6.1.3) a n d show that, whereas all four sentences are
syntactically equivalent at the higher level, only (3 a) and (36) are
A X
B
Mary Mary \
dance
cook fish
(2)
A X
D ! D
Mary Mary !
1
be
be
beautiful child
(3a) (3*)
Fig. 17.
equivalent at the lower level of constituent-structure. Apart from the

fact that we have employed the arbitrary symbols A, B, C, D and E
in place of the traditional names for the parts of speech, the rules that
have just been proposed formalize the traditional statements quoted
above about the structure of various types of simple, declarative
sentences. The question that now arises is whether it is possible to
determine the traditional parts of speech or their functions solely on
the basis of the constituent-structure relations that hold between the
four types of sentences.
The answer to this question is clearly 'yes'. The class A is the class
of nouns, since it is the one constituent class which all sentences have
in common at the highest level of constituent-structure; B is the only
class which combines directly with nouns (under the auxiliary node X)
to form sentences, and is therefore the class of intransitive verbs; since
the members of C combine with nouns (^4), and with no other class,
to form predicates (X), C is the class of transitive verbs; and D is the
copula-class, since it combines with both nouns and with E, which is
therefore the class of adjectives.
This argument rests of course on the specific assumptions incor-
porated in the traditional view of the syntactic function of the parts of
speech. In the construction of a more general theory of the parts of
speech, at least two of these assumptions must be challenged: the
first has to do with the status of the copula or * verb tobe' \ the second
with the universality of the distinction between verbs and adjectives.
7.6.3 The 'verb to be'

It is a well-known fact that in many languages the sentences (3 a) Mary
is beautiful and (3^) Mary is a child would take the form 'Mary
beautiful' and' Mary (a) child'; that is to say, the predicative adjective
or noun would be combined directly with the subject-noun without a
copula. Even in the Indo-European languages the copulative function
of 'the verb to be' appears to be of secondary development. The
'original* state of affairs is illustrated by contemporary Russian:
Marija krasivaja (' Mary is beautiful') and Marija rebenok (' Mary is a
child'), where krasivaja is the feminine form of the adjective in
concord with Marija and rebenok is a noun (in the nominative case).
In Latin and Greek 'the verb to be' was optional in such sentences.
However, if we put them in the past tense (or in some other mood
than the indicative), they would necessarily have the appropriate form
of the 'verb to be' in Russian (byla9 budet, etc.), and also in Latin
(eraty etc.) and Greek (en, etc.).
This fact suggests that the principal function of the copulative
'verb to be' in Russian, Greek and Latin is to serve as the locus in
surface structure for the marking of tense, mood and aspect. (' Locus'
is not a standard technical term. It is introduced to refer to the surface-
structure element which ' carries' the overt marking of some syntactic
distinction.) In other words, 'to be' is not itself a constituent of deep
structure, but a semantically-empty ' dummy verb' generated by the
grammatical rules of Russian, Greek and Latin for the specification of
7-6. THE PARTS OF SPEECH 323
certain distinctions (usually 'carried' by the verb) when there is no
other verbal element to carry these distinctions. Sentences that are
temporally, modally and aspectually 'unmarked' (e.g. 'Mary is
beautiful') do not need the 'dummy' carrier.
This account of the function of the copula in Russian, Greek and
Latin can be generalized for English and other languages, which have
a 'verb to be' in sentences that are 'unmarked', as well as in those that
are 'marked', for tense, mood and aspect.
One of the advantages of regarding the copulative ' verb to be' as a
purely grammatical 'dummy' in English (like the 'dummy auxiliary'
do in such sentences as Do they come regularly?\ He doesn't eat fish,
etc.) is that such noun-phrases as the tall man can be transformation-
ally derived from the phrase-markers underlying such sentences as
the man is tall without the need to delete the copula. Grammars of
English which introduce be into the deep structure of sentences (cf. the
rule X-+D+Y in 7.6.2) will necessarily contain many transfor-
mational rules for the deletion of this element in a variety of positions
in surface structure.
A more important point is that the 'verb to be' in such sentences
as Mary is beautiful (unlike the verb cook in Mary cooks fish) is in
contrast with only a limited set of other 'verbs', notably become. The
occurrence of become rather than be depends upon the selection of the
'marked' rather than the 'unmarked' term in yet another gram-
matical opposition (of stative v. non-stative aspect). We will return
to this point presently (cf. 8.4.7). For the moment, it is sufficient to
have made a general case for the elimination of the 'verb to be' from
the underlying constituent-structure of English. There is no doubt
that a similar case could be made for other languages with a copulative
'verb to be'. We may turn now to the distinction between verbs and
adjectives in general syntactic theory.
7.6.4 Verb and adjective

As we have already observed, adjectives were regarded as a subclass
of 'verbs' by Plato and Aristotle, but as a subclass of 'nouns' by the
Alexandrians and their successors; the tripartite distinction of nouns,
verbs and adjectives (as independent parts of speech) was not estab-
lished until the medieval period (cf. 1.2.5). This difference of attitude
towards the adjective in Greek and Latin is readily explained in terms
of the distinction between deep and surface structure. The principal
reason for grouping the adjective and the noun together in Greek and
Latin is that they are both inflected for number and case. But the
inflexion of the adjective is clearly a matter of surface structure: its
number and case (and also its gender) are derived by the trans-
formational rules of concord from the noun which it modifies. Other-
wise the status of the adjective in Greek and Latin is not strikingly
different from its status in English, where there is no concord between
adjective and noun. Plato and Aristotle considered that the most
typical function of both the adjective and the verb was that of
predication, whereas the most characteristic function of the noun was
that of naming the subject of the predication. It was for this reason
that they grouped the adjective with the verb; and logicians have
taken the same view. On the other hand, since the medieval period,
most grammarians have drawn as sharp a distinction between the
adjective and the verb as they have between the verb and the noun.
We may therefore ask what, if anything, distinguishes the adjective
from the verb in general syntactic theory? For simplicity, we will
illustrate from English.
The two most obvious differences between the lexical classes in
English traditionally referred to as adjectives and verbs both have to
do with the surface phenomenon of inflexion. (1) The adjective, when
it occurs in predicative position, does not take the verbal suffixes
associated with distinctions of tense, mood and aspect; instead, a
'dummy verb* (be, become, etc.) is generated by the grammar to carry
the necessary inflexional suffixes. Thus, Mary is beautiful, Mary
would have been beautiful: Mary dances, Mary would have danced; but
not *Mary beautiful-s or *Mary would have beautiful-ed, or *Mary is
dance or *Mary would have been dance. (2) The verb is less freely
transformed to the position of modifier in the noun-phrase; but when
it does occur in this syntactic position, unlike the adjective, it bears
the suffix -ing. Thus, the beautiful girl: the singing girl; but not *the
beautiful-ing girl or *the sing girl.
In * notional' treatments of the parts of speech, adjectives are
frequently said to denote 'qualities', and verbs to denote either
'actions' or 'states'. But the difference between a 'quality' and a
' state' (if it is not entirely illusory) is less striking than the difference
between an 'action' and a 'state'. One might well wonder, for
example, whether know, exist, happy, young, etc. refer to ' states' or
'qualities'. There is no doubt, however, that know and exist, on the
one hand, and happy and young, on the other, fall together grammati-
cally. This question is decided for us by the criteria discussed in the
previous paragraph. But there are many languages (e.g. Chinese) to
which these criteria do not apply; and linguists tend to say that there
is no adjective v. verb distinction in such languages, but rather a
distinction between stative verbs and verbs of action.
A distinction between stative verbs and verbs of action is also
relevant to English. As we have already seen, there are certain stative
verbs in English which do not normally occur in the progressive form
(cf. 7.5.7): by contrast with these, the majority of English verbs, which
occur freely in the progressive, may be called verbs of 'action*. This
aspectual difference between stative verbs and verbs of action is
matched by a similar difference in English adjectives. Most English
adjectives are stative, in the sense that they do not normally take
progressive aspect when they occur in predicative position (e.g. Mary
is beautiful, not *Mary is being beautiful). But there are a number of
adjectives which occur freely with the progressive in the appropriate
circumstances (cf. Mary is being silly now). In other words, to be
stative is normal for the class of adjectives, but abnormal for verbs;
to be non-stative is normal for verbs, but abnormal for adjectives.
The possibility of free combination with progressive aspect correlates
with a number of other important features of English syntax: most
notably, with the potentiality of occurrence in answer to a question
like What did she do?> What is she doing?. Both Mary danced (that's what
she did) and Mary is being silly (that's what she's doing) are possible
answers to questions of this form, but not *Mary knows Greek (that's
what she does) or *Mary is beautiful (that's what she does).
We talk about * stative verbs' in English (as distinct from adjec-
tives) and 'non-stative adjectives' (as distinct from verbs) because the
aspectual contrast of stative v. non-stative in general coincides with,
but in particular instances is in conflict with, the inflexional dif-
ferences traditionally regarded as being of greater importance in the
definition of the parts of speech. It is, however, the aspectual contrast
which correlates, if anything does, with the notional definition of the
verb and the adjective in terms of 'action* and 'quality*.
7.6.5 The adverb

A typical traditional definition of the adverb might run something
like this: the adverb is a part of speech which serves as a modifier of
a verb, an adjective or another adverb or adverbial phrase.
The first point to notice about this definition is the terminological
one. The Latin prefix ad- (Greek epi-) may be translated as 'attached
to and modifying*. But this is also the sense of 'adjective* (Greek
epithetos). The adjective was the modifier par excellence of traditional
grammar: it was 'attached to' and 'modified' the noun (and, for
reasons that have already been discussed, it was regarded as a type of
noun in the post-Aristotelian period). The adjective was therefore a
nominal modifier (adnominal), and the adverb a verbal modifier
(adverbial). But the definition given above makes reference to the
'modification' of adjectives as well as to the 'modification' of verbs.
The point is that the traditional term 'adverb' (and indeed the
definitions of Dionysius Thrax and Priscian) depended, implicitly,
upon the earlier and wider sense of 'verb'. In other words, it pre-
supposed that 'adjectives' and 'verbs' (in the narrower, more modern
sense) were to be regarded as members of the same major syntactic
class for the purpose of stating their combinatorial properties with
respect to members of other major syntactic classes. We have already
seen that 'adjectives' and 'verbs' have much in common, and that in
many languages (including English) they are correctly brought
together as members of the same deep-structure category.
The second point to notice about the definition of the adverb that
we gave above is this: it implies that the adverb is a recursive category
(more typically than the other parts of speech) in the sense that one
adverb may modify another. For example, extraordinarily and well
are both adverbs (of 'degree' and 'manner', respectively) in sentences
like Mary dances extraordinarily well and Mary cooks fish extra-
ordinarily well; and extraordinarily modifies well in the endocentric
adverbial phrase extraordinarily well. It was on the basis of these
combinatorial possibilities in simple sentences that both Jespersen
and Hjelmslev constructed their theories of the parts of speech
(independently of one another and with certain differences which,
in the context of the present discussion, may be disregarded) some
thirty years ago. We will give an outline of their views presently. But
first we must say a little more about adverbs.
In traditional grammar, adverbs constitute a very heterogeneous
class; and it is doubtful whether any general theory of syntax would
bring together as members of the same syntactic class all the forms
that are traditionally described as * adverbs'. We will restrict our
attention at this point to adverbs of 'manner' (as exemplified by well
and beautifully in Mary cooksfishwell and Mary dances beautifully).
Most adverbs of manner in English (and also in certain other
languages) are distinct from, but morphologically-related to, 'adjec-
tives ' (cf. beautifully: beautiful). Furthermore, they are transformation-
ally related to the corresponding 'adjectives' in a variety of parallel
constructions: cf. Mary is a beautiful dancer: Mary dances beautifully.
Since there would seem to be no possibility of paradigmatic opposi-
tion between the * adverb of manner' and the ' adjective', they are to
be regarded as contextually-determined variants of the same * part of
speech*. The attachment of the adverbial suffix -ly (in English) to
'adjectives' like beautiful (and the 'rewriting* of good as well) is to
be handled by the rules which convert the deep-structure analysis into
the surface structure of sentences. In other words, 'adverbial' refers
to the modification of one verb (in the wider sense of this term) by
another verb, the modifying verb being typically, but not necessarily
(cf. smilingly, etc.), an 'adjective'. Not all 'adjectives' occur in
'adverbial' positions: cf. *The light shone greenly, etc. Conversely,
others occur as modifiers of nouns only in constructions which are
transformationally derived from structures in which the 'adjective'
has an ' adverbial' function: cf. a rapid movement - move rapidly.
But the majority of 'adjectives' in English modify both nouns and
verbs in deep structure.
7.6.6 A ' categorial' interpretation of the parts of speech

In our discussion of the adverb, we made reference to the theories of
the parts of speech put forward some years ago by Jespersen and
Hjelmslev. These theories have tended to be neglected by most
linguists in recent years (as part of the general lack of interest in
'notional' grammar). With the possibility of formalizing the distinc-
tion between deep structure and surface structure in transformational
syntax, these theories have assumed a new importance in general
syntactic theory. They have their roots in traditional grammar, and
they were based on evidence from many languages.
We shall refer to Jespersen's formulation rather than Hjelmslev's
(although, on points of detail, Hjelmslev's is somewhat subtler); but
we will not use Jespersen's terminology, which is in conflict with the
sense attached to particular terms elsewhere in this book and might
lead to confusion. For Jespersen, nouns were categories of the first
degree\ verbs (including 'adjectives') were categories of the second
degree; and adverbs categories of the third degree. This notion of what
we are calling 'degree' is defined in terms of the combinatorial
properties of the categories in question. Each category is modified, in
the most typical simple structures, by a category of ' higher' degree.
Nouns are modified by verbs (including 'adjectives'), which are
therefore adnominal categories; verbs are modified by adverbs, which
are therefore ad-adnominal categories; and adverbs are modified by
other adverbs. No more than three degrees are required for the classifi-
cation of the parts of speech (in any language referred to by either
Jespersen or Hjelmslev), since there is no major category whose
function it is to modify categories of the third degree.
It is worth pointing out here that this theory of * degree' can be
formalized very neatly in terms of categorial grammar (cf. 6.3.1): in
fact, it was implicit in the early development of the notions of
categorial grammar by Lesniewski and Ajdukiewicz. The noun is a
fundamental category; all other parts of speech are derived, complex
categories. Categories of the second degree combine with categories
of the first degree (according to the principles of well-formedness
which Ajdukiewicz called * syntactic connectedness') to form
sentences (or 'propositions'). Categories of the third degree combine
with one another to form categories of the third degree.
For typographical simplicity, let us now introduce a numerical
system of notation for the categorial representation of this notion of
'degree'. (The numerals may be defined to be equivalent to the
'fractional* expressions employed in our earlier references to cate-
gorial grammar.) We will use o (zero) for 'sentence', 1 for 'noun',
2 for 'verb' (including 'adjective'), and 3 for 'adverb'; and we will
use 'primes' to indicate recursion, e.g. 3' ('three prime'), 3"
('three double-prime'), etc. The underlying constituent-structure of
a sentence like Mary dances extraordinarily well is given in terms of
these numerical conventions in Fig. 18. It may be represented,
equivalently, as 0(1 + 2(2 + 3(3 + 3'))). The reader will observe that
the system is assumed to be non-directional: given a complex category
x composed of a pair of fundamental or derived categories, we can
' cancel' the ' denominator' (in the ' fractional' representation) of x
with another category y, whether x and y are adjacent to one another
or not in the surface structure of the sentence, and independently of
the relative sequence of x and y in surface structure. In terms of the
numerical notation, 3' cancels with 3 to yield 3, 3 cancels with 2 to
yield 2, and 2 cancels with 1 to yield o.
We pointed out in an earlier section that (bidirectional) categorial
grammars were weakly, but perhaps not strongly, equivalent to
simple, context-free phrase-structure grammars (cf. 6.3.7). Neither
categorial grammars nor simple phrase-structure grammars are
sufficiently powerful for the total description of the syntax of any
natural language. This has been proved by Chomsky (and others).
So far no one has developed a transformational grammar with a
categorial, rather than a phrase-structure, base-component. (Shau-
mjan's theory comes closest to this conception; but it has certain other
features which make it rather difficult to compare formally with
Chomsky's theory of transformational syntax, and it has not yet been
illustrated in detail.) Nevertheless, the notion of 'degree* has a good
deal of support in the traditional theory of the parts of speech and in
the application of this theory to the description of many languages
1 2
2 3
Mary
3'
dance
good extraordinary
Mary dances extraordinarily well

Fig. 18. A categorial representation of underlying
constituent-structure.
throughout the world. It is noticeable that much of the work now

being published by a number of linguists (Lakoff, Rosenbaum, Fill-
more, and others), though not by Chomsky himself, assumes the
'categorial' identity of the 'verb' and the 'adjective' in the deep
structure of English. It is not inconceivable that further developments
will justify the view that the base-component of a more adequate
transformational grammar of English than is yet available will include
a 'sub-component', the rules of which can be more elegantly formu-
lated in terms of categorial grammar. There are, however, other
features of deep structure which cannot be formalized by categorial
grammars. We shall turn to these presently, drawing upon our
discussion of the secondary grammatical categories and the parts of
speech in this chapter. But first we must refer to an alleged in-
adequacy of categorial grammars.
33 7' GRAMMATICAL CATEGORIES
7.6.7 An alleged inadequacy of categorial grammars

It has been suggested that categorial grammars have one serious
inadequacy by comparison with phrase-structure grammars. The
criticism is this. A categorial grammar will identify Adjectives' and
intransitive verbs' as members of the same major category (e.g.
beautiful and dance in such sentences as Mary is beautiful and Mary
dances\ but will fail to relate 'transitive verbs' and 'intransitive
verbs'. At first sight, this is a damaging criticism. But the relation-
ship between 'transitive verbs' and 'intransitive verbs' is by no
means as straightforward as current transformational theory would
suggest. We must leave that question for a later chapter (cf. section
8.2). At this point we will draw attention to an equally serious in-
adequacy in the base-component formalized along the lines proposed
by Chomsky in his most recent work. The point is that any theoreti-
cally satisfactory solution for phrase-structure grammar could be
extended to categorial grammar; and it would automatically dispose
of the criticism referred to above. Let us therefore turn our attention
briefly to phrase-structure grammars, and the way in which they
formalize the rules of the base-component of a transformational
grammar.
7.6.8 An inadequacy of phrase-structure grammars

Consider the following rules:
(1) X-^NP+Aux+VP
(2) VP-+V+NP
(3) VP-+V
(4) NP-+T+N
Assuming a lexicon of the following kind
N = {many doctor, . . . }
V = {examine, leave, . . . }
T = {the, a, ...}
and the necessary rules for the development of Aux (to handle tense,
mood and aspect), for the assignment of number to the nouns and
for the subsequent handling of subject-verb concord, the above
phrase-structure rules will generate such sentences as The doctor
examines the man (T+N+V+s+T+N) and The man leaves
(T+N+V+s). There are various ways in which the relationship
7.6. THE PARTS OF SPEECH 33 1
between 'transitive' and 'intransitive* verbs can be formalized in
a phrase-structure grammar. The above set of rules does so by making
them both members of the same major category (V) and distinguishing
particular members of V, we will assume, by marking them in the
lexicon with a feature (not shown above) which indicates whether the
verb in question may or must have a following NP: cf. rule (2). This is
essentially the technique developed by Chomsky in Aspects of the
Theory of Syntax. (The earlier system of Syntactic Structures
formalized the relationship between 'transitive' and 'intransitive'
verbs quite differently.)
The point we wish to make here is simply this: the rules given above
fail to formalize the fact that there is an essential, language-indepen-
dent, relationship between iVand NP and between Fand VP. As far as
the formalization of phrase-structure grammars is concerned, it is a
matter of 'accidental' coincidence that linguists will include in their
grammars of different languages rules which always expand NP into a
string of symbols containing N and rules which always expand VP
into a string of symbols containing V. In other words, phrase-
structure grammars fail to formalize the fact that NP and VP are not
merely mnemonically-convenient symbols, but stand for sentence-
constituents which are necessarily nominal and verbal, respectively,
because they have JV and V as an obligatory major constituent.
(Chomsky himself has recognized this particular inadequacy: he has
indicated a possible solution, but this has not yet been developed in
detail, and it is not referred to so far in the published literature.)
What is required, and what was assumed in traditional grammar, is
some way of relating sentence-constituents of the form XP to X
(where X is any major category: N> Vy etc.). It would not only be
perverse, but it should be theoretically impossible for any linguist to
propose, for example, rules of the following form (for the base-
component of English or of any other language):
(1a) X->VP+Aux + NP
(za) NP-+V+VP
(2a) NP-+V
(4a) VP-+T+N
The present system of formalization does not exclude rules like this;
and they are equivalent, not only weakly, but perhaps also strongly,
to the four rules given earlier. (This follows from the principles of
phrase-structure grammar discussed in the previous chapter.)
If the problem can be solved for phrase-structure grammars, it can
be solved also for categorial grammars. To take just the case of
'transitive' and 'intransitive* verbs: any lexical item with the
categorial classification (2n)/n, e.g. a 'transitive' verb, would be
defined by the general principles of the system to be the head of a
phrase analysed by the grammar as Zn. Similarly for 'intransitive'
verbs, which have the categorial classification Zn. However, as we
have already remarked, the question of transitivity is more complex
than we have so far indicated. No current system of transformational
grammar handles all the facts correctly. We return to this point in
the following chapter.
7.6.9 Categories and features

The purpose of the above discussion of categorial and phrase-
structure grammars was not merely to focus attention upon a rather
important difference between the deep-structure combinatorial
properties of the major parts of speech relative to one another, on the
one hand, and their various inflexional and transformational charac-
teristics, on the other.
We have seen that the difference between what are traditionally
distinguished as 'verbs' and 'adjectives' in English is rather complex,
involving differences of aspect ('state' v. 'action') and differences of
inflexion. But we have also seen that the inflexional and aspectual
characteristics do not always coincide. 'Adjectives' and 'verbs' are
similar in the way in which they combine with nouns to form
bracketed strings of constituents in deep structure; and this might
be appropriately referred to as their categorial function. The trans-
formational, aspectual and inflexional differences associated with
particular subclasses of this 'second-degree' category are more
satisfactorily specified in the lexicon by means of features which
differentiate the subclasses.
Current transformational grammars, formalized according to the
system recently proposed by Chomsky, make use of such features, but
only at the lowest level of constituent-structure. Our discussion of the
minor parts of speech and the secondary grammatical categories in
this chapter would suggest that tense, mood, aspect, number,
definiteness, etc., are associated with constituents of various levels.
Such categories cannot be regarded as universal categories of
human language (although they can be defined in universal terms
within the general theory of syntax). Languages vary in respect of
the ' selection ' they make from the total set of secondary grammatical
categories recognized in general syntactic theory: and the way in
which the oppositions within these categories are realized in surface-
structure also varies considerably from language to language. What
may be universal in human language are the combinatorial properties
of the major categories relative to one another (as suggested in the
theories of Jespersen and Hjelmslev). If this is so, we can envisage the
possibility that the base-component of a transformational grammar
for any language will comprise two 'subcomponents'. The first
(whether it is formalized in terms of rewrite rules or not) would be
truly universal and would account for the categorial combination of
lexical items. The second would contain rules associating features of
tense, mood, aspect, number, definiteness, etc., at various levels of the
constituent-structure generated by the categorial subcomponent.
7.6.10 Grammatica est una...

The suggestion that has just been made should be regarded as
extremely tentative: it may not be technically feasible, and it may
rest upon insufficient empirical evidence. But one point should be
stressed. A few years ago the majority of linguists would have
rejected the possibility of constructing a universal theory of gram-
matical categories. This is no longer so. As Chomsky has pointed out:
' modern work... has shown a great diversity in the surface structures
of languages', but 'the deep structures for which universality is
claimed may be quite distinct from the surface structures of sentences
as they actually appear*. It follows that 'the findings of modern
linguistics... are not inconsistent with the hypotheses of universal
grammarians'. Once again Roger Bacon's famous statement about
universal grammar is being quoted with approval by linguists:
' Grammar is substantially the same in all languages, even though it
may vary accidentally' (cf. 1.2.7). The scholastic terms 'substantially'
(secundum substantiam) and ' accidentally' (accidentaliter) may yet be
given an acceptable syntactic interpretation in terms of the formaliza-
tion of the distinction between deep and surface structure.
John Lyons
Online ISBN: 9781139165570
Chapter
8 - Grammatical Functions pp. 334-399

8
GRAMMATICAL FUNCTIONS
8.1 Subject, predicate and adjunct

8. i. i Nuclear and extranuclear constituents
It is a fundamental principle of traditional grammar, and also of much
modern syntactic theory, that every simple, declarative sentence
consists of two obligatory major constituents, a subject and a predicate;
and that it may contain, in addition, one or more adjuncts. Adjuncts
(of place, time, manner, reason, etc.: we shall come back later to the
various kinds of adjuncts) are optional, or structurally dispensable,
constituents of the sentence: they may be removed without affecting
the remainder of the sentence.
To illustrate the way in which these terms are applied, we may
refer to the sentence John killed Bill in Central Park on Sunday. The
subject is John; the predicate is killed Bill; and in Central Park and
on Sunday are adjuncts, of place and time respectively. Either or both
of the adjuncts may be omitted without destroying the grammati-
cality of the sentence: John killed Bill on Sunday, John killed Bill in
Central Park, John killed Bill By contrast, neither ^killed Bill in
Central Park on Sunday nor *John in Central Park on Sunday are
grammatically complete sentences (for the notion of 'incomplete'
sentences, cf. 5.2.3). We will say that the subject and predicate
together form the nucleus of the sentence. The subject and the
predicate are therefore nuclear, and adjuncts extranuclear, con-
stituents.
8.1.2 Topic and comment

It has already been mentioned that, from the time of Plato onward,
the definition of the noun and the verb has been closely associated
with the distinction of subject and predicate (cf. 1.2.5, 7-6-4)- Sapir
was merely repeating the traditional view when he said: ' There must
be something to talk about and something must be said about this
subject of discourse. . .The subject of discourse is a noun... No
language wholly fails to distinguish noun and verb, though in
[334]
8.1. SUBJECT, PREDICATE, ADJUNCT 335
particular cases the nature of the distinction may be an elusive one.'
In this passage, Sapir implicitly defines the subject as the person or
thing about which something is said, and the predicate as the state-
ment made about that person or thing. But this is only one of the
ways in which subject and predicate have been defined by gram-
marians. Since we shall also be considering some of these other
definitions, we will adopt Hockett's now widely accepted terminology
for the notions referred to by Sapir: we wTill call the person or thing
about which something is said the topic, and the statement made
about this person or thing the comment.
Hockett introduces these terms as follows: 'The most general
characterization of predicative constructions is suggested by the terms
"topic" and "comment"...: The speaker announces a topic and
then says something about it. Thus John/ran away; That new book by
Thomas Guernsey\I haven't read yet. [The oblique stroke in the
sentences used as examples indicates the major constituent-structure
break.] In English and the familiar languages of Europe, topics are
usually also subjects and comments are predicates: so in John/ran
away. But this identification fails sometimes in colloquial English,
regularly in certain special situations in formal English, and more
generally in some non-European languages.'
Two points may be made with reference to the passage quoted from
Hockett. First, 'subject' and 'predicate', as syntactic notions, are
distinguished from ' topic' and ' comment' (although they are said to
coincide with 'topic' and 'comment' in the most frequently-used
declarative sentences of 'English and the familiar languages of
Europe'). Second, Hockett appears to imply that the topic neces-
sarily precedes the comment: he goes on to say, in connexion with
his second example, ' That new book by Thomas Guernsey is spoken
first because it specifies what the speaker is going to talk about: it is
the topic of the sentence, though not its subject. The topic is at the
same time the object of the verb haven't read {yet), and the subject of
that verb is /, part of the comment of the whole sentence.'
The topic-comment distinction is frequently glossed (though not
by Hockett) in terms of contextual dispensability or predictability:
the topic, or 'subject of discourse', is described as that element which
is given in the general situation or in some explicit question to which
the speaker is replying; and the comment as that part of the utterance
which adds something new (and thus communicates information to
the hearer). By this criterion, we cannot say what is the topic and
what is the comment in a particular utterance (or indeed whether it
336 8. GRAMMATICAL FUNCTIONS
can reasonably be divided into topic and comment) unless we know
what is contextually 'given'. For example, if John ran away answers
the question, explicit or implicit, 'Who ran away?', then by the
given-new criterion John is the comment, and ran away is the topic.
If the explicit or implicit question is nothing more specific than
'What happened?', then merely the past tense is contextually pre-
dictable, and all the rest of the utterance is 'new'. Only if the state-
ment John ran away answers the explicit or implicit question ' What
did John do ?' is John the topic and ran away the comment according
to the given-new criterion (and more precisely, it is ran away minus
the tense-specification that is 'new'). The terms 'given' and 'new'
are taken from Halliday, who distinguishes various other notions
involved in the analysis of sentences into topic and comment.
Typically, the utterance John ran away would not be used in reply
to an explicit question in which either John or ran away was 'given',
but rather He ran away ('What did John do?') or John {did) ('Who
ran away ?'). In English, and possibly in all languages, the given-new
criterion finds its principal application, not in the determination of
the syntactic structure of sentences, but in the establishment of the
conditions of deletability and pronominal substitution in the 'situ-
ationally-bound', elliptical utterances of connected discourse (for the
distinction of sentences and utterances, cf. 5.1.2).
In many languages, by the use of one word-order rather than
another or the employment of a particular particle, the speaker can
indeed make it clear that he is 'announcing a topic' (not necessarily
'given' in the situation) and then 'say something about it'. This is
only possible to a limited degree in English. It is one (among many)
of the factors involved in the choice of a passive construction (cf.
8.3.3). But it does not operate in the determination of the form oijohn
ran away. Such alternatives as What John did was run away. It was
John who ran away, The one who ran away was John, as well as John
ran away, John ran away, etc. (where the acute accent indicates
contrastive or emphatic stress), are 'marked' for a complex set of
other distinctions (which we will not go into here). John ran away is in
fact structurally ' unmarked' for the distinction of topic and comment:
it exemplifies the most ' neutral' form of the English sentence.
And yet, if we were presented with this sentence, in isolation from
the context in which the corresponding utterance had occurred or
might occur, we would no doubt agree with Hockett (and most
linguists and logicians since the time of Plato) that something is
being said about John, rather than about running away. What moti-
vates our choice in the case of such * unmarked* sentences is an
interesting question. It leads us to the other main approach to the
definition of subject and predicate found in traditional grammatical
and logical theory. It is this second approach which, as we shall see,
also underlies the traditional' notional' definition of the noun as * the
name of a person or thing'; and it may be the only way of defining
both 'subject' and 'noun' (interdependently) that is universally
applicable. We have already hinted at this in the section on the parts
of speech (cf. 7.6.9).
8.1.3 Universals and particulars

At this point we must return to the Aristotelian doctrine of the
'categories' of predication (cf. 7.1.2). It has been mentioned that die
first category of substance was taken to be logically more fundamental
than the remaining accidental properties: substances were persons or
things of which the accidental properties (of quantity, quality,
relation, action, place, state, etc.) could be predicated (or asserted) in
logically well-formed propositions. According to this view, John ran
away, He is in London, My friend is tall, etc., are logically well-formed:
John, he and my friend denote substances (in these instances, persons);
and ran away, is in London, and is tall make predications ('say some-
thing') about these substancespredications of action, place and
quality, respectively.
Now, proper names, as well as pronouns and phrases which
identify a definite person or thing (like John, he and my friend, in the
examples of the previous paragraph) are to be regarded as the most
'substantival'the most truly 'nominal'of the expressions in a
language (hence the traditional term 'substantive' for 'noun'). They
are particular (or 'singular') terms, denoting some definite, individual
substance. Other words and phrases, including indefinite 'common'
nouns (man, book, etc.) and 'abstract' nouns (goodness, beauty, etc.),
as well as verbs, adjectives and adverbs, are universal (or 'general')
terms: they do not of themselves denote individual substances (unless
they are syntactically determined, in the descriptive specification of
an individual, e.g. the man over there), but they denote either a class of
individuals or qualities, states, actions, etc., which may be associated
with individuals.
Some logicians distinguish two kinds of universal terms (and, for
convenience of exposition, we will adopt this terminological distinc-
tion): (i) sortal universals, which serve to group individuals into
classes (whether these classes are thought to be definable on the basis
of some inherent properties of their members or not), and (ii) charac-
terizing universals, which refer to qualities, states, actions, etc. Typical
sortal universals are the 'common' nouns of traditional grammar;
typical characterizing universals are 'abstract' nouns, verbs, adjec-
tives and adverbs.
On the basis of these distinctions, we can formulate the following
important principle of traditional logic: whereas universal terms are
found in both subject and predicate position in well-formed proposi-
tions, particular terms are restricted to subject position. Stock examples
of propositions constructed out of a particular and a universal term
are Socrates is a man (sortal) and Socrates is wise (characterizing); and
of a proposition composed of two universal terms, Men are wise. (We
will not go into the further traditional principle that, of two universal
terms, it is the less specific term that is predicated of the more specific.)
The traditional distinction of particular and universal terms can be
drawn independently of the Aristotelian and scholastic notions of
substance and accidents. It rests upon the recognition, in the per-
ceptual world, of a number of discrete, temporally-enduring ' entities'
(persons, animals and things), of the principles of identification and
classification, according to which these 'entities' may be named (as
individuals) or ' sorted' into classes by means of the lexical conven-
tions of the language in question, and of a set of recurrent properties,
states, actions, etc., which may be associated with the 'entities'.
There is no reason to doubt that (whatever its philosophical status)
this everyday notion of 'entity' is applicable in a sufficient number of
instances in the investigation of the vocabulary and syntactic structure
of various languages, provided that we respect the distinction between
'notional' and 'formal' criteria (cf. 7.6.1).
8.1.4 Congruence of logical and grammatical criteria

It should now be clear why 'subject' and 'noun' are indissolubly
associated in traditional grammatical and logical theory. The most
typical nouns (to which the standard 'notional' definition applies
without any trace of circularity: cf. 7.6.1) are those which denote
individual persons and things. Their normal syntactic position in
sentences containing just one such item combined with a non-noun
(a verb, in the wTidest sense of this term: cf. 7.6.4) is that exemplified
in John ran away. And the term 'subject' was defined in the first
instance on the basis of such sentences.
What this means, in effect, is that the traditional grammarian or
logician, like 'the man in the street*, when confronted with John ran
away and asked the topic-comment question 'What is being said
about what?', will assume (in default of any contextual indications
to the contrary) that the individual person, John, is more likely to be
the focus of the speaker's interest, rather than the activity of running
away. And confronted with Horses are vicious animals or Virtue is
rare, he will say that the topics are horses and virtue, because of their
syntactic parallelism with sentences composed of a particular and a
universal term (John ran away, or John is good), which most clearly
satisfy the conditions for the application of the traditional principles
for determining the subject and the predicate. In other words, the
traditional topic-comment criterion was implicitly determined by the
substance-accidents distinction in the case of simple sentences con-
taining a particular and a universal term; and the application of the
topic-comment criterion to sentences consisting of two universal
terms was determined by their overt grammatical structure.
In the definition of 'subject* and 'predicate* (as in mfcny other
matters) traditional logic and traditional grammar leaned heavily
upon one another. They both made appeal to the Aristotelian doctrine
of the categories of predication, which gave a philosophical basis to
the view that the world is populated with individual persons, animals
and things (substances) and that these substances are either the
initiators or the victims ('agents' or 'patients') of activities and
processes, are endowed with certain qualities, are situated in particu-
lar places at a particular time, are subject to change, and so on. The
degree to which this view of the world is determined in detail by the
grammatical structure of the classical languages is a vexed question,
which we need not go into here. It suffices for the validation of the
notion of 'subject* in general syntactic theory that the categories of
logic and grammar should be seen as necessarily coincident with one
another in the case of simple declarative sentences containing just one
nominal expression.
Every language may be assumed to have, as its most typical sen-
tence-type of minimal syntactic structure, a class of sentences whose
nuclei are composed of a nominal and a verb (the term ' nominal*
is intended to include nouns, pronouns and noun-phrases; and
the term 'verb* is to be understood in the wider sense which also
embraces adjectives: cf. 8.1.1, 7.6.4). The notions of 'subject* and
'predicate* are first defined, as we have seen, with reference to such
sentences. They are then extended to sentences of more complex
syntactic structure. It is in the course of this extension that a certain
conflict may arise between various kinds of logical and grammatical
criteria for identifying the subject.
8.i.s 'Actor' and'goaU

Consider the English sentence John kills Bill, the nucleus of which
consists of two nominals and a verb. In earlier chapters of this book,
we have accepted the traditional view of the syntactic structure of such
sentences, according to which (in the present instance) the subject is
the noun John and the predicate is the phrase kills Bill, composed of
the transitive verb kills with the noun Bill as its object (cf. 6.2.3,7-6-4)-
The main reason for taking this view is that there is a grammatical
parallelism between the noun John in John kills Bill and the subject
in an intransitive sentence and between kills Bill and the predicate of
an intransitive sentence (the terms 'subject' and 'predicate' being
defined in the first instance, as we have seen, in relation to intransitive
sentences, containing just one nominal). In both John kills Bill and
John runs (away) there is concord between the singular noun John
and the verbs kills and runs; and the number of the object-noun Bill is
irrelevant for the rules of concord in English. If pronouns are sub-
stituted for the nouns, the pronoun standing for John will assume the
same form (the 'nominative* case: cf. 7.4.5) in both the transitive
and the intransitive sentence: He kills Bill, He runs away. The
pronoun standing for Bill in the transitive sentence will assume a
different form (the 'accusative'): John kills him. Furthermore, the
two sentences may be co-ordinated in either order and one occurrence
of John deleted from the resultant compound sentence: John kills Bill
and runs away, John runs away and kills Bill. The transformational
rules of co-ordination require that the constituent-structure of John
kills Bill be identical with that of John runs away at the level accounted
for by the rule S-> NP+VP (to give this rule in its customary
formulation: cf. 6.6.4). And there are many other transformational
rules of English which depend upon the identification of John as the
subject and kills Bill as the predicate in John kills Bill: cf. / deplore
John's killing Bill: I deplore John's running away, etc. Finally, both
John kills Bill and John runs away can be extended with that's what
he does, where the ' pro-verb' do stands for either kills Bill or runs
away, but not simply for kills.
This last fact is connected with another aspect of the traditional
theory of subject and predicate, which has not so far been mentioned.
Both intransitive and transitive sentences in English may answer the
implicit question * What does X do ? ', where X is a nominal expression
and do (in its second occurrence in the question) is a * pro-verb' which
brings together intransitive verbs, on the one hand, and transitive
verbs + their objects, on the other. Whenever this condition holds (and
it does not hold for either transitive or intransitive stative verbs:
cf. 7.6.4), the subject may be described as the * actor' (or * agent'). By
contrast, the object-noun in transitive sentences is the 'goal* (or
patient'). The 'notional* interpretation of the subject as * actor* and
the object as 'goal* often conflicts with some of the other criteria
referred to in the previous paragraph. For example, in the sentences
Wealth attracts robbers and Riches attract robbers, the subjects are
wealth and riches (according to the criterion of subject-verb concord),
but robbers is the only noun that could be reasonably described as
fulfilling the role of 'actor*. Nevertheless, for English (and a number
of other languages, including Latin and Greek) there is some truth in
the traditional view that the subject of an active, transitive sentence is
the initiator of the action, and the object the 'patient* or 'goal*. The
structure of the vocabulary reflects this, in that most transitive verbs
tend to occur with an animate noun as their subject in active sentences,
whereas the subject of an intransitive verb and the object of a transi-
tive verb is relatively indifferent to the distinction between animate
and inanimate nouns.
8.1.6 Conflict between criteria

Outside the Indo-European family, there are very many other
languages (Basque, Eskimo, Georgian, etc.), where there is a system-
atic conflict, as it were, between the principal traditional criteria for
the identification of the subject in active sentences containing a trans-
itive verb. We will illustrate this from Eskimo. If the sentences listed
as (1) and (2) below are compared (the inflexional endings have been
shown as distinct from the stemsqimmi-, 'dog*; agna-y 'woman';
taku-y 'see*by means of hyphens), it will be observed that one
noun in each sentence is marked with the suffix -q and the other with
-p (we are not here concerned with the verbal suffixes):
(1) qimtni-p agna-q taku-b-a-a, 'The dog sees the woman*;
(2) qimmi-q agna-p tahu-b-a-ay 'The woman sees the dog*.
We may say that, in transitive sentences, -p marks the ' actor * and -q
the 'goal* of the 'action*. By this criterion, therefore, qimmi-p is the
13 LIT
subject of sentence (i) and agna-p of (2). But in the translations of
sentences like 'The dog runs away' and "The woman runs away'
(with an intransitive verb), the nouns qimmi- and agna- would have
the suffix -q. This syntactic parallelism between the * goal' of a
transitive verb and the subject of an intransitive verb is generally
referred to as ' ergativity \ We shall return to it later in connexion with
the notion of 'transitivity', which we are taking for granted for the
moment (cf. 8.2.3).
Here it is sufficient to point out that the traditional notion of
* subject', as it applies to transitive sentences, is partly dependent
upon the grammatical structure of Latin and Greek (and other Indo-
European languages), in which the following two conditions hold,
(i) One of the two nouns in transitive sentences (and, where the
'notional' category of 'actor' is clearly applicable, it is the noun
which denotes the 'actor') is marked with the same case-inflexion
(the 'nominative': cf. 7.4.5) as the subject of intransitive sentences,
(ii) The number (and person) of the verb is determined by the
subject-noun of intransitive sentences and the 'actor'-noun of
transitive sentences. Both of these conditions may be exemplified
from English, if we use pronouns rather than nouns: He sees them,
They see him, He runs away, They run away. (In Latin and Greek, as
well as in various other Indo-European languages, the case-distinction
manifest in he v. him and they v. them is also shown in non-neuter
nouns: cf. 7.4.3.) As we have seen, in languages with an 'ergative'
construction, the syntactic feature of case (and sometimes concord,
although we have not illustrated this) would determine the 'goal'-
noun of a transitive sentence as the subject. But this conflicts with
the notion of the subject as the 'actor', rather than the 'goal' (or
'patient'); and, in practice, most linguists treat the 'actor'-'goal'
distinction as dominant for languages, like Eskimo, with an ergative
construction. They would say that the object of a transitive verb has
the same case-inflexion as the subject of an intransitive verb.
In English, as also in Latin and Greek, the ' actor'-' goal' criterion
is in systematic conflict with other grammatical criteria for defining
the subject in passive sentences. In the sentence Bill is killed by John,
the 'actor' is John and the 'goal' Bill In this respect, John kills Bill
and Bill is killed by John are identical. But, whereas John is tradition-
ally regarded as the subject of the former, it is Bill that is taken to be
the subject of the latter. The reasons include the following: (i) the
case of the 'goal'-noun in passive sentences is 'nominative' (in
English this can only be shown by means of pronouns: cf. He hits
them. They hit him, He is hit, They are hit); (ii) the 'goal'-noun of a
passive sentence determines the verb as singular or plural, in the
same way as does the subject of either a transitive or intransitive
non-passive sentence (cf. Bill hits them, They hit Bill; Bill runs away,
They run away; Bill is hit, They are hit); (iii) the 'goal'-noun of a
passive sentence may be identified with the * actor'-subject of either
a transitive or an intransitive non-passive sentence for the purpose of
co-ordination (e.g. Bill challenged John to a duel and was killed, Bill
fell downstairs and was killed).
A further reason for saying that the 'goal', rather than the * actor',
is the subject in passive sentences is that the * actor* is an optional,
extranuclear, constituent (it is ' outside' the nucleus, at least in surface-
structure). Bill was killed is a complete sentence, whereas *John
killed is not. The phrase by John in Bill was killed by John is tradition-
ally regarded as an 'agentive' adjunct, syntactically comparable with
the * instrumental' adjunct with a knife in Bill was killed with a knife
(cf. 7.4.5).
8.1.7 Various kinds of subject

We will discuss the relationship between active and passive sentences
below (cf. 8.3.3). At this point, it may be observed that many linguists
have drawn a distinction between the 'grammatical' and 'logical'
subject of passive sentences; saying that in Bill was killed by John the
'grammatical' subject is Bill and the 'logical' (or underlying) subject
John whereas in the corresponding active sentence John killed Bill the
noun John is both the 'grammatical' and the 'logical' subject (and
Bill the object).
The terminological distinction between a 'grammatical* and a
'logical' subject has recently been adopted by Chomsky. He has
further suggested: 'Topic-Comment is the basic grammatical
relation of surface structure corresponding (roughly) to the funda-
mental Subject-Predicate relation of deep structure. Thus we might
define the Topic-of the Sentence as the leftmost NP immediately
dominated by S in the surface structure, and the Comment-of the
Sentence as the rest of the string.' There is some plausibility in this
suggestion. But the topic cannot simply be identified with the
surface-structure, 'grammatical' subject. In terms of Chomsky's
proposal, the phrase this book would (quite reasonably) be identified
as the topic in both the active sentence This book millions of people
have read and the passive sentence This book has been read by millions
12-2
of people: it is only in the passive sentence that it is a ' grammatical'
subject. Furthermore, one might wonder whether the notion of
4
topic' should be restricted to nominal expressions (unless the notion
'topic' is restricted by appeal to the substance-accidents distinction).
In the Russian sentence, Bezal Ivan, 'John ran (away)', the occur-
rence of the verb in initial position might be taken as an indication
that it is the topic of discourse (' announced by the speaker': cf. 8. i .2),
although Ivan is presumably both the ' logical' and the ' grammatical'
subject. And in the Latin sentences Interfectus est Caesar,' Caesar was
killed', it could be maintained that the topic is interfectus (est), which
is identical with neither the 'grammatical' subject Caesar nor the
'logical* subject (some unspecified 'actor').
Consideration of the examples given in this section shows that
there are far more distinctions involved than can be accounted for by
the simple dichotomy of 'grammatical' and 'logical' subjects. Even a
three-way distinction of 'psychological' subject (the topic), 'gram-
matical' subject (in surface structure) and 'logical' subject (in deep
structure) fails to capture all the distinctions which, at one time or
another, have been associated with the notion of 'subject' in gram-
matical and logical theory.
The principal aim of this section has been to show that the distinc-
tion between subject and predicate is universally and clearly applic-
able only in sentences whose nuclei consist of one nominal expression
and an intransitive predicate; and, in such sentences, the definition of
the subject depends ultimately upon the same criteria as those which
define the noun in general syntactic theory. As we have seen, in
passive sentences and in sentences with more than one nominal in
their nucleus, the traditional criteria tend to conflict with one another
in their application to various languages. It is worth pointing out,
however, that most of the more strictly 'formal' criteria that have
been mentioned for distinguishing the subject have to do with such
'surface' phenomena as case, concord and co-ordination. In the
following section, we shall consider the distinction between the
subject and the object of a sentence from a somewhat different point
of view. But first we must briefly consider the notion of 'adjunct'.
8.1.8 Adjuncts
An adjunct is by definition a 'modifier' attached to a 'head', upon
which it is dependent and from which it can be ' detached' without
any consequent syntactic change in the sentence (for the terms
' modifier' and 'head', cf. 6.4.2). We are here concerned with what
are normally regarded as sentence-adjuncts, rather than with the
modifiers of lower-level constituents (such as adjectives modifying
nouns within the noun-phrase, adverbs modifying verbs within the
verb-phrase, etc.). Examples of sentence-adjuncts have been given at
the beginning of this section (in Central Park and on Sunday, in the
sentence John killed Bill in Central Park on Sunday).
Sentence-adjuncts may be of various ranks (clauses, phrases or
words). For example, the clause as soon as his wife arrived, the phrase
three hours later, and the word immediately can all be attached as adjuncts
to the nucleus (which is itself a complete sentence) John left for the
office. Furthermore, adjuncts fall into various classes according to their
semantic function: they may be adjuncts of time, of place, of purpose,
of result, of condition, and so on. Not all these different classes are
manifest at all ranks. But we will not go into these details here. In this
section, we will restrict our attention to adjuncts of time and place,
which have a particularly interesting place in the structure of language.
8.1.9 Adjuncts and complements

The first point to notice about temporal and locative adjuncts is that
at the rank of the word and the phrase (though not at clause rank)
they are frequently identical in internal structure with temporal and
locative complements (for this sense of 'rank', cf. 5.5.1). So far the
term 'complement' has been employed only in relation to nominal or
adjectival expressions which combine with the 'copula' in such
sentences as Mary is a beautiful girl and Mary is beautiful (cf. 7.6.2).
In traditional grammar, the term is used to refer to any word or
phrase (other than the verb itself) which is an obligatory constituent
of the predicate: for instance, the object of a transitive verb (cf. the
ball in John caught the ball). The predicative complement is syntactic-
ally required, in order to 'complete' the structure of the predicate
(hence the term 'complement'). More particularly, the term 'comple-
ment' is used of such 'adverbial' expressions as in Central Park or
on Sunday in sentences like The parade was in Central Park or The
demonstration was on Sunday. The temporal and locative phrases in
these two sentences are obviously not adjuncts (since * The parade was
and *The demonstration was are syntactically incomplete). The
difference between an adjunct and a complement is, in principle,
quite clear: the former is an optional (extranuclear) constituent, and
the latter an obligatory (nuclear) constituent of the sentence.
In practice, the distinction between sentence-adjuncts and predica-
tive complements is often far from clear. As we have just seen, the
same class of words or phrases may occur as a locative or temporal
adjunct in one set of sentences and as a complement (of the copula)
in the other. This fact alone would be of small consequence. But
consider now a sentence like The demonstration occurred on Sunday.
In traditional accounts of English grammar, occur is regarded as an
intransitive verb (which, by definition, combines with a nominal to
form a sentence-nucleus, and requires no complement). This classifi-
cation of occur implies that The demonstration occurred (unlike *The
demonstration was) is a complete sentence, and therefore that on
Sunday is an adjunct. On the other hand, the semantic relationship
between The demonstration was on Sunday and The demonstration
occurred on Sunday would tend to suggest that was and occurred are
elements of the same type, and therefore that on Sunday is a predica-
tive complement in both instances.
It has already been proposed that the copula is not a lexical item in
such sentences as Mary is beautiful, but a purely grammatical
' dummy' serving as the * locus' for the indication of tense, mood and
aspect (cf. 7.6.3). We may now extend the application of the same
principle to the analysis of sentences in which be is found with a
temporal or locative complement (e.g. on Sunday or in Central Park);
and there seems to be no reason why such * verbs' as occur, happen,
take place, etc., should not also be treated as temporal and locative
copulas in such sentences as The demonstration occurred on Sunday,
etc.
8.1.10 Locative and temporal complements

To account for the underlying constituent-structure of sentences like
The demonstration was {occurred, etc.) on Sunday and The parade was
{took place, etc.) in Central Park, the following two rules, (1) and (2)
would seem to be required
(1) S - Nominal + Time
(2) 2 -> Nominal + Place
in addition to rule (3), which underlies at least the intransitive nuclei
(including the 'adjectival' nuclei) of non-locative and non-temporal
sentences
(3) S -> Nominal + Verbal
Some of the implications of this treatment will be examined later
(8.4.2 ff.). But a number of questions arise at this point in con-
nexion with the distinction (if there is any clear distinction) between
adjuncts and complements.
There is an important condition which must be imposed upon the
class of nominals which occur with temporal complements under the
application of rule (1). Such 'sentences* as *John was yesterday, *The
dog occurred on Sunday, etc., must be excluded as ungrammatical.
Let us therefore draw a distinction between what we will call first-
order and second-order nominals in English, and say that only second-
order nominals may occur in sentences whose underlying structure is
Nominal +- Time.
First-order nominals might well be called substantival nominals,
since in the most obvious instances they denote persons, animals,
things or places: in 'notional' terms, they are the most 'noun-like'
nominals (cf. 7.6.1). By contrast, second-order nominals do not denote
'substances'. Some of them may be items listed in the lexicon
(cf. 4.3.1): e.g. accident, event, etc. But the majority can be trans-
formationally derived from sentence-nuclei generated by means of
rule (3). Our example, the demonstration, is of this type.
Once we take note of this characteristic of second-order nominals,
the syntactic distinction between temporal adjuncts and temporal
complements seems to be even less satisfactory. What is the deep-
structure difference, we may ask, between The demonstration took
place on Sunday and They demonstrated on Sunday (not to mention
They held the demonstration on Sunday, etc.)? If there is any difference
at all, it is not illuminated by invoking a constituent-structure
distinction in the deep structure: Nominal + Time, on the one hand,
and Nominal + Verbal + Time, on the other. It would seem rather
that the two sentences have the same underlying constituent-
structure, with or without additional syntactic features determining
the selection of one surface-structure 'version* or the other. This
indicates that we should introduce into the grammar a rule which
transforms the output of S -> Nominal 4- Verbal (this rule for
nominalization is required anyway) and embeds it in the subject-
position of the structure generated by 2 -> Nominal + Time.
Transformationally-derived second-order nominals may also occur
in sentences with a locative complement, as well as in those with a
temporal complement: cf. The demonstration took place in Central
Park, The death of Churchill has occurred in London, etc. It is to be
observed, however, that not only second-order nominals, but also
fully substantival (first-order) nominals, occur with locative comple-
ments : cf. John was at home v. *John was on Sunday.
The difference between first-order and second-order nominals is
also relevant to an interesting asymmetry in the distribution of the
locative and temporal copulas other than be (semantically, the most
' empty* of all copulas in English). Neither *John occurred on Sunday\
nor *John took place at home is possible (except of course by means of
a conceivable, but unusual, 'recategorization* of Johny such that it is
understood as 'The birth of John', etc.and the fact that such
'recategorization' can be effected and interpreted ad hoc depends
upon the difference we are discussing between the two classes of
nominals). The asymmetry that has been referred to lies in the fact
that be is found with second-order nominal subjects with both
locative and temporal predicates, but not with first-order nominal
subjects and temporal predicates; whereas occur, take place, happen,
etc., are restricted to second-order nominal subjects with both
temporal and locative predicates. These distributional facts may be
summarized as follows (using occur to represent the class which also
includes take place, happen, etc.):
Temporal Locative
First-order: *John was on Sunday John was in Central Park
*John occurred on *John occurred in Central
Sunday Park
Second-order: The demonstration was The demonstration was in
on Sunday Central Park
The demonstration The demonstration
occurred on Sunday occurred in Central Park
There may be semantic differences between predicates containing
occur, happen, take place, etc., in various contexts. But these predi-
cates have one thing in common, and it is crucial to the point at issue:
they are non-stative (cf. 7.6.4). Since first-order nominals tend to
denote 'entities' (enduring through some time-span, although they
may move or be moved from one place to another) and second-order
nominals tend to refer to ' events * (with ' punctual' location in space
and time) the fact that occur, etc., are found with the latter, but not the
former, in the surface structure of locative and temporal sentences is
readily explained. The locative and temporal predicates of first-order
nominals are necessarily stative.
But the difference between first-order and second-order nominals
docs not suffice, of itself, to account for the non-occurrence of such
sentences as *John was on Sunday. It is not simply the shortness of
the time-span referred to by on Sunday which renders this sentence
unacceptable, since *Socrates was in the fifth century B.C. is also
impossible. Instead, we would say Socrates lived in the fifth century
B.C. {in the fifth century B.C. being correctly regarded as a predicative
complement in traditional grammar). Similarly, instead of *This
building has been for thirty yearsy we would say This building has
existed for thirty years. This suggests that live and exist (the former
restricted to animate subjects) are the temporal copulas occurring with
first-order nominal subjects. Like be in locative sentences, they are
purely grammatical * dummies'. But the argument will be taken up
again from this point in a later section (8.4.3).
8.1.11 Tense and temporal adjuncts

