Aproaches or Gramamticalization PDF

APPROACHES TO GRAMMATICALIZATION I
TYPOLOGICAL STUDIES IN L A N G U A G E (TSL)

A companion series to the journal "STUDIES IN L A N G U A G E "
Honorary Editor: Joseph H. Greenberg
General Editor: T. Givn
Editorial Board:
Wallace Chafe (Santa Barbara)
Bernard Comrie (Los Angeles)
R.M.W. Dixon (Canberra)
John Haiman (St Paul)
Kenneth Hale (Cambridge, Mass.)
Bernd Heine (Kln)
Paul Hopper (Pittsburgh)
Margaret Langdon (San Diego)
Sandra Thompson
Charles Li (Santa Barbara)

Johanna Nichols (Berkeley)
Andrew Pawley (Canberra)
Doris Payne (Oregon)
Frans Plank (Konstanz)
Jerrold Sadock (Chicago)
Gillian Sankoff (Philadelphia)
Dan Slobin (Berkeley)
(Santa Barbara)
Volumes in this series will be functionally and typologically oriented, cove

ring specific topics in language by collecting together data from a wide vari
ety of languages and language typologies. The orientation of the volumes
will be substantive rather than formal, with the aim of investigating univer
sais of human language via as broadly defined a data base as possible, lean
ing toward cross-linguistic, diachronic, developmental and live-discourse
data. The series is, in spirit as well as in fact, a continuation of the tradition
initiated by C. Li (Word Order and Word Order Change, Subject and Topic,
Mechanisms for Syntactic Change) and continued by T. Givn (Discourse
and Syntax) and P. Hopper (Tense-Aspect: Between Semantics and Pragmatics).
Volume 19:1
Elizabeth Closs Traugott and Bernd Heine (eds)

APPROACHES
TO
GRAMMATICALIZATION
APPROACHES TO
GRAMMATICALIZATION
VOLUME I
FOCUS ON THEORETICAL
AND METHODOLOGICAL ISSUES
edited by
ELIZABETH CLOSS TRAUGOTT

and BERND HEINE
JOHN BENJAMINS PUBLISHING COMPANY

AMSTERDAM/PHILADELPHIA
1991
Library of Congress Cataloging-in-Publication Data

Approaches to grammaticalization / edited by Elizabeth Closs Traugott and Bernd Heine.
p.
cm. -- (Typological studies in language, ISSN 0167-7373; v. 19)
Contents: v. 1. Focus on theoretical and methodological issues - v. 2. Focus on types of
grammatical markers.
Includes bibliographical references and index.
1. Grammar, Comparative and general -- Grammaticalization. I. Traugott, Elizabeth Closs.
II. Heine, Bernd, 1939. III. Series.
P299.G73A6 1991
415~dc20
91-3650
ISBN 90 272 2901 5 (set)/90 272 2895 7 (Vol. I)/90 272 2899 X (Vol. II) (Eur. ; hb; alk. paper)
ISBN 90 272 2902 3 (set)/90 272 2896 5 (Vol. I)/90 272 2900 7 (Vol. II) (Eur. ; pb; alk. paper)
ISBN 1-55619-404-8 (set)/l-55619-400-5 (Vol. I)/1-55619-402-1 (Vol. II) (US; hb; alk. paper)
ISBN 1-55619-405-6 (set)/l-55619-401-3 (Vol. I)/l-55619-403-X (Vol. II) (US; pb; alk. paper)
Copyright 1991 - John Benjamins B.V.
No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any
other means, without written permission from the publisher.
Table of contents
VOLUME I
Foreword
xi
Introduction
Elizabeth C. Traugott and Bernd Heine
Section A: General Method
15
On some principles of grammaticization

Paul J. Hopper
17
On the gradualness of grammaticalization

Frantisek Lichtenberk
37
Serial verbs and the mental reality of 'event': Grammatical vs.

cognitive packaging
T. Givn
81
Grammaticalization as retextualization
Johanna Nichols and Alan Timberlake
129
Section B: Directionality
147
From cognition to grammar: Evidence from African languages

Bernd Heine, Ulrike Claudi and Friederike Hnnemeyer
149
The semantics-pragmatics of grammaticalization revisited

Elizabeth C. Traugott and Ekkehard Knig
189
The de dicto domain in language

Zygmunt Frajzyngier
219
The grammaticalization of rhetorical questions in Tamil

Susan C. Herring
253
VI
TABLE OF CONTENTS
Some grammaticalization changes in Estonian and their implications

Lyle Campbell
The last stages of grammatical elements; contractive and expansive
desemanticization
Joseph H. Greenberg
Substrates, calquing and grammaticalization in Melanesian pidgin
Roger M. Keesing
285
301
315
VOLUME II
Foreword
xi
Introduction
Elizabeth . Traugott and Bernd Heine
Section A: Verbal structure
15
Back to the future

Joan L. Bybee, William Pagliuca, Revere D. Perkins
17
The emergence of perfective aspect in Indo-Aryan languages

Peter Edwin Hook
59
Section B: Argument structure
91
Grammaticization of topic into subject

Masayoshi Shibatani
93
From V/2 to subject clitics: evidence from Northern Italian

John Haiman
The role of motivation in the emergence of grammatical categories:
the grammaticization of subjects
Marianne Mithun
Verbs to promotional suffixes in Sahaptian and Klamath
Noel Rude
Grammaticalization of postpositions and word order in Senufo
languages
Robert
135
159
185
201
TABLE OF CONTENTS
VII
Section : Subordination
225
From postposition to subordinator in Newari

Carol Genetti
227
The evolution of dependent clause morpho-syntax in Biblical

Hebrew
T. Givon
257
Section D: Modality
311
A quantitative perspective on the gramaticization of epistemic

parentheticals in English
Sandra A. Thompson and Anthony Mulac
313
The grammaticization of the German modal particles

Werner Abraham
331
Section E: Spectra of grammaticalization patterns
381
Areal and universal dimensions of grammatization in Lahu

James A. Matisoff
383
Ways to go in Rama: a case study in polygrammaticahzation

Colette Craig
455
Grammaticalization and related changes in contemporary German

Christian Lehmann
493
ABBREVIATIONS
ABL
ABS
ACC
ACT
ADV
AF
AGR
AOR
ART
ASSOC
ASP
AUX
ablative
absolutive
accusative
active, actor
adverb
actor focus
agreement
aorist
article
associative
aspect
auxiliary
BEN
benefactive
CAUS
COMP
COMPL
COMPAR
COND
CONJ
COP
causative
complementizer
complement
comparative
conditional
conjunction
copula
DAT
DEC
DEF
DEM
DET
DIR
DO
DS
dative
declarative
definite
demonstrative
determinator
directional
direct object
directional
subject
dualis
dubitative
durative
DU
DUB
DUR
EMPH
ERG
EXCL
EXCLAM
emphatic
ergative
exclusive
excalamative
FEM
FIN
FOC
FUT
feminine
finite
focus
future
GEN
GEN.OBL
GER
genitive
general oblique
preposition
gerund
IMP
IMPERF
IMPFV
INCL
INCONS
INDEF
INF
INFL
INT
INTR
IO
IRR
imperative
imperfect
imperfective
inclusive
inconsequential
indefinite
infinitive
verbal inflection
intensifier
intransitive
indirect object
irrealis
LOC
locative
MASC
MED
MID.VCE
masculine
medial
middle voice
ABBREVIATIONS
NEG
NOM
NOMI
NONFUT
NONPAST
NP
negative
nominative
nominalizer
nonfuture
nonpast
noun phrase
OBJ
OBL
object
oblique
PART
PARTCP
PARTV
PASS
PAST
PERF
PFV
PL
POS.PURP
POSS
PRE
PREC
particle
participle
partitive
passive
past
perfect
perfective
plural
positive purpose
possessive
verbal prefix
precaution/
PRED
PREP
PRES
PROG
PROH
PURP
negative purpose
predicate
preposition
present
progressive
prohibitive
purposive
Q
QUOT
interrogative
quotative
REAS
RED
REFL
REL
REP
reason
reduplicative
reflexive
relative clause
marker
repetitive
SEQ
SG
SS
STAT
SUB
SUBJ
SUBJUNCT
sequential
singular
same subject
stative
subordinate
subject
subjunctive
TNS
TOP
TRANS
tense
topic
transitive
VB
VOC
verb
vocative
1SG
first person
singular
first person plural
third person
singular
1PL
2SG
Foreword
The papers in these volumes grow out of a symposium on grammaticalization organized by Talmy Givn, and held at The University of Oregon at
Eugene in May 1988. The purpose of this symposium was to "present and
discuss ... a broad range of empirical and theoretical issues concerning the
genesis of morphosyntactic structure". We wish to acknowledge the inspira
tion provided by Talmy Givn's leadership of the symposium.
Many people helped bring these volumes into final shape. Especial thanks
are due to John McWhorter and Gillian Ramchand for their assistance in
commenting on and copy-editing papers, and to Ulrike Claudi and Friederike
Hnnemeyer for preparation of the final materials.
Introduction
Elizabeth Closs Traugott and Bernd Heine
Stanford University and University of Cologne
1.
INTRODUCTION
Although not everyone who uses the term would agree, "grammaticaliza
tion" as we define it refers to that part of the theory of language that focuses
on the interdependence of langue and parole, of the categorial and less
categorial, of the fixed and the less fixed in language. Grammaticalization is
the linguistic process, both through time and synchronically, of organization
of categories and of coding. The study of grammaticalization therefore high
lights the tension between relatively unconstrained lexical expression and
more constrained morphosyntactic coding, and points to relative indetermi
nacy in language and to the basic non-discreteness of categories.
These volumes present a number of diverse theoretical viewpoints on
grammaticalization and give some detailed insights into specific grammaticali
zation phenomena in a number of languages. However, no attempt is made
at completeness; for more comprehensive studies of the field of grammaticali
zation as a whole, see Lehmann (1985), Heine and Reh (1984), Hopper and
Traugott (In progress), Heine et al. (1991). The diversity of perspectives is
manifest in a number of ways. For example, disagreement about how to
approach the subject starts with disagreements about what to call it. Some
authors prefer "grammaticization" or even "grammatization" to "grammati
calization". There is more significant disagreement about whether grammati
calization is primarily a diachronic phenomenon to be studied from a "source
and pathway" perspective, or primarily a syntactic, discourse-pragmatic phe
nomenon, to be studied from the point of view of fluid patterns of language
use across time or at a synchronically segmented moment in time. In so far
as there is any correlation between perspective and terminology, those who
espouse the latter view of the subject tend to call it grammaticization. How
ever, since the majority of participants used "grammaticalization", whatever
ELIZABETH CLOSS TRAUGOTT AND BERND HEINE
their perspective, we have chosen to use this term in the title and in this
introduction. We have not imposed uniform terminology on the contributors,
and both terms will be found in the volumes.
Together the papers suggest answers to a number of questions, some of
which focus on testing the boundaries of grammaticalization, others of which
test the assumptions of linguistic theory in general. We outline here some of
the models of grammaticalization that the authors use, challenge, or develop,
and some of the main issues that the papers highlight.
2.
THE LEXICAL ITEM> MORPHEME MODEL
Early work on grammaticalization viewed it as a diachronic phenomenon.

Although many examples were discussed by nineteenth century scholars (e.g.
Bopp, 1816; Humboldt, 1825; Gabelentz, 1891), the term was apparently first
used by Meillet (1948 [1912]). He defined it as the evolution of grammatical
forms (function words, affixes, etc.) out of earlier lexical forms, and much
subsequent work has been conducted within this framework. Later researchers
have emphasized that morphemes may also arise out of other morphemes.
For example, Lehmann, acknowledging Kuryowicz (1965), says: "Under the
diachronic aspect, grammaticalization is a process which turns lexemes into
grammatical formatives and makes grammatical formatives still more gram
matical..." (1985:303), and Heine and Reh define grammaticalization as:
"...an evolution whereby linguistic units lose in semantic complexity, prag
matic significance, syntactic freedom, and phonetic substance..." (1984:15).
In the present volumes papers by Abraham, Bybee et al., Campbell,
Carlson, Craig, Greenberg, Heine et al., Keesing, Lehmann, Matisoff, Rude,
Traugott and Knig owe much to this perspective, although all have also
been influenced to greater or lesser degree by the discourse > morphosyntax
perspective outlined in 3.
3.
THE DISCOURSE > MORPHOSYNTAX PERSPECTIVE
Alongside the lexical item > morpheme tradition which derives from
Meillet, there has been a more recent tradition associated with Talmy Givn,
Charles Li, Sandra Thompson and others that focuses on the "packaging"
of discourse and evolution of syntactic and morphological structure through
INTRODUCTION
fixing of discourse strategies. For example, Givn (1979:209) characterizes

the process as one of cyclic waves involving
DiscoarseSyntaxMorphologyMorphophonemicsZero
cf. also Bybee (1985), Hopper (1988), Li and Thompson (1974), among others.
Papers in these volumes that develop this morphosyntactic perspective
particularly clearly are those by Genetti, Givn on the development of depen
dent clauses in Hebrew, Haiman, Herring, Hook, Hopper, Lichtenberk, and
Shibatani. It should be noted that in many cases the source item or items
discussed by these authors is not lexical but morphological (a possibility
clearly permitted by the definitions of grammaticalization cited in 2.).
4.
THE GRAMMATICAL CODING PERSPECTIVE ON

GRAMMATICALIZATION
So far we have reviewed issues of grammaticalization that are one way

or another issues related to language change. Yet another perspective on
grammaticalization asks not so much how grammatical coding comes into
being, but what grammatical coding is possible typologically in a language
or in languages, and how it is organized either at one period in time (Givn
on serial verbs, Mithun, Thompson and Mulac) or panchronically (Frajzyngier, Nichols and Timberlake).
The answers to the question whether certain categories must be obligato
rily grammaticalized challenge some of the fundamental assumptions some
linguists bring to linguistic theory, for example, that there are universal
"primitives" such as Subject, Object, Tense (cf. also DuBois, 1985).
5.
THE RELATION OF DIACHRONIC GRAMMATICALIZATION

TO LANGUAGE CHANGE
For those who see grammaticalization as a diachronic process, the ques

tion naturally arises: How can we distinguish grammaticalization from lan
guage change? The answer is that grammaticalization is a kind of language
change, subject to certain general processes and mechanisms of change (see
section 8.), and characterized by certain consequences such as changes in
grammar. These are increased syntacticization in its early stages, and increased
loss of morphosyntactic independence in its later stages, ultimately leading
to zero, i.e. increased morphologization, and phonologization. Like other

changes, grammaticalization spreads gradually across linguistic contexts on
the one hand (Lichtenberk, Nichols and Timberlake), and across social
contexts on the other (cf. Labov, 1972, and much of the sociolingustic
literature on variation and language change in progress). But that does not
mean that all change is grammaticalization. To take an obvious example, the
semantic change from Middle English bede 'prayer' to its modern meaning
of a small spherical object often used in necklaces ('bead'), is an example of
change by metonymy, but not of grammaticalization. Papers in this volume
that particularly address this issue are those by Hopper and Lehmann.
6.
THE UNIDIRECTIONALITY OF GRAMMATICALIZATION
Firmly entrenched in diachronic perspectives on grammaticalization is

the assumption of a cline of unidirectionality; or more specifically of a
pathway that channels change through a limited number of structures that
are minimally different from one another. This does not mean that all unidirec
tionality involves grammaticalization; for example, the well-known tendency
for [sy] to become [] but not vice versa is in itself not an instance of
grammaticalization. However, if it results from morpheme boundary loss,
then it is an instance of a stage in grammaticalization. The question is precisely
what kinds of unidirectionality are necessary or at least typical of grammati
calization, and whether any kinds of reversals of grammaticalization are
usefully considered as examples, albeit rare, of grammaticalization (see 7.
below).
One kind of unidirectionality often associated with grammaticalization
is increase in abstractness. This notion is appealed to in discussions of
metaphorical change (Heine et al. on PERSON > OBJECT > SPACE >
TIME>PROCESS>QUALITY, Traugott and Knig on temporal>
causal>concessive). However intuitively attractive and widely agreed upon
such a cline of abstractness may be, it is essentially a pre-theoretical notion.
Certainly, increased abstractness does not in itself require grammaticalization.
For example, the shift in meaning from grasp in the sense 'seize' to 'under
stand' is not a case of grammaticalization, although it is an example of what
many linguists would consider increased abstractness in meaning.
Much of the literature on unidirectionality characterizes the development
of grammatical from lexical meaning in terms of desemanticization, bleaching,
and emptying or loss of semantic or pragmatic meaning (see the definitions
INTRODUCTION
of grammaticahzation in 2. above). Traugott and Knig challenge this concept

and argue that in its early stages grammaticahzation actually often involves
an increase in pragmatic meaning (though semantic content, strictly speaking,
may be reduced as pragmatic meaning increases). For example, a shift from
temporal to concessive while does not involve primarily bleaching of semantic
meaning, but rather increase of pragmatic significance pertaining to the
speaker's attitude to what is being said. If this is correct, then we need to
modify the standard view of semantic unidirectionality in the process of
grammaticahzation. It is not always a simple case of more > less semantic,
but more often of an implicational hierarchy of the type semantic > pragmatic>less semantic-pragmatic. Note that although this is not a characterization
of bleaching it is nevertheless unidirectional, since the claim is that one would
not expect to find pragmatio>semantio>less semantic-pragmatic, e.g. conces
sive while > temporal. As Traugott and Knig indicate, the meaning changes
outlined here for.grammaticahzation are characteristic of meaning change in
general, and are therefore not diagnostic for grammaticahzation (any more
than bleaching is, for that matter).
Sometimes it has been proposed that the lexical item >morpheme and
the discourse>morphosyntax perspectives come into conflict with respect to
unidirectionality in the earliest stages of grammaticahzation (Herring). How
ever, there is actually no inconsistency when we consider that discourse
presupposes lexicon; discourse uses lexical items in ways that endow them
with pragmatic meaning, and if they have the properties salient to grammati
cahzation outlined above, and are used more frequently, they may well come
to be syntacticized in the kinds of way Givn illustrates. The two formulae
"lexical item>morpheme", "discoursomorphosyntax " can therefore be
combined as: lexical item used in discoursomorphosyntax.
Another kind of unidirectionality often considered characteristic of gram
maticahzation is increase in bondedness. Typically, at the clause level indepen
dent clauses are combined, and a cline may develop from independent clauses
through some kind of loose juxtaposition and coordination to subordination
(cf. Wiegand, 1987). At the phrasal level, forms may become less free and
more bound via grammaticahzation; for example, postpositions become
affixes (Craig, Greenberg). Although not confined to grammaticahzation, one
of its effects is that morphemes undergoing this process move away from
cardinal categoriality, and in their later stages lose the ability to refer and to
associate with the inflectional and derivational trappings of their morphosyntactic category (cf. Hopper and Thompson, 1984). This leads to the emergence
of "linguistic hybrids" which show the properties of several morpheme classes.
Nouns which develop into adpositions, for example, lose nominal characteris
tics such as the ability to be marked for definiteness or number (Genetti), or
verbs which are grammaticalized to case markers tend to lose the ability to
be inflected for person, tense, aspect, and mood. Loss of categoriality may
also lead to the kind of development of multiple grammaticalization chains
in different functional domains that Craig discusses in connection with Rama.
Unidirectionality may suggest a single path of evolution. However,
approaches referring to multiple functional domains (cf. Kemmer 1988) or
to correlated processes seem closer to reality (cf. Heine and Reh, 1984;
Lehmann, 1985; Croft, Forthcoming: Ch. 8). Heine and Reh suggest a tripar
tite classification of correlations of change: Semantic-pragmatic status, gram
matical behavior, and phonological substance. Lehmann uses a bipartite
classification according to paradigmatic and syntagmatic processes, i.e.
according to the alternatives available on the one hand and the effect of
linguistic context on the other. Lehmann's characterization is somewhat
theory-bound and leads to the incorrect claim that grammaticalization
involves reduction in scope (1985:309), despite the fact that the, development
of subordinators from pronouns or verbs of saying, for example, clearly
involves expansion of the scope of these forms. However, its focus on correla
tions is of crucial importance. In these volumes, most of the papers alert us
to correlations.
It is the phenomenon of correlatable parameters that allows us to project
diachronic grammaticalization onto synchronic data. Consider for example,
the grammaticalization of body part nouns. If arrayed simply as a case of
OBJECT>SPACE>TIME>QUALITY the examples of megb in Heine et
al. would not qualify as a case of grammaticalization. They do so precisely
because the semantic changes are correlated with a shift from nominal to
prepositional structure, and with different syntactic privileges of occurrence.
7.
COUNTEREXAMPLES TO UNIDIRECTIONALITY
The discussion so far has focused on unidirectionality, and what kinds

of unidirectionality are characteristic of grammaticalization. Virtually nothing
is exceptionless, and there are of course instances of change in languages that
are counterexamples of tendencies that can be characterized as "less>more
grammatical", "main clause > subordinate clause", etc. In these volumes the
papers by Campbell and Greenberg explicitly raise counterexamples to unidi
rectionality (cf. also Matsumoto, 1988). The well-known phenomenon of
INTRODUCTION
"lexicalization" is also a counterexample, for example when phonological

changes result in morphological loss and the development of idiosyncratic
lexical items, such as the English pairs lie-lay, sit-set, stink-stench, all of
which have their origins in i-umlaut.
It is likely that all these examples are strictly speaking actually not cases
of grammaticalization (although once they have occurred they may be subject
to the generalization, reduction, loss, and other changes typical of grammati
calization). Rather, the examples Campbell and Greenberg cite can be
regarded as instances of reanalysis (Greenberg actually refers to "reinterpretation which.. .goes beyond what is usually called reanalysis in discussions of
grammaticalization"). Nevertheless, since they test the boundaries of what
may be considered bona fide cases of grammaticalization, they provide an
important perspective on the other papers in these volumes.
8.
MECHANISMS OF LANGUAGE CHANGE RELEVANT TO

GRAMMATICALIZATION
A number of mechanisms of language change have already been alluded

to as being relevant to grammaticalization. This is hardly surprising if indeed
grammaticalization is a subset of phenomena occurring in change. Of particu
lar importance for these volumes are: Metaphorical transfer (Heine et al.),
metonymic transfer (Traugott and Knig), reanalysis (Heine et al., Lehmann,
cf. also Langacker, 1977), and analogy (Greenberg, Lehmann, Matisoff).
Another mechanism is the incorporation of grammatical material from the
substrate languages, as illustrated by evidence from Melanesian Pidgin (Keesing). All these are mechanisms that make change possible, but none are
restricted to grammaticalization and all are independent of the unidirectionality associated with grammaticalization.
9.
CONDITIONS LICENSING GRAMMATICALIZATION
A further set of questions for the diachronic perspective on grammaticali

zation pertains to what conditions have to apply for grammaticalization to
take place, in other words, given that a form A exists, what is its potential
for becoming grammaticalized, and how do we know when this is happening?
Experience with cross-linguistic studies of typology and of change have
revealed extensive and robust evidence that only certain lexical classes are
ELIZABETH CLOSS T R A U G O T T A N D BERND H E I N E
likely to become grammaticaUzed. For example, it seems unlikely that wall

paper would become grammaticaUzed. This is not to say that such a change
is impossible, but rather that it is extremely unlikely. What we find in language
after language is that for any given grammatical domain, there is only a
restrictive set of lexical fields, and within them only a restricted set of lexical
items, that are likely to be sources. For example, case markers, including
prepositions and postpositions, typically derive from terms for body parts or
verbs of motion; tense and aspect markers typically derive from specific spatial
configurations; modals from terms for possession, or desire; middles from
reflexives, etc. Furthermore, the paths of change are themselves highly
restricted, and suggest step-by-step developments, not discrete leaps across
semantic or pragmatic domains. Accounts of why these restrictions apply
appeal to cognitive constraints. Whereas some have regarded these cognitive
constraints as hard-wired (i.e. essentially arbitrary, cf. Bickerton, 1981),
authors represented in these volumes regard them as functionally constrained
by salience, communicative strategies, etc. For example, if one is attempting
to express something abstract using the resources already available, or
attempting to express communicative intent, also using resources already
available, then the purposes to which one puts language in part determine
the choices one makes. Therefore, one answer to the question concerning
what potential a form A has for becoming grammaticaUzed is: This depends
on whether its semantic content or the inferences one can draw from it serve
the purposes of creating text, that is, of producing discourse. Ultimately, the
question is, can the form be used to serve the metalinguistic purposes of textconstruction in a natural and therefore easily understood way? It is hard to
imagine a context in which wall paper could be used to serve such a purpose.
The fact that a lexical item A has potential for solving the problem of
expression does not in itself predict that it is a candidate for grammaticalization. We need to look for additional factors. An example is corner, a term
which is used metaphorically in temporal expressions (cf. go round a corner
in time). However, corner does not appear to be grammaticaUzed into tense
or aspect. Why might this be? It has been suggested that the operative
constraint is perceptual salience: Topological spaces only are grammaticaUzed
(Talmy, 1983; Sweetser, 1988). Although the expected end of a process of
grammaticalization is usually zero, in fact there are "graveyards", such as beclefts (Givn) and frozen idioms or lexicalizations (Hopper) that appear to
block or at least delay the process of total loss (see especially Greenberg).
It would seem that the point at which grammaticalization stops may be
determined by the typological characteristics of the language concerned. There
INTRODUCTION
is, for example, a well established channel of grammaticalization leading from

postpositions to nominal case inflections. To our knowledge, however, such
a development has not been documented for languages of the analyticisolating type, where grammaticalization is unlikely to lead to the development
of inflectional morphology. Observations like these suggest that there may be
typological language-internal constraints that block grammaticalization from
proceeding beyond a certain point. Mithun suggests that some very specific
factors such as the internal shape of the grammar of a particular language
may make certain grammatical markers such as subject redundant and may
therefore serve to block their development.
Given that a form A is a candidate for grammaticalization both because
of its semantic context and its salience, a further condition has to apply for
grammaticalization to take place: The form has to be used frequently. The
more grammaticalized a form, the more frequent it is (Bybee et al., Givn
on serial verbs, Hook). The seeds of grammaticalization are therefore in a
correlated set of phenomena: Semantic suitability, salience, and frequency.
Only the third actually leads to grammaticalization and hence to fixing,
freezing, idiomatization, etc.
10. SOME QUESTIONS FOR THE FURTHER STUDY OF

GRAMMATICALIZATION
Among a number of questions still to be answered is what the languageexternal motivation for grammaticalization may be. If semantic suitability,
salience and frequency are among the prerequisites for grammaticalization to
start, then the question still remains as to what motivates the beginning of
the process. Is it discourse-pragmatic pressure, that is, the need to be informa
tive and processable and expressive all at the same time (cf. Langacker, 1977),
the phenomenon of gaps in grammatical paradigms or in the universe of
abstract concepts, a "natural propensity" for signaling metalinguistic relations
in non-lexical ways (cf. Bybee and Pagliuca, 1985), or some other factor or
factors? These questions all point to linguistic problem-solving as a clue to
motivations for grammaticalization. Any claim about problem-solving and
functional purpose raises issues of teleology. Because grammaticalization does
not necessarily happen in any given instance of potential grammaticalization,
considerable caution needs to be taken in proposing teleological explanations.
Clearly we do not know enough yet about communication to argue that
"communicative necessity" motivates the development of grammatical cate-
10
gories. It is primarily through the study of what prevents grammaticalization,

or simply fails to trigger it, that we can begin to hope to answer the question
of motivation.
Also, what are the social and psychological dynamics that are present in
the initial and final stages of grammaticalization? How far is it induced by
individuals, and to what extent does it form an act of human creativity, rather
than being the result of language-internal drifts or forces? Hardly any answers
are available thus far to this type of question, despite extensive sociolingustic
research on the development of morphology. The question in most of this
literature has been posed primarily in terms of variation and continua (e.g.
Bickerton, 1975; Sankoff, 1980: Part II, Rickford, 1987), or of contact (e.g.
Romaine, 1988), but for the most part not in terms of psychological or
cognitive motivation (see, however, Slobin's (1985) suggestions regarding
cross-linguistic operating principles in child language acquisition).
Yet another unsolved puzzle is what motivates the differential speed with
which grammaticalization takes place in different functional domains. Obser
vations on African languages suggest that some kinds of developments pro
ceed faster than others. For example, new categories of tense and aspect have
emerged within a relatively short period, and in some cases a new morphology
evolving along the same grammaticalization pattern is already emerging,
competing with the old one. The development of noun class systems or of
verbal derivation, on the other hand, has been much more conservative:
morphological paradigms found today can be reconstructed as having already
existed in a similar form and function several millennia ago.
Another important question for any theory of language that grammati
calization poses is whether lexical morphemes are by nature different from
grammatical ones. If it is true that diachronically grammatical forms ulti
mately arise out of lexical items (although their immediate sources are often
other grammatical items), does this lead to the logical conclusion that they
cannot be innovated, whereas lexical items can? Among the highly stable
grammatical forms with no certain lexical origin are such forms as the IndoEuropean to of the demonstrative pronoun (later further grammaticalized to
a third person agreement marker). Are we justified in hypothesizing that this
form was once derived from a lexical item, as the diachronic perspective on
grammaticalization would suggest, or does it provide evidence that grammati
cal items can arise full-fledged, and if so, under what conditions? These and
other questions remain for future empirical study.
INTRODUCTION
11
11. ORGANIZATION OF THE VOLUMES

Arranging the many papers in a meaningful way turned out to be a
difficult task, and the result is to some extent arbitrary. In view of the number
and size of the contributions available we decided to divide the work up into
two volumes. This division, too, is to some extent arbitrary.
Volume I focuses on theoretical and methodological issues, and is mainly
concerned with developing the major themes for the study of grammaticalization outlined above.
Section A deals with some new perspectives on the development and
structure of grammatical categories. Hopper discusses five principles which
are relevant for describing grammaticalization. Lichtenberk draws attention
to the role played by diachrony in explaining synchronic facts, and to gradualness in grammaticalization. By looking at the relationships between temporal
packaging and information processing in serial verb constructions, Givn
concludes that grammaticalization is not a gradual process, but rather an
instantaneous one. Nichols and Timberlake develop a panchronic analysis of
morphological realignments resulting from the tension between innovation
and conventionalization.
In Section B, various perspectives on directionality are discussed. The
first two argue for unidirectionality. The papers by Heine, Claudi and Hnnemeyer and by Traugott and Knig deal, among other things, with cognitive
factors in directionality, especially the role of inference in metaphor and
metonymy. Frajzyngier studies directionality of fit in grammar between the
domain of the real world and that of speech. Unidirectional views on gram
maticalization are challenged in a number of different ways. Herring argues
for the independence of the development of grammatical elements out of
lexical items and out of discourse-contexts, and the bi-directionality of some
changes. Contributions by Greenberg and Campbell suggest that typically
unidirectional processes may be reversed under certain conditions. And Keesing shows how areal influence may lead to the elaboration of grammatical
structures through borrowing that is socially directional (from substrate to
superstrate), but not linguistically so.
The papers in Volume II are for the most part detailed studies of distinct
areas of language structure in particular languages or language families.
Section A focuses on the genesis of verbal categories such as tense and
aspect. Bybee, Pagliuca, and Perkins summarize their cross-linguistic findings
on future marking, and Hook discusses the notion of aspectogenesis in some
Indo-Aryan languages.
12
Section is concerned with various aspects of the development of

argument structure. Shibatani resumes earlier discussion on the development
of topic into subject, using evidence from Japanese and Philippine languages.
And Haiman explores further the relationship between word order and the
development of subject clitics. In a rather different vein, Mithun demonstrates
why in some languages, specifically Cayuga (Iroquoian) and Selayarese
(Austronesian), there was no need for the development of a subject category
at all. In the final part of this section, Rude and Carlson show how nonsubject marking arises out of verbal structures in Sahaptian and Klamath
and in Senufo respectively.
Two different patterns in the development of subordinate structures are
discussed in Section C. Genetti describes the transition from case postpositions
to clausal subordinators in Newari (Tibeto-Burman). Givn reconstructs the
transfer from relative clause structure to verbal complement subordination
in Biblical Hebrew.
Section D is devoted to expressions of modality. Thompson and Mulac
present the case of certain English "main clauses" (I think (that)) which are
reinterpreted as epistemic phrases, and Abraham deals with the development
of a set of modal particles in German.
Papers in Section E demonstrate a spectrum of cases of grammaticalization and a diverse array of factors at play. Matisoff investigates the develop
ment of a number of verbal and nominal particles, as well as of verbal
concatenations in the Southeast Asian linguistic area. Craig shows how in
Rama (Chibchan) one verb in different contexts develops multiple grammati
cal functions, one relating to argument structure, the other relating to tense,
aspect and mood; she proposes the term "polygrammaticalization" for this
kind of process. Lehmann discusses a number of emergent grammatical
patterns in modern German which may or may not be instances of incipient
grammaticalization, thereby pushing on the question of how one recognizes
an instance of grammaticalization.
ACKNOWLEDGMENTS
This introduction is clearly inspired by all the participants in the symposium on
grammaticalization. Particular thanks are due to Dan Slobin whose concluding re
marks made the coordination of issues easier, and to Paul Hopper, Frank Lichtenberk,
and Suzanne Romaine for comments on an earlier version.
INTRODUCTION
13
REFERENCES
Bickerton, Derek. 1975. Dynamics of a Creole System. Cambridge: Cambridge Univer
sity Press.
Bickerton, Derek. 1981. Roots of Language. Ann Arbor, MI: Karoma Publishers.
Bopp, Franz. 1816. ber das Conjugationssystem der Sanskritsprache in Vergleichung
mit jenem der griechischen, lateinischen, persischen und germanischen Sprachen.
Frankfurt: Andreische Buchhandlung.
Bybee, Joan. 1985. Morphology. A Study of the Relation between Meaning and Form.
Amsterdam: John Benjamins.
Bybee, Joan and William Pagliuca. 1985. "Cross-linguistic comparison and the devel
opment of grammatical meaning." In Historical Semantics and Historical Word
Formation, Jacek Fisiak (ed.), 59-83. Berlin: de Gruyter.
Croft, William. Forthcoming. Typology and Universals. Cambridge: Cambridge Uni
versity Press.
DuBois, John W. 1985. "Competing motivations." In Iconicity in Syntax, John Hai
man (ed.), 343-65. Amsterdam: John Benjamins.
Gabelentz, Georg von der. 1891. Die Sprachwissenschaft. Ihre Aufgaben, Methoden
und bisherigen Ergebnisse. Leipzig: Weigel Nachfolger.
Givn, Talmy. 1979. On Understanding Grammar. New York: Academic Press.
Heine, Bernd and Mechthild Reh. 1984. Grammaticalization and Reanalysis in African
Languages. Hamburg: Helmut Buske.
Heine, Bernd, Ulrike Claudi and Friederike Hnnemeyer. 1991. Grammaticalization:
a Conceptual Framework. Chicago: University of Chicago Press.
Hopper, Paul J. 1988. "Emergent grammar and the a priori grammar postulate." In
Linguistics in Context: Connecting, Observations, and Understanding, Deborah
Tannen (ed.), 117-34. Norwood: Ablex.
Hopper, Paul J. and Sandra Thompson. 1984. "The discourse basis for lexical cate
gories in universal grammar." Language 60:703-52.
Hopper, Paul J. and Elizabeth Traugott. In progress. Grammaticalization. Cam
bridge: Cambridge University Press.
Humboldt, Wilhelm von. 1825. "ber das Entstehen der grammatischen Formen und
ihren Einfluss auf die Ideenentwicklung." Abhandlungen der Kniglichen Akademie
der Wissenschaften zu Berlin:401-30.
Kemmer, Suzanne. 1988. The Middle Voice: A Typological and Diachronic Study.
Unpublished Ph.D. dissertation, Stanford University.
Kuryowicz, Jerzy. 1965. "The evolution of grammatical categories." Esquisses Lin
guistiques (2):38-54.
Labov, William. 1972. Sociolingustic Patterns. Philadelphia: University of Pennsylva
nia Press.
Langacker, Ronald W. 1977. "Syntactic reanalysis." In Mechanisms of Syntactic
Change, Charles N. Li (ed.), 57-139. Austin: University of Texas Press.
Lehmann, Christian. 1985. "Grammaticalization: Synchronic variation and diachronic
change." Lingua e Stile 20:303-18.
Li, Charles N. and Sandra A. Thompson. 1974. "Historical change and word order:
A case study in Chinese and its implications." In Historical Linguistics I: Syntax,
Morphology, Internal and Comparative Reconstruction, John M. Anderson and
Charles Jones (ed.), 199-218. Amsterdam: North-Holland.
14
ELIZABETH CLOSS T R A U G O T T AND BERND HEINE
Matsumoto, Yo. 1988. "From bound grammatical markers to free discourse markers:
History of some Japanese connectives." In Berkeley Linguistics Society, Proceedings
of the Fourteenth Annual Meeting, 340-51.
Meillet, Antoine. 1948 [1912]. "L'volution des formes grammaticales." In Linguistique
gnrale et linguistique historique, 130-48. Paris: Champion.
Rickford, John. 1987. Dimensions of a Creole Continuum: History, Texts, and Linguistic
Analysis of Guyanese Creole. Stanford: Stanford University Press.
Romaine, Suzanne. 1988. "Pidgins, creoles, immigrant, and dying languages." In
Investigating Obsolescence. Studies in Language Contraction and Death, Nancy
Dorian (ed.), 369-83. Cambridge: Cambridge University Press.
Sankoff, Gillian. 1980. The Social Life of Language. Philadelphia: University of
Pennsylvania Press.
Slobin, Dan I. 1985. "Cross-linguistic evidence for the language-making capacity." In
The Cross-Linguistic Study of Language Acquisition. Volume II. Theoretical Issues,
Dan I. Slobin (ed.), 1157-1256. Hillsdale, N.J.: Erlbaum Associates.
Sweetser, Eve. 1988. "Grammaticalization and semantic bleaching." In Berkeley Lin
guistics Society, Proceedings of the Fourteenth Annual Meeting, 389-405.
Talmy, Leonard. 1983. "How language structures space." In Spatial Orientation:
Theory, Research and Application, Herbert Pick and Linda Acredolo (ed.), 225-82.
New York: Plenum Press.
Wiegand, Nancy. 1987. Causal Connectives in the Early History of English: A Study
in Diachronic Syntax. Unpublished Ph.D. Dissertation, Stanford University.
SECTION A: GENERAL METHOD
On Some Principles of Grammaticization

Paul J. Hopper
Carnegie Mellon University
1.
INTRODUCTION1
The earliest known use of the word grammaticalization2 was apparently

by Meillet, who defined it as "the attribution of a grammatical character to
a previously autonomous word" (1912:131), and noted that in every case
where the ultimate historical source of a grammatical form was known, this
source could be shown to be an ordinary lexical word. While the search for
the origins of grammar was not new (see Lehmann, 1982:1-8), in retrospect
we are tempted to see in Meillet's formulations and in his use of the term
grammaticalization a sense that a specific kind of historical change was at
work which went beyond the neo-grammarian preoccupation with sound
change, Wortgeschichte, and analogy, and that he was concerned to explicate
in broad terms "the history" of "grammar", i.e. the historical antecedents to
structure, in a language. Such an interpretation is certainly sympathetic to
those who view grammatical structure as a self-contained component; and
presumably Meillet, who refers frequently to "le systme grammatical [d'une
langue]", would in theory have agreed. Yet there are also indications in his
writings that what is distinctively grammatical in a language is not its broader
structures but the individual grammatical forms which comprise this structure.
He notes (Meillet, 1925:25) that the specific identity of a language is not its
grammar in the broad sense; French and English are not distinct from one
another in showing a relationship between two nouns by means of a particle,
but by the form and position of the particle (preposed de in French, postposed
-s in English). Meillet's own philological investigations always emphasized
the history of individual grammatical forms, which are referred to unmistak
ably as 'singular facts' (faits singuliers), 'particular facts' (faits particuliers),
'specific features' (traits spcifiques), and 'particular processes of expression
of morphology' (procds particuliers d'expression de la morphologie) (Meillet,
18
PAUL J. HOPPER
1925:22-25). Moreover, these comparative grammatical studies were inter

spersed with etymological works on Indo-European cultural vocabulary with
out any sense that the one or the other aspect of reconstruction and history
was more important. Meillet's work, in other words, stressed the particular
over the general in practically all aspects of change.
It is not that Meillet did not hold a Saussurean view of structure. There
are plenty of references in his work to the grammatical system, and to the
system of [the] language. Yet this system consists almost exclusively of
morphology, and Meillet's diachronic bias rarely allows him to dwell on its
stability and systematicity; almost always when he uses the word 'grammar'
(la grammaire), it is in a phrase which stresses comparison and historical
relationships (grammaire compare, grammaire historique) rather than a syn
chronic system. It is a phenomenon of a later time that linguists came to
accord grammar (i.e. morphology and syntax) a privileged status in the study
of language, and that as a consequence grammaticization came to be seen as
the study of the origins of grammar (rather than of the origins of the array
of grammatical forms constituting its morphology, which is how Meillet
usually presents it). Indeed, in his presentation of grammaticization Meillet
includes such clearly lexical examples as German heute 'today' from a pre
sumed *hiu tagu 'this day (instr.)' alongside such evidently grammatical
examples as Modern Greek tha (future tense formative) from thel hina 'I
want that', showing that what was at issue was not whether this or that form
was "in" the grammar, but the more important question of the processes it
had undergone to attain grammaticization. That these processes were the
same for lexical forms as for forms which were unambiguously (i.e. by general
agreement) 'in' the grammar suggests that for Meillet the notion of grammaticalisation did not rest ultimately on a clear definition of a grammatical system
as such.
The broadened scope of the study of grammaticization as exemplified,
for example, in the work of Givn (1971,1979), Heine and Reh (1984), and
others, has shown that the range of phenomena to be studied is not restricted
to morphology, but includes what Givn has called syntacticization, the fixing
of pragmatically motivated word orders into syntactic constructions and
agreement patterns. (Meillet had in fact presented the change from free word
order in Latin to fixed word order in French (1912:147-8) as an example of
grammaticization, and by implication, in saying that in early Indo-European
"l'ordre des mots avait une valeur expressive, et non syntaxique; il relevait
de la rhtorique, non de la grammaire" (1937:365), made it clear that at least
in some languages word order was to be seen as part of grammar.) The more
ON SOME PRINCIPLES OF GRAMMATICIZATION
19
extensive definition of grammaticization implicit in this work raised the

question of whether, when grammaticization had done its work, there would
in the end be any room left at all for the notion of grammar in the sense of
static structural relationships (Hopper, 1987; 1988a; 1988b). If grammar is
not a discrete, modular set of relationships, it would seem to follow that no
set of changes can be identified which distinctively characterize grammaticiza
tion as opposed to, say, lexical change or phonological change in general.
The only way to identify instances of grammaticization would be in relation
to a prior definition of grammar; but there appear to be no clear ways in
which the borders which separate grammatical from lexical and other phe
nomena can be meaningfully and consistently drawn. Consequently, there
seems to be no possibility of constructing a typology of grammaticization, or
of constructing principles which will discriminate between grammaticization
and other types of change.
Support for such a conclusion comes from research into semantic change.
Traugott (1989) examines the kinds of semantic changes which accompany
grammaticization as well as semantic change in general, and shows that
grammaticization is not distinct from other kinds of semantic change, but
that several tendencies, perhaps to be subsumed under an even smaller number
of much broader tendencies, seem to govern most semantic change whether
thought of as lexical or as grammatical. Later in this paper, I return to the
question of the ultimate validity of the principles which I present here.
2.
GRAMMAR: SOME WORKING ASSUMPTIONS
While a definition of grammar, and hence of grammaticization, is prob

lematical for an individual language taken in isolation, some working assump
tions about grammar are possible from a cross-linguistic perspective. These
assumptions may provide an entry-point for an investigation of emergent
regularities that have the potential for being instances of grammaticization:
a.
Categories which are morphologized might safely be said to be part

of grammar. Bybee (1985) has shown that aspect, number, tense,
and case, among others, occur frequently across languages as affixal
morphology. But some of these may also occur in looser colloca
tions in the form of adverbials and other free elements. Therefore
categories which are commonly morphologized in one language
may be candidates for emergent grammatical constructions in
20
PAUL J. HOPPER
another; an example is Thompson and Mulac's suggestion (these

volumes) that the vernacular English 'I think' has assumed neargrammatical status as an evidential.
b.
Certain types of lexical items are known typically to evolve into

grammaticalized clitics and affixes. Such changes are now so richly
documented that only a few examples need be mentioned: Demon
strative pronouns become articles and class markers; copular verbs
and verbs of motion become aspectual morphemes; nouns denoting
a location become adpositions and eventually case affixes of various
kinds. The detailed list of such changes given by Heine and Reh
(1984:269-81) for African languages is of universal relevance. The
occurrence of certain lexical items in frequent collocations (for
example, when the word 'foot' repeatedly occurs in phrases like 'at
the foot of the hill', etc.) may be prima facie evidence of incipient
grammaticization.
The application of such cross-linguistic generalizations about grammati

cization is a standard, if usually tacit, technique to guide an investigation of
grammaticization in a particular language. Such generalizations are especially
necessary if direct historical data are not available, a situation which may
occur not only for languages which have no historical records, but even for
languages with ample documented histories. Again, the English evidential 'I
think' analyzed by Thompson and Mulac provides a good example: Historical
data on this construction were sparse because it has presumably always been
restricted to the spoken language. Yet many languages possessed morphologized evidentials, and it seemed natural both to look for its pragmatic begin
nings in vernacular English and to focus on an epistemic verb like 'think' as
a possible candidate. But are there any "intra-language" principles by which
we can identify instances of constructions that might be said to be caught up
in grammaticization? Can we discriminate in a single-language (i.e. noncomparative, non-historical) context between accidental collocations and ones
which are moving toward some kind of grammatical status, however defined?
One approach to this question has been suggested by Lehmann in his
1985 paper, in which several concomitants of grammaticization are described.
These are:
Paradigmatization (the tendency for grammaticized forms to be

arranged into paradigms);
21
Obligatorification (the tendency for optional forms to become

obligatory);
Condensation (the shortening of forms);
Coalescence (collapsing together of adjacent forms);
Fixation (free linear orders becoming fixed ones).
Such principles are useful, indeed indispensable, as guides to historical

change, and have repeatedly proven their value in the study of grammaticization. They are, however, characteristic of grammaticization which has already
attained a fairly advanced stage and is unambiguously recognizable as such.
They work best, in fact, when the stage of morphologization has been reached.
It is then usually possible to see that the kinds of changes typified by these
labels have indeed occurred. By the time forms and constructions reach the
stage of grammaticization implicit in being able to apply this typology, the
question whether grammaticization has occurred is usually already answered.
The problem of identifying grammaticization when it is not already obvious
is precisely that the form or construction in question has not yet reached a
stage of being "obligatory", "fixed", and so on. When Lehmann's principles
are applied to the more labile sorts of phenomena that typically confront the
linguist, they are either not applicable (e.g. Paradigmatization), or else are
not distinctively examples of grammaticization (e.g. Coalescence.)
In this paper I will suggest some further principles, ones which (let it be
said at the outset) share some of the same defects as Lehmann's, in that they
also characterize aspects of change in general, and are not distinctive for
grammaticization. In fact, the earlier discussion has made it clear that we
should expect that distinctive "principles of grammaticization" can be formu
lated only to the extent that regularities in language can in principle be
isolated which are unambiguously grammatical. Nevertheless I hope that
these principles will supplement those suggested by Lehmann in being charac
teristic of grammaticization not only at the later, more easily identifiable
stages, but also at the incipient stages where variable phenomena occur, and
where the question more cogently arises as to whether we might speak of
grammaticization. Like Lehmann's principles, too, they are not intended as
novel insights into the nature and course of linguistic change; on the contrary,
they are to a large extent part of the general lore about language change as
it has been elaborated by linguists since the Neogrammarians, or simple
extrapolations from that lore. But they assume a special relevance in the
context of grammaticization as being potentially diagnostic of the emergence
of grammatical forms and constructions out of already available material,
22
PAUL J. HOPPER
and also of different degrees of grammaticization where grammaticization has

already recognizably proceeded.
3.
THE PRINCIPLES
The five principles of grammaticization which I am proposing are as

follows:
(1)
(2)
(3)
(4)
(5)
Layering. "Within a broad functional domain, new layers are

continually emerging. As this happens, the older layers are not
necessarily discarded, but may remain to coexist with and interact
with the newer layers."
Divergence. "When a lexical form undergoes grammaticization to
a clitic or affix, the original lexical form may remain as an autono
mous element and undergo the same changes as ordinary lexical
items."
Specialization. "Within a functional domain, at one stage a vari
ety of forms with different semantic nuances may be possible; as
grammaticization takes place, this variety of formal choices nar
rows and the smaller number of forms selected assume more general
grammatical meanings."
Persistence. "When a form undergoes grammaticization from a
lexical to a grammatical function, so long as it is grammatically
viable some traces of its original lexical meanings tend to adhere
to it, and details of its lexical history may be reflected in constraints
on its grammatical distribution."
De-categorialization. "Forms undergoing grammaticization tend
to lose or neutralize the morphological markers and syntactic
privileges characteristic of the full categories Noun and Verb, and
to assume attributes characteristic of secondary categories such as
Adjective, Participle, Preposition, etc."
These five principles will now be discussed in turn.