All the instances that have been cited here of grammatical sentences
containing what would be traditionally regarded as temporal adjuncts
have tacitly respected a condition of compatibility between the
adjunct and the tense of the sentence. If this condition of compati-
bility is broken, the sentence to which the adjunct is 'attached* is
rendered ungrammatical: e.g. *John killed Bill next week v. John
killed Bill last week. This point need not be laboured. It should be
noted, however, that it casts doubt upon the traditional notion of the
temporal adjunct as syntactically independent of the rest of the
sentence. It is true that the temporal adjunct is in general' detachable*
from a sentence in which it occurs (cf. 8.1.1), but the generative
rules of 'attachment' must make the adjunct and the tense and aspect
of the sentence compatible in the first place. In most current trans-
formational work on English (and on other languages which have
tense), the rules are ordered in such a way that the prior selection of
a particular tense subsequently restricts the choice of a temporal
adjunct; and this may well be the best way of handling the necessary
conditions of compatibility ifl a system of generative rules. The
point is that the temporal adjunct is bound by conditions of compati-
bility to the obligatory (nuclear) category of tense.
We will now take up the discussion of the distinction between
'subject* and 'object* and of various other topics connected with this.
We shall go into considerably more detail than is customary in
general books on linguistic theory. The reason for this is that
problems connected with 'transitivity' and 'voice* are leading to
important revisions in the theory of generative grammar at the
35O 8. GRAMMATICAL FUNCTIONS
present time. Some indication of one possible line of development
will be provided in 8.3.6. But first we must consider the data and
traditional approaches to these questions.
8.2 Transitivity and ergativity
8.2.1 One-place and two-place verbs

At this point it will be convenient to introduce a classification of
verbs in terms of the number of nominals with which they combine
in the nuclei of sentences. According to this classification, we will say
that a verb like die, which requires only one nominal, is a one-place
verb: e.g. the one * place' associated with die is * filled' by John to
form the nucleus of the sentence John died. A transitive verb (e.g. kill)
is a two-place verb, one of the places being filled by the subject and
the other by the object: cf. John killed Bill. Some verbs (e.g. give or
put) are three-place verbs, combining with a subject, a direct object and
an indirect object (or ' directional' locative: cf. 7.4.6): cf. John gave Bill
the book (where the book is the direct object and Bill the indirect object)
or John put the book on the table (where the book is the direct object and
on the table is a directional locative). We will return to three-place
verbs, indirect objects and directional locatives below (cf. 8.2.14).
8.2.2 The term 'transitive9

The traditional 'notional* view of transitivity (and the term itself)
suggests that the effects of the action expressed by the verb 'pass
over' from the 'agent' (or 'actor') to the 'patient' (or 'goal'). There
is no need to emphasize the inappropriateness of the 'notional'
definition of transitivity in respect of many English sentences. As
Robins puts it: 'The weakness of semantic definitions is well illus-
trated here: hit, in / hit you is syntactically a transitive verb, and is
often chosen as an example because the action referred to may
plausibly be said to "pass across" via my fist to you; but hear in
/ hear you is involved in exactly the same syntactic relations with the
two pronouns, and is regarded as a transitive verb, though in this
case, the "action", if any action is in fact referred to, is the other way
round; and who does what, and to whom, in the situation referred to
by the syntactically similar verb in / love you}' As far as it goes, this
criticism of the traditional notion of transitivity (which has often been
expressed) is correct.
8.2. TRANSITIVITY AND ERGATIVITY 351
Once again, however, we must be careful to draw a distinction
between 'formal* and 'notional' definition. It suffices that the
semantic, or 'notional', definition is applicable to the majority of
two-place verbs for the whole of this class of verbs to be called
'transitive*. It was by tacit appeal to this principle (which we have
already discussed in connexion with the traditional notion of the
noun as 'the name of any person, place or thing': cf. 7.6.1) that we
described agna-y 'woman', as the 'actor' and qimmi-9 'dog', as the
'goal' in the Eskimo sentence qimmiq agnap takubaa, 'The woman
sees the dog' (8.1.6). Furthermore, it might be maintained that the
grammatical form of an English sentence like / hear you or / see you
(its parallelism with / hit you, etc.) influences speakers of English to
think of hearing and seeing as activities initiated by the person
' doing' the hearing and seeing. Whether this is a correct account of
perception, from a psychological or physiological point of view, is
irrelevant. If the native speaker of English (and other languages in
which verbs meaning 'hear', 'see', 'smell', etc., are syntactically
parallel with 'notionally' transitive verbs like 'hit' or 'kill') tends to
interpret perception as an activity which 'proceeds' from an 'actor'
to a ' goal', this fact of itself would suggest that there is some semantic
basis for the traditional notion of transitivity. (In fact, see and hear,
as well as love, are not completely parallel with hits kill, etc., in
English. They are 'non-progressive', stative verbs; and, unlike the
'verbs of action', they do not typically occur in sentences answering
a question of the form 'What is X doing?': cf. 7.6.4.) Although the
class of syntactically transitive verbs undoubtedly includes many
verbs (both in English and in other languages) which cannot reason-
ably be said to refer to actions the effects of which ' pass over' from an
' actor' to a ' goal', it is nevertheless true that the traditional * notional'
account of transitivity is clearly applicable to many, if not most,
syntactically (or 'formally') transitive verbs.
8.2.3 The term 'ergative'

In the previous section, various criteria were discussed for deciding
which of the two nominals is the subject of a two-place verb. It was
pointed out that the' actor'-' goal' criterion is in systematic conflict with
criteria of case and concord in languages with an ergative construction,
and far more generally in passive sentences. We will leave passive
sentences out of account for the moment; and we will consider the
relationship between transitivity and ergativity in some greater detail.
The first point to be made is that there are many verbs in English
which may combine with either one or two nominals in sentence-
nuclei (move, change, open, etc.). Consider the following sentences:
(1) The stone moved
(2) John moved
(3) John moved the stone
In (1) and (2) move is intransitive, whereas in (3) it is transitive.
Moreover, there is an important relationship between (1) and (3).
With reference to the information conveyed by (1), we might well ask
'Who moved it?'i.e. 'Who was the "actor" or "agent" responsible
for the movement of the stone ? \ And, if this question is put explicitly,
the answer might be 'John did* (an utterance derived from the
sentence (3) John moved the stone). The term that is generally employed
by linguists for the syntactic relationship that holds between (1) and
(3) is 'ergative': the subject of an intransitive verb 'becomes* the
object of a corresponding transitive verb, and a new ergative subject is
introduced as the 'agent* (or 'cause') of the action referred to. This
suggests that a transitive sentence, like (3), may be derived syntactic-
ally from an intransitive sentence, like (1), by means of an ergative, or
causative, transformation. (The term 'ergative' was coined from a
Greek verb meaning 'cause', 'bring about', 'create'.) It will also be
observed that the causative, or ergative, agent in (3) is an animate
noun: we have already noted the tendency for the subjects of transi-
tive verbs to be animate. We will return to this point.
8.2.4 Causatives
The verb move illustrates one of the ways in which intransitive and
transitive sentences may be related by means of the notion of
causativity: the same verb enters into sentences of both types without
modification of the verb itself. But we also find pairs of different
verbs between which the same syntactic (and semantic) relationship
holds in corresponding intransitive and transitive sentences. Consider
the following two sentences:
(4) Bill died
(5) John killed Bill
In such instances, we may say that the relationship of the transitive to
the intransitive is ' lexicalized'. It is a matter of the lexical structure of
English that we say John killed Bill, rather than *John died Bill The
syntactic and semantic relationship between kill and die is one that
the child learning English must come to appreciate, just as he must
come to appreciate the relationship between the transitive and intransi-
tive uses of the class of verbs including move. For the present we may
call kill and die two * different' verbs. Later, we will consider the
implications of treating them as alternative, syntactically-conditioned,
phonological realizations of the 'same* verb.
In many languages, there is a productive grammatical rule for the
formation of causative verbs. We will illustrate from Turkish and
French. The English sentences (4) and (5) may be translated into
Turkish as
(6) Bill oldu
(7) John Bill-i oldiirdu
Here, the suffix -diir- (which varies in form according to vowel-
harmony: cf. 3.3.13) converts the intransitive stem 67- 'die' into a
transitive stem dl-diir- in the derivation of a sentence like (7). (The
suffix -dii marks past tense; and the -1 attached to the 'goal'-nominal
in (7) marks the object of the transitive verb, when the object is
definiteall proper names and personal pronouns being inherently
definite in Turkish: cf. 7.4.4. Both the tense suffix and the object
suffix are variable in phonological form according to vowel-harmony.)
As another example from Turkish, we may take
(8) Patlicanpiiyor ('The eggplant is cooking')
(9) Ahmet patlicanipipriyor (' Ahmet is cooking the eggplant')
Once again, the two sentences are syntactically related in the same
way as the English sentences (1) and (3), or (4) and (5). (The suffix
-zr- is a less common variant of the causative suffix. Which of the
variants, or 'allomorphs', occurs with a particular verb-stem is in
general determined by a classification of verbal stems for this purpose:
in other words, the distribution of the causative suffixes is not a
matter of free variation, but of lexical conditioning: cf. 5.3.5. The
suffix -iyor- marks the continuous, or 'progressive', aspect. Without
the objective suffixi.e. with patlican rather than patlicani(9)
would be translatable as 'Ahmet is cooking (some) eggplant'. Since
the distinction between a definite and an indefinite nominal is not
marked for the subject in Turkish, (8) is in fact indeterminate
between' Some eggplant is cooking' and the translation given above.)
It will be observed that the English verb cook may be used either
intransitively or transitively. Furthermore, there is an alternative
translation possible for (8): 'The eggplant is being cooked.' The
difference between the intransitive is cooking and the passive (of the
transitive) is being cooked will be discussed below (cf. 8.2.13).
In French, as in English, the relationship manifest in (6) and (7)
is lexicalized in the two different verbs mourir and tuer (cf. Bill est
mort v. John a tue Bill). But (8) and (9) may be translated into French,
as (10) and (11), in a way which illustrates the operation of the
French causative construction.
(10) Les aubergines cuisent
(11) Ahmet fait cuire les aubergines
(Actually, in the present instance, Les aubergines sont en train de cuire,
' The eggfruit is in the process of cooking' as it were, is contextually
more probable than (10). But (10) is possible.) In French, the
intransitive verb cuire is made transitive, not by means of morpho-
logical modification (with a prefix, suffix, infix, etc.), but by use of the
auxiliary 'verb 1 faire ('make', 'do'). This is comparable with the
English make, as exemplified in
(12) John makes the brass shine
which, like
(13) John shines the brass
is syntactically derivable from the intransitive
(14) The brass shines
However, the English construction with make commonly introduces
an implication of force or coercion (though not in the example given
here). This is not generally so of the French construction With faire,
which is far more extensively employed.
Causative constructions, comparable with those illustrated here
from Turkish and French, are extremely common throughout the
languages of the world. They would seem to provide us with a satis-
factory general framework for the discussion of transitivity and
ergativity.
8.2.5 The 'ergative* in the Indo-European languages

Let us now look at the English 'ergative' constructions from the
point of view of case and concord. For reasons which will become
clear immediately, we will use symbols rather than nouns in the model
sentences. The symbols are to be interpreted as follows: A is the
subject of a transitive verb (we will assume that A is necessarily
animate), and B is either the subject of an intransitive verb or the
object of a transitive verb; since B may be animate or inanimate (in
either or both of the syntactic positions in which it occurs) we will
use the subscripts a and i to mark this distinctionBa denotes an
animate and Bi an inanimate nominal. Given these conventions, the
sentences (i)-(3) cited above (8.2.3) m a y ke 'translated' as
(id) Bt moved (cf. It moved)
(2 a) Ba moved (cf. He moved, She moved)
(3 a) A moved Bi (cf. He moved it, She moved it)
There is also a fourth possibility, by virtue of the distinction between
inanimate and animate objects:
(3 b) A moved Ba (cf. He/She moved him/her)
which, for the moment, we will assume is related to (2 a) in the same
way that (30) is related to (ia).
If we now identify the categories of 'neuter* and 'inanimate', in
the Indo-European languages which are traditionally said to have
three genders (cf. 7.3.3), we may say that English is typical of most
Indo-European languages in two respects: (i) Although the case of
Bi in (id) is usually said to be 'nominative' and the case of Bi in
(3a) 'accusative', the occurrence of Bt in subject or object position is
never in fact associated with an inflexional difference in Bt. The
inflexional difference between a' nominative' and an' accusative' form
of nouns and pronouns is relevant only to animate nominals (cf. 7.4.3:
in English, of course, it is relevant only for pronouns). Thus, It
moved, A moved it: He moved, A moved him. (ii) The case of Ba in a
sentence like (2a) is identical with the case of A: cf. He moved, He
moved it/him, etc. Furthermore, if there is number (or gender)
concord between the verb and one of the nominals, Ba in a sentence
like (2 a) determines the number (or gender) of the verb in precisely
the same way as does A in (3 a) or (36), but Ba has no such effect in a
sentence like (3 b): cf. He moves, He moves it /him: They move, They
move it/him. It is mainly because of the features of case and concord
mentioned under (ii) that the Indo-European languages are not
generally regarded as 'ergative' languages (cf. 8.1.6). However, once
we take into account the distinction between animate and inanimate
nouns mentioned under (i), these features of case and concord lose a
good deal of the importance attributed to them in traditional descrip-
tions of the Indo-European languages.
There is some evidence to suggest that the Indo-European system
of case-distinctions did in fact develop from an earlier system in
which the 'nominative' wras an 'agentive', or 'ergative', suffix
(typically s) found only with animate nouns. We will not go into this
question from the diachronic point of view here. It is noticeable,
however, that as far as the case-distinctions of subject and object are
concerned, the difference between, say, English and Eskimo reduces
to the relatively minor fact that in (2 a) Ba has the agentive case in
English, but the non-agentive case in Eskimo. That is to say, if we
think of the 'nominative' in English (and in the Indo-European
languages generally) as the case of the 'actor* (like the Eskimo
suffix -/>), we can introduce a rule into the grammar of English which
has the effect of attaching this case, not only to the ' actor' rather than
the 'goal' with two-place verbs, but also to animate nominals with
one-place verbs. In other words, He moves can be derived by an
obligatory transformation from *Him move (using him, like it, as an
' unmarked' (non-agentive) form, neutral with respect to the distinc-
tion of ' actor' and ' goal').
8.2.6 An îdeaV ergative system

But a distinction between 'agentive' and 'non-agentive' can in fact
be drawn with one-place verbs, 'notionally' at least. Consider a
sentence likt John flew through the air. This might answer either of
two implicit questions: (i)' What did John do ?' or (ii)' What happened
to John?'. If it answers (i), John is an 'agent' (he might be flying an
aeroplane, for instance); if it answers (ii), John is not regarded as an
'agent' (he may have been thrown across the room by someone else).
Some intransitive verbs in English will hardly admit of an ' agentive'
interpretation: cf. John died. Others will hardly admit of interpreta-
tion with a 'non-agentive' animate nominal: cf. John jumped from the
roof. And others will tend to either an 'agentive' or a 'non-agentive'
interpretation with animate nominals, although the less normal
interpretation is also conceivable in certain circumstances. For
example, John fell is probably more normal as a 'non-agentive'
sentence; but the other interpretation is certainly possible. By
contrast, The chimney fell (with an inanimate subject) is necessarily
'non-agentive'.
On the basis of the ' notional' distinction between ' agentive' and
'non-agentive' (relevant only for animate nominals), we might
construct a theoretically 'ideal' system for one-place and two-place
verbs of the kind illustrated by the following ' sentences * (in which he
stands for an 'agentive' and him for a ' non-agentive' animate
nominal: n.b. these are not sentences of English, but pseudo-
English representations of an ' ideal * set of distinctions):
(1 a) It moved
(zb) Him moved
(zc) He moved
(3 a) He moved it
(3 b) He moved him
If these are compared with the four English sentences given in
8.2.5, it wiU t>e s e e n t n a t t n e o n ly difference, as far as the distribution
of the cases is concerned, is that (2 a) of the earlier set has now been
split into {zb) and (zc). This is intended to reflect the possibility, in
principle, of a ' non-agentive' and an 'agentive' interpretation of a
sentence like (2) John moved (8.2.3). This theoretically 'ideal' system
is not realized in English (or in any of the Indo-European languages).
Nor is it realized in languages which are traditionally described as
having an ergative construction. It will be observed, however, that
we can arrive at either the English or the Eskimo system by a simple
'merger' of (zb) and (zc). If the 'merger' takes place in one direction
(zb) Him moved]
/ \ Tj j I -* He moved
(zc) He moved )
the outcome (as far as the category of case is concerned) is the typically
Indo-European system for the masculine or feminine subject of an
intransitive verb. If the 'merger' operated in the reverse direction
(zb) Him moved)
, x rr 7 / -> Him moved
(zc) He moved )
the result would be an ergative construction, characteristic of Eskimo
and many other languages.
8.2.7 Transitivity and animacy
This 'ideal' system for the distribution of an 'agentive' case with
one-place and two-place verbs is not being proposed as a necessary
part of the grammars of English and Eskimo, but merely as an aid to
the understanding of ergativity and transitivity. These two languages
have been taken as examples of what are often regarded as radically
different syntactic types. The general conclusion to be drawn from
the discussion is that in both languages the case-inflexion ,of the
subject of an intransitive verb is of secondary importance. Both
languages agree in the following respects: (i) they distinguish the
' actor' and the ' goal' of transitive sentences by means of a difference
of case (as 'agentive' and 'non-agentive', respectively); (ii) they
employ the same form (the ' non-agentive') for the subject of intransi-
tive verbs as they do for the object of transitive verbs, provided that
the subject of the intransitive verb is inanimate; (iii) they do not
distinguish by case-inflexion between 'agentive' and 'non-agentive'
subjects of intransitive verbs. In each language, the 'agentive' (the
* nominative' in English, the 'ergative' in Eskimo) may be regarded as
the ' marked' term of the opposition of case. The difference between
the two languages lies in the fact that English generalizes the use of
the 'agentive' form to all animate subjects (whether they are
'notionally' agentive or not); and Eskimo generalizes the 'non-
agentive' form. We have not discussed the difference between the
two languages in respect of concord. But it should be clear that this
difference also is a matter of surface-structure, and can be handled in
much the same way as the distribution of case-inflexions.
Although English has been used here to exemplify the use of the
'nominative' and 'accusative' in the Indo-European languages, it is
in many respects an untypical example. First of all, the case-distinc-
tion of ' nominative' v. ' accusative' is not relevant to nouns, but only
to pronouns, in English {he v. him, she v. her, they v. them, who v.
zvhomf etc.). More important, the correlation between 'grammatical'
and ' natural' gender is far greater in English than it is in most of the
other Indo-European languages (cf. 7.3.4). In such languages as
Latin and Greek, as far as the case-inflexions are concerned, the
distinction of' agentive' v.' non-agentive' is ' automatically' extended
to inanimate nouns if they are masculine or feminine in gender. But
this fact does not invalidate the deep-structure distinction between
'agentive' and 'non-agentive' nominals in transitive and intransitive
sentences. In a Greek sentence like lithoipiptousin ap* ouranou>' Stones
fall from [the] sky', lithoi ('stones') is in the 'nominative' case;
although an inanimate noun, it is masculine, and therefore takes the
'nominative' inflexion in subject-position. But lithoi is no more
' agentive' here than is stones in the corresponding English sentence.
Conversely, in the German sentence Das Kind qffnet die Tier, ' The
child opens the door', das Kind ('the child') is neuter in gender: it
cannot therefore take the characteristically 'agentive' inflexion (cf.
Der Mann off net die Tiir> 'The man opens the door'; cf. also Das
Kind sieht den Mann v. Der Mann sieht das Kind, ' The child sees the
man' v. 'The man sees the child'). The deep-structure distinction of
' agentive' v. ' non-agentive' is nevertheless applicable in German to
animate nouns of neuter gender, as well as to masculine and feminine
animate nouns.
Transitivity is bound up, then, with the distinction of animate and
inanimate nominals; and in the 'ideal* system the former may be
either 'agentive' or 'non-agentive' (in both transitive and intransitive
sentences), the latter only 'non-agentive'. The detailed development
of this thesis for particular languages would take us too far from the
main line of argument. We will assume that the general points that
have been made are tenable for at least a considerable number of the
transitive and intransitive sentences of the languages referred to. At
the same time, we must accept that what we are regarding as the
'notional' basis for the system of transitivity has superimposed upon
it in various languages many transitive constructions which do not
satisfy the conditions of the 'ideal' system. For example, Wealth
attracts robbers is a perfectly acceptable transitive sentence of English,
in spite of the fact that wealth is an inanimate noun. It may very well
be that sentences like this (and they are much more common in English
than they are in some other languages, e.g. Greek) should be thought
of as 'parasitic' upon the more 'normal' type of transitive sentences
with an animate subject. However that may be, we shall concentrate
here upon transitive sentences which fulfil the 'normal' conditions.
8.2.8 'Causative9 verbs in English

We have seen that there is a class of verbs in English (to which we
gave the label' ergative verbs': 8.2.3) which occur in both intransitive
and transitive sentences; and that the two-place, transitive construc-
tion is derivable from the one-place, intransitive construction by
means of a causative operation which has the effect of introducing an
'agentive' subject. The example used in the illustrations was move.
Other members of this class of verbs are change, grow, develop, open,
close, start, stop, begin, break, crack, split, tear, etc. They may differ
according to their selection of an 'agentive' or 'non-agentive' subject
in their intransitive usage, but they are 'normal' in that they usually
have an animate subject with the transitive construction. In addition,
there are certain transitive verbs in English, syntactically related as
' causatives' to corresponding intransitive verbs and (from a historical
point of view) derived from them by what were once more or less
productive morphological processes (cf. the Turkish causative verbs
illustrated above: 8.2.4). Examples are lay ('make to lie'),/// ('make
to fall'), etc. (which are, however, more specialized in their selectional
possibilities than the corresponding intransitives). In particular, there
are a large number of transitive verbs morphologically related to
intransitive * adjectival* verbs: enrich {rich), soften (soft), strengthen
(strong), actualize (actual), etc. Most of these morphological patterns
of formation by prefixation and suffixation are no longer productive
in modern English (with the notable exception of -ize, which is
suffixed, not only to verbal stems, but also to nominals: cf. com-
puterize, etc.). These morphological causatives fall between the two
extremes of 'lexicalization', on the one hand (cf. kill: die, discussed
above: 8.2.4), a n ^ the use of the 'same' verb (e.g. move) in both
transitive and intransitive sentences, on the other. By contrast with
'lexical* and morphological causatives, the transitive verbs move,
change, grow, etc., may be described as being derived from the
corresponding intransitive verbs (identical in phonological form, and
for that reason referred to above as the 'same* verbs: 8.2.4) by means
of a morphological process of 'zero modification*. It is worth pointing
out that many of the forms which are traditionally said to occur both
as 'adjectives' and 'verbs' fall within the scope of this classification:
e.g. He warms the milk is related to The milk is warm (which contains
the intransitive, 'adjectival' verb warm) in the same way as He moves
the stone is related to The stone moves. Morphologically, the transitive
verbs move and warm are ' derived' from the intransitive verbs move
and warm by 'zero modification'.
8.2.9 ' Object-deletion'

Apart from the class of verbs exemplified by move, there are various
other classes of verbs in English which would seem to be used
transitively or intransitively. Consider the following sentence:
(15) We never eat at five 6*clock.
The verb eat is here being used without an object. Whether it should
be described as an intransitive verb is, however, a moot point. Many
traditional grammarians would say that it is a transitive verb which
may be employed either with an object or' absolutely'. The' absolute'
usage (without an overt object) is illustrated in (15). If this sentence
is compared with
(16) We never eat caviare at five o'clock
it will be clear that the relationship between the transitive and the
' absolute' usage of eat is quite different from that which holds between
the transitive and the intransitive usage of move. There is no *Caviare
never eats at five o'clock (but rather the passive Caviare is never
eaten at five o'clock) corresponding to (16). It seems reasonable to say
that eat is inherently transitive, but that its object may be deleted (in
the pseudo-intransitive, 'absolute' construction). The deletion of the
object may be contextually determined (and recoverable for the
purpose of semantic interpretation) in particular utterances. But it
may also be a feature of the lexical structure of the language. For
example, / spent the morning painting is susceptible of at least two
distinct interpretations: (i) ' . . .painting pictures' or ' . . .painting the
house (the garden fence, etc.)'. And the 'deletion' of the object in
each case (in the first interpretation it is an ' object of result': cf. 9.5.3)
is determined by the lexical structure of English: there need be no
contextual indication of an object of the verb paint, which the hearer
will feel obliged to infer in order to understand the sentence. In
many languages, there is far less freedom in the matter of 'object-
deletion'. For instance, the Turkish translation of the English
'absolute' sentence He is writing requires a ' dummy' object (morpho-
logically related to the verb 'to write'): Yaziyaz-iyor, 'He is writing
writing', as it were (cf. Mektupyaz-iyor, 'He is writing a letter').
8.2.10 Reflexives
Another class of verbs in English which, prima facie, are used both
transitively and intransitively is illustrated by shave in the following
two sentences
(17) The barber shaved ten men before lunch
(18) He never shaves before lunch
Of these, (17) is a straightforward transitive sentence, with an
'agentive' subject, the barber, and a 'non-agentive' object, or 'goal',
ten men. But (18) is ambiguous: it might mean 'He never shaves
(anyone) before lunch' or 'He never shaves (himself) before lunch'.
The first of these interpretations is accounted for under the term
'object-deletion', discussed in the previous paragraph. It is the
second that concerns us here. Under this interpretation, (18) might
well be described as implicitly reflexive.
A reflexive construction is one in which the subject and object refer
to the same person (or thing). Many languages, like English, have a
set of reflexive pronouns distinguished for person and number
(myself, yourself, himself, etc.); others, like French, German, or Latin,
draw a distinction between reflexive and non-reflexive objects only
in the third person (cf. French, Maman me laveije me lave,' Mummy
is washing me'i 'I am washing myself1; Maman la lave: Maman se
lave, 'Mummy is washing her': 'Mummy is washing herself*); and
there are other languages which use the same reflexive pronoun with
all persons and numbers (e.g. Russian uses either the suffix -sja or
the pronoun sebja, under conditions which we will not go into here).
An explicitly reflexive sentence is one in which the identity of subject
and object is overtly marked, either in the form of the object pronoun
or in some other way (e.g. by the use of a particular suffix, prefix, or
infix, attached to the verb-stem). Thus He killed himself (by contrast
with He killed him) is explicitly reflexive. So too is He washed himself
But it is a characteristic feature of English that many verbs can be used
in reflexive sentences without the occurrence of an object pronoun.
He never shaves before lunch exemplifies this usage; and it is for that
reason described as implicitly reflexive. In fact, shave is very rarely used
in explicitly reflexive constructionsmost commonly perhaps in sen-
tences like / dorit mind shaving others, but I never shave myself, where
the occurrence of myself is required for the contrast with others. (It
may be noticed in passing that, out of context, / never shave myself is
syntactically ambiguous: (i) 'As for me, I never do any shaving'
(object-deletion), (ii) 'As for me, I never shave (myself)' (implicitly
reflexive), (iii) 'I always get someone else to shave me' (explicitly
reflexive). Under the interpretations indicated by the glosses in (i)
and (ii), myself is not the reflexive pronoun, but an emphatic 'adjunct'
to the subject of the sentence, which may occur in various positions
in surface-structure: I myself never shave, I never myself shave, etc.)
The implicitly reflexive construction is commonly found with wash,
which may also be explicitly reflexive; but not, for instance, with dry.
He dried in the sun is interpretable as an intransitive sentence ('non-
agentive') or as a pseudo-intransitive sentence with object-deletion
(e.g. 'He dried the dishes' or 'He dried the clothes'), but probably
not as implicitly reflexive, equivalent semantically to He dried himself
in the sun. Unlike object-deletion of the more general kind, implicit
reflexivity (which may of course be regarded as a special type of
object-deletion) would seem to be restricted in English to a relatively
small class of verbs. Moreover, it is not always clear whether a
particular sentence exemplifies implicit reflexivity or the more
general kind of object-deletion. For instance, change is not only
found as an 'ergative' verb (like move, grow, open, etc.) in corres-
ponding intransitive and transitive sentences, but also in sentences
like
(19) FII just slip upstairs and change before dinner
The interpretation that is intended here is the one suggested by the
insertion of my dress or into a more respectable suit between change and
before dinner, (A sentence like / have changed is of course ambiguous
as between the interpretation we are now considering and the
'ergative' interpretation: ' I have undergone some kind of physical,
intellectual, moral, etc. transformation.') The question is whether the
second clause of (19) is implicitly reflexive. Notice that the following
three sentences are acceptable:
(20) I'll change the baby
(21) Til change the baby's dress
(22) Til change my dress
but not
(23) *Vll change myself
(i.e. explicitly reflexive, in the intended sense). The semantic
parallelism between (20) and (21), on the one hand, and (22) and
I'll change, on the other, might be taken as an indication that I'll
change is implicitly reflexiveautomatically transformed from Til
change myself by deletion of the overtly reflexive object. Alter-
natively, I'll change might be regarded as an instance of object-
deletion of the more general type, related to sentences like (22) as
(15) is related to (16) above (8.2.9). The question would seem to be
syntactically undecidable; and semantically it makes no difference
(by contrast with the two relevant interpretations of a sentence like
I'll wash before lunch).
8.2.11 * Pseudo-intransitive'
So far we have discussed three classes of verbs in English which may
be used (with 'zero modification') in both transitive and intransitive
(or pseudo-intransitive) constructions. The important syntactic
characteristics of verbs belonging to these three classes (exemplified
by move, eat and shave)together with the characteristics of three
other classes, which will be mentioned beloware shown in Table 12.
It will be observed that classes 2 and 3 {eat and shave) differ from
class 1 {move) in a number of ways. In the right-hand side of the table,
Table 12. Some English verbs
Transitive (Pseudo-)Intransitive
A. A
1
r
Subject Verb Object Subject Verb (Object)
I. A: +ag move B: -ag B: +ag move
2. A: +ag eat B: -ag A: +ag eat (4= A)
3- A: H-ag shave B: -ag A: +ag shave ( = A)
4- ^ : +ag walk B: + ag B: +ag walk
5- A: +ag build B: -ag B: -ag build
6. ^ : +ag sell B: -ag B: -ag sell
they are both shown with an entry under ('Object)' and they have
Ay rather than B, in the subject-position. This defines classes 2 and 3
as pseudo-intransitive verbs, since A indicates the subject of the
corresponding transitive construction and the parentheses in the third
column of the right-hand side mark the 'deletion* of the object of the
transitive. The difference between 3 and 2and we have seen that
this cannot always be drawn in particular instanceslies in the
identity (= A) or non-identity (+ A) between the 'deleted* object
and the subject of both the transitive and pseudo-transitive con-
structions. By contrast, class 1 is shown as having the 'ergative'
characteristics discussed above (8.2.3). The distinction between
' + a g ' ('agentive', or 'plus-agentive') and 'ag' ('non-agentive',
or 'minus-agentive') is not crucial to the distinction of 'ergative'
verbs from verbs with 'deletable' objects. Although move, as a one-
place verb, may take either an 'agentive' or 'non-agentive' subject,
provided that the nominal in question is animate (move would there-
fore be shown in the lexicon with the classification ' ag', 'plus-or-
minus agentive': cf. 4.3.3, 8.2.6), it is here assumed that a transitive
sentence like John moved Bill (e.g. to another desk in the office) is more
directly related to the 'non-agentive' interpretation of Bill moved
than to the 'agentive' interpretation. It would seem that the more
usual realization of John: +ag, moved. Bill: +ag (with John as the
'causative' subject and the 'agentive' Bill moved as its 'predicate':
we will come back to the formalization of this notion) is John made
Bill move or John got Bill to move or John had Bill move.
8.2.12 Agentive objects

However, there is at least one class of verbs in English which may be
used with an * agentive' object (cf. class 4 of Table 12). It is exemplified
by walk in the following sentence
(24) John walked the horse
which is in other respects related to
(25) The horse walked
as John moved Bill is to Bill moved. Other members of this highly-
restricted class are gallop, runy jump, etc. One difference between (24)
and the more common type of ' double-agentive' sentence John made
the horse walk would seem to be that (24) implies that John is the
'direct' agent (that he himself led the horse, or rode it). John made
the horse walk does not carry this implication. But the distinction
between a * direct' and an 'indirect' agent cannot be made in this way
with most English verbs. Not only is John made Bill move (like John
made the horse walk) neutral with respect to the distinction of 'direct'
and ' indirect' agent (as applied to John), but so also is John moved Bill.
The most common way of expressing 'indirect' agency in English is
exemplified by John had a house built, which is ' marked' as' indirectly'
agentive. Not only are the limits on the use of the construction
illustrated in (24) unclear: it is also uncertain, in particular instances,
whether the relationship between a pair of corresponding transitive
and intransitive sentences is that exemplified in (24) and (25) or the
more common' ergative' relationship holding between John moved Bill
and Bill moved. One might well hesitate, for example, in respect of
the following two sentences:
(26) The aeroplane flew from London to Paris
(27) The pilot flew the aeroplane from London to Paris
The bird flew through the air has an 'agentive' subject ('What did the
bird do?') and The stone flew through the air has a ' non-agentive'
subject ('What happened to the stone?'), but (26) would seem to be
somewhat indeterminate in this respect. In clear cases, however, the
differences between class 1 and class 4 can be referred to the difference
between 'non-agentive' and 'agentive' nominals as the subject of the
intransitive or the object of the transitive. Apart from this difference,
class 4 verbs are 'ergative'.
8.2.13 Other pseudo-intransitive constructions

Classes 5 and 6 of Table 12 are also rather restricted in their employ-
ment. The constructions in question are illustrated in the following
two sentences:
(28) The house is building
(29) Detergents sell well
Both of these sentences are 'non-agentive', and relatable to transitive
sentences of the most common type {They are building the house, They
sell detergents). But they differ from intransitive 'ergative' sentences
(e.g. The house is moving, Grass grows well, etc.) in that they 'pre-
suppose ' an agent. In this respect, they are similar to passive sentences
in English. In fact, the more normal 'version' of (28) in contemporary
English is probably The house is being built. But (29) exemplifies a
much more common type of construction, and one that cannot be
converted into an equivalent passive 'version' in such a straight-
forward fashion.
According to Halliday: 'This type is especially frequent in simple
present tense (this material washes), particularly in negative potential,
where however the form is don't/won't and not can't: this material
doesn't/won't wash; it is not however restricted to these verbal forms,
and may in fact occur with any tense, especially with certain ~ly
adverbs, as in the clothes washed easily, these books are not going to sell
quickly. Underlying all these is a feature of characterization of the
process as such, either a qualification of it or a generalization about its
feasibility; so that we may call the type "process-oriented" in
contradistinction to the "agent-oriented" type the clothes were
washed.'
This difference between what Halliday has aptly distinguished as
'process-oriented* and 'agent-oriented' sentences can be brought
out by comparing the following three examples:
(30) The books sold quickly
(31) The books were sold quickly
(32) They sold the books quickly
Of these: (30) is definitely 'process-oriented': it would seem to

imply that it was by virtue of some quality of the books that they
were quickly disposed of. ('The books sold themselves', as it were;
and it is noticeable that in certain languages, e.g. French, the equi-
valent of (30) could take the form of a reflexive.) By contrast, (31)
might be described as * agent-oriented': it certainly admits of the
interpretation that it was the (unspecified) agent who was responsible
for the rapidity with which the books were sold. It is, however, (32)
the active sentencewhich carries this implication in a more definite
form. One should perhaps say that sentences like (30) are 'marked'
as 'process-oriented', whereas the passive is 'unmarked* for this
distinction. There are many instances of paired 'process-oriented*
and passive sentences (e.g. She doesn't frighten easily: She isn't easily
frightened) which would seem to be semantically equivalent. The
passive 'version* may occur with an 'agentive' or 'instrumental'
adjunct (cf. 7.4.5): e.g. She isn't easily frightened by burglarsjthreats.
But here also one would tend to the 'process-oriented' interpretation.
And the reason seems to be that the non-past tense in English
(without any 'marking' of aspect) is interpretable as 'habitual'
(cf. 7.5.7): the indefiniteness of the adjunct (burglars, threats)
supports this interpretation in the present instance. By contrast,
She wasn't easily frightened by the burglarsjthreats is more readily
taken as 'agent-oriented'.
The analysis of sentences like (30), containing verbs of class 6, is
somewhat problematical. Although the limits imposed on 'process-
oriented ' sentences are far from clear (and many ' sentences' of this
kind are definitely unacceptable: * The first edition exhausted in three
days, * Caviare never eats at five o'clock, etc.; or doubtful: }*The
music heard well at the back of the hall, etc.), they are nevertheless of
very frequent occurrence in English. Within the general framework
established in this section, they are pseudo-intransitive: But they
differ from verbs of classes 2 and 3, in that their subject in the pseudo-
intransitive usage is the same as their object in the transitive con-
struction. In this respect they are like passive forms of the verb
(cf. 8.3.3).
All pseudo-intransitive sentences (classes 2, 3, 5 and 6 of Table 12)
are to be transformationally derived from transitives. On the other
hand, as our earlier discussion of causative constructions suggests,
true intransitives (cf. classes 1 and 4 of Table 12) may be taken as the
'source' for the generation of transitive sentences. The order of
derivation is therefore intransitivetransitivepseudo-intransitive.
Before taking up the formal consequences of this proposal, we must
briefly consider three-place constructions: i.e. sentence-nuclei with
a subject, a direct object and an indirect object (cf. 8.2.1).
8.2.14 Three-place constructions

The general point to be made is that, as two-place constructions can
be derived from one-place constructions by means of the notion of
'causativity', so three-place constructions can be derived from two-
place constructions by means of a further application of the same
notion. As John moves the stone is to be related, syntactically, to The
stone moves, so (33) is to be related to (34):
(33) John gives the book to Mary,
(34) Mary has the book
For the present, the book may be regarded as the object of have in the
'possessive* sentence (34). The argument of this section is unaffected
by the fact that (34) is not a transitive sentence, as we shall see later
(cf. 8.4.4).
It will be more illuminating to discuss three-place verbs with
reference to a language that has a productive causative construction
applicable both to transitive and intransitive nuclei: and we will use
French for this purpose. The two sentences cited in the previous
paragraph, (33) and (34), may be translated into French as follows
(35) Jean donne le livre a Marie
(36) Marie a le livre
The French causative construction, which has been illustrated above
in relation to the derivation of transitive sentences from intransitive
nuclei (cf. 8.2.4), m a v z^so ^ e applied to transitive nuclei for the
generation of three-place constructions. This is exemplified by
(37) Jean fait manger les pommes aux enfants
in relation to
(38) Les enfants mangent les pommes
(The translation of (37) is 'John makes the children eat the apples';
and of (38) 'The children eat the apples'. The grammatical structure
of (37) may be indicated by means of the quasi-English gloss 'John
makes-eat the apples to the children'.) The structural parallelism
between (35) and (37) is obvious. As le livre is the direct object and
a Marie the indirect object of the 'simple' three-place verb donne
in (35), so les pommes may be regarded as the direct object and aux
enfants the indirect object of the 'complex' three-place verb fait
manger in (37). This parallelism is confirmed by the fact that one may
frequently choose between a 'simple' and a 'complex' three-place
verb in French. For example, the following two sentences are semanti-
cally equivalent as translations of 'John shows the book to Mary*
(39) Jean montre le livre a Marie
(40) Jean fait voir le livre a Marie
(The second of these may be glossed in quasi-English as 'John
makes-see the book to Mary'.) This would suggest that, as kill is the
' lexicalized' two-place causative form of die (cf. 8.2.4), s o montrer
S S S
B: ag Verbx A: +ag
Stage 1 >-
B: ag Yerb2 B: ag
Fig. 19. Transitivity and causativity.
('to show') is the 'lexicalized' three-place causative of voir (*to see').

Similarly, one may frequently choose in French between a 'simple'
and a 'complex' two-place verb: e.g. Jean appelle le docteur ('John
calls the doctor'): Jean fait venir le docteur ('John makes-come the
doctor'). The difference between donner, 'make-have' = 'give', in
(35) and montrer, 'make-see' = 'show', in (39) lies, therefore, in the
difference between obligatory and optional 'lexicalization' in three-
place constructions.
It is a characteristic of the causative construction in French (as also
in certain other languages, including Turkish and Hindi) that the
subject of the underlying two-place nucleus is transformed into an
indirect object with three-place causative verbs, whether 'simple'
(i.e. 'lexicalized') or 'complex' (with the auxiliary faire) whereas the
object of the underlying two-place nucleus remains as the direct
object in the three-place causative construction. The full three-
stage 'cycle' is illustrated in Fig. 19, where B is the subject of the
underlying intransitive nucleus (at stage i), A is the subject of the two-
place transitive nucleus (at stage 2), and C is the subject of the three-place
causative construction (at what, for the moment, we may regard as
the final stage of the * cycle'). The subscript on the symbol ' Verb' at
the right-hand terminal node of each tree corresponds with the
number of nominals which 'fill' the places associated with the verb.
(We will introduce labels for the intermediate nodes presently.) It
will be observed that the figure marks B as * plus-or-minus-agentive'
in all positions (and thus allows for both class 1 and class 4 transitive
sentences of English: cf. 8.2.11). The marking of A as ' plus-agentive'
at stage 3 is more questionable, since it implies that all indirect
objects are 'agentive'. This is certainly not so. But so far we are
treating sentences like iA have By (e.g. Mary has the book) as transi-
tive, with A as the subject and B as the object; and all transitive
sentences have an 'agentive' subject in the system developed above.
8.2.15 Syntactically ambiguous constructions

A further point should be made about the causative construction in
French. Sentences of the following type are ambiguous:
(41) Jean fait manger les enfants.
Although in normal circumstances (41) will be interpreted as 'John
gets the children to eat (their dinner, etc.)', it can also mean 'John
gets the children eaten'. The ambiguity results from the fact that
manger ('to eat') admits of object-deletion (cf. 8.2.9). Underlying (41)
in its more normal interpretation, is the pseudo-intransitive sentence
Les enfants mangent ('The children eat'): the fact that les enfants in
(41) has the form of a direct, and not an indirect, object would
suggest that pseudo-intransitives behave like true intransitives under
the operation of the causative transformation. The rule for the
deletion of the object must therefore operate before the causative
transformation that results in (41). Furthermore, (41), in the inter-
pretation 'John gets the children to eat', cannot be accounted for as
a stage 3 construction derived from a stage 2 nucleus, but rather as a
stage 2 construction which has been derived from another stage 2
construction ('reinterpreted' by the rules of the grammar as a stage 1
nucleus after object-deletion). By contrast, in the interpretation
'John gets the children eaten (e.g. by the lion at the zoo)', (41) is a
stage 3 construction with deletion (or non-realization) of the node
labelled A in Fig. 19. The ambiguity of (41) is somewhat similar
therefore to the ambiguity of such English phrases as the shooting of
the hunters (one of the examples of syntactic ambiguity that has been
discussed by Chomsky). The verb shoot is transitive in X shoots the
hunters (where X stands for the unspecified actor), but pseudo-
intransitive in The hunters shoot (cf. eat, 8.2.9). Since the general
principle for the interpretation of English noun-phrases of the form
the V-ing of NP would seem to be that they are * subjective' if the
verb in question (V) is intransitive, but 'objective* if the verb is
transitive, it follows that pseudo-intransitives operate like true
intransitives in the rule which generates phrases like the shooting of
the hunters from the underlying nucleus of The hunters shoot. It may
be noted in passing that the ambiguity found in (41) is also present
in comparable Turkish sentences (cf. Ahmet focuk-lar-i ye-dir-di>
* Ahmet the-children made-eat*); and the ambiguity found in the
English phrase the shooting of the hunters also occurs, apparently, in
such simple Eskimo sentences as agna-q takubaa (' Someone sees the
woman' or 'The woman sees someone').
It is interesting to note that all these instances of ambiguity
(between an 'objective* and a 'subjective* interpretation) depend
upon the fact that in the constructions in question the 'agentive'
nominal is structurally optional (i.e. an 'adjunct': cf. 8.1.1). The
'objective' interpretation is determined by the occurrence of au lion
in (41)Jean fait manger les enfants au liony 'John gets the lion to eat
the children' or 'John gets the children eaten by the lion'; by the
occurrence of by the gamekeeper with the shooting of the hunters \ by
the occurrence of qimmi-p in the Eskimo sentence (qimmi-p agna-q
takubaay 'The dog sees the woman': cf. 8.1.6). Since the optionality
(or absence) of the 'agentive' is one of the principal characteristics of
passive sentencescf. The hunters were shot (by the gamekeeper)^ The
children were eaten (by the lion)f The woman was seen (by the dog)
there may be some justification for the view that all the ambiguous
constructions mentioned in the previous paragraph are neutral as
between an active ('subjective') and passive ('objective') interpreta-
tion. We must now take up the discussion of the distinction between
active and passive sentences in terms of the traditional category of
'voice*.
8.3 Voice
9
8.3.1 The term 'voice
The term 'voice' (Latin vox) was originally used by Roman gram-
marians in two distinguishable, but related, senses: (i) In the sense of
* sound* (as used in the ' pronunciation' of human language: trans-
lating the Greek term phone)> especially of the ' sounds' produced by
the vibration of the 'vocal cords': hence the term 'vowel' (from
Latin sonus vocalis, 'a sound produced with voice', via Old French
vouel). (ii) Of the 'form' of a word (that is, what it 'sounded' like)
as opposed to its 'meaning' (cf. 7.1.2). The first of these two senses
is still current in linguistics in the distinction between voiced and
voiceless 'sounds' (whether as phonetic or phonological units:
cf. 3.2.4). In the second sense, 'voice' has disappeared from modern
linguistic theory. Instead, the term has developed a third sense,
derived ultimately from (ii) above, in which it refers to the active and
passive 'forms' of the verb. (The traditional Latin term for this third
sense was species or genus. In the course of time, genus was restricted
to the nominal category of 'gender'; and the somewhat artificial
classification of the 'forms' of the different parts of speech in terms
of genera and species was abandoned.) The traditional Greek term
for ' voice' as a category of the verb was diathesis,' state',' disposition',
'function', etc.; and some linguists prefer to use 'diathesis', rather
than 'voice', in this sense of the term. However, the risk of confusion
between the phonetic or phonological sense of 'voice' and its
grammatical sense is very small. In this section we are of course
concerned with voice as a grammatical category.
It is not only the traditional terminology that is confusing, or
potentially confusing, in theoretical discussions of the category of
voice. The Greek grammarians failed to appreciate the true nature of
the distinctions marked by the verbal inflexions which they referred
to as 'active', 'passive' (i.e. signifying the state of 'being acted upon'
or 'suffering the effects of the action') and 'middle'; and this has left
us with a legacy of contradictory statements about the role of voice,
not only in the classical languages, but also in English and other
modern languages, the description of which has been strongly
influenced by traditional grammar. As far as English is concerned,
one finds such statements as the following (by McKerrow): ' If we
were now starting for the first time to construct a grammar of modern
English, without knowledge of or reference to the classics, it might
never occur to us to postulate a passive voice at all. It seems to me
that it is questionable whether in spoken English of to-day there is
really any such thing, and though, as a matter of convenience, it may
be well to retain it in our grammars, I doubt whether it ought to
occupy quite so prominent a position as it sometimes does.' (This
passage, together with statements made about the passive in English
8.3. VOICE 373
by other grammarians, is quoted by Svartvik in a recent compre-
hensive and illuminating study of the whole question.) Paradoxically,
it might also be maintained that what the Greek grammarians said
about the passive voice, and its opposition to the active, is more
directly applicable to modern English than it is to classical Greek!
Although this point will not be developed here in detail, its simple
assertion will serve as an indication that the interpretation of voice
is a matter of considerable controversy, both at the present time and
in the Western grammatical tradition.
8.3.2 'Active' and 'middle' in Greek

Reference was made in the previous paragraph to the three Greek
voices: 'active', 'passive* and 'middle*. As the term suggests, the
middle was thought of as intermediate between the primary opposition
of active and passive (signifying either an 'action', like the active, or
a 'state', like the passive, according to the circumstances or the
inherent meaning of the verb in question). In fact, the passive is not
distinguished from the middle in most of the inflected forms of the
Greek verb; distinct passive inflexions are found only in the future
and the aorist (cf. 7.5.6); of these the former did not develop before
the classical period, and the latter, with certain verbs at least, could
also have a 'middle' sense (what is meant by this we shall come to
presently); finally, the verbal forms that could be used either in
'middle' or passive sentences are far more frequently to be inter-
preted as middle than as passive. In short, the opposition of voice in
Greek is primarily one of active v. 'middle'. The passive was a later
development (as it was in all the Indo-European languages); and it
was at first relatively infrequent.
The implications of the middle (when it is in opposition with the
active) are that the 'action' or 'state' affects the subject of the verb
or his interests. One class of sentences that fall within the scope of
this notion are reflexive sentences (cf. 8.2.10). For example, the
English sentence / am washing myself (or its implicitly reflexive
'version', / am washing) might be translated into Greek as louomai
(v. the active, loud, 'I am washing (something)'). But the middle may
also be used in a transitive sentence with an object that is distinct
from the subject: e.g. louomai khitona, 'I am washing (my) shirt'.
Here, the implication of the use of the middle voice rather than the
active is that the' action' is being carried out by the subject for his own
benefit or in his own interests. It is interesting to note, at this point,
13 LIT
that French has a rather similar construction: cf. not only Je me lave,
' I am washing myself, but also Je me lave une chemise, ' I am washing
(myself) a shirt*. The reflexive implication in sentences like this
might be described as *benefactive' ('for the benefit of', 'in the
interests of): cf. I got myself a new suit. In other instances the sense
of the Greek middle can be best conveyed in English by means of
what may be described, rather clumsily, as a 'causative-reflexive':
cf. mistho, 'I hire' v. misthoumai, 'I get myself hired' (i.e. 'I take
a job', 'I sign on as a mercenary soldier', etc.); diddsko, 'I teach' v.
diddskomai, 'I get myself taught'.
Although one can distinguish other senses of the Greek middle (and
in a fuller treatment various factors would come up for discussion),
what has been said will suffice to illustrate the general character of the
'middle' voice in sentences in which it is opposed to the active. From
the glosses that have been attached to some of the examples it will be
clear that the subject of the 'middle' can be interpreted as 'non-
agentive' or 'agentive', according to the context or the meaning of
the verb; and, if the subject is taken as 'non-agentive', it can also be
identified in certain instances with the object of a corresponding
transitive sentence in the active voice. Under these conditions, the
distinction between the middle and the passive is 'neutralized'.
This point may be illustrated, in fact, from English (where a
number of constructions with the auxiliaries have and get manifest the
same kind of ambivalence). Consider a sentence like
(42) / am getting shaved
in contrast with each of the following two sentences
(43) / am shaving
(44) / am being shaved
If the subject of (42) is taken as 'non-agentive', but as otherwise
'interested' in the 'action', (42) is semantically equivalent to (44)
a passive sentence; if there is an 'agentive' adjunct in the sentence
(e.g. / am getting shaved by the barber), then of course only the passive
interpretation is possible (cf. 8.2.15, on the shooting of the hunters, etc.).
If, however, the subject of (42) is understood as 'agentive', the
sentence is more or less equivalent to (43)a pseudo-intransitive
sentence.
8.3. VOICE 375
8.3.3 The passive