3.1. Layering
Within a functional domain, new layers are continually emerging. As
this happens, the older layers are not necessarily discarded, but may remain
to coexist with and interact with the newer layers.
By a "functional domain", I mean some general functional area such as
23
tense/aspect/modality, case, reference, etc., of the kind which frequently

becomes grammaticized. The term is taken from Givn (1984:32-5). The
Principle of Layering refers to the prominent fact that very often more than
one technique is available in a language to serve similar or even identical
functions. This formal divers;ty comes about because when a form or set of
forms emerges in a functional domain, it does not immediately (and may
never) replace an already existing set of functionally equivalent forms, but
rather the two sets of forms co-exist. They may be specialized for particular
lexical items, particular classes of constructions, or sociolingustic registers;
they may have slightly different meanings, or simply be recognized as "stylis
tic" alternatives. Sometimes layering merely represents a transition from one
technique to another, but it may also be quite stable. The different tech
niques e.g. (for tense/aspect), phonological alternation, affixation, periphra
sis with auxiliary verbs may exemplify quite vividly the different degrees
of grammaticization attained by the different layers.
The tenses and aspects of English provide some good examples. In the
past tense it is possible to distinguish an archaic layer of vowel alternations
in strong verbs like drive/drove, take/took. These formations co-exist with a
more recent layer making use of an apical suffix [t] or [d], as in notice/noticed,
walk/walked; this suffix was itself almost certainly derived from an auxiliary
verb cognate with English do (Lhr, 1984). In addition to the tense formations
with phonological alternation and suffixation, a periphrastic construction
involving do/did appears in questions, negatives, and emphatics. Of particular
interest in English are the many types of aktionsarten represented by construc
tions like:
(1)
(2)
(3)
(4)
(5)
keep (on) +-ing: He keeps (on) signaling to me (iterative)

go on + -ing: He went on asking silly questions (iterative)
be about + to: We were about to get on the plane (prospective)
start+to. They started to laugh (inceptive)
take + to. She took to criticizing his work habits (inceptive)
and others. Nuances of the "future tense" are especially richly represented;
in addition to be about +to already mentioned, there are (Quirk et al.,
1972:87-90):
(6) will: He will be here in half an hour
(7) be going + to. She's going to have a baby
(8) be+-ing. The plane is taking off at 5:20
(9) be + to: An investigation is to take place
24
PAUL J. HOPPER
All the members of these last two groups of forms make use of periphrasis,
which is typical of the most recent layer of grammaticized forms. Ablaut and
affixation, on the other hand, generally represent earlier layers. For example,
English is rich in examples of all three:
a.
b.
Periphrasis. We have used it (newest layer)

Affixation: I admired it (older layer)
Ablaut. They sang (oldest layer)
While different chronological stages may often be identified through such

structural clues as ablaut, affixation, and periphrasis, it is not necessary that
all layers be differentiated historically from one another; often, in fact, the
relative history of different layers cannot be easily inferred and is not docu
mented; to be + to and will both existed in Old English, for example. This
"cluttering" of grammar with functionally similar constructions, possibly
inherited from different times over a vast history, and ever being re-created,
is not easy to reconcile with any picture of a language as a homogeneous,
architectured, and delimited object.
The next principle, Divergence, is conceptually quite close to the principle
of Layering, and might easily be regarded as simply a special case of it.
3.2. Divergence
The Principle of Divergence, or Split, as Heine and Reh call it (1984:57-9
et pass.), refers to the fact that when a lexical form undergoes grammaticization, for example to an auxiliary, clitic or affix, the original form may remain
as an autonomous lexical element and undergo the same changes as any other
lexical items. The Principle of Divergence results in pairs or multiples of
forms having a common etymology, but diverging functionally. The grammat
icized form may be phonologically identical with the autonomous lexical
form, as has happened with the French word pas 'not; negative particle' and
its cognate Pas 'pace, step', or the two may be so distinct that the relationship
is completely opaque, as has happened with the English indefinite article a(n)
and the word one.
Divergence is perhaps to be understood as a special case of Layering.
Layering is somewhat different in that it involves different degrees of grammaticization in similar functional domains, usually of quite different lexical forms,
while Divergence is applicable to cases where one and the same autonomous
lexical item becomes grammaticized in one context and does not become
grammaticized in another. It is, however, not always possible to make this
25
distinction, since multiple divergences may result in several layers. For exam
ple, by a well described sequence of events, the Latin verb habere became a
future tense suffix in Modern French, yielding je chanterai ' shall sing' out
of (ego) cantare habeo 'I have to sing'; it also became the Modern French
lexical verb avoir 'to have'. But avoir itself came to be an auxiliary verb of
the perfect 'm j'ai chant 'I sang, have sung'. Thus j'ai is grammaticized as a
tense-aspect auxiliary beside the ungrammaticized lexical verb j'ai 'I have'
(divergence), but it also constitutes a secondary layer of grammaticization,
being a periphrastic perfect tense-aspect marker beside the older, suffixal
future -erai; schematically:
When the original form survives as a lexical item, we may still think of
divergence as having occurred more than once. We cannot distinguish in
principle between grammaticization and other types of change, nor can we
tell in principle when a change has resulted in grammaticization and when it
has not; consequently there may be no real difference between "divergence"
whose outcome is two different layers of grammaticization and "divergence"
which results in an autonomous lexical form and a cognate grammatical form.
3.3. Specialization
Specialization refers to the narrowing of choices that characterizes an
emergent grammatical construction. It corresponds quite closely to Leh
manns "obligatorification", the loss of choice which occurs when a form is
fully grammaticized. "Specialization", however, more closely matches the
process involved, in that it is only in the final stages of grammaticization that
the use of a form becomes obligatory, and it does not seem appropriate to
think of this evolution as being one in which forms become "increasingly
obligatory". Indeed, to regard the development as being toward the obligatory
use of a form in a construction is perhaps undesirable in that it tends to set
apart grammaticization from other kinds of changes, whereas "specialization"
26
PAUL J. HOPPER
is just one possible kind of change which may or may not lead to grammaticization.
A well-known example of specialization is that of negation in Modern
French, exemplified in:
(10) II ne boit pas de vin 'He doesn't drink wine'
In the ordinary negative clause, the verb is straddled by two negators, ne
preceding the verb and pas following it. Pas is also the general negative
particle, for example in pas beaucoup 'not much', and so on. Historically the
original negator was ne, and nouns like pas 'step, pace' could reinforce the
negation. It can be assumed that the reinforcing noun was once tailored to
the nature of the verb; thus verbs of motion could be reinforced with pas ('he
doesn't go a step'), verbs of giving and eating with me 'crumb' ('he didn't
eat a crumb; he didn't give me a crumb'), and so on. Gamillscheg (1957:753)
lists several such nouns denoting a 'least quantity' of something which could
be used in Old French to reinforce negation:
pas 'step, pace'
point 'dot, point'
mie 'crumb'
gote 'drop'
amende 'almond'
areste 'fish-bone'
beloce 'sloe'
eschalope 'pea-pod'
and others. By the 16th century, the only ones still used with negative force
were:
pas 'step, pace'
point 'dot, point'
mie 'crumb'
goutte 'drop'
Even in the 16th century, pas and point predominated, and by the modern
period these were the only two which were still in use. Of them, it is really
only pas which is a true negator, in that it has overwhelmingly higher discourse
frequency, and is the unmarked negator in the sense of being semantically
neutral (non-emphatic), and participating in a greater variety of constructions
{pas beaucoup but not *point beaucoup, etc.) This means that out of all the
competing forms in Old French which might have won out, only pas has
27
become a general negator. In the spoken language the ne of ordinary verbal

negation is usually dropped {je sais pas don't know'), leaving pas as the
only mark of negation (the stage which might be characterized as "obligatorification").
In his contribution to the present volumes, Hook (Volume II) has
described in detail the process of specialization as it is manifested in the
evolution of compound verbs in Indo-Aryan languages. Compound verbs are
double verbs which are found throughout the language family and might be
seen as evolving toward Main Verb + Auxiliary status. Hook shows in statisti
cal terms a trajectory which fits precisely with general linguistic expectations:
(1)
(2)
(3)
The textual frequency of compound verbs vs. simple verbs varies

in the different languages of the family. In comparable texts there
are about six times as many compound verbs in Hindi as in
Kashmiri, and twice as many in Gujerati as in Marathi. In other
words, grammaticization has proceeded further in Hindi than in
Kashmiri, further in Gujerati than in Marathi, and so on.
Concomitant with increased textual frequency, there is a tendency
to narrow the possible quasi-auxiliary verbs (known as vector verbs)
that can be used with a "main verb". In the Hindi text sample,
with a higher frequency of compound verbs, only 10 different verbs
were used as vector verbs; in the Marathi sample, 14 different verbs
were used as vector verbs.
The five most frequent vector verbs in the Hindi text sample
accounted for 92% of the total number of vector verbs; while in
Marathi the five most frequent vector verbs accounted for only
81% of the total number. Moreover, in both the Hindi and Marathi
text samples, the most frequently occurring vector verb was the
one meaning 'go'. But in Hindi 'go' accounted for 41% of all vector
verbs, while in Marathi 'go' accounted for only 32% of all vector
verbs.
A final observation that can be made about both the Indo-Aryan com
pound verbs and the French negation is that at the same time as one part of
an emergent construction is specialized, another part is, so to speak,
"released". Hook shows that the specialization of the vector verbs in IndoAryan is accompanied by a loosening of the constraints on the "main verb",
which from being restricted to concrete actions increasingly can refer to
mental actions and states. An equivalent loosening presumably also occurred
in the French negators, when pas 'step, pace' ceased to be restricted to verbs
28
PAUL J. HOPPER
of motion, so that the class of verbs which could be negated with ne.. .pas
widened. At an advanced stage of this process combinations of forms settle
into paradigms (Lehmann's "paradigmatization") in which one single auxil
iary form can combine with any stem whatsoever.
3.4. Persistence
The Principle of Persistence relates the meaning and function of a gram
matical form to its history as a lexical morpheme. This relationship is often
completely opaque by the stage of morphologization, but during intermediate
stages it may be expected that a form will be polysemous, and that one or
more of its meanings will reflect a dominant earlier meaning. A good illustra
tion of Persistence is the development in West African languages of object
markers ("accusative cases") out of former serialized verbs like take, described
in Lord 1982. Thus in G (Benue-Kwa) (Lord, 286-288), from which the
following examples are taken, the form k in sentences like:
(11)
k wl m-s
she OBJ book lay-down
'She laid the book down.'
functions as an "accusative case" marker. But k is originally a verb meaning

'take', which is moving toward grammaticization as a casemarker. Sentences
like these are historically of the type 'He took the book [and] laid [it] down,'
and reflect a previous stage of polysemy of k, which could function either
as a full verb 'take' or as an accusative case marker. In G (but not in all
languages of this group), a functional vestige of this polysemy remains: The
accusative marker cannot be used if the verb is (1) an effective verb (i.e. a
verb whose object is produced by the action):
(12) m wl 'She laid an egg.'
BUT NOT: * k wl m 'She k egg lay'
or (2) a verb whose object is experienced rather than affected:
(13) T t v Kk 'Tete saw Koko.'
BUT NOT: *Tt k Kk na 'Tete k Koko saw'
Thus G retains the restriction on the marking of direct objects which derives
from the historical antecedent of this morpheme in the lexical verb 'to take':
Only objects which can be 'taken' are marked morphologically as grammatical
objects of the verb.
29
The importance of making explicit the assumption of Persistence is

discussed by Bybee and Pagliuca (1986) in their study of the "future tense"
in English. When we try to identify ways of expressing the "future tense" in
English, or conversely to characterize the meaning of its most typical mor
pheme, the auxiliary verb "will", we find it necessary to make a variety of
semantic distinctions. Bybee and Pagliuca show that "the differences in the
uses of these future markers (i.e. will, shall, be going to) can be understood
as continuations of their original lexical meanings (117)". While prediction
(the "pure" future) is one of the senses of will in Present-Day English, as in
Bybee and Pagliuca's example:
(14) I think the bulk of this year's students will go into industry
other meanings (called "modal" meanings) also exist, for example:
(15) Willingness: Give them the name of someone who will sign for it and
take it in if you are not at home
(16) Intention: I'll put them in the post today
But these modal meanings were already found in Old English:
(17) Willingness: ... gif he us geunnan wile pt we hine swa
if he us grant
will that we him so
godne
gretan moton
generous greet should
'.. if he will grant that we should greet him who is so generous'
(Beowulf: 346-7)
(18) Intention: Wen ic pt he wille, gif he wealdon mot, In
think I that he will if he prevail should in
pm gup-sele Geata leode etan unforhte
the battle-hall Geat folk eat fearlessly
'I believe that he will, if he should prevail, devour the people of
the Geats without fear'
(Beowulf:442-4)
Bybee and Pagliuca suggest that the "predictive" future develops out of the
intention/promise use of will, and that the future meaning is established in
the Middle English period when inanimates incapable of volition begin to
appear as the subjects of will. When this happened, it did not result in an
across-the-board re-semanticization of will, the predictive future remains only
one of several distinct meanings of will in Present-Day English. All that
PAUL J. HOPPER
happened was that a new meaning was added to an already polysemous form,
and new distributional possibilities were opened for it.
3.5. De-categorialization
One way of characterizing the functional-semantic shift which forms
undergo as they move toward grammaticization is to say that they become
de-categorialized.3 In Hopper and Thompson 1984 a Categoriality Principle
was suggested, according to which the traditional categories of Noun and
Verb were to be viewed as proto-typical instantiations of the basic discourse
functions of identifying participants, especially participants new in the dis
course, and reporting events. It was claimed that the cross-linguistic morphosyntactic peculiarities of nouns and verbs were directly attributable to these
functions. Thus forms were decked out with characteristically noun-like attri
butes, such as articles, case markers, classifiers, and so on, to the degree that
they functioned to identify actual participants, but would lose these attributes
when no identification was involved, e.g. 'drove in the bus' vs. 'went there by
(0) bus'. The theoretical consequence of seeing categories in this way is to
relativize the notion of "category" to discourse, and hence to see membership
in a category as being, not determined in advance for a form, but secondary
to the deployment of the form in discourse; in other words, to replace the
idea of "a category" with one of "degree of categoriality".
The process of grammaticization can usually be seen to involve a loss of
the optional markers of categoriality. The functional counterpart of this "decategorialization" is a loss of discourse autonomy for the form: "noun"-like
forms no longer identify participants in a discourse, "verb"-like forms no
longer report new events. Instead, nouns may appear in secondary roles, such
as adverbial and prepositional ones:
(19) Our thanks were accepted by the mayor thanks to his generosity
(20) His face was pale in (the) face of these new demands
And verbs, too, usually in a participial form, may assume less central
functions:
(21) They saw the Northern Lights Seeing that you have declared
bankruptcy, you can hardly make any new investments
In these new roles, there is a freezing or loss of optionality in morphological
trappings. Thanks in the expression thanks to cannot appear as our thanks to;
31
in (the) face of cannot appear as in that face of seeing that cannot appear
as to see that, having seen that, and so on. Similarly with French pas:
(22)
Ils se sont loigns de trente pas

'they went thirty paces away'
(23) Ils ne fument pas
'they don't smoke'
(24) *ils ne fument un pas
'they don't smoke one step'
And syntactic constraints on the use of a form, for example in co-reference,

no longer hold:
(25)(?) Sitting by the roadside to eat our sandwiches, a heavy shower
drenched us to the skin
(where at least the written language requires that the subject of the main
clause and the understood subject of the participial clause be identical), but
(26) Considering its narrow beam, the boat is remarkably sea-worthy
(where no variety of English requires such identity).
As can be noted from these examples, the loss of prime categoriality
goes hand-in-hand with the tendency for the ex-nouns and ex-verbs to assume
discourse functions. Thus considering in (26), which is perhaps to be described
as a preposition (see Quirk et a l , 1972:328-9 for comparable examples),
functions very much like a concessive conjunction, but has no properties of
the verb to consider. This loss of discourse autonomy is precisely what
Traugott 1982 has described as a central feature of grammaticization: In her
terms, forms as they become grammaticized change their meanings from
"propositional" to "textual", that is, from having a meaning independent of
the text they change to having a meaning or function that is relative to the
text or to some local construction.
Seen in terms of the Categoriality Principle, grammaticization always
entails a loss of noun/verb categoriality. Indeed, categories other than noun
and verb almost always result from a "down-change" of these two primary
categories. Possible exceptions are a small group of adjectives denoting basic
properties like 'big' and 'small' (see Dixon, 1982), and deictics like 'this' and
'there', whose non-deictic sources are often obscure.
32
4.
PAUL J. HOPPER
CONCLUSIONS
My first attempts to isolate and describe "principles" of grammaticization

were undertaken with a somewhat utilitarian impulse, that of identifying
potential instances of grammaticization, possibly before a stage had been
reached at which forms were unambiguously part of the grammar of the
language. For this reason I tended to think of them as "heuristic" principles,
in the sense that with them the linguist would not necessarily begin with
already grammaticized forms and ask about their previous history, but would
be able to select from among the rhetorical resources of texts those recurrent
collocations that were candidates for being, at least marginally, "in" the
grammar of the language. They were, in other words, to be an entree into
the study of grammaticization in a particular language.
But if grammaticization is not already a given, the principles do not in
fact identify it unambiguously. They speak only to the question of "more"
or "less" grammaticized, not to the question of "in" or "out" of grammar.
Worse still, they do not discriminate between processes of change which result
in grammaticization and processes of change which do not result in grammati
cization. Layering, Divergence, Specialization, Persistence, and De-categorialization are not the exclusive domain of grammaticization, but are common
to change in general. If we consider, for example, the difference between the
noun mistress and the titles miss and mrs (misiz), which few would consider
to be an example of grammaticization, we find that it neatly exemplifies all
of the five principles of grammaticization discussed above:
(1)
(2)
(3)
Layering. Mistress, mrs, and miss represent different degrees of

reduction of what was originally the same word. They are different
"layers" of terms of address (to which some would now add a
further layer, ms [miz], intended to be neutral between miss and
mrs).
Divergence. The titular forms mrs, miss, and ms have split off from
the original noun mistress, but the noun form remains with (allow
ing for well-understood semantic changes) something like its origi
nal meaning.
Specialization. Although data on the antecedents to mistress as a
term of address are hard to come by, it seems probable that at one
time mistress was only one of a number of possible titles for women;
others probably included kin terms like mother, status terms like
widow, and familiar terms like gossip ('god-sib').
(4)
(5)
33
Persistence. Clearly the restriction of the title mrs to adult women

reflects the history of the noun mistress, as a feminine form of the
middle-class title master. The splitting-off of the title miss to distin
guish young or unmarried women from married ones dates only
from the 18th century (Jespersen, 1938': 186-7); it too, of course,
continues the historical restriction to females.
De-categorialization. The titles miss, mrs, and ms are de-categorialized in the double sense described above: They usually lack the
ability to take optional morphosyntactic trappings of nouns such
as articles, demonstratives, and possessive pronouns (an exception
must be made when mrs is re-instated as a full noun in the expres
sions 'my/your/his missus' and 'the missus'), and (with the same
exception) they may not alone refer to a discourse participant.
Are such titles to be regarded as part of grammar? Perhaps, but to do so

incurs the danger of defining grammar precisely in terms of principles of
grammaticization such as the ones presented here. The point is that since
grammaticization is always a question of degree, not an absolute, the criteria
which control this gradation are not restricted to grammaticization, but are
simply general criteria of change. The implication of this observation is that
there are no parts (modules, strata, etc.) to a language which are distinct
targets for change, subject to special kinds of changes, and so on; and this
in turn must be seen as an argument against stable holistic structures of
grammar.
We are thereby brought back to Meillet, who, we noted, backgrounded
the project of accounting for "the grammar" of a language, and insisted
instead on the study of the history of particular grammatical forms, in effect
treating grammar as all and only a collection of such forms. His notion of
grammaticalization ("the attribution of a grammatical character to a pre
viously autonomous word") says nothing about forms "entering the gram
mar" of the language, but suggests only that one and the same word is now
lexical, now grammatical in nature. In this core definition, in the same article
in which Meillet creates a new verb grammaticize, it is significant that he also
eschews the noun grammar in favor of the adjective grammatical.
NOTES
1.
I would like to thank the many people who commented on the first version of this paper,
and especially Elizabeth Traugott, Sandra Thompson, Bernd Heine, and Friederike Hn-
34
2.
3.
PAUL J. HOPPER
nemeyer. For the imperfections which have survived their suggestions and criticisms I am
wholly responsible.
Concerning "grammaticization" vs. "grammaticalization" I have no strong feelings. Some
have seen in the -al- form the hint of a suggestion that the resultant forms are "grammati
cal", i.e. part of "the grammar", and they avoid the -al- form for this reason. Allowing
for the slight possibility that such a distinction exists, I here use the form without -al-.
However, the -al- form is used when translating Meillet's term grammaticalisation.
I am indebted to Friederike Hnnemeyer for pointing out the relevance of the work of
Sandra Thompson and myself on Categoriality to grammaticization.
REFERENCES
Bybee, Joan L. 1985. Morphology. A Study of the Relation between Meaning and Form.
Bybee, Joan, and William Pagliuca. 1986. "The evolution of future meaning." In
Papers from the Seventh International Conference on Historical Linguistics, A.
Giacalone Ramat et al. (eds), 108-122. Amsterdam: John Benjamins.
Dixon, R.M.W. 1982. "Where have all the adjectives gone?" In Where have all the
Adjectives gone? and Other Essays in Semantics and Syntax, R.M.W. Dixon (ed.),
1-62. Berlin: Mouton.
Gamillscheg, Ernst. 1957. Historische franzsische Syntax. Tbingen: Max Niemeyer.
Givn, Talmy. 1971. "Historical syntax and synchronic morphology: An archaeolo
gist's field trip." In Papers from the Chicago Linguistic Society 7:394-415.
Givn, Talmy. 1984. Syntax I. Amsterdam: John Benjamins.
Heine, Bernd and Mechthild Reh. 1984. Grammatical Categories in African Languages.
Hamburg: Helmut Buske.
Hook, Peter. 1974. The Compound Verb in Hindi. Ann Arbor, ML: Center for South
Asian Studies.
Hook, Peter. Volume II. "The emergence of perfective aspect in Indo-Aryan lan
guages".
Hopper, Paul J. 1987. "Emergent grammar". In Berkeley Linguistic Society, Papers
of the Thirteenth Annual Meeting, 139-157.
Hopper, Paul J. 1988a. "Emergent grammar and the a priori grammar postulate." In
Linguistics in Context: Connecting, Observation, and Understanding, Deborah Tan
nen (ed.), 117-34. Norwood: Ablex.
Hopper, Paul J. 1988b. "Discourse analysis: Grammar and critical theory in the
1980's." In Profession 88, Phyllis Franklin (ed.), 18-24. New York: Modern
Language Association.
Hopper, Paul J., and Sandra Thompson. 1980. "Transitivity in grammar and dis
course." Language 56:251-299.
Hopper, Paul J., and Sandra Thompson. 1984. "The discourse basis for lexical
categories in universal grammar." Language 60(4):703-52.
Jespersen, Otto. 1955 (1938). Growth and Structure of the English Language. Garden
City, New York: Doubleday.
Lehmann, Christian. 1982. Thoughts on Grammaticalization: A Programmatic Sketch.
35
Kln: Institut fr Sprachwissenschaft, Universitt zu Kln. [Arbeiten des Klner

Universalienprojekts 48].
change." Lingua e Stile 20(3):303-318.
Lord, Carol. 1982. "The development of object markers in serial verb languages." In
Studies in Transitivity, Paul Hopper and Sandra Thompson (eds.), 277-300. Lon
don: Academic Press [Syntax and Semantics 15].
Lhr, Rosemarie. 1984. "Reste der athematischen Konjugation in den germanischen
Sprachen." In Das Germanische und die Rekonstruktion der indogermanischen
Grundsprache, Jrgen Untermann and Bela Brogyani (eds.), 29-90. Amsterdam:
John Benjamins.
Meillet, Antoine. 1912 (1948). "L'volution des formes grammaticales." In Lin
guistique historique et linguistique gnrale, A. Meillet, 130-148. Paris: Champion.
Meillet, Antoine. 1925 (1954). La mthode comparative en linguistique historique. Paris:
Champion.
Meillet, Antoine. 1937 (1964). Introduction l'tude comparative des langues indo
europennes. University of Alabama Press.
Quirk, Randolph, S. Greenbaum, G. Leech and J. Svartvik. 1972. A Grammar of
Contemporary English. London: Longman.
Thompson, Sandra ., and Anthony Mulac. Volume II. "A quantitative perspective
on the grammaticization of epistemic parentheticals in English."
Traugott, Elizabeth. 1982. "From propositional to textual and expressive meanings:
Some semantic-pragmatic aspects of grammaticalization." In Perspectives on His
torical Linguistics, Winfred P. Lehmann and Yakov Malkiel (eds.), 245-271.
Traugott, Elizabeth. 1989. "From less to more situated in language: The uni direction
ality of semantic change." In Papers from the Fifth International Conference on
English Historical Linguistics, Sylvia Adamson, Vivian Law, Nigel Vincent, Susan
M. Wright (eds.). Amsterdam: John Benjamins.
On the Gradualness of Grammaticalization

Frantisek Lichtenberk
University of Auckland
1.
INTRODUCTION
In recent years, a large number of studies of syntactic change have

emphasized the gradual nature of the developments in question, such as
changes in basic word order, the rise of new sentence structures, and the
development of grammatical markers of various kinds; see, for example, Lord
(1973), Li (1975), Chung (1977), Givn (1977), Timberlake (1977), Saltarelli
(1980), Gerritsen (1982), Romaine (1983), Silva-Corvaln (1984), Wiegand
(1984), and Hopper and Martin (1987), among many others, for different
perspectives on the gradualness of syntactic change. Characteristically, such
works are interested in the processes whereby one form and/or function is
transformed into another.
Although it is true that, given enough time, one structure may completely
replace another, it is also true that one commonly finds the old and the new
structure coexisting, often for a considerable period of time. Sometimes the
two are in free variation everywhere; sometimes they are in free variation in
some environments but not in others. Hardly anybody will deny that in every
language there are structures whose use is governed by rigid rules ("rigid" at
least at a given time). At the same time, there are, in every language, structures
whose functions are not easily describable by neat, compact rules. It is the
relation between structures and their functions that may be difficult to capture
by rules, not so much the structures themselves. The latter are relatively
amenable to being characterized by rules. A major virtue of the various
approaches to syntactic change that see syntactic change as gradual is not
just that they show the pervasiveness of variation (this can be demonstrated
by purely synchronic studies), but that they demonstrate the naturalness of
variation. Variation is a necessary consequence of the gradualness of language
change.
The present study falls squarely in the process-oriented, gradualist view
38
F R A N T I s E K LICHTENBERK
of syntactic change. It is concerned with a number of grammaticahzation

processes in To'aba'ita, an Austronesian language spoken on the island of
Malaita in the southeast Solomon Islands. 1 As the papers in the present
collection amply demonstrate, the term 'grammaticahzation' is given different
interpretations by different linguists. The view of grammaticahzation that I
take is this: Grammaticahzation is a historical process, a kind of change that
has certain consequences for the morphosyntactic categories of a language
and thus for the grammar of that language. The prototypical consequences
of grammaticahzation are:
(i) Emergence of a new grammatical category; (ii) Loss of an existing gram
matical category; (iii) Change in the membership of a grammatical category.
All three kinds of change may be historically linked. As I discuss in detail in
Section 2, To'aba'ita has a category of 'verb-like' prepositions, prepositions
that derive from earlier verbs. These prepositions have developed new variants
that belong in a different category. There is evidence that the prepositions
did not develop the innovative forms all at the same time. When the first
innovative form developed, a new formal category came into existence. As
other prepositions developed the innovative forms, the membership of the
new category changed. And if the innovative forms ever completely replace
the old forms, the category of verb-like prepositions will have disappeared
from the language.
The three types of change mentioned above may involve either a lexical
morpheme becoming a grammatical morpheme, for example a verb becoming
a case marker, or an already grammatical element acquiring new properties,
for example a case marker becoming a conjunction or a complementizer.
Many definitions of grammaticahzation characterize grammaticahzation in
terms of the processes, as the development of lexical morphemes into gram
matical ones. I prefer to view grammaticahzation in terms of its consequences,
as a process that leads to certain changes in the grammar of a language. This
emphasis on the effects of grammaticahzation processes rather than on the
processes themselves better reflects what I take to be an important characteris
tic of the grammars of natural languages. The grammars of natural languages
are ultimately historical phenomena, products of historical developments,
grammaticahzation among them. I will return to this point in the concluding
section.
As linguistic elements acquire new properties, they become members of
other categories. Categorial reanalysis is, of necessity, abrupt. One and the
same element cannot be simultaneously a member of two distinct categories
ON THE GRADUALNESS OF GRAMMATICALIZATION
39
(cases of inclusion apart). This, of course, does not preclude the possibility
of different tokens of a morpheme exhibiting properties characteristic of
different categories. This is because a form may be in the process of being
reassigned to a different category; some of its tokens exhibit the old properties,
others exhibit the new properties.
While categorial reanalysis is abrupt, its entry into the language and its
actualization are gradual. There are a number of respects in which a change
in morphosyntactic category may be actualized gradually. Thus a form that
exhibits properties characteristic of a certain, perhaps lexical, category may
begin to lose those properties not simultaneously but one after another. And
the innovative form does not displace the old form overnight. Instead, innova
tive forms typically start out as infrequently used variants, their frequency
increases over time, and ultimately they may completely replace the old forms.
An innovative form does not normally emerge all of a sudden throughout
the language community. It may be more common in some areas than in
others, more common with some speakers than with others. Typically, there
are generational differences. Frequently, there are differences among the
registers. An innovative form may enter a language through one register, and
its spread may not be equally rapid in all the registers. It may also happen
that more than one member of a category is in the process of undergoing the
same change but the individual items have not traveled equally far along the
course of the development.
Finally, we can also speak of gradualness of the acquisition of new
functions by an element provided we make clear what we mean by this kind
of gradualness. What I mean is this: If an element that has a function A
acquires a new function and if subsequently the element that has function
(and possibly still function A) acquires a function C, the change from A
to will be smaller than the change from A to would have been. I will
refer to this generalization as the Principle of Gradual Change in Function.
This principle can be represented schematically as follows:
ABC, not ACB
where , and stand for functions, the arrows signify changes in functions,
and is in some sense less different from A than is.
The principle does not say that changes necessarily take place in some
crosslinguistically definable minimal steps. All it says is that the change from
A to is smaller than the change from A to would have been. The size of
the steps may be language specific. To'aba'ita, like many other Oceanic
languages, makes a formal distinction between positive purpose ('in order
40
FRANTISEK LICHTENBERK
to') and negative purpose/precaution ('in order not to'). As will be discussed
in Section 4, in the history of To'aba'ita there was a change from an ablative
case marker to a negative-purpose/precaution complementizer followed by
another change whereby the negative-purpose/precaution complementizer
became a positive-purpose complementizer. One would not expect a similar
sequence of changes in a language that does not formally distinguish between
positive and negative purpose. Of course, one can speak of gradual change
in function only where a sequence of at least two changes has taken place. If
there is only one change, there is no gradualness. The principle does not
predict that sequences of changes will occur, but it does predict that if they
do, the changes will occur in certain orders and not in others. Needless to
say, in order for the Principle of Gradual Change in Function to be meaning
ful, one must be able to demonstrate for each instance of relevant change
that function is indeed less different from function A than is. Two
examples of gradual change in function will be given later in this paper.
Most of the aspects of gradualness of change in syntactic category
mentioned above are manifested in the To'aba'ita data to be discussed. Due
to the limited nature of the data nothing will be said about geographic and
register-based variation and only little will be said about generational
variation.
The paper is organized as follows: In Section 2,I discuss a set of preposi
tions that derive from earlier verbs. Most of the prepositions have variant
forms, but the relative frequencies of use of the variants are not the same for
all the prepositions. The innovative forms, the result of phonological reduc
tion, are members of a subcategory different from that of the old forms. In
Section , discuss a coordinating conjunction historically related to one of
the prepositions. We will see one manifestation of the Principle of Gradual
Change in Function there. In Section 4, I discuss a set of complementizers
that are historically related to some of the prepositions, and another manifes
tation of the Principle of Gradual Change in Function. In Section 5, I draw
some conclusions about the To'aba'ita data, about grammaticalization in
general and about the value of studies of grammaticalization for our under
standing of the nature of grammars of natural languages.
2.
VERB-LIKE PREPOSITIONS
In this section, I will discuss in detail one type of To'aba'ita prepositions.

These prepositions derive historically from verbs. They have retained some
41
of the formal properties of verbs but have lost others. The prepositions have
been undergoing a new change resulting in the loss of another verbal property.
This innovation a phonological reduction with a morphological conse
quence has been spreading gradually through the category.
2.1. Indexing of direct objects
Before discussing the prepositions, it is necessary first to discuss the
indexing of direct objects of transitive verbs. To'aba'ita is an SVO language,
but direct objects can be fronted to the left of the subject. Common-noun
direct objects that occur in their 'normal', postverbal position are indexed in
the verb by means of the object suffix -a unless the object is backgrounded
(see below). The suffix -a is used regardless of the animacy status of the object
and regardless of its grammatical number: 2
(1)
Wane baa ka
ngali-a kaufa baa ...
man that he:SEQ take -it cover that
The man took the cover, ...'
(2)
...ka
riki- wane 'eri
he:SEQ see -him man that
'... (and) he saw the man.'
(3)
...ka
naare-a
alo baa ki
she:SEQ roast -them taro that PL
'... (and) she roasted the taros.'
(4)
Kera
thaungi-a
ulu wela 'e -ki
they:PFV kill
-them three child that-PL
'They killed the three children.'
Proper-noun objects may but need not be indexed in the verb:

(5)
'O
{riki- riki} ni Fiona?
you(sG):PFV {see-her/see} ART Fiona
'Did you see Fiona?'
Objects that are backgrounded are usually not indexed in the verb
although under some conditions they may be. The factors governing the
backgrounding of objects are rather complex, but basically an object may be
backgrounded for one of two reasons: (i) because it is nonreferential, as in
42
(6) below; or (ii) because its referent is not judged to be important in the
given context, as in (7):
(6)
Kamili'a mili
lio ka
ai
mi
-si
we(EXCL)we(EXCL):PFV look it:sEQ NEG.VB we(EXCL) -NEG
{riki- riki} ta wane
{see -him /see} any person
'We looked but didn't see anybody.'
(7)
Nau kw ai
sau kuki 'a -ku a
I
I:IMPFV wash pot MID.vcE-my
'I am washing pots.' (as a reply to 'What are you doing?', not to
'What are you washing?')
In order to keep the discussion simple, backgrounded objects (of transitive

verbs and prepositions) will be disregarded in what follows. (More detail
about the relation between the importance of the referents of objects in
situations and their indexing may be found in Lichtenberk, 1988a.)
If a verb is not followed by a direct object, because the object has been
fronted, in relativization on the object, or in ellipsis, the suffix -a is used to
index singular objects regardless of their animacy status and also nonhuman
plural objects:
(8)
'Ai ne'e na koki
kasi-
t r e e t h i s FOC We(DU,INCL):iMPFV fell -it
'It's this tree that you and I will cut down.'

(9)
roo basi na thaina -karo'a

'e
thaathaungani-a
two bowREL mother-their(Du) she:PFV make
-them
'the two bows that their mother had made'
(10) Nia ka
ngali-a,
ka
alu-
laa
te'e kaufa
she she:SEQ take -them she:sEQ put-them inside one cover
'She took them [taros mentioned previously] and put them in a
cover.'
If the missing object is plural and human, the suffix -da is used:
(11) To'a na ku
riki-da
ki mai
'i fafo asi nena
people REL I:PFV see -them PL hither at top sea that
'Those are the people I saw by the sea.'
43
(With fronted plural human objects, the suffix -a is also possible, but it is not
common.)
(12) wela na kuki
thaungi-da
child REL we(INCL):IMPFV kill
-them
'the children that we will kill'
(13) Ku riki-da
I:PFV see -them
'I saw them [people].'
There is also a dual object suffix -davo'a 'them (two)', but it is rare; the
plural suffix is normally used instead.
When the object is an independent personal pronoun, it is not indexed
in the verb:
(14) ...kini
'eri ka
'adomi
woman that she:sEQ help me
'... the woman will help me.'
In the third person, there are two options: One is to use an independent
personal pronoun, in which case there is no object suffix on the verb; the
other is to use an object suffix and no independent pronoun; compare (15)
and (16):
(15) Ku
rongo nia
hear him
'I heard him.'
I:PFV
(16) Ku
rongo-a
hear -him
'I heard him.'
LPFV
The precise difference between the two constructions need not concern us
here. Suffice it to say that the independent-pronoun option is used if the
object is to be foregrounded (for example, in contrast). Second, the indepen
dent pronouns are rarely used with nonhuman reference.
2.2. Verb-like prepositions: Similarities to transitive verbs
To'aba'ita has several prepositions that, for reasons to become apparent
below, can be referred to as 'verb-like prepositions'. As will be seen in
44
Section 2.3, not all of To'aba'ita prepositions are verb-like. The following are
the verb-like prepositions:
'ani instrumental; also used as a 'general' preposition with the oblique
objects of many intransitive verbs
fasi (archaic fa 'isi) ablative: 'away from'
uri
allative: 'to, toward'; purpose (as in 'come for (to get) something')
suli prolative: 'along', 'during', 'about' (as in 'speak about something')
b
comitative: 'with' (as in 'John came with Mary.'); cf. faafi below
faafi confective, corresponding to 'with' in 'He came back with a large
parcel.'
With some exceptions, objects of the verb-like prepositions are indexed
in the same ways as are objects of transitive verbs. With the exception of the
comitative and the instrumental/general prepositions, adjacent nonbackgrounded lexical (as opposed to pronominal) objects are indexed by means
of the suffix -a, regardless of their animacy status and their grammatical
number. (As with transitive verbs, proper-noun objects may but need not be
indexed.) The forms of the comitative and the instrumental/general preposi
tions used in these environments will be discussed in detail in Section 2.4. As
will be seen there, the other four verb-like prepositions have parallel optional
variants. Following are a few examples of the indexing of adjacent nonbackgrounded lexical objects with prepositions other than comitative and instru
mental/general:
(17) Si
fanga 'e -ki ka
thada
fasi -a foko
-na
PARTV food that -PL they:SEQ fall from-it mouth -her
'The pieces of food fell from her mouth.'
(18) Thaari baa ka
thamo uri
-a tai si
fanga, ...
girl
that she:SEQ reach toward -it some PARTV food
'The girl reached for some of the food, ...'
(19) Wane 'e
ngata suli -a
wela nia ki
man he:PFV speak about-them child his PL
'The man spoke about his children.'
(20) Keka
oli
na'a faafi-a
roo subi
baa ki
they:SEQ return PERF with-them two k.o. war club that PL
'They went back with the two subi clubs.'
When there is no object adjacent to the preposition, that is, when the
object has been fronted, in relativization on the object, or in ellipsis, the suffix
45
-a is used with all six prepositions to index singular objects regardless of their
animacy status and nonhuman plural objects. When an oblique object is
fronted, the preposition is stranded behind:
(21) Tei na 'oki
lae bii -a?
who? FOC you(sG):iMPFV go with-him
'Who will you go with?'
(22) doo na to'a
ne'eki keki
fiifiru
ki 'ani-a
thing REL people this PL they:IMPFV fight PL with-them
'the things [weapons] the people used to fight with'
(23) Nia ne
nguu suli -a ka
'una
'eri: ...
she FOC:she:PFV sing about-it it:SEQ manner this
'It was her who sang about it like this: ...'
Nonadjacent plural human objects are indexed by means of the suffix
-da. (As with verbs, the dual object suffix -daro'a is used only rarely.)
(24) Kini
lakoo ki ku
dora
'ani
-da
woman that PL I:PFV not know GEN.OBL -them
'Those women I don't know.'
(Again as with verbs, the suffix -a is occasionally used with fronted plural
human objects.)
If the object is an independent personal pronoun, it is not indexed in the
preposition:
(25) Nia 'e
ngali-a naifa fasi
he he:PFV take -it knife from me
'He took the knife away from me.'
In the third person, there are two options: One is to use an independent
personal pronoun, in which case the preposition carries no object suffix; the
other is to use an object suffix and no independent pronoun:
(26) Nau kwai
lae {bii kera/bii -da}
I I:IMPFV go {with them/with-them}
'I will go with them.'
There is another property that, historically at least, most likely relates
the prepositions to transitive verbs: The stems of all of them end in i. Pawley
(1973) has reconstructed *-i as a transitive suffix for Proto-Oceanic, a stage
remotely ancestral to To'aba'ita (see Figure 1 below). The Proto-Oceanic
suffix is not functional in To'aba'ita, but the fact that all six verb-like preposi-
46
tions end in is unlikely to be due purely to chance. At least some of the

forms among the To'aba'ita verb-like prepositions are reconstructible
(whether with a verbal or a prepositional function) for Proto-Oceanic (see
Pawley, 1973).
Figure 1 gives the position of To'aba'ita in the Austronesian family. At
a low level, To'aba'ita is a member of the Cristobal-Malaitan subgroup.
Evidence from other Cristobal-Malaitan languages will be relevant in later
Figure 1. Position of To'aba'ita in the Austronesian family (after Blust (1980), Grace (1955),
Pawley (1972), Lichtenberk (1988b))
47
discussion. Only those Cristobal-Malaitan languages are included in the tree

diagram from which data will be cited.3
None of the six forms, 'ani, fasi, uri, suli, bii, and jaafi, function as verbs
in To'aba'ita. They cannot form predicates, and apart from the object suffixes
they do not take any elements associated with verbs: The subject-aspect and
subject-sequential markers, directionals, nominalizers, and so on. All six
prepositions have prepositional cognates elsewhere in Cristobal-Malaitan (see
Table 2 in Section 2.5), and most of them have verbal cognates in addition.
Fasi (with an archaic variant fa 'isi), the ablative preposition, has verbal
cognates in at least two Cristobal-Malaitan languages: Kwara'ae fa'asi 'leave,
forsake, depart from' and Lau fasi 'lose'.
Suli, the prolative preposition, has verbal cognates in at least two
Cristobal-Malaitan languages: Arosi suri 'follow' and Sa'a sulu 'follow'. Ver
bal cognates with the same meaning are also found outside the CristobalMalaitan group.
B, the comitative preposition, has verbal cognates in at least two
Cristobal-Malaitan languages: Arosi bei 'be one with, be an ally/partner' and
Sa'a pe'i 'be in the company of, be along with'.
Faafi, the confective preposition, also has verbal cognates in at least two
Cristobal-Malaitan languages: 'Are'are haahi 'load, burden, weigh down' and
Kwara'ae faafi 'accuse, impute'.
I have found no verbal cognates of the instrumental/general preposition
'ani or the allative/purpose preposition uri whether in Cristobal-Malaitan or
elsewhere.
Given the parallelism between the To'aba'ita verb-like prepositions and
transitive verbs in terms of the indexing of their objects, the stem-final on
the prepositions (see also Section 2.4 below), and the existence of verbal
cognates of most of the prepositions, one can assume that the prepositions
derive from earlier verbs. I ought to mention though that the main points of
this paper would not be materially affected if it should turn out that it was
the prepositions and not the verbs that were historically primary.
2.3. Other prepositions
Besides the verb-like prepositions, To'aba'ita has two other types of
preposition, which I will call 'noun-like prepositions' and 'bare prepositions'.
The noun-like prepositions index their objects by possessive suffixes, used in
possessive constructions to index the possessor NP. The noun-like preposi
tions are normally preceded by the bare preposition 'i (for which see below).
48
(27) Fale-a
tai fa
'ota
'i fa- wane 'ena
give-them some CLASS betel nut to to-his man that
'Give some betel nuts to the man.'
Compare:
(28) rake -na wane
belly-his man
'the man's belly'
The recipient/benefactive preposition fa will be of interest later (Section 2.6).
(29) Te'e kini
'e
'ono 'i buria -ku
one woman she:PFV sit at behind-my
'A woman sat (down) behind me.'
Compare:
(30)
alinga-ku
ear -my
'my ear(s)'
The bare prepositions do not index their objects:

(31) Kwai lae 'i 'Aoke 'i 'usungadi
I:IMPFV go to A.
in tomorrow
'I will go to Auki tomorrow.'
(32) 'E
fanga mala botho
he:PFV eat
like pig
'He ate like a pig.'
2.4. Reduced variants of the verb-like prepositions
All six verb-like prepositions exhibit two variant forms: 'ani/'ana instru
mental/general, fasi/fasa ablative, uri/ura allative/purpose, suli/sula prolative,
bii/bia comitative and faafi/faafa confective. For convenience, I will refer to
the first member of each pair as the 'i-form' and to the second member as
the '-form'. The i-forms were exemplified in Section 2.2, examples (17)-(26).
They index their complements in the same ways that transitive verbs index
their direct objects. The -forms, on the other hand, do not take any object
suffixes. In fact, as I will argue below, the -forms derive historically from
the -forms with the object suffix -a; that is, 'ana derives from 'ani-, fasa
from fasi-, and so on.
49
The -forms cannot be used in all the environments in which the /-forms
are found. Importantly, they can be used only with adjacent objects and
under the same general conditions that govern the use of the /-forms with the
suffix -a before adjacent objects. With most of the prepositions, both the
/-form with the suffix a and the -form are found in these environments:
(33) Nau ku
rake'iri {uri
-a /ura}
wane
I I:PFV be angry {toward-him/toward} man
'I am angry at the man.'
(34)
Weia 'e
thaka {fasi- fasa} luma
child he:PFV run {from-it/from} house
'The child ran away from the house.'
(35) u'unn {suli -a sula} saiklon Namu

story {about -it/about} cyclone Namu
'story about Cyclone Namu'
(36)
Wela ki keki
'adomi nau {faafi-a /faafa} raa
child my PL they:IMPFV help me {with-it/with} work
ki
my PL
'My children will help me with my work.'
With the comitative preposition, only the -form is possible in these

environments; the /-form is ungrammatical:
(37) Nau ku
thathami-a kwai
ngata {bia /**bii -a}
I
I:PFV want
-it I:IMPFV speak {with/ with-her}
thaari nai
girl
this
'I want to speak with this girl.'
With the instrumental/general preposition, apart from one minor excep
tion, only the -form is possible in these environments; the i-form is ungrammatical:
(38) Nau ku
dora
{'ana
/**'ani
a} kini
na'i
I
LPFV not know {GEN.OBL/ GEN.OBL-HER} woman this
'I don't know this woman.'
The form 'ani-a is possible in one restricted environment. It is optionally
used after the prolative preposition suli-ajsula to express 'about', as in 'speak
50
about something'. 'Ani- is redundant in this context; suli- or sula alone is

sufficient:
(39) Kwai u'unu
{sula 'ani
-a
/suli -a
sula)
I:IMPFV tell story {about GEN.OBL-them/about-them/about}
roo wela
two child
'I will tell a story about two children.'
This redundant use of 'ani-a appears to be archaic, not found in the speech
of the young people.
Although there is no direct historical evidence to appeal to, the conclusion
is inescapable that the -forms are a later development from the -forms with
the suffix -a through the loss of the stem-final ; that is 'ana comes from
'ani-, fasa from fasi-, and so on. If one were to assume that it was the
-forms that were historically primary, one would be confronted with an
unmotivated appearance of the stem-final and with unmotivated use of the
object-marking suffixes on what would earlier have been bare prepositions.
Furthermore, there would be no explanation for the present-day restricted
distribution of the -forms. The converse hypothesis that the -forms devel
oped from the i-forms with the suffix -a does not face any of these problems.
The -forms can be used only with adjacent lexical objects under the
same conditions that govern the use of the suffix -a on the -forms. They are
ungrammatical with independent personal pronouns as objects, which are not
indexed on the i-forms:
(40) Wane 'e
ngata {suli /**sula} 'oe
man he:PFV speak {about/ about} you(sG)
'The man spoke about you.'
Second, the -forms are not used when there is no adjacent object, that
is when the object has been fronted, in relativization, and in ellipsis. An
explanation for this will be given in Section 2.6. Suffice it to say here that the
reason has to do with the contrast between the object suffixes -a and -da in
those environments.
There are two additional pieces of evidence, each having to do with
individual prepositions, that the i-forms are historically primary and the
-forms later innovations. The first piece of evidence concerns the ablative
preposition. The ablative preposition fasi has an archaic, rarely used variant
fa'isi. Now, while the form fasi has an a-variant {fasa), the archaic form fa'isi
does not:
51
(41) Nau ku
ngal-a naifa {fasi- /fasa /fa'isi-a
I
I:PFv take -it knife {from-him/from/from-him
/**fa'isa) wela
/ from} child
'I took the knife away from the child.'
This suggests that by the time the ablative preposition began to develop an
-form, fa 'isi was no longer common and consequently failed to develop an
-form.
The other additional piece of evidence has to do with the diametrically
different frequencies of use of the -forms of the comitative and the confective
prepositions. This will be discussed in detail in Section 2.5.
All of the peculiarities of the use of the -forms can be readily accounted
for if we assume that the -forms derive historically from the /-forms with
the suffix -a through phonological reduction, a process that commonly befalls
grammatical elements. As we will see in Section 2.6, this reduction, the loss
of the stem-final i, has a structural consequence.
2.5. Frequencies of use of the innovative variants

With the comitative preposition and for all practical purposes with the
instrumental/general preposition also only the innovative -forms can be used
with adjacent nonbackgrounded lexical objects; the -suffixed i-forms cannot.
With the remaining prepositions, both variants are grammatical there. How
ever, there are considerable differences among the prepositions with respect
to the relative frequencies of use of the two variants.
To determine the relative frequencies of use of the two variants for each
of the prepositions, I have made a study of a corpus of oral texts. The corpus
consists of 15 texts totaling over 2,800 clauses. Most of the texts are traditional
stories, three are personal reminiscences, and one is an evening prayer. All
but one text come from a male speaker who is probably in his late fifties.
One text, the longest one (over 300 clauses), comes from the man's son, who
is in his early twenties.
Table 1 gives the relative frequencies of use of the -forms and the
-suffixed /-forms with adjacent nonbackgrounded lexical objects for all six
prepositions in the corpus. Note that the frequencies of 100% and 0% can
reflect either a rule of grammar (only one of the two forms is grammatical)
or a statistical property of the corpus (both forms are grammatical, but only
52
one occurs in the corpus). The former is indicated by (for 'categorial'), and
the latter by S (for 'statistical').
The figures in Table 1 demonstrate that the replacement of the a-suffixed
/-forms by the -forms has not progressed equally far for all the prepositions.
The process is complete with the comitative preposition and, for all practical
purposes, also complete with the instrumental/general preposition. The pro
cess is well under way with the aliative/purpo se preposition, less so with the
prolative preposition, and it has barely begun with the ablative and the
confective prepositions.
Table 1. Frequencies of use of the a-forms and the a-suffixed i-forms
before nonbackgrounded lexical objects
tokens
comitative
bia
bii-a
total
8
0
8
100 (C)
0(C)
24
45
69
'ana
'ani-a*
total
159
2
161
98.8
1.2
prolative
34.8
65.2
ablative
fasa
fasia
total
instrumental/general
allative/purpose
ura
uri-a
total
tokens
sula
suli-a
total
9
36
45
20
80
0
22
22
0(S)
100 (S)
confective
1
29
30
3.3
96.7
faafa
faafi-a
total
Grand total of tokens: 335

categorial; S statistical
* 'ani- is found only in an archaic construction following sula; elsewhere,
only 'ana is possible (100% (C))
If we are indeed dealing with a change in progress, we would not be
surprised to find some differences in the relative frequencies of the two variants
between the older and the younger speaker for those prepositions where both
variants are possible and both are relatively frequent (see Table 1). Since the
figures are derived from data that come from only two speakers, and since
53
the numbers of tokens for some of the prepositions are small (especially for
the younger speaker), one cannot draw any conclusions from them; neverthe
less there are differences between the two speakers, and they are in the
expected direction. With the allative/purpose preposition, the younger speaker
uses the innovative -form in 46% of all cases, compared to 30% for the
older speaker; and with the prolative preposition, the figures are 33% for the
younger speaker compared to 18% for the older speaker.
Together, the variation between the innovative -forms and the old
-suffixed i-forms with most of the prepositions and the differences among
the prepositions in the rates of use of their innovative forms are evidence of
a change in progress and of gradualness of the actualization of the change.
The change has been diffusing through the category; it has affected some
members of the category more than others.
There appear to be three factors responsible for the uneven spread of
the innovative forms. One has to do with the age of the innovative form. All
other things being equal, one would predict that the longer an innovative
form has been around, the more frequent it is relative to the old form. There
are case-marking cognates of all six prepositions elsewhere in the CristobalMalaitan group; see Table 2. Only the forms of the prepositions are given in
the table. The functions of the cognates in each set are pretty much the same
Table 2.
Case-marking cognates of the To'aba ita verb-like

prepositions
To'aba'ita
Lau
Kwara'ae
'ani/'ana
'ani/'ana
'ani
Kwaio
'Are'are
Sa'a
Arosi
'ani
'ani/'ana
ani/ana
To'aba'ita
Lau
Kwara'ae
faafi/faafa
fafi
faafi,
fafi7
faafi
Kwaio
'Are'are
Sa'a
Arosi
haaki
suli/sula
suli
suli,
suil4
suri
suri
suli
suri
bii/bia
fai(ni)5
faini,
fani, hain5,6
fe'eni5
bei
pe'
be'i
fasi/fasa
fasi
fa'asi,
fasi, hais6
fa 'asi
uri/ura
uri
54
across the languages, and the minor differences that there may be are irrelevant
to the discussion.
Judging from the available descriptions of the Cristobal-Malaitan lan
guages, only one preposition exhibits an alternation between an i-form and
an -form outside of To'aba'ita; none of the others do. (For Kwaio and
Kwara'ae, the absence of any such alternation has been confirmed to me in
personal communications by R. Keesing and K. Watson-Gegeo, respectively.)
The alternation is found only with the cognates of the To'aba'ita instrumental/
general preposition 'ani/ 'ana. Beside To'aba'ita, the alternation is found in
Lau, 'Are'are and Sa'a. Lau, a very close relative of To'aba'ita, belongs in
the same primary subgroup with To'aba'ita; 'Are'are and Sa'a belong in the
other primary subgroup (Figure 1 in Section 2.2). Since the alternation is
found in both subgroups of Cristobal-Malaitan but not in all the member
languages, its existence in To'aba'ita and Lau on the one hand and in 'Are'are
and Sa'a on the other is most likely the result of independent parallel develop
ments. It is true that one cannot disregard the possibility of independent
parallel developments in To'aba'ita and in Lau, but in the absence of any
such evidence one can assume that the 'ana form is older than the -forms
of the other To'aba'ita prepositions. And as Table 1 shows, 'ana has all but
replaced the -suffixed /-form. The factor of the age of an innovative form is
obviously irrelevant to the comitative preposition, even though there the
-form has completely displaced the -suffixed i-form.
Another factor that, apriori, might be relevant to the rate of replacement
of the old form of a preposition by the innovative form is the frequency of
use of that preposition. Since the innovative forms are reductions of the old
forms, one would predict that, all other things being equal, the more frequent
a preposition is relative to the other prepositions, the higher will be the use
of its innovative form, again relative to the other prepositions. The relative
frequencies of use of the six prepositions vis--vis each other in the corpus
are given in Table 3. The numbers of tokens, taken from Table 1, are aggre
gates of the numbers of occurrences of the -forms and the -suffxed i-forms
before nonbackgrounded lexical objects.
The instrumental/general preposition 'ani-a/'ana is by far most frequent,
accounting for nearly one half of all the verb-like prepositional tokens. It is
the general, not the instrumental function of the preposition that is responsible
for the high frequency of its use. The preposition is used with a large number
of intransitive verbs to introduce their oblique objects; for example, dora 'ana
'not know X', lalakwa 'ana 'not like X', 'ana 'be afraid of X', kwaithathai
'ana 'have X ready', 'be ready with X'.
55
Table 3. Relative frequencies of use of the verb-like

prepositions before nonbackgrounded lexical
objects
tokens
'ani-a/'ana
uri-a/ura
suli- a/sula
fasi-a/fasa
faafi-a
bia
161
69
45
30
22
8
48.1
20.6
13.4
9.0
6.6
2.4
total:
335
100.1
That frequency of use of a preposition is a factor in the rate of use of

its innovative form can be seen from a comparison of the frequencies of use
of the individual prepositions (from Table 3) with the frequencies of use of
their innovative forms (from Table 1). The comparison is given in Table 4.
Table 4.
Comparison of the relative frequencies of use

of the verb-like prepositions and the fre
quencies of use of their -forms
relative frequency
of use (%)
'ani-a/ 'ana
uri-a/ura
suli-a/sula
fasi-a/fasa
faafia
bia
total:
48.1
20.6
13.4
9.0
6.6
2.4
frequency of use
of -form (%)
98.8
34.8
20.0
3.3
0
100
100.1
With one glaring exception, the figures in Table 4 show a direct correla
tion between the frequency of use of a preposition and the frequency of use
of its -form. The higher the frequency of use of a preposition, the more
common the use of its innovative -form relative to the old -suffixed i-form.
The comitative preposition is again an exception. Even though it is least
common of all six prepositions (but see Section 3 below), its innovative form
has completely replaced the old form in the relevant environments.
56
The last factor relevant to the spread of the innovative -forms has to
do with the phonological shape of two of the prepositions. The comitative
preposition is the only one without a medial consonant; moreover, in the
i-form the two vowels adjacent to each other are identical: bii-a. Given the
stress rules of To'aba'ita (see the next paragraph), the first is stressed; the
second is not. The adjacency of the stem-final unstressed to the preceding
stressed must have been conducive to the speedy loss of the stem-final
before another vowel (the suffix -a). Note that the reduction does involve loss
of one of two vowels rather than shortening of a long vowel: bii is phonetically
[mbii], not [mbi:]. The reason why the stem-final was not lost when there was
no object adjacent to the preposition even though the same phonological
conditions obtained will be given in Section 2.6.
The stress factor also helps to account for the slow spread of the -form
of the confective preposition faafi. Although faafa is grammatical, it is rare
in speech, and it does not occur at all in the corpus. The confective preposition
is the only one with a stem of three syllables; the stems of all the others are
disyllabic. When the preposition carries the object suffix a (or -da), stress is
assigned not only to the first syllable, but also to the stem-final syllable:
[]. (In To'aba'ita, the first syllable of a word is always stressed, the last
syllable of a polysyllabic word is never stressed, and there cannot be more
than two unstressed syllables in a row.) It is the presence of stress on the
stem-final that inhibits the loss of the vowel.
The diametrically opposed frequencies of use of the -forms of the
comitative and the confective prepositions are attributable to differences in
the phonological properties of the prepositions, but only on the assumption
that it is the i-forms, not the -forms, that are historically primary. If one
assumed that it was the -forms that were historically primary (bia, faafa),
no such phonologically-based explanation would be available.
There are thus three factors responsible for the uneven spread of the
innovating -forms of the verb-like prepositions in present-day To'aba'ita:
The length of time the innovating form has been in existence, the frequency
of use of the prepositions, and, with two of the prepositions, their phonologi
cal form; but none of the factors is relevant to all six prepositions.
2.6. Emergence of a new syntactic subcategory
The rise of the -forms of the prepositions is the result of phonological
reduction: The stem-final is lost before the suffix a (when the object follows
the preposition). This straightforward phonological process has, however, a
57
structural repercussion. Whereas the old forms are bimorphemic, 'ani-a, uri-a,
etc., the new forms are monomorphemic, 'ana, ura, etc. The monomorphemic
status of the -forms is a consequence of To'aba'ita phonotactics. At the
phonological level, there are no closed syllables in To'aba'ita. (At the phonetic
level, closed syllables may arise through vowel deletion under certain condi
tions, but none of these are relevant to the prepositions.) If the final a of the
-forms were a separate morpheme, this phonotactic condition would be
violated: **'an-a, **ura, etc. Although the condition would not be violated
in the case of the comitative preposition (bi-), there is further evidence that
the -forms of all the prepositions, including the comitative one, are mono
morphemic.
As mentioned repeatedly, the a-forms cannot be used if there is no object
adjacent to the preposition (when the object has been fronted, in relativization,
and in ellipsis). It is here and only here that the two object suffixes -a and
-da contrast, -da is used when the fronted or missing object is human and
plural; -a is used elsewhere:
(42) Tei no
ngali-a naifa {fasi-
/**fasa}?
who? FOc:you(sG):PFV take -it knife {from-him/ from}
'Who did you take the knife from?'
(43) doo na to'a
ki keki
fiifiru
ki {'ani-
/**'ana}
thing REL people PL they:IMPFV fight PL {with-them/ with}
'the things [weapons] the people used to fight with'
(44) Nau kwai
lae {bii da /*bia}
I I:IMPFV go {with-them/ with}
'I'll go with them.'
The exclusion of the -forms from these environments is readily under
stood if we assume that the innovative -forms are monomorphemic. If the
-forms were used without an adjacent object, the functional contrast between
the object suffixes -a and -da would be lost. It is true that there would still
be a contrast between no suffix and -da, but it is well known that languages
prefer relatively heavy and explicit coding of objects in such environments
(see e.g. Givn, 1979). With an adjacent lexical object, on the other hand,
only the suffix -a is possible on the -forms (regardless of the animacy status
and the grammatical number of the object). Since in this environment -a
never contrasts with -da, the loss of the morphemic status of the final does
not lead to a decrease in coding efficiency.
The factor of coding efficiency can be seen as blocking the use of the
58
-forms without an adjacent object only if we assume, once again, that it is

the i-forms, not the -forms that are historically primary. This functional
factor has blocked the spread of the reduced form of the comitative preposi
tion bia into the environments where the two object suffixes do contrast even
though the phonological conditions (lack of a medial consonant, and stress
placement) favors loss of the stem-final there (Section 2.5).
In two respects, the innovative forms of the verb-like prepositions are
like the bare prepositions 'i and mala (Section 2.3). Both types require an
adjacent object, and neither indexes the object. However, they are unlike each
other with fronted objects. With fronted objects, the -variants of the verb
like prepositions cannot be used; instead, the -variants must be used, the
preposition is stranded behind, and the object is indexed:
(45) Tei na 'oki
lae bii-a?
who? you(sG):IMPFV go with-him
'Who will you go with?'
On the other hand, when the object of a bare preposition is fronted, the
preposition must be fronted as well:
(46) Miki
gwee niu
'i 'usungadi
we(EXCL):IMPFV collect coconut in tomorrow
'We will collect coconuts tomorrow.'
(47a) 'I 'usungadi miki
gwee niu
in tomorrow we(EXCL):IMPFV collect coconut
'Tomorrow we will collect coconuts.'
(47b) **'Usungadi miki
gwee niu
'i
tomorrow we(EXCL):IMPFV collect coconut in
The a-variants of the verb-like prepositions form a separate prepositional
subcategory. Although they have verb-like variants, they themselves are not
verb-like. And although they are like the bare prepositions in some respects,
they are unlike them in others.
The reassignment of the verb-like prepositions into another subcategory
is not without precedent in the history of To'aba'ita. There is an earlier case
of categorial reanalysis of a verb-like preposition that involves a change more
radical than the developments discussed above. As mentioned in Section 2.3,
beside the verb-like and the bare prepositions To'aba'ita also has noun-like
prepositions. These prepositions index their objects by means of possessive
suffixes. One of the noun-like prepositions is the recipient/benefactive preposi-
59
tion fa (fe before a high vowel). Like all the other noun-like prepositions, fa
is normally preceded by the bare preposition 'i:
(48) Nau ku fale-a fanga 'i fa-na
wela ki
I
I:PFV give-it food to to-their child PL
gave food to the children.'
Compare:
(49) thaina-na
wela hi
mother-their child PL
'the children's mother'
(50) Nia kuki- raisi 'i fe -
she cook-it rice for for-me
'She cooked rice for me.'
Compare:
(51) thaina-
mother-my
'my mother'
Although today the recipient/benefactive preposition is noun-like, it
ultimately derives from a verb, Proto-Oceanic *pa(n)i8 'give'. In a large
number of Oceanic languages, *pa(n)i has developed into a verb-like recipi
ent and/or benefactive case marker. A detailed discussion of the reflexes of
*pa(n)i can be found in Lichtenberk (1985); here I will give only a brief
summary of its development in To'aba'ita and a few other Cristobal-Malaitan
languages.
*pa(n)i was reflected in Proto-Cristobal-Malaitan as *fani. Most likely,
*fani functioned only as a verb-like preposition, not as a verb. It took the
suffix *-a with adjacent nonbackgrounded lexical objects. (This use of the
suffix is common throughout Cristobal-Malaitan and must be posited for
the protolanguage.) Later, some time after the breakup of Proto-CristobalMalaitan, *fania underwent phonological reduction to *fana. The new form
*fana was structurally ambiguous: It was interpretable as monomorphemic
for the same phonotactic reasons that the -forms of the present-day To'aba'ita prepositions are monomorphemic. 9 Alternatively, it was interpretable
as bimorphemic, consisting of a stem *fa and the third person possessive
suffix *na. *-na (or possibly *-na) as a third-person possessive suffix is of
great antiquity; it can be reconstructed back far beyond the Proto-Cristobal-
60
Malaitan stage. (In To'aba'ita, -na can index singular or plural adjacent object
or possessor NP's; when there is no adjacent object or possessor NP, -na has
only a singular-indexing function; compare examples (48) and (49) above and
(27) and (28) in Section 2.3. The suffix was used in the same way in the
protolanguage.) For example:
(52a) *fana wane
to/for man
'to/for the man'
Or:
(52b) *fa
-na wane
to/for-his man
'to/for the man'
It was the latter analysis that eventually prevailed. Starting out as a verb,
passing through a verb-like prepositional stage and a phonologically and
morphologically reduced stage (it did not index its objects), the etymon ended
up as a noun-like preposition.
3.
A VERB-LIKE PREPOSITION AND A CONJUNCTION
Present-day To'aba'ita has six verb-like prepositions. Beside the verb-topreposition developments discussed above, some of the etyma have undergone
other functional changes. One has developed into a coordinating conjunction;
three others have developed into complementizers. It is the former develop
ment that I will discuss now; the latter will be discussed in Section 4.
To'aba'ita has a coordinating conjunction ma 'and', which has a variety
of uses. It can conjoin noun phrases, verb phrases and clauses, and in the
latter two functions it can be used both in symmetric and asymmetric coordi
nation.
Noun-phrase coordination:
(53) Koro
koki
kwaithathai 'ana
alo
we(DL,INCL) we(DU,INCL):IMPFV be ready
GEN.OBL taro
ma 'ota
ma 'ova
and betel nut and pepper leaf
'Let's get ready (some) taro, betel nuts and pepper leaves.'
61
Symmetric clausal coordination:

(54) Thaari 'eri ka
thathami-a
wela 'eri, ma wela 'eri
girl
that she:SEQ like
-him child that and child that
la'u bo'o thaari 'eri.
ka
thathami-a
he:SEQ like
-her also INT girl
that
'The girl liked the boy [lit.: 'child'], and the boy too liked the girl.'
Asymmetric clausal coordination:
(55) Kera
laungi
'ana foa -na raaraga 'eri, ma
they:PFV decorate oneself with fruit-its tree sp. that and
ka
le'a
'asia '.
it:sEQ be nice very
'They decorated themselves with the fruit of the raaraga tree, and
it looked very nice.'
Ma is not the only 'and' conjunction of To'aba'ita. 'And' coordination
can also be expressed by two forms that also function as the comitative
preposition: bia and b. (Historical connections between comitative case
markers and 'and' conjunctions are, of course, not uncommon crosslinguistically; see, for example, Givn, 1979 and references therein.) The form bia is
used when the conjunct to the right is a lexical NP; b is used when the
conjunct to the right is an independent personal pronoun. The distribution
of the two forms in their coordinating function parallels their distribution in
the comitative-marking function (Section 2.4):
(56) Thaina-na bia nia kero
lae 'ana
uusia, ...
mother-his and father his they(Du):PFV go GEN.OBL market
'His mother and father went to the market, ...'
(57) Tha Maeli b mera
lae ' ro'o
ura
ART Maeli and I
WE(DU,EXCL):PFV go in yesterday for
dee -laa
fish-NOM
'Maeli and I went fishing yesterday.'
Note the dual subject-aspect markers in (56) and (57); the subject NP's clearly
involve coordination.
62
Bia/bii and ma are interchangeable in some contexts:

(58)
Wane 'oro ki {ma /bia} kini

'oro ki ke -si
man many PL {and/and} woman many PL they-NEG
thathami-a
arai
kwao nai.
like
-him big man white this
'Many men and many women don't like this European.'
In the next excerpt from a text, first ma is used to conjoin two NP's, but
in the next clause it is bia that is used to conjoin two NP's coreferential with
the first pair:
(59) Maka nia ma thaina -na kero
keko
lae '.
father his and mother-his they(DU) they(Du):SEQ go PERF
nia bia thaina -na keko

sore-a wela 'eri, ...
father his and mother-his they(Du):SEQ tell-him child that
'His father and mother were about to go. The father and mother
said to their child, ...'
Although ma and bia/bii are interchangeable in certain contexts, there
are major differences between the two conjunctions. First of all, unlike ma,
bia/bii can conjoin only noun phrases; it cannot conjoin verb phrases or
clauses. This prohibition on the use of bia/bii in VP and clausal coordination
is well reflected in the corpus; see Table 5.
Table 5. Distribution of ma and bia in NP coordination
and VP/clausal coordination
NP
coordination
tokens
%
ma
bia*
53
30
63.9
36.1
total:
83
Grand total of tokens :435
VP/clausal
coordination
tokens
%
352
0
100 (C)
0(C)
352
There are no instances of bii in the corpus.

There are no instances of the form b used in coordination in the corpus.
(As mentioned above, bii is used only if the conjunct to the right is an
independent personal pronoun.) As Table 5 demonstrates, approximately 80%
63
of all cases of 'and' coordination (352 out of 435) in the corpus involve VP
or clausal coordination, but not once is bia used there. In fact, this absence
of bia from VP and clausal coordination reflects a rule of To'aba'ita grammar,
not merely a statistical property of the corpus. On the other hand, bia is well
represented in NP coordination.
Second, even though ma and bia/b are often interchangeable in NP
coordination, there is another difference between the two conjunctions. The
difference has to do with the animacy status of the conjoined NP's. Ma is
freely used to conjoin NP's regardless of their animacy status, be they human,
animate (other than human) or inanimate. On the other hand, bia/bii is used
almost exclusively to conjoin human NP's. (Since b requires the conjunct to
its right to be an independent personal pronoun, and since pronouns are
rarely used with nonhuman reference, one would not expect to find many
instances of b conjoining nonhuman NP's in any case.)
The distribution of the two conjunctions according to the animacy status
of the conjoined NP's in the corpus is given in Table 6. Since there are no
cases of coordination of nonhuman animate NP's in the corpus, the animacy
opposition in Table 6 is human vs. inanimate.
Table 6.
Use of ma and bia depending on the animacy

status of the conjoined NP's
human NP's
tokens
%
ma
bia
29
27
total:
56
51.8
48.2
inanimate NP's
tokens
%
24
3
88.9
11.1
27
The figures in Table 6 demonstrate convincingly that while bia is as

common as ma in conjoining human NP's, it is used only infrequently to
conjoin inanimate NP's, where ma is the norm. The numbers of tokens for
the younger speaker are too small to be of significance, but it is interesting
that in his six uses of 'and' he always used bia, never ma, and that two out
of the three uses of bia to conjoin inanimate NP's in the corpus are his.
Bia/bii is used commonly to conjoin human NP's, infrequently to conjoin
inanimate NP's, and not at all to conjoin VP's or clauses. This pattern of
distribution can be readily understood if we take into account the origin of
the conjunction. The conjunction derives historically from the comitative
64
preposition bia/bii. When one considers the animacy status of the objects of
the comitative preposition, one finds that the preposition is used almost
exclusively with human objects. Table 7 is a comparison of the comitative
preposition and the related coordinating conjunction according to the anim
acy status of the objects and the conjoined NP's respectively. In both func
tions, the forms are used predominantly with human NP's. (For reasons to
become apparent in the next paragraph, with the coordinating conjunction it
is the animacy status of the NP to the right of the conjunction that is relevant,
but there are no instances in the corpus of NP coordination where the
conjuncts do not have the same animacy status.)
Table 7.
Comparison of the uses of the comitative prep

osition and the related conjunction according
to the animacy status of the objects and the
conjoined NP's respectively
comitative
tokens
%
human
inanimate
31
2
total:
33
93.9
6.1
coordinating
tokens
%
27
3
90
10
30
Of the two coordinating conjunctions, ma is historically prior. In fact,

ma can be reconstructed with this function at least as far back as ProtoOceanic. Bia/bii is a relatively recent innovation that is beginning to make
inroads into the territory of ma. It is not by accident that it is in strongest
competition with ma where it is, in conjoining human NP's. When the
comitative preposition began to acquire a coordinating function, it was first
so used in an environment that was less different from the typical environment
of its prepositional function than the other potential environments were, that
is with human NP's to its right. Whether bia/bii will ever become common
in conjoining nonhuman NP's and whether it will ever be used in VP and
clausal coordination only future can tell. A priori, there is no reason why it
could not. In the Sa'a language, the cognate of bia/bii is used in VP/clausal
coordination:
(60) E i'o pe'i suke
he sit and beg
'He sat and begged.' (Ivens, 1918)
65
What is important is that bia/bii first became established in human-NP

coordination and not with any other type of conjunct. This is precisely what
the Principle of Gradual Change in Function (Section 1) says. The principle
also predicts that if ever the conjunction comes to be used commonly in VP
and clausal coordination this will happen only after it has been well established
in inanimate-NP coordination; this is because VP and clausal conjuncts are
more different from human-NP conjuncts than inanimate-NP conjuncts are.
(From Ivens' brief account of Sa'a pe'i one cannot tell whether the form is
common both in inanimate-NP and in VP/clausal coordination.) The first
(and so far the only) common coordinating use of biajbii closely reflects the
use of its source.
The acquisition of a coordinating function by what used to be only a
comitative preposition has given rise to a new subcategory of coordinating
conjunctions in To'aba'ita. Bia/bii is different from ma both functionally and
formally. Unlike ma, it is used to conjoin only NP's and predominantly
human NP's, and unlike ma it exhibits two forms, used according to the
category status of the conjunct to its right, lexical or pronominal.
Thus far I have assumed that the coordinating function of bia/bii is an
extension from the comitative prepositional function. It is, of course, conceiv
able, although for reasons to be specified below unlikely, that the coordinating
function developed directly from the original verbal function. Given the
meaning to be reconstructed for the original verb, *ba'i 'be with/in the
company of', one can assume that the typical objects of the verb were human.
Whatever was said above about the development of the coordinating function
from the comitative prepositional function would apply, mutatis mutandis, to
the development of the coordinating function from the verbal function.
Verbal reflexes of the original verb *ba'i have been found only in 'Are'are
and Arosi, neither of which is a member of the same primary subgroup of
Cristobal-Malaitan with To'aba'ita. None of the sister languages of To'aba'ita
in the North-Central subgroup of Cristobal-Malaitan for which data are
available has a verbal reflex of *ba'i although they all have prepositional
reflexes. Although the possibility of independent losses of the verbal reflex
cannot be disregarded, it is more likely that in the history of these languages
the etymon ceased to be used as a verb quite some time ago, before the
development of its coordinating function in To'aba'ita.
The development of bia/bii both in its prepositional and coordinating
functions illustrates a characteristic of grammaticalization called 'persistence'
by Hopper (This Volume). By 'persistence' Hopper means the reflection by a
grammatical element of its own history in terms of its functional and/or
66
formal properties. The comitative preposition and the coordinating conjunc

tion derive from an earlier verb. Given the meaning of the verb 'be with/in
the company of', it is a safe assumption that its direct objects were typically
human. And as we have seen, both the preposition and the conjunction
typically have human NP's in the relevant positions. Second, the original verb
most likely indexed its objects in ways not significantly different from the
ways in which objects of transitive verbs are indexed in present-day To'aba'ita.
This is still reflected in the existence of its reflexes bia and b and in their
distribution.
There is one more comment that needs to be made concerning bia. In
Section 2.5 I said that one of the factors affecting the rate of use of the
innovative -forms of the verb-like prepositions was the frequency of use of
the prepositions: All other things being equal, the higher the frequency of use
of a given preposition, the more frequent its -form. The comitative preposi
tion did not fit this pattern at all. Although, judging by the corpus, the
comitative preposition is least frequent of the six prepositions, bia has com
pletely replaced bii-a in the relevant context. However, since the form bia
functions also as a conjunction, one needs to combine the number of occur
rences of bia in the two functions. Bia occurs in the corpus eight times as a
preposition and 30 times as a conjunction. The total of 38 moves bia up to
the fourth place in Table 3, but this still leaves it behind the aliative/purpose
and the prolative prepositions, with which the process of replacement is
considerably less than half complete (Table 4). This means that it is indeed
the phonological shape of the preposition and the conjunction that is the
decisive factor underlying the rapid spread of the bia form.
4.
VERB-LIKE COMPLEMENTIZERS
4.1. Pairings of prepositional and complementizer functions

Beside the comitative preposition-coordinating conjunction functional
pairing just discussed, there is another kind of functional pairing that one
finds in To'aba'ita. Three of the deverbal etyma originally discussed in
Section 2 function not only as prepositions but also as complementizers. The
prepositions in question are: uri allative/purpose,fasi ablative, and suli prola
tive. The forms of the complementizers are uri, fasi and suli, respectively,
with or without the object suffix -a (see Section 4.2 below). (To'aba'ita also
has a few non-verblike complementizers, one of which is exemplified in
67
example (69) in Section 4.3.) Historical connections between case markers

and complementizers are common crosslinguistically, and their presence, as
such, in To'aba'ita occasions no surprise. What is unusual about the To'aba'ita
case is one of the pairings of the prepositional and the complementizer
functions. The functional pairings are given in Table 8.
Table 8.
Pairings of prepositional and comple

mentizer functions
prepositional
functions
complementizer
functions
uri
allative, purpose
suli
fasi
prolative
ablative
reason,
positive purpose
reason, 'until'
positive purpose
The pairing of allative and purpose prepositional functions with reason

and (positive) purpose complementizer functions (in uri) is common crosslin
guistically. (See further below for positive purpose.) The use of one comple
mentizer to signal both reason and purpose is not unusual. As Thompson
and Longacre (1985) point out, reason and purpose clauses have similar
functions: They both specify explanations for situations. The difference is
that purpose clauses encode as yet unrealized explanatory situations, while
reason clauses encode explanatory situations that may already have been
realized. The pairing of a prolative prepositional function with reason and
temporal-extent complementizer functions (in suli) is not unusual either. What
is surprising is the pairing of an ablative prepositional function with a (posi
tive) purpose complementizer function (in fasi). Both uri and suli have multiple
functions as complementizers, but fasi is used exclusively as a purpose comple
mentizer, more specifically a positive-purpose complementizer. Positive purTable 9.
Frequencies of use of the complementizers in the positive-purpose,

reason and temporal-extent functions
fasi
uri
43
20
total:
63
68.3
31.7
'until'
tokens %
reason
tokens
%
positive purpose
tokens
%
uri
suli
31
3
34
91.2
8.8
suli
1 100
68
pose can also be signaled by uri, but fasi is more common. Uri is used more
commonly to signal reason. Besides uri, reason can also be signaled by means
of suli, but suli is so used only rarely. The frequencies of use of the three
complementizers in the various functions in the corpus are given in Table 9.
Even though the pairing of an ablative prepositional function and a
(positive) purpose complementizer function is unusual, I will argue further
below (Section 4.3) that it is not an unnatural result of a sequence of changes.
Before addressing that problem, there are some other matters to be dealt
with.
4.2. Variant forms of the complementizers
At least two of the complementizers have two variants, one with the
object suffix -a and one without: uri-ajuri reason, positive purpose, and fasi-a/
fasi positive purpose. With the third, reason and temporal-extent comple
mentizer suli, the situation is unclear. In the corpus, the complementizer
always has the form suli-. When the form suli is suggested to speakers, it is
usually though not always rejected. Following are a few examples of the
complementizers:
(61) Fale- ta
si
fanga 'a-na wela nai {fasi- /
give-it some PARTV food to-his child this {pos.PURP-it/
fasi}
ka
bona
POS.PURP} he:sEQ be quiet
'Give some food to the child so that he is quiet.'
(62) Welana'i'e
angi {uri -a j suli-a} 'e
thaofa
child this he:PFV cry {REAS-it/REAS-it} he:PFV be hungry
T h e child cried because he was hungry.'
(63) E
ai
'o
-si lae bii nau uri 'oe
it:PFV NEG.VB yOU(SG)-NEG gO With me REAS yOU(SG)
kwai-na wane ramo
wife-his professional killer
'Don't come with me because you are the wife of a professional
killer [a man who kills in order to collect bounty put up for
someone's death].'
(64)
Wela baa 'e

to'o suli -a to'a
'e -Ki kera
lae naa.
child that he:PFV wait until-it people that-PL they:PFV go PERF
'The child waited until the people had left.'
69
When expressing purpose, uri and fasi may cooccur, in that order:
(65) Nia ka
sifo
uri ta
ia
'i Fafolifua uri
-a
he he:SEQ descend for some fish to Fafolifua POS.PURP-it
fasi
nia kai
ngali-a
mai, hai
na'are
POS.PURP he he:IMPFV take -them hither he:IMPFV roast
-a
'a
-
-them MID.vcE-his
'He went down to Fafolifua for some fish to take back and roast.'
The frequencies of use of the suffixful and the suffixless variants of the
complementizers in the corpus are given in Table 10. There are three instances
of the complex purpose complementizer uri- fasi. Both forms are counted
in their respective categories.
Table 10. Frequencies of use of the suffixful and the suffixless variants of the complementizers
tokens
fasi-a
fasi
3
40
1tokens
%
7
93
uri-a
uri
48
3
total:
43
total
Grand total of tokens: 98
51
tokens
%
94.1
5.9
suli a
suli
total
%
100 (?, S?)
0 (?, S?)
The relative frequencies of use of the two variants are not the same for
all the complementizers. There are no cases of the suffixless form suli in the
corpus. As mentioned above, the grammatical status of this form is unclear.
The figures for fasi and uri are virtually mirror images of each other. Once
again, the numbers of tokens for the younger speaker are too small to be of
significance (4 for fasi and 9 for uri), and at any rate there are no marked
differences between the two speakers.
We are dealing here with another change in progress, this time from an
earlier form with the object suffix -a to an innovative form without the suffix.
The innovation has not affected the three complementizers to the same degree.
The change is nearing completion with fasi, whereas it has only begun to
affect uri. And if the occasional, albeit usually hesitant acceptance of the
sufixless form suli is any indication, the change may be about to begin affecting
this complementizer as well.
There are no significant differences in the forms of the three compie-
70
mentizers, and consequently the differences in the spread of the innovative

forms are not attributable to differences in phonological shape. The frequency
of use of the complementizers may be partly relevant. The reason/temporalextent complementizer suli is by far the least common of the three comple
mentizers (Table 10), and it does not normally occur without the suffix. On
the other hand, the frequencies of use of the other two complementizers are
pretty much the same; nevertheless the frequencies of use of their suffixless
forms are diametrically different.
The factor that does appear to be relevant is the age of the innovation.
Complementizer cognates of the To'aba'ita verb-like complementizers are
found in several Cristobal-Malaitan languages; see Table 11.
Table 11. Complementizer cognates of the To'aba'ita verb
like complementizers in Cristobal-Malaitan lan
guages
To'aba'ita
fasi-a/fasi
positive
purpose
uri-a/uri
reason;
positive
purpose
Lau
fasi
positive
purpose
uri
positive
purpose;
quotative
Kwara'ae
fasi
purpose
(positive,
negative)
suli-a
reason
Kwaio
fa'asi-a
negative
purpose,
precaution
suri-a
reason
Arosi
suli-asulf!)
reason;
'until'
suri-a
reason
Fasi has suffixless cognates in Lau and Kwara'ae, while uri has suffixless
cognates only in Lau, a very close relative of To'aba'ita. The suffixless form
fasi thus appears to be older than the suffixless form uri, and it is much more
71
common. Suli, which does not normally occur without the suffix, has no
suffixless cognates.
Even though the innovation consists in removal of the object suffix, there
is no loss in coding efficiency. Since on the complementizers the suffix a
never contrasts with the plural suffix -da or with its own absence, it is
redundant, and its removal does not result in any loss of information. The
object suffix is also lost in the innovative z-forms of the verb-like prepositions,
but in a different way. There the phoneme a is retained, but it no longer
realizes a morpheme. Although they ultimately derive from the same historical
sources, the innovative prepositional and complementizer forms have
diverged: From fasi- we now have fasa and fasi respectively, and from uri-a
we now have ura and uri respectively. One could hypothesize that this diver
gence in form reflects a movement towards iconicity (one meaning, one form).
This, however, is doubtful because the prepositions too have suffixless forms
fasi and uri (used with independent personal pronouns as objects).
4.3. Pairing of the ablative and the purpose functions

Both fasi and uri are used to signal positive purpose, 'SI in order S2',
not negative purpose or precaution, 'SI in order not S2'. The pairing of an
ablative prepositional function and a positive-purpose complementizer func
tion in fasi is crossHnguistically unusual. When one looks elsewhere in the
Cristobal-Malaitan group, one finds the same pairing in Lau, a very close
relative of To'aba'ita, which has fasi ablative and fasi positive purpose.
In other languages one finds different pairings: Kwaio has fa'asi as an
ablative preposition and fa'asi as a complementizer with a function that
Keesing (1985) terms 'precautionary'. In this function, faasi often corres
ponds to English 'les. In Kwara'ae one finds fa'asi/fasi ablative and fasi
(with 'iri) purpose. According to Deck (1934), Kwara'ae fasi can be used to
signal both positive and negative purpose.
Finally, in at least two languages the etymon also functions as a verb:
Kwara'ae fa'asi 'leave, forsake, depart from' and Lau fasi 'lose'. These verbal
meanings (especially those of the Kwara'ae form) are highly compatible with
the ablative prepositional function. Both contain the notions of separation,
distancing, removal, and thus absence.
It is the ablative-precaution/negative purpose pairings in Kwaio and
Kwara'ae that are the key to understanding the ablative-positive purpose
pairing in To'aba'ita (as well as in Lau and Kwara'ae). Keesing (1985) says
72
that Kwaio fa'asi conveys the meanings 'SI lest S2', 'SI to avoid S2' and is
also used to signal warnings:
(66) Nau ku
aru- 'ubu -na nga kesi fa'asi- wane ta beri -a
I
I:SUBJ put-it inside-its ART case lest -it man DUB steal-it
'I put it in the case lest someone steals it.'
(67) Rugasi-a masari-la -i fa'asi-a ta kee-'
let go -it play -GER-its lest -it DUB bite-you(SG)
'Stop playing with it, or it might bite you.'
(68) '
sia aru-
yOU(SG) NEG p u t - t NEG
- fa'asi-a ta'a
t h e r e - i t s lest
ta
beri -a
-it p e o p l e DUB Steal-t
'Don't put it there, or people might steal it.'

Fa'asi signals negative purpose, precaution or warning: The event of the
first clause serves to avert the potential event of the second clause. Note that
even though the purpose may be negative, the purpose clause is grammatically
positive.
As far as Kwara'ae is concerned, Deck (1934) says that there are several
ways to mark negative purpose. It can be marked by means of sasi ... (kata)
or, more recently, by means of asu ... kata. Deck glosses all of these forms
as 'lest'. The purpose clause is grammatically positive. Negative purpose can
also be marked by fasi 'iri or by 'iri alone. With either of these two forms,
the purpose clause is grammatically negative. As mentioned earlier, fasi (in
combination with other particles, 'iri being one of them) is also used to signal
positive purpose. Although Deck provides examples of negative purpose being
marked by sasi ... kata, by asu ... kata, and by 'iri, and although he gives
examples of fasi marking positive purpose, he gives no examples of fasi
marking negative purpose. It may be that at the time when Deck wrote the
grammar of the language, the negative-purpose marking function of fasi was
relatively uncommon.
In To'aba'ita, fasi signals only positive purpose. To signal negative
purpose or precaution, ada, a non-verblike complementizer, is required:
(69) Riki -a ada 'oko
dekwe-a kuki 'ena
look at-it PREC you(SG):sEQ break-it pot that
'Look out so you don't break the pot.'
Precaution/negative-purpose complementizers are widespread in the
Oceanic languages. Positive-purpose and negative-purpose clauses typically
receive different marking. 10 Although the precaution/negative-purpose com-
73
plementizers of the Oceanic languages often correspond to English 'lest', they

have none of its archaic flavor. They are used commonly, sometimes even
without a preceding clause. In such independent constructions, the comple
mentizers typically signal warning. There is often an implication of precaution,
as in the next example from To'aba'ita:
(70) Ada dani ka
'arungi kulu
' laa
masu'u
PREC rain it:sEQ fall on US(INCL) in inside bush
[E.g. 'Let's stay at home'] 'We might get caught in the rain in the
bush.'
The notions of negative purpose and precaution are related to the notions of
separation, distancing and absence inherent in the meanings 'leave, forsake'
of the ultimate verbal source of the complementizers and in the meaning
'away from' of the related ablative prepositions. One does X in order for Y
not to happen, in order to avert Y.
Most likely, the development of the purpose complementizers took the
following course: In the language ancestral to To'aba'ita, Lau, Kwara'ae and
Kwaio, there was a verb *faasi 'leave, forsake' and an ablative preposition
of the same form. Later, the preposition acquired a new function, that of a
negative-purpose/precaution complementizer. This was followed by another
change in function whereby the complementizer lost its negative force and so
could be used to signal both negative and positive purpose. This stage was
still reflected in the Kwara'ae of over 50 years ago, as described by Deck
(1934). Later, a further change took place in the language ancestral to
To'aba'ita and Lau whereby the complementizer acquired an exclusively
positive-purpose marking function. The negative-to-positive purpose shift
may have been facilitated by the fact that even though the purpose originally
signaled by the complementizer was negative, the purpose clause was gram
matically positive. The function of the complementizer was adapted to the
grammatical structure of its complement.
If the development of an erstwhile verb into a positive-purpose marker
in To'aba'ita (and in Lau) postulated here is indeed the historical one, it is
another instance of gradual change in function. First, there was a change
from an ablative function, which contains the notions of separateness, distanc
ing and absence, into a negative-purpose/precaution marking function, which
too contains these notions. This was followed in the ancestor of To'aba'ita
and Lau by a change whereby those semantic components were removed.
Theoretically at least, the development could have been first from the ablative
function into a positive-purpose marking function, which would have involved
74
loss of the notions of separateness, distancing and absence, followed in

Kwaio by a change into a negative-purpose/precaution marking function,
which would have involved reintroduction of those semantic components.
The Principle of Gradual Change in Function predicts that such a sequence
of shifts in function will not occur.
4.4. Functional shifts among the complementizers
Before leaving the complementizers, one more matter merits discussion.
The complementizer suli serves to signal reason and also temporal extent
('until'). However, it is used only infrequently to signal reason, the comple
mentizer uri being much more common in this function (Table 9, Section 4.1).
At the same time, complementizer cognates of suli coding reason are found
both within (Table 11) and without Cristobal-Malaitan, and this function of
suli is clearly of some antiquity. Why then is suli used so infrequently in
To'aba'ita to signal reason?
The most likely explanation is that suli is being replaced in this function
by uri. Now, uri too has two functions: It signals reason and positive purpose.
In the latter function, it is in competition with fasi. In the absence of direct
evidence bearing on this issue, what will be said below is of necessity specula
tive but not implausible. Originally, positive purpose was marked by uri.
When the reflex of *fa'asi changed from a negative-purpose/precaution com
plementizer into a positive-purpose complementizer, it invaded the territory
occupied by *uri. Gradually, uri has been ceding its purpose-marking function
to fasi, while at the same time assuming the function of a reason comple
mentizer. In doing so, it came into competition with the original reason
marker suli, which has virtually ceded this function to uri.
If this scenario does indeed correspond to historical reality, we have here
a series of shifts in function in a push chain, some of which lead to a decrease
in iconicity (one form, one function), while others are in the direction of
restoring iconicity. When an element develops an additional function, there
is normally a decrease in iconicity because formal changes typically lag behind
changes in function. Thus the acquisition by *faasi of the function of a
negative-purpose/precaution complementizer led to a decrease in iconicity
(one form, more than one function). The change of the negative-purpose/
precaution complementizer into a positive-purpose complementizer also led
to a decrease in iconicity (one function, two forms). This latter stage has been
followed by a gradual withdrawal of the original positive-purpose comple
mentizer uri from this function, which is a change in the direction of restoring
75
iconicity. At the same time, this element has assumed a new function, already
signaled by suli. The result is a decrease in iconicity. However, suli, the
original reason complementizer, has by now almost ceased to serve this
function, which is a development in the direction of restoring iconicity.
5.
SUMMARY AND DISCUSSION
The thread that runs through the discussion of the grammaticalization

processes in To'aba'ita is the gradualness of the implementation of the
changes. One aspect of the gradualness is manifested by the synchronic
alternation between the innovative and the old forms of most of the verb
like prepositions and the verb-like complementizers. Judging by historical
evidence, at least some of the alternations have been in existence for gener
ations.
The innovations have affected the individual members of the two cate
gories to various degrees. In fact, the differences are extreme, ranging from
no or virtually no replacement to complete or virtually complete replacement.
The innovations have been diffusing through the categories lexically. They
have not affected the categories as wholes; rather, they affected certain mem
bers before others, and they have affected the members unevenly. Categorial
reanalyses proceeding by lexical diffusion have been noted by other investiga
tors; see, for example, Li (1975) and Romaine (1983). Here I have attempted
to identify the factors underlying the uneven spread of the innovative forms.
One factor is the frequency of use; another is the phonological properties of
two of the verb-like prepositions. The third factor, the age of the innovative
forms, although relevant to their spread is, of course, irrelevant to the
inception of the innovations.
In the case of the verb-like prepositions, the innovation a phonological
reduction has given rise to a new subcategory of prepositions. Similarly,
the acquisition of a coordinating function by the comitative preposition has
given rise to a new subcategory of coordinating conjunctions. The new
prepositional and conjunction subcategories, while exhibiting new formal
properties have, thus far at least, retained some of the properties functional
and/or formal of their sources. As pointed out by Hopper (This volume),
such functional and formal persistence is characteristic of grammaticalization.
The gradualness of the grammaticalization processes in To'aba'ita, and the
pervasiveness of persistence in grammaticalization in general call into question
those theories that view categorial reanalysis in terms of radical and rapid
76
restructuring (see, for example, Lightfoot, 1979). It is evident that languages,

or rather their speakers, are able to tolerate a great amount of variation
between new and old forms even when the new forms exhibit properties that
are quite unlike those of the preexisting forms.
The grammars of natural languages are never static; in every language
there are always areas that are in flux. Languages may strive toward greater
regularity and iconicity by eliminating anomalies and variation; at the same
time, however, new patterns emerge elsewhere in the grammar, introducing
new anomalies and new variation. Grammars are always noncomplete. In the
grammar of every language there are, at a given time, many rigid regularities;
at the same time, there are also many aspects in every grammar that are not
fully determinate and that are malleable to various degrees. Grammars do
provide patterns for the construction of discourse, but they do not determine
its grammatical form fully. Being noncomplete, they allow speakers a certain
degree of freedom in constructing discourse. For whatever reason, certain
novel patterns become established, which results in a reshaping of the
grammar.
Generative approaches to language typically posit, under various guises,
a dichotomy between grammatical systems or langue on the one hand and
language use or parole on the other, and it is the former that is taken as the
primary or the only object of study. Grammars are taken to be fully selfcontained and determinate systems. However, the relation between grammati
cal systems and language use is not one of strict opposition; rather it is a
symbiotic relation. Grammars shape discourse, and discourse, in turn, shapes
grammars. The merits of studies of grammaticalization lie not only in elucidat
ing specific aspects of the grammars of individual languages but also and
more importantly in revealing an important characteristic of natural
languages: The noncompleteness of grammars.11 Langue and parole are not
neatly separable: in order to better understand one, it is necessary to consider
the other as well. (On this point see also DuBois, 1988).
The gradual nature of grammaticalization is also manifested in sequential
changes in function. Here 'gradualness' has to do with the steps by which
new functions develop from earlier functions one after another. I have formu
lated a Principle of Gradual Change in Function which says that in a sequence
of changes in function, a function that is less different from the original
function will be acquired before a function that is more different from the
original function. The principle has obvious implications for comparative
syntax and syntactic reconstruction. If cognate elements in two or more
languages have different but 'related' functions, it can be used to
77
determine the most likely order of the developments of the functions and to
decide which of the functions, if any, the protolanguage is most likely to have
had (keeping the possibility of independent developments in mind). 12 Certain
sequences of changes are more natural than others.
It is hardly an accident that the innovative forms of the verb-like preposi
tions, without the object suffix -a, have, thus far at least, been introduced
only into those environments where the suffix is not functional, where its
removal does not result in any loss of coding efficiency. Functional factors
can play a role in grammaticalization. If the innovative forms do ever spread
into the environments where the suffix is functional, this will be another
instance of gradual change in function: The forms will have been first intro
duced into that environment where their presence has minimal structural and
functional repercussions.
Finally, the present study has underscored the importance of a historical
perspective for understanding grammars. Unless one considers the origin of
the To'aba'ita conjunction bia/bii 'and', its exclusion from VP and clausal
coordination, its rare use in conjoining inanimate NP's, and the distribution
of the two forms according to the categorial status of the conjunct to the
right will appear to be unmotivated, arbitrary syntactic and semantic restric
tions. And unless one considers the origin of the positive-purpose comple
mentizer/as/, the crosslinguistically unusual fact that the same etymon is also
used as an ablative case marker will remain an oddity.
Grammars are products of their own histories. If one's goal is strictly
language description, limiting oneself to synchronic facts is understandable.
If, however, one seeks to understand why certain aspects of a language system
are the way they are, considerations of diachrony may be crucial. Obviously,
diachronic explanations of synchronic facts are not the only valid kind of
explanation, but if we are serious about trying to understand the properties
of language systems, diachronic considerations have an important role to
play.
NOTES
1.
My research on the To'aba'ita language has been supported by grants from the University
of Auckland Research Fund. I am grateful to a number of participants at the Symposium
on Grammaticalization, especially E. Traugott, for valuable comments on an earlier
version of this paper.
2.
'
Unless specified otherwise, plural pronominal glosses, such as 'we' and 'them', have plural,
not dual, reference. * identifies reconstructions; ** signifies ungrammaticality.
78
3.
The sources of the data are as follows: Capell (1971) (Arosi), Deck (1934), Watson-Gegeo,
pers. comm. (Kwara'ae), Fox (1974), Ivens (1921) (Lau), Geerts (1970) (Are'are), Ivens
(1918), (1929) (Sa'a), Keesing (1985) (Kwaio). The To'aba'ita data come from my own
field notes.
4.
Kwara'ae has a pervasive process of metathesis, hence the variant suil (. Watson-Gegeo,
pers. comm.). See also note 6.
5.
The Lau, Kwara'ae and Kwaio forms exhibit an irregular reflex of the initial consonant,
instead of the expected b. Irregularities in the reflexes of labial obstruents are not
uncommon in Oceanic languages.
6.
K. Watson-Gegeo has informed me that faini is often reduced to jani and fa'asi to fasi.
Kwara'ae is undergoing a change o f / t o h, hence the variants hain and hais respectively
(with metathesis; see note 4).
7.
The formfaafi is from Deck (1934). According to K. Watson-Gegeo (pers. comm.), the
form of the preposition is fafi.
8.
Because of a disagreement among the witnesses concerning the nasal, a doublet has to
be reconstructed.
9.
In Lichtenberk (1985) I say that *fana was interpretable as *fan-a, where *-a was the
object suffix. However, for reasons of phonotactics this could not have been the case.
10.
The prevalence of negative-purpose/precaution complementizers in the Oceanic languages

is no doubt responsible for the presence of a negative-purpose/precaution complementizer
in Melanesian Pidgin (nogut, from English no good).
11.
The notion of the noncompleteness of grammars is related to what Hopper (1987) calls
'emergent grammar'. The notion of emergent grammar is more radical than that of
grammars being noncomplete; it denies any independent existence to grammars. On my
view, at any time certain aspects of grammars are quite fixed whereas others are fluid to
various degrees, but there is never a time where a grammar as a whole is fully determinate.
12.
The principle will be irrelevant to cases where the functions of cognate grammatical
elements belong in distinct domains, such as an andative directional marker ('thither') in
one language and a future tense marker in another language, both arising independently
of each other from an erstwhile verb 'go'.
REFERENCES
Ahlqvist, Anders. 1982. Papers from the 5th International Conference on Historical
Linguistics. Amsterdam/Philadelphia: J o h n Benjamins.
Blust, Robert. 1980. 'Austronesian etymologies." Oceanic Linguistics 19:1-181.
Capell, A. 197L Arosi Grammar. Canberra: Australian National University [Pacific
Linguistics, B-20].
Chung, Sandra. 1977. " O n the gradual nature of syntactic change." In Li 1977:3-55.
Davenport, Michael, Erik Hansen and H a n s F . Nielsen (eds). 1983. Current Topics in
English Historical Linguistics. Odense: Odense University Press.
Deck, N . C . 1934. Grammar of the Language Spoken by the Kwara'ae People of Mala,
British Solomon Islands. Wellington. [Polynesian Society Reprint 5].
79
DuBois, John W. 1988. "Discourse as pattern model for grammar: The possessor=
ergator affiliation." Paper presented at the Symposium on Grammaticalization,
University of Oregon, Eugene, May 1988.
Fisiak, Jacek (ed.). 1984. Historical Syntax. Berlin: Mouton.
Fox, C E . 1974. Lau Dictionary. Canberra: Australian National University. [Pacific
Linguistics, C-25].
Geerts, P. 1970. 're're Dictionary. Canberra: Australian National University [Pacific
Linguistics, C-14].
Gerritsen, Marinel. 1982. "Word order change in Dutch imperative clauses: The
interaction between contextual and syntactic factors." In Ahlqvist 1982:62-73.
Givn, Talmy. 1977. "The drift from VSO to SVO in Biblical Hebrew." In Li
1977:181-254
Grace, George W. 1955. "Subgrouping of Malayo-Polynesian: A report of tentative
findings." American Anthropologist 57:337-339.
Green, R.C. and M. Kelly (eds). 1972. Studies in Oceanic Culture History, Vol. 3.
Honolulu: Bernice P. Bishop Museum [Pacific Anthropological Records, 13].
Hopper, Paul. 1987. "Emergent grammar." Berkeley Linguistics Society 13:139-157.
Hopper, Paul. Volume I. "On some principles of grammaticization."
Hopper, Paul and Janice Martin. 1987. "Structuralism and diachrony: The develop
ment of the indefinite article in English." In Ramat et al. 1987:295-304.
Ivens, Walter G. 1918. Dictionary and Grammar of the Language of Sa'a and Ulawa,
Solomon Islands. Washington: Carnegie Institution of Washington.
Ivens, Walter G. 1921. Grammar and Vocabulary of the Lau Language, Solomon
Islands. Washington: Carnegie Institution of Washington.
Ivens, Walter G. 1929. A Dictionary of the Language of Sa'a (Mala) and Ulawa,
South-East Solomon Islands. London: Oxford University Press.
Keesing, Roger M. 1985. Kwaio Grammar. Canberra: Australian National University
[Pacific Linguistics, B-88].
Li, Charles N. 1975. "Synchrony vs. diachrony in language structure." Language
51:873-886.
Li, Charles N. (ed.). 1977. Mechanisms of Syntactic Change. Austin: University of
Texas Press.
Lichtenberk, Frantisek. 1985. "Syntactic-category change in Oceanic languages."
Oceanic Linguistics 24:1-84.
Lichtenberk, Frantisek. 1988a. "The pragmatic nature of nominal anaphora in To'aba'ita." Studies in Language 12:299-344.
Lichtenberk, Frantisek. 1988b. "The Cristobal-Malaitan subgroup of Southeast Solo
monic." Oceanic Linguistics 28:24-62.
Lightfoot, David W. 1979. Principles of Diachronic Syntax. Cambridge: Cambridge
University Press.
Lord, Carol. 1973. "Serial verbs in transition." Studies in African Linguistics
4:269-296.
Pawley, Andrew K. 1972. "On the internal relationships of Eastern Oceanic lan
guages." In Green and Kelly 1972:1-142.
Pawley, Andrew K. 1973. "Some problems in Proto-Oceanic grammar." Oceanic
Linguistics 12:103-188.
Ramat, Anna G., Onofriu Carruba and Giuliano Bernini (eds). 1987. Papers from the
80
7th International Conference on Historical Lingustics. Amsterdam/Philadelphia:

John Benjamins.
Romaine, Suzanne. 1983. "Syntactic change as category change by re-analysis and
diffusion: Some evidence from the history of English." In Davenport et al.
1983:9-27.
Saltarelli, Mario. 1980. "Syntactic diffusion." In Traugott et al. 1980:183-191.
Shopen, Timothy. 1985. Language Typology and Syntactic Description, Vol. II: Com
plex Constructions. Cambridge: Cambridge University Press.
Silva-Corvaln, Carmen. 1984. "Semantic and pragmatic factors in syntactic change."
In Fisiak 1984:555-573.
Thompson, Sandra A. and Robert E. Longacre. 1985. "Adverbial clauses." In Shopen
1985:171-234.
Timberlake, Alan. 1977. "Reanalysis and actualization in syntactic change." In Li
1977:141-177.
Traugott, Elizabeth C , Rebecca Labrum and Susan Shepherd (eds). 1980. Papers
from the 4th international Conference on Historical Linguistics. Amsterdam/Phila
delphia: John Benjamins.
Wiegand, Nancy. 1984. "Creating complex sentence structure." Berkeley Linguistics
Society 10:674-687.
Serial Verbs and the Mental Reality of 'Event':

Grammatical vs. Cognitive Packaging
T. Givn
Linguistics
Department
University of Oregon, Eugene
1.
THE PHENOMENON*
Verb serialization may be defined ostensively, and somewhat simplemindedly, in the following cross-linguistic terms:
(1)
An event/state that one language codes as a simple clause with a

single verb, is coded in another language as a complex clause with
two or more verbs.
Such a definition systematically excludes from the scope of phenomenon

structures that are coded in all languages by (at least) two verbs, such as:
(2)
Modality-verbs with complements (including purpose clauses):.

John wants to eat an apple
Mary ran to reach the house
(3)
Manipulative-verb with complements (including causatives):

Mary told John to eat an apple
Mary made John eat an apple
(4)
Cognition-utterance verb with complements:.

John knew that Mary lied to him
Mary said that she was sick
She saw him running out of the house
(5)
Main clauses with tightly-bound adverbial clauses:.