The point being made is that the Middle' voice can 'merge' with
the passive under the 'non-agentive' interpretation of the subject;
and this seems to have been the point de depart for the subsequent
development of passive constructions in the Indo-European languages.
Grammars of Latin recognize only two voices, ' active' and ' passive';
but there are many functions of the Latin * passive* which are closer
to those of the Greek 'middle* than they are to those of the passive
voice in its traditional interpretation. Particularly interesting is the
opposition between the active and the 'passive* with verbs like
'move* and 'turn*. We have already seen that the English verb move
is ' ergative': B moves v. A moves B (cf. 8.2.3). I n corresponding Latin
sentences, we find B movetur ('passive') v. A movet B (active, transi-
tive). But B movetur can be translated in two ways: (i) as ' B moves'
(with B either 'agentive' or 'non-agentive'); or (ii) as 'B is moved'
(with an 'agentive', other than B> implied). Only the second of these
interpretations is passive, rather than 'middle'. The first can be
glossed as 'there is movement, and B is affected (whether B is the
cause, or agent, of the movement or not)'.
We have already noted that the 'middle' could be used in Greek
(with certain verbs) in what we would translate into English as
explicitly or implicitly reflexive sentences. Conversely, in many
modern European languages (French, German, Italian, Spanish, etc.)
the reflexive construction is employed in sentences in which one finds
the ' middle' voice in Greek, the ' passive' in Latin, and an ' ergative'
intransitive or a pseudo-intransitive in English (cf. move, shave, sell
in Table 12, p. 364). Furthermore, the Russian passive (except for the
past perfective) has developed from, and to a certain degree is still
'merged' with, the reflexive construction; and in Spanish also (more
strikingly than in the other Romance languages) the reflexive can be
employed in place of the passive in a wide range of sentences.
Limitations of space prevent us from going further into the complex
interrelationships which exist between the 'middle', the reflexive,
the passive, and various kinds of intransitive and pseudo-intransitive
constructions in particular languages. At this point, we must discuss
what is meant by ' passive', an understanding of which we have so far
taken for granted.
Of the following four sentences, it would generally be agreed (in
accordance with traditional views of the structure of English: we have
already seen that certain grammarians have expressed doubts about
13-2
this point, cf. 8.3.1) that (45a) and (46^) are active and (45b) and
(466) are passive; furthermore, that there is some sense in which active
sentences are more 'basic', so that (456) may be regarded as the
passive 'version* of (45a), and (466) the passive 'version* of (46a):
(45 a) John killed Bill
(45 b) Bill was killed by John
(46 a) Bill killed John
(466) John was killed by Bill
This relationship between corresponding active and passive sentences
was traditionally accounted for in some such terms as these:
(i) The object of the active sentence becomes the subject of the
corresponding passive sentence. Thus, Bill is the object of (450) and
the subject of (45 b);John is the object of (46a) and the subject of (466).
(ii) The verb is 'active' in 'form* in the more basic (active)
'version', and 'passive' in 'form' in the less basic (passive) 'version'.
Thus, killed ('active') v. was killed ('passive').
(iii) The subject of the active sentence is not necessarily' expressed'
(overtly represented) in the passive 'version' of the 'same' sentence;
but, if it is 'expressed', it takes the form of an adjunct marked as
'agentive' by means of case-inflexion or by the use of a particular
preposition (cf. 7.4.7). Thus, by John and by Bill in (456) and (466).
The 'agentless' passive sentences corresponding to (456) and (466)
are Bill was killed and John was killed. We have seen that the relation-
ship between corresponding active and passive sentences in English
can be formalized in terms of transformational rules operating upon
the same underlying 'kernel' string (or kernel structure): cf. 6.6.4.
It will be observed, however, that the terms 'active' and 'passive'
were used in two different senses in the traditional formulation of the
three conditions listed in the previous paragraph. In (i) and (iii)
they were applied to sentences, whereas in (ii) they were applied to
the 'forms' of the verb (and, for that reason, put in quotation marks).
The point is that, although the three conditions will coincide in the
clearest instances of the distinction between corresponding active and
passive sentences, they are to some degree independent of one another.
Intransitive 'ergative' sentences (e.g. The stone moved) and various
kinds of pseudo-intransitive sentences (e.g. The books sold quickly,
The house was building, etc.) might be held to satisfy the first of the
three conditions (cf. John moved the stone, John sold the books quickly,
John was building the house, etc.), but not the second (the verb is
8.3. VOICE 377
'active* in 'form': moved, sold, was building, etc., rather than was
moved, were sold, was being built, etc.) or the third (there is no possi-
bility of 'expressing' the subject of the corresponding transitive
'version* by means of the most typical 'agentive' adjunct-construc-
tion : * The stone moved by John, * The books sold quickly by John, etc.).
The possibility of conflict between 'the form of the passive voice*
and its 'function' (the conversion, or transformation, of a sentence in
which the subject is represented as 'acting* into a sentence in which
it is represented as 'being acted upon', or 'suffering the effects of
the action') is well recognized in the theory and practice of traditional
grammarians. In Latin and Greek there are many verbs which are
said to be 'passive (or middle) in form, but active in meaning' (they
are traditionally referred to as ' deponent', since they are thought of as
'laying aside' the 'meaning' normally associated with the 'passive'
voice). An example is the Latin verb sequor,' I follow', which occurs in
active, transitive sentences despite its 'form'. Conversely, though
more rarely, a verb which is 'active' in 'form' may occur in the
corresponding passive ' version' of an active sentence. Consider, for
example, the following two Greek sentences, of which (47 b) is the
normal passive 'version' of (470):
(47a) hoi Athenaioi apokteinousi Sokrdten,' The Athenians kill
Socrates'
(47b) Sokrdtes apothneiskei (hupo ton Athenaion), 'Socrates is
killed (by the Athenians)'
The verb that is translated in (476) as 'is killed' may also be trans-
lated as 'dies' in other contexts; and it is 'active' in 'form'. It will
be observed that, unlike the intransitive 'ergative' sentences and
pseudo-intransitive sentences of English referred to in the previous
paragraph, (476) satisfies both (i) and (iii), but not (ii), of the condi-
tions listed above.
One cannot talk sensibly of a 'conflict' between 'form* and
'function' in particular instances unless there are other instances,
more numerous and taken as typical, in which there is no 'conflict'.
The traditional conception of the distinction between an active and a
passive voice rests upon the assumption that the three conditions
mentioned above will generally coincide. For example, if it were the
case in English that the equivalents of (456) and (466) were
(45 c) *Bill killed by John
(46 c) * John killed by Bill
with by John and by Bill as optional 'agentive' adjuncts; and, if this'
were not merely a feature associated with kill and a relatively small
number of other transitive verbs, but typical of the relationship
between pairs of sentences in which the object of the one 'becomes'
the subject of the other, one would not say that English had an active
and passive voice, but rather that the English verb was 'voice-
neutraF and could occur with either the * actor* or the 'goal' as its
subject. This situation obtains in a number of languages (including
Eskimo: cf. 8.2.15).
8.3.4 'Agentless* sentences

In the statement of condition (iii), it was said that the ('agentive')
subject of the active sentence is not necessarily 'expressed', or overtly
represented, in the passive version. English is in fact rather unusual,
among languages that have a passive voice, in that the 'agentive'
adjunct occurs quite freely. But even in English, sentences like Bill
was killed by John are of less frequent occurrence than sentences like
Bill was killed. If there is any one function that is common to the
passive in all the languages that are customarily said to have a passive
voice (and in certain languages this seems to be its sole function:
e.g. in Turkish), this is that it makes possible the construction of
'agentless' sentences: e.g. Bill was killed.
But there is another way of constructing 'agentless' sentences in
English (apart from the pseudo-intransitive constructions discussed
earlier). This is by means of a 'non-specific' subject in the position
that would be filled by the word or phrase referring to the ' actor' in
a corresponding sentence in which the ' actor' was specified: cf. John
killed Bill v. Someone killed Bill; and Bill was killed by John v. Bill
was killed (by someone). There are therefore five possibilities: active
with a 'specific' agent; active with a 'non-specific' agent; passive
with a ' specific' agent; passive with a ' non-specific' agent (Bill was
killed by someone); and passive without mention of the agent (Bill was
killed). Of these, the passive with a 'non-specific' agent is unusual in
English: but there are circumstances in which it might be found.
There are other languages, however, in which only three or two of
these five possibilities are realized. For example, Turkish has the
structural equivalents of John killed Billy Someone killed Bill and Bill
was killed, but not of Bill was killed by John. (Although it is possible
to add to the passive sentence an adjunct which might be translated
into English, in certain circumstances, as 'agentive', it is not strictly
8.3. VOICE 379
comparable with by John in English sentences like the above.) And
there are languages in which there are no passive sentences, but
rather an opposition between a * specific' and a ' non-specific' active
sentence (cf. John killed Bill v. Someone killed Bill).
In particular languages, there may be a choice between different
kinds of 'non-specific* subjects in 'agentive' position. For example,
the Latin pronouns quidam and aliquis can both be translated as
'someone' in English; but the former might be used when the
* unspecified' person is known to the speaker ('a certain person that
need not be identified'), the latter when he is not ('someone whose
identity is unknown to me'). There is a somewhat similar distinction
in Russian between kto-to and kto-nibudj. And in many languages
there is what might be most appropriately described as a ' dummy-
subject' pronoun. Its function is quite distinct from that of the
'non-specific' pronouns (which are, in any case, not necessarily
restricted to subject position). This may be illustrated by means of
the French on or the German man, in such sentences as On parle
anglais or Man spricht englisch, 'English is spoken' (or, using the
English pronoun they with the same 'dummy' function, 'They speak
English': i.e. 'in this shop', 'in Jamaica', etc.). It is noticeable that
the most usual English translation of such sentences takes the form of
a passive. There are other languages, in which it tends to be reflexive
(or pseudo-reflexive): e.g. Spanish Se habla ingles.
Finally, one should mention what is traditionally described as the
'impersonal' use of the passive with intransitive verbs. So far we have
assumed that the passive voice is defined by the coincidence of the
three conditions listed in 8.3.3. But t n e fifSt f these restricts the
passive to transitive verbs. Although this restriction holds for
English and, on the whole, for most of the Indo-European languages,
the passive is regularly found with intransitive verbs in certain other
languages. For example, corresponding to the Turkish active sentence
(O) Istanbula bu yoldan gider ('He Istanbul-to this route-by go',
i.e. 'He takes this route for Istanbul') there is the 'impersonal'
passive 'version' Istanbula bu yoldan gid-il-ir ('Istanbul-to this
route-by is gone', i.e. 'This is the route for Istanbul' or 'One takes
this road for Istanbul': the same suffix, -*'/-, marks the passive voice
in Turkish, whether the verb is transitive or intransitive). A similar
construction is found, though less commonly, in German: Es wird
heute abendgetanzt ('It will be this evening danced', i.e. 'There will
be dancing this evening'). Occurrences are also attested in Latin
(pugnabatur, ' It was fought', i.e. 'There was some fighting'). Indeed,
it is probable that 'impersonal' forms of the verb were 'merged'
with the * middle' to form the 'passive' of classical Latin (cf. 8.3.3).
' Impersonal' constructions of the kind that have been illustrated
here clearly do not fulfil the conditions given above for the definition
of the passive. For these conditions rest upon the assumption that
passive sentences are derived from active transitive structures. On
the other hand, 'impersonal' sentences are similar to what we may
think of as 'true' passive sentences in two respects: first, they are
'agentless', and we have seen that the most typical function of the
passive is that it makes possible the construction of 'agentless'
sentences; second, they have a 'form' of the verb which is overtly
marked, inflexionally or otherwise, as 'passive'. In languages in
which they occur with 'true' passives they are presumably to be
related to these syntactically.
8.3.5 The formalization of transitivity and voice

The various phenomena that have been discussed in this section
under the rubric of 'transitivity', 'ergativity' and 'voice' present the
linguist with considerable problems, both practical and theoretical.
By ' practical' problems I mean the difficulties of deciding what the
' facts' are in the analysis of particular languages, what are the impli-
cations of the choice of one 'version', or one construction, rather
than another. These problems are formidable enough (although we
have frequently glossed over certain difficulties of interpretation in
our necessarily simplified discussion of the topics that have been
raised). But the theoretical problems are far greater.
In the field of syntactic theory at the present time one thing is
clear. Although great advances have been made in the understanding
and formalization of the relationships between sentences of different
types, none of the models of transformational syntax that have yet
been proposed (the best known of which is that developed by
Chomsky and his associates) is capable of handling the complex
interrelationships between the many kinds of constructions referred
to in this chapter in a systematic and intuitively satisfactory way
(cf. 6.5.7, on 'strong adequacy'). Linguistic theory develops of
course on the assumption that all the ' facts' are in principle formaliz-
able within a consistent general framework. And a good deal of work
is now being published which is directed towards the formalization
of such 'facts' as we have discussed here in an elementary and
informal manner.
8.3. VOICE 381
We shall make no attempt to summarize any of the alternative
technical proposals that have been made or to discuss their merits
and viability. Instead, we will conclude this section by drawing out
some of the theoretical implications of our earlier discussion of
' causativity' (cf. 8.2.4 ff.).
8.3.6 A tentative transformational account of transitives and causatives

Let us assume that the base-component of the grammar of English
contains, amongst other rules, the following:
(i) 2-> Norn + Pred
(The symbols 'Norn* and 'Pred' may be taken as abbreviations for

* Nominal* and * Predicate'.) Both of these rules are recursively
applicable an indefinitely large number of times (cf. 6.2.6).
If rule (i) is applied, followed by the option ' Pred -> Verb' in
rule (ii), the 'output* is the constituent-structure framework for the
nucleus of a simple intransitive sentence; and the noun or noun-
phrase that is inserted in the nominal position is, by definition, the
subject of the sentence (cf. 8.1.4). As we have seen (8.2.6), the subject
of an intransitive verb may be 'agentive' or 'non-agentive': the
nominal will therefore be given the feature 'plus agentive' (' + ag') or
'minus agentive' (' ag') to indicate this option. One-place (intransi-
tive) verbs must be classified in the lexicon according to their
potentiality of combination with nominals in terms of the following
threefold distinction: (a)' either agentive or non-agentive *; (b)' agen-
tive only'; (c) 'non-agentive only*. We will assume that the verbs
move Jumpy die exemplify this classification. Consequently, Bill moved
will be susceptible of interpretation as either *+ag' or < - a g > ;
whereas Bill died will be ' ag', and The horse jumped will be ' +ag'
(cf. 8.2.11). The underlying constituent-structure of the nuclei of these
sentences is represented diagrammatically in Fig. 20.
But the option ' Pred -> S ' may be taken in the first application of
rule (ii). In this case, the symbol ' 2 ' must be expanded again into
'Nom + Pred' by a further (recursive) application of rule (i). And the
conditions now hold for a second application of (ii). If the option
'Pred -> Verb' is taken at this point, the 'output' is the constituent-
structure framework for two-place nuclei,
-,82 8. GRAMMATICAL FUNCTIONS
Pred Nom Pred Nom Pred

[ag] [+ag]
Verb Verb Verb

Bill Bill Horse
cue move jump

(a) (0
Bill died /// moved 77zforcejumped
Fig. 20. Some one-place constructions.
Two alternatives present themselves, when we consider the

question of Tilling* this two-place structure. The first of these
alternatives is to think of the second application of rule (i) as equi-
valent to the development of the predicate constituent generated by
the first application of rule (i), with the nominal constituent generated
by the first application of rule (i) held constant, as it were. This would
have the effect of determining the subject of a two-place nucleus as
syntactically equivalent to the subject of a one-place nucleus. In other
words, under this interpretation of the recursive application of rules
(i) and (ii) the constituent-structure assigned to transitive sentences
(apart from the embedded nodes 'Z* and 'Pred') would be the same
as that assigned by the following two rules:
(a) S-> Nom + Pred
(Verb
(b) Pred
iNom + Verb
This is the approach to the generation of transitive constructions that
has been adopted in most transformational work so far (cf. 7.6.8). But
we have decided, in our discussion of transitivity and ergativity in
this section, that the sense in which the subject of a transitive con-
struction is syntactically equivalent to the subject of an intransitive
construction is largely a matter of such relatively superficial features
8.3. VOICE 383
as case and concord: and these can be handled by later transforma-
tional rules (cf. 8.2.7).
If we wish to capture the general principle noted above, that the
object of a two-place nucleus corresponds to the subject of a one-
place nucleus, we must take the second of the two alternatives. We
must embed the one-place nucleus as the predicate of the two-place
nucleus. This raises certain rather difficult, but probably not in-
superable, technical problems, which we will not go into here. We
will assume that they can be satisfactorily resolved, and consider their
further implications.
For simplicity, we will make use of a rather ad hoc device at this
point. When the one-place nucleus is embedded as the predicate of a
transitive construction, we will assume that a feature * causative* is
attached to the verb (marked as *+caus*). This feature will be
interpreted differently according to various conditions.
Let us first consider the situation in which the embedded nucleus
contains an 'agentive* nominal. There are two possibilities. The verb
that is selected may be marked in the lexicon as belonging to that
small class of English verbs (jump, walk, gallop, etc.: cf. 8.2.12)
which may have an 'agentive' object. In this case, the result is
what would normally be regarded as a transitive construction:
e.g. John jumped the horse. This implies that 'jump: -f caus* is
realized (on the basis of the information given with the one-place
verb jump in the lexicon) as jump: by a * process' of * zero-modifi-
cation* (cf. 8.2.8). But in the majority of instances, the co-occurrence
of 'Nom: + ag* and 'Verb: + caus* in the embedded nucleus will
'trigger off* the application of the rules for the productive causative
constructions with the auxiliary verbs, make, get, have, etc. The
eventual 'output* of the grammar will be such 'overtly* causative
constructions as John made Bill sing, John got Bill to come, etc.
If the embedded nucleus contains a ' non-agentive * nominal, it
fulfils the more normal conditions for transitivity. There are a
number of relevant factors. (1) The verb selected from the lexicon
may be 'ergative*: e.g. move (cf. 8.2.3). ^ n tn ^ s case> ft ls inserted in
the surface structure of the resulting sentence without morphological
modifications (with 'zero modification*). However, since the two-
place verb is an instance of 'Verb: +caus*, the resulting transitive
sentence will be determined by the grammar as the causative of the
corresponding intransitive ('non-agentive*). Thus, John moved Bill
will be given a syntactic analysis which makes it readily analysable,
semantically, as 'John was the agent (directly) responsible for BnTs
(non-agentive) movement*. (2) The verb selected may be one which
has associated with it a morphologically distinct two-place form:
e.g. soft (an * adjectival' verb, in this particular instance: cf. 8.2.8).
The entry in the lexicon for soft must therefore contain the informa-
tion which permits the realization of lsoft: + caus' as soften. (3) The
verb selected may be one which has an associated two-place 'lexi-
calized' form: e.g. die. In this case, the appropriate two-place form
is substituted for 'Verb: +caus\ For example, 'die: + caus' is
realized in surface structure as kill. (4) There may be no corres-
ponding morphological or lexical form, and the verb selected may
fail to satisfy the conditions for 'zero modification' (these conditions,
it may be observed, are far from clear: it is doubtful whether the
'ergative' verbs of English constitute a determinate and closed class).
In this case, 'Verb: + caus' once again 'triggers off' one of the
productive, ' overtly' causative constructions: John made Bill fall,
John made Bill responsible (for the administration of the department),
etc.
Some such approach to the generation of transitive sentences in
English (and other languages) would seem to be required, if we are
to succeed in formalizing the relationships which undoubtedly hold
between the various classes of sentences that have been illustrated.
(Sample underlying constituent-structure analyses are given in
Fig. 21.) At the same time, it is clear that there are many transitive
verbs which do not lend themselves very happily to analysis as
realizations of 'Verb: +caus\ They may be called 'basically transi-
tive' verbs. Examples are eat and read. Little purpose would be
served by treating eat or read as the two-place realizations of' dummy'
one-place verbs, meaning respectively 'be eaten' and 'be read'. The
'non-agentive' one-place sentences Caviare is eaten and The book is
read are not only passive in 'form' (cf. 8.3.3), t>ut> u ^ e BM moved
and unlike Bill moves ('non-agentive'), they presuppose an 'agent'
(which may or may not be specified, of course, in passive sentences).
Since it would be counter-intuitive, and of no use for semantic
analysis, to impose upon 'basically transitive' sentences the syntactic
treatment that has just been outlined for what we may call 'derived'
transitive sentences, one should perhaps allow a third option in
rule (ii) above. If this rule is now amended to
I Verb
Pred-> j Norn + Verb
1;
8.3. VOICE 385
it might be possible to secure the advantages of both alternatives. The
effect of this amendment, it will be observed, is to integrate the more
usual approach to the generation of transitive constructions (cf. rules
(a) and (b) on p. 382) with the proposal that we have made for the
derivation of two-place nuclei from embedded one-place nuclei.
However, this amendment, attractive as it is, increases the technical
Nom Pred Nom Pred Nom Pred

[+ag] 1 [+ag] [+ag]
1 1
1 S 1 I
John John John
Nom Verb Nom Verb Nom Verb

[+ag] [+caus] [-ag] [+caus] [-ag] [+caus"
i i i
i i i
i
i
i
i i
i
Bill sin* Bill move Bill die
(a) (b) (c)
John made Bill sing John moved Bill John killed Bill
Fig. 21. Some two-place constructions.
problems considerably. If these problems could be overcome, the

integration of the two approaches would have the advantage that it
formalizes the following propositions: ' Some transitive constructions
in the language are basic in the sense that they have both a deep-
structure subject and a deep-structure object. Others are derived;
and their object is a transformed, one-place, non-agentive subject.'
This proposition, or at least the second part of it, expresses a point
of view that has often been maintained by linguists working within
the more informal framework of traditional grammar. The first part
of the proposition represents the view of transitivity that has been
formalized by Chomsky in recent work. As we have seen, it is desir-
able that the theory of generative grammar should enable us to
formalize both points of view with respect to sentences containing
different classes of lexical items.
Little need be said now about three-place nuclei. If rules (i) and (ii)
are recursively applicable an indefinitely large number of times, they
will of course generate the constituent-structure framework for nuclei
with any number of nominal positions. In the case of three-place
nuclei, in which either a 'derived' or 'basically' transitive nucleus
has been embedded, we must allow for the transformation of the
subject of the two-place nucleus to the 'indirect object' position in
the three-place construction (cf. 8.2.14). But this 'demotion' of the
two-place subject to the status of adjunct (which is similar to its
'demotion' under the operation of the rules which derive passive
constructions from two-place actives: cf. 8.3.4) is obligatory only with
'lexicalized' three-place causatives: e.g. give, show. Let us assume
that Bill sees the book exemplifies a 'basically' transitive sentence. The
underlying nucleus, in which Bill is the subject and book is the object,
is embedded in a three-place causative sentence in which John is the
subject and *see: + caus' is realized as show: thus John shows the book
to Bill (cf. Fig. 22). However, there are relatively few 'lexicalized'
three-place causatives in the language. (There are far more if we
include such verbs as persuade, which might be regarded as the
' lexicalized' causative corresponding to the two-place believe with a
nominalized sentence as its object; cf. Bill believed that the earth was
flat: John persuaded Bill that the earth was flat. The consideration of
such constructions would introduce additional complications.)
In most instances, an ' overtly' causative construction is employed
at this third stage in the cycle. Suppose, for example, that the verb
selected from the lexicon is die, in the position which has been
assigned two causative-features by the operation of the embedding
rules: 'Verb: +caus, + caus'. There is no lexical or morphological
three-place form which corresponds to kill in the way that kill
corresponds to die. At the third stage, therefore, one of the causative
auxiliaries is introduced. An example of the type of sentence that
might be accounted for in this way is Tom made John kill Bill. It is
interesting to note that, with the causative auxiliaries get and have,
there is an alternative 'active' and 'passive' form of the embedded
two-place nucleus: cf. Tom had John kill Bill v. Tom had Bill killed
{by John). The 'passive' form is structurally similar to the sentences
which occur, typically, with ' lexicalized' three-place causatives, in the
sense that the subject of the two-place nucleus assumes the status of an
adjunct (cf. 8.2.15, on the French causative construction). Finally, it
8.3. VOICE 387
should be noted that the ' overtly' causative construction (in English,
as in many other languages) is indefinitely recursive: cf. Peter made
Harry make Frank... make Tom make John kill Bill. It was for this
reason that rules (i) and (ii) were made indefinitely recursive in
application.
2 S
Pred
book
Verb
see [-f caus]
Bill
[+caus]1
1
1
die
John shows the book to Bill Tom made John kill Bill
Fig. 22. Some three-place constructions.
The proposals that have been made in this section, as far as the
formalization and more technical details are concerned, must be
treated with reserve (cf. 8.3.5). At the same time, it may be pointed
out that there are two respects in which they are similar to other
proposals that are currently appearing in the linguistic journals and
in books on theoretical syntax. First, they presuppose a far more
'abstract' notion of the underlying constituent-structure than was
assumed in the earliest work in transformational grammar: in
particular, they make free use of such 'features' as ' -hag', ' +caus',
etc., which are not constituents in the sense that the auxiliary
element 'Pred' or the terminal element 'Verb' is a constituent, but
are assigned to such constituents for the correct operation of the
transformational rules or the rules of lexical realization. Second, the
proposals that have been made here presuppose or imply that both
the lexical and the syntactic structure of language are in part deter-
mined by such principles as * agency', 'causativity', 'state' (v.
'activity' or 'process'). In this latter respect, theoretical syntax has
to some extent gone back to the assumptions of traditional, ' notional'
grammar (cf. 4.1.3). It has done so, however, without surrendering
(in principle at least) any of the standards of rigour and precise
formulation that were first developed by linguists in conscious
opposition to traditional grammar.
8.4 Existential, locative and possessive constructions
8.4.1 'To be' and1 to have'

It has been suggested above that what is generally referred to as ' the
verb tobe\ in English and other languages, is a grammatical element,
devoid of meaning, which serves only to ' carry' the markers of tense,
mood and aspect in the surface structure of sentences (cf. 7.6.3, 8.2.10).
In this section, we will make the same point with respect to ' the verb
to have9. For simplicity, we will start by assuming that 'being' and
'having' ('possession') are well-defined and universal notions. It may
be pointed out immediately, however, that there are many languages
which have neither a 'verb to be' nor a 'verb to have\ There are
others in which there are many different 'copulas', no one of which
can be put into one-to-one correspondence with English be and have.
8.4.2 Logical analysis ofc the verb to be*

Nowadays, logicians tend to draw a sharp distinction between the
' existential' function of ' the verb to be' and its various ' predicative'
or 'copulative', functions. The philosophical importance of this
distinction lies in the fact that most modern philosophers would say
that existence cannot be predicated of objects in the same sense as
their various attributes, or properties, but is presupposed in the
identification of objects or in any reference to them. We are not
concerned here with the philosophical implications of the distinction
between ' existential' and ' predicative' statements. It is worth noting,
8 4 . LOCATIVE AND POSSESSIVE 389
y
however, that in English 'the verb to be is not used as freely in
existential sentences as it was in Greek: it is such sentences as God is
(where 'the verb to be' is perfectly normal in Greek) that have mainly
interested philosophers. What would generally be described as the
'existential' use of 'the verb to be* in English is not common except
with a locative or temporal complement (cf. 8.1.10). Examples of this
usage are: (i) There are lions in Africa, and (ii) The accident was
yesterday. We shall discuss such ' existential' sentences presently.
Among the 'predicative' uses of 'the verb to bey logicians custom-
arily distinguish: (a) the identification of one entity with another
(a = b: e.g. That man is John); (b) class-membership (bed e.g. John
is a Catholic, ' John is a member of the class of persons characterized
as Catholic'); and (c) class-inclusion (C c= D: Catholics are Christians,
' The members of the class of persons characterized as Catholic are
included among the members of the class of persons characterized as
Christians').
Though logically important, the distinction between class-
membership and class-inclusion does not appear to be of any syntactic
significance in most languages. The distinction between 'charac-
terizing* and 'sortal' sentences (which we have referred to above:
cf. 8.1.3) is, however, of considerable importance: cf. Apples are sweet
and Apples are fruit. The former tend to have an ' adjectival' predicate
(in languages where one can draw a distinction between 'adjectives'
and 'verbs'), and the latter a nominal predicate. The syntactic
analysis of sentences with nominal predicates (e.g. Apples are fruit,
John is a soldier, Mary is still a very young girl) is a very complex
matter: we shall not go into the question here. We will disregard the
differences between various subtypes of nominal predicates and treat
them all like 'adjectival' predicates. For terminological convenience,
we will refer to both classes of sentences (whether they are ' charac-
terizing' or 'sortal', and regardless of any other differences) as
attributive.
A further use of ' the verb to be' is with locative complements. We
have already discussed this construction in sufficient detail (cf. 8.1.10);
and we shall draw upon that discussion in what follows.
8.4.3 Existential and locative sentences

We have distinguished four 'functions' of 'the verb to be\ They are
'existential', 'identifying' (or 'equative'), 'attributive' and 'loca-
tive'. The last three of these are usually grouped together by linguists
as 'copulative' (the 'copula' being a 'link* between the subject and
the predicate in traditional logic); the 'existential' use of 'the verb
to be' is frequently treated as quite different from the other uses.
It has already been pointed out that 'the existential verb to be' is
rare in English without a locative or temporal complement. But in
using the term 'complement', rather than 'adjunct', we are perhaps
begging the question (cf. 8.1.9). The point is that the distinction can
hardly be drawn with respect to existential sentences. Consider the
following two sentences
(1) There are lions in Africa
(2) There is a book on the table
We might be inclined to say that the first is 'existential' and the
second 'locative', on the grounds that the first, but hardly the second,
can be paraphrased with a sentence containing exist: Lions exist in
Africa. And one might add that in Africa is syntactically ' detachable'
(and therefore an adjunct): Lions exist. On the other hand, there is
an obvious structural similarity between (1) and (2). Moreover, from
the point of view of their semantic analysis, existential sentences
might be described as implicitly locative (or temporal). The assertion
that something exists, or existed, requires ' complementation' with a
locative (or temporal) expression before it can be interpreted.
Whether or not this last point is accepted, it remains true that in
many languages, as in English, there are obvious similarities between
locative and existential sentences. There is little or no difference in
meaning between such sentences as Coffee will be here in a moment and
There will be coffee here in a moment: one might suspect that they have
the same deep-structure analysis.
Two further points should be noted. Existential sentences typically
have an indefinite, rather than a definite, subject: this fact raises the
possibility that they should be treated, in a syntactic analysis of their
deep structure, as indefinite locatives (with 'locative', in this context,
being understood to include' temporal': cf. 7.4.6, on the' local' cases).
Secondly, the connexion between existential and locative construc-
tions is supported by the employment of what was originally a locative
(and more particularly a deictic: cf. 7.2.1) adverb in the existential
sentences of a number of European languages: cf. English there (in
there isjare...), French y (in il y a)y Italian ci (in ci sono, etc.),
German da (in ist da, 'is there' or 'exists': cf. das Dasein, 'existence',
i.e. 'the being-there').
8-4- LOCATIVE AND POSSESSIVE 391
8.4.4 Possessive sentences

What is meant by the term * possessive sentence' may be illustrated,
in the first instance, from English:
(3) The book is John's
(4) John has a book
These two sentences are strikingly different in surface structure. In
(3) the subject is the book and the predicate is John's, a noun in the
'genitive*, or 'possessive', case (cf. 7.4.5). In (4) John is the subject;
and the predicate is has a book, with a book as the object of 'the verb
to have\ At the same time, it is clear that (3) and (4) are transforma-
tionally related to the phrase John's book ('the book which is John's*,
'the book which John has'), and therefore to one another. What, then,
is the nature of this relationship ?
In the treatment of the 'genitive' in the previous chapter, we drew
attention to the ' adjectival' function of the genitive in both adnominal
and predicative position: cf. The book is John's: John's book:: the book
is heavy: the heavy book. (It should be noted, in passing, that book in
John's book or the man's book is necessarily definite: 'the book of
John',' the book of the man', as it were. The corresponding indefinite
phrases are a book of John's, one of John's books, etc.) There is a
similar structural parallelism between locative and possessive con-
structions : cf. the book is John's: John's book:: the book is on the table:
the book on the table. (Whether the adnominal word or phrase, e.g.
John's or on the table, occurs before or after the noun it modifies in
surface structure is a secondary consideration.)
In most of the transformational accounts of English syntax so far
published, it has been assumed that phrases like John's book are to
be derived from an underlying structure in which the 'possessive'
noun is the subject of 'the verb have': in other words, it is assumed
that have is a deep-structure verb (like read, etc.), which differs,
however, from the majority of transitive verbs in that (in 'possessive'
sentences) it cannot undergo the passive transformation (*A book is
had by John). There are many reasons for believing that this account
of the relationship between ' /*az>-sentences' and possessive phrases
is incorrect.
First of all, it will be observed that the parallelism between
'adjectival', locative and possessive constructions referred to above
would tend to suggest that in all cases the head of the predicate is
transformed to adnominal position. What is required then is some
39^ 8. GRAMMATICAL FUNCTIONS
rule which derives John has a book from an underlying structure in
which a book is the subject and John is in a predicative relationship
with it. Of the two possessive sentences given above, (3) is more
similar in surface structure than (4) to what we may take to be the
deep structure of (4).
Secondly, it should be noted that relatively few languages exhibit
what we may call *have-sentences': i.e. possessive sentences in which
the 'possessor' is the surface-structure subject of a 'verb to have' and
the 'possessed object* the surface-structure object of this verb. Even
in the Indo-European languages, 'Aaz^-sentences' are of relatively
late and restricted development: they are not found in all the Slavonic
and Celtic languages. Moreover, in the Indo-European languages in
which '/iat^-sentences' occur (Latin, Greek, Germanic, etc.) they
would appear to have developed independently. In many cases (both
in the Indo-European languages and in other languages with a
'verb to have') the possessive use of the verb seems to have developed
from sentences in which it originally meant ' grasp * or ' hold (in the
hand)'.
As Benveniste has pointed out, Latin illustrates the development of
'have-sentences' in the Indo-European languages particularly well.
Consider the following three sentences:
(5) Johannes habet librum (' John-nominative' + ' has' + ' book-
accusative')
(6) Est Johanni liber ('is' + 'John-dative' + 'book-nomina-
tive')
(7) Liber est Johanni(' book-nominative' + ' is' + ' John-dative')
There is no ' definite article' in Latin, and definitcness is not marked
as such in any other way. The difference between (6) and (7) lies
solely in the order of words in surface structure; and the choice
between one order, rather than another, correlates in Latin (although
it is also influenced by other factors) with the selection of one word or
phrase as the 'topic', rather than the 'comment', of the sentence
(cf. 8.1.2). The 'topic' of the sentence is usually definite, whether it is
marked as such in particular languages or not. The normal trans-
lation of (7) is therefore ' The book is John's' (liber being the ' topic');
and the normal translation of (6) is 'John has a book' (not 'John has
the book'). The ' have-sentence' (5), which is a later development from
(6), is translatable as either 'John has a book' or 'John has the book'.
Let us now assume that the development of (5) from (6) is the result
of what was first an optional, 'stylistic', transformation, the effect of
8-4- LOCATIVE AND POSSESSIVE 393
which was to put * John' in surface-structure subject-position. This is
the position which animate nouns normally occupy (as the subject of
either the active or the passive); and animate nouns, rather than
inanimate nouns, tend to be 'topics'.
At first sight, this comparison with Latin and the diachronic
considerations introduced above would seem to have little relevance
to the synchronic description of modern English. Notice, however, that
(3) and (4) must be related to one another in the syntactic analysis of
English, just as (6) and (7) must be related to one another in Latin.
Since *A book is John's is to be excluded (we assume) as ungram-
matical, it is tempting to do so by means of an obligatory transforma-
tion which converts it (or rather the underlying structure) into John
has a book.
We can now relate this proposal to a somewhat similar proposal for
indefinite locative sentences. Although A book is on the table is
acceptable in English, it is a less common sentence than There is a
book on the table. Let us therefore say that what is sometimes called
the 'expletive* ('dummy') use of there in locative sentences is a
syntactic device of English for 'anticipating' the locative phrase in
surface structure. It is noteworthy that the 'expletive' there is not
generally found in locative sentences with a definite subject (e.g. The
book is on the table); and it also serves in ' existential' sentences, which
we have already related to indefinite locatives. Once we take into
account the distinction between definite and indefinite subjects, we
see that there is some parallelism between existential, locative and
possessive sentences in English. This parallelism is even more
striking in certain other languages.
8.4.5 Examples from other languages

Let us consider the following (Mandarin) Chinese sentences:
(8) Shu zdi zhuo-shdng ('book' + 'be' + 'table-top'), 'The
book is on the table'
(9) Zhuo-shdng you shu (' table-top' + ' have' + ' book'),' There
is a book on the table'
(10) Shu shi wo-de ('book'+'be'+ 'me-of'), 'The book is
mine'
(11) Woyoii shu ('me' + 'have' + 'book'), 'I have a book'
It will be observed that there are two 'verbs to be\ zdi and shi, and
one 'verb to have', you, exemplified here, But their distribution is
different from the distribution of the English be and have. (The
'postpositional' shdng is one of a set of such locative forms: in-
herently locative nouns, names of countries, etc., occur in sentences
like (8) and (9) without any such * localizing* form. Similarly,
'alienable' possession requires the particle -de: cf. (10). 'Inalienable'
possession does not: cf. wode shu v. wojid, 'my book* v. 'my home,
family*. For the notions 'alienable' and 'inalienable', cf. 7.4.6.)
Despite their surface-structure differences, the Chinese and the
English constructions can be related as follows. The English have-
transformation operates upon indefinite possessive structures to bring
the 'possessor' into surface-structure position; in Chinese, there is
a similar jyow-transformation, which operates, however, in both
possessive and locative sentences, (9) and (11). In the case of definite
possessive and locative sentences, the syntax of English generates the
'fo-copula'; in Chinese, shi is generated for the one, and zai for the
other. When we translate from one language into the other, we must
take note of the syntactic subclassification of the nouns occurring in
the sentences to be translated. We translate (11) as 'I have a book',
rather than 'There is a book at me', because wo is animate; con-
versely, we translate (9) by means of the existential-locative ' There is
a book on the table', because zhuo is inanimate. In certain gram-
matical descriptions of Chinese, a distinction is drawn between an
'existential' and 'possessive' sense of the 'verb' you: but this is
surely to distort the analysis of Chinese. It is worth noting that (9)
can also be translated into English with a' have-szntznct' in which the
locative is ' anticipated' in surface-subject position and then' repeated'
with a pronoun later in the sentence: The table has a book on it. And
the transformational relationship that holds in Chinese between such
sentences as (8) and (9) is found also in English, over a narrower range
of sentences, between sentences with a locative complement and
corresponding sentences in which the head-noun of the locative
phrase is the surface-subject of such verbs as contain: cf. There is
water in the bottle: The bottle contains water.
Although Russian has a 'verb to have' (imetj), it is not common. The
most usual translation of the four Chinese sentences given above
would be:
(8 a) Kniga na stole (' book' + ' on' + ' table')
(9a) Na stole kniga ('on'+'table'+'book')
(100) Kniga moja ('book' + 'my')
(11 a) U menja kniga (' at' + ' me' + ' book')
8.4- LOCATIVE AND POSSESSIVE 395
It will be observed that in these sentences (unmarked for tense, mood
and aspect) there is neither a 'verb to bey nor a 'verb to have\
U menja in (n a) is structurally similar to locative phrases; and the
preposition u with the 'genitive' (as here) would be translated as 'at'
or 'near' if the noun dependent upon it was inanimate. Although the
modally and temporally marked 'versions' of the above Russian
sentences (e.g. in the past and future 'tense': cf. 7.6.3) would have
the same 'verb to be' (bytj), the negative 'versions' of the possessives
and indefinite locatives fall together in that they have the 'existential'
negative (nety with the 'partitive' genitive). There are very many lan-
guages in which the 'possessive' construction corresponding to (na)
is structurally identical with a locative (e.g. Gaelic, Swahili, Hindi).
In Turkish, it is the possessives and the existential sentences
which are most obviously connected in terms of their surface-
structure similarities. The normal translation of (11 a) is
(nb) Kitab-'tm var ('book-my' + ' existent')
where var is the ' existential verb'. An alternative translation would be
(nc) Ben-de kitap var (' me-locative' 4- * book' -f' existent')
The difference between (11 b) and (11 c) is brought out by translating
the latter as ' I have a book with me' or ' I have a book on me'. Many
languages distinguish between an 'ordinary', or 'general', possessive
and a 'possessive of availability' in this way: and the second is
frequently locative, in terms of the case or preposition used.
Enough exemplification has been given to show that locatives,
possessives and existential sentences are interconnected in a variety
of languages. One has only to consider the range of sentences in
which have occurs in English as the 'main verb' to realize that the
whole category of so-called 'possessive' sentences is suspect. The
one thing that' have-stntznees' have in commonthe one thing that
there is in common in 'Atfw-sentences'is the fact that the 'subject'
of have is brought into a position of prominence in surface structure.
The syntax of these sentences is very complex, from a transforma-
tional point of view; but it seems quite clear that have is not a deep-
structure verb, any more than be is.
8.4.6 The English perfective

Most grammars of contemporary English distinguish between the
'full verb to have' (used in 'possessive' sentences) and the 'auxiliary
verb to have\ which is combined with 'the past participle' to form
the perfective aspect (cf. 7.5.7). Sentences like / have the book and
/ have done the work might seem to be quite distinct from one another
grammatically: but they are diachronically, and to some extent
synchronically, related. The diachronic development can be seen
most clearly with reference to the Latin ' possessive' and the Romance
perfect constructions with 'have*. (The use of 'have' as an 'auxiliary
verb * in the Germanic and Romance languages appears to have been
the result of independent development in each group of languages.
But it is accounted for by the same general principles.) As we have
seen, Est mihi liber ('There is to me a book') was replaced by Habeo
librum ('I have a book') in classical and later Latin. The function of
the dative case (e.g. in mihi) is frequently subsumed under some
general term like 'benefactive* or * dative of interest*. These are
vague terms but they will serve for the purpose of illustrating the
point we are discussing here. The Latin sentence Est mihi liber can
be glossed as ' There is a book with which I am involved or impli-
cated*. In the later stages of Latin, the optional transformation which
brings the * person interested' into surface-structure position (Habeo
librum) becomes obligatory: the 'verb habere' resembles the transitive
verbs of Latin in many respects, and the subject of a transitive verb
is normally 'agentive' (cf. 8.2.7).
The classical Latin for 'I have done the work* (perfective) is Feet
opus. But this could be put into the passive as Opus mihi est factum,
or more commonly with the typical Latin 'agentive adjunct' (a me:
cf. 7.4.5) as Opus a mefactum est. The later Latin equivalent of Feci
opus is Habeo factum opus (cf. Habeo librum), 'The work is done, and
I am involved in its doing (as agent)', i.e. 'I have done the work'.
The origin of the English 'perfect with have9 is similarly explained.
It was at first restricted to transitive verbs, and thus preserved its
relationship with the perfective passive without have (still current in
such sentences as The work is done, The house is built). The subsequent
extension of the 'perfect with have9 to intransitive constructions (e.g.
B has moved parallel with A has moved B) loosens, but does not sever,
the link between the transitive, active with have and the passive without
have. A sentence like John has finished the house is simultaneously
linked with both Johnfinishesthe house and The house isfinished.Current
transformational grammars of English tend to neglect this particular
relationship between the transitive perfective and the passive.
The main point that is being made here, however, is this: it is the
same principle which explains both the diachronic development of
the 'perfect with have' and the 'possessive have'. In both cases, the
8.4. LOCATIVE AND POSSESSIVE 397
/^^-transformation became obligatory, its original function being to
bring the ' person interested' (not necessarily the ' agent') into subject-
position in surface structure. This same principle is still operative in
English. The relationship between / had a book on the table and
There was a book on the table is the same as the relationship between
/ had the work done and The work was done. In neither case can the
surface-subject be regarded as other than the * person interested'.
In a sentence like / had a book stolen (to use one of Chomsky's
examples) the 'person interested' may be either the (indirect) agent
or the ' beneficiary' (' I got someone to steal a book' or * Someone stole
a book from me'). In a sentence like / will have the work done by five
o'clock, the 'person interested' may be either the direct or the indirect
(causative) agent ('The work will be done by me* or 'I will get
someone to do the work'). Whether such sentences should be
described as syntactically ambiguous is, however, a moot point. There
would seem to be no more, and no less, reason for saying that / will
have the work done is derived from two different deep structures than
there is for saying that / built a house by the sea is derived from two
different deep structures. / built a house is also interpretable in two
ways ('I did the building of the house' or 'I got someone else to
build a house for me').
8.4.7 Static, dynamic and causative

All the locative, possessive and attributive sentences so far discussed
in this section (e.g. The book is on the table, John has a book. The book
is valuable) are stative (cf. 7.5.7). It was pointed out, in the treatment
of the 'local' cases in the previous chapter, that the opposition of
'locative' and 'directional' (which is found in the case-systems of
many languages) is a particular instance of a more general aspectual
opposition which might be called static and dynamic, and that, as
locomotion is to location, so acquisition is to possession, and
'becoming' to 'being' (cf. 7.4.6, 7.6.3).
If be and have are not deep-structure verbs, then it would seem
that the following sentences do not contain any verbs in their deep
structure either
(12) John came I went to San Francisco
(13) John got a book
(14) The book became valuable
As be {in San Francisco) is to come/go (to San Francisco), so have (a
book) is to get {a book), and be (valuable) to become (valuable). In each
case we can say that the stative sentence (with be or have) is 'un-
marked', by contrast with the dynamic, which is the 'marked'term
of this particular aspectual opposition.
That this is correctly regarded as an opposition of aspect is clear
from the following considerations. The perfective of the dynamic
implies the (unmarked) imperfective of the static sentence which is in
correspondence with it. Thus John has gone to San Francisco implies
John is in San Francisco, John has got a book implies John has a book,
The book has become valuable implies The book is valuable, (This
particular relationship has led to the virtual equivalence of has got and
has in many dialects of English.) If these pairs of sentences are
treated differently by the base-component of the grammar, then a
significant generalization is lost. For example, certain 'prepositional
phrases' (e.g. in the park, at church) must be analysed as 'locative',
others (e.g. into the park, to church) must be analysed as 'directional';
then the rules must be written in such a way that be does not occur
with 'directionals' and 'verbs of motion' do not occur with 'loca-
tives'. But, if these constructions are distinguished in this way in
deep structure, such facts as the following remain unexplained (except
by ad hoc rules): John went to San Francisco and is still there is an
acceptable sentence, whereas *John went to San Francisco and is still
here is not. The deictic opposition of here/there (with respect to
speaker: cf. 7.2.1) is not relevant to the grammaticality of John is in
San Francisco: it is relevant to the choice between came and went in
John.. .to San Francisco; and it is relevant to the conjunction of a
'directional' and a 'locative' in such sentences as those illustrated
above. These facts can be systematically accounted for, if we say that
the deep structure of e.g. John is in San Francisco and John goes /comes
to San Francisco is identical except for the opposition of static and
dynamic. Each of them is also marked in deep structure for ' proxi-
mity' or 'remoteness' with respect to speaker (here v. there: there are
certain complicating factors, which we need not go into). In com-
bination with the feature' dynamic', the feature ' proximate' generates
come in surface structure and the feature 'remote' generates^.
For each of the three classes of sentences referred to above
locative, possessive and attributivethere is a typical three-place
causative construction (cf. 8.3.6). Corresponding to (i2)-(i4), we have
(12 a) Bill brought I took John to San Francisco
(13 a) Bill gave John a book
(140) Bill made the book valuable
8.4. LOCATIVE AND POSSESSIVE 399
(The last of these three is perhaps a little odd; but it undoubtedly
corresponds to (14) in the way suggested.) The causative construction
is dynamic: and Bill has given John a book implies John has a book,
just as John has got a book does.
Whether the two-place or three-place dynamic locative construction
is used in a particular instance depends upon whether the person or
object that 'engages in locomotion' is 'agentive' or not. The book has
gone to London, though certainly not ungrammatical, is less normal
than John has gone to London; and it may involve ' recategorization'
of the book as 'animate' (cf. 7.4.5). Bill took John to San Francisco (by
contrast with John went to San Francisco or Bill went with John to
San Francisco) marks Bill as 'agentive' and John as * non-agentive \
Finally, it may be observed that possessives and locatives are not
always clearly distinguishable in three-place constructions. It would
be impossible, and it is perhaps unnecessary, to say whether Bring me
the book is 'possessive* or locative (whether me is the 'indirect object*
or 'directional'): we have already drawn attention to the fact that the
same case or preposition is used in many languages for both the
'indirect object' and 'motion to' (cf. 7.4.6). Once again, we notice
the similarity (and perhaps the ultimate identity) of locatives and
possessives. More generally, in this section we have seen that a
considerable number of what are customarily regarded as ' full verbs'
should perhaps be treated as 'surface-structure verbs', each of them
being generated by the grammar to carry distinctions of tense, mood
and aspect, with or without such additional features as ' dynamic' or
'causative'.
John Lyons
Online ISBN: 9781139165570
Chapter
9 - Semantics: General Principles pp. 400-442

9
SEMANTICS: GENERAL PRINCIPLES
9.1 Introductory
9.1.1 The term 'semantics'

Semantics may be defined, initially and provisionally, as 'the study of
meaning*. The term 'semantics' is of relatively recent origin, being
coined in the late nineteenth century from a Greek verb meaning ' to
signify'. This does not mean, of course, that scholars first turned their
attention to the investigation of the meaning of words less than a
hundred years ago. On the contrary, from the earliest times down to
the present day grammarians have been interested in the meaning of
words, and frequently more interested in what words mean than in
their syntactic function. A practical manifestation of this interest is
seen in the innumerable dictionaries that have been produced
throughout the ages, not only in the West, but in all parts of the world
where language has been studied. As we have already seen, the
categories of traditional grammar were to a large extent determined
by their characteristic * modes of signifying' (cf. 1.2.7).
9.1.2 Neglect of semantics in modern linguistics

Many of the more influential books on linguistics that have appeared
in the last thirty years devote little or no attention to semantics. The
reason for this is that many linguists have come to doubt whether
meaning can be studied as objectively and as rigorously as grammar
and phonology, for the present at least. Furthermore, whereas
phonology and grammar quite clearly fall wholly within the province
of linguistics (although the way in which a child learns the phono-
logical and grammatical structure of his language is of considerable
interest to the psychologist), what is commonly referred to as 'the
problem of meaning' might seem to be of equal, if not greater,
concern to philosophy, logic and psychology, and perhaps also to
other disciplines such as anthropology and sociology. Philosophers,
in particular, have always been interested in meaning, since it is
[400]
9-1. INTRODUCTORY 4OI
necessarily involved in such vital and notoriously controversial
philosophical issues as the nature of truth, the status of universal
concepts, the problem of knowledge and the analysis of 'reality*.
9.1.3 Philosophical and psychological interest in meaning

It is not difficult to show why meaning is of interest to philosophers
and psychologists, and why it is regarded as a controversial' problem '.
Consider the innocent-looking question 'What is the meaning of
cowV. It is surely not any particular animal. Is it then the whole class
of animals to which we give the name cow} All cows are different in
some way or other; and, in any case, no one is, or could be, acquainted
with each member of the whole class of cows, and yet we would wish
to say that we know the meaning of cow and are able to use it correctly
in referring to particular animals we have never seen before. Is there
any one property, or set of properties, which distinguishes cows from
all other objects for which we have different words? We soon find
ourselves in the thick of the philosophical controversy between
' nominalists' and' realists' which has endured, in one form or another,
from the time of Plato down to our own day. Have the things to
which we apply the same name some common 'essential' properties
by which we can identify them (as the 'realists' might say) or have
they nothing in common other than the name that by convention we
have learned to apply to them (as the' nominalist' might say) ? And cow
is not a particularly difficult case. We might be prepared to grant that
cows are definable in terms of a biological classification of genera and
species. What about table} Tables are of various shapes and sizes, are
constructed of various materials and are used for various purposes.
But tables, we might say, are at least physically observable and tangible
objects; and we might think it possible to draw up some list of
defining characteristics. What shall we say then about such words as
truth, beauty, goodness, etc. ? Do all the things we describe as' beautiful'
or 'good' have some common property? If so, how do we identify it
and describe it? Shall we say that the meaning of such words as
truth, beauty and goodness is the 'concept' or 'idea' associated with
them in the 'minds' of those who know the language to which the
words belongand, in general, that 'meanings' are 'concepts' or
' ideas' ? If we say this, we shall find ourselves once again in the midst
of philosophical, and psychological, controversy. For many philo-
sophers and psychologists are extremely dubious about the existence
of' concepts', or indeed of the ' mind'. Even if we set these difficulties
402 9- SEMANTICS: GENERAL P R I N C I P L E S
aside, or refuse to look at them, there are others of a more or less
philosophical nature. Does it make sense to say that someone has
used a word to mean something different from what it * really' means ?
Is there such a thing as the 'true' or 'correct' meaning of a word?
9.1.4 The meanings of' meaning'

So far we have talked only of the meaning of words. We also say that
sentences have a meaning. Is 'meaning* being used here in the same
sense? Notice that we often say that sentences and phrases are, or are
not, 'meaningful'; we do not normally say that words are not
'meaningful'. Is it possible then to draw a distinction, and perhaps
many distinctions, between 'being meaningful' and 'having a
meaning'? All these questions, and many other related questions,
have been much discussed by philosophers and linguists. It has
become a commonplace of semantic theory to draw attention to the
numerous meanings of 'meaning'.
In addition to the questions of philosophical interest, there are
questions of more particular concern to the linguist. Philosophers,
like 'the man in the street', usually take 'words' and 'sentences' for
granted. The linguist cannot do this. Words and sentences are first
and foremost units of grammatical description; and they are not the
only grammatical units he will recognize. He must face the general
question of the way in which grammatical units of various kinds are
related to units of semantic analysis. In particular, he must investigate
whether there is a distinction to be drawn between 'lexical' and
'grammatical' meaning.
9.1.5 Inadequacy of current theories of semantics