Having worked, he retired
Fearing for her life, she went back
He came running
82
T. GIVON
Serial-verb constructions have been described most often in non-Euro

pean languages, and have been considered somewhat exotic, given their
perceived paucity in European languages. They can be divided into a number
of more or less distinct types. We will illustrate the most common ones below,
by giving only their English morphemic glosses.
a.
Case-role marking.
In this type, most conspicuous in West-African (Stahlke, 1970; Hyman,
1971; Lord, 1973; Givn, 1975a), South East Asian (Matisoff, 1969;
Goral, 1980) and Oceanic (Durie, 1982; Crowley, 1987) languages,
different serial verbs are used as grammaticalized markers of nominal
case-roles, as in:
(6)
b.
a. she take-stick break

(patient)
'She broke the stick'
b. she walk go-market
(locative)
'She walked to the market'
he work give-her
(benefactive)
'He worked for her'
d. She take-knife cut meat
(instrumental)
'She cut the meat with the knife'
Verb co-lexicalization:
In this type, common in Mandarin (Thompson, 1973), Papua-New
Guinea (Pawley, 1966; 1980; 1987; Bradshaw, 1982;) and Oceanic (Crowley, 1987), two or more verb-stems are co-lexicalized to create a more
complex verbal concept. Such complex concepts or collocations
are just as stable, stereotyped and contextually predictable in charac
terizing conventionalized situations as single words. Thus consider:
(7)
a. She hit-break the glass

'She broke the glass'
b. She frighlen-die him
'She frightened him to death'
c. sleep-perceive
'He dreamed'
d. She eat-perceive the meat
'She tasted the food'
e. He cut-split wood
'He chopped wood'
SERIAL VERBS AND THE MENTAL REALITY OF 'EVENT'
c.
83
Deictic-directional marking:
In this type, common in Tok Pisin (Givn, 1988) and Oceanic (Crowley,
1987), Amerindian (Talmy, 1970), and Tibeto-Burman languages
(DeLancey, 1980), verbs with deictic values, such as 'come' and 'go', are
grammaticalized to impart those deictic values to other motion or transfer
verbs, as in:
(8)
a. He walked he-go
'He walked away (from a reference point)'
b. She took the book she-come
'She took the book toward (a reference point)'
d.
Tense-aspect marking:
This type, sometime indistinguishable from verb complementation (as in
(2) above), is found in -Pisin, West African, South East Asian,
Austronesian and Creoles, inter alia. Aspectual or modal functions are
marked by serial verbs, as in:
(9)
e.
a. He stay work
He is working'
b. He go work
'He will work'
b. He work finish
'He has already worked'
(future)
(perfective)
Evidentiality and epistemic marking:

The last general type, sometime indistinguishable from complementation
(as in (4) above), is found in Iroquois (Mithun, 1986) or spoken English
(Thompson and Mulac, Volume II), inter alia. Here verbs such as 'think',
'say', 'hear' or 'know' become evidential markers, as in:
(10) a.
b.
c.
d.
2.
(durative)
They say she's coming

I understand he's leaving
I think she's home
She's left, y' know
(hearsay)
(hedge)
(inference)
(co-option)
THEORETICAL ISSUES: GRAMMAR VS. COGNITION
The issues raised by the existence of serial-verb constructions, and their

unequal genetic and areal distribution, goes to the very heart of the relation-
84
T. GIVON
ship between grammar and cognition. Note, first, that one may give an
alternative, and seemingly non-ostensive, definition of verb serialization as:
(11) The use of more than one verb in a single clause that codes what
seems to be, at least prima facie, a simple single event.
Definition (11) remains a problematic straw man. On the structural side,
'single clause' is a notion that retains a high potential for circularity. One can
easily define 'clause' as a construction with a single verb at its core. On the
cognitive side, 'single event' is just as susceptible to the very same circular
definition, and linguists are notoriously prone to letting grammatical structure
define what is a 'single event' (cf. Bradshaw, 1982, p. 28; inter alia). Even when
one believes in the ultimate if rough isomorphism between grammatical
and cognitive organization, these circular definitions remain uncomfortable.
The substantive issue alluded to here has, of course, a well known
antecedent in the earlier discussion concerning the formal status of serial-verb
constructions (cf. Stahlke, 1970; Hyman, 1971; Awobuluyi, 1972; 1973; Lord,
1973; Bamgbose, 1974; Li and Thompson, 1973; 1974; Schachter, 1974): Are
they independent or embedded? Conjoined or subordinate? Are the serial
verbs themselves lexical or grammatical morphemes? Verbs or prepositions?
These questions are the formal isomorphs of the substantive issues concerning
the relation between grammar and cognition.
In approaching the relation between grammar and cognition, one may
adopt either one of two extreme positions. These positions may be given as
the two alternative approaches to cross-cultural translation. Consider the
standard 3-line linguistic transcription of an event clause:
Line 1: event clause in the source language
Line 2. morph-by-morph linguistic gloss
Line 3: free meaning translation
Position A, which I will call the extreme universalist position (cf. Katz, 1978),
holds that line-3 is the proper translation of line-l, thus an adequate represen
tation of the cognized event in the source language.
Position B, which I will call the extreme relativist position, 1 holds that
line-2 is the proper translation of line-l, thus an adequate representation of
the cognized event in the source language. Extreme universalists thus argue
from translation to cognition. Extreme relativists argue after Whorf
(1956) from grammar to cognition.
In arguing from grammar to cognition, Pawley (1980, 1987) has staked
out what can be interpreted, at least under one reading, as an extreme culture-
85
relative position. 2 Languages that use serial verb constructions, Pawley seems
to be suggesting, differ fundamentally from those (like English) that don't, in
the way their speakers cognize or package unitaiy 'events'. Speakers of serialverb languages view some of our unitary 'events' as a concatenation of
fragmented sub-events. To support his theoretical position, Pawley quotes
from Grace (1983):
.. .The syntactic function which more than any other seems to hold the
key to human language is that which permits the specification of what I
call "conceptual situation" ... The conceptual situation is a model of
clause-sized chunk of reality or imagined reality. The syntactic mechanisms
involved are primarily those which mark the case-relations of the verb ...
(1983, pp. 7-8; emphases are mine; TG)
In summarizing his own specific observations, Pawley notes:
... We may conclude from the foregoing that there is no universal set of
episodic conceptual events. Indeed, it seems that languages may vary
enormously in the kind of resources they have for the characterization of
episodes and other complex events ... (1987, p. 351)
And further:
.. .Kalam and English do share a body of more or less isomorphic concep
tual events and situations, namely those which both languages may express
by a single clause. This common core presumably reflects certain character
istics of external world and human experience that are salient for people
everywhere. But it is a fairly small core, in relation to the total set of
conceptual situations which English can reduce to a single clause expres
sion... (1987, p. 356)
The intellectual force of Pawley's argument is firmly anchored in a belief
in the iconic relation between grammar and thought. The specific iconism
that concerns us here harken all the way back to Aristotle's conception of
the verb ('predicate') as the code-element at the core of the proposition, and
of the proposition as the code-unit for states and events. 3
Multi-verb sequences thus code multi-propositional sequences, which in turn
code multi-states or multi-events.
Note, in passing, that the opposite view that a cluster of serial verbs
code a single 'event' can be argued on, essentially, the same iconicity
grounds, by invoking other grammatical criteria to determine what is an
'event'. Such an argument may be also found in works by Bradshaw (1982),
Crowley (1987) or Foley and Olson (1985), inter alia.
86
3.
T. GIVON
EMPIRICAL ISSUES: GRAMMAR VS. BEHAVIOR
The empirical bind inherent in the general program of inferring cognition

from grammar, even after one detaches it from Aristotle's empiricism, may
be summed up as follows: One takes for granted the complete isomorphism
between the cognitive package called 'event', and the grammatical package
called 'proposition' (or 'sentence', or 'clause'). Grammatical packaging is of
course relatively easy for the linguist to define. Cognitive packaging is then
left undefined. One winds up then with an inevitable circularity: Grammar is
first used to define cognition, and then is said to correlate with it.
The study whose early results are reported here has been designed to
investigate the correlation between the grammatical and cognitive packaging
of'events', by relying on some well-known phenomena in the temporal packag
ing of linguistic information: The use of delivery rhythm (Just & Carpenter,
1980; Aronson & Ferres, 1983), intonation contours (Kumpf, 1987), and most
particularly, the placement of pauses (Eisler-Goldman, 1958a; 1958b; 1961;
1968; Chafe, 1979; 1987; Tomlin, 1987a). It is widely recognized (see discussion
in Haiman (ed.), 1985a; 1985b; Givn, 1985a; inter alia) that:
(12) The temporal-physical distance between chunks of linguisticallycoded information correlates directly to the conceptual distance
between them.
Observation (12) is indeed one of the major konicity principles underlying
syntactic structure. It may thus be reasoned that if serial-verb constructions
indeed reflect a unique strategy of cognitive segmentation of 'events', then
pauses of the type that characteristically appear in non-serializing lan
guages such as English at the boundaries of finite ('main') verbal clauses,
will also appear in serial-verb languages at serial-verb clause boundaries.
4.
METHODOLOGY
4.1. Languages
In attempting to resolve these issues empirically, a comparative, textbased distributional study was conducted on a sample of both serializing and
non-serializing languages, involving a number of speakers from each language.
For the initial study, 4 Papuan languages in Papua-New Guinea were chosen.
Two Kalam, Alamblak have been reported to have a high and a fair
87
degree of verb serialization, respectively. For two others Tairora and

Chuave no verb serialization had been previously reported. These four are
all strict SOV languages in which serial-verb constructions precede the main/
finite verb. The fifth language, Tok Pisin (Neo-Melanesian Pidgin), has been
shown to have only few albeit conspicuous serial-verb constructions.
In a sense Tok Pisin thus constitutes a base-line for the phenomenon, but
within the same cultural basin. In addition, Tok Pisin is an SVO language in
which serial-verb constructions follow the main/finite verb.
4.2. Text elicitation
4.2.1. The 'chicken story' movie
A 6.25 minute long video movie was prepared, telling the story of a man
and a woman in a rural non-Western setting. The actors were black Africans
speaking Swahili. Clothing, demeanor and activities were designed in consul
tation with two experienced New Guinea anthropologists. 4 The following is
a short synopsis of the story:
A man walks toward a tree, leans his farming implements on it and goes on
to chop wood with an axe. A woman appears and walks to him. After some
conversation she takes the wood, moves aside and collects some more wood,
then carries it all away. The man quits his chopping, collects his tools and
walks off toward a grove. The scene shifts to the woman coming around a
small shed. She unloads her wood, lights a fire, fetches water from a barrel
and sets a pot of water to boil. She disappears behind the shed and comes
back carrying a chicken. She tries to slaughter it with a knife but clumsily
bungles the job and the chicken escapes. After some perfunctory chasing, the
woman goes back to the house, brings out some bread and cheese, makes a
sandwich, wraps it up and leaves with it. The scene shifts to the man working
the field. The woman arrives and offers him the package. They sit down, the
man unwraps the food, rejects it, throws it back at the woman, then chases
her around the tree with his hoe.
The movie was designed to present many simple transitive activities, with
characteristic agents using conventional instruments, manipulating prototypi
cal patients, and moving to and from natural locations. The choice was
designed to maximize the opportunity for using the most common serial-verb
constructions.
88
T. GIVN
4.2.2. Elicitation procedures and speakers

Following loosely the methodology of Chafe (1980), the movie was
shown, using portable battery-run video equipment, to 4-8 speakers of each
of the five languages. They were instructed, in their native language, to watch
the movie once, then during the second presentation to describe orally exactly
what they were seeing. From each speaker we obtained two types of descrip
tion of what they saw: (i) During the presentation ('on-line'); and (ii) after
the presentation ('post-viewing'). The somewhat artificial nature of the 'on
line' text is of course acknowledged. However, it tends to maximize the
amount of pauses in running text, given the relatively slow pace of the action.
We reasoned that if any elicitation condition would force pauses upon serialverb constructions, this condition would. Both descriptions were taperecorded. The texts elicited from the four most fluent speakers of each
language were subjected to further analysis. Whenever possible, a genderbalanced sample was chosen.5
4.3. Transcription of texts
The collected texts (two texts per four speakers per language) were
transcribed, translated and double-glossed, by a cooperative effort of literate
native speakers and linguists fluent in the respective languages.6 In one case
(Tok Pisin) the transcription and translation was done partly by a nonnative. 7 The output transcripts for each language give the verbatim text, the
morph-by-morph gloss, and a free English translation.
4.4. Phonetic processing of tapes
The recorded tapes of all the analyzed texts were subjected to sound
wave extraction, performed at the Phonology Laboratory, Linguistics Dept.,
UC Berkeley.8 The continuous sound-wave record of the various texts was
plotted on paper (at 10 mm per second). The paper-plotted record discrimi
nates well between periods of phonation and silence.
4.5. Pause measurement
All pauses exceeding 100 milliseconds were measured and marked on the
paper-record of the sound-wave. By simultaneously comparing the transcripts,
the running tapes and the paper-record of the sound-wave, the pauses were
89
marked in the appropriate place in the transcribed texts. These pause-marked

texts constituted the input to the quantitative analysis.
5.
RESULTS OF QUANTITATIVE ANALYSIS
5.1. Preamble
In this section the results of our quantitative analysis are presented
language by language. Before presenting the tabulated results for each lan
guage, a grammatical analysis of verb serialization in the language is given.
In this paper we present in detail the results from Tok Pisin, Kalam and
Tairora.
5.2. Tok Pisin
5.2.1. Verb serialization in Tok Pisin
Tok Pisin is a strict SVO language in which serial-verb constructions
invariably follow the main/finite verb. There are only two sub-types verb
serialization in Tok Pisin, both involving grammaticalization. The first sub
type utilizes the two verbs 'come' and 'go' in deictic-directional functions, as
in:
(13) Deictic-directional serial clauses:
a. ...i-wokabaut i-go...
(INTRANSITIVE)
PRED-mOVe PRED-gO
'.. .he went away (from a reference point)...'

b. ...tromwe i-go...
throw
(TRANSITIVE)
PRED-gO
'...(he) threw (it) away...'

. ...i-wokabaut i-...
(INTRANSITIVE)
PRED-move PRED-come
'.. .she moved toward (a reference point)...'
d. .. .em karim i-...
(TRANSITIVE)
she carry PRED-come
'...she brings (it) over (toward a reference point)...'
90
T. GIVN
The second sub-type involves the use of stap ('be') as a durative aspect
and pinis ('finish', without the predicate-marker i-) as the completive aspect,
as in:
(14) Aspectual verb serialization:
a. ...em brukim i-stap...
(DURATIVE)
he break PRED-be
'... he keeps breaking (it)...'
b. ...em wokim paya pinis... (COMPLETIVE)
she make fire finish
'...she gets the fire started...'
Main/finite clauses in Tok Pisin were divided into two types: (i) Those
with zero conjunction, and (ii) those with an overt conjunction, most com
monly na, tasol or . illustrate, respectively:
(15) a. Zero conjunction:
...em kisim sospan, wokabaut i-go...
she get saucepan move
PRED-go
'...she takes a saucepan, she goes away...'
b. Overt conjunction:
.. .em putim pinis, na em kisim tamiok...
he put finish, and he get
axe
'.. .he drops it down, and he picks up a/the axe...'
The majority of finite/main clauses in our Tok Pisin texts do not differ
morphologically from the serial-verb clauses, since they are by-and-large
unmarked for tense-aspect (except for the occasional post-posed aspectualserial constructions, as in (14), above). The presence vs. absence of the
predicate-marker i- is no structural clue either, since neither in main/finite
clauses nor in serial-verb clauses is its presence fully predictable. For the
purpose of text measurements, we further divided serial-verb constructions
tentatively into three types:
(a) Simple ones, either deictic-directional or aspectual, as in (13) and (14)
above.
91
(b) Causative serial-verb constructions, as in:

(16) ...em lait nau paya i-...
she light now fire PRED-come-up
'.. .she gets the fire going...'
(c) Relative-clause constructions, as in:
(17) ... na i-go
long hap [we]
man blong-en i-stap...
and PRED-go to place [where] man GEN-her PRED-
'.. .and she goes to where her man is...'
Category (b) above, infrequent in our Tok Pisin texts, is also found in some
of the Papuan languages studied. Category (c) is rather infrequent in our
texts, and probably should not have been counted as 'serial'.
5.2.2. Quantitative results: Tok Pisin
5.2.2.1. On-line text
a.
Serial verb density

In the combined on-line text (over all 4 speakers) we counted 98 serialverb clauses of all sub-types, distributed over 534 main/finite clauses. The
ratio of serial verbs over main clauses in these combined texts is thus 98/534 =
0.183. We shall refer to this ratio as the SV-density index. (For the combined
post-view texts index was 62/371 =0.167) This is clearly a low value, less than
one serial construction per five main/finite clauses. It corroborates our deci
sion to use Tok Pisin as the base-line of the verb serialization scale.
b.
Pause distribution
In Table 1, below, we present the overall distribution of pauses associated
with the three non-serial categories:
(i)
(ii)
(iii)
Mid-clause pauses following lexical items;

Pauses following main/finite clauses with zero conjunction;
Pauses following main/finite clauses with overt conjunction.
In Table 2, below, we present the overall distribution of pauses associated

with serial-verb constructions in the on-line text:
For comparing the frequency of pauses associated with (i.e 'following')
serial-verb construction with some expected base-line, the ratio of the mid-
92
T. GIVN
Table 1 Overall distribution of pauses associated with non-serial categories; Tok Pisin,
on-line
Pause size
Inter clause
Mid-clause
0-conjunction
N
overt conjunction
N
0.1>
0.1-0.4
0.5-0.7
0.8-1.0
1.1-1.5
1.6- <
NA
98
40
16
5
3
NA
60.4
24.6
9.8
3.0
1.8
32
53
67
38
30
83
12.5
20.7
26.1
14.8
11.7
32.4
90
52
47
33
24
33
32.3
18.7
16.9
11.8
8.6
11.8
total:
162
100.0
256
100.0
278
100.0
Table 2
Speaker
Distribution of pauses preceding serial-verb constructions; Tok Pisin, on-line

Simple
Causative
Relative
Total
I
II
III
IV
23
13
18
28
3
4
1
/
/
7
1
/
26
24
20
28
total:
82
98
preceded
by pause
0
0
0
2(.3,.4)
2
clause lexical pauses (following nouns, adjectives or adverbs) was computed

as a fraction of the total number of the main/finite clauses. There are some
problems associated with this choice, and possible correction factors that may
bring it into range are discussed further below. Nonetheless, this ratio serves
as some base-line for comparison. For our on-line Tok Pisin text, this ratio
is 162/534 = 30.3%. That is, there is a 30.0% probability that a main/finite
clause will display a mid-clause lexical pause.
In Table 3, below, we plot the probability of a pause occurring with all
categories:
93
Table 3 Probability of pauses in all the categories; Tok Pisin, on-line

serial-verb
clause
2/98 = 2.0%
mid-clause
post-lexical
finite/main
clause
overt conjunction
finite/main clause
zero conjunction
188/278 = 67.6%
224/256 = 87.5%
162/534 = 30.0%
Given that the duration of the conjunctions na, tasol and is 0.2, 0.3
and 0.3 seconds on the average, respectively, one should consider that category
as actually displaying 100% pause probability. The results display vividly the
order-of-magnitude difference between serial and main/finite clauses in
Pisin. Further, the probability of a serial construction being separated by a
pause from its main clause is much lower than the probability of a normal
lexical pause occurring at mid-clause positions.
If pause probabilities are any indication of cognitive packaging, serialverb constructions in the on-line Tok Pisin texts behave as if they are either
co-lexicalized or co-grammaticalized with the main verbal clause. One way or
another, they are tightly-packaged into the main/finite clause.
Adjacency and the potential for co-lexicalization or co-grammaticalization

To assess the potential for co-lexicalization or co-grammaticalization of
serial-verb clauses, we counted the frequency of serial verbs directly following
the main/finite verb in the clause. The results, for our on-line text, are given
in Table 4, below.
Table 4
Probability of a serial verb directly following a main verb; Tok Pisin, on-line
Speaker
I:
Speaker
II:
Speaker
III:
Speaker IV:
19/24 = 79.1%
13/24 = 54.5%
14/20 = 70.0%
23/28 = 82.1%
total:
69/96= 71.8%
The significance of the absolute values here is to some extent tainted by

the following consideration: The bulk of the serial-verbs in transitive clauses
(see (13b,d), above) have the direct object intervening between them and the
main verb. And just about 100% of serial-verbs in intransitive clauses (see
(13a,c), above) directly follow the main verb. Thus:
94
T. GIVN
(18) a. ...i-wokabaut i-go...
(INTRANSITIVE, ADJACENT)
PRED-mOVe PRED-gO
'...she goes away...'

b. ...em layt nau paya i-...
she light now fire PRED-come-up
'.. .she lights the fire up...'
(TRANSITIVE,)
NON-ADJACENT)
As we shall see further below, in Kalam, an SOV language, serial-verbs

display a much higher frequency of adjacency to another verb stem thus
a higher potential for co-lexicalization or co-grammaticalization.
5.2.2.2. Post-view text
a.
Serial verb density

For the combined post-view texts (over 4 speakers), the ratio of serialverbs per main/finite clauses was: 62/371 =0.167. This falls closely within the
range of the on-line text.
b.
Pause distribution
In Table 5, below, the pause distributions for non-serial categories in the
post-view Tok Pisin texts are given.
Table 5
Overall distribution of pauses associated with non-serial categories; Tok Pisin,

post-view
Pause size
Mid-clause
Inter-clause
0-conjunction
overt conjunction
N
0.1>
0.1-0.4
0.5-0.7
0.8-1.0
1.1-1.5
1.6-<
NA
70
16
11
11
4
NA
62.5
14.3
9.8
9.8
3.5
49
44
39
20
14
3
29.0
26.0
23.0
11.8
8.3
1.8
83
36
49
13
14
2
41.1
17.8
24.2
6.4
6.9
0.9
total:
112
100.0
169
100.0
202
100.0
In Table 6, below, the distribution of pauses associated with serial-verb

constructions in the post-view text is given.
95
Table 6 Distribution of pauses preceding serial-verb constructions; Tok Pisin, post-view

Speaker
Simple
Causative
Relative
Total
I
II
III
IV
14
11
22
12
1
/
/
/
/
1
1
/
15
12
23
12
1(.2)
0
0
3(.1,.2,.4)
total:
59
62
preceded
by pause
In Table 7, below, we give the probability of pauses being associated

with the various categories in the post-view text.
Table 7 Probability of a pause associated with all categories; Tok Pisin, post-view
serial-verb
clause
4/62 = 6.4%
mid-clause
post-lexical
finite/main clause
overt conjunction
finite/main clause
zero conjunction
112/371=30.1%
119/202 = 58.9%
120/169 = 71.0%
With minor variation in the absolute values, the results closely follow those
seen for the on-line text of Tok Pisin.
.
Adjacency and potential for co-lexicalization or co-grammaticalization

In Table 8, below, we give the probabilities of a serial verb being adjacent
to (i.e. directly following) a main verb.
Table 8 Probability of a serial-verb directly following a main verb; Tok Pisin, postview
Speaker
I:
Speaker
II:
Speaker :
Speaker IV:
9/14 = 64.3%
4/12 = 33.3%
9/23 = 39.1%
10/12 = 83.3%
total:
32/61 = 52.4%
Again, the wide variation in ratios across speakers is attributable to the

percentage of transitive constructions involved (see discussion above).
96
T. GIVON
5.3. Kalam
5.3.1. Verb serialization in Kalam
Kalam is a rigid OV, clause-chaining language, typical of the bulk of
Papuan Highlands languages. Serial-verb clauses always precede the main
verb within the main/finite clause. The clause-chaining system further divides
Kalam main clauses into three major types, according to various morphosyntactic criteria:
a.
b.
'Fully finite' chain-final clauses, with tense-aspect, mood and subject

agreement marking and no switch-reference morphology
'Medium-finite' different-subject (DS or 'switch reference') medial
clauses, with tense-aspect, subject agreement and switch-reference
marking
'Least finite' same subject (SS or 'non-switching') medial clauses, with
no tense-aspect or subject agreement morphology, but with one marked
mode (irrealis), and two cataphoric temporal categories (see below).
Both types of medial clauses are further marked for their cataphoric ('anticipa
tory') temporal relation vis-a-vis the following clause: Either simultaneous or
sequential. Further, the different subject (DS) sequential verb form also marks
the anticipated person/number agreement of the subject of the following
clause. A full description of Kalam morphology may be found in Pawley
(1966). The most common categories of main clauses occurring in our texts
9
are:
(19) a. Final clauses in perfective aspect.
...bi-nak
ak spet ominal dand sand-ip...
man-your DEF spade two
carry leave-PERF/3s
'...The man carries away two spades...'
b. Final clause in durative aspect.
...mon kamb ak yupiri-sap...
wood heap DEF gather-DUR/3s
'...she's gathering the wood-pile...'
Medial simultaneous-DS (immediate-past),
(followed by medial-sequential-SS):
.. .kikaruk am-nak-nin,
nuk kimb-iy...
chicken go-IPAST/3s-siM/DS she leave-SEQ/ss
'...the chicken having gone away, she leaves...'
97
d. Medial sequential-DS (remote-past),

(followed by medial-simultaneous-immediate-past DS) :
...ny-,
dand korip owak-nin...
give-RPAST-SE/DS she carry house come-IPAST/3s-sIM/DS
'.. .(he) having given it to her, she brings it over to the house...'
e. Medial sequential-SS:
.. .nyaip nyilung ak d-iy,
konam tik-iy...
knife small DEF take-SE/ss throat cut-SE/ss
'...she picks a knife and cuts its throat...'
f.
Medial simultaneous-SS:
...kikaruk gok tangiy-ying a-sp-ay
akan...
chicken some walk-sI/ss say-PRES-3PL Q
'...(I wonder if it's) chicken walking and making noise...'
A small fraction of medial-SS clauses in our recorded narratives are irrealis

clauses, as in:
(20) ...nying mal-ning gi-sap...
water fill-IRR/ss do-PRES
'...she intends to fill (it with) water...'
Technically speaking, such irrealis-SS clauses should not have been counted
as main clauses, since they are embedded complements of a main verb, a
category we have excluded across the board (see section 1, above). However,
in quite a few cases, only a single verb appears, so that no complementation
exists, as in:
(21) ...mindak kind nuk pik-juw-ning...
later
back her hit-dislocate-iRR/ss
'...later he aims to dislocate her back...'
For the sake of uniformity, all irrealis-SS clauses were therefore counted as
medial-SS clauses.
Two non-serial clause-type was not considered main/finite independent
clauses. Both are embedded constructions, so that one does not expect them
to display the same pause probability as main clauses. The first is relative
clauses, as in:
(22) ...ki-yon
g-ak
ak d-iy...
there-up do-IPAST/3s DEF take-SE/ss
'...he took those that he had already done...'
98
T. GIVON
The second is complements of the verb 'say'. They are, characteristically,

inflected for first-person-subject agreement, technically thus marked as direct
quote, as in:
(23) ...'korip bi-ndon
am-jak-sip-yin
ang-ak...
h o u s e there-OVer gO-COMPL-PRES-I Say-IPAST/3S
'...(she) wanted to return to the house...'

(lit.: '...(she) said: "I'm returning to the house"...')
Serial verbs in Kalam have been described extensively by Pawley (1966,
1976, 1987). Structurally, we define them here (with one exception, see further
below), as verb stems that appear with neither medial nor final verbal mor
phology. They are thus most typically bare stems. One can further sub-divide
them into three types:
(a)
(b)
(3)
Lexical
Tense-aspect
Causative-lexical
Type (a) is numerically predominant in our texts. Kalam has only 95 or

so lexical verb stems. Many conventional states, processes, events or actions,
whose number far outstrip 95, thus require lexical coding by often long
verb combinations, often with no intervening elements. To illustrate this from
our texts, consider:
(24) a. ...d- d-am
mon kinyin bi-yan
ay-sap...
take-go take-go tree base there-down put-PREs/3s
'...he is carrying and putting (them) at the base of the tree...'
b. ...tuw band dand sand-ip...
axe piece carry leave-PERF/3s
'...he carries the axe away...'
...mon kamb ak
d-
-...
wood heap DEF take-come put-PERF/3s
'...he brings and puts it (in) the wood-pile...'
d. ...gok
ak
tapin timb-rik-sap...
t h o s e DEF tOO
chop-CUt-PRES/3s
'...he's cutting those (pieces) too...'

e. ...konay timb-rik tip-pang
yok-sap...
much very chop-cut chop-break throw-PREs/3s
'...he's chopping and cutting and throwing much more...'

f.
99
...mon tip-pang
kom h g-ip...
wood chop-break roll crush do-PERF/3s
'...he cuts, chops rolls and crushes the wood...'
Type (b), found only sparsely in our texts, involves the grammaticalization of verbs as the completive aspect. In our texts, this involves primarily
the verb jak 'rise', 'arrive', as well as, more infrequently, the verb d 'take'.
Thus consider:
(25) .. .bi-nen
ap-jak-sap...
there-up come-coMPL-PREs/
'...he's arriving up there...'
Jak can be still used as a lexical verb, as in:
(26) .. .kun ak kosond
gunap jak-ip...
that DEF kunai-grass some rise-PERF/3s
'...some kunai-grass seems to be growing there...'
Type (c) is somewhat aberrant in that it involves the use of more finite
medial-verb morphology. Nonetheless, it represents a clear case of stable
lexical compounding, quite parallel to the Mandarin Chinese resultative verb
compounds (Thompson, 1973). In our Kalam texts, this type occurs with all
speakers in exactly the same context: Describing the lighting of a fire or the
heating/boiling of water, in both instances using the combination of the
transitive verb (d)-angiy 'light' in the first clause, and the intransitive yin
'burn', 'boil', 'heat up' in the second. Thus consider:
(27) ...mon dangiy-ek
yin-imb
wood take-light-RPAST/sE/Ds/3s burn-PERF/3s
'...she lights the wood...'
(Lit.: '...she lights the wood and it burns...')
Other fixed collocations of this kind are found in our text only sporadically,
as in:
(28) ...wong ak yim-ek
ar-an-jap...
garden DEF plant-RPAST/3s/Ds/3s up-grow-PREs/3s
'...he is planting (his) garden...'
(Lit.: '...he's planting the garden and it grows...')
One must note, finally, that on purely semantic/functional grounds at
least some medial-sequential-SS verbs (i.e. with the suffix -iy) can be consid-
100
T. GIVON
ered 'serial'. Typically, the verb d 'take' is involved, probably the most
common serial verb in our texts. Thus, the first medial verb clause in the two
expressions below may be considered 'serial':
(29) a. ...tiy-tawel bap d-iy
-kom g-iy...
tea-towel piece take-SE/ss wrap-wrap do-SE/ss
'.. .she takes a tea-towel and wraps it in it...'
'.. .she wraps it in a tea-towel...'
b. ...timb-rik-iy
man-man g-iya-k...
chop-CUt-SE/SS like-like
do-3PL-RPAST
'...(where) they had cut and chopped wood...'

'...(where) they had chopped the wood...'
Since sometimes it is hard to make decisions about what verbal clauses should
be included in this type, and since this type is relatively infrequent in our
Kalam transcripts, it was decided to go by structural criteria here, and not
count such instances as serial (but rather as medial-SS clauses).
5.3.2. Quantitative results: Kalam
a.
Serial verb density

The ratio of serial verbs per main clause in the combined on-line Kalam
texts was: 656/711=0.911. This is over 5-times the density of serial verbs
observed in Tok Pisin, thus clearly marking Kalam as a highly-serializing
language.
b.
Pause distribution
The pause distribution for the non-serial categories in Kalam, for the
combined on-line texts, is given in Table 9, below.
The distribution of pauses associated with serial-verb constructions are

given in Table 10, below. Only two sub-types were observed: Bare-stem (col
lapsing the 'lexical' and 'aspectual' types) and causative-lexical. Pauses
appeared only with the 'lexical' category. Of the total serial-verb sample of
656, 614 (93.6%) were bare-stem, overwhelmingly 'lexical', and 16 (4.4%)
'causative'.

Table 9
Overall distribution of pauses associated with non-serial categories; Kalam,

on-line
Pause size
Mid-clause
N
0.1>
0.1-0.4
0.5-0.7
0.8-1.0
1.1-1.5
1.6-<
total:
101
NA NA
38
48.7
19.2
15
6
7.7
8
10.2
14.1
11
78
100.0
Inter-clause
Medial-SS
Medial-DS
Final
N
%
18.6
17.7
7.6
9.7
14.6
32.0
100.0
121
18
7
4
4
4
76.6
11.4
4.4
2.5
2.5
2.5
19
5
/
1
1
2
/
3.6
3.6
7.1
98
93
40
51
77
168
158
100.0
28
100.0
525
67.8
17.8
The pause distribution in the serial categories in the combined on-line

texts in Kalam is given in Table 10, below.
Table 10 Distribution of pauses following
serial-verb constructions; Kalam,
on-line
Pause size
0.1>
0.1-0.4
0.5-0.7
0.8-1.0
1.1-1.5
1.6- <
627
18
4
3
3
1
95.6
2.7
0.6
0.4
0.4
0.1
total:
656
100.0
In Table 11, below, we give the probability of pauses associated with all
categories.
As can be seen, the probability of a pause following a serial-verb clause
is the lowest. As in Tok Pisin, it is lower than the probability of mid-clause
lexical pauses. The three main-clause type scale according to their degree of
'morphological finiteness': The least-marked medial-SS clauses (predomi-
102
Table 11
T. GIVN
Probability of pauses in the various categories; Kalam, on-line
Serial
verbs
29/656 = 4.4%
Mid-clause/
lexical
Medial-SS
clauses
Medial-DS
clauses
Final
clauses
78/711 = 10.9%
37/158 = 23.4%
9/28 = 32.1%
427/525 = 81.3%
nantly sequential-SS) show the lowest pause probability; the medium-marked

medial-DS clauses show a higher pause probability; and the most marked
final clauses show a pause probability within the range observed for main
clauses in Tok Pisin (87.5% there, 81.3% for Kalam).
.

The results of this measurement for the Kalam on-line texts are given in
Table 12, below. In Kalam we compare in this case serial verbs with the two
medial-clause categories. Either of those, given their below-50% pause prob
ability, have a certain potential for directly preceding another verb stem.
Further, we considered for this comparison only the tokens of these two
categories that were not followed by a pause.
Table 12 Probability of a verb in serial and medial clauses not
being followed by another verb stem; Kalam, on-line
Serial
85/656=12.9%
No-pause-medial-SS
No-pause-medial-DS
39/132 = 29.5%
14/26 = 53.8%
As one can see, only 12.9% of Kalam serial verbs fail to be followed by
another verb stem. This probability more than doubles for no-pause medialSS verbs, and exceeds 50% for medial-DS verbs. This gradation, like that of
the pause probability, closely follows the scale of 'degree of finiteness'.
d.
Adjacency to a preceding object noun: Another grammaticalization

potential
The 'classical' serial-verb languages of West Africa and S.E. Asia (Matisoff, 1969; Givn, 1975a) show an overwhelming serial-verb type so far not
represented in our sample the case-marking type. While not predominant,
this type may be in fact present in Kalam, at least potentially. It is limited,
at least in our texts, almost exclusively to the use of two verbs d 'take'
following both accusative and instrumental objects, and dand 'carry' following
accusative objects. 10 To illustrate these, consider:
SERIAL VERBS AND THE MENTAL REALITY OF
'EVENT'
103
(30) a. ...bin-ak
spet omnal d-...
man-DEF DEF spade two
take-come
'...the man brings over two spades...'
b. .. .nyaip nyiluk ak di
timb-rk-iy...
knife small DEF take chop-cut-SE/ss
'...she sliced it with the knife...'
...tuw band ak dand sandi-p...
axe piece DEF carry leave-PERF/3s
'...he carries the axe away...'
To assess the potential for these two verbs to develop a grammaticalized casemarking usage, we calculated the probability of either direct or instrumental
objects being preceded by 'take' or 'carry'. Only non-zero full NPs were
considered. The results of this measure are given in Table 13, below.
Table 13 Probability of an accusative or instrumental object
NP being directly followed by the serial verbs 'take'
or 'carry'; Kalam, on-line
Speaker
accusative and
instrument NPs
directly followed
by serial verbs
total accusative and

instrument NPs
I
II
III
IV
21
28
22
18
42.0%
53.8%
45.9%
50.0%
50
52
48
36
100.0
100.0
100.0
100.0
total:
89
47.8%
186
100.0
The probability of the case-role of an accusative or instrumental object

being marked by a serial verb ranges around 50%. General considerations
suggest that grammaticalization requires a considerably higher frequency.11
Another indication that the grammaticalization potential for case-mark
ing is not fully realized in Kalam comes from analyzing the actual cases where
accusative or instrumental objects are, or are not, followed by 'take' or 'carry'.
Most typically, an object is followed by a serial verb when the concrete
104
T. GIVON
meaning of 'take/ carry with the hand' is still quite literal. Examples of this
kind are:
(31) a. ...spet omnal dand sand-ip...
spade two
carry leave-PERF/3s
'...he carries two spades away...'
'...he carries two spades and leaves...'
b. ...mon kamb di
yok-ip...
wood bunch take throw-PERF/3s...
'...she throws the firewood down...'
'...she takes the firewood and throws it down...'
In contrast, serial verbs typically do not follow an object when the sense of
'take/carry with the hand' is not natural, as in:
(32) a. ...nying
water
'...she's
(into
songi-sap...
pour-PRES
pouring the water
the pan, from the bucket)...'
b. .. .biy pik-ay-iyn
ong-iy...
woman his her hit-put-EXHORT/I say-SE/ss
'.. .he aims to hit his wife...'
'...nungumiy hoe ak d-iy
wong g-amb...
husband hoe DEF take-SE/ss garden do-IPAST
'...the husband was working the garden with the hoe...'
Example (32c) is particularly instructive. It appears at the end of a post-view
narrative, when the scene had just shifted back to the husband who was in
the middle of hoeing. The instrumental 'hoe' is marked by 'take', even though
only the consequence of taking 'hold' is present. But the accusative
'garden' is unmarked, since it is not under any circumstances hand-held.
Given the much higher potential of both 'take' and 'carry' in Kalam to
co-lexicalize with the following verb, (about 88% adjacency, see table 12,
above), their grammaticalization potential as case-role markers is probably
as of yet unrealized.
105

a.
Serial verb density

The ratio of serial-verb per main clause in the combined post-view texts
in Kalam was 357/321 = 1.112 per clause. This is even higher than in the on
line text (0.911), though the difference is probably not significant.
b.
Pause distribution results

The pause distribution for non-serial categories in the combined postview Kalam texts are given in Table 14, below.
Table 14
Pause size
Overall distribution of pauses associated with non-serial categories; Kalam,

post-view
Mid-clause
N
0.1 >
0.1-0.4
0.5-0.7
0.8-1.0
1.1-1.5
1.6-<
total:
NA NA
68.2
15
13.6
3
13.6
3
4.5
1
/
/
22
100.0
Inter-clause
Medial-SS
Medial-DS
Final
N
74
38
26
4
2
1
51.0
26.2
17.9
2.7
1.4
0.6
17
15
9
1
1
/
39.5
34.9
20.9
2.3
2.3
/
38
44
29
14
4
4
28.6
33.0
21.0
10.5
3.0
3.0
145
100.0
43
100.0
133
100.0
Of the total of 357 serial-verbs in the post-view text, 11 were of the 'causative'
type (3.0%). The rest were overwhelmingly of the 'lexical' type, the only type
that allowed any of the 21 observed pauses (5.8%). The pause distribution
for the serial-verb constructions (combined) is given in Table 15, below. In
Table 16, below, we give the probability of pauses associated with all cate
gories, serial as well as non-serial.
The only discernible change from the on-line results (Table 11, above) is the
considerably larger probability of pauses associated with medial clauses, both
SS and DS. This again underscores the consistently lower probability of
pauses associated with serial-verb clauses.
106
T. GIVN
Table 15 Distribution of pauses following

serial-verb constructions; Kalam,
post-view
Pause size
0.1>
0.1-0.4
0.5-0.7
0.8-1.0
1.1-1.5
1.6-<
336
15
3
2
/
1
94.14.2
0.8
0.5
/
0.2
total:
357
100.0
Table 16 Probability of pauses in the various categories; Kalam, post-view

Serial
verbs
21/357 = 5.4%
Mid-clause/
lexical
Medial-SS
clauses
22/321 = 6.8%
71/145 = 48.9%
Medial-DS
clauses
Final
clauses
26/43 = 60.4% 95/133 = 71.4%

In Table 17, below, we give the results of the adjacency measure for
serial, no-pause medial-SS and no-pause medial-DS clauses for the combined
post-view Kalam texts. The results virtually duplicate those of the on-line
text (Table 12), again marking the serial verbal clause as the most likely
candidate for co-lexicalization or co-grammaticalization.
d.
Adjacency to a preceding noun: Another grammaticalization potential

In Table 18, below, we list the results of this measure for the post-view
Kalam texts.
Table 17 Probability of a verb in a serial or medial clause not
being followed by another verb stem; Kalam, post-view
Serial
52/357=14.5%
No-pause medial-SS No-pause medial-DS

23/74 = 31.0%
13/17 = 76.4%
107
Table 18 Probability o f an accusative or instrumental object

NP being directly followed by the serial verb 'take '
or 'carry'; Kalam, post-view
Speaker
Accusative and
instrument NPs
directly followed
by serial verb
Total accusative and

instrument NPs
I
II
III
IV
12
11
15
20
52.1%
47.8%
60.0%
54.0%
23
23
25
37
100.0
100.0
100.0
100.0
Total
58
53.7%
108
100.0
The results fall closely within the roughly-50% range observed for the on
line text, above.
5.4. Tairora
5.4.1. Verb serialization in Tairora
Tairora is a clause-chaining SOV language of the Gorokan group, TransHighlands phylum. The available descriptions are confined to verbal and
nominal morphology (Vincent, 1973a, 1973b) and make no mention of verb
serialization. Still, serial-verb constructions turn out copiously in our Tairora
texts, though not at the same high density observed in Kalam. The clausechaining system of Tairora involves no rigid morphological separation
between chain-medial and chain-final clauses. Rather, main clauses are
divided as follows:
(a) Tense-marked ('finite') clauses:
with full tense-aspect marking & subject agreement; further sub-divided
into:
(i)
DS-marked. Cataphoric subject-switching clauses
(ii) Non DS-marked
108
T. GIVON
(b) Non tense-marked ('non-finite') clauses:

with only (3rd person) subject pronominal agreement, tacitly interpreted
then as cataphoric-SS marking
In the 3rd sg. pronoun/agreement predominant in our text, the same suffix
-ro, is thus used in a double function. If the clause is non-finite (b), -ro simply
marks 3rd person subject agreement, and tacitly also cataphoric-SS. This may
be seen in the main verbs in (33), (34) and (35) below. However, if the clause
is highly-finite, thus marked for tense-aspect and its own subject agreement,
-ro then marks switch subject cataphoric-DS to 3rd person singular in
the following clause (see (35) below). Three marked tense-aspects are found
in our texts: The irrealis -re appears in mostly embedded complements, as in:
(33) ...vainti vi-va
naaho kaVa vara-re-va vi-ro...
man DEM-SUBJ garden work take-IRR-3s go-3s/ss
'...the man goes to work his garden...'
Together with another complement type (following 'say' verbs) these irrealis
clauses are counted as neither main nor serial clauses. An example of the
second type is:
(34) ..."Te iha te'a-ke vata-uro" ti-ro...
I
wood cut-ASP put-PAST/I say-3s/ss
'..."I've cut and stacked wood' he said...'
The second marked tense-aspect found in our Tairora texts is the anterior
-ira/-ina. It is confined obligatorily to embedded background clauses, primar
ily relative clauses, as in:
(35) ...saavori paepae vata-ira
vi-vare-ro...
shovel knife put-ANT/3s go-take-3s/ss
'.. .he goes back to where he put the shovel and knife...'
Again, we count such clauses as neither main nor serial-verb clauses, since
they are syntactically embedded. Finally, all finite DS-clauses in our text are
marked with the past-perfective aspect -iva (3rd person singular subject
agreement form), as in:
(36) ...v-iva-ro,
vainti-vano iha mini ke-ro...
go-PAST/3s-Ds/3s man-suBJ wood there leave-3s/ss
'...she having left, the man now leaves the wood there...'
109
We divide the serial-verb constructions in Tairora into three types.

(a) Simple bare-stem
(b) Aspectual
(c) ss-marked serial verbs
Category (a) is the predominant one in our texts, with examples such as:
(37) a. ...ihai utu vare-ro...
knife pick take-3s/ss
'...she takes the knife...'
b. ...kara vi-ra
vara an-ira...
food DEM-OBJ take come-
'...the food that she had brought (him)...'
...vi-ra
vuru ami-ro...
DEM-OBJ bring give-3s/ss
'...she brought-and-gave it (to him)...'
Category (b) is infrequent, involving the use of the verb -vai 'be' as a durative
aspect, as in:
(38) ...vi-ra-ma
iha
vara-vai-ro...
there-OBJ-EMPH wood take-be-3s/ss
'...there he's taking some wood...'
Category (c), again numerically infrequent, reflects an attempt to assess a
phenomenon already observed in the study of Kalam: That some seriallooking verbs are nonetheless marked as 'semi-finite' SS-clauses. We thus
recognize a potential gradation between bare-stem and SS-marked serial
constructions, and have chosen to assess the behavior of this 'semi-finite'
category in Tairora. Bruce (1985) made somewhat similar observation con
cerning gradation of serial-verb constructions in Alamblak. An example of
this category is:
(39) a. ...iha
vare-ro maa -ini ani-ro...
wood take-3s/ss home-LOC come-3s/ss
'...she carried the wood home...'
b. ...vi-ra-ma
an-ira-nte-ro
vara ani-ro...
DEM-OBJ-EMPH come-ANT-return-3s/ss take come-3s/ss
'...there she's bringing it back...'
110
T. GIVON
In (39a), the combination 'take.. .come' ( ='carry to') is used with an SSmarked semi-finite vare- 'take'. In (39b) the frozen combination 'come-ANTreturn-' is used with that semi-finite morphology, while 'take' is used as barestem, in serial combination with 'come'.
5.4.2. Quantitative results: Tairora
a.
Serial-verb density
The ratio of serial verbs per main clause for the combined on-line texts
in Tairora was 408/596 = 0.684 per clause. This falls between the Tok Pisin
and Kalam values.
b.

In Table 19, below, we present the distribution of pauses associated with
the various non-serial categories.
Table 19
Pause
Overall distribution of pauses associated with non-serial categories; Tairora,

on-line
Mid-clause
size
0.1>
0.1-0.4
0.5-0.7
0.8-1.0
1.1-1.5
1.6<
Total
Inter-clause
Finite-DS
Finite-SS
NA
28
22
6
14
11
NA
34.5
27.1
7.4
17.3
13.6
50
11
5
4
2
6
64.1
14.1
6.4
5.1
2.5
7.7
131
46
37
47
85
172
25.3
8.9
7.1
9.1
16.4
33.2
81
100.0
78
100.0
518
100.0
The pause distribution for the three serial categories is given in Table 20,
below.
111
SERIAL VERBS A N D THE MENTAL REALITY OF 'EVENT'
Table 20
Distribution of pauses following serial-verb constructions; Tairora, on-line

Serial-verb type
Pause
size
Aspectual
Semi-finite-SV
Bare-stem-SV
0.1 >
0.1-0.4
0.5-0.7
0.8-1.0
1.1-1.5
1.6-<
51
100.0
73
100.0
279
3
98.2
1.0
total:
51
/
/
/
/
/
/
/
/
/
/
100.0
/
/
/
/
/
73
/
/
/
/
/
100.0
0.3
0.3
284
100.0
As can be seen, the 'semi-finite' category, with SS-clause morphology, is just

as devoid of pauses as the others. In Table 21, below, we give the pause
probabilities computed for all categories in the on-line text.
Table 21
Probability of pauses in all categories; Tairora, on-line
Serial clauses
Mid-clause/lexical
Finite-DS clauses
Finite-SS clauses
5/408 = 1.2%
81/596=13.6%
28/78 = 35.9%
387/518 = 74.7%
Much as in Tok Pisin and Kalam, the serial-verb category displays the lowest
pause probability, followed by mid-clause lexical pauses. The fact that finiteDS clauses show a lower probability of pauses than finite-SS clauses in Tairora
at first seems to reverse the trend seen in Kalam. However, in Tairora the
finite-SS clauses combine chain-medial and chain-final positions. Thus, the
higher pause-probability for this category probably reflects the expected
higher pause-probability of chain-final clauses, and thus may be in fact
compatible with the Kalam results.

The results of the measure of adjacency to a following verb-stem, for the
112
T. GIVON
combined Tairora on-line texts, are given in Table 22, below. The finite-SS
and finite-DS main-clause categories include only the tokens that were not
followed by a pause.
Table 22
Probability of a verb not being followed by another verb-stem; Tairora, on

line
No-pause
finite-SS
Serial-verb clauses
Aspectual
0/51 = 0.0%
SV-bare
SV-finite
73/284 = 25.7%
11/73 = 5.0%
No-pause
finite-DS
101/131 = 77.1% 42/50 = 84.0%
The results clearly separate the behavior of serial-verb clauses from that of
finite clauses. Further, they also show that the 'semi-finite' serial-verb cate
gory, with SS-marking morphology, if anything shows a higher potential for
co-lexicalization with another verb than the main ('bare-stem') serial category.

a.
Serial-verb density
The ratio of serial verbs per main clauses in the combined post-view
Tairora texts was 264/317 = 0.832.
b.

The results of this measure, for the combined post-view text in Tairora,
are given in Table 23, below. The pause-distribution for the three serial-verb
categories is given in Table 24, below. The probability of pauses in all cate
gories in the post-view texts in Tairora is given in Table 25, below. The results
closely follow the trend seen in the on-line text.

The results of the measure of adjacency to a following verb-stem, for the
various categories in the combined post-view text of Tairora, are given in
Table 26, below. The results are again rather consistent with those given
above for the on-line text in Tairora, again separating the behavior of serialverb constructions from that of main-finite clauses.
113
SERIAL VERBS A N D THE MENTAL REALITY OF 'EVENT'
Table 23 Overall distribution of pauses associated with non-serial categories; Tairora,

post-view
Pause
Mid-clause
size
0.1>
0.1-0.4
0.5-0.7
0.8-1.0
1.1-1.5
1.6<
total:
Inter-clause
Finite-DS
Finite-SS
NA
29
5
7
3
/
NA
65.9
11.3
15.9
6.8
/
55
15
13
6
2
1
59.8
16.3
14.1
6.5
2.2
1.1
109
41
35
25
11
4
48.4
18.2
15.5
11.1
4.9
1.8
44
100.0
92
100.0
225
100.0
Table 24 Overall distribution of pauses associated with serial categories; Tairora, postview
Pause
size
Serial-verb type
Aspectual
N
Bare-stem
Finite
0.1>
0.1-0.4
0.5-0.7
0.8-1.0
1.1-1.5
1.6<
17
/
/
/
/
/ -
100.0
/
/
/
/
/
39
1
1
1
/
/
92.8
2.4
2.4
2.4
/
/
194
7
2
1
/
1
86.2
3.4
0.9
0.4
/
0.4
total:
17
100.0
42
100.0
205
100.0
Table 25 Probability of pauses in all the categories; Tairora, post-view

Serial clauses
Mid-clause/lexical
Finite-DS clauses
Finite-SS clauses
14/264 = 5.3%
44/317=13.9%
37/92 = 40.2%
116/225 = 51.5%
114
T. GIVON
Table 26
Probability of a verb in a serial or medial clause not being followed by another

verb stem; Tairora, post-view
No-pause
finite-SS
Serial-verb clauses
Aspectual
0/17 = 0.0%
6.
SV-bare
SV-finite
49/205 = 23.9%
8/42=19.0%
No-pause
finite-DS
71/120 = 59.1% 45/57 = 78.9%
OVERALL COMPARISON
In this section we compare the quantitative results for Tok Pisin, Kalam
and Tairora.
6.1. Serial-verb density
Tok Pisin and Kalam represent the two extremes of our 5-language
study, with Tok Pisin presenting the lowest density of serial verbs per verbal
clause, and Kalam the highest. Tairora represents a mid-point between the
two. The results of this comparison are summarized in Table 27, below.
Table 27 Average number of serial
verbs per verbal clause:
Three-language comparison
Text type
Tok Pisin
Tairora
Kalam
on-line
0.183
0.684
0.911
post-view
0.167
0.832
1.112
6.2. Pause distribution

The three languages compared here, with serial-verb 'density' varying
from around 0.1 per main clause (Tok Pisin) to around 1.0 per main clause
(Kalam), are amazingly consistent in showing the same sharp order-of-magnitude differences between pause probabilities associated with main/finite
115
clauses and those associated with serial-verb constructions. The 3-language

comparison, for both on-line and post-view texts, is given in Table 28, below.
Table 28
Pause probability in various clause-types: Cross-language

comparison
Language/
text
Tok Pisin
on-line
post-view
Tairora
on-line
post-view
Kalam
on-line
post-view
Pause probability
Serial verb
clauses
Mid-clause
lexical
Range for various

main-clause types
2.0%
6.4%
30.0%
30.1%
67.6-87.5%
58.9-71.0%
1.2%
5.3%
13.6%
13.9%
35.9-74.7%
40.2-51.5%
4.4%
5.4%
10.9%
6.8%
23.4-32.1-81.3%
48.9-60.4-71.4%
While the absolute values vary from language to language, the scalar
relation remains amazingly consistent. Serial-verb constructions have pause
probabilities in the range of 1-6%, while finite clauses have pause probabilities
in the range of 20-80%. Further, the most frequent main-finite clause-type
in all three languages is the one displaying the highest values, at the range of
50-90% pause probability. The main-clause types with lower pause prob
ability in Tairora and Kalam in fact represent an intermediate point on the
scale of 'finiteness' or 'independence' of clauses. In Tok Pisin, the main-clause
type with the lower pause-probability is the one displaying an overt conjunc
tion which itself counts as a temporal gap. This clause-type thus displays
100% pause-probability.
Why are the pause-probabilities of serial-verb constructions consistently
lower than random mid-clause lexical pauses? One can think of two alternative
explanations. First, the difference may be due to the different base-levels
chosen for the computation. For serial-verbs, we computed the percent of
pauses out of the total sample of serial-verbs. For mid-clause lexical pauses,
we computed the percent of pauses out of the total sample of finite clauses.
If one assumes that on the average a main clause in connected discourse has
2-4 non-verbal lexical words (nouns, adjectives, adverbs), a correction factor
of 2-4 may be applied to the average values of the mid-clause lexical pause
116
T. GIVON
category. Such a correction will bring the pause probability of this category
newly expressed now as mid-clause pauses per non-verbal lexical words
roughly into line with serial-verb pause probability.
An alternative explanation may be that serial verb stems, in Kalam as
elsewhere, tend to be either co-lexicalized or grammaticalized. As such, they
have presumably become part of a larger word. Pauses associated with serial
verb stems would thus be mid-word pauses, whose probability is bound to
be much lower than that of between-word pauses.
In sum then:
(i)
(ii)
The probability of serial-verb constructions showing a pause is

much lower, by a clear order of magnitude, from the pause prob
ability associated with typical main clauses; and
The pause probability of serial-verb constructions falls within the
probability range of mid-clause pauses associated with lexical
words, or is even lower, i.e. falling within the range of the prob
ability of mid-word pauses between grammatical morphemes.
6.3. The effect of the on-line text-elicitation technique on pause distribution

One may recall that we proposed to employ the on-line elicitation tech
nique in order to maximize the potential for pauses. If one examines again
Table 28, above, our expectations m this regard seem to have been borne
out but only for one conspicuous category: main/finite clauses. In order
to foreground this aspect of the results, we re-cast the data of the pauseprobabilities (Table 28, above) in Table 29, below, using as the most represen
tative category of finite/main clauses, the most frequent and morphologi
cally most finite clause type.
The trends evident in these results may be summarized as follows:
(a) The probability of pauses associated with main/finite clauses indeed
increases when the less-than-natural text-elicitation technique of on-line
narration is used.
(b) This contrasts sharply with the virtual insensitivity of mid-clause lexical
pauses to the effect of on-line elicitation.
(c) It contrasts even more sharply with the apparent (though probably not
significant) decrease in pause probability for serial-verb constructions
under the on-line condition.
117
Table 29 The effect of the on-line text elicitation on percent of pauses in the three main
categories
Language
Pause probability
Serial-verb
clause
Mid-clause/
lexical
Most common
main clause
Language
post-view
on-line
post-view
on-line
post-view
on-line
Tok Pisin
Tairora
Kalam
6.4%
5.3%
5.4%
2.0%
1.2%
4.4%
30.0%
13.9%
6.8%
30.1%
13.0%
10.9%
71.0%
51.5%
71.4%
87.5%
74.7%
81.3%
Once again, sharp difference are observed between the temporal packaging
behavior of serial-verb constructions as compared with that of main/finite
verbal clauses.
6.4. Adjacency to another verb-stem and the potential for co-lexicalization or
co-grammaticalization
In Table 30, below, we present the cross-language comparison of this
measure, in terms of the probability of a serial or medial-clause verb being
adjacent to another verb stem.
Table 30 Probability of adjacency of verbs to other verb stem:
Cross-language comparison
Language/
text
Tok Pisin
on-line
post-view
Tairora
on-line
post-view
Kalam
on-line
post-view
Serial verb
clauses
Mid-clause
lexical
No-pause
main-clause type
71.8%
52.4%
/
/
/
/
79.5%
77.6%
32.9%
40.9%
16.0%
17.1%
87.1%
85.5%
71.5%
69.0%
46.2%
24.6%
118
T. GIVON
The results suggest that in both Kalam and Tairora, both of which have
a much higher serial-verb density than Tok Pisin, the potential is very high
for serial verb stems to either co-lexicalize with other verb stems and yield
complex lexical verbs (the most frequent option), or to co-grammaticalize (as
tense-aspect markers; a much less frequent option). The adjacency probabili
ties, and thus the potential for co-lexicalization or co-grammaticalization, are
lower for non-serial chain-medial verbs with SS (same subject) morphology,
and lowest for medial verbs with DS (different subject) morphology.
One emphasizes, as Pawley (in personal communication) does, that high
adjacency probability is only one necessary but not sufficient factor
contributing to the potential for co-lexicalization or co-grammaticalization.
The other, equally necessary, factor is the frequency of specific recurrent
combinations ('collocations'). Quantitative data on this are not yet available
for our sample. What is more, the texts elicited in this research depict such
a restricted domain of experience, that one would hesitate to consider the
frequency distribution of specific collocations in them as indicative of their
overall frequency in the speakers' verbal-cognitive behavior.
6.5. Adjacency to object or instrument nouns: Potential for

grammaticalization as case-markers
In the one language we took this measure, Kalam, we found that accusa
tive and/or instrumental objects have the probability of 47.8% (in the on-line
text) and 53.7%o (in the post-view text) of being followed by either 'take' or
'carry'.
This probability, while quite high, in our view falls considerably below
the frequency range where a potential for grammaticalizing serial verbs as
the accusative or instrumental case-markers may be viable. This contrasts
sharply with 85-87% adjacency to another verb in Kalam, or even the
7779%o adjacency to another verb in Tairora. Both of the latter probabilities
approach the range of near-categoriality that makes co-lexicalization (or cogrammaticalization) a viable potential.
Finally, since only a small portion of our serial-verb samples in either
Tairora (durative) or Kalam (completive) involve grammaticalization as tenseaspect, one may tentatively conclude that the bulk of serial-verbs in those
two languages are either co-lexicalized with other verb-stems, or at the very
least exhibit a strong potential for such co-lexicalization. This potential is
apparent from two factors measured in our study:
119
(a) The systematic lack of clause-final pause that tends to characterize finite
lexical verbs in text; and
(b) The very high frequency of serial verbs being adjacent, in text, to other
verb stems.
To the extent that such adjacent verb stems are indeed co-lexicalized, they
jointly code stereotyped chunks of experience for which the verbal lexicon
does not contain the appropriate single verbal stem.
7.
SOME TENTATIVE CONCLUSIONS
7.1. Empirical methodology

The languages we report on here differ widely in the density of verbserialization: From the 0.1-0.2 ratio of serial-verbs per main clause in Tok
Pisin, to the 0.9-1.2 ratio in Kalam. Serialization type is also rather different.
In Tok Pisin it is predominantly grammaticalized. In Kalam and Tairora it
is primarily lexical. These languages also represent a considerable typological
cleavage: Tok Pisin is an SVO language, with no clause-chaining, and postverbal serial constructions. Kalam and Tairora are OV languages, with clausechaining, and pre-verbal serial clauses. Given such diversity, our results are
rather consistent. One may conclude, tentatively, that our methodology at
the very least produces consistent results. The temporal packaging of serialverb constructions in the languages studied here contrasts sharply with the
packaging accorded main/finite verbal clauses.
Another important feature of our methodology is that it is non-circular,
relying on neither grammar nor intuitive free translation to infer cognition.
The iconic correlation between temporal packaging and cognitive packaging
is widely documented in both experimental psychology and linguistics.
To the extent that one accepts this, our results constitute grounds for
rejecting tentatively at least an extreme, perhaps straw-man, version of Pawley's (1980; 1987) implicit hypothesis. Serial-verb constructions in Kalam (as
in Tok Pisin and Tairora) display pause probabilities that fall within or below
the range characteristic of pauses associated with lexical words within the
clause. These pause probabilities diverge dramatically from those characteris
tic of inter-clausal transition. The fact that events coded in English by singleverb clauses are coded in a language like Kalam by multi-verb clauses, by
itself reflects no deep differences in event cognition.
120
T. GIVN
7.2. Grammar and culture: Typological vs. cognitive variability

Our results suggest, tentatively, that serial verb constructions do not
represent a different cognitive way of segmenting reality. The significance of
this grammatical phenomenon is thus not primarily cross-cultural or cognitive,
but rather typological. It represents the kind of typological variability one
finds in most areas of the grammar, where different languages perform
roughly-similar speech-processing tasks by slightly different though often
related structural means. The range of cross-language typological variabil
ity in grammar is far from unconstrained. But whatever the range is, it does
not necessarily imply a corresponding range of cross-culture cognitive vari
ability.
Does grammar mirror cognition? Paradoxically, yes and no, depending
on which feature of grammar one considers. The feature selected by Pawley,
that of lexical verbhood, turns out to be less cognitively significant as well as
less universal. One must note, however, that our study points out to a strong
correlation between pause-distribution and another grammatical feature
the degree of finitenes s of verbal clauses, or grammatical verbhood. Consis
tently, the more finite or morphologically prototypical a verb is, the
higher is the probability of a pause being associated with its clause. Serial
verbs thus reveal themselves, with great cross-language consistency, to be
rather non-prototypical verbs. They lack most grammatical trimmings of
verbhood; they are not coded as typical verbs, but rather as stripped-down
stems. It is thus not lexical verbhood per se, but rather grammatical verbhood,
that turns out to reflect cognition. And at least in this case, the reflection
tends toward the universal rather than the language/culture-specific.
Whether primarily a device for enriching the grammar (as in Tok Pisin),
or for enriching a limited verbal lexicon (as in Kalam), verb serialization
should be viewed within the context of the typology cross-language coding
variability, rather than the typology of cross-culture cognitive diversity. This
does not close the door on the possibility that some grammatical features
may turn out to reflect the latter type of diversity. But verb serialization is
probably not one of those.
Thus, for example, Slobin (1987) and his associates have recently reported
a series of empirical studies that pursued a similar issue. 12 They focused on
the converse clause of the Sapir-Whorf hypothesis not language as a
reflection of thought, but rather language as a constraint on thought:
.. .users of markedly different grammars are pointed by their grammars
towards different types of observations and different evaluations of exter-
121
nally similar acts of observation, and hence are not equivalent as observers
but must arrive at somewhat different views of the world... (Whorf,
1940/1956, p. 221)
Slobin's experimental results suggest that some facets of grammatical organ
ization indeed constrain the way members of different speech communities
i.e. cultures organize their thinking for speaking. What these studies suggest
is that cognitive organizationfor the purpose of verbal communication is more
language-specific, thus distinct from human-universal cognitive organization.
It is more rigidly constrained by the available grammatical categories of the
language. Within such a framework, Pawley's original intuitions may indeed
find a more natural context.
8.
POSSIBLE EXTENSIONS
8.1. Other types of verb serialization

One major type of verb serialization, the case-marking type, is almost
entirely absent in our sample of Papua-New Guinea languages. This is the
type that provoked the original controversy about the status of serial-verb
constructions in the early 1970s. The areal distribution of this type is both
wide and significant: A vast expanse of the Niger-Congo family in West
Africa (Stahlke, 1970; Hyman, 1971; Awobuluyi, 1971; 1973; Lord, 1973;
Bambose, 1974; Schachter, 1974; Givn, 1975a); a vast area of South-East
Asia, including Chinese (Li and Thompson, 1973; 1974), Mon Khmer, AustroTai and Tibeto-Burman (Matisoff, 1973; Goral, 1980) languages; a wide area
of Austronesian (Schtz, 1969; Bradshaw, 1982; Durie, 1982; Crowley, 1987;
Foley and Olson, 1985); sporadic reports in the Caribbean Creoles (Huttar,
1981). It would be useful to be able to extend our methodology to representa
tive languages of this major serial-verb type.
8.2. The clause-type continuum
As noted above, there exists in Papuan languages, a clear continuum of
clause-types, graded by degree of finiteness. In terms of temporal packaging,
serial-verb clauses, on the one hand, and prototypical main/finite clauses, on
the other, behave as two extreme points on this scale: The former as co-lexical
stems (or grammatical morphemes) within a clause; the latter as full-fledged
122
T. GIVON
independent clauses. However, chain-medial verbs exhibit pause probabilities

and adjacency probability somewhere between the two extreme poles. This
scalarity resurrects, in a better empirical context, the early suggestions made
by Ross (1972, 1973; see also Givn, 1980) concerning the continuum of
finiteness (or 'nouniness') of verbal clauses. The methodology developed in
this study has turned out to be rather sensitive to such gradations. It may be
easily extended to probe the temporal packaging behavior of other dependent
clauses:
(a)
(b)
(c)
(d)
various types of adverbial clauses

verb complements
various types of conjunction
restrictive vs. non-restrictive REL-clauses
Continuum phenomena of one type or another have been noted in most of

these clause-types (Givn, 1980a for verb complements; Givn, 1980b for
conjoined clauses). The extension of our methodology to these new areas will
yield better understanding of the nature of inter-clausal relations and the
grammatical sub-systems that code it cross-linguistically.
8.3. Gradual vs. instantaneous grammaticalization
Our pause-distribution and stem-adjacency measurements lend them
selves also to the study of another well-known continuum, that of grammati
calization. Several earlier studies of verb serialization (Hyman, 1971; Lord,
1973; Givn, 1975a) suggested that grammaticalization was a gradual process.
That is, for example, that serial-verb constructions developed gradually from
independent, finite verbal clauses, through various stages of reduced finiteness
toward eventual full grammaticalization.
But there are reasons to suggesting the exact opposite perspective: That
cognitively, grammaticalization is not a gradual process, but rather an
instantaneous one. It involves the mental act of the mind recognizing a sim
ilarity relation and thereby exploiting it, putting an erstwhile lexical item into
grammatical use in a novel context. The minute a lexical item is used in a
frame that intends it as grammatical marker, it is thereby grammaticalized.
Taking this second perspective, one must distinguish rigorously between
the two aspects of grammaticalization:
(a)
(b)
functional analogical extension (semantics, pragmatics)

structural code adjustment (phonology, morpho-syntax)
123
In diachronic change, as has been widely suggested, structural adjustment

tends to lag behind creative-elaborative functional reanalysis (Givn, 1971;
1975a; 1979; ch. 6; Lord, 1973; Heine and Reh, 1984; Heine and Claudi,
1986). The studies purporting to show the gradual nature of grammaticalization have, in all likelihood, reported on the protracted structural adjustment
at the code level, adjustments that follow sometime long after the
original developments at the functional level.
This view, of instantaneous grammaticalization at the functional level
and gradual adjustment at the structural level, is somewhat divergent from
that presented in Bybee et al. (Volume II), where it is suggested that '.. .formal
and semantic reduction proceed in parallel...' If 'in parallel' means 'at the
same gradual pace', the two positions are indeed rather different.
Until very recently, this type of argument could have only proceeded in
an empirical vacuum. The methodology employed in this study can be easily
adapted to pursue this issue in a more meaningful way. Temporal-intonational
packaging is the oldest, most subtle, most iconic and most ubiquitous element
in syntactic structure. It takes place almost automatically and is extremely
sensitive to the cognitive dimensions of information processing and chunking.
Our quantified methodology can probably detect the early, functional, onset
of grammaticalization long before its more conventionalized structural corre
lates come on line.
ACKNOWLEDGEMENTS
*
The research reported here was supported in part by a grant from the National
Endowment for the Humanities (1985-87; 1988-1990; 'Serial verbs and the
mental reality of "event"'); in part by a grant from the John Simon Guggenheim
Foundation (1986; 'The Pragmatics of Human Language'); in part by a Fullbright lectureship (1986; 'The American Indian: Past and present'); and in part
by a grant from the Deutsche Forschungsgemeinschaft (1987; 'Lectures on
Pragmatics'). The National Science Foundation has rather consistently declined
to support this research. I am also indebted to the SIL organization in Ukarumpa, Papua-New Guinea for much help and encouragement during my stay
in Papua-New Guinea.
For the original inspiration for this project, for much continuous help and
many helpful comments, I am indebted to my good friend Andy Pawley. Helpful
comments on earlier presentations were also made by the faculty and students
of the Anthropology Department, Auckland University; the Linguistics Depart
ment, Australian National University; the Institut fr Afrikanistik, Universitt
zu Kln; the Linguistics Colloquium, University of Oregon; participants in the
124
T. GIVON
Third Pacific Linguistics Conference, University of Oregon, 1987; and partici
pants in the Symposium on Grammaticalization, Eugene, Oregon, M a y 1988.
NOTES
1.
While it is easy to document the position of extreme universalists, such as Chomsky or

Katz, extreme relativists of the Whorfian mold are harder to pin down. The only one I
am aware of who makes no apology for his bold convictions is my friend Pete Becker.
The extreme relativist position I describe here may thus be somewhat of a straw-man.
2.
Pawley (in personal communication) has objected to my interpretation of his position.

The reader may wish to judge for her/himself.
3.
See The Categories and De Interpretatione, in Ackrill (tr. and ed., 1963). For Aristotle,
thoughts ('affectations of the soul') coded real-world states or events, and propositions
('words') in turn coded thoughts. Most of us nowadays, certainly including Grace and
Pawley, feel safer in allowing that propositions code mental representations of states or
events, whose exact isomorphic matching with some presumed 'reality' remains to be
determined.
4.
With special thanks to Edward and Bambi Schieffelin.
5.
Speakers that produced steadier verbal descriptions and spoke louder were preferred.
Women under those settings tended sometimes to speak too softly or produce less verbal
description. Only in one case (Alamblak) did this necessitate using 3 men and only 1
woman in the sample.
6.
Acknowledging here the generous help and hospitality of Lyle and Helen Scholz (Kalam),
Alex and Lois Vincent (Tairora), Pat and Melenda Edmiston (Alamblak) and Robin and
Ruth Thurman (Chuave). The Kalam transcripts were also inspected and corrected by
Andy Pawley, to whom special thanks are due for comments, suggestions and encour
agement.
7.
Acknowledging here the generous help of Linda Cruz Givn. By definition, the Pidgin
speakers were not native. Three of them spoke Tok Pisin since childhood (Two women
from the Sepik region; one man from Morobe province). One man (from an off-shore
island in Milne Bay province) learned the Pidgin as adult in the Highlands. All four
speakers were recorded at Ukarumpa, Eastern Highlands province. All were tri-lingual
(English, Pidgin and their respective native language).
8.
Acknowledging here the generous technical help of John Ohala and Maricela Amador.
9.
Lyle Scholz's Kalam writing system (slightly adjusted) is followed here. In this system,
the predictable ('epenthetic') vowel [i] is written as /'i/, and the high-front vowel [i] as
/iy/. The rest of the system is transparent.
10.
dand may itself be, historically, a composite of d 'take' plus *and, with the latter possibly
a verb at some earlier time. This is obviously a speculation. Pawley (in personal communi
cation) suggests that dand is not a verb at all, but rather an adverbial meaning 'in the
hand, carrying'. This may be synchronically the case. My own feeling is that diachronically
it is of verbal origin.
11.
Generative grammarians will of course insist on 100%. The study of curves of diachronic
12.
125
change, i.e. grammaticalization, suggests that somewhere between 75-80% speakers begin
to treat the lopsided frequency as a categorial phenomenon. For discussion, see Givn
(1985b).
See also Berman and Slobin (1987).
REFERENCES
Ackrill, J.L. (ed.). 1963. Aristotle's Categories and De Interpretatione. Oxford:
Clarendon Press.
Aronson, Doris and Steven Ferres. 1983. "A model for coding lexical categories during
reading." Journal of Experimental Psychology: Human Perception and Performance
9:700-25.
Awobuluyi, Oladele. 1971. '"Splitting verbs' in Yoruba." Annales de l'Universit
d'Abidjan, Srie H, Fascicule Hors Srie 1:151-64.
Awobuluyi, Oladele. 1973. "The modifying serial construction: A critique." Studies
in African Linguistics 4:87-111.
Bamgbose, Ayo. 1974. "On serial verbs and verbal status." Journal of West African
Languages 9:17-48.
Berman, R. and D. I. Slobin. 1987. "Five ways of learning how to talk about events:
Crosslinguistic studies in children narratives." Berkeley Cognitive Science Report
46.
Bradshaw, J. 1982. Word Order Change in Papua-New Guinea Austronesian Lan
guages. University of Hawaii at Manoa, Ph.D. dissertation.
Bruce, L. 1985. Serialization: The Interface of Syntax and Lexicon. MS.
Chafe, Wallace L. (ed.). 1980. The Pear Stories: Cognitive, Cultural, and Linguistic
Aspects of Narrative Production. Norwood, N.J.: Ablex [Advances in Discourse
Processes 3.].
Chafe, Wallace L. 1987. "Cognitive constraints on information flow." In Tomiin
1987b. 21-51.
Chafe, Wallace L. and Johanna Nichols (eds). 1986. Evidentiality: The Linguistic
Coding of Epistemology. Norwood, N.J.: Ablex.
Crowley, Terry. 1987. "Serial Verbs in Paamese." Studies in Language 11:35-84.
DeLancey, Scott. 1980. Deictic Categories in the Tibeto-Burman Verb. Indiana Uni
versity, Ph.D. dissertation.
DuBois, John W. 1987. Constraint and Strategy: Transitivity in the Management of
Information Flow. University of California, Los Angeles, MS.
Durie, Mark. 1982. Clause Crunching in Oceanic. MS.
Foley, William A. and Mike Olson. 1985. "Clausehood and verb serialization." In
Nichols and Woodbury 1985. 17-60.
Fujimura, Osamu (ed.). 1973. Three Dimensions of Linguistic Theory. Tokyo: TEC
Company.
Givn, T. 1971. "Historical syntax and synchronic morphology: An archaeologist's
field trip." In Papers from the Seventh Regional Meeting of the Chicago Linguistic
Society, 394-415.
Givn, T. 1975a. "Serial verbs and syntactic change: Niger-Congo." In Li 1975.
47-112.
126
T. GIVN
Givn, T. 1975b. "Focus and scope of assertion: Some Bantu evidence." Studies in
African Linguistics 6:185-205.
Givn, T. 1979. On Understanding Grammar. New York/San Francisco/London:
Academic Press.
Givn, T. 1980a. "The binding hierarchy and the typology of complements." Studies
in Language 4:333-77.
Givn, T. 1980b. Ute Reference Grammar. Ignacio, Colorado: Ute Press.
Givn, T. 1985a. "Iconicity, isomorphism and non-arbitrary coding in syntax." In
Haiman 1985a:187-219.
Givn, T. 1985b. "Function, structure and language acquisition." In Slobin
1985:1005-27.
Goral, D. 1980. Verb Concatenation in South East Asian Languages: A Crosslinguistic
Study. University of California, Berkeley, PhD dissertation.
Grace, George. 1983. "The linguistic construction of reality." Ethnolinguistic Notes
(Honolulu) 3(4).
Haiman, John (ed.). 1985a. Iconicity in Syntax. Amsterdam/Philadelphia: John Benja
mins [Typological Studies in Language 6].
Haiman, John. 1985b. Natural Syntax: Iconicity and Erosion. Cambridge: Cambridge
University Press.
Heine, Bernd and Mechthild Reh. 1984. Grammaticalization and Re analysis in African
Heine, Bernd and Ulrike Claudi. 1986. On the Rise of Grammatical Categories: Some
Examples from Maa. Berlin: Dietrich Reimer.
Huttar, George L. 1981. "Some Kwa-like features of Djuka syntax." Studies in African
Hyman, Larry M. 1971. "Consecutivization in Fe? Fe?." Journal of African Languages
10:29-43.
Just, Marcel A. and Patricia A. Carpenter. 1980. "A Theory of reading: From eye
fixation to comprehension." Psychological Review 87(4).
Kumpf, Lorraine . 1987. "The use of pitch phenomena in the structuring of stories."
In Tomiin 1987b:189-216.
Leben, William R. (ed.). 1974. Papers from the Fifth Annual Conference on African
Linguistics [Studies in African Linguistics, Suppl. 5].
Li, Charles N. and Sandra A. Thompson. 1973. "Serial verb constructions in Mandarin
Chinese: Subordination or coordination?" In You Take the High Node and I'll
Take the Low Node: Papers from the Comparative Syntax Festival 96-103. Chicago:
Chicago Linguistic Society.
Li, Charles N. and Sandra A. Thompson. 1974. "Co-verbs in Mandarin Chinese:
Verbs or prepositions?" Journal of Chinese Linguistics 2:257-78.
Li, Charles N. (ed.). 1975. Word Order and Word Order Change. Austin/London:
University of Texas Press.
Lord, Carol. 1973. "Serial verbs in transition." Studies in African Linguistics 4:269-96.
McKaughan, N. (ed). 1973. The Languages of the Eastern Family of the East New
Guinea Highlands Stock Vol. 1. Seattle: University of Washington Press.
Matisoff, James A. 1969. "Verb concatenation in Lahu: The syntax and semantics of
'simple' juxtaposition." Acta Linguistica Hafniensia 12.2: 171-206.
Matisoff, James A. 1973. A Grammar of Lahu. University of California, Berkeley: UC
Press.
127
Mithun, Marianne. 1986. "Evidential diachrony in Northern Iroquoian." In Chafe

and Nichols 1986.
Nichols, Johanna and Anthony C. Woodbury (eds.). 1985. Grammar Inside and Outside
the Clause. Cambridge: Cambridge University Press.
Pawley, Andrew. 1966. The Structure of Kalam: A Grammar of A New Guinea
Highlands Language. University of Auckland, MS.
Pawley, Andrew. 1976. On Meeting a Language That Defies Description by Ordinary
Means. University of Auckland, MS.
Pawley, Andrew. 1987. "Encoding events in Kalam and English: Different logics for
reporting experience." In Tomlin 1987b:329-60.
Ross, John Robert. 1972. "The category squish: Endstation hauptwort." In Papers
from the Eighth Regional Meeting of the Chicago Linguistic Society 316-28. Chi
cago: Chicago Linguistic Society.
Ross, John Robert. 1973. "Nouniness." In Fujimura 1973:137-257.
Schachter, Paul. 1974. "A non-transformational account of serial verbs." In Leben
1974:253-70.
Schtz, Albert J. 1969. Nguna Grammar. Honolulu: University of Hawaii Press [Oce
anic Linguistic Publications 5].
Slobin, Dan I. (ed.). 1985. The Crosslinguistic Study of Language Acquisition Vol. 2:
Theoretical Issues. Hillsdale/New Jersey/London: Lawrence Erlbaum.
Slobin, D. I. 1987. "Thinking for speaking." In Berkeley Linguistics Society, Proceed
ings of the Thirteenth Annual Meeting, J. Aske, N. Beery and H. Filip (eds), 435-44.
Stahlke, Herbert F.W. 1970. "Serial verbs." Studies in African Linguistics 1:60-99.
Talmy, Leonard. 1970. The Semantic Structure of English and Atsugewi. University
of California, Berkeley, Ph.D. dissertation.
Thompson, Sandra A. 1973. "Resultative verb compounds in mandarin Chinese: A
case for lexical rules." Language 49:361-79.
Thompson, Sandra A. and Anthony Mulac. "A quantitative perspective on the grammaticization of epistemic parentheticais in English." Volume II.
Tomiin, Russell S. 1987a. "Linguistic reflections of cognitive events." In Tomiin
1987b:455-79.
Tomiin, Russell S. (ed.). 1987b. Coherence and Grounding in Discourse: Outcome o f a
Symposium, Eugene, Oregon, June 1984. Amsterdam/Philadelphia: John Benjamins
[Typological Studies in Language 11].
Vincent, A. 1973a. "Tairora verb structure." In McKaughan 1973.
Vincent, A. 1973b. "Notes on the Tairora Noun Morphology." In McKaughan 1973.
Whorf, Benjamin Lee. 1956. Language, thought, and Reality. John B. Carroll, (ed.)
Cambridge, Mass.: M.I.T. Press.
Grammaticalization as Retextualization
Johanna Nichols and Alan Timberlake
University of California, Berkeley
1.
INTRODUCTION
The study below describes an instance of change in grammar which is

not grammaticalization in the extensional sense in which the term is commonly
used. As we read the recent literature, the term is apparently, by a wide
margin of preference, understood to refer to changes in morphemes a
lexical word becomes a grammatical morpheme, or a partially grammatical
morpheme becomes more grammatical. While we would not wish to dispute
the fact that this occurs, the character of this scenario, both synchronically
and diachronically, may be less straightforward and obvious than is usually
assumed. Synchronically, the putative end point of this change the state
of grammaticalization probably cannot be defined. Diachronically, the
received notion of grammaticalization appears to make an implicit claim that
the natural direction of change is towards ossification, towards idiomatization,
towards a kind of semiotic entropy. But this is only half of the story. Change
in grammar is a constant trade-off between, on the one hand, the innovation
of novel tokens of text that did not exist before and, on the other hand, a
constant pressure to regularize, to idiomatize, to conventionalize such inno
vations.
The change described below is the gradual extension of the use of
instrumental case (at the expense of the nominative) in Old Russian with
predicative nouns, these being nouns that predicate a property of a major
argument of the host predicate. 1 The instrumental case is otherwise used for
a variety of functions, including to express the instrument or means of an
action. Except for trivial sound changes, the morphemes involved have
remained stable. The change, then, is not so much a change in morphemes
as in syntactic patterns. Further, it is change whose source was quite likely a
syntactic idiom which was fully grammaticalized at the outset. Examination
130
JOHANNA NICHOLS AND ALAN TIMBERLAKE
of this change leads us to suggest a view of grammatical networks that is not

synchronic or diachronic in orientation, which we accordingly term
panchronic.
On a panchronic view, any grammatical network will have areas that are
relatively rigid (in which case usage is syntagmatically predictable) and areas
that are less rigid (in which case usage is not syntagmatically predictable, and
hence is associated with covert textual values). The relatively fixed areas of a
grammatical network are then exemplars with characteristic properties; exem
plars are pieces, perhaps minimal units, of text. Because a grammar based on
exemplars necessarily underspecifies usage, new tokens of text may arise by
extrapolating from given examplars. These new tokens of usage can be
conventionalized as exemplars (grammaticalized, if you will), setting the stage
for more innovations. This dual process of innovation and conventionaliza
tion we term retextualization.
2.
DESCRIPTIVE NOTIONS
To analyze predicatives, we distinguish three semantic levels. The lowest

is the predicative syntagm, consisting of three parts: the predicative itself (we
consider only nouns, among other reasons because they develop the predica
tive instrumental much earlier than adjectives); the host predicate, various
past-referring forms of the verb 'be' in the following; and the controller, the
referent of the argument of the host predicate of which the predicative noun
is predicated. At the next level, the syntagm is oriented according to the
temporal-modal coordinates of the text, a level we call aspectuality. The
highest level we will call textuality, the relationship between the current
sentence and the adjacent text.
At each level, one can distinguish three semantic values. The predicative
construction itself can identify a unique individual, describe an individual, or
report its functioning in a certain capacity. Aspectually, the predicative state
can be continuative (no end point specified), inceptive (initial end point speci
fied), or durative (both inception and cancellation specified). Textually, we
distinguish between digressive (explication or expansion of the current infor
mation of the text), transitional (sequential narrative of developments), and
culminative textuality (stating resolutions of sequential developments); other
typologies are possible. The levels are not always clearly distinguishable,
precisely because there are relations of similarity that hold across the different
levels; for example, as we will illustrate shortly, a function predicative (one
GRAMMATICALIZATION AS RETEXTUALIZATION
131
which states that an individual acts in a certain capacity for a certain interval
of time, under certain conditions) is most naturally used with durative aspectuality (which states that in fact the predicative relation began, persisted for an
interval of time, and then was cancelled); in turn a durative predicative is
naturally used with transitional or culminative textuality (Timberlake, 1989).
3.
OLD RUSSIAN PREDICATIVE INSTRUMENTAL
Before describing the earliest attested stage, we should acknowledge the

problems that arise in tracing the diachronic development of usage, problems
which have not been mitigated by sufficient nose-to-the-grindstone philology.
Given that much of the impetus for case selection is covert, textual, and
pragmatic, one can never be sure, in attempting to compare usage across
time, that the respective speakers were trying to communicate the same
message. Additionally, there is the problem that written Russian from earlier
periods was inevitably some compromise between the sacral Slavonic language
(in its East Slavic variant) and a secular, more indigenous East Slavic (eventu
ally specifically Russian) register, closer to the spoken language. One can
never be sure a priori whether a given snippet of text is more Slavonic than
Russian (or vice versa), or whether it is consistently one or the other in all
respects.2 Hence one can never be sure, in comparing texts across time, that
the texts are Russian (or Slavonic) to the same degree. Having conceded that
the task of investigating historical semantics of grammatical morphemes is in
principle impossible, we proceed to sketch what seems a plausible path of
development.
In Old Russian ( = OR) chronicles, the earliest stage of attestation, we
distinguish three contexts of usage.3 At this (rather broad) stage, the instru
mental is attested rarely: The instrumental appears only with function nouns
nouns that report functions, roles, occupations (e.g. csar 'tsar; secular
leader', vladyka '(church)leader', voevoda 'commander', r 'monk', er
nica 'nun'). 4 The instrumental never appears with either descriptive nouns
those that describe an individual by a set of properties or identificatory
nouns those that identify a unique individual. (1) illustrates a descriptive
(national) noun, and (2) a purely descriptive noun; (3) illustrates both descrip
tive and identificatory types, and (4) an identificatory predicative.*
(1)
U Jaroplka e ena gr ekini NOM b IF

'The wife of Jaropolk was IMPERF a Greek
NOM '
132
JOHANNA NICHOLS A N D ALAN TIMBERLAKE
(2)
I be nesyt bluda, privodja sobe mubsky eny, i device rastljaja.

Be IMPERF bo enoljubc NOM, j akoe i Soloman.
'And he never tired of sin, inasmuch as he took other men's wives
for himself and defiled virgins. For he was IMPERF a philanderer
NOM, like Solomon.'
(3)
... otc e b IMPERF ima l NOM LJ, b IMPERF
Dobrynja ui NOM Volodimeru

'... the father of the two was IMPERF Malko Ljubchanin NOM, thus
was IMPERF Dobrynja uncle NOM to Volodimer.'
(4)
Be bo tgda voda tekui vzdl gory Kyevbskyja, podolii ne

sdjljudie, gor;grade b IMPERF Kyev NOM
'For there was water flowing along the Kiev Mountain, and people
did not settle in the valley, but on the mountain; that city was
IMPERF K i e v .'
Descriptive and identificatory nouns occur exclusively with the imperfect of

'be' with continuative aspectuality, which reports a state that extends in both
directions from the temporal-modal reference. Further, this context usually
is textually digressive, since it explicates or expands on the current state of
information in the text. Examples (1-4) are all of this type.
Function nouns are likewise nominative in this context. Continuative
aspectuality is forced by punctual time adverbials that localize the temporalmodal reference. Thus peree 'formerly' in (5) imparts the reading 'there was
a certain time, earlier than now, around which the predicative state held'.
(5)
V lto 6864 prestavisja Ivam episkop Rostovskii, toblIMPERF

peree arximandrit NOM U svjatogo Spasa na Moskv
'In the year 6864 Bishop Ivan of Rostov died, he who was IMPERF
formerly the archimandrite NOM at the Church of the Savior in
Moscow.'
The continuative sense of imperfect 'be' can be implicit:

(6)
I rodi Adam Kaina Avelja; be IMPERF Kaim ratai NOM, a Avel

pastuxNOM
'And Adam sired Cain and Abel. Cain was IMPERF a tiller NOM of
the soil and Abel a shepherd .'
Thus, in the first context imperfect of 'be' with continuative aspectuality

the predicative noun may be either function or non-function, but the nomi
native is used exclusively.
133
The next context, maximally distinct from the context just discussed, has
two characteristics: the noun must be a function noun, and aspectuality is
durative. Accordingly, the predicative state has different values over each of
three time intervals: the state begins, persists, and then is cancelled. Examples
are (7), (8), and (9).
(7)
A bylPERF arxiepiskopom INS Feofilb do vzjatia Novagrada 6 let

'And Feofil was PERF [ = had been?] archbishop INSTR six years until
Novgorod was seized.'

(8)
A Davidb be IMPERF vladykoju INSTR 17lt prestavisja fevralja 5,

i poloim ivpritvor svjatyja Sofi podle Kliment a
'David was IMPERF the leader INSTR for seventeen years and he died
on February 5, and they laid him in the portico of Saint Sofia
alongside Clement.'
(9)
Prestavisja arxiepiskop Novgorodkij Semeon. Bystb AOR vlady

koju INSTR 5lt 3 msjaca
'Archbishop Simon of Novgorod died. He was AOR (church)leader
INSTR for five years and three months.'
To compare, in (5) above (continuative context), what is reported is that there

was an earlier time at which an individual held a certain function, while in
(7-9) (durative contexts) the total span of time over which the individual held
an office is specified.
Durative adverbs (and the instrumental of predicative nouns) occur with
all three past-tense forms of 'be', probably with somewhat different textuality:
(7), with the perfect form, reports one event as transitional to another ('up
to the taking of Novgorod'); (8), with imperfect, is transitional; (9), with
aorist, summarizes the total experience of the predicative state, and may well
be textually digressive. The durative context requires instrumental, indepen
dent of the textuality of the predicative sentence. The periphrastic pluperfect
can be considered a subcontext of the durative, in which the instrumental is
likewise required:
(10) U Jaroplka e ena grekini be, i bjae IMPERF byla PERF erniceju
INSTR, jue be IMPERF privelb otc ego Svjatoslavb, i vbda ju za
Jaropblka, krasy radi lica eja
'Jaropolk had a Greek wife, who had been IMPERF(PERF) a nun
INTR, whom his father Svjatoslav had brought and he married her
to Jaropolk, on account of the beauty of her face.'
134
A third context, aorist of 'be' in its inceptive sense, is the most interesting
because it allows variation. In this context, the nominative is usual, as
illustrated in (11), (12), and (13):
(11) Pri sm bo starci Feodosii prestavisja, bysthAOR Stef igumen
NOM, i po Stefan Ni, semu starcju i ee suju. Edinoju emu
stojaju na utrhni, i vide osbla, stojaa na igumeni meste
'With this the elder Theodosius died, and it happened that Stefan
was AOR [ = became?] abbot NOM, and after Stefan, Nikon, which
elder is still alive. Once when he was at matins, he saw an ass
standing in the place of the abbot.'
(12) Umrju e Iisusu, byst AOR sudia NOM V nego mesto ljuda
'Once Jesus died, in his place Judas was AOR [ = became?] the judge
.'
(13) Po Aleksandr carstvova Maksimijan, let 6; se pree pastyrb NOM

b IMPERF i potom voim NOM bystb AOR, potom r NOM b
IMPERF
'After Aleksander reigned Maksim, for six years; he was IMPERF

formerly pastor NOM and then. was AOR [ = became?] a warrior NOM
and then he was IMPERF tsar .'
The instrumental, however, is possible, as in (14):
(14) V lto 6567. Izjaslav i Svjatoslav i Vbsevolod vysadia stryja
svoego Sudislava knjazja is poruba, sedevalet 20 4, zavodive i
krstu, bystb AOR r INSTR
'In the year 6567, Izjaslav, Svjatoslav, and Vsevolod liberated their
uncle Prince Sudislav, who had sat for 4 and 20 years, from prison,
converting him to Christianity, and he was AOR [ = became] a monk
INSTR.'
In this context the selection of case is not syntagmatically conditioned. With

either case, the aorist means that the state came into existence at some specific
time in the past, so the aspectuality is inceptive in all the examples above.
The selection of case is correlated with different textualities. In (14) the
inception of the predicative state is presented as the culmination of the
temporally prior events. In the other examples modal sequencing (causality,
resultativity) is absent or downplayed. In (11), the inception of abbot-hood
for Stefan is an accidental fact reported as a transition between the stories of
Theodosius and Nikon, who (as monks who saw visions) are central to the
135
narrative of this passage. The sense of (11), then, is something like 'what
happened next is that Stefan became abbot' rather than 'what Stefan did next
was to become abbot'. (12), though it reports a change of state, is identificatory; it reports who became the judge rather than what happened to Judas.
(13) is perhaps the most subtle. In (13) there is a single hero, Maksim, who
has had various functions in his life. But the middle fact that at some
point he became a warrior is not a narrative fact; it is not reported as a
result of the preceding state nor as a condition for the following state. The
three states reported are merely various properties characterizing this indivi
dual at different phases of his life. Note in (13) that the inceptive aorist is
coordinated with two imperfect forms of 'be' in their continuative sense.
Thus, the instrumental in this third context signals only the most virulent
form of sequentialized narrative, which we termed culminati ve.5
In a panchronic model of grammatical networks, we need perhaps two
exemplars representing the two contexts in which case usage is syntagmatically
predictable. In (15), the notation ' x / / means both and y occur freely, and
'x
y' is to be read as 'x motivates . Exemplar (15)<a> is the context of
continuative aspectuality with nominative (function or non-function nouns);
continuative aspectuality motivates digressive textuality. Exemplar (15) is
the durative context with instrumental of function nouns; durativity motivates
transitional or culminati ve textuality.
(15) Grammatical Network (Old Russian)
<a> . . . b IMPERF Kyev NOM/gr ekini NOM/episkop NOM
'...was IMPERF Kiev NOM/a Greek woman NOM/bishop NOM
i. identificatory/descriptive/function predicative
ii. continuative aspectuality
iii. digressive textuality
 . . . b IMVERF/byst AOR/byl PF voevodoju INSTR 6 lt
'...was IMPERF/was AOR/was PERF commander INSTR six years'
i. function predicative
ii. durative aspectuality
iii. transitional/culminative textuality
<b'> ... byst AOR voevodoju INSTR
'...was AOR [ = became] a commander INSTR'
i. function predicative noun
ii. inceptive aspectuality
iii. culminative textuality
136
As noted above, the syntagm with an aorist of 'be' in its inceptive sense shows
variation in case. For this context, it may well have been that there was no
clear examplar. Instead, the values associated with case choice were derived,
as it were, on the hoof, by comparing this context to the fixed exemplars
(15)<a-b>. In particular, use of instrumental in (15)<b'>, byst AOR voevodoju
INSTR 'he became a commander INSTR', is apparently an innovative extrapola
tion on the basis of the conventionalized exemplar of (15). Speakers might
have reasoned as follows. Under durativity ((15)), both the inception and
the cancellation of the state are mentioned, or at least understood; it is
bounded at two places on a single axis. Inceptive aspectuality reports only
one of those boundaries, the initial one, and is therefore a weaker form of
aspectuality. But if inceptive aspectuality is combined with culminative textuality, then the resulting semantic constellation is analogous to durativity:
Whereas durativity is double bounding on a single axis, inception plus culmi
native textuality is bounding on two axes. This derivation is encoded in (15)
in the statement that (15)
(15)<b'>.
This reconstruction of the grammar and usage of OR relies on two
principles governing the organization of synchronic grammatical networks
and indirectly also diachrony. On the one hand, it must evidently be possible
to compare semantic operations across different levels. In the example above,
a certain value for textuality (the role of the sentence in the surrounding text)
is taken to be comparable to a certain value for the more local notion of
aspectuality. On the other hand, there is evidently a kind of conservation of
semantic operations involved here in the highly variable use of the instrumen
tal with the aorist, the context of (15)<b'>. The instrumental is justified only
if a threshhold of limitation has been reached (durative or inceptive aspectual
ity plus culminative textuality).
In our exposition of OR, we have assumed that the variable usage in
context (15)<b'> was derived actively by comparison with other, more stable
patterns. That point is difficult to prove. Prevailing analytic methods prejudice
one to the view that any distribution of morphemes (such as nominative and
instrumental here) should be completely determined; and given the rather
finite evidence, it would be possible to write a deterministic grammar of case
choice for predicatives in OR. We suspect, however, that the corpus of texts
can never be completely determined by a grammar. No grammar could ever
specify completely all the possible contextual variations that could, and do,
arise in text; if a grammar did, it would be unlimited. But even if one were
to construct a completely deterministic grammar for OR, it must be recognized
that the context with the aorist of 'be' is the only context with variation, and
137
that the variation attested in this context is dependent on the properties of

other, more fixed patterns (those of (15)<a-b>). Thus, even a deterministic
grammar will have to distinguish between primary contexts, in which usage
is relatively fixed, stable and syntagmatically predictable, and secondary
contexts, in which usage is variable, paradigmatic and, crucially, derivative
of usage in primary contexts.
To summarize at this interim point, we have suggested that grammar be
viewed as an active process that manipulates a relatively limited network of
exemplar constructions. To determine the meaning of a token of case selection
in context, a given syntactic construction is compared to available patterns.
If the match is not complete, as it is not in context of the aorist of 'be', a
meaning is assigned to the construction analogous to the meanings of fixed
exemplars.
4.
PREHISTORY OF THE PREDICATIVE INSTRUMENTAL
Let us now work backwards and forward from the OR situation. Looking
backwards in time, we note that in Old Church Slavic ( = OCS), the most
archaic attested form of Slavic,6 the predicative instrumental with 'be' is not
attested in the more archaic codices. It appears to be attested only in the
syntactically most innovative document (the Codex Suprasliensis), and then
not in the past tense, but only in the future or with participles.7 This indicates
that the predicate instrumental with 'be' was just in the process of being
innovated at the time when the earliest attestations arose and that it is of
recent (not Indo-European) origin. The most appealing of the various hypoth
eses that have been made about the origin of the predicate instrumental is
that it derives from a predicative-like structure (often called appositive) with
a semantically autonomous verb of motion, position, or location; the sense
is 'go/sit/arrive in the capacity of; qua' This construction, illustrated in (16)
and (17), is attested in OR; it occurs only with the instrumental of nouns,
never with the nominative (a related construction with adjective, coincidentally also illustrated in (17), does take the nominative):
(16) ...glagoljue: kto vzydet PRES S nami voevodoju INSTR na Xananeja
ssisja s nimi?
'...saying: who will go PRES with us as commander INSTR against
the Canaanites to fight with them?'
138
(17) V to e leto ide AOR Nzdylo Pextinyb voevodoju INSTR na Luky;

f...] ubia 40 mu, a eny i deti poimaa, a sami priidoa
AOR na Luky vse zdorovy NOM
'In that year Nezdylo Pextinych went AOR to Luky as commander
INSTR [...] they killed forty men, and took their wives and children,
while they themselves all arrived AOR at Luky healthy .'
Although the construction type ide AOR voevodoj INS 'he went AOR as com
mander INSTR' is not attested in OCS probably because the relevant idiom
is never demanded by the limited content of the canonical texts it is most
likely a Common Slavic construction which antedated the predicative instru
mental with 'be' and provided the source for it. 8 Reconstructing backwards
from OCS and OR, then, we can assume a relatively late stage of prehistoric
Slavic in which the instrumental was used (not in paradigmatic opposition to
the nominative) with a restricted set of autosemantic verbs, a usage which
provided the source for the subsequent development of the instrumental with
'be'. 9
How could the extension from the construction with autosemantic verbs
to the copular construction have taken place? There are two, or perhaps
three, points of interest. First, the host predicate. The host predicate describes
an activity and imposes an activity sense on its predicative, the sense of
'functioning in a certain capacity'. When the instrumental is extended to 'be',
the host predicate is imputed an activity sense ('be, acting as').
Second, the original construction of 'go, acting as' implies the constant
possibility that the state could have the opposite polarity. The state is
imputed to hold only over the time at which the activity reported by the
host predicate holds; the state may well not have held before that and may
not hold after that time. In OR, the predicative instrumental with 'be' is
entrenched in durative contexts. Whereas with the original idiom ide AOR
voevodoju INSTR inception and cancellation of the state are merely presump
tions, with be IMPERF voevodoju INSTR 6ltthe state is necessarily cancelled.
As the instrumental is extended from 'go' to 'be', durativity becomes
obligatory.
Third, if we assume a general pattern of incremental development, it
would be natural to hypothesize that the predicate instrumental appeared
first as an 'optional' variant associated with some covert textual value. The
instrumental may have implied that the predicative state culminated the
episode, (8) above being an example. The relationship between instrumental
139
with autosemantic verbs and extrapolated instrumental with 'be' is sketched

in (18).
(18) Grammatical Network (Prehistoric Slavic)
<a> ...be IMPERF Kyev NOM/grekini NOM/episkop NOM
'...was IMPERF Kiev NOM/a Greek woman NOM/bishop NOM'
. identificator/descriptive/function predicative

... ide AOR voevodoju INSTR
'...went AOR as commander INS'

. 'go functioning as'
iii. digressive/transitional/culminative textuality
<b'> ... b IMPEKF/byst AOR/by1 PERF voevodoju INSTR 6 lt
'...was IMPERF/was AOR/was PERF commander INSTR six years'
i. 'be' = 'function as'

iii. culminative textuality
Thus, the first extension of the predicative instrumental to 'be' in be IMPERF
voevodoju INSTR 6lt'he was IMPERF [ = acted as] commander INSTR for six
years' can be motivated on the basis of the existing pattern of ide AOR
voevodoju INS 'he went AOR as commander INSTR'. All that is required is the
imposition of an activity sense on 'be' and a tightening of the belt with respect
to aspectuality (the scenario of inception, state holding, and cancellation
becomes actual in the narrative, not merely a typical associated scenario) and
perhaps also textuality as well.
Once this textual token b IMPER voevodoju INSTR 6letis itself adopted
as an exemplar of the grammatical network, it can serve as the model for
further deviations. If (hypothetically) b IMPERF voevodoju INSTR 6letwas
first used only in sequential narrative (recall (8) above), that condition can
be subsequently eliminated (as it is in (9) above). And while in the hypothe
sized b IMPERF voevodoju INSTR letb both temporal boundaries are overtly
specified both the inception and the cancellation the condition on
temporal boundaries can subsequently be weakened and the instrumental
could then come to be used in contexts where only one of the boundaries is
in focus, namely the inception (with the aorist of 'be'). As we observed above,
however, in this (presumably relatively new) context for the predicative instru-
140
mental, the choice of case is correlated with further textual values, digression
vs. sequential narrative.
5.
MIDDLE RUSSIAN PREDICATIVE INSTRUMENTAL
Projecting forward from Old Russian, we can sketch the further develop
ment of the predicative instrumental in later Russian. 10 A stage we might
term Middle Russian can be described as minimal extensions of the OR
situation.
To judge by the infrequent examples in the late seventeenth century
descriptive prose of Kotoixin, 11 usage has changed somewhat from OR.
Descriptive nouns still occur in the nominative (only one example):
(19) I kotorye vory byli PT ljudi NOM bogatye i oni ot svoix bed otkupalis
And whatever thieves were PT rich people NOM, they could buy
their way out of their troubles.'
Function nouns
merely allowing
the instrumental
one would have
now require the instrumental in the past tense, instead of

it as an option (6xx instrumental/6xx total). Thus, in (20),
of a function noun is used digressively, where in Old Russian
expected nominative:
(20) Posle crja Ivana Vasilevia ostalisja dva syna byl PT crem
INSTR, a drugoismaterij otdelen byl
After Tsar Ivan Vasievich there remained two sons. One was PT
tsar INSTR, the other was isolated with his mother.'
(20) is not durative. It may be inchoative ('he was the one who would become
the tsar'), though that interpretation is not necessary - the sense may be
simply 'he is the one who was the tsar'. Even if (20) is taken as inchoative,
it is still textually digressive. The instrumental in (20) can be contrasted with
an identical context a little earlier (end of the fifteenth century):
(21) Korol e sestru svoju vzjat, i so dvema synmi, v Ugorskuju zemlju
Budin. Edin pri kraleve syne ivet, a drugij byl PT U Varadinskogo
biskopa NOM
'The king took his sister, along with her two sons, back to the
Hungarian land, to Buda. One lives with the son of the king, while
the other was PT a bishop NOM with Varadin.'
141
The comparison of (20) with (21) shows that, by the seventeenth century, the
instrumental had become nearly obligatory for function nouns, regardless of
aspectuality or textuali ty.
In Middle Russian, one begins to see a new distinction emerge in predica
tive nouns. Quasi-function predicative nouns nouns that vaguely hint at
activities, even if they are not culturally sanctioned roles begin to allow
the instrumental. (22) vs. (23) is close to a minimal pair:
(22) i eeli postroitsja domom kakoi prikaznoi lv, obolgu crju i mnogie
carskie krivdy uinjat, to butto bylPTposulnikNOM i zloimatel
NOM
'and if an some official should build a house, people will lie to the
tsar and will cause many lies, that, supposedly, he was PT a bribe
giver NOM or a bribe-taker .'
(23) tx vorov pytajut nakrpko, vprjam li t ljudi na kotoryx oni
govorjat s nimi v tom vorovstv tovaryami INSTR ili stanovikami
INSTR i oberegalikami INSTR byli PT i ne naprasno l govorjat
po nasertke
'and these thieves are questioned thoroughly as to whether exactly
the people they are accusing in this crime were PT comrades INSTR
or accessories INSTR or lookouts INSTR, or whether they are not
accusing them without justification, out of malice.'
(22), which reports a generalized property, has continuative aspectuality. In
contrast, (23) reports a property restricted to a particular occasion (cf. v tom
vorovstv 'in that act of thievery').
The exemplars of the grammatical network of the seventeenth century,
based on the data available, can be summarized as in (24). Here only a single
form of 'be', labelled preterite, is given, since by this time the etymological
perfect form was the universal past tense. Descriptive and identificatory nouns
still occur only in the nominative ((24)<a>); pure function nouns now occur
only in the instrumental (so that now the aspectuality and textuality are
irrelevant). A quasi-function noun allows variation in case; instrumental
motivates durativity, as stated in (24)<b'>.
(24) Grammatical Network (Seventeenth Century)
<a> ... byli PT ljudi NOM bogatye
'...were PT rich people NOM'
. non-function predicative
142
 ... l PT r INSTR

'...was PT the tsar INS '
i. function predicative
ii. continuative/inceptive/durative aspectuality
<b') ... byli
PT
oberegalikami
INSTR
'...were PT lookouts INSTR'
i. quasi-function predicative
Usage in the seventeenth century differs somewhat from that of OR:
Pure function nouns, which previously allowed variation (in contexts of
inceptive aspectuality), now virtually require the instrumental, regardless of
aspectuality or textuality. At the same time, quasi-function nouns now allow
the instrumental in the specfic context of durative aspectuality. The overall
effect has been to fix usage in one domain and develop variation in another.
We suspect that this is characteristic of change in grammatical networks:
Contexts in which usage is optional become contexts in which usage is
obligatory, and contexts in which only one alternative had been possible
become contexts of optional usage. The process of extending the instrumental
is a cyclical process of generating new tokens of text that are analogous but
not identical to preexisting exemplars followed by conventionalization of
these new tokens as exemplars themselves. In each instance, the new tokens
of text differ minimally, by the adjustment of one or two parameters, from
the preexisting patterns. In this way change is (as has long been noted) highly
incremental and close to imperceptible to the users of the language.
6.
CONCLUSIONS
Against the background of this description, let us return to general issues.