No one has yet presented even the outlines of a satisfactory and
comprehensive theory of semantics. This point must be made clear
in any discussion of the subject. However, the fact that no systematic
theory of semantics has so far been developed does not imply that
no progress at all has been made in the theoretical investigation
of meaning. A brief account will be given below of the more
important contributions that have been made by linguists and
philosophers in recent years.
We have already defined semantics, provisionally, as 'the study of
meaning'; and this definition reflects the one point of agreement
among semanticists. As soon as we come to consider particular treat-
mcnts of the subject, we are confronted with a bewildering variety
of approaches to the definition and determination of 'meaning'.
Distinctions are made between 'emotive' and 'cognitive' meaning,
between 'significance* and 'signification', between 'performative'
and 'descriptive' meaning, between 'sense' and 'reference', between
'denotation' and 'connotation', between 'signs' and 'symbols',
between 'extension' and 'intension', between 'implication', 'entail-
ment' and 'presupposition', between the 'analytic' and the 'syn-
thetic '; and so on. The terminology of the subject is richand rather
confusing, since it is used without any high degree of consistency
and uniformity between different authors. It is inevitable therefore
that the terms introduced in the present chapter will not necessarily
carry the same implications as the same terms employed in other
treatments of semantics.
We will begin with a brief exposition and some criticism of the
traditional approach to the definition of meaning.
9.2 Traditional semantics

9.2.1 Naming things
Traditional grammar was founded on the assumption that the word
(in the sense of 'lexeme': cf. 5.4.4) was the basic unit of syntax and
semantics (cf. also 1.2.7, 7.1-2). The word was a 'sign' composed of
two parts: we shall refer to these two components, for the purpose of
this discussion, as the form of the word and its meaning. (It will be
recalled that this is but one of the senses in which the term 'form' is
used in linguistics; the 'form' of a word as a 'sign', or lexical item,
must be distinguished from the particular 'accidental', or inflexion-
ally-variant, 'forms' which it assumes in sentences: cf. 4.1.5). Very
early in the history of traditional grammar, the question arose of the
relationship between words and the 'things' they referred to or
'signified'. The Greek philosophers of the time of Socrates, and
following them Plato, set this question in the terms in which it has
generally been posed ever since. For them the semantic relationship
holding between words and ' things' was the relationship of' naming';
and the further question arose as to whether the ' names' we give to
'things' were of 'natural' or 'conventional' origin (cf. 1.2.2). In the
course of the development of traditional grammar, it became custom-
ary to distinguish between the meaning of a word and the 'thing',
or 'things', which were 'named' by it. As the distinction was
404 9- S E M A N T I C S : GENERAL P R I N C I P L E S
formulated by the medieval grammarians: the form of a word (the

vox-part of a dictio) signified 'things' by virtue of the 'concept'
associated with the form of the word in the minds of the speakers of
the language; and the * concept', looked at from this point of view, was
the meaning of the word (its significatio). This we will take to be the
traditional view of the relationship between words and * things'. We
have already seen how this view was made the basis, in principle, for
the philosophical definition of the * parts of speech' according to their
'modes of signifying' (cf. 1.2.7). Without going further into the
details of the traditional theory of 'signification', we should mention
the fact that the terminology employed did not entirely eliminate the
possibility of equivocation, or confusion, in the application of the
term 'signify': the form of a word could be said to 'signify' both
the 'concept' under which 'things' were subsumed (by 'abstraction'
from their 'accidental' properties) and also the 'things' themselves.
And there was of course considerable philosophical disagreement as
to the relationship between 'concepts' and 'things' (in particular,
and most notably, the disagreement between the 'nominalists' and
the 'realists': cf. 9.1.3). We may neglect these philosophical dif-
ferences for the present.
Meaning (Concept)
WORD
Form * * Referent
Fig. 23.
9.2.2 Reference
At this point it will be useful to introduce a modern term for ' things'
in so far as they are ' named', or ' signified', by words. This is the term
referent. We will say that the relationship which holds between words
and things (their referents) is the relationship of reference: words refer
to (rather than 'signify', or 'name') things. Granted the distinction
of form, meaning and referent, we can give a familiar diagrammatic
9-2. TRADITIONAL SEMANTICS 405
representation of the traditional view of the relationship between
them in terms of the * triangle of signification' (sometimes referred to
as 'the semiotic triangle') in Fig. 23. The dotted line between the
form and the referent is intended to indicate that the relationship
between them is indirect: the form is related to its referent through
the mediating (conceptual) meaning associated with both indepen-
dently. The diagram makes clear the important point that in tradi-
tional grammar the word results from the combination of a particular
form with a particular meaning.
9.2.3 Synonymy and homonymy

On the basis of this view of the nature of the word, we can account for
the traditional semantic classification of words in terms of synonymy
and homonymy. The 'ideal* language, one might say (as did the
anomalists: cf. 1.2.3), would be one in which each form had only one
meaning and each meaning was associated with only one form. But
this 'ideal' is probably not realized by any natural language. Two, or
more, forms may be associated with the same meaning (e.g. hide:
conceal, big: largewe will assume that they do in fact have the same
meaning): in which case the words in question are synonyms. And
two, or more, meanings may be associated with the same form
(e.g. bank: (i)' of a river', (ii)' for the deposit of money'): in which case
the words are homonyms. If the language is one for which the ortho-
graphy is at variance with, or unrelated to, the phonology, then one
may of course distinguish further between homography (e.g. lead,
in (i) a dog's lead and (ii) made of lead) and homophony (e.g. meat, meet;
sow, sew: cf. 1.4.2). The important point to notice here is that
homonyms in the traditional view are distinct words: homonymy is
not difference of meaning within one word. In principle, the associa-
tion of two or more meanings with one form is sufficient to justify
the recognition of two or more words. This follows from the tradi-
tional view of the word.
9.2.4 Multiple meaning

The recognition of a distinction between sameness and difference of
meaning does not take us very far in semantics. It seems clear that
some meanings are 'related1 in a way that others are not. The fact
that this is so disturbs the symmetry of the simple opposition between
synonyms and homonyms. How different must the meanings
14 LIT
406 9. SEMANTICS: GENERAL PRINCIPLES
associated with a given form be before we decide that they are
sufficiently different to justify the recognition of two, or more,
different words? In their attempts to demonstrate the 'natural' origin
of language, the Greeks introduced a number of principles to account
for the extension of a word's range of meaning beyond its 'true', or
'original', meaning (cf. 1.2.2). The most important of these principles
was metaphor ('transfer'), based on the 'natural' connexion between
the primary referent and the secondary referent to which the word
was applied. Examples of 'metaphorical' extension might be found
in the application of such words as mouth, eye, head, foot and leg to
rivers, needles, persons in authority, mountains and tables, respec-
tively. In each instance there is discernible some similarity of shape
or function between the referents. Various other types of 'extension'
or 'transference' of meaning were recognized by the Greek gram-
marians, and have passed into traditional works on rhetoric, logic and
semantics. Meanings that are more or less clearly 'related' in
accordance with such principles are not traditionally regarded as
being sufficiently different to justify the recognition of distinct words.
The traditional semanticist would not say that the mouth of a river
and the mouth as a part of the body are homonyms; but rather that
the word mouth has two related meanings. We have, therefore, in
addition to synonymy and homonymy, what has come to be called in
the more recent development of traditional semantics multiple
meaning (sometimes called polysemy). The distinction between
homonymy and multiple meaning is evident in the organization of
the dictionaries we customarily use: what the lexicographer has
classified as homonyms will be listed as different words, whereas
multiple meanings will be given under one entry.
The distinction between homonymy and multiple meaning is, in
the last resort, indeterminate and arbitrary. Ultimately, it rests upon
either the lexicographer's judgement about the plausibility of the
assumed ' extension' of meaning or some historical evidence that the
particular ' extension' has in fact taken place. The arbitrariness of the
distinction between homonymy and multiple meaning is reflected in
the discrepancies in classification between different dictionaries; and
this arbitrariness has been increased, rather than lessened, by the
development of sounder methods of etymology in the nineteenth
century. To give an example: most modern dictionaries of English
recognize as two different words (i) ear, referring to a part of the body,
and (ii) ear, referring to parts of such cereal plants as wheat, barley,
etc. It so happens that these two 'words' have developed from words
which in Old English differed in form as well as meaning (i) eare;
(ii) ear. But how many speakers of English know this? And, if they
do, what effect does their knowledge of this fact have upon their use
of the language? It would clearly be wrong to assume that ear is two
words for those (including lexicographers of English) who know the
history of the language and one word for the rest of usunless it
were in fact discovered that those who know the history of the
language use words like ear differently from those who do not know
the history of the language. If we discovered that this was in fact the
case, we should have to say that the two groups were speaking slightly
different languages. Any historical knowledge we might have about
the development of the meanings of words is in principle irrelevant
to their synchronic use and interpretation (cf. 1.4.5). The distinc-
tion of the synchronic and diachronic in semantics is, however,
subject to the same general limitations as it is in phonology and
grammar.
9.2.5 Antonymy
There is a further traditional category of 'relatedness of meaning'
which must be mentioned here. This is antonymy, or * oppositeness
of meaning*. For the better-known European languages at least,
there are a number of dictionaries 'of synonyms and antonyms'
available, which are frequently used by writers and students to
'extend their vocabulary* and achieve a greater 'variety* of 'style*.
The fact that such special dictionaries are found useful in practice is
an indication that words can be more or less satisfactorily grouped
into sets of synonyms and antonyms. There are two points that
should be stressed, however, in this connexion. First, synonymy and
antonymy are semantic relations of a very different logical nature:
'oppositeness of meaning* (love: hate, hot: cold, etc.) is not simply the
extreme case of difference of meaning. Second, a number of distinc-
tions have to be drawn within the traditional concept of ' antonymy *:
dictionaries of 'antonyms* are only successful in practice to the
degree that their users draw these distinctions (for the most part
unreflectingly). Both of these points will be taken up later (cf. 10.4).
There is little of value that can be extracted from traditional theoretical
discussions of 'antonymy*.
There are a number of criticisms that have been made against
traditional semantics in recent years by both linguists and philo-
sophers. The most important of these criticisms will nowT be discussed.
14-2
9.2.6 Conceptualism and mentalism

We have already referred to philosophical and psychological con-
troversies about the status of ' concepts' and * ideas' in the 'mind'
(cf. 9.1.3). Traditional semantics makes the existence of 'concepts'
basic to the whole theoretical framework, and therefore (almost
inevitably) encourages subjectivism and introspection in the investi-
gation of meaning. To quote Haas: 'an empirical science cannot be
content to rely on a procedure of people looking into their minds, each
into his own'. This criticism presupposes the acceptance of the view
that semantics is, or ought to be, an empirical science, which as far
as possible avoids commitment with respect to such philosophical
and psychological disputes as the distinction of 'body' and 'mind'
and the status of 'concepts'. This view will be accepted in the
discussion of semantics given in these chapters. It should be stressed,
however, that the methodological renunciation of 'mentalism' does
not imply the acceptance of 'mechanism', as some linguists have
suggested. Bloomfield's 'mechanistic' and 'positivist' definition of
the meaning of a word as a full ' scientific' description of its referent
is more detrimental to progress in semantics than the traditional
definition in terms of 'concepts', since it gives preferential treatment
to the relatively small set of words in the vocabularies of natural
languages which refer to 'things' describable, in principle, by the
physical sciences. Moreover, it rests upon two tacit, and unjustified,
assumptions (i) that a 'scientific' description of the referents of these
words is relevant to the way in which the words are used by speakers
of the language (most of whom have little knowledge of ' scientific'
description); and (ii) that the meaning of all words is ultimately
describable in the same terms. Indeed, it could be argued that
Bloomfield's proposal (which has also been made by others) depends
upon a 'realist' view of the relationship between language and 'the
world' which is not very different from that of many ' conceptualists':
at the very least, it tends to promote the view that since there is, for
example, a word intelligence, there is also something to which it refers
(and this 'something' will presumably be satisfactorily described by
'science' in due course); that since there is a word love, there is also
something to which this refers; and so on. The position that should be
maintained by the linguist is one that is neutral with respect to
'mentalism' and 'mechanism'; a position that is consistent with both
and implies neither.
9.2.7 'Ostensive9 definition

Implicit in the previous paragraph is a further criticism of traditional
semantics (and also of certain modern theories). We have already
seen that the term 'meaning' itself has many 'meanings' in ordinary
usage. When we put to someone the question ' What is the meaning
of the word #?' in everyday (non-philosophical and non-technical)
discussion, we receive, and are not surprised to receive, answers of a
form which varies with the circumstances of the situation in which we
ask the question. If we are inquiring about the meaning of a word in
a language other than our own, our question will normally be answered
by translation. (All sorts of semantically interesting problems are
involved in 'translation', which we may neglect for the moment:
cf. 9.4.7). More revealing for our present purpose is the situation in
which we ask about the meaning of words in our own language (or
another language we 'know* at least 'partially'the whole notion of
'knowing a language completely' is of course chimerical). Suppose we
were to ask the meaning of cow in the improbable (but, for the
argument, favourable) situation in which there were a number of
cows present in a neighbouring field. We might be told: ' Do you see
those animals over there? They are cows.' This method of conveying
the meaning of the word cow includes an element of what philosophers
refer to as ostensive definition. (An ostensive definition is one which
defines an object by 'pointing' to it.) But ostensive definition, of
itself, is never sufficient, since, first of all, the person interpreting the
'definition' must know in advance the significance of the gesture of
'pointing' in this context (and must know that 'definition', rather
than something else, is intended) and, more important, he must
correctly identify the object that is 'pointed' to. In the case of our
hypothetical example, the words those animals limit the possibility
of misinterpretation. (They do not eliminate it entirely; but we will
assume that the' definition' of cow has been satisfactorily interpreted.)
The theoretical importance of this over-simplified, and rather un-
realistic, example is twofold: first of all, it makes clear the difficulty
of explaining the meaning of any word without using others to limit
and make more explicit the 'scope' of 'ostension' (it suggests that it
may be impossible to determine, and perhaps also to know, the
meaning of one word without also knowing the meaning of others to
which it is 'related'as for instance cow is to animal)) secondly,
ostensive definition is only applicable to a relatively small set of
words. Consider the futility of trying to explain the meaning of true,
beautiful, etc., in this way! The meaning of such words is typically

explained, not always successfully, in terms of synonyms (whose
meaning is assumed to be known already to the person putting the
question) or by means of rather lengthy definitions of the kind given
in dictionaries. Once again, this emphasizes the inevitable 'circu-
larity ' of semantics: there is no one point in the vocabulary from which
you can start and from which you can derive the meaning of the rest.
This question of * circularity * will come up for discussion later
(cf. 9.4.7).
9.2.8 Context
Another feature of the everyday situations in which we inquire about
the meaning of words is that we are frequently told that ' it depends
on the context'. ('Give me the context in which you met the word;
and Til tell you its meaning/) It is often impossible to give the
meaning of a word without 'putting it in a context'; and dictionaries
are useful in proportion to the number and diversity of the 'contexts'
they cite for words. Frequently, and perhaps most typically of all,
the meaning of a word is explained by giving a 'synonym' with an
indication of the 'contextual' limitations governing the use of the
word in question {addled: 'bad (of eggs)'; rancid: 'bad (of butter)';
etc.). Facts such as thesethe diversity of the ways in which, in
practice, we state the meaning of words, the 'circularity' of the
vocabulary, and the relevance of 'context'are not given full
theoretical recognition in traditional semantics.
9.2.9 'Meaning' and 'use'

At this point mention may be made of Wittgenstein's famous and
influential slogan: 'Don't look for the meaning of a word, look for
its use.' The term ' use' is in itself no clearer than the term ' meaning';
but the substitution of the one for the other has the effect of diverting
the semanticist from the traditional tendency to define 'meaning' in
terms of 'signification'. Wittgenstein's own examples (in his later
work) show that he thought that the ' uses' to which words were put
in language were of many different kinds. He did not put forward, or
claim to be putting forward, a theory of the 'use' of words as a
theory of semantics. But we are perhaps justified in extracting from
Wittgenstein's rather programmatic statement the following prin-
ciples. The only empirical control we have upon the study of language
is the * use' of language-utterances in the multifarious situations of
everyday life. Expressions such as ' the meaning of a word' and ' the
meaning of a sentence (or proposition)' are dangerously misleading in
that they tempt us to go out looking for the ' meanings' they have and
to identify as their 'meanings' such entities as physical objects,
'concepts' in the 'mind' or 'states of affairs' in the physical world.
We have no direct evidence about the understanding of utterances,
only about their misunderstandingwhen something 'goes wrong' in
the process of communication. If, for instance, we say to someone bring
me the red book that is on the table upstairs and he brings us a book of a
different colour, or a box, or goes off downstairs in search of the book,
or does something totally unexpected, we might reasonably say that
he has ' misunderstood' the whole or some part of the utterance (other
explanations are of course possible). If he does what is expected (goes
off in the right direction and comes back with the right book) we
might say that he has correctly understood the utterance. What we
mean is that (in a case like this) there is good prima facie evidence of
a 'behavioural' kind to suggest that he has not misunderstood it. It
may be that if we continued, rather persistently, to test his 'under-
standing' of the words bring or red or book, we should come to a
point where something he did or said would show that his 'under-
standing' of these words is somewhat different from ours: that he
draws from utterances containing them implications that we do not
(or conversely, that we draw implications that he does not), or that
he uses them to refer to a slightly different class of objects and
actions. Normal communication rests upon the assumption that we
all' understand' words in the same way; this assumption breaks down
from time to time, but otherwise 'understanding' is taken for
granted. Whether we have or have not the same 'concepts' in our
'minds' when we are talking to one another is a question that cannot
be answered otherwise than in terms of the ' use' we make of words
in utterances. It would probably be true, but rather pointless, to say
that everyone 'understands' a particular word in a slightly different
way. Semantics is concerned with accounting for the degree of uni-
formity in the 'use' of language which makes normal communica-
tions possible. Once we abandon the view that the 'meaning' of a
word is what it 'signifies', we shall quite naturally recognize that
relationships of different kinds have to be stated in accounting for
'use'. Two of the 'factors' to be discussed later will be distinguished
as reference (which has been mentioned already) and sense.
412 9- SEMANTICS: GENERAL PRINCIPLES
9.2.10 Indeterminacy of meaning

As a further consequence of the abandonment of the view that the
' meaning' of a word is what it' signifies', and that what it * signifies' is
* transferred' (in some sense) from speaker to hearer in the process of
communication, we shall be more ready to accept that it is both un-
necessary and undesirable to assume that words have a fully-
determined meaning. As we have seen, the way in which language is
used in normal situations can be explained on the much weaker
assumption that the speakers of the language in question are in
sufficient agreement about the 'use* of words (what they refer to,
what they imply, etc.) to prevent 'misunderstandings'. This point
must be borne in mind whenever we talk of 'the meaning* of words
and sentences. We shall take it for granted throughout the following
sections of these chapters on semantics.
9.3 * Meaningfulness'
9.3.1 'Having meaning' and 'significance'
It was pointed out above (9.1.4) that, although we commonly say
that sentences or phrases are, or are not, 'meaningful', we do not
normally say that words are not 'meaningful' (for the present we shall
continue to adopt the traditional view, according to which words are
the minimal 'meaningful' units of language; 'word' is here being
used of course in the sense of lexeme: cf. 5.4.4). This fact in itself
suggests that the term ' meaningful' may be employed in two distinct
senses. We will assume that this is so and, for convenience and clarity,
introduce a terminological distinction between having meaning and
significance (or being significant). In terms of this distinction, we will
say that words have meaning, whereas phrases and sentences may or
may not be significant. It should be observed that this statement leaves
open the possibility that otaer units besides words may have meaning;
and it does not imply that there is no connexion between ' having
meaning' and 'significance'. Traditional semantics (and a number of
modern theories) confuse the distinction that is being drawn here by
talking of 'signification' in both cases.
In this section it will be argued that having meaning (in the sense
in which this notion is to be defined) is logically prior to 'meaning':
in other words, that we must first of all decide whether a particular
element has meaning before we ask what meaning it has; further-
9.3. ' M E A N I N G F U L N E S S ' 413
more, although this may appear paradoxical at first sight, that it is

possible for an element to have meaning without having any parti-
cular meaning.
9.3.2 The situational context

We will start with an intuitive and undefined notion of context.
Every (spoken) utterance occurs in a particular spatiotemporal
situation which includes the speaker and hearer, the actions they are
performing at the time and various external objects and events. As
we have already seen, there may be * deictic * features of the utterance
which make reference to the situation in which it occurs (cf. 7.2.1).
The hearer will not be able to understand an utterance unless he
interprets these * deictic' elements correctly by reference to the
relevant features of the situation. However, the context of an utterance
cannot simply be identified with the spatiotemporal situation in
which it occurs: it must be held to include, not only the relevant
objects and actions taking place at the time, but also the knowledge
shared by the speaker and hearer of what has been said earlier, in so
far as this is pertinent to the understanding of the utterance. It must
also be taken to include the tacit acceptance by the speaker and hearer
of all the relevant conventions, beliefs and presuppositions * taken for
granted* by the members of the speech-community to which the
speaker and hearer belong. The fact that it is in practice, and perhaps
also in principle, impossible to give a full account of all these ' con-
textual* features should not be taken as a reason for denying their
existence or their relevance. But it may be interpreted as an argument
against the possibility of constructing a complete theory of the
meaning of utterances. (The reader will have observed that we are at
present talking of utterances, not sentences: cf. 5.1.2.)
9.3.3 'Having meaning* implies choice

On the basis of this intuitive notion of * context' we can now define
having meaning for utterances. An utterance has meaning only if its
occurrence is not completely determined by its context. This definition
rests upon the widely-accepted principle that * meaningfulness implies
choice'. If the hearer knows in advance that the speaker will inevitably
produce a particular utterance in a particular context, then it is
obvious that the utterance gives him no information, when it occurs:
no 'communication' takes place. Complete utterances will generally
have meaning, since the speaker might in the last resort have remained
silent. But there are some socially-prescribed utterances which are
highly, if not wholly, determined by their contexts; and such
utterances are theoretically interesting in a number of different
respects. Let us assume for the sake of the argument that How do you
do? is the sole socially-prescribed utterance in the context of being
formally introduced to someone, and that it is mandatory in such
situations. If this is so, then it seems quite reasonable to say that
How do you do? has no meaning. All that needs to be said about this
utterance in a semantic description of English utterances is that it is
' used' in the situations in question. It would be futile to insist that it
must 'mean* something over and above its 'use'. But let us now
assume that, although it is the sole socially-prescribed utterance in the
context of introduction, it contrasts with silence (or a nod of the head,
a smile, a surly look, etc.), in the sense that the person being intro-
duced has all these * choices' open to him. Then, by the definition
given above, the result of each * choice' has meaningit may com-
municate something to the other person; and we can sensibly go on
to ask what meaning each of the potential * actions' has by contrast
with the others.
9.3.4 Relevance of non-linguistic behaviour

It is worth while drawing out some further implications of the
principle of * choice* on the basis of this simple example. First of all,
utterances interact with, and may be in semantic contrast with, non-
linguistic behaviour (including silence, facial expressions and gestures).
The utterance How do you do?, although itself mandatory in the
context, may be pronounced in a variety of ways'politely',
'casually', 'scornfully', 'condescendingly', etc.and these different
'modalities' of the utterance may be 'expressed' by 'tone of voice' or
accompanying gestures (or both simultaneously). The question that
now arises (and it is relevant to all utterances, not merely to those that
are socially prescribed for particular contexts) is whether we should
say that features such as 'tone of voice' and gestures (of anger, con-
descension, politeness, etc.) have meaning. In principle, the answer is
clear. If such features of the utterance are completely determined (in
the sense that the speaker exercises no control over themno
' choice'), they have no meaning. If, on the other hand, he deliberately
wishes to express his anger, his impatience or his 'good breeding',
then these' facts' are indeed' communicated' by him, and the features
9.3. 'MEANINGFULNESS' 415
of the utterance which serve this purpose have meaning in terms of
the definition given above. That the hearer may 'infer* the same facts
even when they are not' communicated' by the speaker (and may not
be able to tell whether he is intended to 'infer' them or not) does not
affect this question in the slightest. It is quite inappropriate to widen
the notion of' communication' to include all the ' information' which
can be 'inferred' from an utterance by the hearer. The principle of
'choice* determines whether utterances and features of utterances
have meaning or not.
9.3.5 Quantifiability of' having meaning9

The second point to be made about having meaning is that it is in
principle quantifiable with respect to ' expectancy' (or probability of
occurrence) in context. From this point of view, having no meaning
is merely the limiting case of complete 'predictability'. Any utterance
(or feature of the utterance) that is not completely 'predictable'
(determined by its context) may be more or less probable than either
silence or some other utterance (or feature of the same utterance)
with which it is in contrast in the system of communication. And the
less probable a particular element is, the more meaning it has in that
context ('element' should here be taken to refer to all the results of
'choice', including silence, permitted by the system of communi-
cation for particular contexts). To return to our simple example: if
How do you do? contrasts with, but is far more probable than, silence
(or any other 'element') in the context of introductions, then it has
less meaning than silence in that context. It might even be reason-
able in such cases to say that the socially-prescribed utterance is
'unmarked', and has meaning only in the rather vacuous sense of
contrast (without the communication of any positive 'information'),
whereas silence is ' marked' and may serve a positive communicative
function. This would appear to be an intuitively satisfying statement
of the relationship between the two 'behavioural' possibilities (on
the assumption that the facts are as they have been stated). In any
case, it is certainly in accord with general, everyday usage to say that
the ' meaningfulness' of utterances, and parts of utterances, varies in
inverse proportion to their degree of 'expectancy' in context. And
this is the sense of 'meaningfulness' that is here being explicated in
terms of having meaning. Although it may be possible to say that one
element has more meaning than another in a particular context on the
basis of their relative 'probabilities of occurrence', it is clear that the
precise quantification of having meaning would depend upon our
ability to identify the contextual features which determine the
* probabilities of occurrence*. (Strictly speaking, we should not talk
of ' probabilities of occurrence* and * variation in inverse proportion*
unless we are able to determine and compute the relevant condi-
tioning factors.) It is very unlikely that it will ever be possible to
quantify having meaning in this precise sense. But this is of less
importance than one might expect since, as we shall see, what
meaning elements have in a given context is unrelated to how much
meaning they have relative to the elements with which they are in
contrast. The point to be stressed here is that the question what
meaning an element has arises only in the case of those elements
which do in fact have meaning (in the sense in which we have defined
this notion) in the contexts in which they occur. Although this point
has been illustrated so far only for complete utterances of a * ritual-
istic* character, we shall generalize it later in the light of the distinc-
tion between utterances and sentences drawn in a previous chapter
(cf. 5.i.a).
9.3.6 ' Behaviourism * in semantics

There are two further points to be made about such socially-
prescribed utterances as How do you do?. They tend to be iready-
made*, in the sense that they are learned by native speakers as
unanalysed wholes and are clearly not constructed afresh on each
occasion on which they are used in what, following Firth, we may
refer to as 'typical repetitive events in the social process*. Since they
have this character, it would be possible to account for them in a
'behaviouristic' framework: the utterances in question could reason-
ably be described as 'conditioned responses* to the situations in
which they occur. This fact should not be overlooked by the seman-
ticist. There is a good deal of our everyday use of language which is
quite properly described in 'behaviourist* terms, and can be attri-
buted to our 'acting out* of particular 'roles* in the maintenance of
socially-prescribed, 'ritualistic* patterns of behaviour. With regard
to this aspect of their use of language, human beings behave like
many animals, whose ' systems of communication * are made up of a
set of'ready-made utterances' which are used in particular situations.
The more typically human aspects of language-behaviour, which
depend upon the generative properties of language and also upon the
semantic notions of having meaning, reference and sense, are not
9-3- 'MEANINGFULNESS' 417
plausibly accounted for by extending to them the 'behaviourist'
notions of 'stimulus' and 'response'. It is nevertheless true that
human language includes a 'behavioural' component. Although we
shall say no more about this in what follows, we give theoretical
recognition to it here.
9.3.7 ' Phatic communion'

Mention should also be made in this connexion of that aspect of
language-behaviour to which Malinowski applied the term 'phatic
communion'. He was drawing attention to the fact that many of our
utterances cannot rightly be said to have as their sole, or primary,
function the communication or seeking of information, the giving of
commands, the expression of hopes, wishes and desires, or even the
'evincing of emotion' (in the loose sense in which semanticists
frequently employ this last phrase), but serve to establish and
maintain a feeling of social solidarity and well-being. Many ' ready-
made ' utterances like How do you do?, which are socially prescribed in
particular contexts, may serve this function of 'phatic communion'.
But there are many others which are more or less freely constructed
by native speakers, and which simultaneously communicate informa-
tion and serve in 'phatic communion'. An example might be It's
another beautiful day, said (we will suppose) as the opening utterance
in a conversation between customer and shopkeeper. Quite clearly
this utterance is not primarily intended to 'convey' to the shop-
keeper some information about the weather; it is an instance of
'phatic communion'. At the same time, it does have a meaning,
different from the meaning of innumerable other utterances that
might have occurred in the same context and would have served the
purposes of 'phatic communion' equally well; and the next 'move'
in the conversation will generally be related to the particular utterance
on the basis of this meaning. We must therefore distinguish between
that aspect of the ' use' of utterances which may be referred to their
function in 'phatic communion' and that part of their 'use* which is
to be distinguished as their meaning (if they have meaning in terms
of our definition). In saying this, we recognize that, even when both
of these aspects are present, either one or the other may be the
dominant part of the 'use* of the utterance. Malinowski overstated
his case when he said that the communication of information was one
of the 'most derivate and specialized functions' of language.
9.3.8 Extension of 'having meaning' to all linguistic units

So far we have illustrated the notion of having meaning only in
relation to whole utterances regarded as unanalysable wholes. For the
present, we shall continue to talk of utterances, rather than sentences,
and we shall continue to draw upon the intuitive notion of 'context';
but we will now generalize the notion of having meaning in terms of
the following principle: any linguistic element which occurs in an
utterance has meaning only if it is not completely determined
('obligatory') in that context.
It is evident that the notion of having meaning (as it is here
defined) is applicable at all levels in the analysis of utterances,
including the phonological. There will be many contexts, for instance,
in which the words lamb and ram might equally well occur in other-
wise identical utterances. Since these utterances, we will assume, are
different in meaning (the referents of lamb and ram are different and
the implications Contained' in the utterances in which they occur
are, in general, different), the phonemes /I/ and /r/ not only have
meaning, but have a different meaning in these utterances. And there
are other utterances, containing words other than lamb and ram,
where the difference in meaning might be carried solely by the
phonological opposition of /I/ and /r/. As we saw in an earlier chapter
(cf. 3.1.3), the phonological structure of particular languages rests,
ultimately, upon this differential property of phonemes (more
strictly, of their * distinctive features') within certain limits imposed
by the complementary principle of phonetic similarity. There is
therefore good reason for applying the notion of having meaning even
at the level of phonological analysis. It is worth noting, however,
that in the case of phonetically-distinguishable, but * similar',
speech-sounds, having meaning necessarily implies having a different
meaning in at least some contexts. This is not so at the 'higher' levels
of analysis. In languages in which the speech-sounds [I] and [r] occur
but never distinguish utterances, we say that they are in comple-
mentary distribution or free variation (in other words, that they are
alternative phonetic realizations of the same phonological unit:
cf. 3.3.4). In those contexts in which speech-sounds, elsewhere
recognized as distinct phonological units, have the same meaning we
may quite reasonably say that they are 'synonymous*. Examples
might be the initial vowels of the alternative pronunciations of
economics (contrast the differential value of the same vowels in beat:
bet, etc.) or the stress-patterns of controversy: controversy.
9.3. 'MEANINGFULNESS* 419
Although the semanticist must give theoretical recognition to the
principle of having meaning at the phonological level, he will not
generally be concerned further with 'the meaning* of phonological
units. The reason is that phonological units never have reference and
do not contract any semantic relationships other than sameness and
difference of meaning. Furthermore, sameness of meaning when it
holds between phonological units (phonological 'synonymy', as
illustrated above) is sporadic and unsystematic. It has to be stated in
terms of alternative realization rules for particular words; once these
rules have been given, there is nothing more of interest to be said.
Generally speaking (the qualification is required for cases of ' sound-
symbolism*a semantically interesting phenomenon which limita-
tions of space prevent us from discussing further: cf. 1.2.2), 'the
meaning* of a given phonological unit is simply its difference from
every other phonological unit (if there are any) that might have
occurred in the same context.
9.3.9 Restricted contexts

We have now reached a stage in the argument at which we must
draw upon the distinction between utterances and sentences (cf. 5.1.2).
Two points must be borne in mind. When we use language to com-
municate with one another, we do not produce sentences, but
utterances; those utterances are produced in particular contexts and
cannot be understood (even within the limits imposed above on the
interpretation of the term 'understanding*: cf. 9.2.9) without a
knowledge of the relevant contextual features. Furthermore, in the
course of a conversation (let us suppose that it is a conversation) the
context is constantly developing, in the sense that it 'takes into
itself* from what is said and what is happening all that is relevant to
the production and understanding of further utterances. The limiting
case of contexts which have not ' developed * in this sense would be
those in which the participants in a conversation do not draw upon
their previous knowledge of one another or the 'information*
communicated in earlier utterances, but where they share the more
general beliefs, conventions and presuppositions governing the
particular ' universe of discourse * in the society to which they belong.
Such contexts, which we will refer to as restricted contexts, are com-
paratively rare, since most utterances depend for their understanding
upon the information contained in previous utterances. We must not
lose sight of the relationship between utterances and particular contexts.
The second point to be made is this: since sentences are never
produced by speakers (sentences being theoretical units established
by linguists to state the distributional limitations upon the occurrence
of classes of grammatical elements), there can be no direct relation-
ship between sentences and particular contexts. At the same time,
utterances have a grammatical structure which depends upon their
'derivation' from sentences; and the grammatical structure of
utterances is, or can be, semantically relevant. This is particularly
clear in the case of 'syntactic ambiguity* (cf. 6.1.3). Moreover
(except for such 'ready-made* expressions as How do you do?)
utterances are produced by speakers and understood by hearers on
the basis of the regularities of formation and transformation deter-
mined for sentences by the rules of the grammar. At the present time,
neither linguistics nor any of the other sciences concerned with the
' mechanisms' underlying the production of utterances is in a position
to make any very definite statements about the way in which a know-
ledge of the abstract relationships holding between grammatical
elements in sentences interacts with contextual features of various
kinds to effect the production and understanding of utterances in
which the observable 'correlates' of these grammatical elements
occur. That there is some interaction between the grammatical
structure of the language and the relevant contextual features would
seem to be a fact; and we must take account of it.
Since we cannot in general identify either the actual elements
' selected' by the speaker in the production of utterances or all the
relevant features of particular contexts, we can make it a matter of
methodological decision to do what linguists have generally done in
practice; and that is to handle the semantic relationships between
utterances in terms of the semantic relationships holding between the
sentences from which they are assumed to be 'derived' when they
are produced by native speakers in restricted contexts. (The notion
of 'restricted context' must still be retained, since, as we shall see
below, the semantic relationships that hold between sentences cannot
be stated without at least this degree of' contextualization': cf. 10.1.2.)
The features of particular contexts will then be invoked (in what must
be, for the present at least, a rather ad hoc manner) to account for the
'residual' semantically-relevant aspects of utterances. What has been
represented here as a matter of conscious, methodological decision
should not be taken to imply any priority of the grammatical over the
contextual in the psychological processes of producing and under-
standing utterances.
9.3 ' M E A N I N G F U L N E S S ' 421
9.3.10 Deep-structure elements have meaning in sentences

We may now apply the notion of 'having meaning* to the gram-
matical elements from which sentences are generated by the rules
determining the formation and transformation of their bases (cf.
6.6.1). Since having meaning implies 'choice', it follows that no
elements introduced into sentences by means of obligatory rules can
have meaning in the sense of the definition. (Such ' dummy' elements
as do in Do you want to go? have no meaning: cf. 7.6.3.) Furthermore,
if we assume that all the 'choices' are made with respect to the
selection of elements in ' deep' structure (these elements being either
'categories' or 'features': cf. 7.6.9), it is clear that the notion of
having meaning is not bound to units of any particular rank. First of
all, the distinction of such units as morphemes, words and phrases in a
particular language is to some degree based on 'surface' structure
(6.6.1); and, secondly, there are many 'grammatical categories'
(tense, mood, aspect, gender, number, etc.: cf. 7.1.5) which may or
may not be realized by morphemes or words, but which constitute
systems of 'choices' in sentences. Whether or not one can draw a
sharp distinction between 'lexical' and 'grammatical' meaning, in
terms of what meaning the elements have, is a question we shall
discuss later (cf. 9.5.2). For our present purpose, it is sufficient to
point out that the notion of having meaning applies equally to elements
of both kinds in the 'deep' structure of sentences. Moreover, this
notion is taken into account, explicitly or implicitly, in all recent
linguistic theory. Classes of elements (denoted by either auxiliary or
terminal symbols: cf. 6.2.2) are established at each point of 'choice'
in the generation of sentences.
It follows from what has just been said that no element in a
sentence will have meaning unless it is a member of one of the
syntactically-determined classes in the 'deep' structure of the
sentence: and it is this fact which justifies the almost universally
accepted assumption of linguists, logicians and philosophers that the
set of elements which have meaning in a particular language is, at
least to a very high degree, coextensive with the members of the
terminal 'constituents' and 'features' of that language. However, it
does not follow that every 'constituent' and 'feature' will have
meaning in every sentence in which it occurs. This is an important
point which has occasionally been overlooked by linguists, and
therefore merits some further discussion.
The whole question turns on the distinction between grammatical
42a 9- SEMANTICS: GENERAL P R I N C I P L E S
and semantic acceptability. As we saw in an earlier chapter (cf. 4.2.12 ff.),
grammaticality is that part of the acceptability of utterances which can
be accounted for in terms of the rules of formation and transformation
specifying the permissible combinations of the distributional classes
of elements ('categories' and 'features') in sentences. It is generally
recognized that there will be innumerable sentences generated by the
grammar of any language which are unacceptable in various ways;
and it is traditional to describe at least one type of unacceptability by
saying that the sentences in question are 'meaningless' or 'non-
sensical'. Let us suppose for example that the following sentences are
generated by a grammar of English (and are therefore grammatical):
(a) John drinks milk (beer, wine, water, etc.)
(b) John eats cheese {fish, meat, bread, etc.)
(c) John drinks cheese (fish, meat, bread, etc.)
(d) John eats milk (beer, wine, water, etc.)
Let us further assume that all these sentences are generated with the
same structural description: that the verbs drink and eat, and the
nouns milk, beer, wine, water, cheese, fish, meat, bread, etc., are not
distinguished in the lexicon by means of any relevant syntactic
feature. It seems clear that, in some sense of the terms ' acceptable'
and 'unacceptable', utterances derivable from the sentences grouped
under (a) and (b) are acceptable, whereas utterances derivable from
the sentences grouped under (c) and (d) are unacceptable (in ' normal'
circumstances). Whether this kind of acceptability and unaccept-
ability is to be described by reference to criteria of ' meaningfulness'
(in the sense of this term which we propose to distinguish as ' signifi-
cance') is a question to which we shall return presently. The point
being made here is that the sets of elements which can occur and have
meaning as the verb and the object of these sentences are far smaller
subsets of the elements whose occurrence is permitted by the rules
of the grammar. Once again the limiting case is that of an element
whose occurrence is wholly determined by the context of the other
elements in the sentence. An example of complete determination at
this level might be the occurrence of teeth in / bit him with my false
teeth. As we shall see later (cf. 9.5.3), this sentence manifests a
semantically interesting type of syntagmatic 'presupposition' which
is normally latent, but which is made overt when 'syntactic support'
is required for a 'modifier' (in this instance, false). If the word teeth
never occurred in sentences other than those in which it is completely
9.3 ' M E A N I N G F U L N E S S ' 423
determined by its context, it would not have meaning in English; and
the semanticist would have nothing to say about it.
The purpose of this argument has been to demonstrate the way in
which the notion of having meaning both can, and must, be trans-
ferred from the more 'concrete' instances in which it applies to
grammatically-unstructured whole utterances, on the one hand, and
to utterances which differ minimally in respect of their phonological
structure, on the other, to the more 'abstract* plane on which it
applies to the more important far more numerous class of sentences
generated by the rules of grammar. The notion of having meaning is
validated by its reflection of the intuitively-satisfying principle that
'meaningfulness implies choice* in particular contexts. Its trans-
ference to the more 'abstract' plane rests upon a methodological
decision, the motivation for which is twofold: first, it recognizes the
fact that the particular contextual features which influence the
production and interpretation of utterances can only be handled in an
ad hoc manner; and second, it satisfactorily relates the semantic
interpretation of sentences to their syntactic description. Given that a
particular element has meaning over a certain range of sentences, we
can then ask what meaning it has; and this question can be answered
in various ways, as we shall see in the next section.
9.3.11 ' Significance'

We must now look briefly at the notion of 'significance' (cf. 9.3.1). At
first sight, one might be tempted to identify significance with total
acceptability, relative to particular contexts for utterances and relative
to the more generalized restricted contexts for sentences. But we have
already seen that there are many 'layers' of acceptability ('above' the
grammatical) which, although they are often described loosely as
'semantic', can be distinguished from what is traditionally described
as 'meaningfulness' or 'significance' (cf. 4.2.3). Certain utterances
might be condemned as 'blasphemous' or 'obscene'; some might be
acceptable in certain uses of language (prayers, myths, fairy-stories,
science-fiction, etc.), but unacceptable in everyday conversation.
Little purpose would be served by attempting to define 'significance'
in such a way that it covered all these various 'dimensions' of
acceptability. For example, suppose it were the case that, although the
verb die was freely used in combination with animate nouns, including
the names of persons, in English, there was a generally-respected
taboo prohibiting its use in combination with my father, my mother,
424 9* SEMANTICS: GENERAL PRINCIPLES
my brother and my sister (i.e. with reference to members of one's own
immediate family): so that My father died last night, but not His
father died last night, would be considered unacceptable. It seems
quite evident that the correct explanation of the unacceptability of
My father died last night is one which allows us to say, first of all, that
it is * meaningful', since, if it occurred despite the taboo, it would be
understood (indeed it could be argued that the taboo depends upon
the possibility of understanding it) and, secondly, that the semantic
relationship between My father died last night and His father died last
night is identical with that between My father came last night and His
father came last night, etc. Traditionally, the significance of gram-
matically well-formed sentences is accounted for in terms of certain
general principles of compatibility between the 'meanings' of their
constituent elements. One might say, for instance, that John eats milk
and John drinks bread are not significant, because the verb eat is com-
patible only with nouns (as its object) which denote consumable
solids and the verb drink with nouns which denote consumable
liquids. (Notice that in terms of this statement John eats soup might
be regarded as semantically anomalous, being rendered ' socially
acceptable' by a particular convention external to the generalizable
rules for the interpretation of English sentences.) There are great
difficulties attaching to the notion of significance (one might wish to
argue, for instance, that John eats milk is 'meaningful', though the
circumstances in which it might be employed are somewhat unusual).
Nevertheless, the traditional explication of this notion in terms of
'compatibility' seems to be basically sound. We shall discuss some
more recent formulations of this notion in a later section (cf. 10.5.4).
9.4 Reference and sense

9.4.1 Reference
The term ' reference' was introduced earlier for the relationship which
holds between words and the things, events, actions and qualities they
'stand for' (cf. 9.2.2). It was pointed out that, under certain circum-
stances, the question 'What is the meaning of the word x1y can be
answered by means of 'ostensive' definitionby pointing to, or
otherwise indicating, the referent (or referents) of the word (cf. 9.2.7).
There are certain philosophical difficulties attaching to the precise
definition of the notion of' reference' which we need not be concerned
with here. We shall assume that the relationship of reference (some-
94 REFERENCE AND SENSE 425
times described as * denotation') is essential to the construction of any
satisfactory theory of semantics: in other words, that there is a sense
in which at least certain items in the vocabularies of all languages can
be put into correspondence with 'features' of the physical world.
Acceptance of this assumption does not imply acceptance of the
view that reference is the semantic relationship to which all others
can be reduced; nor does it imply that all the items in the vocabulary
of a language have reference. As it will be understood here, * reference'
necessarily carries with it presuppositions of 'existence* (or 'reality')
which derive from our direct experience of objects in the physical
world. To say that a particular word (or other item which has meaning)
' refers to an object' implies that its referent is an object which ' exists'
(is 'real') in the same sense in which we say that particular persons,
animals and things 'exist'; and also, that it would be possible, in
principle, to give a description of the physical properties of the object
in question. This notion of 'physical existence' may be taken as
fundamental for the definition of the semantic relationship of
reference. The application of the terms 'existence' and 'reference'
may then be extended in various ways. For instance, although there
are no such objects (we will assume) as goblins, unicorns or centaurs,
we can quite reasonably ascribe to them a fictional or mythical
'existence' in a certain kind of discourse; and we can therefore say
that the words goblin, unicorn or centaur have reference in English
when the language is used in discourse of this kind. Similarly, we can
extend the application of the terms 'existence' and 'reference' to
such theoretical constructs of science as atoms, genes, etc.; and even
to quite abstract objects. It is important to notice, however, that the
source of these' analogical' extensions of the notions of' existence' and
'reference' is to be found in their fundamental, or primary, appli-
cation to physical objects in the 'everyday' use of language.
It follows from this interpretation of the notion of reference that
there may be many items in the vocabulary of a language which do
not stand in a relationship of reference to anything outside the
language. It may be, for example, that there is no such thing as
intelligence or goodness to which the words intelligent and good refer,
although it would always be possible for a psychologist or a philo-
sopher to postulate the existence of such entities within the framework
of a particular theory of psychology or ethics, and even to claim that
their ' reality' is demonstrable by means of some kind of ' ostensive'
definition. The fact that there may be disagreement, at various levels
of sophistication, about the 'reality' of certain alleged 'objects' does
not affect the general principle that reference presupposes existence.
To insist that all lexical items must refer to something would be futile
if it meant that, in certain instances, no evidence could be brought
forward for the existence of that * something' other than the fact that
there was a lexical item * referring' to it.
Two further points may be mentioned in connexion with the notion
of reference. By accepting the view that certain lexical items refer to
objects and properties of objects outside language we are not logically
committed to the assumption that all the objects denoted by a parti-
cular term form a * natural class', in the sense that they belong
together' independently of the 'convention' tacitly accepted by the
speech-community in question to group them together 'under' a
common term: in other words, the position maintained here is
compatible with either 'nominalism' or 'realism' in philosophical
semantics. Second, the reference of a lexical item need not be precise
and fully-determined, in the sense that it is always clear whether a
particular object or property falls within the scope of a given lexical
item: we have already seen that no such assumption is required in
order to account for the 'understanding' of utterances in the normal
process of communication (cf. 9.2.9). It is frequently the case that the
'referential boundaries' of lexical items are indeterminate. For
example, the precise point at which one draws the line between the
reference of hill and mountain, of chicken and hen, of green and blue,
and so on, cannot be specified. But this does not mean that the notion
of reference does not apply to such words. It is a characteristic of
languages that they impose a particular lexical 'categorization' upon
the world and draw the boundaries 'arbitrarily', as it were, at
different places. As we shall see, this is one of the reasons why it is
often impossible to establish lexical equivalences between different
languages. The fact that the 'referential boundaries' are 'arbitrary'
and indeterminate does not normally lead to misunderstanding, since
the 'precise' classification of an object 'under' one lexical item or
another is very rarely relevant; and when it is, we have resort to
other systems of identification or specification. For example, if we
wish to refer to one of two persons each of whom might be appro-
priately denoted by either girl or woman, we can distinguish them by
name, by relative age, by the colour of their hair, by what they are
wearing, and so on. Although the reference of girl' overlaps' that of
woman, the two words are not synonymous; their relative position
along the dimension of age is fixed, and there will be many cases in
which only one of the two words is appropriate. Far from being a
9 4 - REFERENCE AND SENSE 427
defect, as some philosophers have suggested, referential imprecise-
ness ' of the kind illustrated makes language a more efficient means of
communication. Absolute ' preciseness' is unattainable, since there is
no limit to the number and nature of the distinctions one might draw
between different objects; and there is no virtue in being obliged to
draw a greater number of distinctions than is necessary for the
purpose in hand.
9.4.2 Sense
We must now introduce the notion of 'sense'. By the sense of a word
we mean its place in a system of relationships which it contracts
with other words in the vocabulary. It will be observed that, since
sense is to be defined in terms of relationships which hold between
vocabulary-items, it carries with it no presuppositions about the
existence of objects and properties outside the vocabulary of the
language in question.
If two elements can occur in the same context, they have meaning
in that context; and we may go on to ask what meaning they have.
As we have seen, one part, or component, of the meaning of certain
elements may be described in terms of their reference. Whether the
two elements have reference or not, we can ask whether they have the
same meaning or not in the context, or contexts, in which they both
occur. Since sameness of meaning, synonymy, is a relation which holds
between two (or more) vocabulary-items, it is a matter of sense, not
reference. For reasons which we need not go into here, we may wish
to say that two items have the same reference, but differ in sense;
and we certainly wish to say that items may be synonymous, even if
they have no reference. It may be assumed that, for items which have
reference, identical reference is a necessary, but not sufficient,
condition of synonymy.
Theoretical discussions of synonymy are often vitiated by two
unjustified assumptions. The first is that two elements cannot be
'absolutely synonymous' in one context unless they are synonymous
in all contexts. This assumption is sometimes supported with an
appeal to the distinction between 'cognitive' and 'emotive' meaning.
But this distinction is itself far from clear. It is undeniable that the
choice of one item rather than another by a particular speaker may be
influenced by the difference in their 'emotive associations'. How-
ever, this does not mean that these 'emotive associations' are
always relevant (even if they are shared by other members of the
speech-community). And it cannot simply be made a matter of
assumption that words always carry with them the' associations' which
derive from their use in other contexts. We will therefore reject the
assumption that words cannot be synonymous in any context unless
they are synonymous in all contexts.
The second assumption commonly made by semanticists is that
synonymy is a relation of identity holding between two (or more)
independently-defined senses. In other words, the question whether
two words, a and b, are synonymous is reduced to the question
whether a and b denote the same entity, their sense. In the approach
to semantics being outlined here, there will be no need to postulate
the existence of independently-defined senses. Synonymy will be
defined as follows: two (or more) items are synonymous if the
sentences which result from the substitution of one for the other have
the same meaning. This definition clearly rests upon a prior notion of
* sameness of meaning' for sentences (and utterances). We shall
return to this question presently. The point being made here is that
the relation of synonymy is stated as holding between lexical items and
not between their senses. The synonymy of lexical items is part of
their sense. To put the same point in a more general form: what we
refer to as the sense of a lexical item is the whole set of sense-relations
(including synonymy) which it contracts with other items in the
vocabulary.
9.4.3 Paradigmatic and syntagmatic relations of sense

There are many other sense-relations besides synonymy. For instance,
husband and wife are not synonymous, but they are semantically-
related in a way that husband and cheese, or hydrogen, are not; good
and bad are different in sense, but more similar than good and red,
or round\ knock, bang, tap and rap are related in a way that knock and
eat, or admire, are not. The relationships illustrated here are para-
digmatic (all the members of the sets of semantically-related terms can
occur in the same context). Terms may also be related to one another
syntagmatically: cf. blond and hair, bark and dog, kick and foot, etc.
(For the general distinction of paradigmatic and syntagmatic relations,
see 2.3.3.) Whether these syntagmatic and paradigmatic sense-
relations can be defined in terms of their 'distance* from synonymy
along a scale of sameness and difference of sense (as some semanticists
have suggested) is a question we will not go into here: an alternative
treatment will be presented in the following chapter. For the
94- REFERENCE AND SENSE 429
present, we will simply assume that at least some vocabulary-items
fall into lexical systems, and that the semantic structure of these systems
is to be described in terms of the sense-relations holding between the
lexical items. This statement is intended as a more precise formula-
tion of the principle that ' the meaning of each term is a function of
the place it occupies in its own system' (cf. 2.2.1, with reference to
Russian and English kinship terms).
9.4.4 Semantic fields

In recent years, there has been a good deal of work devoted to the
investigation of lexical systems in the vocabularies of different
languages, with particular reference to such fields (or domains) as
kinship, colour, flora and fauna, weights and measures, military ranks,
moral and aesthetic evaluation, and various kinds of knowledge, skill
and understanding. The results obtained have conclusively demon-
strated the value of the structural approach to semantics, and have
confirmed the pronouncements of such earlier scholars as von Hum-
boldt, de Saussure and Sapir to the effect that the vocabularies of
different languages (in certain fields at least) are non-isomorphic: that
there are semantic distinctions made in one language which are not
made in another; moreover, that particular fields may be categorized
in a totally different way by different languages. This fact is expressed
in Saussurean terms by saying that each language imposes a specific
form on the a priori undifferentiated substance of the content-plane
(cf. 2.2.2, 2.2.3). As a n illustration of this notion we may take (as
substance) the field of colour and see how it is determined, or
'informed', in English.
9.4.5 Colour-terms
For simplicity, we will first of all consider only that part of the field
which is covered by the words red, orange, yellow, green and blue.
Each of these terms is referentially imprecise, but their relative
position in this lexical system is fixed (and as a set they cover the
greater part of the visible spectrum): orange lies between red and
yellow, yellow between orange and green, and so on. It is part of the
sense of each of these terms that they belong to this particular lexical
system in English and that they contract relationships of contiguity
(or, more precisely perhaps, 'betweenness') relative to one another
in the system. It might appear that the notion of sense is unnecessary
43 9- S E M A N T I C S : GENERAL P R I N C I P L E S
here, and that an account of the reference of each of the colour-terms

would be sufficient as a description of their meaning. Consider,
however, the conditions under which one might come to learn, or be
said to know, the reference of these words. The child learning
English cannot first of all learn the reference of greeny and then
subsequently the reference of blue or yellow, so that at a particular time
he could be said to know the reference of one, but not the other. (It is
true that he might learn, ostensively, that green referred to the
colour of grass, or the leaves of a particular tree, or one of his mother's
dresses: but the reference of green is wider than any particular
instance of its application, and a knowledge of its reference involves
knowledge also of the boundaries of its reference.) It must be
supposed that over a certain period the child gradually learns the
position of green with respect to blue and yellow, of yellow with
respect to green and orange, and so on until he has learnt the position
of each of the colour-terms with respect to its neighbour in the lexical
system and the approximate location of the boundaries of the area in
the continuum of the field covered by each term. His knowledge of the
meaning of the colour-terms necessarily involves therefore a knowledge
of both their sense and their reference.
The field covered by the five colour-terms considered so far can be
regarded as an undifFerentiated (perceptual or physical) substance
upon which English imposes a particular form by drawing boundaries
in it at particular places and giving to the five areas thus recognized a
particular lexical classification (as red, orange, yellow, green and blue).
It has often been shown that other languages impose a different form
upon this substance by recognizing a different number of areas
within it and drawing the boundaries at different places. To refer to
an example used above: the Russian words sinij and goluboj together
cover roughly the same area as the English word blue; they refer to
distinct, but contiguous, colours and are co-ordinate in the system
with the words zelenyj and zeltyj (' green' and * yellow')they are
not to be regarded as terms which refer to different shades of one
colour, in the way that crimson and scarlet, with other terms, sub-
divide the area covered by red in English (cf. 2.2.3).
The relationship between colour-terms and their meaning is not
quite as straightforward as we have so far represented it. The
difference in the reference of red, orange, yellow, green and blue can
be described in terms of their variation in hue (the reflection of light
at different wavelengths). Physicists recognize two other variables in
the analysis of colour: luminosity, or brightness (the reflection of more
or less light), and saturation (the degree of freedom from dilution
with white). The ranges of colour denoted by black, grey and white
in English differ mainly in respect of luminosity; and there are other
common colour-terms whose reference must be specified according
to all three dimensions of variation; e.g. brown refers to a range of
colour that is between red and yellow in hue, of relatively low lumi-
nosity and saturation; pink to a colour that is reddish in hue, of fairly
high luminosity and fairly low saturation. Consideration of these facts
might lead us to say that the substance of the field of colour is a three-
dimensional (physical or perceptual) continuum.
Even this statement is an over-simplification. Not only do languages
differ in the relative weight they give to the three dimensions of hue,
luminosity and saturation in the organization of their systems of
colour-terms (e.g. both Latin and Greek seem to have given greater
weight to luminosity than to hue); there are languages in which
distinctions of colour are made according to quite different principles.
In a classic paper on this subject, Conklin has shown that the four
main ' colour-terms' of Hanunoo (a language of the Philippines) are
associated with lightness (generally covering white and the light tints
of other English colours), darkness (including English black, violet,
blue, dark green and the dark shades of other colours), 'wetness'
(usually correlating with light green, yellow and light brown, etc.)
and 'dryness' (usually correlating with maroon, red, orange, etc.).
That the distinction between ' wetness' and ' dryness' is not simply a
matter of hue (' green' v. ' red': this might appear to be the distinction
on the basis of the most frequent English translations of the two terms
in question) is clear from the fact that' a shiny, wet, brown section of
newly cut bamboo' is described by the term which is generally used
for light green, etc. Conklin concludes that * colour, in a western
technical sense, is not a universal concept'; that the oppositions in
terms of which the substance of colour is determined in different
languages may depend primarily upon the association of the lexical
items with culturally-important features of objects in the natural
environment. In the case of the Hanunoo words one of the dimensions
of the system would seem to derive from the typical appearance of
fresh, young ('wet', 'succulent') plants. It is worth noting in this
connexion that dictionaries of English frequently define the main
colour-terms with respect to typical features of the environment
(e.g. blue might be said to refer to the colour of the clear sky, red to
the colour of blood, and so on).
43* 9- SEMANTICS: GENERAL PRINCIPLES
9.4.6 Semantic 'relativity'

The field of colour has been discussed at some length, because it is so
often used as an example to demonstrate the way in which the same
substance may have a different form imposed upon it by different
languages. We have now seen that even in the case of colour there is
good reason to doubt whether one can reasonably postulate an a
priori identity of 'content-substance*. Conklin's account of the
categories of 'colour' in Hanunoo should certainly warn us against
assuming that the linguistically-relevant dimensions of the substance
of colour are necessarily those selected as criterial by the physical
sciences. The general conclusion to be drawn is that the language of a
particular society is an integral part of its culture, and that the lexical
distinctions drawn by each language will tend to reflect the culturally-
important features of objects, institutions and activities in the society in
which the language operates. This conclusion is supported by a number
of recent studies of various fields in the vocabularies of different
languages. Since the natural environments in which different societies
live, not to mention their institutions and patterns of behaviour, are so
diverse, it is extremely doubtful whether one can talk profitably about
semantic structure as the imposition of form upon an underlying
(perceptual, physical or conceptual) substance common to all languages.
As Sapir has said: 'The worlds in which different societies live are
distinct worlds, not the same world with different labels attached.'
Even if it is admitted that different societies live in 'distinct
worlds' (and we shall return to this point presently), it might still be
maintained that each language imposes a particular form upon the
substance of the 'world' in which it operates. Within certain limits
this is true (as we have seen, for example, in the case of colour-terms).
It is important to realize, however, that lexical systems do not
necessarily presuppose an 'underlying' substance. Let us assume for
the sake of the argument that the words honesty, sincerity, chastity,
fidelity, etc., fall into a lexical system with virtue. The structure of
this system can be described in terms of the sense-relations that hold
between its members. From this point of view, the question whether
there are any 'substantial' correlations between the lexical items and
identifiable traits of character or patterns of behaviour is irrelevant.
If there are any such correlations they will be described in terms of
reference, not sense. In short, the applicability of the notion of
substance in semantics is determined by the same postulate of
'existence' as the notion of reference (cf. 9.4.1).
The assertion that ' the worlds in which different societies live are
distinct worlds* is frequently interpreted as a proclamation of
linguistic 'determinism'. Whether Sapir (or von Humboldt before
him and Whorf after him) believed that our categorization of the
world is totally determined by the structure of our native language is
a question we need not go into here. It is generally agreed that
linguistic determinism, interpreted in this strong sense, is an un-
tenable hypothesis. However, our earlier acceptance of the view that
particular languages reflect in their vocabulary the culturally-
important distinctions of the societies in which they operate commits
us to a certain degree of linguistic and cultural Relativity'. We must
therefore account for the undeniable fact that it is possible to achieve
a knowledge of the structure of the lexical systems in languages other
than our own, whether in learning them for practical purposes or for
the investigation of their vocabularies. Translation from one language
to another clearly depends upon this possibility.
9.4.7 Cultural overlap

Cultures (in the sense in which this term is used by anthropologists
and sociologists) are not coterminous with languages. For example,
many of the institutions, customs, articles of dress, furniture and
food, etc., to be found in France and Germany are found also in
England: others are peculiar to each country, or to a particular region
or social class in each country. (The relationship between language
and culture is of course far more complex than this simplified state-
ment would suggest: political boundaries do not coincide with
linguistic boundaries, even if we grant that there is some validity in
the notion of a unified speech-community; cultural identities may be
found between different social classes in different countries; and
so on.) In general, it may be assumed that there will be a greater or
less degree of cultural overlap between any two societies; and it may
be the case that certain features will be present in the culture of all
societies. Practical experience of learning foreign languages (in the
normal conditions in which these languages are used) suggests that
we quickly identify certain objects, situations and other features in the
area of cultural overlap and learn the words and expressions that
apply to them without difficulty. The meaning of other words and
expressions are learned less readily and their correct use comes, if
ever, only with long practice in speaking the language. We may give
theoretical recognition to these facts of experience by saying that
entry is made into the semantic structure of another language in
the area of cultural overlap; and that, once we have broken into the
circle by means of the identification of items in this area (cf. 9.4.7,
on the inevitable 'circularity' of semantics), we can gradually
develop and modify our knowledge of the rest of the vocabulary
from within, by learning the reference of the lexical items and the
sense-relations that hold between them in the contexts in which
they are used. True bilingualism implies the assimilation of two
cultures.
9.4.8 'Application'
When items of different languages can be put into correspondence
with one another on the basis of the identification of common features
and situations in the cultures in which they operate, we may say that
the items have the same application. The reasons for using this term
rather than * reference' are twofold. First of all, it has been used of
the relationship which holds between situations and expressions
which occur in them (e.g. of the relationship between Excuse me,
Thank you, etc., and the various identifiable situations in which these
utterances occur); this is clearly not a relationship of reference.
Secondly, we wish to allow for the semantic identification of lexical
items which have no reference: one might wish to say, for example,
that the English word sin and the French word pichi have the same
application, although it might be difficult, or impossible, to establish
this fact in referential terms. It may well be that the second of these
reasons for the introduction of the notion of application will disappear
with the construction of a comprehensive and satisfactory theory of
culture. At the present time, the notion of application, like the
process of translation, rests rather heavily upon the intuitions of
bilingual speakers. This does not mean that the notion'is without any
objective foundation, since bilingual speakers tend to be in agreement
about the application of most words and expressions in the languages
they speak.
Nothing has been said in this section about the way in which para-
digmatic and syntagmatic sense-relations are established. Before we
discuss the question we must consider the possibility of extending the
notions of reference and sense to grammatical, as well as lexical, items.
9- SEMANTICS: GENERAL PRINCIPLES 435
9.5 * Lexical' and i grammatical' meaning

9.5.1 ' Structural meanings'
In our discussion of 'grammatical categories', we made reference to
the traditional, 'Aristotelian' view that only the major parts of speech
(nouns, verbs, 'adjectives', and adverbs) were 'meaningful' in the
full sense of this term (they ' signified' the ' concepts' which were held
to constitute the 'matter' of discourse) and that the other parts of
speech contributed to the total meaning of sentences by imposing
upon the' matter' of discourse a certain grammatical' form' (cf. 7.1.3).
Strikingly similar views have been maintained by many opponents of
traditional grammar.
For example, Fries draws a distinction between 'lexical' and
' structural' meanings which almost exactly reflects the ' Aristotelian'
distinction of 'material' and 'formal' meaning. The major parts of
speech have 'lexical' meaning; and this is given in the dictionary
associated with the grammar. By contrast, the distinction between the
subject and the object of a sentence, oppositions of definiteness, tense
and number, and the difference between statements, questions and
requestsall these distinctions are described as 'structural meanings'.
("The total linguistic meaning of any utterance consists of the lexical
meanings of the separate words plus such structural meanings... It is
the devices' that signal structural meanings which constitute the
grammar of a language.')
At least three different kinds of semantic function are subsumed by
Fries under the term 'structural meaning': and other linguists have
employed the term 'grammatical meaning' (in contrast with 'lexical
meaning') in the same sense. The three different kinds of 'meaning'
are: (1) the 'meaning' of grammatical items (typically the minor
parts of speech and the secondary grammatical categories); (2) the
'meaning' of such grammatical 'functions' as 'subject-of, 'object-
oV or 'modifier-oP; (3) the 'meaning' associated with such notions
as 'declarative', 'interrogative' or 'imperative' in the classification of
different sentence-types. It is important to distinguish these various
kinds of 'grammatical meaning'; and we will discuss each briefly in
turn.
9.5.2 Lexical and grammatical items

Various criteria have been proposed for the distinction of grammatical
and lexical items. The most satisfactory (and the only one we will
43^ 9* SEMANTICS: GENERAL PRINCIPLES
mention here) has been formulated by Martinet, Halliday and others
in terms of paradigmatic opposition within either closed or open sets
of alternatives. A closed set of items is one of fixed, and usually small,
membership: e.g. the set of personal pronouns, tenses, genders, etc.
An open set is one of unrestricted, indeterminately large, member-
ship ; e.g. the class of nouns or verbs in a language. In terms of this
distinction we can say that grammatical items belong to closed sets,
and lexical items to open sets. This definition corresponds quite
well with the traditional distinction between the major parts of speech,
on the one hand, and the minor parts of speech and secondary gram-
matical categories, on the other. Unlike a number of other definitions
that have been suggested, it is not restricted to languages of one
morphological 'type' (e.g. 'inflecting' languages: cf. 5.3.6). For the
moment, we will assume that it is correct, and that on the basis of the
distinction between closed and open sets all the elements introduced
into the deep structure of sentences can be classified as either
'grammatical' or 'lexical'. The question that now arises is whether
there is any difference, in principle, between the meaning of gram-
matical and lexical items.
The first point to notice is that lexical items are traditionally said
to have both 'lexical' and 'grammatical meaning' (both 'material'
and 'formal meaning': cf. 9.5.1). To employ the terminology of
scholastic, 'speculative' grammar: a particular lexical item, e.g. cowy
not only 'signifies' a particular 'concept' (the 'material', or 'lexical',
meaning of the item in question), but it does so according to a
particular 'mode of signifying', e.g. as a 'substance', a 'quality',
an 'action', etc. (cf. 1.2.7, 7-1-1)- Although linguists rarely express
themselves in these terms nowadays, this general conception of the
difference between the ' lexical' and ' grammatical' meaning of lexical
items is still current. Moreover, it would seem to have a certain
validity.
For example, there is a well-known poem by Lermontov, which
begins: Beleet parus odinokij... This is difficult, if not impossible, to
translate into English, because it depends for its effect upon the fact
that in Russian 'being white' can be 'expressed' by means of a
'verb* (as in belyj = 'white', which in temporally, aspectually and
modally unmarked sentences would be employed, normally, without
'the verb to be\ cf. 7.6.3). Parus odinokij is translatable as 'a lonely
sail' (parus being the noun, and odinokij an 'adjective'). In traditional
terms, the 'verb' represents 'being white' as a 'process' or 'activity',
the 'adjective' as a 'quality' or 'state'. The difference between the
9-5- 'LEXICAL' AND 'GRAMMATICAL' MEANING 437
choice of the * verb' rather than the 'adjective' in the present instance
can be brought out in English only by means of a rather inadequate
circumlocution, such as 'There is a lonely sail which stands out
(or even, 'shines forth') white (against the background of the sea
or sky). . .'. Problems of this nature are familiar enough in trans-
lating from one language to another. We are here concerned with
the theoretical question: can one say that there is a particular ' gram-
matical meaning' associated with each of the major parts of speech?
We have already seen that the difference between the 'verb' and
the 'adjective' in general syntactic theory is a complex matter: in
some languages, no such distinction can be drawn at all; in others,
there are many syntactic features associated with the distinction, and
they may be in conflict in particular instances (cf. 7.6.4). But the
principal criterion, and the one which reflects the traditional distinc-
tion between 'activity' and 'quality', is the aspectual distinction
between 'dynamic' and 'static' (cf. 8.4.7). In Russian this difference
of 'grammatical meaning' is 'superimposed' upon the 'lexical
meaning' which is common to both the 'verb' beletj and the 'adjec-
tive' belyj. To this extent, the traditional theory of the 'modes of
signifying' is correct: it must be reformulated, of course, within a
more satisfactory theory of syntactic structure.
At the same time, one must not lose sight of the general principle
that 'having meaning implies choice'. If the language that is being
described allows the option of either a 'verbal' or an 'adjectival'
expression (to restrict ourselves to the distinction illustrated), then
the employment of either the one or the other comes within the
scope of the semantic analysis of the language. One may then ask
whether the two 'modes' of expression have the same meaning or
not; and, if they differ in meaning, one may go on to ask how they
differ semantically. If this difference can be referred to some deep-
structure grammatical distinction (e.g. 'dynamic' v. 'static'), it
is appropriately called 'grammatical meaning'. But this does not
imply that the selection of a 'verb' rather than an 'adjective' is
always associated with a difference of 'grammatical meaning'. In
many instances, a particular 'lexical meaning' is necessarily associ-
ated with one, rather than another, part of speech. In short, on this
question, as on many others, linguistic theory should strike a balance
between 'notional' and 'formal' grammar (cf. 7.6.1). It should not
be maintained that 'denoting an activity' is part of the 'meaning'
of every 'verb' or 'denoting a quality' part of the 'meaning' of
every 'adjective'.
15 LIT
Lexical items are traditionally said to have both ' lexical' (' material')
and 'grammatical' ('formal') meaning. Grammatical items are
generally described as having only 'grammatical' meaning. In the
previous chapter, we saw that a certain number of items that occur in
the surface structure of sentences as * verbs' can be interpreted as the
' lexical realizations' of aspectual, causative and other * grammatical'
distinctions. Whether these suggestions were correct or not is a
question that we may leave on one side. In the present state of
syntactic theory, the distinction between grammatical and lexical
items is somewhat indeterminate. The reason is that the distinction
between open and closed sets of alternatives can only be applied with
respect to the positions of' choice' in the deep structure of sentences;
and, as we have seen, there is considerable room for disagreement as to
where these positions of 'choice' are.
The main point that must be made here is that there seems to be
no essential difference between the 'kind of meaning' associated with
lexical items and the ' kind of meaning' associated with grammatical
items in those cases where the distinction between these two classes of
deep-structure elements can be drawn. The notions of 'sense' and
'reference' are applicable to both. If there is any generalization that
can be made about the meaning of grammatical elements (and it will
be recalled that certain 'purely grammatical' elements have no
meaning: cf. 8.4.1), it would seem to be that grammatical 'choices'
have to do with the general notions of spatial and temporal reference,
causation, process, individuation, etc.notions of the kind that were
discussed in chapters 7 and 8. However, we cannot assume in advance
that such notions, even if they are clearly identifiable, will necessarily
be ' grammaticalized', rather than ' lexicalized', in the structure of any
particular language.
9.5.3 The 'meaning' of grammatical 'functions'

The second class of phenomena in the structure of English to which
Fries (and others) have applied the term 'structural meaning' (or
'grammatical meaning') is exemplified by such notions as 'subject',
'object' and 'modifier'. Fries's book was written before the develop-
ment of modern transformational syntax, and he was concerned solely
with surface structure (from a rather restricted point of view).
Consequently, much of what he has to say about these ' functional'
notions, though correct, is hardly relevant to semantic analysis. This
is also true of most modern linguistic theory.
9.5. ' L E X I C A L ' AND 'GRAMMATICAL' MEANING 439
It is quite clear that certain deep-structure grammatical relation-

ships between lexical items and combinations of lexical items are
relevant to the semantic analysis of sentences. Chomsky has suggested
that it is the 'functional* notions 'subject-oP, ' direct-object-oP,
4
predicate-of' and * main-verb-oP which constitute the principal
deep-structure relations between lexical items; and Katz, Fodor and
Postal have recently tried to formalize the theory of semantics by
means of a set of 'projection rules' operating upon lexical items that
stand in such relationships within sentences (cf. 10.5.4). Such notions
as 'subject', 'predicate' and 'object' were discussed in the previous
chapter; and we saw that their formalization in general syntactic
theory is not quite so clear as Chomsky has assumed. It follows that
the status of the 'projection-rules' which interpret sentences on the
basis of these notions is also somewhat dubious.
In our discussion of 'transitivity' and 'ergativity', we pointed out
that many of the 'direct objects' of English sentences could be
generated by embedding one-place constructions as the ' predicates'
of two-place constructions, and introducing a new 'agentive' subject.
But we also saw that there were other two-place, transitive construc-
tions which could not be satisfactorily generated in this way. This
fact alone would suggest that the relation 'direct-object-oP cannot
be given a single interpretation in the semantic analysis of sentences.
In traditional grammar, many different kinds of 'direct object' were
distinguished. Of these, one may be mentioned here, since (whatever
its status in the theory of syntax) it is undoubtedly of considerable
importance in semantics. It is the 'object of result' (or 'effect').
The 'object of result' may be illustrated with reference to the
following two sentences:
(1) He is reading a book
(2) He is writing a book
In (1) the book that is referred to exists prior to, and independently of,
its being read; but the book referred to in (2) is not yet in existence
it is brought into existence by the completion of the activity described
by the sentence. By virtue of this difference, the book in (1) is tradi-
tionally regarded as an' ordinary' object of the verb is reading, whereas
the book in (2) is described as an 'object of result'. From the semantic
point of view, any verb that has as its object an 'object of result'
might be appropriately described as an 'existential causative'. The
most common ' verb' in English that falls into this class is make; and
we have already pointed out that this is also a 'causative auxiliary'
15-2
(cf. 8.3.6, 8.4.7). This same 'verb' serves, like do, as a 'pro-verb' in
interrogative sentences. A question like What are you doing? carries
fewer presuppositions about the 'predicate' in the sentence which
answers it (the verb can be transitive or intransitive, but it must be
an 'action' verb: cf. 7.6.4). By contrast, What are you making? pre-
supposes that the ' activity' in question is ' resultative' and has, as its
aim or term, the 'existence' of some 'object'. This difference, it
should be noticed, is not so striking in a number of European
languages as it is in English. (For example, Qu'est-ce que tu fais?,
in French, might be translated as either 'What are you doing?' or
'What are you making?'.) But this does not mean that the difference
between 'ordinary' objects and 'objects of result' is irrelevant in
these languages.
The importance of this notion of ' existential causative' lies in the
fact that there is frequently a high degree of interdependence between
a particular verb, or class of verbs, and a particular noun, or class of
nouns, in sentences that manifest the 'object of result' construction.
For example, no satisfactory semantic analysis of the noun picture
could be given which did not state its syntagmatic relationship with
such verbs as paint and draw; conversely, the fact that these verbs
may take the noun picture as an 'object of result' is to be stated as
part of their meaning.
This notion of syntagmatic interdependence, or presupposition, is
of considerable importance in the analysis of the vocabulary of any
language (cf. 9.4.3). It is of far wider applicability than can be
illustrated here. There are presuppositions holding between particular
classes of nouns and verbs, where the noun is the subject of the verb
(e.g. bird:fly,fish:swim); between 'adjectives' and nouns (blond: hair,
addled: egg); between verbs and 'ordinary' objects (drive: car);
between verbs and nouns in an 'instrumental' relationship (bite:
teeth, kick: foot); and so on. Many of these relationships between
particular classes of lexical items cannot be stated, except by means of
an ad hoc set of 'projection rules', within the framework of trans-
formational syntax outlined by Chomsky.
Since there is as yet no fully satisfactory syntactic framework
within which to state the various sense-relations in terms of which the
vocabulary of languages are structured, we shall make no attempt to
formulate a set of 'projection rules' operating upon deep-structure
grammatical relations. In the following chapter, we shall discuss a
number of particularly important paradigmatic relations between
classes of lexical items; and we shall do so in a fairly informal way.
9.5. 'LEXICAL* AND 'GRAMMATICAL' MEANING 441
We assume that these relations could be stated more elegantly on the
basis of a more satisfactory account of deep-structure grammatical
relations.
9.5.4 The ' meaning' of' sentence- types'

The third class of * meanings' which are commonly described as
'grammatical' may be exemplified with reference to the difference
between * declarative', * interrogative' and 'imperative' sentences. In
recent transformational theory, there has been a tendency to intro-
duce into the deep-structure phrase-markers of sentences such
grammatical elements as 'question-marker' and 'imperative-marker',
and then to formulate the rules of the transformational component
in such a way that the presence of one of these ' markers' will * trigger
off' the appropriate transformational rule. We are not concerned
here with the syntactic advantages of this formulation of the dis-
tinction between various 'sentence-types', but with its semantic
implications.
It has been suggested (by Katz and Postal) that these ' markers' are
semantically similar to the lexical and grammatical elements which
occur as constituents of the nuclei of sentences. For instance, the
' imperative-marker' would be listed in the dictionary with an entry
'that represents it as having roughly the sense of "the speaker
requests (asks, demands, insists, etc.) that"'. But this suggestion
rests upon a confusion in the term 'meaning'. It is to eliminate such
confusions that semanticists have drawn distinctions between 'sense',
'reference' and various other kinds of 'meaning'. If we continue to
employ the term 'meaning' for all kinds of distinguishable semantic
function, we can say quite reasonably that there is a difference of
'meaning' between corresponding statements, questions and com-
mands (which are not necessarily ' expressed' as declarative, interro-
gative and imperative sentences, respectivelybut for simplicity we
may neglect this fact). But the question whether two lexical items have
' the same meaning' or not is normally interpreted with respect to the
notion of synonymy: sameness of sense. This is a paradigmatic
relation: i.e. a relation which either holds or does not hold between
items that occur in the same context in the same 'sentence-type'. In
the following chapter, we shall see that the notion of 'synonymy'
with respect to x and y can be explicated in terms of the set of impli-
cations which 'follow' from two sentences which differ only in that
where one has x the other has y. But these considerations simply do
not apply to corresponding declarative and interrogative or impera-
tive sentences (e.g. You are writing the letter v. Are you writing the
letter? or Write the letter). Although corresponding members of
different * sentence-types * can be said to differ in ' meaning', they
cannot be said to differ in sense. There is no point in trying to
formalize the theory of semantics in such a way that the * meaning*
of a 'question-marker* or * imperative-marker' can be described in
the same terms as the 'meaning* of lexical items.
John Lyons
Online ISBN: 9781139165570
Chapter
10 - Semantic Structure pp. 443-481

10
SEMANTIC STRUCTURE
10.1 Introductory
i o. i. i The priority of sense-relations
In this chapter we shall be concerned with the notion of sense (as
distinct from both reference and application: cf. 9.4.1-9.4.8). We have
already seen that the vocabulary of a language will contain a number
of lexical systems the semantic structure of which can be described in
terms of paradigmatic and syntagmatic sense-relations; and we have
stressed that these relations are to be defined as holding between lexical
items and not between independently-determined senses (cf. 9.4.2).
This last point is of considerable theoretical and methodological
importance. It is one of the cardinal principles of 'structuralism', as
developed by de Saussure and his followers, that every linguistic item
has its 'place' in a system and its function, or value, derives from the
relations which it contracts with other units in the system (cf. 2.2.2-
2.2.9). Acceptance of the structural approach in semantics has the
advantage that it enables the linguist to avoid commitment on the
controversial question of the philosophical and psychological status
of 'concepts' or 'ideas' (cf. 9.2.6). As far as the empirical investi-
gation of the structure of language is concerned, the sense of a lexical
item may be defined to be, not only dependent upon, but identical
with, the set of relations which hold between the item in question and
other items in the same lexical system. The nature of these sense-
relations will be discussed in this chapter.
The methodological significance of the structural approach to the
definition of sense may be illustrated by means of a comparison with
the proposal made by Russell and other modern logicians for the
definition of such notions as length, weight, shape, etc. In traditional
logic the question 'Is x the same length as y}' was generally inter-
preted as if it were secondary to and dependent upon questions of
a quite different logical structure: 'What is the length of #?' and
'What is the length of yV (length being conceived as a property that
objects might have more or less of). In practice, the length of an object
is determined by comparing it with some conventional standard.
[443 ]
IO
444 - SEMANTIC STRUCTURE
When we say, for example, that x is exactly one metre long, we are
asserting that if it were compared with the platinum-iridium bar kept
at the International Bureau of Weights and Measures, x would be
found to be equal in length to the distance between the two lines
marked on the bar (the fact that since i960 the metre is internationally
defined by means of more complex, but more reliable, physical
measurements does not affect the point being made). In other words,
the question 'What is the length of #?' is answered by means of a
procedure which yields an answer to a question of the form 'Is x
the same length as zV (z being the standard). Given two objects
x and y, we can compare them directly with one another or in-
directly by reference to some third object z (the platinum-iridium
bar in Paris, a ruler that has been calibrated in accordance with some
agreed standard of measurement, etc.). In either case, 'What is the
length of xV is dependent upon, and indeed reducible to, a set of
questions of the form 'Is x the same length asjy?' There is no other
way, empirically, of determining the length of x\ this being so, Russell
proposed that length should in fact be defined in terms of the relation
'having the same length as*. (We need not go into the details of
Russell's formulation of the definition here. The general principle is
independent of this.)
Just as ' having the same length' is a relation which holds between
two objects (and not between the 'lengths' inherent in them), so
'having the same sense'or synonymyis a relation which holds
between two lexical items (and not between the 'senses' associated
with them in the minds of the speakers: cf. 9.2.6). The definition of
sense is far more complex than the definition of length (or weight,
etc.) since there is more than the relation of sameness and difference
involved. But there would seem to be no more reason to postulate a
set of ' senses' associated with the lexical items in a system than there
is to postulate a set of 'lengths' inherent in physical objects. The
question 'What is the sense of #?' (and the answer to this question, it
will be recalled, is only one part of the answer to the question ' What
is the meaning of #?') is methodologically reducible to a set of
questions each of which is relational: 'Does sense-relation J?^ hold
between x and yV
10.1.2 ' Analytic* and 'synthetic' implication

The notion of sense is frequently discussed by philosophers in
connexion with the distinction between synthetic and analytic state-
IO.I. INTRODUCTORY 445
ments. This distinction may be put as follows: a synthetic statement
is one which is true 'contingently'as a matter of empirical fact
which might have been otherwise; an analytic statement is one that is
1
necessarily' true, and its truth is guaranteed by (i) the sense of its
constituent elements and (ii) the syntactical rules of the language. To
take a standard example: the sentence All bachelors are unmarried
might be regarded as analytic on the grounds that bachelor and un-
married are semantically-related in such a way that the truth of the
sentence is guaranteed.
The validity of the notion of analyticity is open to dispute; and it is
possible that it is philosophically indefensible in the form in which it is
generally discussed. Fortunately, the semantic analysis of language as
it is used in everyday discourse need not wait upon the solution of the
philosophical problems attaching to the distinction between con-
tingent and necessary truth. What the linguist requires is a pragmatic
concept of analyticityone which gives theoretical recognition to the
tacit presuppositions and assumptions in the speech-community and
takes no account of their validity within some other frame of reference
assumed to be absolute or linguistically and culturally neutral. It was for
this purpose that we introduced earlier the notion of the restricted con-
text. Any statements that are made in this chapter about the semantic
relations that hold between sentences by virtue of the sense of the
lexical items in them are to be interpreted in the light of this notion.
Sense-relations are stateable within a framework which includes the
notion of implication. This notion may be introduced here by way of
the prior concepts of explicit assertion and denial. We will assume
that in all languages it is possible to establish rules of correspondence
between affirmative and negative sentences; and that the corres-
pondence between a particular affirmative and a particular negative
sentence is accounted for by the grammar of the language. Thus the
negative sentence John is not married corresponds to the affirmative
sentence John is married. We will now say that a negative sentence
explicitly denies whatever is explicitly asserted by the corresponding
affirmative sentence; and on the basis of this notion of explicit
assertion and denial we can construct the semantically more interesting
notion of implicit assertion and denial, or implication. One sentence,
S-L, is said to imply another, S2symbolically, Sx => S2if speakers
of the language agree that it is not possible to assert explicitly Sx and
to deny explicitly S2. And Sx implicitly denies S2S1 implies not
S2:S1 => ~ S2if it is agreed that the explicit assertion of Sx makes
impossible, without contradiction, the explicit assertion of *S2.
446 10. SEMANTIC STRUCTURE
It should be stressed that implication, in the sense in which it has
been defined here, is in principle objectively testable. This does not
mean of course that all speakers will necessarily agree that one sentence
implies another. As we have already seen, what is normally meant by
'understanding* utterances can be quite well accounted for without
making the assumption that all speakers of a language will draw from
a given utterance exactly the same set of implications (cf. 9.2.9).
What may be assumed is that there is a sufficiently large overlap in the
implications that hold for different speakers to prevent misunder-
standing in the majority of instances in which they communicate with
one another. Semantic theory must allow for a certain degree of
indeterminacy in the number and nature of the implications that hold
between the sentences of a language.
10.2 Synonymy
10.2.1 A stricter and a looser sense of 'synonymy'
One may distinguish a stricter and a looser interpretation of the term
'synonymy*. According to the stricter interpretation (which is the
one most commonly found in contemporary semantic theory) two
items are synonymous if they have the same sense. It is this inter-
pretation of synonymy that we shall be discussing in the present
section.
The looser interpretation may be illustrated by means of a quota-
tion from Roget's Thesaurus: 'Suppose we take the word " n i c e " . . .
Under it [in the Index] we will see.. .various synonyms representing
different shades of meaning of the word "nice".' The 'synonyms'
given for nice in the Index are savoury, discriminative, exact, good,
pleasing, fastidious and honourable. Each of these words itself appears
in one of the lists of ' synonyms* in the main body of the text. For
instance, turning to the section in which pleasing occurs we find ' an
array of literally dozens of equivalents... expressing every possible
shade of meaning*. So too for good, exact, etc. The thesaurus there-
fore provides us with ' an array of hundreds of words and expressions
which are at our disposal to use instead of.. ."nice" with which we
started*. All these words and expressions are 'synonymous* with nice
under the looser interpretation of the notion of synonymy.
IC.2. SYNONYMY 447
10.2.2 Proposals for the quantification of synonymy

It is sometimes suggested that synonymy is a matter of degree; that
any set of lexical items can be arranged on a scale of similarity and
difference of sense, so that, for example, a and b might be shown to be
identical in sense (strictly synonymous), a and c relatively similar in
sense (loosely synonymous), a and d less similar in sense, and so on.
Various proposals have been made in recent years for the quantifi-
cation of 'synonymy' along these lines. We shall not discuss any of
these proposals here. Even if it were showrn that one or other measure
of similarity of sense were empirically reliable (applicable by different
scholars at different times and consistent in its results) and succeeded
in bringing together, as more or less * synonymous', items which the
native speaker felt'belonged together', we should still be left with the
problem of accounting for the differences between the 'synonyms'.
(It may be worth pointing out that the practical utility of reference
works such as Rogefs Thesaurus depends upon a prior knowledge of
the language on the part of the person using them. Unless he can
himself distinguish correctly between the hundreds of 'equivalents'
that he is given for nice he can hardly be said to have them 'at his
disposal'.) There is no reason to believe that, if b and c are shown to
be 'equidistant' in sense from a, they are themselves synonymous
and related semantically to a in the same way. Suppose, for example,
that both mother and son were shown to be 'equidistant' from father
according to one of the proposed measures of similarity of sense. How
would we interpret this result? We should clearly not wish to say that
mother and son were even ' loosely' synonymous. The sense-relation-
ship between father and mother is patently, and describably, different
from that which holds between father and son. In short, there is no
obvious way of deriving the various sense-relations which are known
to be of importance in the organization of the vocabulary from a
measure of relative 'synonymy'.
10.2.3 ' Total synonymy^ and ''complete synonymy*

It is a widely-held view that there are few, if any, 'real' synonyms in
natural languages. To quote Ullmann: 'it is almost a truism that total
synonymy is an extremely rare occurrence, a luxury that language can
ill afford.' As argued by Ullmann this view rests upon two quite
distinct criteria: 'Only those words can be described as synonymous
which can replace each other in any given context without the
IO. SEMANTIC STRUCTURE
slightest change either in cognitive or emotive import.' The two
conditions for 'total synonymy* are therefore (i) inter changeability in
all contexts, and (ii) identity in both cognitive and emotive import.
We will discuss the validity of the distinction between ' cognitive' and
'emotive' sense presently. For the moment we may take it for
granted.
The condition of interchangeability in all contexts reflects the
common assumption that words are never synonymous in any
context unless they can occur (and have the same sense) in all
contexts. We have already referred to and rejected this assumption
(9.4.2). Like all sense-relations, synonymy is context-dependent: we
will return to this point. The main objection to the definition of
synonymy proposed by Ullmann (and others) is that it combines two
radically different criteria and prejudges the question of their inter-
dependence. It will be helpful to introduce a terminological distinc-
tion at this point. Granted the validity of a distinction between
'cognitive' and 'emotive' sense, we may use the term complete
synonymy for equivalence of both cognitive and emotive sense; and
we may restrict the term total synonymy to those synonyms (whether
complete or not) which are interchangeable in all contexts. This
scheme of classification allows for four possible kinds of synonymy
(assuming that only two values are attributed to each of the variables):
(1) complete and total synonymy; (2) complete, but not total; (3)
incomplete, but total; (4) incomplete, and not total. It is complete
and total synonymy that most semanticists have in mind when they
talk of 'real' (or 'absolute') synonymy. It is undoubtedly true that
there are very few such synonyms in language. And little purpose is
served by defining a notion of ' absolute' synonymy which is based
on the assumption that complete equivalence and total interchange-
ability are necessarily connected. Once we accept that they are not, and
at the same time abandon the traditional view that synonymy is a
matter of the identity of two independently-determined 'senses', the
whole question becomes much more straightforward.
10.2.4 'Cognitive' and 'emotive' meaning

Many semanticists invoke the distinction between 'cognitive' and
'emotive' (or 'affective') meaning in their discussions of synonymy.
The terms themselves clearly reflect the view that the use of language
involves two or more distinguishable psychological 'faculties'the
intellect, on the one hand, and the imagination and the emotions, on
10.2. SYNONYMY 449
the other. One of the points that is frequently emphasized, both in
technical treatments of semantics and also in the more popular works
on the subject, is the importance of 'emotive' factors in linguistic
behaviour. It is often said that, by contrast with the vocabulary of
scientific and technical discourse, the words of 'everyday language'
are charged with emotional 'associations', or 'connotations', over
and above their primary, purely 'intellectual' meaning.
There is no need to discuss here the psychological validity of the
distinction between the various mental 'faculties' upon which the
semantic distinction of 'cognitive' and 'non-cognitive' meaning was
originally based. The term 'cognitive' meaning is employed by many
scholars who would not necessarily subscribe to the view that the
'intellectual' is sharply distinct from the 'affective'. As far as the
actual use of language is concerned, it is undoubtedly true that one
word may be preferred to another because of its different emotive or
evocative associations. But the extent to which this is of importance
varies considerably from one style or situation to another. For
instance, Ullmann cites as examples of English words which are
cognitively, but not emotively, synonymous liberty: freedom, hide:
conceal. It is not difficult to think of occasions when a speaker or
writer might deliberately use one rather than the other of these
synonyms and make his choice on the basis of these ' connotations'
which the words are likely to evoke. But there are also many contexts
in which either one or the other might be used without any notice-
able difference of effect. It would be wrong to assume that the
emotive connotations of a word are always relevant to its employment.
A more important point is the following. The distinction between
'cognitive' and 'non-cognitive' synonymy is drawn in various ways
by different authors. But in all cases it is 'cognitive' synonymy w7hich
is defined first. No one ever talks of words as being 'emotively', but
not 'cognitively synonymous'. This fact of itself would be sufficient
to suggest that 'emotive', or 'affective', is being used as a catch-all
term to refer to a number of quite distinct factors which may influence
the selection of synonyms on particular occasions or in particular
contexts. What is required is an account of these factors in terms
appropriate to them. No useful purpose is served by employing the
undoubtedly relevant category of 'emotive' (or 'affective') connota-
tions for anything that does not come within the scope of ' cognitive'
meaning.
Some of the factors which influence or determine our choice between
' cognitively' synonymous words and expressions have nothing to do
with sense, reference or anything else that might reasonably be called
'meaning'. Many people deliberately refrain from using the same
word more than once in the same utterance, if they can avoid it.
Others consciously or unconsciously follow the practice of choosing a
shorter word in preference to a longer word, a more 'everyday' wTord
rather than a 'learned' word, an 'Anglo-Saxon' word instead of a
Latin, Greek or Romance word, and so on. In writing verse, the
particular phonological constraints imposed by the metre or rhyme
introduce yet other non-semantic factors.
There are also factors which, though they might well be described
as 'semantic', have to do with the situational or stylistic acceptability
of particular forms rather than with their sense or reference. We have
already seen that there are many 'dimensions' of acceptability that
would need to be accounted for in a complete description of linguistic
behaviour (cf. 4.2.3). We will say no more about these other determi-
nants of full acceptability here, since we are concerned with the more
general principles of semantic structure. It seems preferable to
restrict the term 'synonymy' to what many semanticists have
described as ' cognitive synonymy'. This is the convention that we will
adopt for the remainder of this chapter. As a consequence we shall
have no further use for the distinction between ' complete' and ' in-
complete' synonymy.
10.2.5 Synonymy defined in terms of bilateral implication

Synonymy may be defined in terms of bilateral implication, or
equivalence. If one sentence, Sl9 implies another sentence, 5 2 , and if
the converse also holds, Sx and S2 are equivalent: i.e. if S => 5 2 and
if S2 => Sv then 5X s= S2 (where ' = ' stands for 'is equivalent to'). If
now the two equivalent sentences have the same syntactic structure
and differ from one another only in that where one has a lexical item,
x, the other has jy, then x and y are synonymous. An alternative way of
formulating the definition of equivalence would be as follows: if Sx
and S2 each implies the same set of sentences, then they are equiva-
lent to one another. The difficulty with a definition of this form,
however, is that it falls foul of the principle that the set of sentences
implied by any given sentence is indeterminate (cf. 9.2. .0). If we
define equivalence in terms of bilateral implication, we may assume
that sentences wThich imply one another also imply the same set of
other sentences, unless and until this assumption is proved false in
particular instances.
10.2. SYNONYMY 451
10.2.6 Synonymy and 'normaV inter changeability

In traditional semantics, synonymy has generally been regarded as a
relationship holding between lexical items; and the definition that has
just been given takes this view. It is of course possible to extend the
application of the term 'synonymy* so that it also covers groups of
lexical items that are brought together in a particular syntagmatic
construction, as well as single lexical items. One might well say, for
example, that the phrases female fox and male duck are synonymous
with the lexical items vixen and drake, respectively. But it is impor-
tant to notice that, in making this statement, one is assuming that the
phrases and the lexical items are indeed interchangeable in the normal
use of the language. By contrast, *male cow and bull, and *female bull
and cow, are not normally interchangeable (it may be assumed), even
though one can easily imagine a situation in which the simplest way of
explaining the meaning of bull (to someone who knew the meaning
of cow and male) was by means of the normally unacceptable sentence
A bull is a male cow. The reason why ^female bull and mmale cow are
semantically unacceptable is that neither bull nor cow, unlike fox and
duck, is 'unmarked' for the distinction of sex (cf. 2.3.7, o n dog and
bitch). So much would be undisputed by all semanticists. But the
condition of 'normal' interchangeability is here intended to exclude
many semantically-compatible (significant) groupings of lexical items,
as well as such semantically-incompatible phrases as *male cow. The
phrase mature female bovine animal (which might be given as a
dictionary definition of cow) is undoubtedly well-formed, both gram-
matically and semantically. But it is probably far less 'normal' a
phrase than even the semantically ill-formed *male cow. The native
speaker of English would not 'normally' construct such a phrase as
mature female bovine animal and use it interchangeably with cow in
his everyday use of the language. The question of synonymy therefore
does not arise in the case of the lexical item cow and the phrase
mature female bovine animal. Alternatively, one might say that the
most interesting question that arises in instances of this kind is not
whether the relationship of synonymy holds, or how to account for it
if it does, but why it is that a lexical item like cow and a phrase like
mature female bovine animal are not in fact freely interchangeable.
Many semanticists have failed to see the importance of this question.
We will return to it later in connexion with 'ccmponcntial analysis'
(cf. 10.5.5).
10.2.7 Context-dependent synonymy

One final point may be made about synonymy: more than any other
sense-relation, it is context-dependent, and in a theoretically interesting
way. It is evident that it is not of itself a structural relationship. All
instances of synonymy could be eliminated from the vocabulary with-
out affecting the sense of the remainder of the lexical items. The * im-
poverished' vocabulary would offer fewer opportunities for stylistic
variety, but everything that could be said with the larger vocabulary
could also be said with the smaller synonymy-free vocabulary.
Although synonymy is not essential to the semantic structure of
language, it arises in particular contexts as a consequence of the more
fundamental structural relations, hyponymy and incompatibility (to
be discussed in the following section). It frequently happens that the
distinction between two lexical items is contextually neutralized. For
instance, the difference between the marked term bitch and the un-
marked term dog is neutralized in a context, like My has just had
pups, which determines the animal referred to as female. All sense-
relations are in principle context-dependent, but contextually-deter-
mined synonymy is of particular importance. It is clear that it can
be brought within the scope of the general principle that the same
information may be conveyed in language either syntagmatically or
paradigmatically (cf. 2.3.8). One can say either Fm flying to New York
or Fm going to New York by air, either Fm driving to New York or
Fm going to New York by car. In the one case the distinction is made
by the paradigmatic choice of the verbs fly and drive, in the other by
the syntagmatic modification of the more general verb go. If a
particular lexical item is very frequently modified syntagmatically in a
particular way, this may have the effect, diachronically, of transferring
the distinction from the syntagmatic to the paradigmatic and making
the overt syntagmatic modification redundant. It is presumably this
phenomenon which accounts for the development in the sense of the
verb starve. At one time, it meant something like 'die' (cf. the
genetically-related German sterben) and earlier 'to be stiff', but it was
commonly modified syntagmatically with of hunger and thus acquired
the sense it now possesses in modern standard English. (In certain
areas of Northern England the typical syntagmatic modification was,
and still is, with cold, so that Fm starving is roughly equivalent to the
standard English Fm freezing.) The history of the vocabulary of
English, and doubtless of all languages, is full of examples of semantic
' specialization' of this kind.
10.2. SYNONYMY 453
It is important to realize that the contextual determination of a
lexical item may be probabilistic rather than absolute. For instance,
the substitution of buy for get in Til go to the shop and get some bread
would not generally be held to introduce any additional implications:
buy and get would normally be taken as synonymous in this context.
The standard conventions and presuppositions of the society are such
that, unless there is some evidence to suggest the contrary, it will be
assumed that what is obtained from a shop is obtained by purchase.
At the same time, it must be admitted that get is not necessarily
synonymous with buy (even with the syntagmatic support of from the
shop). The example also illustrates the further point that there is no
sharp distinction to be drawn between the probabilistic determination
of synonymy by other lexical items in the same utterance and the
determination of synonymy by the features of the situation in which
the utterance occurs. If one says Pmjust going to get some bread as one
steps into a shop, the context-dependent synonymy of get and buy is no
weaker than it would be if the words from the shop occurred in the utter-
ance. Not only is it no weaker, it is no different in kind, since the same
set of cultural presuppositions determine the implications in both cases.
10.3 Hyponymy and incompatibility

10.3,1 Hyponymy
Hyponymy and incompatibility are the most fundamental para-
digmatic relations of sense in terms of which the vocabulary is
structured. Although they are very largely interdependent, we shall for
convenience discuss them separately.
The term 'hyponymy' is not part of the traditional stock-in-trade
of the semanticist; it is of recent creation, by analogy with' synonymy'
and 'antonymy'. Although the term may be new, the notion of
hyponymy is traditional enough; and it has long been recognized as
one of the constitutive principles in the organization of the vocabulary
of all languages. It is frequently referred to as * inclusion*. For
example, the * meaning' of scarlet is said to be 'included* in the
'meaning' of red; the 'meaning* of tulip is said to be 'included* in
the 'meaning* of flower; and so on.
This relationship, the 'inclusion* of a more specific term in a more
general term, has been formalized by certain semanticists in terms of
the logic of classes: the class of entities referred to by the word flower
is wider than and includes the class of entities referred to by the word
tulip; the class of entities that may be truly described as scarlet is
I0
included in the class of entities that may be truly described as red;
and so on. It will be observed that this formulation of the relationship
of * inclusion' rests upon the notion of reference (since it operates with
classes of 'entities' which are named by lexical items). One reason for
preferring to introduce the new technical term 'hyponymy* is simply
that it leaves 'inclusion* free for the theory of reference and its
formalization in terms of class-logic. We have already seen that it is
desirable to draw a theoretical distinction between sense and
reference. It is important to realize that hyponymy, as a relation of
sense which holds between lexical items, applies to non-referring
terms in precisely the same way as it applies to terms that have
reference.
A more important reason for preferring to use an alternative to
'inclusion* is that 'inclusion* is somewhat ambiguous. From one
point of view, a more general term is more 'inclusive* than a more
specific termflower is more inclusive than tulipsince it refers to a
wider class of things. But from another point of view, the more
specific term is more ' inclusive*tulip is more inclusive than flower
since it carries more 'bits' of information, more 'components* of
'meaning* (cf. 2.4.3, 10.5.1). The difference in the point of view
from which one may consider' inclusion * corresponds to the difference,
in traditional logic and in certain theories of semantics, between the
extension and the intension of a term. The extension of a term is the
class of entities to which the term is applicable or refers; the intension
of a term is the set of attributes which characterize any entity to
which the term is correctly applied. Extension and intension vary
inversely in relation to one another: the greater the extension of a
term, the less its intension; and conversely. For example, the extension
oi flower is greater than that of tulip, since the former term refers to
more things; on the other hand, the intension of tulip is greater than
that of flower, since the characterization or definition of tulips must
make reference to a wider set of attributes than those which suffice
to characterize flowers. It may be mentioned in passing that certain
semanticists, notably Carnap, have attempted to draw the distinction
between sense and reference in terms of the logical distinction
between intension and extension. We have taken the view that the
difference between sense and reference is of a quite different order
(cf. 9.2.2, 9.4.1, 9.4.2). Confusion is avoided by the employment of a
neutral, non-metaphorical term like 'hyponymy*. We will say that
scarlet, crimson, vermilion, etc., are co-hyponyms of red, and tulip,
violet, rose, etc. co-hyponyms oi flower. Conversely, we will say that
IO.3. HYPONYMY AND INCOMPATIBILITY 455
red is super ordinate with respect to its hyponyms (the more obvious
Greek-based term 'hyperonym' is not sufficiently distinct acousti-
cally from 'hyponym' in English).
Hyponymy may be denned in terms of unilateral implication. (For
instance, X is scarlet will be taken to imply X is red; but the converse
implication does not generally hold.) In the most typical instances, a
sentence containing a superordinate term will imply either (i) the
disjunction of sentences each containing a different member of a set
of co-hyponyms, or (ii) a sentence in which the co-hyponyms are
semantically ' co-ordinated', as it were. Both of these possibilities may
be illustrated with / bought someflowers.This sentence might imply
the disjunction of I bought some tulips, I bought some roses, I bought
some violets, etc. (By 'disjunction* in this context is meant the choice
of one from a set of alternatives: if p implies the disjunction of q,
r and s, then/) implies either q or r or s.) It might also imply a sentence
like / bought some roses and some tulips, or / bought some violets and some
tulips, etc. It is of course one of the most convenient features of the
principle of hyponymy that it is enables us to be more general or
more specific according to circumstances. It would be quite in-
appropriate to say that someflowersis either imprecise or ambiguous (as
between 'some roses', 'some tulips', etc., on the one hand, and 'some
roses and some tulips', 'a rose and some tulips', etc. on the other).
10.3.2 Synonymy as symmetrical hyponymy

Although a superordinate term does not generally imply its hyponym,
it is frequently the case that the situational context or the syntagmatic
modification of the superordinate term will determine it in the sense
of one of its hyponyms. This is the source of context-dependent
synonymy (cf. 10.2.7). ^ ^ ** suggests the possibility of defining the
relationship of synonymy as symmetrical hyponymy: if x is a
hyponym of y and if y is also a hyponym of x (i.e. if the relationship is
bilateral, or symmetrical), then x and y are synonyms. Drawing upon
the terminological distinction made in set-theory and the logic of
classes, we may refer to the relationship of unilateral, or asymmetrical,
implication which holds between tulip zxAfloweras proper hyponymy.
All hyponymy is transitive, in the sense that if the relation holds
between a and b and also between b and c, then it also holds between
a and c. Synonymy, as a special case of hyponymy, has therefore the
additional property that it is a symmetrical relation (it holds between
a and b and between b and a). And for purely formal reasons it may
45^ IO. SEMANTIC STRUCTURE
be defined also as reflexive: every lexical item is substitutable for, and
is synonymous with, itself in the same context. (Synonymy is there-
fore an equivalence-relation in the mathematical sense of this term.)
10.3.3 Absence of superordinate terms

The main point to be made about the relation of hyponymy as it is
found in natural languages is that it does not operate as compre-
hensively or as systematically there as it does in the various systems of
scientific taxonomy (in botany, zoology, etc.). The vocabularies of
natural languages tend to have many gaps, asymmetries and indeter-
minacies in them. For instance, there is no superordinate term in
English of which all the colour-words are co-hyponyms. (Logicians
frequently cite as an example of analytic implication If it is red, then
it is coloured. But this implication does not in fact generally hold for
all colour-terms in normal English usage. The adjective coloured
is in contrast with white in certain contextsin sorting out the
laundry, in the classification of people according to their race, etc.
and with transparent in others: e.g. There was some coloured liquid in
the bottleone might also wonder whether coloured is in contrast with
white, as well as with transparent, in contexts of this kind.) Similarly,
there is no more general adjective of which square and round are
co-hyponyms. On the other hand, there are many words that are
commonly regarded as lexical items whose application is so general
that they might well be treated as grammatical * dummies * in ' deep'
syntactic analysis: e.g. come/go, person, thing, event, etc. At this point,
there is a high degree of correspondence between syntax and semantics
(cf. 9.5.2).
10.3.4 Hierarchical structure of vocabulary