What the example of OR shows, at the very minimum, is that all changes in
grammar do not follow the scenario whereby lexical words become grammati
cal morphemes. The changes here involve morphemes that, both phonologically and in their other syntactic uses, are remarkably stable over time. The
scenario of grammaticalization in its usual extensional sense appears to
involve two things: It presupposes that the distinction between grammatical
and non-grammatical phenomena is clear; and it suggests that the basic
143
direction of development will be towards ossification, towards idiomatization,

towards entropy. Both of these assumptions we have disputed. The synchronic
situation in OR (and again in the seventeenth century, and now in the
twentieth century) is a complex blend of syntagmatic predictability and textual
value; usage at every stage is simultaneously 'grammaticalize and fluid. The
transition from one stage to another involves fixing usage in certain contexts
and opening up variation in others.
One may also discern in the panchronic model sketched above an alterna
tive to the model for change proposed some years ago by Andersen (1973).
In Andersen's model of change, change is set in motion by (speaking meta
phorically) a cataclysmic reanalysis of surface data {abduction, in Andersen's
terms); the actual, day-to-day, incremental mapping out of the change takes
place by gradually removing restrictions on the usage in order to make usage
accord with the underlying change in grammar {deduction, in Andersen's
terms).
This model embodies four claims we find dubious. First, the initial
abductive leap must contain the final goal of the change. 12 In the problem
at hand, that would be tantamount to the assumption that the first usage of
be IMPERF voevodoju INSTR was in effect a decision to implement a complete
usage of predicative instrumental in the language. This claim is difficult to
argue with, but we see no reason to assume that the complete history of the
predicate instrumental was ordained in 863 AD. Second, this model requires
that there be well-defined universal principles, accessible to every language
learner, that define what concrete rules of usage are consistent with the
underlying grammatical decision. This implication is likewise hard to evaluate;
it would be hard to state what the most ideal implementation of a distinction
in predicative case should be. We have pointed out that there are analogies
across different levels of the grammatical network. Specifically, the fact that
case may signal a distinction of narrative vs. digressive textuality (in some
subcontext) is analogous to the distinction of continuative as opposed to
durative or inceptive aspectuality. But these analogies can be derived without
recourse to a universal notion of the ideal predicative construction. Third,
Andersen's model also assumes, like all generative models, that usage can be
determined by grammar. This is likewise an assumption that makes us ner
vous. Short of specifying all sentences and contexts, a grammar can never
fully determine output. A fourth issue, likewise elusive but of considerable
interest, is whether innovations arise only through the abductive reanalysis
of previous data on the part of the addressee, as Andersen's model claims, 13
or whether it is not possible that innovations arise, by extrapolation, directly
144
in speaker's production. We suggested above, for example, that the original

idiom ide AOR voevodoju INSTR '[he] went as commander' might have given
rise directly to be IMPERF voevodoju INSTR 6lt'[he] acted as commander six
years' without a prior abduction that reinterpreted the conditions of usage
for the instrumental.
What we have done here is attempt to illustrate a dynamic, panchronic
model of grammar oriented around exemplars. Text tokens that do match
any exemplar are always possible, and even likely; if these are conventional
ized, they themselves become exemplars. In this model, the forces of innova
tion and conventionalization (idiomatization) are simultaneously always
present, at every diachronic stage. They imply each other. Rather than focus
ing attention exclusively on grammaticalization (as it seems to us the grammaticalization literature does), we have emphasized the interactive nature of
the innovative and idiomatizing forces.
ABBREVIATIONS
PT = preterite
PRES = present (including perfective 'non-past')
NOM = nominative
INSTR = instrumental
Notes
*
Uncharacteristically, the verb 'be' distinguished three simple past tenses, at least in certain
persons and numbers: e.g. 3d sg. bjae, b, and bystb. The first was clearly imperfect, the
third aorist, and the intermediate form functionally probably an imperfect (though see
van Schooneveld 1959). The identification of the fourth past-referring tense form byb
(originally a participle that demanded another form of the verb 'be' as an auxiliary) as
perfect holds only for the oldest period. This form eventually takes on the general function
of past tense in Russian. We have glossed it in OR as perfect; by the seventeenth century,
when this historical perfect has almost completely displaced the aorist in written Russian,
it seems appropriate to label the form simply preterite.
OR examples OR are cited from Polnoe sobranie russkix letopisej [ = PSRI] (directly
or with an indication of the secondary source) or axmatov, A.A. (ed.) 1916 [1969].
Provest 'vremennyx let, 1: Vvodnaja as. Tekst. Primeanija. The Hague/Paris: Mouton
[Slavic Printings and Reprintings 98] [Pov. vr. 1]. Examples from Kotoixin are cited from
the edition of Pennington 1980, with certain simplifications (no distinction between
multiple graphemes for single phonemes, and no notation of superscripted letters).
Citations:
(1): Pov. vr. l., 90
(2): Pov. vr. l, 96
145
(3): Pov. vr. /., 81

(4) Pov. vr. /., 63
(5) Letopisec Rogoskij, PSRL 15:64; Schaller, 1975:180
(6) Pov. vr. /., 109
(7) Novg. vt. letopis', PSRL 30:200; Schaller, 1975:178
(8) Erm. letopis', PSRL 23 (pril. 1):165; Schaller, 1975:178
(9) Patokova 1929:5
(10) Pov. vr. let., 90
(11) Pov. vr. /., 243
(12) Pov. vr. l., 120
(13) Letopis' Avraamki, PSRL 16:18; Schaller, 1975:181
(14) Pov. vr. l., 206
(16) Xronograf 1512, PSRL 22:92; Schaller 1975:175
(17) Tverskaja letopis', PSRL 15:290
(19) Kotoixin, f. 147r-v
(20) Kotoixin, f. 27
(21) Ja. S. Lur'e, Poves o drakule. Issledovanie i podgotovka tekstov.
(22) Kotoixin, f. 226 v
(23) Kotoixin, f. 169 r-v
1.
The general framework and modern usage are described in Nichols 1981.
2.
This reduces the value of the standard strategic move of assuming that if a given passage
is Slavonic (respectively, Russian) in some respects, it will be so in all respects, including
case selection.
3.
Our discussion relies on the investigations of Patokova (1929), Busch (1960:177-191),

and Schaller (1975:171-186), as well as our own partial examination of primary sources.
With considerable philological quesiness, we lump together chronicles that are by no
means comparable in terms of age of composition and/or age of compilation. The most
archaic portions of the chronicles were compiled by the eleventh century, but these
portions were presumably revised as additions were made; copies date from considerably
later.
4.
Busch (1960:182, fn. 36) suggests that the instrumental is restricted to newly borrowed
terms, as if their stylistic coloring were relevant. The instrumental was used with voevoda
'secular (war)lord' and vladyka '(church)leader', which are etymologically Slavic, though
their senses may nevertheless be cultural borrowings.
5.
As this description should show, it would not be possible to say unambiguously whether
the instrumental is syntagmatically conditioned or syntagmatically free; it is both, in a
complicated mix. This remark is addressed to an implicit or putative definition of 'grammaticalize as syntagmatically conditioned. Though this is only one possible way of
defining grammaticalized, we suspect that the complexity documented here would cause
discomfort for any other definition of 'grammaticalized'.
6.
To repeat, the relationship of OCS to OR written language is complex. Genetically, OCS

is some manner of early written South Slavic (perhaps with a pan-dialectal character),
and in this sense did not directly antecede Russian, which developed out of the East
Slavic dialect zone of Slavic. Yet at the same time, because of the cultural and historical
priority of OCS, the norms of this language constituted the basis for 'Old Russian' texts
from which we are obliged to take evidence about older stages of East Slavic.
146
7.
8.
9.
10.
11.
12.
13.

Bauerov, 1963; Vaillant, 1964:190.
One could go back further and derive this use from the 'instrumental of comparison',
but we are not interested in pursuing the origin further back than this construction.
As suggested by various scholars: Potebnja, 1888/1958; Fraenkel, 1925; Patokova, 1929;
Nichols, 1973(143, 209).
There is a general assumption that the situation described above was relatively stable for
an extended period of time, perhaps up to the sixteenth century, but that observation
may be artefactual it may result from the preponderance of conservative, ecclesiastical
texts until the sixteenth and seventeenth centuries, when (rather suddenly) texts written
in a more vernacular style appear.
In this text written in 1666-67, a largely non-historical description of court mores,
predicative nouns with the past tense of 'be' (a total of ten examples) come up accidentally
in the occasional historical passages.
Cf.: "this means that, before a change begins to be manifested little by little, its end result
is already given in the underlying representations" (p. 788).
Cf.: "the source of abductive innovations is to be found in distributional ambiguities in
the verbal output from which the new grammar is inferred" (p. 789).
REFERENCES
Andersen, H. 1973. 'Abductive and deductive change." Language 49:765-793.
Bauerov, M. 1963. "Bespredlonyj tvoritenyj pade v staroslavjanskom jazyke." In
Josef Kurz (ed.), 1963:287-311.
Busch, U. 1960. Die Seinstze in der russischen Sprache. [ = Slavisch-Baltisches Seminar
der Westflischen Wilhelms-Meisenheim am Glan Universitt Mnster 4.].
Fraenkel, E. 1925. "Der prdikative Instrumental im Slavischen und Baltischen und
seine syntaksischen Grundlagen." Archiv fr slawische Philologie 40:77-117.
Kurz, Jozef (ed.), Issledovanija p Sintaksisu Star oslavjanskogo Jazyka. Prague.
Nichols, J. 1973. The Balto-Slavic Predicate Instrumental: A Problem in Diachronic
Syntax. Ph.D. dissertation, University of California, Berkeley.
Nichols, J. 1981. Predicate Nominals: A Partial Surface Syntax of Russian. Berkeley,
CA. [UCPL 97].
Pennington, A. E. (ed.). 1980. Grigorij Kotoixin. O Rossii v carstvovanie Alekseja
Mixajlovia. Text and Commentary. Oxford: Clarendon.
Patokova, O. V. 1929. "K istorii razvitija tvoritenogo predikativnogo v russkom
literaturnom jazyke." Slavia 8:1-37.
Potebnja, A. A. 1888/1958. Iz zapisok po russkoj grammatike, 1-2. Kharkov.
Schaller, H. 1975. Das Prdikatsnomen im Russischen. Eine bescreibend-historische
Untersuchung. Kln/Wien: Bhlau [Slavistische Forschungen, 18].
Thelin, Nils B. (ed.). 1990. Aspect in Discourse. Amsterdam/Philadelphia: John Benja
mins [PXB NS5].
Timberlake, A. 1989. "The aspectual case of predicative nouns in Lithuanian texts."
In Nils B. Thelin (ed.) 1990.
Vaillant, A. 1964. Manuel du vieux slave 1: Grammaire. Paris,
van Schooneveld, H. 1959. A Semantic Analysis of the Old Russian Finite Preterite
System. The Hague: Mouton [Slavistic Printings and Reprintings, 7].
SECTION : DIRECTIONALITY
From Cognition to Grammar Evidence from

African Languages
Bernd Heine, Ulrike Claudi and Friederike Hnnemeyer
University of Cologne
1.
INTRODUCTION1
Grammaticalization problems have been a topic in linguistics almost

since its beginnings. Scholars like Bopp (1816), Wilhelm von Humboldt
(1825), von der Gabelentz (1891) and Meillet (1912) have laid the foundation
for a research field which has attracted a remarkable amount of scholarly
attention during the last two decades. Kurylowicz (1965:52) has provided the
by now classical definition of the term: "Grammaticalization consists in the
increase of the range of a morpheme advancing from a lexical to a grammatical
or from a less grammatical to a more grammatical status, e.g. from a derivative
formant to an inflectional one." Roughly the same definition has been used
by other scholars2 and will be adopted here as well. A number of alternative
terms like reanalysis (see 3.1), syntacticization (Givn, 1979:208ff.), semantic
bleaching (see 2.3.1), semantic weakening (Guimier, 1985:158), condensation
(Lehmann, 1982:10/11), reduction (Langacker, 1977:103-107), subduction
(Guillaume, 1964:73-86), etc. are occasionally used as synonyms or nearsynonyms, but in most cases they refer to specific aspects, like the semantic
or syntactic characteristics of grammaticalization. Grammaticalization has
also been referred to as "grammaticization" (Givn, 1975:49; Bolinger,
1978:489; Bybee and Pagliuca, 1985).3
What is common to all definitions of grammaticalization is, first, that it
is conceived as a process. Although this process can be interpreted synchronically as well (cf. Li, 1975a; Lehmann, 1982:2; Heine and Claudi, 1986) there
is wide agreement that it forms essentially a diachronic phenomenon. Second,
most scholars treat grammaticalization as a morphological process, as one
which concerns the development of a given word or morpheme. A third
150
BERND HEINE, ULRIKE CLAUDI AND FRIEDERIKE HNNEMEYER
characteristic which is implicit in these definitions and has frequently been

mentioned as an intrinsic property of the process is that grammaticalization
is unidirectional, i.e. that it leads from a "less" to a "more grammatical" unit
but not vice versa. A few counterexamples have been cited (e.g. Kahr, 1976;
Jeffers and Zwicky, 1980) but have been refuted by Lehmann (1982:16-20).
Although some examples can be adduced to show that the process may be
reversed, that is, that "degrammaticalization" may in fact take place, such
examples are less usual and will be ignored in the remainder of this paper. 4
While abundant data on grammaticalization processes have become
available during the last 15 years there are a number of problems that remain
unsolved. In the present paper we wish to look at two of these problems in
more detail, one that concerns the causes of grammaticalization and another
one that relates to its implications for language structure. It will be argued
that grammaticalization is the product of a specific type of conceptual manipu
lation and can immediately be accounted for with reference to this manipula
tion (see 2.3, 2.4).
Grammaticalization may be influenced by various factors, like contact
between languages, interference between the written and the spoken form of
a given language, the socio-cultural context, or overall typological develop
ments. 5 All these factors will remain out of consideration here since we will
be confined to the most basic structure of grammaticalization.
2.
THE COGNITIVE SETTING
2.1. An underlying principle

One of the main claims made here is that underlying grammaticalization
there is a specific cognitive principle called the "principle of the exploitation
of old means for novel functions" by Werner and Kaplan (1963:403). By
means of this principle, concrete concepts are employed to understand, explain
or describe less concrete phenomena. In this way, clearly delineated and/or
clearly structured entities are recruited to conceptualize less clearly delineated
or structured entities, non-physical experiences are understood in terms of
physical experiences, time in terms of space, cause in terms of time, or abstract
relations in terms of kinetic processes or spatial relations, etc.6
According to this view, grammaticalization can be interpreted as the
result of a process which has problem-solving as its main goal, its primary
function being conceptualization by expressing one thing in terms of another.
F R O M COGNITION TO G R A M M A R
151
This function is not confined to grammaticalization, it is the main characteris

tic of metaphor in general (cf. Lakoff and Johnson, 1980).
This view might suggest that the development of grammatical structures
is motivated, e.g. by unfulfilled communicative needs, or by the presence of
cognitive contents for which there do not exist adequate linguistic designa
tions. While this may be the case, we will not deal with this point here but
rather reserve it for a separate discussion. It may suffice to note that various
students of grammaticalization have pointed out that new grammatical devices
have developed despite the existence of old, functionally equivalent structures
(cf. Bybee, 1985), and Knig (1985:280) for instance has drawn attention to
the principle of creative language use according to which there is a constant
attempt to express the same (grammatical) meaning in other words.
2.2. Source structures
It would seem that there is a limited number of basic cognitive structures
forming the input or source of grammaticalization. We will now try to
characterize this input. In the following sections (2.3 and 2.4) we will discuss
the relation between input and output, which, as we claim, is metaphorical
in nature, and section 3 will be concerned with the output or target of
grammaticalization.
2.2.1. Concepts
Concepts which enter into grammaticalization processes in the vast
majority of cases refer to concrete objects, processes or locations, although
there are a few deictic and interrogative concepts in addition (cf. Traugott,
1982:246).
The term "source concept" is to be understood as a relative notion.
Some entity is a source concept only with reference to some other, more
"abstract", concept which may itself be the source of another, even more
"abstract", concept. For instance, a concrete object like a body part {back)
may serve as a source concept for space {three miles back) which again may
be the source for time {three years back) (see 2.3.3).
Source concepts have been described as "fundamental elements (symbolic
and deictic in function) in a typical speech situation". 7 They are of frequent
and general use (cf. Bybee and Pagliuca, 1985:72), although it is likely that
their frequent use is due to their being "fundamental elements".
Linguistically, source concepts are coded ultimately as lexemes. These
152
have much in common with what in lexicostatistics is called the basic vocabu
lary, i.e. lexemes which are less subject to replacement than others. Both
include for instance body part items like 'head', 'breast', 'back', 'belly', 'hand',
'foot', natural phenomena like 'earth' and 'sky', some human items like
'person', 'father', 'mother' and 'child', process verbs like 'come', 'give', 'take/
hold', posture verbs like 'stand', 'sit', a mental process verb like 'say', or
quantifiers like 'one' or 'many', or basic demonstratives (cf. Swadesh, 1951;
Gudschinsky, 1956). Both have in common as well that they include items
which are largely culture-independent, that is, they tend to be conceived in a
similar way across linguistic and ethnic boundaries. There are however some
remarkable differences, as we shall see below.
Source concepts may be said to refer to some of the most elementary
human experiences; they are typically derived from the physical state, behavior
or immediate environment of man. What appears to make them eligible as
such is the fact that they provide "concrete" reference points for human
orientation which evoke associations and are therefore exploited to under
stand "less concrete" concepts (see 2.3). The human body for instance offers
a convenient pool of reference points for spatial orientation. Parts of the
body are recruited for instance as source concepts for the expression of
grammatical concepts because of their relative location: 'back' or 'buttock'
for the space behind, 'breast', 'chest', 'face', 'eye', or even 'head' for the front,
'belly', 'stomach' or 'heart' for inside, 'head' for above, and 'anus' or 'foot'
for below.8 Body parts like 'liver' all belong to the basic vocabulary of
lexicostatistics but do not seem to form reference pints for spatial orientation
and, hence, do not evoke associations relevant for the expression of spatial
concepts, or any other grammatical concepts for that matter. 9
But location is not the only characteristic of body parts which is
exploited; there are some alternative associations as well. For instance, the
association between holding an object in one's hand and owning that object
has led to the development of the body part 'hand' as a marker of possession
in some West African languages (cf. Claudi and Heine, 1986; see 2.2.2 below),
and the observation that the head as the center of intellectual activity is
responsible for human behaviour might have induced the choice of 'head' as
a reference point for some more abstract concept, CAUSE/PURPOSE, which
again has triggered the grammaticalization of 'head' as an adposition and/or
complementizer of cause/finality. One should also mention that various body
parts as well as the term 'body' itself have provided the source for the
development of reflexive pronouns in many African languages (Keith Allan,
Derek Nurse, p.c.; Essien, 1982, Awolaye, 1986).
FROM COGNITION TO G R A M M A R
153
With regard to processes, source concepts refer to some of the most basic
human activities, like 'do/make', 'take/hold', 'finish' or 'say', or movements
like 'go', 'come', 'leave' or 'arrive'. Furthermore, a number of items specifying
a position or state are among the most common source concepts, typically
coded linguistically as state verbs, such as 'be/exist', 'be at', 'sit', 'stand', 'lie
(down)', 'stay/live'. Some concepts expressing desire ('want/like') or obligation
('shall', 'ought to') provide source items as well in a number of languages.
On the other hand, there are some verbs figuring in the basic vocabulary list
of lexicostatisticians which one might consider as candidates for source con
cepts but which nevertheless are not. These include 'eat', 'drink', 'hear', 'sing',
'hit', 'die' and many others.
2.2.2. Propositions
In addition to the source concepts there are some more complex struc
tures which we tentatively refer to as source propositions. These propositions
express states or processes which appear like source concepts to be
basic to human experience and can be rendered by means of linguistic predica
tions typically involving two participants. Perhaps the most common of these
predications are:
(1)
(2)
(3)
(4)
"X
"X
"X
"X
is at Y"
moves to/from/along Y"
is part of Y"
does Y"
(locational proposition)
(motion proposition)
(part-whole proposition)
(action proposition).
Source propositions describe in an elementary way where an object is, where

it moves from or to, how one object is related to others, and what one does.
It may be useful to distinguish between propositions that are static or timestable (see Givn, 1979:320-321), like (1) and (3), and those that are dynamic,
like (2) and (4), or between propositions having a spatial dimension, like (1)
and (2), and those that do not, like (3) and (4).
Each of these propositions may give rise to differing grammatical struc
tures. The locational proposition (1) is employed for instance to develop
verbal aspects or moods like progressive or intentional when the constituent
represented by Y is filled with a nominalized process verb. 10 Thus in many
languages a construction "X is at/in/on Y" has been reinterpreted as meaning
"X is doing Y". Constructions of this type, labelled PP-periphrasis in Heine
and Reh (1984:115), have developed into progressive and similar aspects in
over one hundred African languages. 11 That such constructions may also
154
give rise to grammatical categories expressing intention may be surprising,

but has been recorded for Ewe: 12 In this language, a structure "X is at Y's
place" has acquired the meaning "X is about/intends to do Y" and has been
grammaticalized to a verbal aspect referred to in grammars of this language
as "ingressive" or "intentional". 13 But the locational model has also given
rise, e.g. to expressions of verbal possession (see below). In this case, X is
understood as the possessed and Y as the possessor participant of the loca
tional relationship. Underlying this type of conceptualization there appears
to be an implication of the kind "what is at Y's place 14 belongs to Y".
While one source proposition may give rise to differing grammatical
categories, one and the same grammatical category may as well be derived
from entirely different propositions. Verbal possession ('to have, own') for
instance may not only be derived from the locational model, but also from
the action model, the latter being based on a metaphorical transfer from "Y
takes/seizes X" to "Y owns X". The former is the case for instance in Ewe,
where verbal possession ("Y has/owns X") goes back to a construction "X
is in Y's hand"), as can be seen in the following sentence:
(5)
le
as-nye
house be at hand-POSs:lsG
have a house.'
Examples for the action proposition as the source for verbal possession
can be found in a number of Eastern Cushitic languages, where a clause like
"Y has/owns X" is rendered linguistically as "X seizes Y", as can be seen in
the following sentence from Waata, an Oromo dialect spoken at the Kenyan
coast:
(6)
ani mn
k'aw a
1SG house seize
have a house.'
More research is needed on the exact role source propositions of this

kind play in the process of grammaticalization. This applies in particular to
their interrelationship with the source concepts discussed in the previous
section. What is important to note is that such propositions may, but need
not, be present in the process of grammaticalization. They are never present
for instance when a concept turns up as the dependent constituent, rather
than the head, of a construction undergoing grammaticalization (see 3.1).
FROM COGNITION TO GRAMMAR
155
2.2.3. Source structures and target structures

One of the most urgent problems in the analysis of grammaticalization
concerns the relationship between input and output, i.e. between the source
and the target within this process. This problem entails in particular the
following questions:
(1)
(2)
(3)
What source concepts and/or propositions give rise to which gram

matical concepts?
Given some grammatical category, is it possible to unambiguously
define its non-grammatical source?
To what extent are the source and target structures, as well as the
relationship holding between them, universally determined?
A number of data relevant to these questions have been presented during

the past decade and valuable generalizations attempted. But we are still far
from being able to answer these questions with a degree of certainty approach
ing prediction. It is by now well established that one source concept can give
rise to more than one grammatical category and, conversely, a given grammat
ical category may be historically derived from more than one source concept
or structure. An example from So, a Kuliak language of northeastern Uganda,
may illustrate this. The verb ac 'come' of this language has developed on the
one hand into a verbal derivative suffix -ac ("venitive") denoting movement
towards the speaker or deictic centre. On the other hand, it has become an
auxiliary and eventually a verbal proclitic denoting future tense. There is
however a second verbal proclitic denoting future, g, derived from another
motion verb, g 'go'. 1 5 Thus, we find in this example instances of both one
source giving rise to two grammatical categories and of one grammatical
category having two different lexical concepts as its source. An additional
question that arises therefore is:
(4)
Is it possible to define the factors determining the choice between

alternative sources available and, conversely, under what condition
does a source concept develop, respectively, into grammatical cate
gory A rather than B, or into rather than A, or into both?
2.3. Transfer
2.3.1. Some prerequisite notions: "Bleaching" and "abstraction"
In a number of works, grammaticalization has been described under
labels like "semantic bleaching" (Givn, 1975; Lord, 1976:183/189), "seman-
156
tic depletion" (Lehmann, 1982:127), "semantic weakening" (Guillaume,

1964:73-86; Guimier, 1985:158), "desemanticization" (Heine and Reh,
1984)16 or "generalization or weakening of semantic content" (Bybee and
Pagliuca, 1985:59/63). In these works grammaticalization is viewed from the
perspective of the source concept which carries the "full meaning" whereas
the output of the process is interpreted as an impoverished form, one that is
emptied of, or has bleached out, the semantic specificities of its source.
Usually, this view implies that the process concerned acts like a filtering
device which sifts out anything except the semantic core. In this way, complex
meanings are reduced to less complex, but more grammatical, contents.
While the "bleaching" view, as we will call it, captures one important
aspect of grammaticalization, it would seem that it ignores certain other
characteristics of this process, and of the nature of the resulting structures.
In the process of grammaticalization, the source meaning may disappear
completely (cf. Traugott, 1980:48), or it may be replaced by what appear
from a synchronic point of view to be totally unrelated meanings or
functions, like the French negative markers pas, personne, point and rien
which go back to nouns denoting, respectively, 'step', 'person', 'point' and
'thing'. Given the right context, grammaticalization may take directions which
are difficult to reconcile with the "bleaching view".
That the output of grammaticalization is "more abstract" than its input
has been pointed out in a number of discussions (see Zirmunskij, 1966:83;
Traugott, 1980:46-47; Lehmann, 1982:128),17 and we will also use the term
"abstraction" to describe the nature of grammatical concepts with reference
to their respective source.
Among the many types of uses to which the term "abstraction" has been
put there are three which are of particular interest to our discussion. One,
called generalizing abstraction, consists in reducing the number of distinguish
ing features of a concept to its most "central characteristics", or "nucleus".
This type of abstraction evokes taxonomic reasoning, and it appears to be
present for instance in biosystematic folk taxonomies of the kind described
by Berlin, Breedlove and Raven (1973; 1974) where taxa are classified into a
specific (e.g. cork-oak), generic {oak), life form {tree) and unique beginner
rank {plant), thereby being increasingly emptied of their distinguishing fea
tures. A second type, isolating abstraction, separates one particular property
or feature which is not necessarily the "core" or "nucleus characteristic" of
that concept. 18
Both generalizing and isolating abstraction appear to be present as well
when grammaticalization is analyzed in terms of bleaching: Lexemes become
157
more "abstract" by losing their semantic specificities and being increasingly

reduced to their respective core meaning (generalizing abstraction) or to one
particular part.of their meaning (isolating abstraction). Abstraction of both
types implies that its output is necessarily part of its input, i.e. that what
happens in the course of grammaticalization is that concepts are merely
reduced in their intensional, and expanded in their extensional, content.
Another type is called metaphorical abstraction (cf. Schneider, 1979). This
type appears to be more complex and, hence, is more difficult to describe. It
serves to relate "more abstract" contents with more concrete contents across
conceptual domains, where the latter form the metaphorical vehicles for the
former. Some of the typical distinctions characterizing this relation are sum
marized in Table 1.19
It is abstraction of the metaphorical, rather than any other, type which
underlies grammaticalization. Abstraction of this kind concerns the way we
understand and conceptualize the world around us. Objects which are close
to us, are clearly structured and clearly delineated, are less "abstract" than
objects which are more distant, less clearly structured and/or delineated. And
abstraction also relates to referentiality or manipulability in discourse. Objects
which refer, which are autonomous speech participants, are less "abstract"
than those that show a low degree of referentiality or manipulability (cf.
Givn, 1982; Hopper and Thompson, 1984).20
2.3.2. Metaphor
We have argued elsewhere that the process underlying grammaticalization
is metaphorically structured (Claudi and Heine, 1986; cf. Sweetser, 1982,
1987; Fleischman, 1989),21 leading from source structures as described in
section 2.2 to grammatical structures. The relevant process may be described
in terms of a few basic categories which can be arranged lineally in the
following way:22
PERSON > OBJECT > PROCESS > SPACE > TIME > QUALITY
Each of these categories can be viewed as representing a domain of conceptual
ization which is important for structuring experience. The relationship among
them is metaphorical, i.e. any of them may serve to conceptualize any other
category to its right. The above arrangement of categories may therefore be
interpreted as consisting of a number of what we have proposed to call
categorial metaphors,23 like SPACE IS AN OBJECT or TIME IS SPACE,
where the first category forms the topic and the second the vehicle within a
158
Table 1
Vehicle and topic in metaphorical abstraction
IDEATIONAL
VEHICLE
TOPIC
Clearly delineated, compact

physical (visible, tangible,
etc.)
thing-like objects
sociophysical interactions
fuzzy, diffuse
non-physical, mental
process
space
individual
absolute or quantified
magnitude
autonomous
TEXTUAL
"real world"
less discourse-based
qualities
mental processes
(Sweetser, 1982:503)
state
time, cause, manner
mass, class,
non-countable
relative magnitude
(cf. Talmy 1986:4)
relational
new
presentational
independent of context
"world of discourse"
more discourse-based,
or "speaker-based"
(Traugott 1986:540f.)
non-referential
circumstantial
participant
old
relational
context-dependent
expressive
non-expressive.
referential
central participant
INTERPERSONAL
metaphorical equation. In many languages for instance the lexeme for the
body part 'back' is used as a metaphorical vehicle to express a spatial concept,
'behind' ( = SPACE IS AN OBJECT), and the latter again serves as a vehicle
for a temporal concept, 'after' ( = TIME IS SPACE). 24
The arrangement of categories is unidirectional, it proceeds from left to
right and can be defined in terms of "metaphorical abstractness" (see above),
where a given category is "more abstract" than any other category to its left
and "less abstract" than anything to its right. This is in agreement with our
claim that grammaticalization is the result of a problem-solving strategy
according to which concepts that are more immediately accessible to human
159
experience are employed for the expression of less accessible, more abstract
concepts (see 2.1).
There are a number of conceptual domains which are difficult to locate
along the scale of basic categories presented above. One of them is predicative
possession. At the present stage of research we will assume that possessive
concepts belong to the domain of QUALITY, in particular because of the
following observations:
(1)
(2)
The first observation is negative: possession does not show any of

the salient characteristics of any of the categories to the left. For
example, it does not have thing-like, spatial, or process-like con
tours it is most appropriately understood as a non-physical,
time-stable entity, in the same way as qualities and states are.
A survey of possessive constructions in African languages (see
2.2.2) suggests that among the most prominent domains for the
expression of possession are SPACE ("Y is at X's place" > " X
owns Y") and PROCESS ("X seizes/holds Y " > " X owns Y"). In
view of the unidirectionality principle underlying grammaticalization we must assume that possession is located to the right of these
two categories, which serve as a source for possessive concepts.
And since, apart from TIME, there is only one major category to
the right of SPACE and PROCESS, possession most likely is a
member of that category, QUALITY.
The distinction of cognitive categories like the ones figuring in the

metaphorical chain presented above is reflected in various aspects of language
structure. Interrogative pronouns for instance tend to be structured lexically
in a way that largely mirrors this distinction. Thus, all African languages
known to us have separate pronouns for the categories PERSON ('who?'),
OBJECT ('what'), PROCESS ('what?'), SPACE ('where?'), TIME ('when?')
and QUALITY ('how?'). Note that OBJECT and PROCESS tend to have
identical pronominal expressions a fact that might be suggestive of a
special metaphorical relationship between these two categories, 25 which is
also reflected, e.g. in the grammaticalization of complement constructions
involving the coding of a verb as the nominal complement of a transitive
(auxiliary) verb.
Second, there appears to be some kind of correlation between these
metaphorical categories and the division of both word classes, or sub-classes,
on the one hand and constituent types on the other. Thus, the following
prototypical correlations may be established:
160
Category
Word type
Constituent type
PERSON
OBJECT
PROCESS
SPACE
TIME
QUALITY
human noun
non-human noun
verb
adverb, adposition
adverb, adposition
adjective, adverb
noun phrase
noun phrase
verb phrase
adverbial phrase
adverbial phrase
modifier.
Third, it would seem that the various hierarchies which have been iden
tified, e.g. as determinants of word or constituent order, are structured in a
way which is suggestive of a similar or even identical underlying cognitive
patterning (cf. Allan, 1987). For instance, the case hierarchy proposed by
Givn (1984:174) might be correlated with the metaphorical chain distin
guished here roughly in the following way:
Case function
agent
benefactive
dative
accusative
locative
instrument and others
Category
PERSON
OBJECT
SPACE
QUALITY
Similarly, structures such as the personal, social status and role hierar
chies (see Allan, 1987:57ff.) appear to imply a basic pattern according to
which there is a human category preceding a non-human one, which again is
followed by more abstract categories relating to non-physical, quality-like
referents.
2.3.3. Chaining
In the preceding section we have interpreted the process underlying
grammaticalization as a problem-solving strategy whereby "abstract" con
cepts are described or understood in terms of less "abstract" concepts. This
161
process is metaphorical in nature and involves a transfer in discrete steps

from one cognitive domain to another.
In the present section we will argue however that the process is gradual
and continuous rather than discrete. Once again we will use the concept
BACK to exemplify this point. Our example concerns the noun megb 'back'
from Ewe (see also 3.3).26
The development from a body part noun ('back') to a prepositional and/
or adverbial entity ('behind, back') has been alluded to above. In accordance
with the structure of categorial metaphors discussed in 2.3.2 this development
may be reconstructed for the Ewe lexeme megb as involving essentially the
following categories:
OBJECT > SPACE > TIME > QUALITY
Thus, in sentence (7) megb denotes a body part, hence a concept of the
OBJECT category, while in (8) it expresses a locative content (SPACE), either
as an adverb, as in (8a) or a postposition, as in (8b), and in (9) a temporal
content (TIME). Finally, in (10) it exhibits yet another meaning, 'mentally
retarded', which neither denotes a thing-like nor a spatio-temporal concept
but rather one of the QUALITY category:
(7)
(8)
-p
megb f
BJE
3SG-POSS back be cold
'His back is cold.'
a. le megb SPACE
3sG is house DEF behind
'He is at the back of the house.'
b. no megb SPACE
3SG stay behind
'He stays back.'
(9)
k le -megb
TIME
3SG die be 3sG-behind
'He died after him.'
(10)
tsi
megb QUALITY
3SG remain behind
'He is backward/mentally retarded.'
While this interpretation in terms of metaphorical transfers accounts for much

of the "polysemy" of megb, there remain some problems. One relates to
162
BERND HEINE, U L R I K E CLAUDI A N D F R I E D E R I K E H N N E M E Y E R
semantic ambiguity. In quite a number of sentences in which this lexeme is

used it may refer at the same time to more than one of the categories
distinguished above. In sentence (11), for instance, megb may in the same
way denote a body part ('back') and the 'back part' of an inanimate item,
and in (12) it may mean either the 'back part' of an object or the 'space
behind' that object. Sentence (13) again may have either a spatial or a
temporal meaning. Finally, sentence (14) ( = (10)) has both a temporal and a
qualitative significance.
(11) megb keke-cl
le -s
back broad-INDEF be poss-hand
(a) 'He has a broad back.'
(b) 'Its backside is broad.'
(12) dzra
-
p megb d
prepare house-DEF poss back ready
'Prepare the back wall of/the place behind the house!'
(13)
le megb n-m
3sG be behind PREP-1 SG
(a) 'He is behind me (spatially)'

(b) 'He is late ( = he could not keep pace with me).'
(14)
ts
megb
3sG remain behind
(a) 'He remained behind/is late.'
(b) 'He is mentally retarded.'
It would seem that these cases of overlapping meaning are not coinciden
tal but rather form an integral part of the development from a lexeme to a
grammatical morpheme. Thus, the categories OBJECT, SPACE, TIME and
QUALITY are not completely separated from one another. Their relationship
should be rendered graphically rather as something like (15).
(15)
Overlapping of this kind offers nothing extraordinary; it is a common feature
of grammaticalization processes and has been described in Heine and Reh
(1984:57-59) under the label split.27 The transfer of megb from a "thing
like" (OBJECT) to a spatial entity (SPACE) for instance does not necessarily
163
imply a sudden replacement of the former by the latter but rather a stage
where, at least for some time, the former coexists side by side with the latter,
the result being overlapping.
What appears superficially as a chain of discrete, though overlapping,
categories, can however equally well be interpreted as representing a contin
uum without any clear-cut internal boundaries. We noted that sentences (7)
through (10) are suggestive of the presence of distinct categories like OBJECT,
SPACE, TIME or QUALITY, yet it is equally possible to isolate conceptual
entities intermediate between these categories. The OBJECT-SPACE chain
for instance consists of at least four such entities, as the following sentences
show. In (16), megb denotes the body part 'back' (OBJECT/PERSON) 28
and in (17) it is transferred from the human (or animal) body to other physical
items meaning 'back part' (OBJECT). In (18), megb is further transferred
from the 'back part' of X to the 'place behind' X, i.e. it denotes a spatial
entity which is still conceived as an object (OBJECT/SPACE). Finally, in
(19), it refers to a purely spatial concept (SPACE). Thus, there are two
intermediate points between the nominal meaning 'back of body' and the
adverbial meaning 'behind'.
(16) = (7)
-p
megb fa
3SG-POSS back
be cold
'His back is cold.'
(17)
e k c -
p megb ny m ?
2sG see house-DEF poss back
nice DEM Q
'Do you see that nice back wall of the house?'
(18)
xc-
megb le ny
house-DEF back be nice
'The place behind the house is nice.'
(19) = (8a) le xc-a

megb
3sG be house-DEF behind
'He is at the back of the house.'
In a similar way, intermediate points can be identified between the
SPACE and the TIME categories: As sentence (20) shows, megb is first
conceptualized as a temporal object ('the time after'; OBJECT/TIME) before
it appears as a purely temporal concept (TIME), as in (21). The range of
conceptual distinctions expressed by this lexeme has now increased to seven
and can be represented graphically as (22):
164
(20)
dz le nuu- p megb
3sG leave be food-DEF poss behind
'He left after the eating time.'
(21) = (9)
k le -megb
3sG die be poss-behind
'He died after her.'
(22)
OBJECT/ > OBJECT > OBJECT/ > SPACE

PERSON
SPACE
>OBJECT/>TIME
TIME
>QUALITY
These are but a few examples suggesting that there is no discontinuity

between categories like OBJECT and SPACE. Many more intermediate enti
ties could be identified if more contexts were considered. It is hoped that
these few examples may suffice to give an impression of the continuum nature
of the process from OBJECT to SPACE, or of any other categories. The
conclusion to be drawn from such observations is that an analysis in terms
of discrete metaphorical jumps captures only one aspect of the process. There
is both discontinuity and continuity, or metaphorical transfer and gradual
extension, involved in grammaticalization. The presence of such divergent
cognitive activities may be the result of an interplay between conceptualtaxonomic behaviour on the one hand and pragmatic-textual strategies on
the other.
2.4. Context-induced reinterpretation (CIR)
The observations made in the preceding sections suggest that grammati
calization consists of both a discrete and a continuous component. We shall
now argue that the former is metaphoric in nature and largely free from
discourse pragmatic constraints, whereas the latter appears to be metonymic
and depends strongly on the linguistic and extra-linguistic context. Consider
the following sentences:
(23)
(24)
(25)
(26)
From Cologne to
From Cologne to
He was asleep all
To get to Vienna,
Vienna it is 600 miles

Vienna it is 10 hours by train
the way/all the time from Cologne to Vienna
you travel from morning to evening
165
The prepositions from and to have a locative meaning in (23) but a

temporal one in (26). We are dealing here with another instance of the
SPACE-to-TIME metaphor (cf. 2.3.2). In (24) and (25), however, the situation
is not all that clear: These prepositions can be understood either spatially or
temporally, or else as being neither clearly spatial nor temporal. The preposi
tional meaning in these two sentences appears to mark an intermediate stage,
or one out of many possible intermediate stages, in the transition from SPACE
to TIME. If we accept this position then it follows that we are dealing here
with a type of conceptual contiguity which is to be described more appropri
ately in terms of a metonymic relationship than in terms of a metaphoric
"jump" (cf. Traugott and Knig, These volumes).
What appears to be responsible for the rise of metonyms is a discourse
pragmatic manipulation of concepts whereby these are subjected to contextual
factors in utterance interpretation (Sperber and Wilson, 1986:1). We will refer
to this process as context-induced reinterpretation (CIR). This means, for
instance, that, given the right context, spatial concepts may licence temporal
implicatures. The result is, for instance, that the prepositions from and to in
(23) may receive a temporal interpretation like the one underlying (24) or
(25). Once this interpretation becomes conventionalized, the ground is pre
pared for yet another interpretation according to which a concept that may
be understood either spatially or temporally is interpreted exclusively as a
temporal entity in contexts like the one underlying (26).
This analysis has been inspired by Traugott and Knig (these volumes)
who argue that metonymy in such cases serves the "strengthening of informativeness" (see also Traugott, 1987). According to them, semantic change of
this kind contrasts with that involving metaphor in that it is "associated with
solving the problem of being informative and relevant in communication",
whereas metaphor "is correlated with solving the problem of representation."
It would seem that metaphor and metonymy form different components
of one and the same process, i.e. grammaticalization, leading from "concrete"
to "abstract", grammatical, concepts. On the one hand, this process is made
up of a scale of contiguous entities which are in a metonymic relationship to
one another. On the other hand, it contains a smaller number of more salient
and discontinuous categories, such as SPACE, TIME, or QUALITY. The
relationship between these categories, which we have discussed briefly in 2.3.2,
is metaphoric but can also be described as being the result of a number of
metonymie extensions. In spite of their differing nature, the metonymie and
the metaphoric component both co-exist in the process of grammaticalization
and have the following structure in common,
166
BERND HEINE, U L R I K E CLAUDI A N D F R I E D E R I K E H N N E M E Y E R
A ,
which suggests that in the transition from a conceptual entity A to there
is an intermediate stage (,) where the preceding and the succeeding entity
co-exist side by side.29 The presence of this intermediate stage, which has
been described in the literature on grammaticalization under labels such as
split (Heine and Reh, 1984:57), is responsible in language structure for some
kinds of both ambiguity and free variation.
This view differs slightly from that of Traugott and Knig who argue
that metaphor and metonymy correlate with shifts to different types of
grammatical function:
'Metaphor is largely correlated with shifts from meanings situated in the
external described situation to meanings situated in the internal evalua
tive, perceptual, cognitive situation, and in the textual situation. Meton
ymy is largel', correlated with shifts to meanings situated in the subjective
belief-state or attitude toward the situation, including the linguistic one"
(Traugott and Knig, This Volume).
According to the claim made here, both metaphor and metonymy are part
and parcel of one and the same process, grammaticalization, although in the
case of a particular grammatical function one of them may be more prominent
than the other. The development of concessive, causal and reference markers
discussed by Traugott and Knig (This volume) are examples suggesting that
conversational inferences lead to metonymy and may provide the main param
eter for conceptual shift, e.g. from a temporal to a causal interpretation. It
would seem, however, that even in the case of these examples metaphor is
involved. This can be demonstrated by looking at the following sentences
cited by Traugott and Knig (This volume) to exemplify a transition of the
conjunction since from a temporal marker, as in (27), via a temporal marker
having a causal implicature (28) to a purely causal marker (29).
(27) I have done quite a bit of writing since we last met
(28) Since Susan left him, John has been very miserable
(29) Since you are not coming with me, I will have to go alone
It would seem that underlying this causal inference from a temporal expression
there is a TIME-to-CAUSE metaphor whereby a sequence of events in time
is used metaphorically to refer to a sequence of events in a causal relationship.
Once the implication "what happens earlier is the cause of what happens
thereafter" becomes conventionalized the result is a shift from a metaphorical
category of TIME to one which is more "abstract", like that of CAUSE.
167
Linguistically, the result is that a complement of time turns into a complement

of reason.
No attempt is made here to justify that an assumption of the kind: "since
X happens earlier than Y, X must be the cause of Y" may give rise to a
metaphor which forms part of a more general parameter of conceptualization.
This topic will be reserved for a separate treatment. 30 Furthermore, we will
not deal here with the interrelationship between the metaphoric and the
metonymic component in the process of grammaticalization. It might turn
out that it is the former which is responsible for defining the direction of
conceptual change, but this is an issue that requires much further research.
3.
GRAMMAR
In the preceding section we were dealing with the cognitive base of

grammaticalization, most of all with the metaphorical manipulation of con
cepts. We will now be concerned with the question as to how this affects
language structure.
3.1. Reanalysis
Perhaps one of the most spectacular effects conceptual manipulation of
the type discussed here has on language can be seen in the reanalysis of
linguistic structures. The term "syntactic reanalysis" has been used for a
number of different phenomena. The most elaborate treatment of it is that
of Langacker (1977) who defines it as "change in the structure of an expression
or class of expressions that does not involve any immediate or intrinsic
modification of its surface manifestation" (1977:59). What we consider here
as reanalysis essentially falls under what Langacker (1977:79) calls "syntactic/
semantic reformulation". 31
In some other works "reanalysis" has been used as a near-synonym for
grammaticalization (cf. Lord, 1976:179), i.e. for the development from lexical
to grammatical entities. Heine and Reh (1984:95ff.) propose to separate
reanalysis from grammaticalization, essentially because of the unidirectionality principle, which is an inherent property of the latter but not necessarily
of the former. 32
The term reanalysis has also been employed to refer to what Heine and
Reh (1984:110) called constituent-internal reanalysis, i.e. to one specific form
of the more general process of reanalysis which has the effect of re-defining
168
constituent boundaries. 33 This process turns a structure like (i) into a new
structure (ii):
(i)
(ii)
(,)
A (B,C)
In this paper we will confine the discussion of reanalysis to cases where

conceptual manipulation of the type considered above causes a linguistic
structure to be treated as some alternative structure. In this way, pragmatic
constructions may be reanalyzed as syntactic structures and syntactic or
morphosyntactic structures as differing morphosyntactic structures. 34 Some
of the more common types of syntactic reanalysis have been discussed in
Heine and Reh (1984:95-111), where however no attempt was made to look
into its causes. On the basis of the observations made above we will now try
to account for reanalysis.
There is a remarkable amount of iconicity between cognitive and linguis
tic patterning. We have drawn attention to the fact that the metaphorical
categories distinguished in 2.3.2 are not only reflected in the lexical structure,
but that there is also a correspondence between these categories and word
classes (cf. Claudi and Heine, 1986:301).
Now, once a transfer from one category to another takes place this is
likely to affect the status of the word type used to express that concept as
well. For instance, when an entity of the OBJECT category serves to conceptu
alize an entity of the category SPACE or TIME then this is likely to trigger
a linguistic development from a nominal to an adverbial word, either an
adverb or an adposition. Thus, the change from a noun 'back' to a postposi
tion 'behind' might be viewed as the result of an activity which aims at
restoring a one-to-one relationship between cognitive and linguistic structure.
This process is paralleled by reanalysis, whereby a given noun phrase, e.g.
'back of the mountain', is reanalyzed as a prepositional phrase, 'behind the
mountain'.
Typically, reanalysis accompanies grammaticalization, i.e. when a given
morpheme is grammaticalized not only its own pragmatic or syntactic position
is affected, but also that of the constituent it belongs to or, conversely, when
reanalysis takes place this is likely to involve the grammaticalization of at
least one morpheme within the structure undergoing reanalysis. In such cases,
both grammaticalization and reanalysis are the result of one and the same
strategy, namely the one which aims at expressing more "abstract" concepts
in terms of less "abstract" ones. On the other hand there is reason to assume
169
that while both appear like inseparable twins they have, nevertheless, to be
kept strictly apart, especially because of the following considerations:
(1) Whereas grammaticalization is essentially a unidirectional process,
reanalysis is not, as is shown in Heine and Reh (1984:95ff).
(2) There exist cases of both grammaticalization without reanalysis and
reanalysis without grammaticalization. The former occurs, for instance, fre
quently when the grammaticalized unit is the dependent unit of the constituent
affected by grammaticalization (see 3.2). Thus, when a demonstrative turns
into a definite article {this man>the man), the definite article into a nongeneric article, etc. (see Greenberg, 1978), or the numeral 'one' into an
indefinite article {one man>a man; Givn, 1981) then we are dealing with the
grammaticalization of a demonstrative or numeral, but no reanalyis is
involved: The syntactic status of the determiner head phrase remains
unchanged. The latter is the case, for instance, when two coordinate clauses
are reanalyzed as a new structure: Main clause subordinate clause. This
process may imply that the morphology used for linking the two clauses is
grammaticalized from a coordinating to a subordinating device. However,
when there is no linking morphology involved then we are dealing with a
case of reanalysis without grammaticalization. This would be the case for
instance in a sentence like She went to bed, she was tired, consisting on the
surface of a sequence of two main clauses. The interpretation this sentence
is likely to receive is that the second clause is reanalyzed as a subordinate
clause, a causal complement of the "first clause. In cases like these, we are
dealing with reanalysis without grammaticalization since there is no morpho
logical material to be grammaticalized (see below).
While one may argue that we are dealing with an instance of grammati
calization when a main clause is reanalyzed as a subordinate clause (Traugott,
p.c.), there are also cases of opposite developments where subordinate clauses
are reanalyzed as main clauses, and where the possibility that such develop
ments constitute instances of grammaticalization can be ruled out. We may
exemplify this from Teso, which 35 has a morphological sex gender system
and a VSO syntax. In the following sentence, however, it exhibits an SVO
word order: 36
(30) mam petero e-koto
ekirjok
not Peter 3sG-want dog
'Peter does not want a dog.'
This sentence is historically derived from the complex sentence *e-mam
petero e-koto ekinok 'It is not Peter (who) wants a dog', consisting of a main
170
clause (e-mam petero) and a subordinate clause (e-koto ekirjok). The SVO
order of this sentence is due to the fact that the main verb -mam 'not to be'
was grammaticalized to a negation marker. This single instance of grammati
calization was responsible for a number of instances of reanalysis, like the
following:
a.
b.
d.
e.
The complex sentence was reanalyzed as a simple sentence.

The subordinate clause was reanalyzed as the main clause.
The subject of the erstwhile main clause was reanalyzed as the
subject of the new sentence.
Due to the grammaticalization of the verb -mam 'not to be', the
main clause was reanalyzed as a grammatical marker.
The former VSO structure was reanalyzed as SVO, with the effect
that Teso has introduced an SVO word order in negative clauses.
Examples like this, where one instance of grammaticalization triggers a

series of reanalysis, are not difficult to come by. The way in which the various
manifestations of reanalysis in such examples are interrelated would seem to
require a separate treatment.
3.2. Dependency
Throughout this paper a distinction between a cognitive and a linguistic
domain is maintained, based on the assumption that certain linguistic struc
tures require an analysis that transcends the limits of linguistic methodology.
In doing so, we have ignored the question as to where the boundary between
the two domains is to be located. In the present section we wish to look at
one issue that has been mentioned repeatedly in previous sections (see 2.2.2,
3.1). This issue concerns dependency relations.
There are some developments that do occur when the unit undergoing
grammaticalization forms the governing, but never when it forms the depen
dent member of the construction in question. The presence of source proposi
tions like the ones presented in 2.2.2 or of reanalyis structures (3.1) is confined
to constructions where the grammaticalized unit forms the governing member
or head. Thus, when in a given language the noun 'back' is grammaticalized
to a preposition or postposition ('behind', 'after') then this involves both a
source proposition, in this case a part-whole proposition, and reanalysis. The
effect of grammaticalization in this case is that a genitive construction like
back of the mountain is reanalyzed as an adverbial phrase behind the mountain.
Nothing of this sort happens when the unit undergoing grammaticalization
171
forms the dependent member, as is the case, e.g. when the demonstrative
'this' is grammaticalized to a definite marker ('the'), turning a phrase: Demon
strative noun (this man) into a phrase: Definite article noun (the man).
There is neither a source proposition involved nor is there any discernible
reanalysis pattern.
This suggests that dependency forms a parameter which is of immediate
relevance to our discussion. Whether a given entity governs or is governed
by another entity is likely to determine its fate in the process of metaphorical
use and of grammaticalization. Since Tesnire (1959) at the latest, depen
dency relations have become a central concern of linguistic analysis. Their
equivalent in cognitive psychology however is less clear. There is reason
enough to assume that they are equally relevant to cognition as to language
structure.
3.3. Grammaticalization chains
Both transfer and context-induced reinterpretation are responsible for
what turn up in language structure as grammaticalization chains, which we
interpret as a somehow frozen result of conceptual manipulation. Grammati
calization chains reflect linguistically what has happened on the way from
more concrete to abstract contents, and they make it possible to reconstruct
that process.
It may be useful to distinguish "grammaticalization chain" from a similar
term, "grammaticalization channel", which has found some currency during
the past decade (Givn, 1979; Lehmann, 1982; Heine and Reh, 1984). Both
refer to the same phenomenon, though highlighting differing aspects of it.
The latter term relates mainly to alternative ways or paths of grammaticaliza
tion. In Heine and Reh (1984:113) grammaticalization channels are therefore
described as alternative options available to languages for introducing a new
grammatical category. While in that usage channels are defined with reference
to their endpoint, they may equally be viewed from their starting point, or
source. This is the case, for instance, when one tries to establish what a given
conceptual entity, say a body part, may develop into, like a grammatical
marker of time, manner, purpose, cause, etc. In such cases, the term "gram
maticalization channel" refers to different development lines of one and the
same source concept.
Grammaticalization chains 37 on the other hand concern the internal
structure of channels, or parts thereof; they relate to the cognitive and linguis
tic nature of these channels. They can in the same way be interpreted as
172
relationship patterns or processes, as synchronic or diachronic phenomena,

or even as dynamic, panchronic entities. Lehmann (1982:26) claims that the
relationship among elements figuring in a grammaticalization channel is a
diachronic one.
Within a diachronic perspective one might characterize grammaticaliza
tion as a process turning an item, in most cases a lexical item, from a "less
grammatical to a more grammatical status" (Kurylowicz, 1965:52). With the
term "grammaticalization chain" we wish to draw attention to a few charac
teristics of this process. These characteristics concern in particular overlapping
and asymmetry.
To demonstrate what these terms stand for we will come back to the
development of the Ewe lexeme megb 'back' which was discussed briefly in
section 2.3.3. As we saw there, this lexeme stands for at least seven differing
conceptual entities constituting a chain of increasing "abstractness" extending
from a concrete, visible/tangible entity, a body part, to a non-physical entity
expressing a quality, 'backward, mentally retarded'. The presence of such a
chain has considerable linguistic implications; it is immediately reflected, e.g.
in morphosyntactic behavior. This behavior is summarized in (31) and can
be described in the following way. As a noun denoting a human body part
(OBJECT), megb shows all the properties that concrete nouns in Ewe have.
It may form the head of a noun phrase (NP) and take any kind of nominal
qualifiers. When used as the head of a genitive construction it is linked with
the genitival modifier by means of the "alienable" possessive marker p. 38 At
the other end of the chain megb appears as a grammaticalized word which
has turned from a nominal into an adverbial entity, may no longer accept
any qualifiers, or be associated with the genitive marker p.
Between these two extremes there are a number of intermediate stages
showing a similar structure as those of the conceptual chain. The transition
from one stage to the next is gradual, first involving the case morphology:
Once megb no longer refers to a human body part but is still conceived as
an OBJECT concept, the use of the genitive marker becomes optional (stage
B). In stages and D, where megb appears, respectively, as a SPACE and
a TIME concept but is still treated as OBJECT, it may optionally form the
head of an adverbial phrase (AP) although it behaves like a noun which may
take qualifiers like adjectives, demonstratives, etc. Stages E and F, where
megb no longer has OBJECT-like features, are marked by two linguistic
innovations: It may no longer be associated with the genitive marker p and,
although it can still be found occasionally with nominal (possessive) modifiers,
it has turned from a noun into either an adverb or a postposition.
173
Table 2. The morpho-syntax of megb

Category
Gloss
Word class
(N = noun,
A = adverb,
= post
position)
Constituent type
NP = head of a
noun phrase,
AP = head of an
adverbial
phrase)
Morphology
(p = presence
of the possessive
marker p)
OBJECT/
PERSON
'back of
body'
NP
OBJECT
'back
part'
NP
/-
OBJECT/
SPACE
'place
behind'
NP/AP
P/-
OBJECT/
TIME
'time
after'
NP/AP
P/-
SPACE
'behind'
N/A/P
AP
TIME
'after'
N/A/P
AP
QUALITY
'retarded'
AP
This transition from a full-fledged noun to an adverb or postposition

shows some properties which appear to be characteristic of grammaticalization processes. The first is that it is marked by overlapping: There is always
a stage where the preceding and the following structure co-exist side by side
as optional variants, before the former gives way to the latter. In stage
overlapping concerns the case morphology, in stages and D the constituent
type and in E and F the word class involved. Thus, we meet the same type
of chaining in morphosyntactic structure that we observed in the case of the
conceptual structuring (cf. 2.3.3).
In cases of overlapping the categorial status of the items involved is
difficult to describe. In stages E and F of Table 2, for instance, megb is not
made up of a combination of the properties of nouns, adverbs and postposi
tions although it exhibits properties of all three word classes; megb rather
is, and has in fact been described in traditional grammars as, a linguistic
174
"hybrid" which is elusive of the established linguistic taxonomy and may best
be characterized as an entity which is no longer a noun but not yet quite an
adverb or postposition, just somewhere in between.
While in cases like this the linguistic situation appears to be an immediate
reflection of conceptual manipulation, there is also a decisive difference:
Conceptual chaining precedes morphosyntactic chaining. This difference is
somehow predictable since it can be derived from the nature of the process
concerned. As we have tried to demonstrate above, grammaticalization is the
result of conceptual manipulation, and cognitive re-structuring therefore pre
cedes linguistic change. In Table 2, for instance, we notice that the lexeme
megb has a spatial or temporal significance in stages and D, respectively,
but is still encoded as a noun, and even in stages E and F it still has nominal
traits although it turns up as a pure spatial or temporal concept, respectively.
This means that the iconicity between conceptual and linguistic structure
alluded to in 3.1 is consistently interfered with. The result is asymmetry
between cognitive and linguistic structure, which appears to be another salient
characteristic of grammaticalization chains. 39
As we will see below, these characteristics of grammaticalization chains
are immediately relevant to linguistic description.
3.4. Grammaticalization and discourse role
Chains like the one presented above display a predictable correlation
with the discourse pragmatic parameter of referentiality/manipulability: The
lexeme megb is maximally referential when used as an OBJECT-like entity
and minimally referential when associated with the QUALITY category. It
would seem, however, that this relationship is complex and that not much is
gained by reducing cognitive-ideational phenomena of the kind under con
sideration to pragmatic-textual interpretation, or vice versa.
We may use another example from Ewe to illustrate this point. Our
example involves the lexeme tsu 'man, adult male', which can be regarded
as a "prototypical noun". Depending on the respective context, this lexeme
displays a remarkable range of semantic and morphosyntactic variation.
Sentences (31) to (33) are characteristic of some of the uses which are associ
ated with rjtsu. In (31), tsu denotes a concrete noun which exhibits all
characteristics of a "good noun": It fills the syntactic slot of a noun and may
take any of the qualifiers which commonly associate with nouns, like number
markers, determiners, etc. In (32) it also behaves syntactically like an object
noun but is not animate as in (31) and may not take any qualifiers, and in
175
(33) it occupies the position of an object noun but is no longer "nominal in

meaning" and takes modifiers like tc 'very', which are governed by adjec
tives and adverbs but not by nouns, and hence behaves more like an adjective
or adverb than a noun.
(31) me le rjtsu ny d d-m
1SG COP man nice INDEF want-PROG
am looking for a nice man.'
(32)
de rjtsu la
me
n-m
put man body POSTP PREP-1 SG
'He has given me courage.'
3SG
(33) wc rjtsu tc
3sG do man very
'He behaved very bravely.'
The case of rjtsu is in no way peculiar or idiosyncratic; the Ewe dictio
nary is full of cases of a similar nature. The strategy of lexicographers dealing
with them is to list the various uses of the relevant lexeme as "polysemes" or
"homonyms" and leave it at that (cf. Westermann, 1905:410/411). One way
of accounting for such cases is suggested by Hopper and Thompson (1984)
who analyse variations in the use of nouns and verbs in terms of their
respective discourse roles. In sentence (31), for instance, rjtsu may be said to
form an example of a highly salient discourse participant which is autonomous
and manipulable and represents a prototypical noun, hence its capacity to
associate with the whole range of nominal morphology. In (32) and, even
more so, in (33), on the other hand, rjtsu may be interpreted as a nonmanipulable, dependent and non-individuated entity which is low in categorial
status and therefore lacks the morphological trappings characteristic of proto
typical nouns.
The present approach may be viewed as complementing that of Hopper
and Thompson (1984). Rather than looking at the discourse value it focusses
on the conceptual manipulation of linguistic units. As we have seen above,
concrete, visible/tangible objects are employed to conceptualize less concrete
entities. We have proposed a number of cognitive categories like PERSON,
OBJECT, SPACE, etc. which stand in a metaphorical relationship governed
by cognitive distance. This relationship is essentially unidirectional, where less
distant categories tend to be employed to understand and/or describe more
distant categories. In this way, concepts associated with the category PER-
176
SON may serve to describe concepts of a higher degree of cognitive distance,

like OBJECT or QUALITY. Exactly this appears to have happened in the
case of tsu. In sentence (31) this lexeme stands for a concrete, human
concept. Sentence (32) is an example where rjutsu no longer refers to a human
being but rather to a quality associated with that concept, although being
expressed like an OBJECT-like entity. In (33) finally tsu designates purely
a quality, thereby competing with word classes normally employed by Ewe
speakers to encode qualities, like adjectives and state verbs. Thus it would
seem that the conceptual range of this lexeme includes three cognitive cate
gories, PERSON, OBJECT and QUALITY, and, depending on the relevant
category it is associated with in a given context, it shows a differing morphosyntactic behaviour: It takes the full range of nominal morphology when
referring to the category PERSON but lacks nominal characteristics when
referring to an OBJECT-like entity. Finally, when standing for a QUALITY
concept it has more in common with adjectives than with nouns.
The view adopted here differs however at least in one aspect from that
proposed by Hopper and Thompson (1984). A lexeme like tsu no doubt
experiences what they refer to as decategorialization when used in certain
contexts: It loses the ability to act as an autonomous discourse participant,
it loses in manipulability and "cardinal categoriality" and it no longer behaves
like a prototypical noun. But these losses are counter-balanced by the fact
that there are also gains: In the same way as rjutsu ceases to refer to an
OBJECT-like unit, a visible, tangible entity, it acquires the contours of an
alternative cognitive domain, that of qualities, and its loss in noun-like
properties tends to be compensated by a gain in properties of other word
classes like those of adjectives or adverbs. 40 The lexeme tsu certainly
constitutes a "more important" and a "more salient" discourse participant
when used as a referential unit but it might be cognitively more complex and
expressive when used non-referentially.
3.5. Problems of grammatical description

We have interpreted grammaticalization as a result of the transfer from
concrete to less concrete conceptual domains, a process that is essentially
metaphorical in nature. This leads to the emergence of grammaticalized
structures as the frozen, linguistic product of that cognitive activity. In the
present section we will look into the question how all this affects synchronic
grammar.
177
3.5.1. 'Back' in So
To start with, let us look at another example involving the lexeme 'back',
this time from So, a Kuliak language spoken in northeastern Uganda. So is
a VSO language distinguishing three cases, absolute (unmarked), dative (-Vk)
and ablative (-, -, -a). Our example is confined to the ablative (ABL) case
marker, but we could have taken any other case instead. The use of this
marker is obligatory with certain verbs. The verb nk 'be somewhere' for
instance requires its locative complement to be in the ablative case. There is
however some variation as to when and where case has to be marked. In (34)
below, three largely synonymous sentences are presented, all being translated
as 'He is behind the mountain'.
(34) a. nk
be
b. nk
nks
ca su-o
sg
3SG back-ABL mountain
ca s-o sg-o
ca s sog-o
These sentences differ only in the fact that the ablative case is marked
on 'back' in (34a), on both 'back' and 'mountain' in (34b) and on 'mountain'
in (34c).
While (34) offers a case where differing morphophonemic structures
express essentially one and the same meaning, the opposite case exists as well,
where several meanings correspond to one form only, like in (35), where
'back' may either denote a body part (a) or a spatial concept, a prepositional
entity. Whenever (a) obtains then 'back' forms the head of a genitive construc
tion, while in the case of (b) it has the function of a preposition within an
adverbial phrase.
(35) nks cc su-o
m
be fly back-ABL girl
a. 'There is a fly on the back of the girl.'
b. 'There is a fly behind the girl.'
A grammar of So has to account, inter alia, for questions like the
following:
a.
Why are there three optional variants in (34) expressing much the
same meaning?
178
BERND HEINE, ULRIKE CLAUDI AND F R I E D E R I K E H N N E M E Y E R
b.
d.
e.
Why can case in (34) be marked either on the head or on the

modifier or on both? Are there any linguistic clues to explain this
situation?
Why is (35) semantically ambiguous?
Why does the lexeme s have two rather divergent meanings,
denoting on the one hand a body part 'back' and on the other
hand a preposition 'behind'?
Why is case marking on s obligatory when it denotes a noun but
optional when it denotes a preposition?
It would seem that answering such questions is hardly possible using any
of the models of language description that are on the market. What we are
dealing with here are structures which are the immediate result of conceptual
manipulation leading from a lexical to a grammatical entity, and sentences
(34) and (35) represent differing stages of this process. As we have demon
strated above, this process does not proceed straight from one category to
another but rather involves overlapping, i.e. a stage where the former meaning
still exists while a new meaning is introduced. The result is semantic ambiguity,
as can be observed in (35), where the morphosyntax is still that of the first
stage, while semantically the first stage (35a) co-exists side by side with the
second stage (35b). (35b) also exhibits another characteristic of grammaticalization chains, asymmetry (see 3.3): Whereas the meaning has shifted from
body part noun to preposition, the morphology is still that of a noun, i.e.
conceptual transfer has not yet affected the morphosyntax.
Asymmetry is even more pronounced in example (34). This sentence
represents a stage where the conceptual transfer from the body part noun
'back' to the preposition 'behind' has been concluded, yet morphosyntax has
not quite kept pace with this process. Thus, while s clearly has the function
of a preposition it still retains the case morphology of a noun in (34a) and
(34b), and only in (34c) the case marker is eliminated.
Sentence (34) exemplifies yet another feature of grammaticalization
chains, which has been described by Heine and Reh (1984:98ff.) under the
label adjustment and which relates to the strategy of restoring a kind of
iconicity between semantic and morphosyntactic structure: With the reanalysis
of the genitive noun phrase su-o sg 'back of the mountain' as a prepositional
phrase 'behind the mountain' the erstwhile head noun 'back' is reanalyzed as
a preposition and the modifying genitive noun 'mountain' as the "semantic
head" of the emerging prepositional phrase. (34) exemplifies the three major
stages of the adjustment process: (34a) represents the initial stage where the
179
erstwhile head noun still shows the case morphology. (34b) marks the typical
overlapping stage, where the case morphology is still on the erstwhile head
noun but has also been introduced on the new, "semantic head". Iconicity is
restored in (34c), where the preposition exhibits no more case marking, which
is now entirely confined to the "new head". Thus, example (34) exhibits a
case shift from the preposition to the noun governed by that preposition,
whereby the intermediate stage is marked by a kind of case agreement, where
the case marker is suffixed to both the preposition and the noun.
An additional problem a descriptive linguist is confronted with when
dealing with sentences like (34) and (35) is the following. In (34) we have
isolated a "preposition" which has either the shape s-o or s. The morphosyntax of this lexeme is complex. It has both nominal and prepositional character
istics, but there are reasons to suggest that it forms neither of these but rather
a word class intermediate between nouns and prepositions. The same lexeme
shows two different meanings in (35), both a nominal and a prepositional
meaning. It would equally be possible to define the meaning of s-o in (35)
as one that contains semantic features of both a body part noun and a spatial
preposition but which cannot be reduced to an element that combines the
semantics of both, rather constituting a unit intermediate between a noun
and a preposition. We have alluded to this problem above (see 3.3); it may
suffice to note here that the continuum-like nature of grammaticalization
chains makes it difficult to maintain a heuristic approach to linguistic descrip
tion which takes discrete word classes for granted. A good part of language
behavior takes place between, rather than within, linguistic categories.
3.5.2. Some conclusions
The discussion about the lexeme s in So concerns only a small segment
of a grammaticalization chain, like the one sketched in more detail in section
3.3. Chains like these are not isolated instances occurring in some exotic
languages, rather they may be observed in any language and have to be
accounted for in a theory of language description. Such a theory has to take
into consideration the following kind of observations which are immediately
derived from grammaticalization phenomena, as we have seen in the preceding
sections:
a.
Grammaticalization can be conceived as a process mapped onto

language structure. It is hard to understand the structure without
understanding the process that has given rise to it.
180
BERND HEINE, ULRIKE CLAUDI AND FRIEDERIKE HUNNEMEYER
b.
c.
d.
e.
f.
The dynamics of this process are reflected, e.g. in the form of

grammaticalization chains figuring in synchronic language struc
ture. An "adequate" description of that structure has to provide
information in particular on:
the number and types of grammaticalization chains occurring in
that language,
the number of members involved in each of these chains,
the type of contexts associated with each member,
the semantic, morphological and syntactic properties of each
member.
A not insignificant part of what turns up in grammar as polysemy
or homophony represents different members of one and the same
grammaticalization chain. Members which are close to one another
in such a chain tend to be interpreted as polysemes while more
distant members are likely to be interpreted as homonyms. For
instance, within the grammaticalization chain presented in section
3.3, the meanings 'back of body' and 'back part (e.g. of a house)'
are likely to be regarded as polysemes, while any of these meanings
and the meaning 'mentally retarded' might be regarded as
homonyms.
In addition to their discrete characteristics, grammaticalization
chains behave like continua without clear-cut boundaries, and they
are therefore elusive to a taxonomic approach to language descrip
tion in terms of discontinuous taxa like constituent types, word
categories, or morpheme classes.
As we have seen above (3.3) overlapping, where an earlier stage
coexists side by side with a subsequent stage, is an intrinsic property
of grammaticalization chains. Overlapping has various implications
for synchronic grammar. One effect is that it creates ambiguity,
whereby either one form shows several meanings or else one mean
ing is associated with two or more different forms.
Since conceptual shift precedes morphosyntactic and phonological
shift, the result is asymmetry between meaning and form. All
languages we are familiar with show examples of morphemes or
constructions which have acquired a new meaning or function
although they still retain the old morphosyntax. In this way, gram
maticalization may, and in fact frequently does, interfere with the
iconicity structure holding between form and meaning.
4.
181
Summary
One main purpose of this paper was to look at one of the major problems
of grammaticalization studies, the question of what motivates or causes
grammaticalization processes. We concentrated on some cognitive aspects of
grammaticalization and argued that the rise of grammatical categories is the
result of what we call conceptual manipulation as a problem-solving strategy
which serves to understand or describe more "abstract" concepts or concep
tual domains in terms of more concrete ones. This process is metaphorical in
nature and can most appropriately be described by means of categorial
metaphors on the one hand and context-induced reinterpretation on the other.
Transfer and the chaining processes are responsible for what turns up in
language structure as grammaticalization chains one of the most striking
characteristics of grammaticalization processes, interpreted as the result and
linguistic reflection of that cognitive activity.
Assuming that language as a whole is metaphorically structured (cf.
Mauthner, 1901; Lakoff and Johnson, 1980) and that grammaticalization and
grammaticalization processes are in no way exceptional, the question arises
how "true" metaphors are distinguished from those leading to grammatical
structures. We have drawn attention to the "source structures", including
source concepts and source propositions, which can be characterized as
referring to basic human activities, providing reference points for human
orientation and being capable of evoking relevant associations. Moreover,
source concepts seem to be of a very general, unspecified content.
Apart from questions such as what motivates grammaticalization and
what constrains the input of grammaticalization such as its source concepts,
other topics have been addressed, among them the relationship between
metaphor and context, the nature of chaining, and the way grammaticalization
affects synchronic language structure.
Traugott and Knig (this volume) argue that grammaticalization involves
metonymy. This is supported by our own observations. The main claim made
in the present paper is that, rather than forming mutually exclusive cognitive
activities, metaphor and metonymy are both present in the development of
grammatical categories; they form complementary aspects of this process.
NOTES
1.
For valuable comments we are indebted to Keith Allan, Derek Nurse, Talmy Givn,
Elizabeth Traugott, Joan Bybee, Fritz Serzisko and Eithne Carlin. Our gratitude is also
182
BERND HEINE, U L R I K E CLAUDI AND F R I E D E R I K E H U N N E M E Y E R

due to Kossi Tossou for information on his mother tongue, Ewe and to the Deutsche
Forschungsgemeinschaft (German Research Society) for having sponsored this research.
2.
Cf. Lehmann (1982:v): "From the diachronic point of view, it (i.e. grammaticalization;
authors' note) is a process which turns lexemes into grammatical formatives and renders
grammatical formatives still more grammatical." In a similar way, grammaticalization is
described in Heine and Reh (1984:15) as a process or "evolution" "whereby linguistic
units lose in semantic complexity, pragmatic significance, syntactic freedom, and phonetic
substance, respectively. This is the case for instance when a lexical item develops into a
grammatical marker."
3.
For a discussion of this term see Lehmann (1982:9).
4.
Cf. Givn's discussion of English "up" which appears to experience a kind of degrammaticalization to a process verb (Givn, 1975:96). See also Campell and Greenberg, this
volume.
5.
Concerning a discussion of such factors, see Traugott (1982:265).
6.
For more details see section 2.3.2. From a slightly different perspective, Traugott (1980:54)
comments the meaning change to be observed in grammaticalization in the following
way: "The speaker needs to specify a new relation, or to strengthen one that already
exists but has become eroded .. .The exigencies of having to be clear direct the speaker
to the most concrete term possible."
7.
Cf. Traugott (1982:246). Kuryiowicz (1964:245) notes: "Fundamental categories directly

based on the speech-situation (ego or hic, nunc) are the starting-point of the elaboration
of higher (grammatical) categories."
8.
Svorou (1986:526) has shown that there are three kinds of nouns developing into locative
adpositions:
(1) Body part nouns: head, heart, anus, mouth, face, neck, ear, forehead, back, loins,
rib, body, breast, chest, blood, foot, waist, belly, and stomach.
(2) Object-part nouns: front, edge, top, back, bottom, side, flank, end, middle, entrance,
circumference, outside, interior, exterior, upper space, space in between.
(3) Environmental landmarks: field, ground, canyon, sky, house.
9.
One may wonder however why body parts like 'nose', 'hair', etc. are apparently never
exploited as a concept for spatial orientation.
10.
Nominalization can be achieved by morphological forms like infinitives, gerundials,

participles, nomina agentis and the like.
11.
The English progressive construction might as well owe its origin to the locational
proposition.
12.
Ewe, also called Gbe, is a West African Niger-Congo language spoken in Togo and
eastern Ghana.
13.
Thus, the Ewe sentence me le yi-yi g (I be going intentional) 'I am about to go, I intend
to go' is historically derived from *me le yi-y gb (I be go-go-nominalization area/place)
'I am at the place/area of going'.
14.
The phrase "at Y's place" may be rendered in individual languages as "at Y's home",
"in Y's hand", etc.
15.
So has a third future marker, - for which there is no etymology. Note that the ac future
is largely confined to the Tepes dialect whereas the g future occurs in the Kadam dialect
183
of So. What the So verbs ac and g have in common, in spite of their opposite deictic
content, is that both imply a goal case.
16.
Whereas Heine and Reh (1984) define desemanticization essentially as a shift from a
"lexical" to a "grammatical" meaning, Greenberg (these volumes) uses this term for a
process whereby a given morpheme loses its grammatical (or lexical) meaning, thereby
becoming a "functionally empty" segment.
17.
"Since the initial meaning is richer, more specific, it is also more palpable, more accessible
to the imagination ("anschaulich") and, in this sense, more concrete; whereas the meanings
of strongly grammaticalized signs, such as "of", "will" or "and", do not yield mental
images, cannot be illustrated and are, in this sense, more abstract" (Lehmann, 1982:128).
18.
This is only one of the ways in which "isolating abstraction" has been used; various other
ways have been proposed as well.
19.
A number of these distinctions are closely interrelated, some may be viewed as expressing
essentially the same thing in different ways. The division into an ideational, a textual and
an interpersonal domain is based on Halliday (1970:143).
20.
This point will be further taken up in 3.4.
21.
As we shall see below, this is but one aspect of the process, though perhaps the most
salient one. Furthermore, it is important to note that we are dealing here with only one
of various types of metaphor, one that serves to describe or understand conceptually
complex phenomena in terms of less complex phenomena. This type has been discussed
under the label "conceptual metaphor" by Lakoff and Johnson (1980). Thus, both
expressive metaphors, which serve to enrich the expressiveness of an utterance, or taboo
metaphors, which serve to conceal or obscure reality (cf. Claudi and Heine, 1986:299),
remain out of consideration here.
22.
Note the rearrangement of the categorial metaphors PROCESS and SPACE in compari
son to former publications, cf. Claudi & Heine, 1986: 301.
23.
Concerning the term "categorial metaphor" and the way it is to be distinguished from
the conceptual metaphors proposed by Lakoff and Johnson (1980) see Claudi and Heine
(1986). The former represent a more "abstract" level of analysis in that each of them
includes a cluster of conceptual metaphors.
24.
In a number of languages an additional metaphor, QUALITY IS TIME, has been applied,

with the result that the lexeme for 'back' has also acquired the meaning 'mentally back,
retarded' (see below).
25.
In this context, Keith Allan (p.c. of 28.9.1987) draws attention to the fact that PROCESS
implies a proposition and is itself a sort of abstract object.
26.
Ewe is a tone language having an analytic-isolating morphosyntax. It has SVO as its

basic word order, though SOV order occurs in the progressive and ingressive aspects (see
Heine and Reh 1984:188-190), and a possessor-possessed syntax.
27.
"A characteristic of virtually all developments is that when a given linguistic unit undergoes
a certain process then it does not do so in all its uses; it tends rather to be retained in its
former status as well, so that there are two coexisting forms of that unit: One that still
represents the old status and another that marks the new status resulting from grammaticalization" (Heine and Reh, 1984:57).
28.
The labels "OBJECT/PERSON" stand for an OBJECT concept which is typically associ
ated with human beings, though less typically also with animals.
184
BERND HEINE, U L R I K E CLAUDI A N D F R I E D E R I K E HNNEMEYER
29.
Elizabeth Traugott (p.c.) points out that this structure presents a fundamental principle
of all language change.
30.
Note that the TIME category may give rise to categories other than CAUSE, like
CONDITION or QUALITY, see Heine (1990).
31.
In Langacker's terminology, reanalysis is further classified into resegmentation and refor

mulation.
32.
One of the effects of grammaticalization is for instance that it turns governing into
governed constituents. This process may trigger an opposite development of the erstwhile
governed to a governing constituent (cf. Heine and Reh, 1984:95; 104-105). Both develop
ments involve reanalysis but only the former grammaticalization (see below).
33.
Langacker (1977:64) refers to constituent-internal reanalysis, perhaps more appropriately,

as boundary shift, which he treats as one form of resegmentation, the other forms being
boundary loss and boundary creation. This type of reanalysis has been observed in a
number of languages but will not be further considered here since it does not seem to be
a property directly relevant to grammaticalization.
34.
In Heine and Reh (1984:95), the term was applied only to syntactic and pragmatic, but
not to morphological structures; this restriction is not maintained here.
35.
Teso is an Eastern Nilotic language of the Nilo-Saharan family spoken in western Kenya
and eastern Uganda.
36.
The sentence is taken from Hilders and Lawrence (1956:XIX); see also Heine and Reh
(1984:104-105).
37.
Grammaticalization chains must not be confused with grammaticalization scales, which

according to Lehmann (1982:26) are descriptive constructs involving functionally similar
signs.
38.
Concerning the distinction "alienable"/"malienable" in Ewe see Claudi and Heine

(1986:316), Claudi and Heine 1989.
39.
This observation has been made already in earlier works. Givn (1975:86) for instance
notes: "It is highly unlikely that a verb would change suddenly into a preposition by all
semantic, morphological and syntactic criteria at once. One thus expects to find, for a
long time, many different types of intermediate cases in the language, where by some
criteria a 'particle' is already a preposition, while by others it is still 'a verb' In particular,
morphological and syntactic behavior is likely to lag behind the mofe progressive semantic
re-analysis, and thus quite often represent vacuous relics of the older semantic situation."
40.
In accordance with the terminology proposed by Hopper and Thompson (1984) one
might talk in this case of "recategorialization" which follows decategorialization.
REFERENCES
Allan, Keith. 1987. "Hierarchies and the choice of left conjuncts (with particular
attention to English)." Journal of Linguistics 23:51-77.
Awolaye, Yiwola. 1986. "Reflexivization in K w a languages." In Gerrit J. Dimmendaal
(ed.) 1986:1-14.
Berlin, Brent, Dennis E. Breedlove and Peter H. Raven. 1973. "General principles of
185
classification and nomenclature in folk biology." American Anthropologist

75:214-242.
Berlin, Brent, Dennis E. Breedlove and Peter H. Raven. 1974. Principles of Tzeltal
Plant Classification: An Introduction to the Botanical Ethnography of a MayanSpeaking Community in Highland Chiapas. New York: Academic Press.
Bolinger, Dwight. 1978. "Intonation across languages." In Greenberg et al. (eds).
1978:471-524.
Bopp, Franz. 1816. ber das Conjugationssystem der Sanskrit sprache in Vergleichung
mit jenem der griechischen, lateinischen, persischen und germanischen Sprachen.
Frankfurt (M.): Andreische Buchhandlung.
Bybee, Joan L. 1985. "On the nature of grammatical categories." Paper presented at
the Second Eastern States Conference on Linguistics, Buffalo, October 3, 1985.
Bybee, Joan L. and William Pagliuca. 1985. "Cross linguistic comparison and the
development of grammatical meaning." In Fisiak (ed.) 1985:59-83.
Claudi, Ulrike and Bernd Heine. 1986. "On the metaphorical base of grammar."
Studies in Language 10(2):297-335.
Claudi, Ulrike and Bernd Heine. 1989. "On the nominal morphology of 'Alienability'
in some African languages." In Paul Newman and Robert D. Botne (eds). Dor
drecht: Foris.
Croft, William, Keith Denning, and Suzanne Kemmer (eds). 1990. Studies in Typology
and Diachrony. Amsterdam and Philadelphia: John Benjamins [Typological Studies
in Language 20].
Dimmendaal, Gerrit J. (ed.). 1986. Current Approaches to African Linguistics Vol. 3.
Dordrecht/Cinnaminson: Foris [Publications in African Languages and Linguistics
6].
Essien, Okon E. 1982. "The so-called reflexive pronouns and reflexivization in Ibibio."
Studies in African Linguistics 13(2):93-108.
Fisiak, Jacek (ed.). 1985. Historical Semantics, Historical Word Formation. The Hague:
Mouton.
Fleischman, Suzanne. 1989. "Temporal distance: A basic linguistic metaphor." Studies
in Language. 13(1):1-50.
Gabelentz, Georg von der. 1891. Die Sprachwissenschaft: Ihre Aufgaben, Methoden
und bisherigen Ergebnisse. Leibzig: Weigel Nachfolger.
Givn, Talmy. 1975. "Serial verbs and syntactic change: Niger-Congo." In Li (ed.)
1975:47-112.
Givn, Talmy. 1981. "On the development of the numeral 'One' as an indefinite
marker." Folia Linguistica Historica 2:35-54.
Givn, Talmy. 1982. "Logic vs. pragmatics, with human language as the referee:
Toward an empirically viable epistemology." Journal of Pragmatics 6:81-133.
Givn, Talmy. 1984. "Direct object and dative shifting: Semantic and pragmatic case."
In Plank 1984:151-82.
Givn, Talmy. 1989. Mind, Code and Context: Essays in Pragmatics. Hillsdale, N.Y.:
Erlbaum.
Greenberg, Joseph H. 1978. "How does a language acquire gender markers?" In
Greenberg et al. (eds). 1978(3):47-82.
Greenberg, Joseph H., Charles A. Ferguson, and Edith Moravcsik (eds). 1978. Univer
sals of Human Language, 4 Vols. Stanford: Stanford University Press.
Gudschinsky, Sarah 1956. "The ABC's of lexicostatistics (Glottochronology)."
Word 12:175-210.
186
Guillaume, Gustave. 1964. Langage et Science du Langage. Paris, Nizet and Quebec:
Presses de l'Universit Laval.
Guimier, Claude. 1985. "On the origin of the suffix -ly." In Fisiak (ed.) 1985:155-170.
Halliday, M.A.K. 1970. "Language structure and language function." In Lyons (ed.)
1970:140-165.
Heine, Bernd. 1990. "The Dative in Ik and Kanuri." In Croft et al. (eds) 1990:
129-149.
Heine, Bernd and Ulrike Claudi. 1986. On the Rise of Grammatical Categories. Some
Hilders, J.H. and J.C.D. Lawrance. 1956. An Introduction to the Ateso Language.
Kampala: The Eagle Press.
Hopper, Paul J. and Sandra Thompson. 1984. "The discourse basis for lexical cate
gories in universal grammar." Language 60:703-752.
Humboldt, Wilhelm von. 1825. "ber das Entstehen der grammatischen Formen und
ihren Einflu auf die Ideenentwicklung." Abhandlungen der Kniglichen Akademie
der Wissenschaften zu Berlin 1825:401-430.
Jeffers, Robert J. and Arnold M. Zwicky. 1980. "The evolution of clitics." In Traugott
et al. (eds) 1980:221-231.
Kahr, Joan . 1976. "The renewal of case morphology: Sources and constraints."
Working Papers on Language Universais 20:107-151.
Knig, Ekkehard. 1985. "Where do concessives come from? On the development of
concessive connectives." In Fisiak (ed.) 1985:263-282.
Kurylowicz, Jerzy. 1964. The Inflectional Categories of Indo-European. Heidelberg:
Carl Winter.
Kuryiowicz, Jerzy. 1965. "The evolution of grammatical categories." In Kuryiowicz
1975:38-54.
Kuryiowicz, Jerzy. 1975. Esquisses Linguistiques II. Mnchen: Fink
Lakoff, George and Mark Johnson. 1980. Metaphors We Live by. Chicago/London:
University of Chicago Press.
Langacker, Ronald W. 1977. "Syntactic reanalysis." In Li (ed.) 1977:57-139.
Lehmann, Christian. 1982. Thoughts on Grammaticalization: A Programmatic Sketch.
Vol.1. Kln: Universitt zu Kln, Institut fr Sprachwissenschaft [Arbeiten des
Klner Univer salien-Projekt s 48].
change." Lingua e Stile 20(3):303-318.
Lehmann, Winfred P. and Y. Malkiel (eds). 1982. Perspectives on Historical Linguistics.
Li, Charles N. 1975a. "Synchrony vs. diachrony in language structure." Language
51(4):873-886.
Li, Charles N. (ed.). 1975b. Word Order and Word Order Change. Austin/London:
University of Texas Press.
Li, Charles N. (ed.). 1977. Mechanisms of Syntactic Change. Austin/London: Univer
sity of Texas Press.
Lord, Carol. 1976. "Evidence for syntactic reanalysis: From verb to complementizer
in Kwa." In Chicago Linguistics Society, Papers from the Parasession on Diachronic
Syntax, 179-191.
187
Lyons, John (ed.). 1970. New Horizons in Linguistics. Harmondsworth: Penguin Books
Ltd.
Mauthner, Fritz. 1901. Beitrge zu einer Kritik der Sprache. Stuttgart: Cotta.
Meillet, Antoine. 1912. "L'volution des formes grammaticales." Reprinted in Meillet,
A. 1921-36. Linguistique historique et linguistique gnrale, Vol 1, 130-148. Paris:
Klincksieck.
Newman, Paul and Robert D. Botne (eds). 1989. Current Approaches to African
Linguistics Vol. 5. Dordrecht/Providence: Foris [Publications in African Languages
and Linguistics 8].
Norvig, Peter and George Lakoff. 1987. "Taking: A Study in lexical network theory."
In Berkeley Linguistics Society, Proceedings of the 13th Annual Meeting, 195-206.
Planck, Frans (ed.). 1984. Objects: Towards a Theory of Grammatical Relations.
London: Academic Press.
Schneider, Wolf. 1979. Wrter machen Leute. Magie und Macht der Sprache. Reinbeck:
Rowohlt.
Sperber, Dan and Deirdre Wilson. 1986. Relevance: Communication and Cognition.
Cambridge, Mass.: Harvard University Press.
Svorou, Soteria. 1986. "On the evolutionary paths of locative expressions." In Berkeley
Linguistic Society, Proceedings of the 12th Annual Meeting, 515-527.
Swadesh, Morris. 1951. "Diffusional cumulation and archaic residue as historical
explanations." Southwestern Journal of Anthropology 7:1-21.
Sweetser, Eve E. 1982. "Root and epistemic modals: Causality in two worlds." In
Berkeley Linguistics Society, Proceedings of the 8th Annual Meeting, 484-507.
Sweetser, Eve E. 1987. "Metaphorical models of thought and speech: A comparison
of historical directions and metaphorical mappings in the two domains." In Berke
ley Linguistics Society, Proceedings of the 13th Annual Meeting, 446-459.
Talmy, Leonard. 1986. "The relation of grammar and cognition." Paper No. 165,
Series A, July 1986. Duisburg: L.A.U.T.
Tesnire, Lucien. 1959. Elements de syntaxe structurale. Paris: Editions Klincksieck.
Traugott, Elizabeth Closs. 1980. "Meaning-change in the development of grammatical
markers." Language Sciences 2(1):44-61.
Traugott, Elizabeth Closs. 1982. "From propositional to textual and expressive mean
ings: Some semantic-pragmatic aspects of grammaticalization." In Lehmann and
Malkiel(eds) 1982:245-271.
Traugott, Elizabeth Closs. 1986. "From polysemy to internal semantic reconstruc
tion." In Berkeley Linguistics Society, Proceedings from the 12th Annual Meeting,
539-550.
Traugott, Elizabeth Closs. 1987. "On the rise of epistemic meanings in English; an
example of subjectification in semantic change." Language 65:31-55.
Traugott, Elizabeth Closs, Rebecca LaBrum, and Susan Shepherd (eds.). 1980. Papers
from the Fourth International Conference on Historical Linguistics. Amsterdam:
John Benjamins.
Werner, Heinz and Bernard Kaplan. 1963. Symbol-Formation. An Organismic-develop
mental Approach to Language and the Expression of Thought. New York/London/
Sydney: Wiley and Sons.
Westermann, Diedrich. 1905. Wrterbuch der Ewe-Sprache. Berlin: Dietrich Reimer.
Zirmunskij, V. M. 1966. "The word and its boundaries." Linguistics 27:65-91.
The Semantics-Pragmatics of
Grammaticalization Revisited
Elizabeth Closs Traugott and Ekkehard Knig
Stanford University and Freie Universitt Berlin
1.
INTRODUCTION
"Grammaticalization", as used in this paper, refers primarily to the

dynamic, unidirectional historical process whereby lexical items in the course
of time acquire a new status as grammatical, morpho-syntactic forms, and in
the process come to code relations that either were not coded before or were
coded differently (Givn, 1979; Forthcoming). 1 The study of grammaticaliza
tion challenges the concept of a sharp divide between langue and parole, and
focuses on the interaction of the two. It also challenges the concept of
categoriality, and takes as central the concept of a continuum of bondedness
from independent units occurring in syntactically relatively free constructions
at one end of the continuum to less dependent units such as clitics, connectives,
particles, or auxiliaries, to fused agglutinative constructions, inflections and
finally to zero (cf. Bybee, 1985:11-12; Lehmann, 1985:304).
There has recently been considerable interest in the semantic-pragmatic
processes involved in early stages of grammaticalization, specifically the devel
opment of lexical items into clitics, particles, auxiliaries, connectives, etc.
Traugott (1982) suggested that the main path of change at this early stage of
grammaticalization is: Propositional (>textual)>expressive. More recently
she has revised this formulation and specified the shift as one from meanings
grounded in more or less objectively identifiable extralinguistic situations to
meanings grounded in text-making (for example connectives, anaphoric mark
ers, etc.) to meanings grounded in the speaker's attitude to or belief about
what is said, and has shown that it is part of the larger mechanism of semantic
change in general (Traugott, 1989; 1990). In these papers she refers to princi
ples of inferencing, but with only very general sketches of the kinds of
190
ELIZABETH CLOSS TRAUGOTT AND EKKEHARD KNIG
inferencing that are involved. Our purpose here is to show that different kinds
of inferencing are at work, depending on the particular kind of grammatical
function that is evolving. We will argue that the development of markers of
tense, aspect, case and so forth involve primarily metaphoric inferencing (as
is widely accepted, cf., among others, Bybee and Pagliuca, 1985; Sweetser,
1988; Heine et al., These volumes). By contrast, the kind of inferencing that
is dominant in the development of connectives, specifically causals such as
since, concessives such as while, and preference markers such as rather (than),
is strengthening of informativeness as a conversational implicature becomes
conventionalized. It is the latter kind of inferencing that will be the major
focus of this paper. Of course, metaphor and strengthening of informativeness
are not inconsistent with each other, but rather can be regarded as complemen
tary kinds of pragmatic processes, provided we analyse metaphor as involving
a kind of inferencing (cf. Levinson, 1983; Sperber and Wilson, 1986).2
There has been some difficulty in the past in thinking about the semanticspragmatics of grammaticalization because there has been an assumption since
at least Meillet (1948 [1912]) that grammaticalization involves semantic
weakening, also known as bleaching. Heine and Reh, for example, define
grammaticalization as: an evolution whereby linguistic units lose in semantic
complexity, pragmatic significance, syntactic freedom, and phonetic sub
stance, respectively. (1984:15) (cf. also Givn, 1975; 1979) From this point of
view, grammaticalization is a kind of impoverishment, or deficit as Leh
mann puts it, a process whereby signs lose their integrity (1985:307). Certainly,
bleaching can occur, but, we would argue, most clearly only in the later stages
of grammaticalization, for example in the development of the main verb do
into a dummy auxiliary in Standard English,3 in the development of third
person pronouns into agreement markers, or more generally in the process
of paradigmatic fixation.
Bleaching and grammaticalization must be uncoupled if we are to under
stand the semantic-pragmatic processes of early stages of diachronic grammat
icalization. Sweetser (1988) rightly argues that in cases of grammaticalization
where image-schematic metaphoric transfer occurs, there is less elaboration
of the source meanings than in lexical change, but the grammatical meaning
is added; therefore "bleaching" is an inappropriate concept. With reference
to examples such as the development of future go, she says:
.. .we lose the sense of physical motion (together with all its likely back
ground inferences). We gain, however, a new meaning of future prediction
or intention together with its likely background inferences. We thus
cannot be said to have merely "lost" meaning; we have, rather, exchanged
SEMANTICS-PRAGMATICS OF GRAMMATICALIZATION REVISITED
191
the embedding of this image-schema in a concrete, spatial domain of

meaning for its embedding in a more abstract and possibly more subjective
domain. (Sweetser, 1988:392).
She also elaborates on the proposal that grammatical meaning is inherently
topological and schematic, but that lexical meaning need not be. For example,
grammatical relations may express topological relations on a linear parameter,
but not precise distances between points on the scale. Or they may express
the relative spatial position of two objects, such as sequence and adjacency,
but not precise angles, e.g. comer in time does not grammaticalize, although
front-back, up-down and other spatial relations do (cf. Talmy, 1983; Traugott,
1985a). Sweetser argues that when a lexical item becomes grammaticalized,
only the topological image-schema is transferred in metaphorical mappings:
the meaning transfer "... is to a fairly abstract, topological domain...so there
is less fleshing-out of meaning. However, the meaning of the new domain
itself is added." (Sweetser, 1988:393)
In the present paper we will be focusing on what is added in the process
of grammaticalization, most particularly on strengthening of the expression
of speaker involvement. Consider, for example, the development of the phrase
pa hwile pe 'at the time that' into the temporal connective while. Here the
textual meaning is strengthened and pragmatic functions pertaining to meta
linguistic text-building are added. Later temporal while developed into the
concessive while in the sense 'although', which construes a world that has no
reference in the described situation, but only in the speaker's world of belief
about coherence, especially about correlations between situations or eventuali
ties. At this stage the pragmatics of the speaker's belief-state is strengthened.
The fundamental process we see at work is a principle of informativeness
or relevance, essentially the principle: Be as informative as possible, given the
needs of the situation (cf. Atlas and Levinson, 1981). This contextualized
view of informativeness recognizes that there is a trade-off between speakers'
tendencies to say no more than they must (i.e. the principle of economy),
given assumptions about hearers' willingness and ability to be cooperative,
and hearers' tendencies to select the most informative among possible compet
ing interpretations. It is a view coherent with Lehmann's claim that "every
speaker wants to give the fullest expression to what he means" (1985:315),
but puts bounds on it. If Lehmann's principle were unbounded, presumably
each meaning would be expressed by a different form. But we know this not
to be true. One form may mean several things, as do while, since and rather
(than). The principle of informativeness or relevance adopted here demands
192
only that the contribution is as informative as required, and presupposes that

more will be read in, cf. Levinson (1983:146-7) and:
...human cognitive processes...are geared to achieving the greatest pos
sible cognitive effect for the smallest amount of processing effect...to
communicate is to imply that the information communicated is rele
vant. . .the principle of relevance is enough on its own to account for the
interaction of linguistic meaning and contextual factors in utterance inter
pretation. (Sperber and Wilson, 1986:1)4
From a historical perspective, the principle of informativeness and relevance
presumably drives speakers to attempt to be more and more specific through
grammatical coding, and most especially to invite hearers to select the most
informative interpretation. It does not, however, require a teleological move
ment to one-meaning-one-form that an unbounded principle of expressiveness
would require.
2.
SOME THEORETICAL AND METHODOLOGICAL

ASSUMPTIONS
Before proceeding, a few assumptions should be stated. A fundamental

assumption of all work on language change is that only linguistic phenomena
that can be identified in contemporary languages should be reconstructed (the
"uniformitarian principle", cf. Romaine, 1982); therefore only inferences that
are attested in a modern language may be reconstructed.
As is also true of all linguistic change, the processes outlined are possible
and not necessary. For example, there was no necessity for temporal while to
develop the concessive sense of 'although'; indeed, in German, the cognate
weil developed a causal, not a concessive, meaning even though the conversa
tional implicatures relevant for concessiveness must have been present in
some contexts.
A further assumption is that all languages as we know them have both
semantic and pragmatic meanings. Both are equally present, although prag
matic meanings may be expressed in different ways at different times in
different languages (e.g. by intonation, particles, clitics, word order, etc.). The
claim in this paper that pragmatic meanings are grammaticalized later than
non-pragmatic (or propositional) ones does not mean that a language can at
one stage have only propositional meanings; it means simply that, given a
form X, the pragmatic polysemies associated with it, if any, will have devel-
193
oped later than the propositional ones, that is, they will have been coded at
a later stage. That is, meanings typically shift from what is said to what is
meant, not vice versa. To put it another way, concessive meanings will not
precede temporal ones (concessive / temporal), although the reverse can be
true (temporal>concessive).
In seeking to account for this kind of unidirectionality, we need to ask
ourselves what could motivate it. It is this question that leads us to a theory
of inference and of informativeness and relevance. As we will show, inferences
from temporal to concessive are well-attested and comprehensible, but an
inference from concessive to temporal is not (how would we infer a temporal
from Although you like your oysters raw, I prefer them cooked?) The inferences
in question are typically not strictly deductive to the logical form, but rather
abductive, that is, they are inferences to the best explanation of why the
sentence might be true or relevant in the context (for the theory of abduction,
see Peirce, 1955; Givn, 1989; Hobbs et al., 1988; and, within the historical
domain, Andersen, 1973).
Essential to the present study is the distinction between those pragmatic
meanings that are conventionalized (coded, whether lexically, grammatically,
or prosodically) and those that are inferred in context, largely through conver
sational processes of meaning-specification. Even where absence of coding is
concerned, many inferences are largely predictable. For example, conjoined
clauses without any connective (i.e. without any coding of coherence) are
likely to be interpretively enriched as having some coherence simply because
they are uttered in sequence. Typically the relationship inferred, if the clauses
are action/event clauses, will be that of temporal sequence, as in la), but
other relationships can be inferred if the sequence is temporally and logically
inconsistent, as in lb) (examples adapted from Blakemore, 1987:113):
(1)
a. The road was icy. She slipped

b. She slipped. The road was icy
If a grammatical form is present, e.g. and, because, you see, this element will
further 'constrain the relevance of the proposition it introduces' (Blakemore,
1987:130). But this constraint is not absolute; being an instruction to hearers
how to interpret, the grammatical form may itself be augmented, as was
while. New inferences may arise from it which may themselves be understood
in terms of conversational and conventionalized inference.
The approach taken here is that distinct new polysemies of a form are
new conventional meanings. The germ of this idea is to be found in Grice's
statement that it is possible 'for what starts life.. .as a conversational implica-
194
ture to become conventional' (1975:58). Although analyses may differ concern

ing when polysemy does or does not occur, it is generally accepted that out of
context after is not polysemous in English; however, in context it allows vari
ous inferences about immediacy of precedence. By contrast, out of context
since is polysemous between temporality and causality. We argue that the
polysemies of since have arisen through the conventionalizing of earlier conver
sational inferences. The inferences in question are presumably stereotypical
ones, since only standard inferences can plausibly be assumed to have a lasting
impact on the meaning of an expression or to function cross-linguistically.
3.
THE CONVENTIONALIZING OF CONVERSATIONAL

IMPLICATURES
3.1. Inferred causation

One of the best known instances of an abductive inference involving
pragmatic strengthening is the classical fallacy post hoc ergo propter hoc (cf.
Geis and Zwicky, 1971; Atlas and Levinson, 1981; and Horn, 1984:33).
Consider such well-known instances in PDE 5 as:
(2)
a. After we heard the lecture we felt greatly inspired ( + > because

of the lecture we felt greatly inspired)
b. The minute John joined our team, things started to go wrong
( + >because J joined out team, things started to go wrong)
In (2) the inferences are conversational, i.e. not part of the meaning of any
particular element in the utterance. They are relevance-based, or strengthen
informativeness because they embellish the relation between After the lecture
or The minute John joined our team and the rest of the utterance, and provide
an interpretation of why the speaker thought it was relevant to include these
temporal facts.
Consider by way of contrast the temporal and the causal meanings of
since in 3):
(3)
a. I have done quite a bit of writing since we last met (temporal)

b. Since Susan left him, John has been very miserable (temporal,
causal)
Since you are not coming with me, I will have to go alone
(causal)
195
d. Since you are so angry, there is no point in talking with you