Many semanticists have been attracted by the possibility of describing
the vocabulary of the language in terms of a hierarchical, taxonomic
classification working from the most general to the more specific
categories. We have already mentioned Rogefs Thesaurus, the most
famous attempt to analyse the vocabulary of English in this way; and
we shall return to the question of the hierarchical structure of the
vocabulary in the section devoted to the principles of ' componential
analysis* (cf. 10.5.1; also 4.3.3).
The most important factor in the hierarchical organization of the
vocabulary by means of the relation of hyponymy is the structure of
the culture in which the language operates and in which it serves as
the principal medium for communication. It is a truism that words
referring to artefacts cannot be defined except in relation to the
purpose or normal function of the objects they refer to: e.g. school,
'a building where children are taught', house, 'a building where
people live'. But this is true of the vocabulary as a whole, which is not
only * anthropocentric' (organized according to general human
interests and values), but 'culture-bound* (reflecting the more
particular institutions and practices of different cultures). Part of
what I have referred to as the semantic anisomorphism of different
languages (cf. 2.2.1) is accounted for by the fact that individual
languages vary considerably in the extension of 'roughly equivalent*
terms. It is often possible to identify (in terms of their application:
cf. 9.4.8) the hyponyms of a certain term in one language with
lexical items in another language without being able to find an
equivalent for the superordinate term. As an instance of this pheno-
menon, we may consider the word demiourgds in Greek.
Among the hyponyms of demiourgos (which is usually translated as
'craftsman', 'artisan') we find a large number of terms, including
tdkton, iatros, auletes, skutotomos, kubernttes. For each of these there is
a satisfactory English equivalent for the purpose of translating the
works of the classical authors: 'carpenter', 'doctor', 'flute-player',
'shoemaker', 'helmsman'. But there exists no word in English that is
superordinate to all the translation-equivalents of demiourgos without
being also superordinate to other words that are not translation-
equivalents of demiourgos. The distinction between arts, crafts, trades,
professions, and so on, is not relevant to the meaning of demiourgos.
Anyone who had a culturally-recognized occupation which required
specialized knowledge or training was a demiourgos. The meaning of
this word can only be described in terms of its hyponyms and in
terms of the sense-relations it bears to other words in Greek (in
particular, to the verb epistasthai, ' to know (as a result of study or
training)'). In fact, the translation of many of its hyponyms rests
implicitly upon the decision to treat certain classes of people and their
'professional' activities as culturally-equivalent. We identify the
application of the English word doctor and the Greek word iatros by
virtue of our decision to treat the cultural, or social, function of those
denoted by these words as equivalent; and this decision involves the
tacit recognition that many of the activities characteristic of the
'doctor' and the 'iatros' are culture-bound and irrelevant to what we
regard as their 'culture-invariant' function. All translation from one
IO. SEMANTIC STRUCTURE
language to another involves decisions of this kind. It is a sound
methodological principle that sense is not held invariant in trans-
lation (so that there is no synonymy between words of different
languages) but a greater or less degree of equivalence in the * appli-
cation* of words. And at the present time semantic theory can do
little more than appeal to the bilingual speaker for intuitive judge-
ments of equivalence in the area of 'cultural overlap' (cf. 9.4.7).
10.3.5 Incompatibility
Incompatibility can be defined on the basis of the relationship of
contradictoriness between sentences. If one sentence, Sly explicitly or
implicitly denies another sentence, S2, then Sx and S2 are contra-
dictory (Sx and S2 are explicitly contradictory if S negates S2
syntactically, otherwise they are implicitly contradictory: cf. 10.1.2).
If S2 and Sx are implicitly contradictory sentences of identical deep-
syntactic structure, and if they differ only in that where one has the
lexical item x the other has y, then x and y are incompatible.
To take a simple, and familiar, example from the colour-terms in
English. If someone says Mary was wearing a red hat, this will be
understood as implicitly denying Mary was wearing a green (blue
white, yellozv, etc.) hat. And the substitution of any one of the terms
in the set green, blue, white, yellow, etc., for red would also be taken as
implying the denial of Mary was wearing a red hat. The colour-terms
therefore form a set of incompatible lexical items.
This is obvious enough. What has not always been quite so clear to
semanticists is the fact that the incompatibility of red, green, etc., is
not a secondary consequence of the sense which each of them has
(independently as it were) but is necessarily involved in learning and
knowing the sense of each of the terms in the set. As we have already
seen, the colour-terms taken together exhaust a referential continuum;
and learning where to draw the boundaries within the continuum for
a particular term, say blue, is dependent upon the knowledge that on
either side of the boundary is 'not blue' (cf. 9.4.5). In principle, it is
perhaps conceivable that the reference of one of the colours could be
learned without knowing the items referring to the areas of the
continuum beyond the boundaries of 'blue' (i.e. by contrasting blue
explicitly with not.. .blue). One could conceive of the language being
learned in an environment which did not provide instances of colour
at all 'points' in the continuum. But in practice, one may assume, the
reference and sense of the most common colour-terms is learned more
or less simultaneously, with continual adjustment of the boundaries
until they approximate to the norm for the speech-community. Further
lexical differentiation is then possible on the basis of hyponymy,
red being 'subdivided* into crimson, scarlet, and so on. But the further
differentiation will vary considerably as between individual speakers.
Those whose profession or interests require them to draw more
numerous distinctions of colour will develop a very rich colour-
terminology. But they will do so subsequently to the acquisition of the
'grosser* distinctions characteristic of the non-specialized vocabu-
lary of the community as a whole.
10.3.6 Incompatibility and difference of sense

Incompatibility is to be distinguished from mere difference of sense.
This is particularly clear in the case of incompatible co-hyponyms of a
superordinate term, which are different within some 'dimension* of
similarity of sense. For instance, crimson and soft are different in
sense, but not incompatible: both adjectives may be applied to the
same object without contradiction. On the other hand, crimson and
scarlet are similar in sense (their similarity being stateable as co-
hyponymy with respect to red), but incompatible. The 'higher-level'
incompatible terms red, green, blue, are also similar in sense, although
there is no superordinate term of which they are co-hyponyms.
The distinction between incompatibility and difference of sense is
less clear in other instances; notably in the case of words which denote
physical 'objects* (whether 'natural* or manufactured). The words
chair and table are incompatible (we will neglect the theoretically-
uninteresting complications introduced by the consideration of dual-
purpose furniture), but we might be inclined to say, and no doubt
correctly, that the meaning of the one could be learned independently
of the other. Of course, we would not say that anyone knew the
meaning of table if he used the word to refer to objects which other
speakers of English described as 'chairs*. The question is whether
there is any 'dimension* of sameness prior to the distinction of the
two incompatible terms. The same question can be put in relation to
the words door and window. In the case of table and chair there is the
superordinate term furniture; there is no such term which brings
together door and window. But the existence or non-existence of a
superordinate seems to be of relatively small importance here. And
when we consider such pairs of words as chair and cow (or a host of
other lexical itemsto use Lewis Carroll's example, shoe, ship,
sealing wax or cabbage and king) which semantically have nothing in
common other than the fact that they denote physical entities, there
is little point in distinguishing between incompatibility and difference
of sense. It is in the case of sets of lexical items which give structure to
a continuum that the relation of incompatibility is of crucial im-
portance in both the learning and the U9e of language. And it would
be a mistake to think that the distinction between incompatibility
and mere difference does not apply at all to the lexical classification
of words which denote persons, animals and physical objects. One
has only to think of such sets as tree, shrub, bush, etc., to see that the
distinction is of importance here also.
One final point should be made in connexion with the notions of
hyponymy and incompatibility. We have repeatedly stressed the
principle that the same semantic distinctions can be made either para-
digmatically or syntagmatically. To give yet another example,
English draws a paradigmatic distinction between brother and sister.
Turkish does not: the word karde is ' unmarked * with respect to the
distinction of sex, but may be * marked' by syntagmatic modification
if one wishes to make clear the sex of the person referred to:
kizkardes, * sister' (' girl-brother', as it were). Other languages make
a paradigmatic distinction between 'eldest son', 'younger son', etc.
10*4 Antonymy, complementarity and converseness

10.4.1 ' Oppositeness' of meaning
Antonymy, or 'oppositeness of meaning', has long been recognized
as one of the most important semantic relations. However, it has been
the subject of a good deal of confusion, partly because it has generally
been regarded as complementary to synonymy and partly because
most semanticists have failed to give sufficient attention to different
kinds of 'oppositeness'. Synonymy and antonymy, as we shall see,
are sense-relations of a very different kind. For simplicity, we will
distinguish terminologically between three types of 'oppositeness';
and we will reserve the term' antonymy' for just one of the three types.
A fuller treatment of ' opposites' would draw more distinctions than
we have space for here.
10.4.2 Complementarity
The first relation of' oppositeness' to be discussed is that which holds
between such pairs of words as single: married, male: female, etc. We
IO.4. ANTONYMY 461
will use the term complementarity for this, saying that single and
married, or male and female, are complementaries. It is characteristic
of such pairs of lexical items that the denial of the one implies the
assertion of the other and the assertion of the one implies the denial
of the other: ~x => y andjy ==> ~x. Thus, John isn't married implies
John is single \ and John is married implies John is not single. In the
case of those pairs for which we are reserving the term * antonymy'
(e.g. good: bad, high: low), only the second of these implications holds:
y z> ^x. John is good implies the denial of John is bad; but John is
not good does not imply the assertion of John is bad.
Complementarity may be regarded as a special case of incompati-
bility holding over two-term sets. The assertion of one member of a
set of incompatible terms implies the denial of each of the other
members in the set taken separately (red implies ~blue, ~ green,
etc.); and the denial of one member of a set of incompatible terms
implies the assertion of the disjunction of all the other members
(~red implies either green or blue or...). In a two-term set of in-
compatible terms, there is only one other member. Conjunction and
disjunction therefore fall together: * both y and z' and * either y or z'
amount to the same thing \iy and z have the same value. And from
this fact there follow the particular conditions of complementarity
mentioned above. It would be erroneous, however, to suppose that
complementarity is merely the limiting case of incompatibility with
the set of incompatible terms reduced, accidentally as it were, to two.
Dichotomization is a very important principle in the semantic
structure of language. We will take up this point below.
Everything that has been said so far about complementarity and
the implications between lexical items which determine thi9 relation
presupposes the applicability of the complementary terms. The use
of the dichotomous terms married and single presupposes the ap-
plicability of whatever might be the culturally accepted criteria of
' marriageability \ John isn't married is hardly less anomalous semanti-
cally than The stone isn't married, if the person referred to as John is
not in fact 'marriageable' (by virtue of age and other criteria).
A further point should be noticed in connexion with comple-
mentary terms. Although it is normally the case that the denial of the
one implies the assertion of the other and the assertion of the one
implies the the denial of the other, it is generally possible to ' cancel'
either or both of these implications. But this fact should not be
taken as sufficient to invalidate the normal usage of complementary
terms. The point may be made more clearly perhaps by taking the
complementaries male and female as illustrative of the general prin-
ciple of * normality* as it is intended to be understood here. Granted the
applicability of the distinction of sex, there is a first-level, normal
dichotomy into male and female; and this dichotomy reflects the
assumption that a number of different biological and behavioural
characteristics will 'normally* be associated in the same person or
animal. There are, however, many cases where the dichotomous
classification is unsatisfactory either biologically or behaviourally,
and then the terms hermaphrodite or homosexual are available to take
account of these 'abnormalities'. Most of the complementary terms
in the everyday vocabulary of languages would seem to operate in the
same way within the framework of the relevant presuppositions,
beliefs and conventions subsumed under the notion of 'restricted
context* (cf. 9.3.9). As Moravcsik has pointed out, in a paper devoted
to the discussion of the philosophical distinction of the analytic and
the synthetic, it is not difficult to think of circumstances in which one
might wish to assert of the same person that he was both a bachelor
and married (or neither single nor married). This situation might arise,
if the person in question was not in fact married according to the law
and customs of the society, but nevertheless lived and behaved in a
way characteristic of people to whom the term married is applied
'normally* (living regularly with one woman, having children by her
and maintaining a home, etc.). The fact that it is possible to 'cancel'
some of the implications of the first-level dichotomous classification
means that in such cases the implications can only be regarded as
'normally*, and not 'absolutely*, analytic. But this principle holds for
sense-relations in general.
Not only is it possible to conceive of situations in which the
assertion of one term does not necessarily imply the denial of its
complementary, but it is also possible to qualify a complementary
term, 'abnormally*, with more or less. One can say, for instance, that
one person is more married than someone else (implying that his
behaviour is more typical of what is 'normally* characteristic of
married men). This is perhaps unusual, but it is a possibility which
semantic theory should allow for. What is involved is the qualifi-
cation of one or more of the presuppositions which determine the
'normal* interpretation of the term in question. However, in their
'normal* usage complementary terms are not qualifiable, or gradable,
in this way.
IO.4. ANTONYMY 463
10.4.3 Antonymy
The relation to which we are giving the name antonymy (to the
exclusion of other kinds of 'oppositeness') may be exemplified by the
terms big and small in English. It is characteristic of antonyms of this
class, 'opposites' par excellence, that they are regularly gradable.
Grading (in the sense in which the term is being employed hereit is
borrowed from Sapir, to whom we shall refer presently) is bound up
with the operation of comparison. The comparison may be explicit or
implicit. Explicitly comparative sentences fall into two types, (i) Two
things may be compared with respect to a particular * property', and
this i property' predicated of the one in a greater degree than it is of
the other: e.g. Our house is bigger than yours. (2) Two States' of the
same thing may be compared with respect to the * property* in
question: e.g. Our house is bigger than it used to be. Actual utterances
(taken out of context) may be ambiguous as between the two types of
explicit comparison: e.g. Our house is bigger, which is presumably
derived from a sentence of either one type or the other by the deletion
of the phrase or clause introduced by than. But they are still explicitly
comparative, and can only be interpreted if the other term of the
comparison is recoverable from the context.
Both types of explicit comparison may be combined in the same
sentence: e.g. Our house is bigger than yours used to be, He is taller than
his father was. But the semantic interpretation of these more complex
comparative sentences does not seem to introduce any additional
problems. In fact, each of the two simpler types of explicit comparison
may be subsumed under a more general formula which also covers the
more complex sentences:
Comp {([NP,, xjT. + M^ + A J ([NP* *]T,. + M, + An)}
In this formula, iNPi stands for * noun-phrase' (denoting the thing
or things being compared), 'x' stands for the particular lexical item
which is graded (in English, this is generally realized with the suffix
~er, e.g. bigger, or with the word more preceding the uninflected
adjective, e.g. more beautiful), *T' stands for * tense', ' M ' for 'mood'
and 'A' for 'aspect'. The subscripts distinguish the different values
which may be assumed by the noun-phrase and by the markers of
tense, mood and aspect. In terms of the formula, the sentence Our
house is bigger than yours used to be might be analysed as:
Comp {{[Our house, big]Tn(m.past
0 + \ ) ([Your house, big]Tpasi + M0 + Ahabi(ual)}
464 IO. SEMANTIC STRUCTURE
This analysis is not definitive, but merely illustrative of the variable
factors involved. As we have already seen, the analysis of tense, mood
and aspect in English is a complicated matter (cf. 7.5.8). For simplicity
of exposition, we have shown tense, mood and aspect as independent
variables in the formula: from the syntactic point of view this is quite
unsatisfactory, but it does not affect the point being made here. The
subscript o ('zero*) indicates the 'unmarked* term in a category;
the other subscripts are self-explanatory. The corresponding non-
comparative sentences are Our house is big (modally and aspectually
unmarked and non-past) and Your house used to be big (modally un-
marked, habitual in aspect and past). The reason why the semantic
analysis of the comparative sentence Our house is bigger than yours used
to be does not proceed by way of a prior semantic analysis of the
syntactically-embedded sentences Our house is big and Your house is
big will occupy us presently.
In the case of our model sentence, Our house is bigger than yours
used to bey the two noun-phrases are different (NP1 does not equal
NP2) and so are the values of T and A (Tt * T, and Am =J= An).
The two simpler types of explicit comparison can be derived from
the formula by imposing a condition of identity either between NPX
and NP2 or between i and j \ k and /, and m and n. In Our house is
bigger than yours the second, but not the first, identity holds (i = j \
k = I and m = n, but NPX 4= NP2). In Our house is bigger than it
used to be it is the other way round (NP1 = NP2; but, although k = /,
i 4= j and m 4= n). If both identities hold simultaneously the result
is of course a contradictory sentence: Our house is bigger than it is.
Given this formal framework, we can state the most important
defining characteristic of the relation of antonymy. If x and y are
antonyms, then a comparative sentence containing x of the form
(i) Comp^NP,, x]Ti + Mk + Am) ([NP2, *]T, + M! + An)}
both implies and is implied by a corresponding comparative sentence
containing y:
(ii) Comp {([NP2, y] T, + M, + An) ([NPV y] T, + Mk + A J }
To exemplify: Our house is bigger than yours used to be both implies and
is implied by Your house used to be smaller than ours is; Our house is
bigger than yours implies and is implied by Your house is smaller than
ours; and Our house is bigger than it used to be implies and is implied
by Our house used to be smaller than it is {now). The English words big
and small are therefore antonyms in a range of contexts illustrated by
these sentences.
IO.4. ANTONYMY 465
10.4.4 * Implicitly graded' antonyms

We may now consider sentences in which antonyms are not explicitly
graded. First of all, it may be observed that the denial of the one does
not imply assertion of the other. Our house is not big does not imply
Our house is small (although Our house is big does imply Our house is
not small). This is a fact well-known to logicians; and it distinguishes
antonyms from complementaries. More important, however, is the
fact that sentences containing antonyms are always implicitly, if not
explicitly, comparative. This was pointed out many years ago by
Sapir, in a passage that deserves to be quoted in full:
* Such contrasts as small and large, little and much, few and many,
give us a deceptive feeling of absolute values within the field of
quantity comparable to such qualitative differences as red and green
within the field of color perception. This feeling is an illusion,
however, which is largely due to the linguistic fact that the grading
which is implicit in these terms is not formally indicated, whereas it
is made explicit in such judgements as " There were fewer people
there than here" or "He has more milk than I". In other words,
many, to take but one example, embodies no class of judgements
clustering about a given quantity norm applicable to every type of
experience, in the sense in which red or green is applicable to every
experience in which color can have a place, but is, properly speaking,
a purely relative term which loses all significance when deprived of its
connotation of "more than" and "less than". Many merely means
any number taken as a point of departure. This point of departure
obviously varies enormously according to context/ Sapir goes on to
observe, later in the same article: 'contrasting qualities are felt as of
a relatively absolute nature, so to speak, and good and bad, for
instance, even far and near, have as true a psychological specificity as
green and yellow. Hence the logical norm between them is not felt as
a true norm but as a blend area in which qualities graded in opposite
directions meet. To the naive, every person is either good or bad; if
he cannot be easily placed, he is rather part good and part bad than
just humanly normal or neither good nor bad.'
The importance of this insight into the nature of antonyms should
not be underestimated. Many pseudo-problems have arisen in logic
and philosophy as a consequence of the failure to appreciate that such
words as big and small, or good and bad, do not refer to independent,
* opposite' qualities, but are merely lexical devices for grading as
'more than' or 'less than* with respect to some implicit norm. Plato
was troubled, for instance, by the fact that, if one asserted of X that
he was * taller than* Y but Shorter than' Z, one appeared to be
committed to the simultaneous predication of the two 'opposite'
qualities ' tallness' and ' shortness' of the same personthat X was
both tall and short. A similar pseudo-problem is exemplified by
sentences such as A small elephant is a large animal If small and large
are regarded as merely incompatible, or complementary, terms, this
sentence should be contradictory (cf. *A male elephant is a female
animal). But it is not; and, however we choose to formalize the rules or
principles of semantic interpretation, what should be formalized by
the rules is quite clear. The implicit' size-norm' for elephants is not
necessarily the same as the implicit' size-norm' for animals taken as a
whole class. The semantic analysis of A small elephant is a large animal
should take something like the following form: 'An elephant which
is small-rather-than-large by comparison with the norm relevant for
elephants is (nevertheless) large-rather-than-small by comparison
with the norm relevant for animals'.
It is because explicitly ungraded antonyms are understood as
implicitly graded with reference to some relevant norm that a
comparative sentence such as Our house is bigger than yours (or Our
house is bigger than yours used to be) cannot be satisfactorily analysed,
from the semantic point of view, on the basis of the analysis of the
syntactically-embedded sentences Our house is big and Your house is
(or used to be) big. A sentence like Our house is big is, semantically, a
comparative: 'Our house is bigger than the normal house.'
The implicit grading of antonyms also accounts for the fact that
there is no contrast between the two members of a particular pair in
'unmarked' questions (and in various other syntactic functions). For
instance, the sentence How big is it? does not presuppose that the
object of the inquiry will be classed as * big' rather than ' small', but is
completely open, or' unmarked', as to the expectations of the inquirer.
It may be regarded as equivalent to ' Is it big or small?'. The question
brings into the discussion a scale recognized by the participants as
relevant and asks that the object be measured, as it were, along this
scale. The first-level measurement is in terms of the dichotomy ' big-
rather-than-smair or 'small-rather-than-big' (by comparison with
the norm). If thefirst-leveldescription as big or small is not sufficiently
precise for the purpose, it is always possible to put the further,
'marked' questions How big is it? or How small is it? (which differ in
stress and intonation from the * unmarked' How big is it?this
IO.4. ANTONYMY 467
difference is summarized, for the present purpose, in the acute accent
on the word how? in the * marked' questions). The * marked'
questions How big is it? and How small is it? carry with them the pre-
supposition that the object in question has already been placed towards
one end of the scale rather than the other, and seek further specification
of the place of the object on the scale relative to the relevant 'size-
norm*.
The opposition between antonyms is 'neutralized', not only in
'unmarked' questions of the kind illustrated in the previous para-
graph, but also in various nominalizations: What is the width of the
river? Everything depends upon the height, etc. The nouns narrowness
and lowness would not occur in such contexts. In general, only one of
a pair of antonyms will occur in ' unmarked' contexts {big, high, wide,
good, tall, etc.); and it is worth observing that many of the nominal-
izations of these ' unmarked' forms are irregular in English (cf. big:
size, high: height, wide: width, etc.) by contrast with the less frequent
'marked' forms (small: smallness, low: lowness, narrow: narrowness,
etc.). The fact that the distinction between antonyms is neutralized
in certain syntactic positions contributes, no doubt, to our feeling
that one antonym has a 'positive', and the other a 'negative',
polarity. We tend to say that small things 'lack size', rather than
large things 'lack smallness'. And, in general, the 'unmarked'
antonym is used for what is felt as 'more than', rather than 'less
than', the norm.
10.4.5 Converseness
The third sense-relation which is frequently described in terms of
' oppositeness' is that which holds between buy and sell or husband
and wife. We will use the term converseness to refer to this relation.
The word buy is the converse of sell, and sell is the converse oibuy.
Although antonymy and converseness must be distinguished, there
is a parallelism between the two relations. As NPX bought NP^from
NP2 implies, and is implied by, NP2 sold NPS to NPt so NP1 is bigger
than NP2 implies, and is implied by, NP2 is smaller than NPV In
both cases the lexical substitution of one term for the corresponding
antonym or converse is associated with a syntactic transformation
which permutes the noun-phrase, NPX and NP2, and also carries out
certain other' automatic' changes in the selection of the appropriate pre-
position (or case-inflexion, in other languages). It may be observed that
this ' permutational' feature is also characteristic of the relationship
between corresponding active and passive sentences: NPX killed
NP2 implies, and is implied by, NP2 was killed by NPV In English
it is possible to form passive sentences in which the 'surface* subject
is identical with the 'indirect object' of the corresponding active
sentence. John's father gave him a book is related semantically to both
(i) John was given a book by his fatherp, and (ii) John received a book from
his father. In many languages (including French, German, Russian,
Latin, etc.) the 'indirect object' cannot be transformed into the
' surface' subject of a passive sentence in this way; and John was given
a book by his father would be translated (to use French for exempli-
fication) as either Le pere de Jean lui a donne un livre ('John's father
gave him a book') or Jean a re$u un livre de son pere ('John received
a book from his father').
The consideration of the verbs for 'marry' in various Indo-
European languages is illuminating from the point of view of the
relation of converseness. (When we say that all these verbs are
'equivalent in meaning' we are, of course, invoking the notion of
'application' and 'cultural overlap': cf. 9.4.8. It is only 'rough'
equivalence anyway, as we shall see.) The English verb marry is
symmetrical, or reciprocal, in that NPX married NP2 implies, and is
implied by, NP2 married NPV (We are not here talking of the transi-
tive, or 'causative', verb exemplified in The priest married them and
They were married by the priest, but of the verb that occurs in such
sentences as John married Jane ox Jane married John.) In a number of
languages, including Latin and Russian, there are two lexically-
distinct, converse, verbs or verb-phrases. In Latin, for instance,
nubere is used if the subject of the (active) sentence is a woman and
in matrimonium ducere (' to bring into wedlock') if the subject is a man.
In Greek, the active of the verb gamdn is employed for the man and
the middle (or occasionally the passive) of the same verb for the
woman: it is as if one were to say in English John married Jane but
Jane got herself married to John ('middle') or Jane was married by
John (passive). These three possibilities illustrate the way in which
'the same relationship' between two persons or things may be
expressed by means of a symmetrical 'predicator' (like marry), by
lexically-distinct ' predicators' (like nubere and in matrimonium
ducere) or by the ' grammaticalization' of the asymmetry according to
the syntactic resources of the language (as with gamein).
The vocabulary of kinship and social status provides many instances
both of symmetry and of converseness. NPX is NP2s cousin implies,
and is implied by, NP2 is NPx's cousin, but TVPj is NP2s husband
IO.4- ANTONYMY 469
implies, and is implied by, NP2 is NP^s wife. Converseness also
intersects with complementarity (of sex), so that NP is NP2s father
implies either NP2 is NP^s son or NP2 is NPfs daughter, NPX is NP2s
niece implies either NP2 is NPxys uncle or AT 2 is NP^s aunt, and
so on.
Other lexical items are * permutationally' related in the same way
as converse terms, although they do not imply one another. For
example, NPX asked (NP2)... 'expects', rather than implies, NP2
answered (NP X )...; and NP2 answered (NPJ... 'presupposes9 NP
asked (NP2).... Similarly, NPX offeredNPZ to NP2'expects* the dis-
junction of the complementary sentences NP2 accepted NP3 and NP2
refused NPS. 'Expectancy' and 'presupposition' of this kind are
ordered with respect to temporal sequence: this is not so, it should be
noted, in the case of such converse terms as give and receive.
10.4.6 A parallelism between antonymy and complementarity

We have noted the parallelism between converse terms and explicitly
graded antonyms (and the purely grammatical transformation by
which active and passive sentences are related). It is no less important
to stress the parallelism between antonymy and complementarity.
They are alike in that the assertion of a sentence containing an
antonymous or complementary term implies the denial of a corres-
ponding sentence containing the other antonym or complementary.
This being so, one might envisage the elimination from the vocabu-
lary of all instances of both antonyms and complementarity. Instead
of John is single, one could say, equivalently, 'John is not married';
and instead of The house is small and The house is big, ' The house is
less big' and 'The house is more big' ('than the norm' being under-
stood). The fact is that we do not; and this, as Sapir pointed out in
the article referred to above, is one of the facts which 'so often
renders a purely logical analysis of speech insufficient or even
misleading'.
The existence of large numbers of antonyms and complementary
terms in the vocabulary of natural languages would seem to be
related to a general human tendency to 'polarize' experience and
judgementto 'think in opposites'. Although we have distinguished
between complementaries, such as single and married, and antonyms,
such as good and bad (and it is important to draw this distinction), it
is noticeable that the difference between them is not always clear-cut
in the ' logic' of everyday discourse. If the answer ' No' is given to the
LIT
16
question 'Was it a good film?', this will probably be understood to

imply, 'It was a bad film', unless the person answering the question
goes on to qualify his denial and make clear, as it were, that he is not
content to make his judgement in terms of the polarized contrast of
good and bad. It may well be therefore that the gradability of antonyms
(though not their implicit reference to some accepted norm of com-
parison) is ' psycholinguistically' secondarythat is, something which
speakers are conscious of and utilize only when a first-level dicho-
tomization into 'yes' and 'no' is insufficient.
10.5 Componential analysis and universal semantics

10.5.i Preliminary discussion
What is meant by the term 'componential analysis' in semantics is
best explained by means of a simple exampleone that has often
been used for this purpose by linguists. Consider the following sets
of English words:
(1) man woman child
( 2 ) bull cow calf
(3) rooster hen chicken
(4) drake duck duckling
(5) stallion mare foal
(6) ram ewe lamb
On the basis of our intuitive appreciation of the sense of these words
we can set up such proportional equations as the following:
man: woman: child:: bull: cow: calf
This equation expresses the fact (and for the moment we may assume
that it is a fact) that, from the semantic point of view, the words
many woman and child, on the one hand, and bull, cow and calf, on
the other, all have something in common; furthermore, that bull and
man have something in common, which is not shared by either cow
and woman or calf and child; that cow and woman have something in
common, which is not shared by either bull and man or calf and child;
that calf and child have something in common that is not shared by
either bull and man or cow and woman. What these different groups
of words have in common we will call a semantic component. (Other
terms have also been used in the literature: 'plereme', 'sememe',
IO.5. COMPONENTIAL ANALYSIS 471
'semantic marker*,'semantic category', etc.; references will be found
in the notes.)
Let us now introduce some elementary arithmetical considerations.
Given a numerical proportion (what the Greek mathematicians and
grammarians called an 'analogy': cf. 1.2.3) f t n e general form
a:b::c:d
where the first of the four expressions divided by the second is equal
to the third divided by the fourth, we can factorize the proportion
into what for the present purpose we may call its 'components';
and we can then refer to each of the four expressions as the product
of a pair of components. (We have already made use of this parallel
in our discussion of the distributional definition of the morpheme:
cf. 5.3.3.) For example, from the proportion
2:6::io:3O
we can extract the components 1, 2, 3 and 10. The proportion can
then be restated as
(2x 1)1(2 x 3)::(iox I ) : ( I O X 3)
where 2 is analysed as the product of 2 and 1; 6 as the product of
2 and 3; and so on. In this instance, three of the components are
prime numbers, 1, 2 and 3; the fourth, 10, is not. However, in the case
of numerical proportions we can always discover whether a given
number is a prime or not; and, if it is not, we can determine its
ultimate componentsthe set of prime numbers in terms of which it
can be factorized. For the present purpose, we may assume that the
process of factorization rests upon the availability of all the relevant
proportions. For instance, if we have available the further proportion
1:2::5:10, we could factorize 10 into the prime numbers 2 and 5;
and we could then express our original proportion as
(2XI):(2X3)::((2X5)XI):((2X5)X3)
Each of the four expressions is now restated as the product of its

ultimate components.
Let us now apply these considerations to the analysis of the English
words given above. From the proportion man:woman::bull:cow, we
can extract four components of sense: we will refer to these as
(male), (female), (adult-human), (adult-bovine). At this stage of the
analysis, if one were actually analysing the words on the basis of
proportional equations, (adult-human) and (adult-bovine) would be
16-2
regarded as single components. But as soon as we restate the pro-
portion man: woman: child:: bull: cow: calf as
(male) x (adult-human): (female)
x (adult-human): (non-adult-human)
:: (male) x (adult-bovine): (female)
x (adult-bovine): (non-adult-bovine)
we can extract the further components (adult) and (non-adult). No
one of these components, it should be observed, is assumed to be an
ultimate component (a * prime'): it is conceivable that, by bringing
forward for comparison other words of English and setting up further
proportions, we should be able to factorize (human) or (male) into
'smaller' semantic components, just as we factorized 10 into 5 and 2.
Eventually we might hope to describe the sense of all the words in the
vocabulary in terms of their ultimate semantic components. Assuming
that the proposed analysis of the few English words given above is
correct as far as it goes (and we will presently consider what' correct'
means here), we can say that the sense of man is the product of the
components (male), (adult) and (human); that the sense of mare is the
product of (female), (adult) and (equine); and so on.
The componential approach to semantics has a long history in
linguistics, logic and philosophy. It is inherent in the traditional
method of definition by dividing a genus into species and species into
subspecies; and this method of definition is reflected in most of the
dictionaries that have ever been compiled for particular languages,
and in the organization of such works as Roget's Thesaurus (cf. 10.2.1).
A number of attempts have been made in recent years to formalize
these traditional principles of semantic analysis. We may begin by
discussing some of the more important assumptions upon which
current componential theories of semantics are based or with which
they are frequently associated. The first is the assumption that the
semantic components are language-independent, or universal.
10.5.2 The alleged universality of semantic components

It has frequently been suggested that the vocabularies of all human
languages can be analysed, either totally or partially, in terms of a
finite set of semantic components which are themselves independent
of the particular semantic structure of any given language. According
to this view (which has been a commonplace of philosophical and
linguistic speculation since the seventeenth century) the semantic
components might be combined in various ways in different languages
(and thus yield ' senses' or ' concepts' unique to particular languages),
but they would themselves be identifiable as the ' same' components
in the analysis of the vocabularies of all languages. To quote Katz, who
has put forward this view in a number of recent publications:
'Semantic markers [i.e. semantic components] must.. .be thought of
as theoretical constructs introduced into semantic theory to designate
language invariant but language linked components of a conceptual
system that is part of the cognitive structure of the human mind.'
Little need be said about the alleged universality of semantic
components, except that it is an assumption which is commonly made
by philosophers and linguists on the basis of their anecdotal discussion
of a few well-chosen examples from a handful of the world's languages.
Chomsky has suggested: * It is surely our ignorance of the relevant
psychological and physiological facts that makes possible the widely
held belief that there is little or no a priori structure to the system of
" attainable concepts 'Y The first point that should be made about
this remark is simply that the belief that there are few, if any,
'universal, language-independent constraints upon semantic features
[i.e. semantic components]' is probably most widely-held among those
linguists who have had some experience of the problems of trying to
compare the semantic structure of different languages in a systematic
fashion: many have tried, and failed, to find a set of universal
components. The second point is that, although Chomsky's own work
contains a number of interesting, and probably correct, observations
about certain classes of lexical items (e.g. * proper names, in any
language, must designate objects meeting a condition of spatio-
temporal contiguity', 'the color words of any language must sub-
divide the color spectrum into continuous segments', 'artifacts are
defined in terms of certain human goals, needs and functions instead
of solely in terms of physical qualities'), such observations do not go
very far towards substantiating the view that there is 'some sort of
fixed, universal vocabulary [of semantic components] in terms of
which [possible concepts] are characterized'.
It may well be that future developments in semantics, psychology,
physiology, sociology, anthropology, and various other disciplines,
will justify the view that there are certain 'language invariant but
language linked components of a conceptual system that is part of the
cognitive structure of the human mind', as Katz has suggested. Such
empirical evidence as there is available at the present time would tend
to refute, rather than confirm, this hypothesis.
16-3
I0
10.5.3 Componential analysis and conceptualism

It is obvious that the value of componential analysis in the description
of particular languages is unaffected by the status of the semantic
components in universal terms. It should also be realized that
componential theories of semantics are not necessarily * conceptualist',
or 'mentalistic'. This point is worth stressing, since not only Katz
and Chomsky, but also Hjelmslev, Jakobson, and many others who
have advocated a componential approach to semantics, have done so
within a philosophical and psychological framework which takes it
for granted that the sense of a lexical item is the ' concept' associated
with this item in the 'minds' of the speakers of the language in
question. For example, Katz introduces the notion of semantic
components (or 'semantic markers') as follows: 'Consider the idea
each of us thinks of as part of the meaning of the words "chair",
"stone", "man", "building", "planet", etc., but not part of the
meaning of such words as "truth", "togetherness", "feeling",
"shadow", "integer", "departure", etc.the idea that we take to
express what is common to the meaning of the words in the former
group and that we use to conceptually distinguish them from those in
the latter. Roughly, we might characterize what is common to our
individual ideas as the notion of a spatially and contiguous material
thing. The semantic marker (Physical Object) is introduced to
designate that notion.'
We have already suggested that semantic theory should avoid
commitment with respect to the philosophical and psychological status
of 'concepts', 'ideas' and the 'mind' (cf. 9.2.6). Here it is sufficient
to observe that what Katz has to say about the difference between the
two groups of words can be stated without employing the term
'concept' or 'idea'. The first group of words denote things which are,
or can be, described in English as 'physical objects' (the expression
'physical object' is of course itself made up of English words); the
second group of words do not. Whether the correct application of the
first group of words to their referents presupposes that the speaker
has some 'idea' of 'physical object' in his 'mind' is a psychological
question which we may leave on one side. The important question for
the linguist is whether there are any facts pertaining to the accept-
ability or unacceptability of sentences, or to the relations of impli-
cation which hold between sentences, which can be described by
assigning to all the words of the first group a distinctive semantic
component, which we will agree to call' (physical object) \ The answer
to this question carries no implications whatsoever for the dispute
between various schools of philosophy and psychology about the
status of 'mental concepts'.
10.5.4 Apparent advantages of the componential approach

At first sight, the componential approach to semantics would seem to
have one striking advantage over other approaches: in terms of the
same set of components one can answer two different questions. The
first question has to do with the semantic acceptability of syntagmatic
combinations of words and phrases: whether a given combination is
to be generated as significant or excluded as meaningless. The second
question is this: what is the meaning (i.e. the sense) of a particular
combination of lexical items? We will take each of these questions in
turn.
We have said that the significance of grammatically well-formed
sentences (and parts of sentences) is traditionally accounted for in
terms of certain general principles of ' compatibility' between the
'meanings' of their constituent lexical items (cf. 9.3.11). One way of
stating this notion of semantic 'compatibility' is to say that the
relevant semantic components of the lexical items in the syntagmatic
combination generated by the syntax must not be contradictory. Let
us assume, for example, that the word pregnant contains a component
which restricts it to the modification of nouns which contain the
component '(female)'. On the basis of this fact ('modification' being
interpreted by the syntactic rules of the language) such phrases as
the pregnant woman or a pregnant mare would be generated as signifi-
cant and such phrases as the pregnant man or a pregnant stallion would
be excluded as meaningless ('uninterpretable'). Whether such phrases
as the pregnant duck are significant would presumably be decided with
reference to further components of sense associated with the wrord
duck and further restrictions imposed upon the combinability of
pregnant with nouns.
There is no doubt that this is an elegant way of accounting for the
combinatorial restrictions which hold between lexical items in
particular grammatical constructions. It is to be noted, however, that
any comprehensive treatment of the significance of sentences in such
terms presupposes an adequate syntactic analysis of sentences and
satisfactory rules for the semantic interpretation of the relevant
grammatical relations. The example that has just been given, which
involved the 'modification' of a noun by an 'adjective', is one which
has never been regarded as particularly troublesome by semanticists.
Its formalization within the framework of current syntactic theory is
trivial by comparison with the problem of formalizing the vast
majority of the relations of semantic 'compatibility' which hold in
the sentences of any language. In the last few years there has been a
remarkable concentration of interest upon the problems attaching to
the formalization of different relations of semantic 'compatibility'
(notably by Katz, Weinreich and Bierwisch). So far the results are
not impressive, despite the sophistication of the formal apparatus
that has been developed; and it would seem that progress in this area
is dependent upon the construction of a more appropriate theory of
syntax than is yet available.
The second question that componential analysis sets out to answer
is 'What meaning does a given sentence or phrase have?' The general
answer to this question is that the meaning of a sentence or phrase is
the 'product' of the senses of its constituent lexical items; and the
sense of each lexical item is the 'product' of its constituent semantic
components. The meaning of a sentence or phrase is therefore
determined by 'amalgamating' all the semantic components of the
lexical items according to a set of 'projection rules' which are
associated with deep-structure grammatical relations. It was sug-
gested in the previous paragraph that current syntactic theory does
not yet provide us with a satisfactory account of many of the relevant
deep-structure grammatical relationships; and this was the main
burden of our discussion of 'grammatical functions' in chapter 8.
It follows that we are at present unable to interpret the term' product'
(or 'compositional function'to employ the more technical term)
in the proposed definition of the meaning of a sentence or phrase as
'the product of the senses of its constituent lexical items'.
At the same time, it is clear that many of the semantic relations
discussed in the previous chapter might be reformulated within a
componential theory of semantics. Synonymy, hyponymy, in-
compatibility and complementarity are obviously definable in terms
of the semantic components of the lexical items in question. (For a
componential approach to the definition of these relations the reader
is referred to the works cited in the notes.) What must be stressed,
however, is the fact that the componential analysis of lexical items
rests upon the prior notion of 'implication' with respect to the
assertion and denial of sentences. Componential analysis is a tech-
nique for the economical statement of certain semantic relations
between lexical items and between sentences containing them: it
cannot claim to circumvent any of the problems of indeterminacy
that were discussed above in connexion with ' understanding' and
'analytic implication' (cf. 10.1.2).
10.5.5 The 'cognitive realityy of semantic components

The most interesting work so far published in the field of compo-
nential semantics has come, not from philosophers and linguists, but
from anthropologists; and they have recently devoted considerable
attention to what they have called the 'cognitive validity', or 'reality',
of semantic components. It was this question that we had in mind,
when we said earlier that we should have to examine what was meant
by 'correct' in the context of componential analysis (cf. 10.5.1).
A good deal of the anthropological discussion makes reference to
the analysis of the vocabulary of kinship in various languages. It has
been shown, for example, that one can analyse the most common
kinship terms of English in various ways. (In particular, brother and
sister can be regarded as having the same component, ' direct line of
descent', as father and mother or son and daughter, as against cousin,
which shares the component 'collateral' with uncle and aunt, and
with nephew and niece; alternatively, brother and sister can be analysed
as having the same component, 'co-lineal', as uncle and aunt or
nephew and niece, as against cousin, which has the component 'ab-
lineal'.) The question is which, if any, of the various possible analyses
is 'correct'. Each of them is self-consistent; each of them distin-
guishes every member of the lexical system from every other member
of the system; and each of them is 'predictive', in the sense that it
provides the anthropologist with a means of deciding, with respect to
any member of the family, what his relationship is to other members of
the family in terms of the lexical system. But each of the alternative
analyses rests upon a different set of proportional equations:
either
father: mother:: son: daughter:: brother: sister
or
uncle: aunt:: nephew: niece:: brother: sister
It is therefore the ' cognitive validity' of one set of proportions, rather
than the other, which should decide the question of 'correctness' (if,
indeed, this question is decidable). As far as the anthropological
analysis of kinship is concerned, the ' cognitive validity' of a particular
proportion is determined, presumably, by the social status and roles
assigned to the different classes of family-relatives in the society; and
10. SEMANTIC STRUCTURE
this might well be reflected also in the linguistic ' intuitions' of the
members of the community.
But we can also consider the question of' correctness' from a more
strictly linguistic point of view. Let us return, for this purpose, to the
simple illustration of componential analysis with which we began this
section. We assumed the validity of the following proportions
man: woman: child:: bull: cow: calf
bull: cow: calf:: rooster: hen: chicken
etc.
On the basis of these proportions, we * extracted* the semantic
components (male) v. (female), (adult) v. (non-adult), (human) v.
(bovine) v. (equine) v. . . . (sheep). We may now ask what is the
linguistic status of these components.
At first sight, the opposition of the contradictory components
(male) and (female) looks satisfactory enough. If we know that
someone is an adult, male, human being, then we know that the word
man, rather than woman or child, is appropriately applied to him; if
we know that a particular domestic fowl is an adult female of a given
species, then we know that hen, rather than rooster or chicken, is the
appropriate term of reference; and so on. But one might maintain
that to differentiate man and woman, rooster and hen, etc., in terms of
the sex of their referents is to give priority to but one of the many
linguistically-relevant features which distinguish them. If one asks
a young child (most of whose utterances are perfectly acceptable and
manifest the same semantic relationships, as far as one can judge, as
the utterances of his elders) what is the difference between men and
women, he might answer by listing a whole set of typical charac-
teristicsthe kind of clothes they wear, how their hair is cut, whether
they go out to work or stay at home and look after the children, etc.
A totally unrelated set of criteria might be proposed for the dif-
ferentiation of rooster and hen, of bull and cow, and so on. Why should
one suppose that sex is the sole criterion even in adult speech ? And
how far is it true to say that woman: child: :cow: calf:: hen: chicken, etc. ?
Obviously, there is a certain class of sentences, the semantic
acceptability or unacceptability of which can be accounted for in
terms of this proportional equation: That woman is the mother of this
child, That hen is the mother of this chicken, etc. v. That man is the
mother of this child, That zooman is the father of this child, That woman
is the mother of this calf, etc. And the grammatical phenomenon of
gender in English is partly determined by the sex of the referent.
But this does not mean that (male) and (female) are the sole semantic
features which differentiate the complementary terms man v. woman,
bull v. cow, etc. The status of such components as (adult) v. (non-
adult) is even more dubious: once again, there are sets of semantically-
acceptable or semantically-unacceptable combinations that can be
accounted for in terms of this opposition, but there are others that
cannot.
The problem is undoubtedly related to the anthropologist's
problem of 'cognitive reality'. Consider, for example, a society in
which the role of men and women is so different that there are very
few activities in which both will engage. Assume now that there are
two lexical items in the vocabulary of that language, which can be
translated into English as ' man' and * woman' on the basis of their
reference to male, adult human beings and female, adult human
beings, respectively. Knowing this fact about the reference of the two
lexical items, the linguist could apply these terms appropriately to
men and women. He would be fairly sure that the translation of
English sentences such as ' The man gave birth to a child' (assuming
that there is a term that can be satisfactorily translated as' give birth to')
would be semantically-unacceptable. But there might be an enormous
range of other sentences, including 'The man cooked a meal', 'The
woman lit a fire', etc., which are equally unacceptable. Our own
cultural prejudices and our own taxonomic classification of the physical
world should not be taken as valid for the analysis of either the culture
or the language of other societies, still less of any alleged ' conceptual
system that is part of the cognitive structure of the human mind'.
A further point should be made. It is one of the concomitant
dangers of componential analysis that it tends to neglect the difference
in the frequency of lexical items (and therefore their greater or less
'centrality' in the vocabulary) and the difference between lexical
items and semantic components. For example, it is often suggested
that brother and sister can be replaced with the 'synonyms' male
sibling and female sibling. But this is true only in the context of
anthropological or quasi-anthropological discussion. The words
brother and sister are extremely common words, known presumably
to all speakers of English, whereas sibling is a technical term, coined
for the convenience of anthropologists; and most English speakers
probably do not know it. The fact that there is no common super-
ordinate term for the two complementaries brother and sister is
prima facie evidence that the opposition between the two terms is
semantically more important than what they have in common.
480 IO. SEMANTIC STRUCTURE
Similarly, the fact that there is a term horse, which has as its hyponyms
the complementaries stallion and mare, is relevant to the analysis of
the structure of the English vocabulary. Any theory of semantics
which encouraged us to believe that the phrase adult male elephant
stood in exactly the same semantic relationship to elephant as stallion
does to horse would be unsatisfactory.
Componential theories of semantics do not necessarily fall victim
to inadequacies of this kind. But there has been surprisingly little
attention devoted to a discussion of the relationship between lexical
items like male or adult and semantic components like (male) or
(adult). One cannot avoid the suspicion that the semantic components
are interpreted on the basis of the linguist's intuitive understanding of
the lexical items which he uses to label them.
10.5.6 Concluding remarks

Limitations of space prevent us from going further into the details of
recent componential work in semantics. If our treatment of the
subject has been somewhat negative, it should be realized that this
has been by deliberate decision. I have tried to draw attention to
some of the assumptions upon which componential theories of
semantics are frequently basedin particular, the assumption that
semantic components are universal. We have seen that the notion of
componential analysis rests upon the establishment of proportional
equations with respect to the sense of lexical items. The important
question, which is not always considered, is the degree to which these
proportions are 'cognitively valid*. It is too often assumed that these
proportions can be set up simply on the basis of introspection.
Componential analysis has, however, made considerable contribu-
tions to the development of semantics. Apart from anything else, it
has brought the formalization of syntax and the formalization of
semantics (or some aspects of semantics) closer together than they
have been in the past. That linguists are once again seriously con-
cerned with the relations between syntax and semantics is due very
largely to the impact made by the work of Katz and Fodor, taken up
more systematically within the framework of ' an integrated theory of
linguistic descriptions' by Katz and Postal, and further developed
by Katz in a number of subsequent publications. Although Katz and
Postal tended to minimize the value of previous work in componential
analysis, they were right to insist upon the importance of specifying
the form of the 'projection rules' and the manner of their operation
IO.5. COMPONENTAL ANALYSIS 481
* within the context of explicit generative linguistic descriptions*. This
had not been attempted before.
In a previous chapter, we raised the possibility of a rapprochement
between * formal' and 'notional' grammar (cf. 4.3.4); and much of
our subsequent discussion of 'grammatical categories' and 'gram-
matical functions' (in chapters 7 and 8) would tend to suggest that
further progress in the formalization of syntax depends upon this
rapprochement. We may conclude the present work in the certain
expectation that the next few years will see the publication of a
number of books and articles directed towards this goal.
It is not unlikely also that a greater concentration of interest upon
the theory of semantics will bring linguists back to the traditional
view that the syntactic structure of languages is very highly deter-
mined by their semantic structure: more especially, by the ' modes of
signifying' of semantically-based grammatical categories (cf. 7.6.10).
If this development does take place, one must be careful not to
assume that linguistic theory has merely retreated to the position held
by traditional grammarians. All future grammatical and semantic
theory, however traditional its aims miojht be, must meet the rigorous
demands of twentieth-century, 'structural' linguistics. Revolutions
may be followed by counter-revolutions; but there can be no simple
restoration of the past.
10.5.7 Envoi
Ohe iam satis est, ohe libelle,
iam pervenimus usque ad umbilicos...
iam lector queriturque deficitque,
iam librarius hoc et ipse dicit:
'ohe iam satis est, ohe libelle'.
MARTIAL
John Lyons
Online ISBN: 9781139165570
Chapter
Notes and references pp. 482-488

NOTES AND REFERENCES
(Full titles of books and articles will be found in the bibliography)
Notes to chapter 1, pp. 1-52

1.1.4 Quotation from Bloomfield: Language, 3.
1.2 Traditional grammar: Many of the standard textbooks of linguistics are
unsympathetic and unreliable in their references to traditional grammar
(e.g. Bloomfield, Language) Jespersen, Language; de Saussure, Cours). The
following may be recommended for further introductory reading: Robins,
Short History; Dinneen, Introduction; Arens, Sprachwissenschaft; Kuken-
heim, Esquisse; Leroy, Grands Courants; Mounin, Histoire; Sandys, Short
History.
1.2.1 Origins of Greek grammatical theory: cf. Forbes, * Greek pioneers';
Roller, ' Anfange'. On the Stoics, the importance of whom is frequently
underestimated, cf. Barth, Stoa; Barwick, Probleme; Pohlenz, Begriindung.
1.2.3 Analogy and anomaly: For the suggestion that Varro misrepresented
the issues, cf. Fehling, 'Varro'; also Dihle, 'Analogie'.
1.2.6 Roman period: For the general background, cf. Collart, Varron;
Barwick, Remmius Palaemon.
1.2.7 Medieval period: For the general philosophical background, cf.
Gilson, History; Copleston, History. On medieval linguistic theory: cf.
Dinneen, Introduction, 126-50; Robins, Short History; Bursill-Hall,
4
Medieval theories'; Roos, *Sprachdenken'. More specialized works:
Grabmann, Mittelalt. Geistesleben; Heidegger, Kategorienlehre; Roos,
Modi Significandi.
1.2.8 There is an interesting account of seventeenth-century 'rationalist'
linguistics in Chomsky, Cartesian Linguistics. I may have underestimated
the originality of the Port Royal grammarians.
1.2.10 The Indian tradition: cf. Belvalkar, Systems; Misra, Descriptive
Technique; Allen, Phonetics; Brough, 'Theories'; Bloomfield, Review of
Liebisch; Emeneau, ' India'; Renou's Introduction to Wackernagel and
Debrunner, Grammatik.
1.3 Comparative philology: The standard work is Pedersen, Linguistic
Science. It is notoriously deficient in its treatment of subjects lying outside
the principal concerns of the ' Neogrammarians', and may be supple-
mented with Jespersen, Language, 32-99; Leroy, Grands Courants, 17-60;
Mounin, Histoire, 152-212; Arens, Sprachwissenschaft, 139-337.
1.4 Modern linguistics: This section may be read in conjunction with
Gleason, Introduction, 1-13; Hockett, Course, 1-11, 569-86; Robins,
General Linguistics, 1-17; Hall, Introductory, 3-35; Martinet, Elements,
9-51; Lepschy, Linguistics, 17-41; Dinneen, Introduction, 1-19; Hill,
[482]
NOTES AND REFERENCES 483
Introduction, 1-12; Sapir, Language, 3-12; Malmberg, Nouvelles Tendances;
Householder, forthcoming book.
1.4.1 Ferdinand de Saussure: De Saussure's own notes have recently been
published, cf. Godel, Sources; and a critical edition of the Cours is now
appearing, cf. Engler, Cours. For a convenient summary of de Saussure's
views, cf. Dinneen, Introduction, 195-212; also Lepschy, Linguistica, 42-8
(includes a comprehensive, up-to-date bibliography).
1.4.2 Priority of the spoken language: cf. Abercrombie, Elements, 119;
Uldall, ' Speech'; Vachek, ' Some remarks'; Pulgram, * Graphic and
phonic*.
1.4.3 On language-planning: cf. Haugen, * Linguistics'; Hoenigswald,
'Proposal'.
1.4.7 Langue and parole: Saussure, Cours, 23 fT. (Introduction, chapter 3);
cf. Lepschy, Linguistics, 45-6; Dinneen, Introduction, 196-9; Leroy,
Grands Courants, 108-10. Chomsky relates his own distinction of 'com-
petence ' and * performance' to de Saussure's distinction of * langue' and
* parole', cf. Aspects, 4.

2.1.3 'Double articulation': cf. Martinet, Linguistique Synchronique, 1-35;
also Palmer, ' Hierarchy'; Hockett, ' Linguistic elements'.
2.1.4 'Level', 'plane', 'expression', 'content': The terminology used at
this point is that of Hjelmslev and his followers, cf. Hjelmslev, Pro-
legomena', Spang-Hassen, ' Glossematics'. The term 'level' is employed
elsewhere in this book in a very general sense (to include what Hjelmslev
calls a 'plane'). Attention is also drawn to 5.5.
2.2 Substance and form: cf Saussure, Cours, 155 fT. (part 2, chapter 4).
2.2.7 Arbitrariness of substantial realization: cf. Householder, 'Unique-
ness '.
2.3 Paradigmatic and syntagmatic relations: De Saussure himself does not
employ the term 'paradigmatic', but 'associative'. Furthermore, de
Saussure's 'associative relations' include non-paradigmatic relations of
various kinds: cf. Cours 170 fT. (part 2, chapter 5). The term 'paradig-
matic' is due to Hjelmslev, cf. Robins, General Linguistics, 78. (It is an
unfortunate choice of term, since it occasionally leads to confusion with
'paradigm': cf. Index.)
2.3.7 * Marked' and 'unmarked': these terms are associated particularly
with the 'Prague School' (cf. Vachek, Reader). For a recent discussion:
cf. Greenberg, Language Universals.
2.4 Statistical structure: cf. Cherry, Human Communication', Miller,
Language; Roberts, Statistical Linguistic Analysis; Mandelbrot, ' Theory',
Malmberg, Nouvelles Tendances, 278-304; Apostel et al., Logique.
2.4.1 Functional load: cf. Martinet, Economic, Hockett, 'Quantification'.
2.4.2 Information-theory: There is an elementary account in Gleason,
Introduction, pp. 373-90.
484 NOTES AND REFERENCES
2.4.7. On ' homonymics': cf. Orr, Words, esp. 91-160; Ullmann, Principlesy
144-52. (Orr and Ullmann give references to the work of Gillie>on and his
followers.)
2.4.9 Information about the frequency of English consonants is derived
from Roberts, Statistical Linguistic Analysis. The most commonly-used
list of word-frequencies is Thorndike and Lorge, Teacher's Word-Book.

3.2 Phonetics: cf. Abercrombie, Elements; Jones, Outline; Pike, Phonetics;
Ladefoged, Elements; Fant, Acoustic Theory. See Addenda (1), p. 489.
3.3 Phonology: cf. Robins, General Linguistics, 121-79; Martinet, Elements,
52-96; Hockett, Manual; Trubetzkoy, Grundziige; Jones, Phoneme;
Malmberg, Structural Linguistics, 30-127. See Addenda (2), p. 489.
3.3.9 'Grimm's law': cf. Fourquet, Mutations.
3.3.11 For recent developments in distinctive feature theory, cf. Jakobson,
Fant and Halle, Preliminaries; Jakobson and Halle, Fundamentals;
Chomsky and Halle, * Controversial questions'; Fudge, ' Phonological
primes'.
3.3.12 Prosodic analysis: cf. Robins, General Linguistics, 157-68. A number
of ' prosodic' articles are being reprinted in Palmer, Prosodic Analysis.
3.3.13 Turkish phonology: cf. Waterson, * Some aspects'; Lyons, 'Pho-
nemic and non-phonemic'. For a treatment in terms of generative
phonology: cf. Lees, Phonology.

4.1.3 Quotation from Jespersen, Philosophy, 55.
4.1.5 Chomsky's use of the term 'formal' {v. 'substantive'): cf. Aspects,
27-30.
4.2 Formal grammar: This section draws heavily upon Harris, Methods,
and Chomsky, Syntactic Structures.
4.2.3 Russell's discussion of 'grammatical', but 'meaningless', sentences:
Inquiry, 170 ff.
4.2.4 'Common core': the term comes from Hockett, Course, 332. For a
good account of systematic variety in language, cf. Halliday, Mclntosh
and Strevens, Linguistic Sciences, 75-110; Quirk, Use of English, 50-63,
79-96, 154-73. On dialects, cf. Mclntosh, Introduction; also Bach,
Mundartforschung; Dauzat, Geographie.
4.2.5 On 'acceptability' and 'grammatically', cf. Chomsky, Aspects, 10-
15-
4.2.14 For the distinction between discovery procedures, decision pro-
cedures and evaluation procedures, cf. Chomsky, Syntactic Structures,
49-60.
4.3.3 Quotation from Chomsky, Aspects, 79-80.

5.1.2 Quotations from Harris: Methods, 14.
5.2 The sentence: cf. Robins, General Linguistics, 190-292; Hockett,
Course, 199-208; Fries, Structure, 9-53.
5.2.1 Bloomfield's definition: from Language, 170.
5.2.5 'Ready-made* utterances: cf. Saussure, Cours, 172.
5.2.7 Phonological criteria: Hill, Introduction, 3366., attempted to define
the sentence solely in phonological terms. Once we distinguish between
utterances and sentences (and many linguists have failed to draw this
distinction) there is no reason to expect that phonological criteria should be
definitive. For a discussion of Hill's approach, cf. Haas, 'Linguistic
structures'.
5.3 The morpheme: cf. Robins, General Linguistics, 201-13; Bazell,
'Problem'; Linguistic Form, 51-64; Nida, Morphology.
5.3.6 Morphological typology: cf. Bazell, Typology. Ratios of morphemes
to words: cf. Greenberg, as cited in Mohrmann, et al., Trends 1, 119.
5.3.7 On Turkish: cf. Godel, Grammaire; Swift, Grammar.
5.3.8 On Latin: cf. Householder, 'Descriptive analysis'; P. H. Matthews,
'Inflexional component*; Hill, Introduction, 441-82.
5.3.9. Quotation from Sapir: Language, 124.
5.4 The word: cf. Robins, General Linguistics, 193-201; Rosetti, Linguistica,
11-46; Togeby,' Mot'; Sapir, Language, 24-41; Bazell,' Historical source'.
5.4.7 'Potential pause': cf. Hockett, Course, 166-9 (n.b. Hockett employs
the term ' lexeme' in a different sense from the sense used here).
5.4.8 Semantic definitions of the word: cf. Ullmann, Principles, 504.
5.4.9 'Minimal free form': cf. Bloomfield, Language, 178; also Robins,
General Linguistics, 194; Bazell, Linguistic Form, 67-8; Hockett, Course,
168-9.
5.5 'Rank': cf. Halliday, 'Categories'; Pike, Integrated Theory, Longacre,
' Fundamental insights'; Lamb, Outline; Postal, Constituent Structure;
P. H. Matthews, ' Concept of rank'; Huddleston, ' Rank and depth'. See
Addenda (3), p. 489.

6.1.2 Bloomfield: cf. Language, 161 ff.
6.1.3 Grammatical ambiguity of this kind: cf. Wells, 'Immediate con-
stituents '; Hockett, ' Two models'; Chomsky, ' Syntactic Structures', 28.
6.2 Phrase-structure grammar: cf. Chomsky, 'Three models'; Syntactic
Structures; Postal, Constituent Structure; Bach, Introduction; B. Hall,
Foundations. More technical papers in Luce et al., Readings, 75 ff.,
Handbook, 269 ff. Also Gross, ' Equivalence'.
6.3 Categorial grammar: cf. Bar-Hillel, Language, 99 ff.; also Lambek,
'Calculus'. Attention is drawn to 'dependency grammar', which has
certain similarities: cf. Hays, ' Dependency theory'; Tesniere, Elements.
6.4 Endocentric and exoccntric: cf. Bloomfield, Language, 194 ff., 235 f.;
Hockett, Course, 184 ff.; Harris, Methods, 275 f.; Robins, General Lin-
guistics, 234 f.
6.5.4 Concord and government: cf. Hockett, Course, 214 ff.
6.5.5 Context-sensitivity: cf. Chomsky, Syntactic Structures, 28.
6.6 Transformational grammar: cf. Chomsky, Syntactic Structures', Topics',
Harris, * Co-occurrence'; * Transformational theory'; Bach, Introduction;
Shaumjan, Strukturnaja; Lamb, Outline; Koutsoudas, Writing Grammars.
6.6.1 Deep and surface structure: cf. Hockett, Course, 246 ff.; Chomsky,
Aspects, 16; Cartesian Linguistics, 31 ff.
6.6.2 Transformational ambiguity: cf. Chomsky, Syntactic Structures, 88 ff.
6.6.3 'Diversification': cf. Lamb, Outline, 17.
6.6.4, 6.6.6 Quotations from Chomsky, Syntactic Structures, 42; Aspects,
98.
6.6.9 Recent developments: cf. Chomsky, Aspects; Topics; Katz and Postal,
Integrated Theory; Fillmore, 'Embedding transformations'. For a critical
review of these developments: cf. P. H. Matthews, Review of Chomsky's
Aspects.

My main sources for this chapter on 'grammatical categories' and for
chapter 8 on ' grammatical functions' have been works (listed in the biblio-
graphy) by Bally, Benveniste, Bloomfield, Boas, Brondal, Entwistle, Gray,
Hjelmslev, Jakobson, Jespersen, Kurylowicz, Martinet, Meillet, Vendryes.
7.2 Deictic categories: cf. esp. Kurylowicz, Inflexional Categories; also
Fillmore, ' Deictic categories', 220 ff.
7.3.5 Swahili: cf. Ashton, Grammar.
7.3.8 'Classifiers': For Chinese, cf. Chao, Primer. For Vietnamese, cf.
Emeneau, Studies; Honey, 'Word classes'; Thompson, Grammar. For
Turkish and Uzbek, cf. Waterson, 'Numeratives'.
7.4 Case: The classic treatments are Hjelmslev, Categoric, and Jakobson,
* Kasuslehre'. Two recent studies of particular interest are Fillmore,
' Modern theory'; ' Case for case'. For Turkish: cf. Swift, Grammar;
Godel, Grammaire. For Finnish: cf. Sauvageot, Esquisse.
7.5.1 Tense: cf. Jespersen, Philosophy, 254 ff.
7-5-3 Quotation from Palmer: English Verb, i n . Quotation from Hockett,
Course, 237.
7.5.5 Aspect: For Russian, cf. Ward, Russian Language, 227 ff. Quotation
from Hockett, Course, 237.
7.5.7 The term ' imitative' I have taken from Strang, English Structure, 146.
7.5.8 The following recent publications (and many others) discuss the
categories of tense, mood and aspect with reference to English: Joos,
English Verb; Ota, Tense and Aspect; Palmer, English Verb; Ehrman,
Modals; Crystal, 'Specification'; Mclntosh, 'Predictive statements'.
7.6 Parts of speech: cf. Lyons, 'Notional theory'.
7.6.6 References to Jespersen, Philosophy; Analytical Syntax; to Hjelmslev,
Principes. References to recent transformational work: cf. esp. Lakoff,
Irregularity, Deep and Surface; Fillmore, * Case for case'; Postal and
Rosenbaum, Sentence Formation.
7.6.10 Quotation from Chomsky, Aspects, 118.

Main sources for this chapter: see covering note to chapter 7.
8.1.2 Quotations from Sapir: Language, 119; Hockett, Course, 201. For
'given' and 'new': cf. Halliday, * Notes'.
8.1.3 For * sortal' and ' characterizing' universals: cf. Strawson, Individuals.
8.1.7 Eskimo: cf. Hill, Introduction, 419-40.
8.1.8 Quotation from Chomsky: Aspects, 221. For a summary of earlier
views: cf. Sandmann, Subject; Sechehaye, Structure.
8.2.2 Quotation from Robins: General Linguistics, 266.
8.2.3 Ergative: cf. Allen, * Transitivity'; Hockett, Course, 235; Martinet,
'Ergatif; W.K.Matthews, * Ergative'; Uhlenbeck, 'Agens'; Vaillant,
'Ergatif'.
8.2.13 Quotation from Halliday: * Notes', JL, 3 (1967), 47.
8.3.1 Reference to Svartvik: Voice, 2.
8.3.5 For alternative proposals: cf. Anderson, * Ergative'; Fillmore,
* Case for case'; Halliday, * Notes \
8.4 Existential, locative and possessive: cf. Lyons, 'Note'; Graham,
4
Being'; Kahn, ' Greek verb'; Benveniste, ' tre'; Bendix, Componcntial
Analysis.
8.4.6 Chomsky's example: cf. Aspects, 21.

(Much of chapters 9 and 10 is a more systematic and more up-to-date version
of what appears as Part 1 of Lyons, Structural Semantics.) There is a vast
literature on semantics; and there is no book which even claims to cover the
whole field. The most comprehensive treatment of previous work in
semantics, from the purely linguistic point of view, is Ullmann, Principles;
cf. also Kronasser, Handbuch; Nida, Translating. For philosophical seman-
tics: cf. Bar-Hillel, 'Logical syntax'; Cohen, Diversity; Warnock, English
Philosophy; Wells, 'Meaning*. A number of important articles are reprinted
in Caton, Philosophy; Linsky, Semantics. For a criticism of twentieth-century
philosophical semantics: cf. Katz, Philosophy. For psychological semantics:
cf. Brown, Words; Bruner et al.t Thinking; Carroll, Language and Thought;
Miller, Language. For a popular account of philosophical semantics: cf.
Salomon, Semantics. Also Brower, Translation; Austin, Words.
9.1.1 The term ' semantics'; cf. Bral, Essai; Read, ' Account \
9.2.6 Conceptualism and mentalism: Quotation from Haas,' Denning', 74.
BloomfielcTs approach: cf. Language, 139 ff.; Fries, * Meaning \ In my view,
the issue has been confused, rather than clarified, by Chomsky, Aspects,
193 f.; Katz, * Mentalism \
9.2.9 * Meaning* and 'use': cf. Wittgenstein, Investigations; Wells,
* Meaning and use'. Misunderstanding: it might be argued that the notion
of * understanding * is on a par with * misunderstanding * with respect to my
argument. Whether or not this is so, there is a considerable methodological
advantage in assuming that misunderstanding is more directly testable.
Most theories of semantics are vitiated by their assumption that * under-
standing ' is a matter of the transference of the same * content' from one
person's * mind' to another.
9.3.6 Quotation from Firth: Papers, 203. 'Behaviourism' in semantics:
my argument should not be taken as suggesting that the theory of seman-
tics can be satisfactorily developed on the basis of stimulus-response theory
(cf. my criticisms of behaviourist semantics in Lyons, Structural Semantics,
2f.).
9.3.7 Malinowski's notion of 'phatic communion': cf. 'Problem', 309 ff.
9.4 Reference and sense: This distinction goes back to Frege, 'Sinn';
a somewhat similar distinction is found in Russell, ' Denoting'. My
account is based very largely on Carnap, Meaning; and Quine, Logical
Point of View; cf. also Bar-Hillel, 'Logical syntax'.
9.4.4 Semantic fields: cf. Guiraud, Semantique, 68 ff.; Ohman,' Theories';
Ullmann, Principles, 152 ff.
9.4.5 Reference to Conklin: 'Hanun6o'; cf. also Lenneberg and Roberts,
Experience; Carroll, Language and Thought, 95, 107 ff.; Andre", Etude.
9.4.6 Quotation from Sapir: Selected Writings, 162. On linguistic 'deter-
minism': cf. Carroll, Language and Thought, 106 ff.; Black, 'Relativity'.
9.5.1 Quotation from Fries: Structure, 66.
9.5.2 Reference to Martinet, Elements, 117 (section 4.19); Functional View,
50 f.; to Halliday,' Categories';' Lexis'; Mclntosh and Halliday, Patterns,
5 f. (For criticism of Halliday's distinctions between ' grammar' and ' lexis'
and between 'contextual* and 'formal* meaning: cf. Lyons, 'Firth's
theory', 298 f.)
9.5.3 Reference to Chomsky: Aspects, 71.
9.5.4 Reference to Katz and Postal: Integrated Theory, 74 ff.

10.1.1 Definition of'length', etc.: cf. Reichenbach, Elements, 210.
10.1.2 'Analytic' and 'synthetic': cf. Cohen, Diversity, 153 ff.; Staal,
' Analyticity'.
10.2.1 Quotation from Roget: 'How to use this book', p. vi.
10.2.2 Quantification of synonymy: For a more interesting proposal than
most, cf. Sparck Jones, Synonymy; cf. also Needham,' Automatic classifi-
cation '.
10.2.3 Quotation from Ullmann; Principles, 108-9.
John Lyons
Online ISBN: 9781139165570
Chapter
Addenda pp. 489-489

10.2.4 'Cognitive' v. 'emotive': cf. Ullmann, Principles, 96 ff.; Salomon,
Semantics, 27 ff. For a full and critical account, cf. Henle, Language, 121-72
10.3.1 'Hyponymy': As far as I know, this term was first employed by
Bazell, ' Logical syntax'but in a somewhat different sense.
10.3.2 The definition of synonymy in terms of hyponymy is proposed in
Staal, ' Analyticity', 78.
10.3.4 The Greek word demiourgds is one of many related words discussed
in Lyons, Structural Semantics.
10.4.1 ' Oppositeness': One of the best discussions is in Ogden, Opposition.
10.4.2 Reference to Moravcsik: * Analytic*.
10.4.4 Quotation from Sapir: Selected Writings, 122. Discussion of such
sentences as A small elephant is a large animal: cf. Weinreich, ' Explora-
tions ', 422 ff.; Katz, * Recent issues', 184 ff.
10.5 Componential analysis: cf. Bendix, Componential Analysis', Bierwisch,
'Hierarchie'; Bolinger, *Atomization'; Burling, 'Cognition'; Conklin,
'Lexicographical treatment'; Ebeling, Linguistic Units; Goodenough,
'Componential analysis'; Hallig and Wartburg, ' Begriffssystem';
Hjelmslev, Prolegomena) Katz, Philosophy, 'Recent issues'; Kiefer,
' Semantic relations'; Lamb,' Sememic approach'; Lounsbury,' Semantic
analysis'; 'Structural Analysis'; Weinreich, 'Explorations'.
10.5.2 Quotation from Katz: 'Recent issues', 129. Quotations from
Chomsky: Aspects, 160; 29.
io5-3 Quotation from Katz: 'Recent issues', 129.
10.5.5 * Cognitive reality' ;cf. Burling,' Cognition'; Romneyand D'Andrade,
' Cognitive aspects'; Wallace and Atkins, ' Kinship terms'.
ADDENDA
1 It was not correct to say that the values of the cardinal vowels are
defined by the IPA (p. 104). The IPA established a set of divisions, or
ranges, within the articulatory continua. Daniel Jones identified a number
of fixed points within these ranges and took these points as ' cardinal'.
2 In the section on neutralization in phonology (3.3.5), it should be
pointed out that there is a third possible way of describing the ' facts'. If we
abandon the condition of biuniqueness (so that the same speech-sound is
always associated with the same phoneme), we can say that /d/ is realized as
[t] in word-final position and as [d] elsewhere and that /t/ is realized as [t]
generally. Cf. Fudge, ' phonological primes'.
3 There is a greater difference between Halliday's 'ranks' and Lamb's
'strata' than I have suggested in 5.5.1. It is perhaps only with respect to the
distinction of morphology and syntax that they are strictly comparable.
John Lyons
Online ISBN: 9781139165570
Chapter
Bibliography pp. 490-505

BIBLIOGRAPHY
This list includes all the works referred to in the notes. Periodicals are cited
according to the conventions of the Linguistic Bibliography of the Permanent
International Committee of Linguists, published annually (Utrecht and
Antwerp: Spectrum). The following abbreviations are used:
AL Acta Linguistica. Revue Internationale de linguistique structurale.
Copenhague.
ALH Ada Linguistica Academiae Scientiarum Hungaricae. Budapest.
AmA American Anthropologist. Menasha, Wise.
ArchL Archivum Linguisticum. Glasgow.
BPTJ Biuletyn Polskiego Towarzystva Jfzykoznawczego IBulletin de la
Societe polonaise de Linguistique. Wroclaw and Krakow.
BSE Brno Studies in English. Brno.
BSL Bulletin de la Societe de Linguistique de Paris. Paris.
BSOAS Bulletin of the School of Oriental and African Studies, University of
London. London.
CFS Cahiers Ferdinand de Saussuret Geneve.
CJL Canadian Journal of Linguistics / Revue Canadienne de Lin-
guistique. Toronto.
C & M Classica et Mediaevalia. Revue danoise de philologie et d'histoire.
Copenhague.
CPh Classical Philology. Chicago.
CQ The Classical Quarterly. London.
CR The Classical Review. London.
FL Foundations of Language. International journal of language and
philosophy. Dordrecht, Holland.
GK Gengo Kenkyu. (Journal of the Linguistic Society of Japan.)
Tokyo.
Glotta Glotta. Zeitschrift fur griechische und lateinische Sprache.
Gottingen.
IF Indogermanische Forschungen. Zeitschrift fur Indogermanistik und
allgemeine Sprachwissenschaft. Berlin.
IL Indian Linguistics. Journal of the Linguistic Society of India.
Poona.
IJAL International Journal of American Linguistics. Baltimore.
IRAL International Review of Applied Linguistics in Language Teaching /
Internationale Zeitschrift filr angewandte Linguistik in der Sprach-
erziehung. Heidelberg.
JAOS Journal of the American Oriental Society. New Haven, Connecticut.
JL Journal of Linguistics. The Journal of the Linguistics Association
of Great Britain. London and New York.
[490]
BIBLIOGRAPHY 491
KNf Kwartalnik Neofilologiczny. Warszawa.
KZ Zeitschrift fiir vergleichende Sprachforschung auf dem Gebiete der
indogermanischen Sprachen. (Begriindet von A. Kuhn.) Gottingen.
Lg. Language. Journal of the Linguistic Society of America. Baltimore.
Lingua Lingua. International Review of General Linguistics / Revue
internationale de linguistique ge"ne>ale. Amsterdam.
MSLL Monograph Series on Languages and Linguistics, Georgetown Uni-
versity. Washington, D.C.
NTS Norsk Tidsskrift for Sprogvidenskap. Oslo.
PF Prace filologiczne. Warszawa.
PMLA Publications of the Modern Language Association of America. New
York.
RLing Revue de Linguistique. Bucarest.
7.4 Southwestern Journal of Anthropology. Albuquerque, N.M.
SL Studia Linguistica. Revue de linguistique gn6rale et compared.
Lund.
SS Slovo a Slovesnost. Praha/Prague.
StGram Studia Grammatica. Berlin.
TCLC Travaux du Cercle Linguistique de Copenhague. Copenhague.
TCLP Travaux du Cercle Linguistique de Prague. Prague.
TLP Travaux Linguistiques de Prague. Prague.
UCPL University of California Publications in Linguistics. Berkeley and
Los Angeles.
VJa Voprosy Jazykoznanija. Moskva.
Word Word. Journal of the Linguistic Circle of New York. New York.
Abercrombie, David. Elements of General Phonetics. Edinburgh: Edinburgh

University Press and Chicago: Aldine, 1966.
Abraham, S. and Kiefer, F. A Theory of Structural Semantics. (Janua
Linguarum, 49.) The Hague: Mouton, 1965.
Ajdukiewicz, Kasimierz. 'Die syntaktische Konnexitat.' Studia Philosophica
(Warszawa), 1 (1935), 1-28.
Allen, W. Sidney. Phonetics in Ancient India. London, 1951.
On the Linguistic Study of Languages. (Inaugural lecture.) Cambridge:
Cambridge University Press, 1957. (Reprinted in Strevens, Five
Inaugural Lectures, pp. 3-26.)
* Structure and system in Abaza.' TPhS (1956), 127-76.
* Transitivity and possession.' Lg. 40 (1964), 337-43.
Anderson, John M. 'Ergative and nominative in English.' JL> 4 (1968).
Andre", J. tude sur les Termes de Couleur dans la Langue Latine. Paris,
1949.
Apostel, L., Mandelbrot, B. and Morf, A. Logique, Langage et Theorie
dyInformation. Paris: Presses Universitaires de France, 1957.
Arens, H. Sprachwissenschaft. Freiburg and Miinchen: Karl Alber, 1955.
Ashton, E. O. Swahili Grammar. London, 1944.
Austin, J. L. How To Do Things With Words. Cambridge, Mass.: Harvard
University Press, 1962.
17-3
492 BIBLIOGRAPHY
Bach, A. Deutsche Mundartforschung: Ihre Wege, Ergebnisse und Aufgaben.
2nd ed. Heidelberg: Carl Winter Universitatsverlag, 1950.
Bach, Emmon. An Introduction to Transformational Grammars. New York:
Holt, Rinehart and Winston, 1964.
Bally, Charles. Linguistique Generate et Linguistique Francaise. Paris: Ernest
Leroux, 1932.
Bar-Hillel, Y. Language and Information. Reading, Mass.: Addison-Wesley
and Jerusalem: Jerusalem Academic Press, 1964.
'Logical syntax and semantics.' Lg. 30 (1954), 230-7. (Reprinted in
Language and Information.)
Barth, P. Die Stoa. Stuttgart, 1946.
Barwick, K. Remmius Palaemon und die Rdmische Ars Grammatica. Leipzig,
1922.
Probleme der Stoischen Sprachlehre und Rhetorik. (Abhandl. der Sachs.
Akad. Wiss. zu Leipzig, Phil.-hist. Klasse, 49, 3.) Berlin: Akademie
Verlag, 1957.
Bazell, Charles E. Linguistic Form. Istanbul: Istanbul Press, 1953.
Linguistic Typology. (Inaugural lecture.) London: School of Oriental and
African Studies, 1958. (Reprinted in Strevens, Five Inaugural Lecturesy
pp. 29-49.)
'Logical and Linguistic syntax.* Litera (Istanbul), 2 (1955), 32-4.
' On the historical source of some structural units.' In Misceldnea Homenaje
I, A Andre Martinet. La Laguna, 1957.
'Syntactic relations and linguistic typology.' CFS, 8 (1949), 5-20.
'On the neutralization of syntactic oppositions.' TCLC, 5 (1949), 77-86.
(Reprinted in Hamp et al.y Readings.)
'On the problem of the morpheme.' ArchL, 1 (1949), 1-15. (Reprinted in
Hamp et al.y Readings.)
'The correspondence fallacy in structural linguistics.' Studies. . .English
Department, Istanbul University, 3 (1952), 1-41. (Reprinted in Hamp
et al.t Readings.)
'The sememe.' Litera (Istanbul), 1 (1954), 17-31. (Reprinted in Hamp
et al.y Readings.)
'Three misconceptions of grammaticalness.' MSLLt 17 (1964), 3-9.
Bazell, Charles E., Catford, J. C , Halliday, M. A. K. and Robins, R. H.
(eds.). In Memory of J. R. Firth. London: Longmans, 1966.
Belvalkar, K. Systems of Sanskrit Grammar. Poona, 1915.
Bendix, E. H. Componential Analysis of General Vocabulary: The Semantic
Structure of a Set of Verbs inEnglishy Hindiy and Japanese. (Part 2 oilJALy
32.) Bloomington: Indiana University and The Hague: Mouton, 1966.
Benveniste, fi. Problemes de Linguistique Generate. Paris: Gallimard, 1966.
' " t r e " et "avoir" dans leurs fonctions linguist!ques.' BSL> 55 (i960).
(Reprinted in Problemes.)
Bierwisch, M. 'Eine Hierarchie syntaktisch-semantischer Merkmale.'
StGramy 5 (1965).
Black, Max. 'Linguistic relativity: The views of Benjamin Lee Whorf.'
Philosophical Review, 68 (1959), 228-38.
BIBLIOGRAPHY 493
Bloch, Bernard and Trager, G. Outline of Linguistic Analysis. Baltimore:
Waverley Press, 1942.
Bloomfield, Leonard. Language, New York: Holt, Rinehart and Winston,
1933 and London: Allen and Unwin, 1935. (References are to the British
edition.)
Review of Liebisch Konkordanz Pdnini-Candra in Lg. 5 (1929), 267-76.
Boas, Franz. Race, Language and Culture. New York: Macmillan.
Bolinger, Dwight L. 'The atomization of meaning.' Lg. 41 (1965), 555-73.
* Linear modification.' PMLA, 67 (1952), 1117-44.
Breal, Michel. Essai de Semantique. Paris: Hachette, 1897. Translated as
Semantics: Studies in the Science of Meaning. New York: Dover, 1964.
Bright, William (ed.). Sociolinguistics. (Proceedings of the U.C.L.A. Socio-
linguistics Conference, 1964.) The Hague: Mouton, 1966.
Brondal, Vigo. Les Parties du Discours. Copenhague: Munksgaard, 1948.
Brough, John. 'Theories of general linguistics in the Sanskrit grammarians.'
TPhS (1951), 27-46.
Brower, Reuben, A. (ed.). On Translation. Cambridge, Mass.: Harvard
Brown, Roger W. 'Language and categories.' Appendix to Bruner et al.y
Thinking.
Words and Things. Glencoe, 111.: Free Press, 1958.
Bruner, Jerome S., Goodnow, J. J. and Austin, G. A. A Study of Thinking.
New York: Wiley, 1956.
Burger, A. * Sur le passage du systeme des temps et des aspects de l'indicatif
du latin au roman commun.' CFS, 8 (1949), 19-36.
Burling, Robbing. 'Cognition and componential analysis.' AmA> 66 (1964),
20-8.
Bursill-Hall, Geoffrey, 'Medieval grammatical theories.' CJL, 9 (1963),
39-54-
'Notes on the semantics of linguistic description.' In Bazell et aL, In
Memory, pp. 40-51.
Capell, A. Studies in Sociolinguistics. The Hague: Mouton, 1966.
Carnap, Rudolph. Meaning and Necessity. 2nd ed. Chicago: University of
Chicago Press, 1956.
Carroll, John B. Language and Thought. Englewood Cliffs, N.J.: Prentice-
Hall, 1964.
Catford, J. C. A Linguistic Theory of Translation. London: Oxford Uni-
versity Press, 1965.
Caton, Charles E. (ed.). Philosophy and Ordinary Language. Urbana, 111.:
University of Illinois Press, 1963.
Chao, Y. R. Mandarin Primer. Cambridge, Mass.: Harvard University
Press, 1948.
Cherry, Colin. On Human Communication, Cambridge, Mass.: M.I.T. Press,
1957. (Reprinted New York: Science Editions, 1959.)
Chomsky, Noam. Syntactic Structures. The Hague: Mouton, 1957.
Current Issues in Linguistic Theory. The Hague: Mouton, 1965.
Aspects of the Theory of Syntax. Cambridge, Mass.: M.I.T. Press, 1965.
494 BIBLIOGRAPHY
Chomsky, Noam, Topics in the Theory of Generative Grammar. The Hague:
Mouton, 1966. (Also in Sebeok, Current Trends, vol. 3.)
Cartesian Linguistics. New York and London: Harper and Row, 1966.
* Three models for the description of a language.' 1956. Reprinted in Luce
et al.y Readings.
Chomsky, N. and Halle, M. ' Some controversial questions in phonological
theory.' JL, 1 (1965), 97-138.
Cohen, L. Jonathan. The Diversity of Meaning. 1st ed. London: Methuen,
1962. (2nd ed., 1966.)
Collart, Jean. Varron, Grammarien Latin. Paris, 1954.
Conklin, H. C. 'Hanun6o color categories.' SJA, 11 (1955), 339-44.
(Reprinted in Hymes, Language in Culture, pp. 189-92.)
'Lexicographical treatment of folk taxonomies.' In Householder and
Saporta, Problems, pp. 119-41.
Copleston, Frederick. A History of Philosophy. Volumes 2 and 3. London:
Burns Oates, 1950 and 1953.
Crystal, David. 'Specification and English tenses.' JL, 2 (1966), 1-34.
Curry, Haskell B. 'Some logical aspects of grammatical structure.' In
Jakobson, Structure of Language, pp. 56-68.
Dauzat, A. La Geographie Linguistique. Revised ed. Paris: Flammarion,
1943-
Deese, James. The Structure of Associations in Language and Thought.
Baltimore: Johns Hopkins Press, 1965.
Diderichsen, P. 'Morpheme categories in modern Danish.' TCLC, 5 (1949),
134-55.
Dihle, A. 'Analogie und Attizismus.' Hermes, 85 (1957), 170205.
Dinneen, Francis P. An Introduction to General Linguistics. New York: Holt,
Rinehart and Winston, 1967.
Diver, William. 'The system of agency in the Latin noun.' Word, 20 (1964),
178-96.
Dover, K. J. Greek Word Order. Cambridge: Cambridge University Press,
i960.
Ebeling, C. L. Linguistic Units. (Janua Linguarum, Series Minor, 12.) The
Hague: Mouton, i960.
Ehrman, Madeline. The Meaning of the Modals in Present-Day American
English. The Hague: Mouton, 1966.
Elson, Benjamin and Pickett, V. B. An Introduction to Morphology and
Syntax. Santa Ana, Calif.: Summer Institute of Linguistics, 1962.
Emeneau, M. B. Studies in Vietnamese (Annamese) Grammar. (UCPL, 8.)
Berkeley and Los Angeles: University of California Press, 1951.
* India and linguistics.' JAOS, 75 (1955), 145-53.
Engler, Rudolph. Cours de Linguistique Generale de F. de Saussure: Edition
Critique. Wiesbaden: Otto Harrassowitz. (Appearing in fascicules, from
1967.)
Entwistle, William J. Aspects of Language. London, 1953.
Fant, Gunnar. Acoustic Theory of Speech Production. The Hague: Mouton,
i960.
BIBLIOGRAPHY 495
Fehling, Detlef. ' Varro und die grammatische Lehre von der Analogie und
der Flexion.' Glottay 35 (1956), 214-70 and 36 (1957), 48-100.
Fillmore, Charles J. 'The position of embedding transformations in a
grammar.1 Word, 19 (1963), 208-31.
'Deictic categories in the semantics of "come".' FL, 2 (1966), 219-27.
'Towards a modern theory of case.' (Project on Linguistic Analysis,
Report No. 13.) Ohio State University, 1966.
' The case for case/ (To be published in the Proceedings of the 1967 Texas
Conference on Language Universals, edited by Bach, E. and Harms,
R.J.)
Firbas, Jan.' Thoughts on the communicative function of the verb in English,
German and Czech/ BSE, 1 (1959), 39-63.
Firth, J. R. Papers in Linguistics, ig34~igî. London: Oxford University
Press, 1951.
'Ethnographic analysis and language.' In Raymond Firth (ed.), Man and
Culture. London, 1957, pp. 93-118.
Fleisch, H. 'Esquisse d'une histoire de la grammaire arabe.' Arabica (Leiden)
4 (1957), 1-22.
Fodor, J. A. and Katz, J. J. (eds.). The Structure of Language: Readings in
the Philosophy of Language. Englewood Cliffs, N.J.: Prentice-Hall,
1964.
Forbes, P. B. R. 'Greek pioneers in philology and grammar.' CR> 47 (1933),
105-12.
Fourquet, J. Les Mutations Consonantiques du Germanique. Paris, 1948.
Frake, C. O.' The diagnosis of disease among the Subanun.' AmA, 63 (1961),
113-32. (Reprinted in Hymes, Language in Culture.)
Frege, G. 'Uber Sinn und Bedeutung' (1892). Translated as 'On sense and
reference* in Peter Geach and Max Black, Translations from the Philo-
sophical Writings of Gottlob Frege. Oxford, 1952.
Fries, C. C. The Structure of English: An Introduction to the Construction of
English Sentences. New York: Harcourt Brace, 1952 and London:
Longmans, 1957. (References are to the British edition.)
'Meaning and linguistic analysis.' Lg. 30 (1954), 57-68.
Fudge, E. C. 'The nature of phonological primes.' JL, 3 (1967), 1-36.
Gabelentz, Georg von der. Die Sprachwissenschaft. Leipzig, 1891.
Geach, Peter T. Reference and Generality: An Examination of Some Medieval
and Modern Theories. Ithaca, N.Y.: Cornell University Press, 1962.
Ghizzetti, Aldo (ed.). Automatic Translation of Languages. (Papers presented
at a NATO Summer School, Venice, July 1962.) Oxford, London,
New York, etc.: Pergamon Press, 1966.
GilH6ron, Jules and Roques, Mario. Etudes de Geographie Linguistique.
Paris: Champion, 1912.
Gilson, fitienne. History of Christian Philosophy in the Middle Ages. London,
1955. (Translated from La Philosophie au Moyen Age. Paris: Payot,
1922. Revised edition, 1962.)
Gleason, H. A. An Introduction to Descriptive Linguistics. 2nd revised
edition. New York: Holt, Rinehart and Winston, 1961.
496 BIBLIOGRAPHY
Godel, Robert. Grammaire Turque. Geneve: Librairie de l'Universit6, 1945.
Les Sources Manuscrites du Cours de Linguistique Generate de Ferdinand de
Saussure. Geneve: Droz and Paris: Minard, 1957.
Goodenough, Ward H. * Componential analysis and the study of meaning.'
Lg. 32 (1956), 195-216.
Grabmann, M. Mittelalterliches Geistesleben, 3 vols. Munich, 1956.
Graham, A. C. '"Being" in Classical Chinese.' In Verhaar, The Verb 'Be'
and its Synonyms. Part 1, pp. 1-39.
Gray, Louis H. The Foundations of Language. New York: Macmillan, 1939.
Greenberg, Joseph. Essays in Linguistics. Chicago: University of Chicago
Press, 1957.
(ed.), Universals in Language. Cambridge, Mass.: M.I.T. Press, 1963.
Language Universals. (Janua Linguarum, Series Minor, 59.) The Hague:
Mouton, 1966.
Greimas, A. J. Semantique Structurale. Paris: Larousse, 1966.
Gross, Maurice. * On the equivalence of models of language used in the fields
of mechanical translation and information retrieval.' In Ghizzetti,
Automatic Translation, pp. 123-37.
Guiraud, Pierre. La Semantique, Paris: Presses Universitaires de France,
1955-
Haas, William. 'On defining linguistic units.' TPhS (1954), 54-84.
'Linguistic structures.' Word, 16 (i960), 251-76.
Hall, Barbara. Mathematical Foundations for Language. New York and
London: Harper and Row, forthcoming.
Hall, Robert A. Introductory Linguistics. Philadelphia: Chilton, 1964.
Idealism in Romance Linguistics. Ithaca, N.Y.: Cornell University Press,
1964.
Halle, M. 'Phonology in a generative grammar.' Word, 18 (1962), 54-72.
(Reprinted in Fodor and Katz, Structure of Language, pp. 324-33.)
Halliday, M. A. K.' Categories of the theory of grammar.' Word, 17, 241-92.
'Syntax and the consumer.' MSLL, 17 (1964), n - 2 4 .
'Lexis as a linguistic level.' In Bazell et al., In Memory, pp. 148-62.
'Notes on transitivity and theme in English.' JL, 3 (1967) and 4 (1968).
(To be published in three parts.)
Halliday, M. A. K., Mclntosh, A. and Strevens, P. D. The Linguistic
Sciences and Language Teaching. London: Longmans, 1964.
Hallig, R. and Wartburg, W. von. Begriffssystem als Grundlage fiir die
Lexicographie. (Abhandlungen der deutschen Akademie der Wissen-
schaften zu Berlin, 4.) Berlin, 1952.
Hamp, Eric P., Householder, F. W. and Austerlitz, R. Readings in Lin-
guistics II. Chicago and London: Chicago University Press, 1966.
Harris, Zellig S. Methods in Structural Linguistics. Chicago: University of
Chicago Press, 1951. (Reprinted as Structural Linguistics, 1961.)
'Discourse analysis.' Lg. 28 (1952), 18-23 and 474-94.
'Distributional structure.' Word, 10 (1954), 146-94.
'Co-occurrence and transformation in linguistic structure.' Lg. 33 (1957),
283-340.
BIBLIOGRAPHY 497
String Analysis of Sentence Structure. The Hague: Mouton, 1962.
' Transformational theory.' Lg. 41 (1965), 363-401.
Haugen, Einar. 'Linguistics and language-planning.' In Bright, Socio-
linguistics.
Hays, David G. Introduction to Computational Linguistics. (Mathematical
Linguistics and Automatic Language Analysis, 2.) New York: American
Elsevier, 1967.
'Dependency theory: A formalism and some observations.' Lg. 40 (1964),
Heidegger, M. Die Kategorienlehre und Bedeutungslehre des Duns Scotus.

Tubingen: Mohr, 1916.
Henle, Paul (ed.). Language, Thought and Culture. Ann Arbor, Mich.:
University of Michigan Press, 1958.
Hill, Archibald A. Introduction to Linguistic Structures: From Sound to
Sentence in English. New York: Harcourt, Brace and World, 1958.
Hjelmslev, Louis. Principes de Grammaire Generale. (Kng. Danske Viden-
skabernes Selskab., Hist.-Filolog. Medd., 16, i.) Copenhague: Host &
Son, 1928.
La Categorie des Cas. Etude de Grammaire Generale. (Acta Jutlandica
Aarsskrift for Aarhus Universitet, 9.) Aarhus: Universitets-forlaget, 1935.
Essais Linguistiques. (TCLC, 12.) Copenhague: Nordisk Sprog- og
Kulturforlag, 1959.
Prolegomena to a Theory of Language. (Translated from the Danish, 1943,
by Francis J. Whitfield.) Bloomington, Ind.: Indiana University, 1953.
Hockett, C. F. A Course in Modern Linguistics. New York: Macmillan, 1958.
A Manual of Phonology. Bloomington, Ind.: Indiana University Press,
1955.
' T h e quantification of functional load: a linguistic problem.' (Memoran-
dum RM-5168-PR.) Santa Monica, Calif.: RAND Corporation, 1966.
' T w o models of grammatical description.' Word, 10 (1954), 210-33.
(Reprinted in Joos, Readings.)
'Linguistic elements and their relations.' Lg. 37 (1961), 29-53.
' Grammar for the hearer.' In Jakobson, Structure of Language, pp. 220-36.
Hoenigswald, Henry M. Language Change and Linguistic Reconstruction.
Chicago: University of Chicago Press, 1960.
' A proposal for the study of folk linguistics.' In Bright, Sociolinguistics.
Hoijer, Harry. Language in Culture. Chicago: University of Chicago Press,
1954-
Honey, P. J. 'Word classes in Vietnamese.' BSOAS, 18 (1956), 534-44.
Householder, F. W. ' O n linguistic terms.' In Saporta, Psycholinguisticsy
pp. 15-25.
' O n some recent claims in phonological theory.' JL, 1 (1965), 13-34.
' O n the uniqueness of semantic mapping.' Word, 18 (1962), 173-85.
'Descriptive analysis of Latin declension.' Word, 3 (1947), 48-58.
Householder, F. W. and Saporta, S. (eds.). Problems in Lexicography.
(Publications of Indiana Research Center in Anthropology, Folklore
and Linguistics, 21.) Baltimore, 1962.
498 BIBLIOGRAPHY
Huddleston, R. D. 'Rank and depth.' Lg. 41 (1965), 586.
Hughes, John P. The Science of Language: An Introduction to Linguistics.
New York: Random House, 1962.
Humboldt, Wilhelm von. Uber die Verschiedenheit des Menschlischen Sprach-
baues. Berlin, 1836. Republished, Darmstadt: Claasen and Roether,
1949.
Hymes, Dell (ed.). Language in Culture and Society: A Reader in Linguistics
and Anthropology. New York: Harper and Row, 1964.
Jakobson, Roman. Selected Writings I: Phonological Studies. The Hague:
Mouton, 1962.
'Beitrag zur allgemeinen Kasuslehre.' TCLP, 6 (1936), 240-88. (Re-
printed in Hamp et al.f Readings.)
'Boas's view of grammatical meaning/ AmA, 61 (1959), 139-45.
'Linguistics and communication theory.' In Jakobson, Structure of
Language, pp. 245-52.
' O n linguistic aspects of translation.' In Brower, Translation, pp. 232-9.
' Implications of language universals for linguistics.' In Greenberg,
Universals.
(ed.). On the Structure of Language and its Mathematical Aspects. (Pro-
ceedings of 12th Symposium on Applied Mathematics.) Providence,
R.I.: American Mathematical Society, 1961.
Jakobson, R. and Halle, M. Fundamentals of Language. The Hague: Mouton,
1956.
Jespersen, Otto. The Philosophy of Grammar. London: Allen and Unwin,
1929.
Language, Its Nature, Development, and Origin. London: Allen and
Unwin, 1922.
Analytic Syntax. Copenhagen: Munksgaard, 1937.
Jones, Daniel, An Outline of English Phonetics. 8th ed. Cambridge: Heffer,
1956. (First edition 1918.)
The Phoneme: Its Nature and Use. Cambridge: Heffer, 1950.
Joos, Martin (ed.). Readings in Linguistics. Washington, D.C.: American
Council of Learned Societies, 1957. (Republished as Readings in
Linguistics I. Chicago and London: Chicago University Press, 1966.)
The English Verb. Madison and Milwaukee: University of Wisconsin
Press, 1964.
Kachru, Yamuna. Introduction to Hindi Syntax. Urbana, 111.: University of
Illinois, 1966.
Kahn, C. H. ' T h e Greek verb "to b e " and the concept of being.' FL, 2
(1966), 245-65-
Katz, J. J. The Philosophy of Language. New York: Harper and Row, 1966.
'Recent issues in semantic theory.' FL, 3 (1967), 124-94.
'Mentalism and linguistics.' Lg. 40 (1964), 124-37.
Katz, J. J. and Fodor, J. A. ' T h e structure of a semantic theory.' Lg. 39
(1963), 170-210. (Reprinted in Fodor and Katz, Structure of Language.)
Katz, J. J. and Postal, P. M. An Integrated Theory of Linguistic Descriptions.
(Research Monographs, 26.) Cambridge, Mass.: M.I.T. Press, 1964.
BIBLIOGRAPHY 499
Kiefer, F. 'Some semantic relations in natural languages.* FL, 2 (1966),
228-40.
Kollar, H. 'Die Anfange der griechischen Grammatik.' Glotta, 37 (1958),
5-4-
Koutsoudas, Andreas. Writing Transformational Grammars. New York:
McGraw-Hill, 1967.
Kronasser, H. Handbuch der Semasiologie. Heidelberg: Carl Winter Uni-
versitatsverlag, 1952.
Kukenheim, L. Esquisse Historique de la Linguistique Francaise et de ses
Rapports avec la Linguistique Generale. Leiden, 1962.
Kuroda, S.-Y. 'Causative forms in Japanese.' FLy 1 (1965), 30-50.
Kurylowicz, J. Esquisses Linguistiques. Wroclaw-Krakow: Polska Akademia
Nauk, Komitet Jezykoznawczy, i960.
The Inflectional Categories of Indo-European. Heidelberg: Carl Winter
Universitatsverlag, 1964.
* Derivation lexicale et derivation syntaxique.' BSLy 37 (1936), 79-92.
(Reprinted in Esquissesy pp. 41-50; also in Hamp et al.y Readings IIy
pp. 42-50.)
' Ergativnost' i stadial'nost' v jazyke.' Izv. Akad. Nauk SSSRy 5 (1946),
387-93. (Reprinted in Esquisses, pp. 95-103.)
'La notion de l'isomorphisme.' TCLC, 5 (1949), 48-60. (Reprinted in
Esquisses, pp. 16-26.)
'Les temps composes du roman.' PFy 15 (1931), 448-53. (Reprinted in
Esquisses.)
Ladefoged, Peter. Elements of Acoustic Phonetics. Chicago: University of
Chicago Press and Edinburgh: University of Edinburgh Press, 1962.
Lakoff, G. On the Nature of Syntactic Irregularity. (Report No. NSF-16,
Mathematical Linguistics and Automatic Translation.) Cambridge:
Mass.: Harvard University Computation Laboratory, 1965.
Deep and Surface Grammar. Cambridge, Mass.: M.I.T. Press, forth-
coming.
Lamb, Sydney M. Outline of Stratificational Grammar. Washington, D.C.:
Georgetown University Press, 1966.
'On alternation, transformation, realization and stratification.* MSFOu,
17 (1964), 105-22.
'The sememic approach to structural semantics. In Romney and
D'Andrade, Transcultural Studies, pp. 57-78. (Special publication of
AmAy 66 (1964), 57-78.)
Lambek, J. 'On the calculus of syntactic types.' In Jakobson, Structure of
Language, pp. 166-78.
Lees, Robert B. The Grammar of English Nominalizations. Bloomington,
Ind.: Research Center in Anthropology, Folklore and Linguistics,
i960.
The Phonology of Modern Standard Turkish. (Indiana University Publi-
cations: Uralic and Altaic Series, 6.) Bloomington, Ind., 1964.
Leisi, Ernst. Der Wortinhalt: Seine Struktur im Deutschen und Englischen.
Heidelberg, 1953.
500 BIBLIOGRAPHY
Lenneberg, Eric H. and Roberts, John M. The Language of Experience:
A Study in Methodology. (Indiana University Publications in Anthro-
pology and Linguistics.) (Memoir 13 of IjfAL.) Bloomington, Ind.,
1956.
Lepschy, Giulio C. La Linguistica Strutturale. Torino: Einaudi, 1966.
Leroy, Maurice. Les Grands Courants de la Linguistique Moderne. Bruxelles:
Presses Universitaires de Bruxelles and Paris: Presses Universitaires de
France, 1964.
Linsky, Leonard (ed.). Semantics and the Philosophy of Language. Urbana,
111.: University of Illinois Press, 1952.
Longacre, R. E. * Some fundamental insights of tagmemics.' Lg. 41 (1965),
65-76.
Lounsbury, Floyd G. 'A semantic analysis of the Pawnee kinship usage.'
Lg- 32 (1956), 158-94-
'The structural analysis of kinship semantics.* In Proceedings of the Ninth
International Congress of Linguists (edited by H. G. Lunt), pp. 1073-93.
The Hague: Mouton, 1964.
Luce, R. D., Bush, R. R. and Galanter, E. (eds.). Handbook of Mathematical
Psychology. Volume 2, chapters 9-14. New York and London: Wiley,
1963.
(eds.). Readings in Mathematical Psychology. Volume 2. New York and
London: Wiley, 1965.
Lyons, John. Structural Semantics. (Publications of the Philological Society,
20.) Oxford: Blackwell, 1963.
* Firth's theory of "meaning".' In Bazell et al., In Memory, pp. 288-302.
'Phonemic and non-phonemic phonology.' IJAL, 28 (1962), 127-33.
'Towards a "notional" theory of the "parts of speech".'JX, 2 (1966),
209-36.
'A note on possessive, existential and locative sentences/ FL, 4 (1967).
Mclntosh, Angus. An Introduction to a Survey of Scottish Dialects. Edin-
burgh: Nelson, 1952.
'Patterns and ranges.' Lg. 37 (1961), 325-37. (Reprinted in Mclntosh and
Halliday, Patterns of Language, pp. 183-99.)
'Predictive statements.' In Bazell et al., In Memory, pp. 303-20.
Mclntosh, Angus and Halliday, M. A. K. Patterns of Language: Papers in
General, Descriptive and Applied Linguistics. London: Longmans, 1966.
Magnusson, R. Studies in the Theory of the Parts of Speech. (Lund Studies in
English, 24.) Lund: Gleerup and Copenhagen: Munksgaard, 1954.
Malinowski, B. 'The problem of meaning in primitive languages.' Supple-
ment 1 to Ogden and Richards, Meaning of Meaning, pp. 296-346.
Malmberg, Bertil. Structural Linguistics and Human Communication.
Heidelberg: Springer, 1963.
Les Nouvelles Tendances de la Linguistique. Paris: Presses Universitaires
de France, 1966. (Translated from the Swedish, 1962.)
Mandelbrot, B.' On the theory of word-frequencies and on related Markovian
models of discourse.' In Jakobson, Structure of Language, pp. 190-
219.
BIBLIOGRAPHY 5OI
Martinet, Andre*. Elements deLinguistique Genet ale. Paris: Armand Colin, i960.
(English translation, Elements of General Linguistics. London: Faber, 1964.)
Economie des Changements Phonetiques. Berne: A. Francke, 1955.
A Functional View of Language. Oxford: Clarendon Press, 1962.
La Linguistique Synchronique: Etudes et Rechetches. Paris: Presses Uni-
versitaires de France, 1965.
'Le genre f^minin en indo-europe*en: examen fonctionnel du probleme.'
BSLy 52 (1956), 83-95.
'La construction ergative et les structures e'le'mentaires de l'^nonceV
Journal de Psychologie Normale et Pathologique (1958), pp. 377-92.
(Reprinted in Linguistique Synchronique, pp. 206-22.)
Matore*, G. La Methode en Lexicologie. Domaine Frangais. Paris: Didier, 1953.
Matthews, G. H. Hidatsa Syntax. (Papers in Formal Linguistics, 3.) The
Hague: Mouton, 1965.
Matthews, P. H. 'Transformational grammar.' ArchL, 13 (1961), 196-209.
'Problems of selection in transformational grammar.' JL, 1 (1965), 35-47.
'The inflexional component of a word-and-paradigm grammar.' JL, 1
(1965), 139-71-
'The concept of rank.'^L, 2 (1966), 101-10.
'Latin.' Lingua, 17 (1967), 163-81.
Review of Chomsky, Aspects, in JL, 3 (1967), 119-52.
Matthews, W. K. 'The ergative construction in modern Indo-Aryan.'
Lingua, 3 (1953), 391-406.
Meillet, Antoine. Linguistique Historique et Linguistique Generale. 2 vols.
Paris: 1926 and 1938.
Miller, George A. Language and Communication. New York: McGraw-Hill,
Misra, Vidya Niwas. The Descriptive Techniques of Pdnini. (Janua Linguarum,
Series Practica, 18.) The Hague: Mouton, 1967.
Mohrmann, Ch., Sommerfelt, A. and Whatmough, J. Trends in European and
American Linguistics igjo-ig6o. Utrecht: Spectrum, 1961.
Mohrmann, Ch., Norman, F. and Sommerfelt, A. Trends in Modern
Linguistics. Utrecht: Spectrum, 1963.
Moravcsik, J. M. E. 'The analytic and the nonempirical.' Journal of Philo-
sophy, 62 (1965), 4I5-29-
Morris, Charles W. Signs, Language, and Behavior. Englewood Cliffs, N J . :
Prentice-Hall, 1955.
Signification and Significance. Cambridge, Mass.: M.I.T. Press, 1964.
Mounin, Georges. Histoire de la Linguistique des Origines au XXe Siecle.
Paris: Presses Universitaires de France, 1967.
Needham, R. M. 'Automatic classification in linguistics.' The Statistician,
17 (1967), 45-53-
Nida, Eugene A. Morphology: A Descriptive Analysis of Words. 2nd ed.
Ann Arbor, Mich.: University of Michigan Press, 1949.
A Synopsis of English Syntax. Norman, Okla.: Summer Institute of
Linguistics, i960. (Republished, The Hague: Mouton, 1966.)
Towards a Science of Translating. Leiden: Brill, 1964.
5O2 BIBLIOGRAPHY
Ogden, C. K. Opposition. London: Kegan Paul, 1932.
Ogden, C. K. and Richards, I. A. The Meaning of Meaning. 8 th edition.
London: Routledge and Kegan Paul, 1946. (First edition, 1923.)
Ohman, Suzanne. Wortinhalt und Weltbild. Stockholm, 1951.
'Theories of the "linguistic field"/ Word, 9 (1953), 123-34.
Olsson, Yngve. On the Syntax of the English Verb: With Special Reference to
1
Have a Look' and Similar Complex Structures. (Gothenburg Studies in
English, 12.) Gothenburg, Stockholm and Uppsala: Almqvist and
Wiksell, 1961.
Orr, John. Words and Sounds in English and French. Oxford, 1953.
Ota, Akira, Tense and Aspect of Present-Day American English. Tokyo:
Kenkyusha, 1963.
Palmer, F. R. 'Linguistic hierarchy.* Lingua, 7 (1958), 225-41.
' " Sequence' and "order".' MSLL, 17 (1964), 123-30.
A Linguistic Study of the English Verb. London: Longmans, 1965.
(ed.). ? Prosodic Analysis. London: Oxford University Press, forthcoming.
Pedersen, Holger. Linguistic Science in the Nineteenth Century, trans. J. W.
Spargo. Cambridge, Mass.: Harvard University Press, 1931. (Repub-
lished as The Discovery of Language. Bloomington, Ind.: Indiana
University Press, 1959.)
Pike, Kenneth L. Language in Relation to a Unified Theory of Human
Behavior. (Janua Linguarum, Series Maior, 24.) 2nd revised edition.
The Hague: Mouton, 1967.
Phonetics. Ann Arbor, Mich.: University of Michigan Press, 1943.
Pohlenz, M. Die Begriindung der Abendldndischen Sprachlehre durch die Stoa.
Gottingen: Vandenhoeck und Ruprecht, 1939.
Porzig, W. Das Wunder der Sprache. Berne, 1950.
Postal, Paul M. Constituent Structure: A Study of Contemporary Models of
Syntactic Descriptions. Bloomington, Ind.: Indiana University Publi-
cations in Folklore and Linguistics, 1964.
Postal, P. M. and Rosenbaum, P. S, English Sentence Formation: Recent
Advances in Transformational Analysis. Reading, Mass.: Addison-
Wesley, forthcoming.
Pulgram, E. * Graphic and phonic systems: figurae and signs.' Word, 21
(1965), 208-2^..
Quine, Willard V. From a Logical Point of View. Cambridge, Mass.: Harvard
Word and Object. Cambridge, Mass.: M.I.T. Press, i960.
Quirk, Randolph. The Use of English. (With supplements by A. C. Gimson
and J. Warburg.) London: Longmans, 1962.
Read, A. W. 'An account of the word "semantics".' Word, 4 (1948), 78-97.
Reichenbach, Hans. Elements of Symbolic Logic. London and New York:
Macmillan, 1947.
Roberts, A. Hood. A Statistical Linguistic Analysis of American English.
(Janua Linguarum, Series Practica, 8.) The Hague: Mouton, 1965.
Robins, R. H. General Linguistics: An Introductory Survey. London:
Longmans, 1964.
BIBLIOGRAPHY 503
A Short History of Linguistics. London: Longmans, 1967.
* The development of the word class system of the European grammatical
tradition.' FLy 2 (1966), 3-19.
Roget, Peter M. Roget's Thesaurus. Abridged edition, with additions, by
J. L. Roget and S. R. Roget. Harmondsworth, Middlesex: Penguin
Books, 1953. (Original edition, 1852.)
Romney, A. K. and D'Andrade, R. G. 'Cognitive aspects of English kin
terms.' Am A, 66 (1964), 146-70.
Roos, H. ' Sprachdenken im Mittelalter.' C & M, 9 (1949).
Die Modi Significandi des Martinus von Dacia. Miinster, 1952.
Rosengren, Inger. Semantische Strukturen: Eine Quantitative Distributions -
analyse einiger Mittelhochdeutscher Adjektive. (Lunder Germanistische
Forschungen, 38.) Lund: Gleerup und Copenhagen: Munksgaard, 1966.
Rosetti, A. Linguistica. (Janua Linguarum, Series Maior, 16.) The Hague:
Mouton, 1965.
Russell, Bertrand. An Inquiry into Meaning and Truth. London, 1940.
'On denoting.' Mindy 14 (1905), 479-93.
Salomon, Louis B. Semantics and Common Sense. London and New York:
Holt, Rinehart and Winston, 1964.
Sandmann, M. Subject and Predicate. (Edinburgh University Publications,
Language and Literature, 5.) Edinburgh, 1954.
Sandys, J. E. History of Classical Scholarship. Cambridge, 1903.
Sapir, Edward. Language: An Introduction to the Study of Speech. New York:
Harcourt, Brace and World, 1921.
Selected Writings in Language^ Culture and Personality. (Edited by D. G.
Mandelbaum.) Berkeley, Calif.: University of California Press, 1949.
Saporta, S. (ed.). Psycholinguistics: A Book of Readings. New York: Holt,
Rinehart and Winston, 1961.
Saussure, F. de. Cours de Linguistique Generale. 5th edition. Paris: Payot,
I
955 (First edition 1916.) (English translation, by Wade Baskin,
Course in General Linguistics. New York: Philosophical Library, 1959.)
Sauvageot, Aur^lien. Esquisse de la Langue Finnoise. Paris: Klincksieck, 1949.
Schuchardt, H. Hugo Schuchardt Brevier, 2nd ed. (Edited by L. Spitzer.)
Halle a. S., 1928.
Sebeok, Thomas. Current Trends in Linguistics. Volume 3: Theoretical
Foundations. The Hague: Mouton, 1966.
Sechehaye, Albert. Essai sur la Structure Logique de la Phrase. Paris: Honore*
Champion, 1926.
Shaumjan, S. K. Strukturnaja Lingvistika. (Akademija Nauk S.S.S.R.,
Institut Russkogo Jazyka.) Moskva: Izdatel'stvo 'Nauka', 1965. (An
English edition is in preparation.)
Spang-Hanssen, H. ' Glossematics.' In Mohrmann, Sommerfelt and
Whatmough, Trends.
Sparck Jones, K. Synonymy and Semantic Classification. Cambridge: Cam-
bridge Language Research Unit, 1964.
Strang, Barbara M. H. Modern English Structure. London: Edward Arnold
and New York: St Martin's Press, 1962.
504 BIBLIOGRAPHY
Strawson, P. F. Individuals: An Essay in Descriptive Metaphysics. London:
Methuen, 1959.
Strevens, P. D. (ed.). Five Inaugural Lectures. Oxford: Oxford University
Press, 1966.
Sturtevant, Edgar H. An Introduction to Linguistic Science. New Haven,
Conn.: Yale University Press, 1949.
Linguistic Change. Chicago: University of Chicago Press, 1961.
Sturtevant, William C. 'Studies in ethnoscience.' Am A, 66 (1964), 99-
124.
Svartvik, Jan. On Voice in the English Verb. (Janua Linguarum, Series
Practica, 63.) The Hague: Mouton, 1966.
Swift, Lloyd B. A Reference Grammar of Modern Turkish. (Indiana Uni-
versity Publications, Uralic and Altaic Series, 19.) Bloomington:
Indiana University and The Hague: Mouton, 1963.
Teeter, Karl V. 'Descriptive linguistics in America: Triviality vs. Irrele-
vance.' Word, 20 (1964), 197-206.
Tesniere, L. Elements de Syntaxe Structurale. Paris: Klincksieck, 1959.
Thompson, Laurence C. A Vietnamese Grammar. Seattle, Washington:
University of Washington Press, 1965.
Thorndike, E. L. and Lorge, I. The Teacher's Word Book of 30,000 Words,
New York: Columbia University Press, 1944.
Togeby, Knud. 'Qu'est-ce qu'un mot.' TCLC, 5 (1949), 99-111.
Trier, Jost. Der Deutsche Wortschatz im Sinnbezirk des Verstandes. Heidel-
berg: Carl Winter, 1931.
Trubetzkoy, N. S. Grundztige der Phonologie. Prague: Cercle Linguistique
de Prague, 1939. (French edition, Principes de Phonologie, trans. Jean
Cantineau. Paris: Klincksieck, 1949.)
Uhlenbeck, C. C.' Agens und Patiens im Kasussystem der indogermanischen
Sprachen.' IF, 12 (1901), 170-2.
Uldall, H. J. Outline of Glossematics ( = TCLC, 10). Copenhagen, 1957.
* Speech and writing.' AL, 4 (1944), 11-16. (Reprinted in Hamp et al.,
Readings, pp. 147-51.)
Ullmann, Stephen. The Principles of Semantics. 2nd ed. Glasgow: Jackson
and Oxford: Blackwell, 1957.
Semantics: An Introduction to the Science of Meaning. Oxford: Blackwell,
1962.
Vachek, Josef. A Prague School Reader in Linguistics: Studies in the History
and Theory of Linguistics. Bloomington, Ind.: Indiana University Press,
1964.
'Some remarks on writing and phonetic transcription.' AL, 5 (1945),
88-93. (Reprinted in Hamp et al., Readings, pp. 152-7.)
Vaillant, Andre*. 'L'ergatif indo-europe"en.' BSL, 37 (1936), 93-108.
Vendryes, Joseph. Le Langage. Paris: Albin Michel, 1923. English trans-
lation, Language. London: Kegan Paul and New York: Knopf, 1931.
Verhaar, John W. M. (ed.). The Verb ' Bey andits Synonyms. (Foundations of
Language, Supplementary Series.) Volume 1. Dordrecht: D. Reidel,
1967.
BIBLIOGRAPHY 505
Vogt, Hans. Esquisse d'une Grammaire de Georgien Moderne. In NTS, 9
(1938), 5-114 and 10 (1939), 5-188.
Wackernagel, J. Vorlesungen ilber Syntax. Basel: Emil Birkhauser, 1920.
Wackernagel, J. and Debrunner, A. Altindische Grammatik. (With an
Introduction by Renou.) 1957.
Wallace, F. C. and Atkins, J. ' The meaning of kinship terms.' AmA, 62
(i960), 58-80.
Ward, Dennis. The Russian Language Today. London: Hutchinson, 1965.
Warnock, G. J. English Philosophy Since igoo. Oxford, 1958. (Home
University Library of Modern Knowledge, no. 234.) Oxford, 1958.
Wartburg, Walther von. Einfiihrung in Problematik und Methodik der Sprach-
wissenschqft. 2nd ed. Tubingen: Max Niemeyer, 1962. French trans-
lation, Problemes et Methodes de la Linguistique. Paris: Presses
Universitaires, 1963.
Evolution et Structure de la Langue Franfaise. 5th ed. Berne: Francke, 1958.
Waterson, N. ' Some aspects of the phonology of the nominal forms of the
Turkish word.' BSOAS, 18 (1956), 578-91. (Reprinted in Palmer,
Prosodic Analysis.)
* Numeratives in Uzbek.' In Bazell et al., In Memory, pp. 454-74.
Weinreich, U. * Explorations in semantic theory.' In Sebeok, Current Trends,
vol. 3.
Wells, Rulon S. 'Immediate constituents.' Lg. 23 (1947), 81-117. (Re-
printed in Joos, Readings, pp. 186-207.)
* Meaning and use.' Word, 10 (1954), 235-50.
Whorf, Benjamin L. 'Grammatical categories.' Lg. 21 (1945), 1-11. (Re-
printed in Language, Thought and Reality, pp. 87-101.)
Language, Thought and Reality: Selected Papers. (Edited by John B.
Carroll.) New York: Wiley, 1956.
Wittgenstein, L. Philosophical Investigations. Oxford: Blackwell and New
York: Macmillan, 1953.
Ziff, Paul. Semantic Analysis. Ithaca, N.Y.: Cornell University Press, i960.
Zipf, G. K. Human Behavior and the Principle of Least Effort. Cambridge,
Mass.: Addison-Wesley, 1949.
ADDENDA
Apresjan, Ju. D. EksperimentaVnoje Issledovanije Semantiki Russkogo Glagola.
Moskva: Izdatel'stvo 'Nauka', 1967.
Chomsky, N. 'Remarks on nominalization.' To appear in Jacobs and Rosen-
baum (eds.), Readings in English Transformational Grammar.
Householder, F. W. Forthcoming book. Englewood-Cliffs, N. J.: Prentice-
Hall.
Ruwet, N. Introduction a la Grammaire Generative. (Recherches en Sciences
Humaines, 22.) Paris: Plon, 1967.
TABLE OF SYMBOLS
AND NOTATIONAL CONVENTIONS
* asterisk: (i) 'reconstructed form', 31
(2) ungrammatical, or unacceptable, expression, 142
( ) parentheses: semantic component ('marker'), 471
{ } brace brackets: (1) morpheme, 184
(2) extensional definition of a class, 77
[ ] square brackets: (1) phonetic transcription, 60, 101
(2) grammatical feature, 166
/ / obliques: (1) expression-elements, 60
(2) phonemic transcription, 101
italics: orthographic representation (or transcription), 60
CAPITALS: lexeme, 197
+ plus-sign: (1) concatenation, 91, 116
(2) positive value of binary variable, 123
minus-sign: negative value of binary variable, 123
< 'is less than', 192
> 'is greater than', 92
= equals-sign: (1) 'is equal (equivalent) to', 77
(2) identity of reference, 389
=t= 'is not equal (equivalent) to', 77
epsilon: class-membership, 161, 389
-> arrow: (1) 'develops diachronically into', 31
(2) 'is to be rewritten as', 162
<= inclusion-sign: 'is included in', 389
=> inclusion-sign: 'includes', used for 'implies', 445
= equivalence-sign: bilateral implication, 450
[506]
INDEX OF PROPER NAMES
Abercrombie, 483, 484 255-69, 329-33, 343, 37i, 380, 385,

Ajdukiewicz, 227, 328 439, 44O, 473-4, 482, 483, 484, 485,
Allen, 482, 487 486, 487, 488, 489
Anderson, 487 Cicero, 13, 16
Andre, 488 Cohen, 487, 488
Apollonius, 12 Collart, 482
Apostel, 483 Conklin, 431-2, 488, 489
Arens, 482 Copleston, 482
Aristotle, 6, 11, 14, 270-4, 289, 323-4 Cratylus, 4
Ashton, 486 Crystal, 486
Austin, 487
Dante, 17
Bach, 484, 485, 486 Darwin, 33
Bacon, 16, 333 Dauzat, 484
Bally, 486 Dihle, 482
Bar-Hillel, 227, 485, 487, 488 Dinneen, 482, 483
Barth, 482 Dionysius, 12, 13, 319, 326
Barwick, 482 Donatus, 14-15, 17
Bazell, 485, 488 Durkheim, 51
Belvalkar, 482
Bendix, 487, 489 Ebeling, 489
Benveniste, 392, 486, 487 Ehrman, 486
Bierwisch, 476, 489 Emeneau, 482, 486
Black, 488 Engler, 483
Bloch, 131 Entwistle, 486
Bloomfield, 2, 172-3, 176, 202-1, 211- Erasmus, 17
12, 230, 233-5, 3" 408, 482, 485,
486, 487 Fant, 484
Boas, 486 Fehling, 482
Bolinger, 489 Fillmore, 329, 486, 487
Breal, 487 Firth, 416,488
Brondal, 486 Fodor, 439, 480
Brough, 482 Forbes, 482
Brower, 487 Fourquet, 484
Brown, 487 Frege, 488
Bruner, 487 Fries, 435, 438, 485, 487, 488
Burling, 489 Fudge, 484
Bursill-Hall, 482
Gilli^ron, 90, 484
Caesar, 13 Gilson, 482
Carnap, 388 Gleason, 482, 483
Carroll, J. B., 140, 459, 487, 488 Godel, 483, 485, 486
Carroll, Lewis, 120 Goethe, 195
Caton, 487 Goodenough, 489
Chao, 486 Grabmann, 482
Cherry, 483 Graham, 487
Chomsky, 51, 136-7, I54~5> *57, l 6 2, Gray, 486
164-5, 169, 215, 234, 237-9, 244, 253, Greenberg; 483, 485
[507]
508 INDEX
Grimm, 27-8, 123-6, 484 Lepschy, 482, 483
Gross, 485 Leroy, 482, 483
Guiraud, 488 Lesniewski, 227, 328
Linsky, 487
Haas, 485, 487 Longacre, 485
Hall, B., 485 Lounsbury, 489
Hall, R., 482 Luce, 485
Halle, 484 Lyons, 484, 486, 487, 488, 489
Halliday, 206, 336, 366, 484, 485, 487,
488 Mclntosh, 484, 486, 488
Hallig, 489 McKerrow, 372
Harris, 131, 155, 157, 172, 247-8, 484, Malinowski, 417, 488
485, 486 Malmberg, 483, 484
Haugen, 483 Mandelbrot, 483
Hays, 485 Martinet, 482, 483, 484, 486, 487,
Heidegger, 482 488
Henle, 488 Matthews, P. H., 485, 486
Herder, 24 Matthews, W. K., 487
Hill, 482, 485, 487 Meillet, 486
Hjelmslev, 326-8, 333, 474, 483, 486, Miller, 483, 487
489 Misra, 482
Hockett, 121, 131, 241, 247-8, 311, 315, Moravcsik, 462, 489
335-6, 482, 483, 484, 48S, 486, 487 Mounin, 482
Hoenigswald, 483
Honey, 486 Needham, 488
Householder, 483, 485 Nida, 485, 487
Huddleston, 485
Humboldt, 24, 429, 433 Ogden, 489
Ohman, 488
Jakobson, 120, 474, 484, 486 Orr, 484
Jespersen, 134, 305, 326-8, 333, 482, Ota, 486
484, 486
Jones, Daniel, 103, 104, 484 Palmer, 316, 483, 484,486
Jones, Sir William, 24 Panini, 19-20
Joos, 486 Pedersen, 482
Petrarch, 16
Kahn, 487 Pike, 206, 484, 485
Katz, 439, 441, 473-7, 480, 486, 487, Plato, 3, 6, I O - I I , 323-4, 334, 33^,
488, 489 401, 466
Kiefer, 489 Pohlenz, 482
Roller, 482 Postal, 439, 441, 480, 485, 486
Koutsoudas, 486 Priscian, 14-15, 326
Kronasser, 487 Protagoras, 10-11
Kukenheim, 482 Pulgram, 483
Kurytowicz, 486
Quine, 488
Ladefoged, 484 Quirk, 484
Lakoff, 329, 486
Lamb, 207, 485, 486, 489 Rask, 26
Lambek, 227, 485 Read, 487
Lear, 120 Reichenbach, 488
Leas, 484 Renou, 482
Leibniz, 273 Richelieu, 18
Lenneberg, 488 Roberts, 483, 484, 488
INDEX 509
Robins, 350, 482, 483, 484, 485, 486, Tesniere, 485
487 Thompson, 486
Roget, 488 Thorndike, 484
Romney, 489 Togeby, 485
Roos, 482 Trubetzkoy, 120, 484
Rosenbaum, 329, 486
Rosetti, 485 Uhlenbeck, 487
Russell, 140, 484, 488 Uldall, 483
Ullmann, 447-9, 484, 485, 487, 488
Salomon, 487, 488
Sandmann, 487 Vachek, 483
Sandys, 482 Vaillant, 487
Sapir, 191, 334-5, 4^9, 433, 4^3, 4^5, Varro, 8, 13, 482
469, 483, 485, 487, 488, 489 Vendryes, 486
Saussure, 38, 45-6, 51, 56, 58-9, 67, 139, Verner, 29, 126
176, 177, 184, 271, 443, 482, 483, 485 Virgil, 14
Sauvageot, 486
Scherer, 28 Wackernagel, 482
Sechehaye, 487 Wallace, 489
Shaumjan, 248, 329, 486 Ward, 486
Spang-Hanssen, 483 Warnock, 487
Sparck Jones, 488 Waterson, 484, 486
Staal, 488 Weinreich, 476, 489
Strang, 486 Wells, 485, 487
Strawson, 487 Whorf, 433
Strevens, 484 Wittgenstein, 410, 487
Svartvik, 383, 487
Swift, 486 Zipf, 94
INDEX OF SUBJECTS
ablative, 289 ff. analysis (v. synthesis), 158ff.,217, 234

absolute (v. relative) point of reference, analytic (v. synthetic) implication,
299 444ff.,462
abstract noun, 337 ff. analytic language-type: cf. isolating
abstraction, 404 anaphora: cf. reference
abstractness of linguistic theory, 70 Anglo-Saxon, 17
accent, 29 animal communication, 416
acceptable, 137ff.,150, 172, 421 ff. animate noun, 165ff.,293 f., 296, 341,
accidence, 195, 198 355 ff, 357
accidents, 16, 198, 271 ff., 320, 333 anomaly: cf. analogy
accusative, 174, 223, 290 ff., 340ff.,355 anthropocentric organization of vocabu-
acoustic phonetics, n o f. lary, 457
acquisition, 301, 397 ff. anthropology, 400, 477 f.
active (v. passive), 12, 257, 296ff.,343, antonymy, 407, 460 ff.
372, 468 aorist, 314 f.
active (v. stative): cf. stative, state apical, 105
actual (v. potential) words, 119 f. application, 434, 457ff.,468
adequacy, strong, 231, 246 ff., 261, 380 Arabic, 19, 22
adequacy, weak, 246ff.,261 arbitrariness of realization, 63 f.
adjective, n , 147, 289 ff., 296, 316, archiphoneme, 116 f.
323 ff., 326, 391 Aristotelian philosophy, 198, 270 ff.,
adjunct, 334, 343, 344 ff-, 37*, 39 289, 337 ff.
adnominal, 295 ff., 326, 391 Armenian, 18
adverb, 12, 278 f., 297, 325 ff. article (definite v. indefinite), 279, 392
affective: cf. emotive articulation, 102
agent, 295 ff., 34ff->35 ff. articulator, 105
agent-oriented, 366 f. articulatory phonetics, 120 ff.
agentive, 295 f., 343, 356, 374, 399 Aryan, 24
agentive objects, 365ff.,383 aspect, 300, 313 ff., 397 ff.
agentless sentences, 378 ff. aspiration, 113 f., 123 ff.
agglutinating, 187 ff. assertion, 445 ff.
agreement, 148, 239 associations (or connotations), 427 f.,
Alexandrians, 8-9, 12, 313 448 f.
Algonquian, 22 attitude of speaker, 307 ff.
alienable (v. inalienable), 301, 394 attraction, 279 f.
allomorph, 184ff.,353 attributive, 297, 389, 397 ff.
allophone, 101, 112 ff. auditory phonetics, 102
alphabetic writing, 39 auxiliary symbol, 230, 320, 421
Altaic, 22 auxiliary verb, 256, 306 ff., 354, 374
alternative rules, 217 f
alveolar, 105 back vowel, 103, 128 ff.
ambiguity, 212 ff., 249ff.,252, 370, 420 Bantu, 22, 284
American Indian languages, 22, 199, 'barbarisms', 13
299, 3 i i base-component (of transformational
analogical creation, 30 f. grammar), 255
analogy, 6-8, 30-3, 36-8, 182, 405, 471 base form, 20
analysable, 257 ff., 262 f., 268 basic vocabulary, 26
[510]
INDEX 511
Basque, 17, 341 comitative, 295 ff.
behaviourism, 416 f. command, 307
benefactive, 374, 396 comment: cf. topic
bilabial, 105, 121 ff. common core, 140
bilingualism, 434, 458 common gender, 287
binary systems, 85-7, 127 common noun, 337 f.
bit (binary digit), 85, 454 communication-theory: cf. information-
blasphemy, 140, 423 theory
borrowing, 25-6, 30, 32 comparative grammar, 22
bound forms, 201 comparative philology, 21 ff., 33
bracketing, 211 ff., 249 comparison, 436 ff.
Breton, 21 competence (v. performance), 51
broad transcription, 100 complement, 320, 345 ff., 390
Bulgarian, 204 complementarity (of meaning), 460 ff.,
476
cancellation, 227 ff. complementary distribution, 70, 73, 112
Cantonese, 41 completion of action: cf. perfective
cardinal vowel, 103 f. complex sentence, 178, 224 f., 265 f.,
case, 12, 270 ff., 289 ff., 340 ff., 344, 320
354 component, semantic, 168, 454, 474 ff.
categorial grammar, 227 ff., 327 ff. componential analysis, 474 ff.
category, 270 ff., 332, 337, 421 composition, 53, 171, 207
causative, 296, 352 ff., 359 ff., 381, 398, compound sentence, 265 f.
439 computers, 86, 160
Celtic, 21, 392 concatenation, 209, 228
certainty, 308 concept, 401 ff., 408, 443 f., 473 ff.
change, 34, 300 conceptualism, 408, 474 ff.
channel (of communication), 87 ff., 96 concord, 239 ff., 280, 342 f., 354 ff.
characterize, 146, 156 conditional probabilities, 91 ff.
chess (Saussure's analogy), 46, 59, 61 congruence of grammar and semantics,
Chinese, 22, 41, 76, 188, 273, 283, 288, 166 ff.
301, 324, 393 congruence of levels, 180, 205
choice, 89, 413 f., 421, 437 conjoining, 265
circularity of semantics, 410, 434 conjugation, 195, 272
circularity of traditional definitions, conjunction, 11, 266
318 ff. connotation, 403, 449
class-inclusion, 389, 453 f. consonant, 102, 104 ff.
class-logic, 453 f. constant (v. variable), 237 ff., 261 f.
class-membership, 389 constituent, 145, 209 ff., 249, 265 ff.,
classical fallacy, 9 268, 321
classicism, 9, 23 constituent string, 265
classification (v. generation), 234 construction: cf. syntagm
clause, 53, 170 f., 170, 225, 266, 345 content-plane, 54, 56
clear lateral, 112 content-substance, 56, 429 ff.
close vowel, 103 context, 74, 91 ff., 143, 173, 235 ff,
closed (v. open) set of alternatives, 436, 307, 339, 361, 410, 4i3, 419 , 427 f-
438 context of situation, 413
cognitive (v. emotive) meaning, 403, qontext-dependent synonymy, 448,
427, 448 ff 452 f-, 455
cognitive reality, 477 ff. context-free rules, 226, 235 ff., 242 ff.
cohesion, internal, of words, 202 ff. context-sensitive rules, 235 ff., 244 ff.,
collective, 282 262
colour-terms, 56-8 contingent (v. necessary), 301, 445
combinatorial relations, 67, 210 continuum, 103, 430 f., 458, 460
INDEX
convention, 4, 7, 272, 403 diminishing returns, 152 ff., 160, 164

converseness (of meaning), 467 ff. directional (v. locative), 300, 302 ff.,
coordinating conjunction, 266 35O, 397 ff.
coordination, 178, 222, 233, 266, 340 ff., discontinuous constituent, 223 f.
344 discovery, 151, 157
copula, 320, 322 ff., 346 if., 349, 388 discreteness of elements, 67
corpus, 138, 144 distinctive feature, 120 ff., 126 f., 185
correctness, 9, 42-4, 245 ff. distribution, 70 ff., 143 ff., 157, 182 f.,
correlation, 53 212, 231 ff.
correspondences between languages, 26 diversification, 253 ff.
cost-effectiveness, 160 domination, 259, 266
countable, 281 ff. dorsum, dorsal, 105
cultural overlap, 433 f., 458, 468 double articulation of language, 54, 142
cultural words, 26 double-base: cf. generalized
culture and language, 431 ff., 457 f., 479 doublets, 30, 47
Czech, 21, 205 dual number, 283
dummy, 324, 346, 349, 393, 421, 456
dark lateral, 112 durative, 315
dative, 289 ff. Dutch, 21, 35
decidability, 156 dynamic (v. static), 300, 397 ff., 437
decision-procedure, 156
declarative, 307 ff., 441 f. element, 272
declension, 190, 195, 272, 289 f. elliptical: cf. incomplete
deep structure, 207, 247 ff., 254, 265, embedding, 225, 253, 265 f., 347, 381 ff.
268 f., 281, 288, 303, 319, 322, 324, emotive: cf. cognitive
333, 347, 39O ff., 421 ff., 437, 47 empiricism, 424 ff., 443 f.
definite (v. indefinite), 276, 296 f., 353, endocentric, 231 ff., 234, 241, 296 f.
39O, 392 English, 21, passim
degree, 327 ff. entail, 403
deictic, deixis, 275 ff., 300, 305, 390, 413 epistemology, 15, 400 f.
deletion-rule, 237, 269 equative, 389
demonstrative pronoun, 240, 278 ff., equivalence (of distribution), 70
300 equivalence (of sense), 450
denial, 445 ff. equivalence, strong, 226 f.
denominator, 229 equivalence, weak, 226 f.
denotation, 403, 425 ergative, 342, 350 ff.
dental, 104, 121 ff. Eskimo, 188, 299, 341 ff., 378
dependent, 231, 235, 241, 344 essence, 198, 401
deponent, 377 etymology, 4-5, 406
derivation (v. inflexion), 195 f. event, 348
derived phrase-marker, 268 evolution of language, 23, 32-3, 191
derived sentence, 173 f. exclusive first person, 277 f.
derived structure, 250 existence and reference, 425 f., 432
descriptive (v. prescriptive) grammar, existential, 388 ff.
7, 42-4 exocentric, 231 ff., 241, 296
determinism, 432 ff. explanation, 267
diachronic, 45 ff., 90, 356, 396, 407 explicit, 136, 156
diacritics, 109 exponent, 184
dialect, 34-5 expression (v. content), 54
diathesis, 372 expression-elements, 61 ff., 73 ff., 81 ff.,
dichotomization, 461, 466, 469 f. 120
dictio, 272, 404 extension: cf. intension
dictionary, 53, 158, 400, 406 exterior (v. interior) case, 299 ff., 303
difference of meaning, 459 extranuclear: cf. nucleus
INDEX 513
factorization, 183, 471 gerund, 249 f.
families of languages, 21-2 given (v. new), 335 ff.
feature, grammatical, 164 f., 332, 421 glossocentric definitions, 319
feedback, 90, 111, 160 glottal stop, 115
feminine, 11, 168, 283 ff., 293 goal, 341 ff., 350 ff.
field theory of semantics, 429 ff. Gothic, 24, 27-9
figures of speech, 13 government, 239 ff.
Fijian, 283 grading, 463
Finnish, 22, 299 ff., 301, 303 grammar, 9, 53-4, 131, 133 ff., 146,
Finno-Ugrian, 22 passim
first-order nominals, 347 ff. grammatical (v. lexical), 317, 346, 402,
form (v. matter), 198, 271 ff. 421, 435 ff.
form (v. meaning), 53, 136, 377, 403 ff. grammatical (v. meaningful), 133, 137,
form (v. substance), 54 ff., 99, 136, 184, 140 ff., 145, 148 ff., 154 f., 172, 349,
198 f., 429 ff. 43i f-
form (of words), 194,1966^,272,372,403 graphic substance, 62 ff.
formal (y. informal), 136 f. Greek, 4-12, 13, 17, 21, 24, 26, 27-9,
formal {v. intimate, familiar), 280 31, 41, 179, 205, 254, 297, 304 f->
formal (v. notional), 134 ff., 147 f., 152, 312 f., 314, 322, 358, 373 ff., 431, 457
179, 270, 318 f. Greek grammarians, 10 ff., 283
formal (v. semantic), 134 ff. Grimm's law, 27-30, 123 ff.
formal (v. substantial), 136 f., 184
formal (v. substantive), 136 f. habitual, 309, 315, 367
formalism, 51 Hanunoo, 431
formalization, 136, 157 having meaning, 402, 412 ff., 418 ff.,
formant, n o 437
frame, 143 head (of construction), 233, 344
free forms, 201 hearer, 275 ff., 307
free variation, 72 f., 114 f., 353 Hebrew, 17, 19, 22, 23
French, 17-18, 21, 40-1, 47-8, 56, 58, hierarchical classification, 164 ff., 456 f.
239 f., 241, 279 f., 286 f., 310, 354, High German, 28, 29
368, 379 high vowel, 128 f.
French Academy, 17-18 Hindi, 56, 302, 369, 395
frequency of occurrence, 83 ff. historical approach to language, 23
frequentative, 315 homeostatic systems, 90
fricative, 104 Homeric poems, 9
front vowel, 103, 128 ff. homography, 39-40, 405
function, 194, 232, 266, 377, 438 ff. homonymic conflict, 90 f.
functional load, 81-4, 95, 287 homonymy, 7, 39, 287, 450 ff.
fusional: cf. inflecting homophony, 39-40, 405
future, 403 ff., 310 f. honorific, 279 f.
Hopi, 311
Gaelic, 21, 395 humanism, 16
gender, 10, n , 12, 239 f., 270 ff., Hungarian, 22, 205
283 ff., 293, 355, 372 hyponymy, 452, 453 ff., 476
generalized transformation, 265
generative, 139 f., 146, 155 f., 209 ff., Icelandic, 17
234, 266, passim idea, 401 ff., 443, 474
genitive, 290 ff., 390 idealization, 50, 140
genus, 372 identification, 389
Georgian, 341 ideographic writing, 39
German, 21, 35, 116, 297 f., 358 f., 379 idioms: cf. ready-made utterance,
Germanic languages, 21, 24, 27-9, 124, schema
396 imitative sounds: cf. onomatopoeia
5H INDEX
immediate constituents, 209 ff. isolating, 187 ff., 206
imperative, 307 ff., 441 f. isomorphic systems, 55, 59, 60, 126,
imperfective: cf. perfective 429, 457
impersonal, 379 f. Italian, 21, 287 f.
implication, 403, 445 ff. iterative, 315
implicit assertion or denial, 445
inalienable: cf. alienable Japanese, 302
inanimate: cf. animate Junggrammatiker: cf. Neogrammarians
inceptive, 315
inchoative, 315 kernel sentence, 256
inclusion, 70, 234 kernel string, 256 ff., 376
inclusive first person, 279 f. Kikuyu, 22
incompatibility, 452, 458 ff., 476 kinship vocabulary, 55, 429, 468, 477
incomplete (or elliptical) sentence,
174 ^ labelled brackets, 217
indefinite: cf. definite labial: cf. bilabial
indeterminacy of grammar, 152 ff., labiodental, 105
477 langue (v. parole), 51-2, 139, 176
indeterminacy of semantics, 412, 426 larynx, 103
Indian grammar, 19-20, 196 Latin, 13-16, 17, 21, 26, 27-9, 30, 31-2,
indirect discourse, 174, 253 f. 58, 76, 174, 179, 189 ff., 205, 223 f.,
indirect object, 290, 295 ff., 302, 350 241, 249, 253, 278 f., 290 ff., 304 f.,
individual: cf. universal 312 f., 322, 392, 431
Indo-European, 21-2, 25, 27 laws of sound-change: cf. sound laws
Indo-Iranian, 21, 31, 124, 284, 291, layers of conditioning, 95 ff.
295 ff., 3O4, 307, 314, 322, 342, least effort, 90
354 ff., 395 length, syntagmatic, 80-1, 90, 94
infinitive, 174 level, 54, 184, 206
infix, 354 lexeme, 197 ff., 289, 403, 413
inflecting (v. inflexional), 187 ff., 191 lexical conditioning, 185 ff., 353
inflexion, 12, 133, 148, 190 f., 241, 289, lexical selection, 164
324 lexical substitution, 149, 161 ff., 243
inflexional, 191 lexical system, 429 ff.
informant, 138 lexicalization, 352, 354, 369, 438
information-content, 84, 454 lexicon, 149, 158, 161 ff., 318
information-retrieval, 149 library-classification, 159
information-theory, 84 ff. linear, 145, 209
inherently transformational, 261 f. liquid sound, 5
initial symbol, 216 literary language, 42
initiator of action: cf. actor loan-words: cf. borrowing
instrumental, 295 ff., 301, 343 local (v. grammatical), 295 ff., 298
intension (v. extension), 403, 454 locative, 291 f., 345 ff., 388 ff.
intention, 308 locative adjunct or complement, 345 ff.,
interchangeability, 451 ff. 389 f.
interior: cf. exterior locomotion (v. location), 300, 397 ff.
International Phonetic Alphabet, 104 ff. locus, 322, 346
interrogative, 307 ff. logarithm, 85 f.
intimacy, 280 logic, 11, 15, 71, 270 ff., 337 ff., 388 f.
intonation, 132, 177, 179 logical subject, 343 f.
intransitive, 12, 251 ff., 296, 320 ff., 332 long components, 105 ff.
introspection, 408 low vowel, 128 f.
intuition, 154, 478
IPA = International Phonetic Alphabet Macedonian, 204
Irish, 17, 21 machine-translation, 159
INDEX 5*5
macroscopic view of language-change, nasal, 104, 106, 121 ff.
49 f. nasalization, 104
major part of speech, 273 f., 436 natural gender, 283 ff., 357
Malay, 192 nature (v. convention), 4, 403, 406
Mandarin Chinese, 41, 393 Neogrammarians, 28 ff., 37
manner, adverb of, 326 f. nesting of constructions, 233
marked (v. unmarked), 79, 126, 306, neuter, 167 f., 283 ff., 293
307, 314, 323, 336, 356, 415, 45i, neutralization, 115 ff., 126, 253 ff.
460 new: cf. given
masculine, 11, 156 ff., 283 ff., 293 noise (in information-theory), 87 ff.,
mass noun, 282 96
mathematics and linguistics, 71 nominalism, 401, 404, 426
matrix string, 265 ff., 268 nominalization, 265, 347
matter (v. form), 198, 271 ff., 435 ff. nominative, 223, 290 ff., 301, 340 ff.
meaning: cf. semantics non-past tense, 305 f., 367
meaning, grammatical v. lexical, 435 ff. nonsense, 140, 422
meaningful, meaningless: cf. significant, normative: cf. prescriptive
having meaning Norwegian, 204
mechanism (v. mentalism), 408 notional grammar, 134 ff., 147, 155,
medieval grammar, 14-16 166 f., 270, 317 ff., 350 ff.
membership of grammatical classes, noun, 10, 11, 19, 147, 227, 279 ff,
147 ff, 318 f. 289 ff, 317, 320 ff., 337 ff., 357
Menomini, 311 noun-phrase, 213 f., 235
mentalism, 308, 474 f. nucleus, nuclear, 334 ff., 339 f-, 343>
mention: cf. use 35O ff.
metaphor, 5, 406 number, 12, 148, 239 ff., 243 ff., 263,
metaphysics, 15, 198 270 ff., 277, 288 ff.
microscopic: cf. macroscopic numerator, 229
middle voice, 372 ff.
mind, 401 ff., 408, 443 f., 474 object, 148, 290, 293 ff., 320, 340 ff.,
minimum free form, 201 f. 350 ff, 348 ff.
minor part of speech, 273, 436 object of result, 361, 439 f.
misunderstanding, 411 f. object-deletion, 360 ff., 303 f., 370
modal: cf. mood objective, 249 ff., 295 ff.
mode: cf. mood obligation, obligative, 208 ff.
modes of signifying, 14, 271 ff., 400, obligatory rules, 218 ff., 259, 268, 356,
404, 436 ff., 481 393, 397
modify, 178, 210 ff., 233, 296 f., 304, obscenity, 140, 423
325 f., 344, 422, 438, 475 f. Old Church Slavonic, 41
Modistae: cf. scholastic grammarians Old High German, 24
mood, 307 ff., 316 ff. Old Norse, 24
morph, 183 ff. one-place verbs, 350 ff.
morpheme, 170, 180 ff., 206 ff. onomatopoeia, 5, 119
morphographemic, 265 open set, 436, 438
morphology, 194 ff. open vowel, 103
morphophonemic, 116, 264 opposite in meaning: cf. antonymy
motion: cf. locomotion opposition: cf. contrast
multidimensional phonology, 131 optative, 308
multiple meaning, 405 ff. optional rules, 218 ff., 257, 259, 268,
392
names and naming, 5, 338, 403 ff. oral, 104
naming of grammatical classes, 147, ordering of rules, 220 f., 268
317 ff. organs of speech, 102 ff.
narrow transcription, 100 orientation, 275, 299, 303
516 INDEX
orthographic word, 199 Port Royal, 17, 273
ostensive definition, 409 f. positivism, 32-3, 408
overlapping distribution, 70 possession, 290, 295 ff., 300 f., 368,
391 ff.
paradigm, 6, 190, 195, 272, 289 f. possibility, 308
paradigmatic (v. syntagmatic) relations, postposition, 302 f., 394
70 ff., 117, 428 f., 460 postpositive article, 204
parole: cf. langue potential pause: cf. pause
parsing, 210 power of grammars, 261
participant, 276 Prague School, 116, 126 f.
participle, 12, 249 f. predicate, 11, 19, 148, 171, 210 ff., 320,
particle, 20 333 ff-, 337 ff-, 439 *"
particular: cf. universal predication, 270 ff., 337 ff.
parts of speech, 3, 11, 12, 13, 15, predictive, 155, 297, 316
19-20, 147 ff., 179, 270 f., 317 ff., prefix, 354
404 preposition, 12, 20, 290, 297 f., 301,
passive, 12, 257 ff., 261 f., 296 ff, 336, 302 ff., 319, 398
342 ff, 351, 371 ff., 375 ff., 468 prescriptive (or normative) grammar, 7,
past tense, 304 ff. 42-4
patient, 341 ff., 35<>ff- present tense, 304 ff.
pause, potential, 180, 199 f. presupposition, 403, 422, 440, 462
perception, 351 primitive languages, 44, 199
perfective, 313 ff., 395 ff, 398 primitive (undefined) terms, 137^
performance (v. competence), 51 172
permutation, 258, 268 probability of occurrence, 84 ff., 222,
Persian, 21 415
person, 270 ff., 276 ff. procedural linguistics, 157
personal pronouns, 275 ff. production (t>. recognition) grammar,
personification, 251, 277, 298 158, 230
phatic communion, 417 process-oriented, 366 f.
philology, 22 progressive aspect, 315 f., 351, 353
philosophy, 400 ff., 444 ff. progressive conditioning, 93
phonematic unit, 129 projection-rule, 439 f., 476, 480
phoneme, 100, 112 ff. projective, 155
phonetic spelling, 197 promissive, 310
phonetics, 22, 34, 99, 101 ff. pronoun, 12, 173, 275 ff., 283 ff.
phonic substance, 62 ff. proper (v. improper) inclusion, 234
phonological conditioning, 184 ff. proper noun, 337
phonology, 53-4, 99, 112 ff., 179 f., property, essential (defining), 401
184 ff., 204 f., 265, 268, 418 proportion: cf. analogy
phrase, 53, 148, 170 f., 206 ff., 214, 232, prosodic analysis, 127 ff., 185
266, 345, 412 prosody, 129
phrase-marker, 259, 266 Provencal, 17
phrase-structure, 215 ff., 255, 321 ff., pro-verb, 340, 440
328, 330 proverbs, 177
pitch, 103 proximate (v. remote), 276, 305 f.,
place: cf. local, locative 398
place (point) of articulation, 104 ff. pseudo-intransitive, 252, 363 ff., 370
plane, 54 psycholinguistics, 160, 400 ff., 420
plereme, 470 psychological subject, 314
plosive: cf. stop punctual, 315, 348, 351
plural, 148, 281 ff. punctuation, 40
Polish, 21 purity* of language, 9, 42-4
polysemy, 496 putative, 310
INDEX 517
quality, 324 f. Sanskrit, 19-20, 21, 24-5, 27-9, 31, 41,
question, 307 ff., 435, 441 188, 273, 297 ff-
schema, 178
race and language, 24 scholastic grammarians, 14-16, 198,
rank, 171, 178, 181, 205 ff., 345 271, 436
rationalism, 17, 24 scientific study of language, 1, 22
ready-made utterances, 177, 416, 420 Scottish Gaelic, 21
realism (v. formalism), 51 second-order nominals, 347 ff.
realism (v. nominalism), 401, 404, 408, segmentation, 181 ff.
426 semantic category, 471
realization, 60, 184, 196, 280 semantic change, 401, 407
recategorization, 277, 282, 299, 316, 348 semantic component: cf. component
Received Pronunciation, 115 semantic marker: cf. component
recognition (v. production) grammar, semantic structure, 55, 429
158, 230 semantics, 53-4, 134 ff., 140 f-, 143,
reconstruction, 31, 33 154 f., 166 ff., 200 f., 268 f., 400 ff.,
rection, 241 passim
recursive, 221 ff., 262, 268, 326, 381, sememe, 470
386 semiotic, 405
recycling, 225 Semitic, 22
redundancy, 85 ff. sense, 403, 411, 427 ff., 443 ff.
reference (anaphoric or pronominal), sense-relations, 427 ff., 443 ff.
233, 275 ff., 283 ff. sentence, 52, 53, 170ff.,206, 227, 334 ff.,
reference (v. sense), 404 ff., 410, 419, 401, 412 f., 419 f., passim
424 ff., 434, 454 sentence-type, 145, 178 f., 218, 441 f.
reflexive, 361 ff., 373 ff. sequence, 76 ff., 209 f.
regressive conditioning, 92 set-theory, 71
regular, 6, 119 f. sex and gender, 283 ff., 293
related languages, 21-2 short components, 105 ff.
relations, structural, 50, 59 sign, linguistic, 403
relative (v. absolute) point of reference, significant (meaningful), 137, 140 f.,
299 154, 172, 402 f., 412, 422, 423 ,
relativity: cf. determinism 475 ff.
remote: cf. proximate signification, 15, 271 ff., 400 ff., 410
Renaissance, 16-17 simple sentence, 178, 225
represent, 184 single-base: cf. singulary transformation
resemblance between languages, 26 singular, 148, 281 ff.
restricted context, 419 singulary transformation, 265, 268
rewriting rules, 161, 210, 216 ff., 235 ff. Sino-Tibetan, 22
ritual utterance, 139, 416 f. Siouan languages, 301
Roget's Thesaurus, 446 f., 456, 472 situation of utterance, 275 ff., 300, 413,
role, 276, 280 434
Roman grammarians, 13 slang, 90 f.
Romance languages, 21, 279, 396 Slavonic languages, 21, 283, 294, 313
romanticism, 23 Sophists, 10
root, 20, 195 f. sortal universals, 337 ff.
rounded vowel, 103, 128 ff. sound, 53
Rumanian, 204 sound-change, 27 ff.
Russian, 21, 55, 56-7, 5$, 204, 279 f., sound-law, 28-33
294, 297 ff., 3Oi, 3i3ff- 322, 375, sound-symbolism, 5, 419
394, 43O Spanish, 21, 287, 375, 379
spatiotemporal, 275, 299, 413
safety-margin (between ranges of real- speaker, 275 ff., 307, 398
ization), 68, 90 species, 372
518 INDEX
specific (v. non-specific), 378 f. syncretism, 298

spectrograph, n o synonymy, 7, 405 ff., 410, 418 f., 427 f.,
speculative grammar, 14-16, 179, 271 441, 446 ff., 455, 576
speech, 38, 62 ff., 197 syntagm, 210
speech-sound, 99 f., 112 syntagmatic relations, 70 ff., 117 f., 452,
speech-synthesis, n o 460
spelling, 197 syntax, 133, 194, passim .,
spirant: cf. fricative synthesis, 158 ff., 217, 234
spoken language: cf. speech synthetic language-type, 188 ff.
state (t;. action), 300, 314 ff., 324 ff., 372 Syrian, 19
statement, 435, 441 system, 50, 52, 59
statistical methods, 81 ff. systematic correspondence: cf. corre-
stative, 315 f., 324 ff., 341, 348, 351 spondence between languages
status, 276, 280
stem, 196 taboo, 423 f.
stimulus and response, 416 f. Tagalog, 192
Stoics, 5, 8, n , 12 taxonomy, 456 f.
stop, 104 temporal adjunct, complement, 345 ff.,
stratum, 206 f. 389 f.
stress, 132, 336, 418 tense (as a grammatical category), 11,12,
string, 209 148, 270ff.,289, 304ff.,316ff.,349 ff.
structural analysability, 257 ff., 262 f. terminal string, 216, 223
structural description, 146, 149 terminal symbol, 216, 421
structural meaning, 435 terminology of linguistics, 1
structural position, 82 thing signified, 403 ff.
structural pressure, 37 thought and language: cf. concept, idea,
structuralist, 157 mind
structure, 36-8, 501, 53 three-placeverb,35o,368ff.,386f.,397ff.
subcategorization: cf. subclassification Tibetan, 22
subclassification, 150 ff., 164 ff. time: cf. tense, temporal
subject, n , 19, 148, 171, 210 ff., 290, tone, 76, 132
293 ff., 320, 334 ff., 337 ff-, 35O ff., topic (v. comment), 334 ff., 392
438 f. transformation, 155, 224, 247 ff., 260 ff.,
subject of discourse, 275 ff. 296 f., 327, 347, 376, 380
subjective, 249 ff., 295 ff., 303 transitional probabilities, 93
subjectivism, 408 transitive, 12, 251 ff., 296, 320 ff., 332,
subjunctive, 310, 312 f. 340 ff., 350 ff., 439
subordination, 178, 233, 311 ff. translation, 409, 431, 434, 437, 457 f.
substance (v. accidents), 16, 198, 271 ff., trans instantiation, 60
333 tree-diagram, 161, 211
substance (v. form), 54 ff., 99, 184, 198, trial number, 283
43O f- Turkish, 22, 128 ff., 132, 188 f., 205,
substantive noun, 347 276, 278 f., 290 ff., 299, 302, 353, 361,
substantive universal: cf. formal 369, 37i
substring, 259 two-place verbs, 350 ff.
suffix, 196, 264, 354 types of languages, 187 ff., 436
supplementary rule, 223 f. typology, 191
surface structure: cf. deep structure
Swahili, 22, 284 ff., 394 underlying structure, 250
Swedish, 21, 204 understanding, 411 f., 419, 426, 446
syllabic writing, 39 unidimensional phonology, 131
syllable, 132 universal (v. particular, individual)
symbol, 403 terms, 337 ff-
synchronic, 45 ff., 407 universal grammar, 14-16, 270
INDEX
universal semantics, 472 ff. voiced (v. voiceless), 104, 121 ff.
universe of discourse, 419, 426 vowel, 102, 103 f.
unrounded vowel, 103, 128 ff. vowel harmony, 128 ff., 205, 353
usage, 7, 8, 272 vox, 272, 372
use and meaning, 410 f., 414
use and mention, 201 Welsh, 21
utterance, 52, 171 f., 176, 336, 413, 419, wish, 308
423, passim word, 53, 68-70, 142, 170 f., 180, 194 ff.,
206, 272, 402, 412, passim
variable (v. constant), 161, 237 ff., 261 f. word-boundary, 264
Vedic hymns, 1920 word-formation: cf. derivation
velar, 105, 121 ff. word-order, 76, 203 f., 223, 297, 302 ff.
verb, 10, 11, 19, 147, 280, 319, 323 ff., writing, 38-9, 62 ff., 196 f., 199
436 f. writing systems, kinds of, 39
Verner's law, 29-30, 125 f.
Vietnamese, 188, 283 zero, 193, 237
vocabulary, 54 ff., 427 ff., 433 zero-modification, 363
vocative, 290 Zipf's law, 89 f.
voice (as a phonetic term), iO2ff., 106,372 Zulu, 22

Introduction To Theoretical Linguistic Lyons PDF

Uploaded by

Copyright:

Available Formats

Introduction To Theoretical Linguistic Lyons PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Introduction To Theoretical Linguistic Lyons PDF

Uploaded by

Copyright:

Available Formats

What is the overall structure and content of the book?

What is the overall structure and content of the book?

What are some of the main topics covered in linguistics?

What are some of the main topics covered in linguistics?

INTRODUCTION TO

Cambridge University Press 1968

First published 1968

ISBN o 521 09510 7

Transferred to digital printing 2001

i Linguistics: The Scientific Study of Language

2 The Structure of Language

3 The Sounds of Language

4 Grammar: General Principles

9 Semantics: General Principles

Notes and references 482

Table of symbols and notational conventions 506

Index of proper names 507

Index of subjects 510

Explanatory notes, bibliographical

Introduction to Theoretical Linguistics

Book DOI: http://dx.doi.org/10.1017/CBO9781139165570

Online ISBN: 9781139165570

Hardback ISBN: 9780521056175

Paperback ISBN: 9780521095105

PREFACE TO THE 1995 EDITION pp. ix-xiv

Chapter DOI: http://dx.doi.org/10.1017/CBO9781139165570.001

Cambridge University Press

Cambridge, February 1995 J.L.

Introduction to Theoretical Linguistics

Book DOI: http://dx.doi.org/10.1017/CBO9781139165570

Online ISBN: 9781139165570

Hardback ISBN: 9780521056175

Paperback ISBN: 9780521095105

1 - Linguistics: The Scientific Study of Language pp. 1-52

Chapter DOI: http://dx.doi.org/10.1017/CBO9781139165570.002

Cambridge University Press

1.1.3 Objective approach to language

1.2 Traditional grammar

1.2.2 ' Nature' and ' convention'

1.2.3 Analogists and anomalists

i .2.4 Alexandrian period

The assertion that the language of Plato is a 'purer' form of Greek

1.2.5 Greek grammar

1.2.6 The Roman period

1.2.7. Medieval period

1.2.8 The Renaissance and after

1.2.9 Wider influence of the Greco-Roman tradition

1.2.1 o The Indian tradition

1.3 Comparative philology

1.3.2 'Scientificy linguistics

1.3.3 Evolutionary point of view

1.3.4 A wider range of languages

1.3.6 Discovery of Sanskrit

1.3.7 Importance of Indo-European languages

1.3.9 * Grimm's law '

(Notice that 8 in the transcription of Gothic denotes the voiceless

i.3.10 The 'Junggrammatiker'

1.3.11 ' Verner }s law' and other ' sound-laws '

Sanskrit bhratar- pitdr-