(causal)
With since, when both clauses refer to events, especially events in the past,
the reading is typically temporal, as in (3)a. When one clause refers to a nonpast event or to a state, the reading is typically causal, as in (3)c. and (3)d.,
but the causal reading is not required, as (3)b. indicates. The contrastive
readings in (3)b. signal polysemy, i.e. conventionalized meanings, not just
conversational ones.
It has often been pointed out that temporal-causal polysemies can arise
historically through changes in the status of inferences (cf. Geis & Zwicky,
1971:565; Abraham, 1976; Braunmller, 1978). The following examples are
all instances of grammatical markers that appear to have undergone a change
from 'temporal' to 'causal' (the temporal itself may be derived from an
original spatial, as in the case of consequently):
(4)
Eng. since, consequently, Gm. infolgedessen (therefore); Fr. puisque

(<Lat. posteaquam 'after', 'ever since'); Sp. pues (<Lat. post);
Swed. eftersom, emedan (cf. <medan 'during); Rum. din moment ce
'from the moment on', 'because'; Du. want 'because' (<PGmc
'when + then'); Eston, parast 'after, because of' (cf. parastlouna
'afternoon'); Finn, koska 'when', 'because', 'as'.
A look at some of the data in the history of English can begin to give
us some insight into some of the contexts in which the change of since
(originally sippan) came about. In OE texts between ca.850 and 1050 A.D.
sippan as a preposition was used almost exclusively to mean 'after' (Mitchell,
1985:Par.2670) cites a total of three likely examples of a causal reading); the
standard causal was the originally deictic for pm pe 'for that that'. As a
connective sippan meant 'from the time that', that is, it marked the lower
temporal boundary of the event in the main clause, and signalled an overlap
with some point in an earlier event.
In certain contexts, however, the modern reader may detect a causal
implicature. For example, Mitchell (1985:11, 352) cites as a putative example
of causal sippan:6
(5)
Or. 156.11
pa, sippan
he irre ws & gewundod,
then, after/since he angry was and wounded,
foices
he of slog
micel ps
he slaughtered much of-that troop
196
When we read this sentence out of its larger context, the participle gewundod,
being perfective and involving change of state, seems to favor a temporal
reading, but the adjective irre favors a causal one, since it appears to express
state. Nevertheless, a temporal reading with the adjective is plausible if we
assume that ws is inceptive/resultative or perfective, i.e. 'had come to be'
rather than 'was'. In other words, the adjective in this context can be interpre
ted as expressing a contingent rather than a general state (the latter would
more probably have been expressed by bi7). And indeed we find that this
sentence occurs in the context of a narrative concerning the legendary Pyrrhic
victory. In the battle a Roman soldier wounds an elephant in the navel; this
elephant, having become enraged on being wounded, wreaks mayhem on the
army. Sippan in (5), then, must be interpreted as a temporal (indeed it partially
translates the Latin adverbial postquam 'afterward'), and is therefore not a
conclusive example of the conventionalizing of causative inference in OE.
One virtually indisputable example is:
(6)
Bo. 36 104.26
Ac ic pe wille nu giet getcan pone weg...
But I thee will now still teach
that way...
sian ongitst purk
mine lare
hwt sio
since thou seest through my teaching what that
soe gesl bi, & hwr hie bi
true happiness is, and where it is
'But still I will now teach you the way.. .since through my teaching
you see what true happiness is, and where it is'
Here sippan translates the Latin causative quoniam; but even without the
Latin original we must assume it is causative since the context is non-narrative:
The stative perception/mental verb 'see, undertand' introduces an aspectual
generic clause signalled by the verb bi instead of the contingency verb is.
We may hypothesize that the occurrence of sippan in contexts which did
not block a causal interpretation, such as (5), came to be conventionalized
by frequent use. By the fifteenth century, i.e. very late ME, the form is attested
frequently in stative and other non-completive environments where the tempo
ral reading is blocked, as in (6). In other words, although the causal inference
is detectable in OE, enough other examples are undecidable that we cannot
establish that the causal inference had truly become conventionalized until
the fifteenth century.
The formulation post hoc ergo propter hoc suggests that it is strict
sequence of events, event1 followed by event2, that gives rise to causal implica-
197
tures. And indeed it does account for some causal inferences from temporals,
as in the case of After John entered the room, Bill jumped out of the window.
However, as we have seen, states are more likely to give rise to causal
interpretations than sequences of events, so a more appropriate formula
would be state1 relevant to state2. Furthermore, the traditional formulation
post hoc ergo propter hoc taken literally might lead one to wonder why it is
that some temporals that seem to be stereotypically non-sequential, or at least
can be used both sequentially and non-sequentially, can be interpretively
enriched to causal relations. Consider the conversational implicatures in the
following PDE sentences:
(7)
a. I couldn't work when the television was on.

b. I can't sleep now that I am alone.
Historically, too, we frequently encounter the conventionalizing of causals

that were formerly not necessarily sequential:
(8)
weil ( < O H G dia wila so 'so long as'); Lat. dum 'when, as long as,
because'; Fr. quand when, because'; Finn, 'when, while, as,
since, because'; Eston, 'while, as, since, because'.
It appears then that what is needed for a causal inference to arise is

partial temporal overlap, not sequence. More specifically, the meaning 'from
the time that' licenses (but does not require) causal implicatures to become
conventionalized. A good example is OE nu, which was used as an adverb
meaning 'now'; as a connective it almost always has a causal meaning. We
can postulate that a context such as is found in PDE (7b), where one clause
(the matrix) expresses a state, was the entry-point for the shift from
temporal > causal. Consider:
(9)
AECHom I, 24 350.21
Efne nu pu
eart gehled ne synga pu
heononfor
Even now thou art saved not sin
thou henceforth
'Now you are saved, don't sin from now on.' ( + > because you
are saved, don't sin from now on)
This may be interpreted as 'note that as of the present time you are/have
become saved, and treat it as the starting point of future behavior', with an
inferential enrichment + >'because you note this, therefore from this time
forth modify your behavior'. In (9) causality may be only a conversational
implicature, since temporal relations are clearly distinguished (past participle
gehled, future adverb heononfor). However, it is presumably conventional-
198
ized in the following, where there is no tense change and indeed the context
implies that there has been no change of state.
(10) AECHom I. 26.378.6
Untwylice
pu
lyhst pt u god sy, nu
Unquestionably thou liest that thou god art, because thou

nast
manna gepohtas
not-knowest men's thought
'Without question you are lying when you say that you are God,
because you do not know the thoughts of men.'
The adverbial nu 'from this time forth' is sufficiently informative to
convey a relation between the clause it initiates and the main clause, but
invites the inference that more was meant; why else would the speaker have
mentioned the temporal relation at all? Note that (10) does not code an
external causal relationship that can be regarded as true in the world, but
rather an epistemic causal augmenting the inference that the speaker has
evidence for the proposition (cf. Sweetser, 1990; Blakemore, 1987:139 analyses
causals of this type as indicating dependent relevance, i.e. relevance between
propositions).
As would be expected in any kind of language change, the temporal
meaning that the causal adverbials and connectives discussed here originally
had was not lost immediately, but was typically maintained alongside the
causal meaning for a long time (for detailed studies of coexistence of
polysemies in these volumes, cf. papers by Abraham and by Heine et al.).
In the instance of since, the meanings have coexisted since the twelfth
century. In other instances, either the temporal or the causal meaning has
disappeared as a separate polysemy. An example of the latter is causal now.
An example of the loss of the temporal is provided by consequently, formed
in the fifteenth century from Latin via Old French in both the temporal
and causal senses. The temporal meaning, exemplified in (11a), survived
until the seventeenth century, but was eventually replaced by the causal, as
in (l1b):
(11) a. 1450 (cl405) Purvey Determ. 175/181 (MED)
If it is leuefful to preche pe naked text to pe pupel, it is
It it is lawful to preach the straight text to the pupil, it is
also lefful to write it to hem &
consequentliche,
also lawful to write it to him and afterwards
by proces of tyme, so
al pe Bibil
by process of time, similarly all the Bible
199
b. 1447 Bokenham Sts. 519 (MED)

Than folwyth it thus ful
consequently That thou
Then follows it thus completely with-reason That you
clepyst crist thy god
call
Christ your god
What is important here is that an older meaning that has been lost, such as
temporal consequently, does not seem to be recoverable by inferences; how
ever, if a later meaning such as causal now is lost, it will still exist as a
conversational inference, as in (7b). This is exactly what we would expect if
the hypothesis of the unidirectionality of inferences alluded to in Section 2.
is correct.
3.2. From concomitance to concessivity
We turn now to another example: The concessives. Among the major
sources in the development of concessive connectives are expressions whose
original or at least earlier, meaning was 'concomitance' or 'cooccurrence',
and certain negative expressions (cf. Knig, 1985; 1986; 1988; Forthcoming):
(12) a. connectives originally expressing simultaneity or temporal over
lap: while, still, yet; Gm. zugleich, indes (sen), dennoch; Fr.
cependant; Turk, iken 'while', 'although'; Indonesian sekali-pun
'although' ( < ' a t the same time + even'); Hawaiian oiai 'while',
'meanhile', 'although'; Jap. nagara 'while', 'although'
b. connectives originally expressing simple cooccurrence or con
comitance, or even similarity: EME withal (<'along with the
rest'), all/just the same; Gm. gleichwohl, bei all; Dan. evenwel;
Rum. cu toate ca 'with all (things) that'); Turk, bununla beraber
'together
with
this');
Hopi
naama-hin
'although'
(<'together'+'thus'); Quileute -e 'with, although'
negation of asymmetry (e.g. negation of opposition, negation
of negative correlation): notwithstanding, nevertheless, nonethe
less; Fr. n'empeche que; Gm. nichtsdestoweniger
As in the case of causal connectives, many of the concessive connectives
have retained their original meaning alongside the concessive implicature that
has become part of the conventional meaning. Sometimes it may still constrain
the environments in which a form may occur. The fact that, for instance,
while is still awkward for some speakers in contexts expressing anteriority of
200
one event to another shows that this conjunction has not entirely lost its
original meaning of temporal overlap:
(13) ?While our business was extremely successful last year, this year
does not look too promising.
(contrast although which is acceptable for all speakers in this context).
PDE examples such as those in (14) illustrate the fact that expressions
of simultaneity, concomitance, or correlation can synchronically be amplified
and interpreted as expressions of concessivity:
(14) a. He can play the Beethoven sonatas and he is only seven years
old
b. It is midnight and Mary is still working
It is difficult to find a method that is effective and, at the same
time, inexpensive
d. Not having any money, all the same I went into this expensive
restaurant
There are so many things going on simultaneously and there are so many
things cooccurring that mere cooccurrence or concomitance of two situations
(states, in particular), is rarely highly relevant information. Nevertheless, there
are some contexts in which concomitance may be highly relevant and worthy
of pointing out. One of these contexts is where there is a general incompatibil
ity between the two situations, i.e. where one situation does not normally
cooccur with the other. Thus in (14a) the situation of being seven years old
is assumed not normally to cooccur with the ability to play the Beethoven
sonatas. What may occur 'normally' is of course not always fully agreed on.
In the case of (14b), the speaker uses and to draw attention to coocurrence
and thereby invites the hearer to interpret the speaker as meaning that in the
speaker's view it is not normal for Mary (or people in general) to work when
it is midnight.
While originated in OE in an adverbial phrase translatable as 'at the
time that' consisting of the dative distal demonstrative, the dative noun hwile
'time', and the subordinator e, a highly explicit coding of simultaneity, cf.
(15) ChronA(Plummer) 913.3
& wicode pr pa hwile pe man pa burg
worhte
and camped there that time that one that fortress worked-on
& getimbrede
and built
'and camped there while the fortress was worked on and built'
201
This phrasal expression was reduced by late OE to the simple conjunction

wile:
(16) ChronE(Plummer) 1137.36
t lastede pa
[xix] wintre wile Stephne was king
that lasted those 19
winters while Stephen was king
In the process, the precise specification of simultaneity signalled by the
demonstrative was lost, allowing for other, less precise, inferences to play a
part. Among such inferences is that the conditions specified in the subordinate
clause serve not only as the temporal frame of reference for those in the main
clause, but also as the grounds of the situation (the disasters lasted nineteen
years because Stephen was king). Such an inference to grounds of the situation
is dominant over temporality in some examples dating from the later 14th
century:
(17) 1375 Barbour's Bruce 1.60 (OED)
Thar my cht succed na female, Quhill foundyn my cht be ony male
'No female was able to succeed while any male could be found.'
(cf. as long as, at the same time as, which also came to have concessive
inferences). In some languages, among them German, this inference to the
grounds of the situation, and thence to cause, has become the main extension
of 'while'. In English yet another augmentation led to the inference of surprise
overlap in time or surprise relations between event and ground, hence to the
adversative, concessive meaning. Probable instances of the coding of surprise
and hence concessivi ty appear in the early seventeenth century, cf.:
(18) 1617 Sir W. Mure Misc. Poems xxi.23 (OED)
Whill others aime at greatnes boght with blod, Not to bee great thou
stryves, bot to bee good
'While others aim at greatness that is bought with blood, you strive
to be not great but good.'
This could be interpreted as a statement about simultaneous behaviors.
However, there is a strong inference, reinforced by the inversion in the second
line, that it is unusual not to be blood-thirsty.
To sum up, although the formal semantic analysis of the concessive
presents many puzzles that have not been solved satisfactorily, it is intuitively
plausible to consider concessive relations of the type:
202
(19) a. although p, q
as presupposing a conditional:
(19) b. if p, then normally ~ q
Thus, for (14b) If it is midnight, then normally she would not still be working,
for (18) If others strive to be great, then normally you would not strive to be
(just) good, not great.
The meaning of general incompatibility makes the concessive meaning a
favored target for certain types of negative expressions. Before we turn to the
grammaticalization of negative terms to express concessivity, consider:
(20) a. He stays home although he is not sick
Here the negative concessive amounts to an assertion of the positive
(20) b. He stays home although he is well
because sick is a semantically negative adjective standing in a complementary
relationship to the positive well. In either case, the logical meaning is actually
If he stays home, then normally he isj should be sick/not well. (20a) is a more
oblique way than (20b) of expressing incompatibility because of the multiple
layers of negation it involves: although he is not sick = then normally he is not
(not sick).
In several Indo-European languages (and possibly other languages,
though we have not discovered any to date), we find that negated semantically
negative terms have been grammaticalized to concessives. Notwithstanding,
for example, involves the negative of 'against' + 'stand' (OE wi was the term
for opposition that retained its old meaning in a few fixed expressions like
this concessive and fight with the enemy in the sense of 'fight against the
enemy'). To say something is 'not opposing' amounts to asserting that it is
congruent, correlated, etc. Similarly, nonetheless nevertheless presumably
coded relationships of the kind:
(21) p, not the less q (despite the expectation that q is less valued)
(22) p, not the less frequently q (despite the expectation that less fre
quent is less valued)
What is not less (nor more), and not less frequent (nor more so), is equal in
quantity or frequency, therefore correlated. The various negative terms, there
fore, logically compute as members of the same semantic field as positive
203
terms like just/all the same. Such positive terms assert parity, but imply
expected disparity.
(23) p, in exactly the same way q (despite the expectation that q is
inversely proportional to p)
Whether the phrases are negative or affirmative, in each case the more
literal meaning has been augmented as the phrase has become grammaticalized. In other words, the more literal meaning has been specialized to a
limited set of adverbs or conjunctions that can express concessivity (cf.
Hopper on specialization, These volumes). In their earlier histories, when
they were fully lexical, they were members of relatively open sets where their
literal meanings prevailed. For example, in Middle English never the less
contrasted with phrasal adverbs like never the wiser/better/worse/more/nearer,
but unlike them became specialized as a clausal connector denying the antici
pated presupposition connected with the clause. An early example is:
(24) c1330R. Brunne Chron. 61 (OED)
Neverpeless to William he 3eld him wele his bone
'Nevertheless he fittingly granted William his prayer.'
Nonetheless (often spelled none the less) contrasted in its literal meaning with
none the more/wiser/worse/better right up until the nineteenth century, when
the concessive use arose (OED); all of these except the first survive in PDE,
but are not part of the same paradigm as nonetheless either in meaning or in
syntactic position. Likewise, all the same was a member of a lexical set
including all the more better richer, etc., but, along with just the same came
to take on a contrastive, concessive function in the nineteenth century. One
of the earliest examples cited in the OED is:
(25) 1845 Disraeli Sybil vi.iv (OED)
What you say is well worth attention; but all the same I
feel we are on the edge of a regular crisis.
3.3. From temporals to preference to denial markers
Another instance of a standard interpretive augmentation that has left
its imprint on the conventional meaning of related expressions in a variety
of languages is provided by the historical development of preference adverbs
and connectives like rather (than) from temporal adverbs, which may them-
204
selves have derived from spatial terms (cf. Dieterich and Napoli, 1982). Some
examples are:
(26) Eng. sooner (than), rather (than) ( < O E hrapor 'sooner'); Gm. ehe,
eher (als), bevor; Span, antes (<Lat. ante 'before'); Fr. plutt
(<plus tt 'more soon'); Russ. skoree 'more quickly, rather';
Ruman, mai curnd 'earlier, rather'; Finn. pikemmin (cf. pika 'at
once'), ennen 'before, rather'. In PDE the preferential reading of
soon (er) coexists with the temporal reading in PDE, cf.
(27) a. Bill died sooner than Mary (temporal)
b. Bill would sooner die than marry Mary (preference)
It is most frequently encountered in contexts of volitional will/would, and
invites the inference that q is something the subject does not want (to do).
Sometimes, however, a term for 'earlier' can be truly ambiguous with a
temporal (at least within the bounds of an intonational sentence). Quirk et
al. (1985:112 note b.) for example, cite:
(28) He'd sit alone in the dark before he'd watch television
One reading gives the temporal habitual (every evening he sat in the dark
before watching TV), and another the preference (he prefers sitting in the dark
to watching TV).
In the case of sooner, both the temporal and the preference sense date
back to early ME, when the comparative sooner came to be attested (there
are no examples of the comparative in OE, and only a few of sona, all in the
temporal sense). We can therefore not gain insight from sooner into the
development of the preference reading. However, when we turn to hrapor,
(the comparative of hrpe 'quickly, soon'), we can trace how 'sooner, more
quickly' came to mean 'preferably', via the inference that 'the sooner the
better'. The temporal meaning of rather essentially disappeared by the ENE
period. Rather than in PDE has a preference reading and, in addition, what
can be called a metalinguistic denial meaning (Thompson, 1972; Dieterich
and Napoli, 1982). Consider:
(29) a. He recites rather than sing
b. He recites rather than sings
(29a), with the non-finite verb sing illustrates the preference rather than.
As pointed out by Thompson (1972:242-3) the preference rather than "presup
poses that there is a preference on the part of the subject for the situation of
205
the main clause over that of the adverbial clause", thereby excluding such
locutions as:
(29) c. *It rains rather than snow
Furthermore, "the action of the first clause will render the second clause
unnecessary or impossible to carry out". Therefore (29a) implicates:
(29) d. He recites because he is unable to sing/so that he does not have
to sing
By contrast (29b), with the finite verb sings, is the metalinguistic rather than,
and expresses the speaker's preference for the formulation He recites and
denial of a presupposed or expressed He sings. In essence (29b) means
(29) e. I prefer to say/assert that he recites and deny that he sings
In French the contrast between the equivalent two connectives is coded
by the distinction not between non-finite and finite verb but between absence
or presence of the negative ne. The latter codes speaker's denial. Thus (30a)
parallels (29a) and (30b) parallels (29b):
(30) a. II recite plutt qu'il chante
b. Il recite plutt qu'il ne chante
A brief look at the development of rather than shows at least one path
by which such readings can arise from a temporal. In OE we find hrapor
meaning 'sooner, earlier':
(31)
AECHOM I 25 356-28
iohannes ws hraor mannum cu urh his mrlican
John
was sooner to-men known through his splendid
drohtnunga ponne crisi wre: for pan e he ne
conversations than Christ was: for that that he not
teowde his godcundan mihte r
pam e he ws
manifest his divine
might before that that he was
prittig geara on pre menniscnysse
thirty years in that human incarnation
'John (the Baptist) was known to men sooner than Christ through
his splendid preachings, because Christ did not reveal his divine
power before he was thirty years old.'
Sometimes hrapor is used in a context not of events but of states, and here
the adverb has the meaning 'more' (note the inference in (31) that John was
206
known more because he was known sooner). An example is the translation

provided by AElfric in his Grammar of the Latin magis 'more':
(32) AEGram 241.2
magis swypor: magis hoc uolo quam Ulud swyor oe
magis more: magis hoc volo quam illud more or
hraor ic wille pis ponne t
rather I want this than that
In modal contexts of wish, expectation, etc. when the comparand was present
hraper ponne was augmented to express preference, cf.
(33) AELS (Agnes) 301
wiste past seo dohtor e drihten hfde gecoren hraor
knew that that daughter who Lord had
chosen rather
wolde sweltan ponne ceorlian
wanted to-die than to-take-a-husband
'knew that his daughter, who had chosen the Lord, would rather
die than get married'
As is typical of preference rather than, there is a negative implicature, in this
case 'die so that she would not have to get married'. This kind of inference
of rejection of one alternative presumably allowed strengthening of the con
nective to express rejection of a metalinguistic alternative, particularly in
declarative contexts. In (34) the intentional force of the subject (we) is embed
ded in a complement to mater 'reason', and potential ambiguity between
preferred action by the subject and preferred choice of words can be inferred:
(34) c1380 Wyclif Sel. Wrks. I.409. (OED)
We have litil mater for to laughe, but rather for to morne
In essence this is ambiguous between: 'There is little reason for us to laugh;
we should prefer to mourn' and 'There is little reason for us to laugh; more
correctly speaking we should mourn'. Later we find unambiguous metalinguis
tic denial of one alternative with non-human subjects as in:
(35) 1611 Shakespeare, Tempest II.i.60
The rarity of it is.. .That our garments, being.. .drench'd in the sea,
hold notwithstanding their freshness and glosses, being rather newdy'd than stain'd with water.
The implicature that the clause introduced by rather than is unnecessary or
impossible to carry out has been augmented to express preference for the
207
formulation our garments were new-dy'd and denial that they were stained.
The metalinguistic preference reading has been conventionalized and in PDE
requires syntactically contrastive contexts such as (29a) and (29b).
4.
STRENGTHENING OF INFORMATIVENESS VS. METAPHORIC

PROCESSES
We have argued that certain kinds of grammaticalization are instances

of strengthening of informativeness, or the conventionalizing of conversa
tional inferences. We turn now to discussion of which kinds of grammaticali
zation in its early stages involve this kind of change.
So far we have not discussed examples of another kind of inference,
metaphor. Metaphor has traditionally been regarded as central to semantic
change in general. Although definitions of metaphor vary, most share certain
concepts in common, especially understanding and experiencing one kind of
thing in terms of another, and directionality of transfer from a basic, usually
concrete, meaning to one more abstract, cf. Sapir (1977), Lakoff and Johnson
(1980), Claudi and Heine (1986). Although metaphorization has traditionally
been recognized primarily in lexical change, recently many arguments have
been put forward that the development of grammaticalization is also strongly
motivated by metaphoric processes, cf. Sweetser (1988, 1990), Bybee and
Pagliuca (1985), Claudi and Heine (1986), Heine and Claudi (1986), Heine et
al. (These volumes). For example, Claudi and Heine say of the "vehicle of a
metaphor and the lexeme undergoing grammaticalization" that they:
.. .are governed by an arrangement of conceptualization.. .which is unidi
rectional and proceeds from concrete to abstract, and from concepts
which are close to human experience to those that are more difficult to
define in terms of human cognition. (1986:328)
Claudi and Heine discuss the development of body part terms into locatives,
of spatials into temporals, etc. in terms of conceptual metaphors such as
SPACE IS AN OBJECT, TIME IS SPACE.
Examples of spatio-temporal metaphors in the process of grammaticali
zation are widely known and will be only touched on here. They include the
use of GO for future (I'm going to go), COME for perfect (Fr. je viens de le
faire), BE AT/BE IN for progressive, cf. Traugott (1978), Fleischman (1982),
Bybee (1987), and of verbs of motion for case, cf. Givn (1975).
208
In each case, more concrete concepts come to serve as models for more
abstract ones and metaphor is clearly at work. What the changes just cited
have in common is one of three tendencies that have been identified for
semantic change in general, both lexical and grammatical (Traugott, 1989:34),
which can be expressed as follows:
(36) Semantic-pragmatic Tendency I:
Meanings based in the external described situation > meanings
based in the internal (evaluative/perceptual/cognitive) situation
By "internal" situation is meant the situation as perceived or understood by
a sentient being, not necessarily the speaker. For example, temporal relations
are "internal" in the sense that they have fewer physical correlates. The
extension of the originally spatial preposition fter to the temporal preposi
tion fter in Old English is an instance of Tendency I: A shift from reference
to a concrete, physical situation to reference to a cognitive, perceptual situa
tion. When the spatial term is itself derived from a body part, which is often
the case (cf. BEHIND), Tendency I may operate twice, once from
OBJECT > SPACE, and then again from SPACE > TIME.
Other well-known examples of metaphoric changes in the grammaticali
zation of spatial terms include the development of adverbs or prepositions
into clause connectives, for example of Old English prepositional fter
'following behind, later' to the Middle English subordinating after. These are
examples of a second tendency in semantic change, identified by Traugott
(1989:35) as
(37) Semantic-pragmatic Tendency II.
Meanings based in the described external
situation > meanings based in the textual situation
or
internal
where 'textual' means 'cohesive'. When fter became a temporal connective,

it underwent Tendency II and shifted to a marker of textual, cohesive rela
tions. The metaphor presumably at work here is that events expressed in
clauses are treated as objects with spatio-temporal correlates.
But is metaphor the right or the only process to adduce for all cases of
grammaticalization? Recently some nervousness about the universality of
metaphor in grammaticalization has been expressed. For example, Heine et
al. (These volumes) acknowledge that metaphor is not the only process
involved; and although Bybee and Pagliuca (1985) cite metaphor as the
process involved in the development of deontic from 'agent-oriented' modal
ities, e.g. must, have to, in Bybee (1987) metaphor is abandoned as an explana-
209
tion of the development the hypothetical uses of the modals could and would8
It is difficult to see in what sense a causal is an analog of a temporal, or a
concessive of a coocurrence relation. We have suggested that strengthening
of informativeness and conventionalizing of conversational inferences are the
prime processes at work in the development of causals, concessives, and
preference/denial connectives.
These three cases of grammaticalization involve not only Tendency II
but also the third tendency in semantic change identified in Traugott (1989:35):
(38) Semantic-pragmatic Tendency III
Meanings tend to become increasingly situated in the speaker's
subjective belief-state/attitude toward the situation
Causals, concessives and particles of denial are all essentially expressions of
speaker attitude to the relationship of elements within the proposition or of
propositions to each other, as well as of the compatibility of those relations.
Temporal sian, which is a textual marker, comes via Tendency III to express
the speaker's view of a causal relation between states of affairs. Similarly,
temporal while, also a textual marker, comes by Tendency III to express the
speaker's surprise at the relation between two propositions.
In the case of hrapor a temporal adverb came by Tendency I to have a
preference (evaluative) meaning; the textual connective hrapor panne, devel
oped by Tendency II, could express either a temporal (the older) or a prefer
ence (the newer) meaning. The shift from the preference to the denial
connective occurred through the operation of Tendency III. As a preference
marker, rather than connects clauses the subjects of which must be human
but not necessarily the speaker, cf. Benveniste's (1973) 'sujet d'nonc'. By
contrast, in the denial reading, the connective connects clauses the subjects
of which can be any noun, but the subject of the denial is the speaker
(Benveniste's 'sujet enonciation'). We may note further that the development
of the denial meaning is an example of shifts from non-epistemic to epistemic
meanings such as have been noted in Shepherd (1982), Bybee and Pagliuca
(1985), Hanson (1987), Traugott (1989). To give just one example, must in
the epistemic sense of 'I conclude that' derived from the obligative sense of
'ought to' by strengthening of conversational inferences and subjectification.
If I say She must be married in the obligation sense, I invite the inference that
she will indeed get married. This inference is of course epistemic, pertaining
to a state of affairs that is anticipated to be true at some later time. When
this epistemic inference is conventionalized its origin in the speaker's sub-
210
jective belief-state is strengthened as well (cf. also Abraham, These volumes,

on the development of epistemic meanings in German modal particles).
Strengthening of informativeness has no well-established place in the
taxonomy of semantic changes, which includes metaphor, metonymy, exten
sion, restriction, among others (Anttila, 1972). One approach would be to
add strengthening of informativeness to that taxonomy. However, to treat it
as something separate from the other processes is to miss some generalizations.
We suggest that strengthening of informativeness is in fact a type of
metonymy.
Metonymy is usually paired with metaphor as a type of semantic change.
However, it has not been assigned a status equivalent to metaphor. For
example, Dirven speaks of metaphor as a "major associative leap" but of
metonymy as a "minor process" (1985:98). In the tradition deriving from
Jakobson and Halle's (1956) classic distinction between metaphor as choice
functioning on the paradigmatic axis versus metonymy as association and
sequence functioning primarily on the syntagmatic axis, metaphor is thought
to lead to homogeneity and coherence, metonymy to juxtaposition and poten
tial incoherence (Sapir, 1977:4). Another view of the difference between
metaphor and metonymy that is more useful to us gives metonymy a more
important role. This is Anttila's suggestion that metaphor concerns semantic
transfer through a similarity of sense perceptions, and is therefore analogical
and iconic, while metonymy is semantic transfer through contiguity and is
therefore indexicai (1972:142). By "indexicai", Anttila means that metonymy
points to semantic relations in certain contexts.
Three types of context have been much discussed:
a.
b.
Contiguity in socio-physical or socio-cultural experience: e.g. (i)

Lat. coxa 'hip' > Fr. cuisse 'thigh' (the parts of the body are spa
tially contiguous in the physical world) (Ullmann 1964:218); (ii)
boor 'farmer' > 'crude person' (association of behavior with a cer
tain person or class of persons); (iii) Lat. lingua
'tongue'>'language' (association of activity with an enabling
factor) (Kronasser, 1952:29); (iv) concern 'interest, solicitous
regard' > 'matter that concerns' (association of a mental state with
its object or cause (Stern, 1968:376)),
Contiguity in the utterance (that is, collocation), often ending in
ellipsis, cf. painting by Picasso>a Picasso (Anttila, 1972:142);
French ne.. .pas >pas
c.
211
Synecdoche, or the part-whole relation, e.g. redbreast > 'robin'

(Ullmann, 1964:219), and most especially body part changes, cf.
Wilkins (1980) on FINGERNAIL > FINGER > HAND.
As these examples suggest, the contiguities and associations usually cited with
reference to metonymy tend to be concrete. The main exceptions have to do
with behavioral-judgmental associations (cf. boor). However, some examples
of metonymy have also been cited in the literature on semantic change. Most
pertinent to our discussion is Brinton's suggestion with regard to the develop
ment of English have as an auxiliary verb, that "Rather than bleaching, the
semantic change involved in the development of the perfect seems to have
been metonymic" (1988:102).
We propose extending the notion of metonymy from traditional concrete
and overt contexts to cognitive and covert contexts, specifically the pragmatic
contexts of conversational and conventional inference. The contiguity
involved is based in the discourse world. The 'indexing' involved is the
pointing to relevance that conversational inferences about stereotypical situa
tions entail. With regard to the development of the causal meaning of sippan,
the following hypothesis can be made: An originally conversational implicature arising in the context of communication of temporal sequence came to
be associated with sippan 'from the time that' and then came to be a conven
tional implicature pointing to or indexing cause, somewhat as assumptions
about the behavior of farmers came to be associated with boor. Similarly, an
originally conversational implicature that a marker of simultaneity would not
be used unless there was something remarkable about that simultaneity came
to index the surprise factor and thus the concessive. If this is an acceptable
analogy between pragmatic strengthening and metonymy, then it should be
noted that the development of conventional implicatures is a case of synecdo
che (part > whole).
The germ of the idea put forward here is to be found in Stern (1968
[1931]). Stern views permutation (his term for metonymy) as resulting from
"a word [being] used in a phrase where a notion in some way connected with
its meaning is liable to form an element of the context" (1968:353). Stern
rejects a hypothesis he attributes to Leumann that permutations result from
"a difference between the meaning intended by the speaker and that compre
hended by the hearer" on the grounds that the speaker must be assumed to
know his native language (1968:360). In other words, he rejects the view that
the change results from speaker's improper processing (or inadequate learn
ing). He suggests instead that permutations result from "striving to fulfill as
212
ELIZABETH CLOSS T R A U G O T T A N D E K K E H A R D K N I G
adequately as possible the symbolic and communicative functions of speech"

(1968:359); in other words, speaker's communicative intent and desire to be
fully expressive is central. He goes on to list under examples of permutation
the development of the logical meanings of considering, supposing, of Gm.
weil, and concessive while.
5.
CONCLUSION
In discussing the principle of exploiting old means for novel functions,

and the recruitment of concrete for more abstract terms, Heine et al. (These
volumes) suggest that:
...grammaticalization can be interpreted as the result of a process which
has problem-solving as its main goal, its primary function being conceptu
alization by expressing one thing in terms of another. This function is
not confined to grammaticalization, it is the main characteristic of meta
phor in general.
In other words, semantic change in general, not just grammaticalization, can
be interpreted as problem-solving. The authors identify one principal problem:
That of representing members of one semantic domain in terms of another,
in other words, metaphor. But in semantic change (including the process of
grammaticalization) there is a second problem: The search for ways to regulate
communication and negotiate speaker-hearer interaction. We have suggested
that this is a kind of metonymic change, indexing or pointing to meanings
that might otherwise be only covert. The main direction of both types of
problem-solving is toward specification. Metaphorical change involves speci
fying one, usually more complex, thing in terms of another not present in the
context. Metonymic change involves specifying one meaning in terms of
another that is present, even if only covertly, in the context. In the changes
we have discussed, the metonymie change is from less to more informative,
that is, in the direction of explicit coding of relevance and informativeness
that earlier was only covertly implied; in other words, like metaphor, it is a
case of pragmatic inferencing and strengthening, but along a different axis of
relations.
We have argued that semantic change in the early stages of grammaticali
zation does not necessarily involve bleaching; on the contrary, it usually
involves specification achieved through inferencing. Second, the inferencing
is of two kinds: Metaphor and metonymy, which, as would be expected, are
213
not totally inseparable, but which correlate with shifts to different types of
grammatical function. Metaphor is largely correlated with shifts from mean
ings situated in the external described situation to meanings situated in the
internal evaluative, perceptual, cognitive situation, and in the textual situa
tion. Metonymy is largely correlated with shifts to meanings situated in the
subjective belief-state or attitude toward the situation, including the linguistic
one. Note this formulation is couched in terms of preference rules (cf. Jacken
doff, 1983 with reference to rather different semantic domains). Therefore it
is not inconceivable that metonymy might operate in the domain typical of
metaphor or vice versa.9
An example of the difference can be highlighted by examination of the
development of the words prefer (ence) and rather. Prefer is an evaluative
verb derived (already in Latin) by Tendency I from the Latin spatial verb
prae-fere 'to bear before, carry in front' (prae 'in front, before', fere 'carry').
Here there is a metaphorical transfer from the domain of motion in space to
mental evaluation. A fairly concrete image schema (Sweetser, 1988) is involved
and the process is fairly obviously metaphorical. In the case of rather, an
image-schema involving transfer of 'sooner' to 'preferably' is harder to con
ceive, since the concepts are both somewhat abstract, and conceivably meton
ymy may have been at work even here. But in the case of the extension of
the .textual preference rather (by Tendency II) to the metalinguistic assertive
rather (by Tendency III), the notion of image-schema becomes totally prob
lematic. If there is mapping from preference for a situation to preference for
an expression, it is very indirect note the shift from syntactic subject to
speaker subject, among other things. Here indexing of speaker's belief-state
and spelling out of communicative relevance, rather than mapping, is at work.
In sum, while metaphor is correlated primarily with solving the problem
of representation, metonymy and conventionalizing of conversational mean
ings are associated with solving the problem of expressing speaker attitudes,
including metalinguistic overlays of meaning.
NOTES
1.
Elizabeth Traugott thanks Suzanne Kemmer and Stephen Levinson for comments on an
earlier version of this paper. Bernd Heine and Peter Hook raised issues that helped clarify
the arguments presented here. Of course, none of them are responsible in any way for
the content of this paper.
Parts of this paper are based on Knig and Traugott (1988) and Traugott (1988).
214
ELIZABETH CLOSS TRAUGOTT A N D E K K E H A R D K N I G

Ekkehard Knig gratefully acknowledges the financial support received from the
Deutsche Forschungsgemeinschaft.
2.
Sperber and Wilson (1986) treat metaphor under relevance; the relationship of inferences
of metaphorical and conversational relevance to the proposition in question is, however,
rather different, see Section 5.
3.
Fischer (Forthcoming), based on Denison (1985), suggests a pragmatically motivated

development of do via narrative completive constructions.
4.
Important to discussions of relevance and of strengthening of inferences is the current

debate whether there is only one kind of pragmatic inference, characterized as relevance
by Sperber and Wilson (1986), or two kinds, characterized as the mutually interactive
Q-Principle and R-Principle by Horn (1984; 1989). The Q-Principle is "make your contri
bution sufficient, say as much as you can" (given R), while the R-Principle is "make your
contribution necessary, say no more than you must" (given Q). Horn suggests that the
Q-Principle, which puts upper bounds on implicata and invites hearers to infer no more
and possibly less than they hear, blocks formations such as nall from not all, since nall
would be no more informative than some (1989:256). The R-Principle, which puts lower
bounds on implicata, and invites hearers to infer more than they hear, accounts, among
other things, for the development of modal auxiliaries from main verbs, and the develop
ment of speech act inferences (Horn, 1984:30-31). Since all examples of grammaticalization discussed here involve strengthening of relevance, this debate is not pursued here.
Its relevance for grammaticalization will be the focus of another paper.
5.
Abbreviations and approximate dates for stages of English are: Present Day English
(PDE) 1700 on; Early Modern English (ENE) . 1500-1700; Middle English (ME)
1100-1500; and Old English (OE) 600-1100.
6.
All examples from OE are cited in the form in Healey and Venezky (1980). Others
are from the Oxford English Dictionary (OED) and the Middle English Dictionary
(MED).
7.
There is no categorial contrast between wesan and beon in OE, but in general beon is
used to express future, generic, and stative 'be' (cf. Spanish ser), wesan to express
contingency, activity, and past (narrative) tense (cf. Spanish estar). The two verbs merged
in the suppletive paradigm of be during the ME period.
8.
Cf. also Frajzyngier (1986) on the non-usefulness of metaphorical explanations for the
development of switch reference. Frajzyngier suggests that the development of switch
reference is 'not a metaphorical extension and not even a syntactic reanalysis. It is rather
a logical conclusion about the utilization of certain functional properties of a morpheme
in a certain environment.' While switch reference may be a case of non-metaphorical
inference, it is problematic to think of the change as a response to new environments,
since this approach presupposes use of a word in a new environment where its usual
meaning would be inappopriate. We argue here that semantic-pragmatic change can occur
only when it is licensed by inferences from extant contexts.
9.
Heine et al. (These volumes) suggest that metonymy operates quite widely. The develop
ment of switch-reference discussed by Frajzyngier (1986) may be an example of metonymic
change in the development of a textual marker, cf. Ft. 8.
215
REFERENCES
Abraham, Werner. 1976. "Die Rolle von Trugschlssen in der Diachronie von Satzkonnektoren." In Opuscula Slavica et Linguistica, H.P. Pohl and N. Salnikow
(eds), 11-72. Festschrift fr A. Issatschenko. Klagenfurt.
Abraham, Werner. Volume IL These volumes. "The grammaticization of the German
modal particles."
Andersen, Henning. 1973. "Abductive and deductive change." Language 49:765-93.
Anttila, Raimo. 1972. An Introduction to Historical and Comparative Linguistics. New
York: Macmillan.
Atlas, Jay D. and Stephen C. Levinson. 1981. "/-clefts, informativeness, and logical
form." In Radical Pragmatics, Peter Cole (ed.), 1-61. New York: Academic Press.
Ben veniste, Emile. 1973. Indo-European Language and Society trans, by Elizabeth
Palmer. Coral Gables, Florida: University of Miami Press.
Blakemore, Diane. 1987. Semantic Constraints on Relevance. Oxford: Blackwell.
Braunmller, Kurt. 1978. "Remarks on the formation of conjunctions in germanic
languages." Nordic Journal of Linguistics 1:99-120.
Brinton, Laurel J. 1988. The Development of English Aspectual Systems. Cambridge:
Cambridge University Press.
Bybee, Joan. 1985. Morphology: A Study of the Relation Between Meaning and Form.
Bybee, Joan. 1987. "The grammaticization of tense and aspect." Paper presented at
the LS A Winter meeting, San Francisco, December.
Bybee, Joan and William Pagliuca. 1985. "Cross-linguistic comparison and the devel
opment of grammatical meaning." In Historical Semantics: Historical Word Forma
tion, Jacek Fisiak (ed.), 59-83. Berlin: de Gruyter.
Claudi, Ulrike, and Bernd Heine. 1986. "On the metaphorical base of grammar."
Studies in Language 10:297-335.
Denison, David. 1985. "The origins of periphrastic do: Ellegard and Visser reconsid
ered." In Papers from the Fourth International Conference on English Historical
Linguistics, Roger Eaton, Olga Fischer, Willem Koopman, and Frederike van der
Leek (eds), 45-60. Amsterdam: John Benjamins.
Dieterich, Thomas G. and Donna Jo Napoli. 1982. "Comparative rather." Journal of
Dirven, Ren. 1985. "Metaphor as a basic means for extending the lexicon." In The
Ubiquity of Metaphor, Wolf Paprott and Ren Dirven (eds), 85-119. Amsterdam:
John Benjamins.
Fischer, Olga. Forthcoming. "Middle English Syntax." In Cambridge History of
English Vol. II, Norman Blake (ed.). Cambridge: Cambridge University Press.
Fleischman, Suzanne. 1982. The Future in Thought and Language: Diachronic Evidence
from Romance. Cambridge: Cambridge University Press.
Frajzyngier, Zygmunt. 1986. "Grammaticization through analysis: A case of switch
reference." In Proceedings of the Second Annual Meeting of the Pacific Linguistics
Conference.
Geis, Michael L. and Arnold M. Zwicky. 1971. "On invited inferences." Linguistic
Inquiry 11:561-6.
Givn, Talmy. 1975. "Serial verbs and syntactic change: Niger-Congo." In Word
216
Order and Word Order Change, Charles N. Li (ed.). Austin: University of Texas
Press.
Givn, Talmy. 1989. "Modes of knowledge and modes of processing: The routinization
of behavior and information." Chap. 7 of Mind, Code and Context: Essays in
Pragmatics, 237-67.
Grice, H. Paul. 1975. "Logic and conversation." Syntax and Semantics III: Speech
Acts, Peter Cole and Jerry Morgan (eds), 41-58. New York: Academic Press.
Hanson, Kristin. 1987. "On subjectivity and the history of epistemic expressions."
Chicago Linguistic Society 23:132-47.
Healey, Antonette and Richard Venezky. 1980. A Microfiche Concordance to Old
English. The Dictionary of Old English Project, Centre for Medieval Studies,
University of Toronto.
Heine, Bernd and Ulrike Claudi. 1986. On the Rise of Grammatical Categories; some
Heine, Bernd, Ulrike Claudi, and Friederike Hnnemeyer. Volume I. "From cognition
to grammar Evidence from African Languages."
Hobbs, Jerry R., Mark Stickel, Paul Martin, and Douglas Edwards. 1988. "Inter
pretation as Abduction." MS.
Hopper, Paul J. Volume I. "On some principles of grammaticization."
Horn, Laurence R. 1984. "Toward a new taxonomy for pragmatic inference: Q-based
and R-based implicature." In Meaning, Form, and Use in Context: Linguistic
Applications, Deborah Schiffrin (ed.), 11-42. Georgetown University Round Table.
Washington: DC: Georgetown UP.
Horn, Lawrence R. 1989. A Natural History of Negation. Chicago: University of
Chicago Press.
Jackendoff, Ray. 1983. Semantics and Cognition. Cambridge, Mass.: MIT Press.
Jakobson, Roman and Morris Halle. 1956. Fundamentals of Language. The Hague:
Mouton.
Knig, Ekkehard. 1985. "Where do concessives come from? On the development of
concessive connectives." In Historical Semantics: Historical Word-formation, Jacek
Fisiak (ed.), 263-82. Berlin: De Gruyter.
Knig, Ekkehard. 1986. "Conditionals, concessive conditionals and concessives: Areas
of contrast, overlap and neutralization." In On Conditionals, Elizabeth Traugott,
Alice ter Meulen, Judith S. Reilly, and Charles A. Ferguson (eds), 229-46. Cam
Knig, Ekkehard. 1988. "Concessive connectives and concessive sentences: Crosslinguistic regularities and pragmatic principles." In Explaining Language Univer
sals, John Hawkins (ed.), 145-66. Oxford: Blackwell.
Knig, Ekkehard. Forthcoming. "Concessive relations as the dual counterpart of
causal relations." In Universal Semantics Semantic Universals, Dietmar Zaefferer
(ed.). Dordrecht: Foris.
Knig, Ekkehard and Elizabeth Traugott. 1988. "Pragmatic strengthening and
semantic change: The conventionalizing of conversational implicature." In Under
standing the Lexicon: Meaning, Sense and World Knowledge in Lexical Semantics,
Werner Hllen and Rainer Schulze (eds), 110-24. Tbingen: Max Niemeyer Verlag.
Kronasser, Heinz. 1952. Handbuch der Semiasiologie. Heidelberg: Carl Winter.
217
Lakoff, George and Mark Johnson. 1980. Metaphors we Live by. Chicago: Chicago
University Press.
change." Lingua e Stile 20:303-18.
Levinson, Stephen 1983. Pragmatics. Cambridge: Cambridge University Press.
Meillet, Antoine. 1948 [1912]. "L'volution des formes grammaticales." In Linguistique
gnrale et linguistique historique, 130-48. Paris: Champion.
Mitchell, Bruce. 1985. Old English Syntax. London: Oxford University Press, 2 Vols.
Peirce, Charles Sanders. 1955. Philosophical Works of Peirce, Justus Buchler (ed.).
New York: Dover.
Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik. 1985. A
Comprehensive Grammar of the English Language. London: Longman.
Romaine, Suzanne. 1982. Socio-historical Linguistics: Its Status and Methodology.
Cambridge: Cambridge University Press.
Sapir, J. David. 1977. "The anatomy of metaphor." In The Social Use of Metaphor:
Essays on the Anthropology of Rhetoric, J. David Sapir and J. Christopher Crocker
(eds). Philadelphia: University of Pennsylvania Press.
Shepherd, Susan C. 1982. "From deontic to epistemic: An analysis of modals in the
history of English, creoles, and language acquisition." In Papers from the Fifth
International Conference on Historical Linguistics, Anders Ahlqvist (ed.), 316-23.
Sperber, Dan and Deirdre Wilson. 1986. Relevance. Communication and Cognition.
Cambridge, Mass.: Harvard University Press.
Stern, Gustav. 1968 [1931]. Meaning and Change of Meaning: with Special Reference
to the English Language. Bloomington: Indiana University Press.
Sweetser, Eve. 1988. "Grammaticalization and semantic bleaching." In Berkeley Lin
guistics Society, Proceedings of the Fourteenth Annual Meeting, 389-405.
Sweetser, Eve. 1990. From Etymology to Pragmatics. Cambridge: Cambridge Univer
sity Press.
Talmy, Leonard. 1983. "How language structures space." In Spatial Orientation:
Theory, Research and Application, Herbert Pick and Linda Acredolo (eds), 225-82.
New York: Plenum Press.
Thompson, Sandra A. 1972. "Instead of and rather than clauses in English." Journal
of Linguistics 8:237-49.
Traugott, Elizabeth 1978. "On the expression of spatio-temporal relations in
language." In Universais of Human Language, Vol. III: Word Formation, Joseph
H. Greenberg, Charles A. Ferguson, and Edith Moravcsik (eds), 369-400. Stan
ford: Stanford University Press.
Traugott, Elizabeth 1982. "From propositional to textual and expressive meanings;
some semantic-pragmatic aspects of grammaticalization." In Perspectives on His
torical Linguistics, Winfred P. Lehmann and Yakov Malkiel (eds), 245-71. Amster
dam: John Benjamins.
Traugott, Elizabeth . 1985. "'Conventional' and 'Dead' metaphors revisited." In
The Ubiquity of Metaphor: Metaphor in Language and Thought, Wolf Paprott and
Ren Dirven (eds), 17-56. Amsterdam: John Benjamins.
Traugott, Elizabeth C. 1985b. "Conditional markers." In Iconicity in Syntax, John
Haiman (ed.), 289-307. Amsterdam: John Benjamins.
Traugott, Elizabeth 1986. "From polysemy to internal semantic reconstruction."
In Berkeley Linguistics Society, Proceedings of the Twelfth Annual Meeting, 539-50.
218
Traugott, Elizabeth . 1988. "Pragmatic strengthening and grammaticalization."

Berkeley Linguistics Society, Proceedings of the Fourteenth Annual Meeting, 406-16.
Traugott, Elizabeth 1989. "On the rise of epistemic meanings in English: An
example of subjectification in semantic change." Language 65:31-55.
Traugott, Elizabeth C. 1990. "From less to more situated in language: The unidirectionality of semantic change." In Papers from the Fifth International Conference
on English Historical Linguistics, Sylvia Adamson, Vivien Law, Nigel Vincent, and
Susan Wright (eds). Amsterdam: John Benjamins, 496-517.
Ullmann, Stephen. 1964. Semantics: An Introduction to the Science of Meaning. Oxford:
Blackwell.
Wilkins, David. 1980. Towards a Theory of Semantic Change. Ph.D. dissertation,
Australian National University (Forthcoming at SUNY Press).
The De Dicto Domain in Language

Zygmunt Frajzyngier
University of Colorado at Boulder
1. PURPOSE AND SCOPE OF THE PAPER

I take the notion of grammaticalization in both diachronic and syn
chronic sense. Although I am interested in how grammatical morphemes
emerged from individual lexical items, I am more interested in the question
of what semantic and pragmatic functions are encoded in the grammatical
system of language. The means of grammatical encoding, including the lexical
sources of grammaticalization, will emerge as a by-product of such an
approach. The purpose of the present paper is to show that languages make
a distinction between the domain of reality and the domain of speech and
also to show how this distinction is encoded. In the study of language the
domain of speech as opposed to the domain of reality has only been implicitly
invoked, mainly for the distinction between anaphoras and deictics, in both
formal and functional studies. However, certain notions to be discussed in
this paper have been raised in one form or another by other scholars, especially
in connection with the development of definite articles (Greenberg, 1985).
Traugott, 1980 makes a distinction between 'the world being talked about'
and 'the speaker's organization of that world in the act of speaking' (Traugott,
1980:47). In various papers one may also find scattered usage of the terms
'linguistic', 'metalinguistic', in reference to some antecedent of the anaphora,
but these terms have too wide a scope (cf. Dubois et al., 1973:317) to be of
use in the present paper. I propose that the distinction between the domain
of reality and the domain of speech is marked in at least two areas of language
structure: In the marking of embedded clauses and in the system of reference,
where languages differentiate between sets referring to the domain of reality
and sets referring to the domain of speech. Usually, the set referring to the
domain of reality encodes more distinctions than the set referring to the
domain of speech.
220
ZYGMUNT FRAJZYNGIER
Using the notion of the domain of speech, one can explain functional
syncretisms and provide an explanation for the similarity of various mor
phemes considered hitherto as unrelated. The explanation that I will provide
will account for several of the phenomena discussed by Anderson and Keenan,
1985 as a 'relativization of deixis'. Although I draw data and examples from
a variety of languages, I do not want to claim that the distinction is universal.
It is entirely possible that not all languages encode this distinction in their
grammatical systems.
For the sake of brevity I will refer to the domain of speech as the domain
de dicto and to the domain of the real world as the domain de re. The domain
de dicto in the present paper includes the hypothetical mood. Justification for
this inclusion will be provided later in the paper. The terms de dicto and de
re are used here with the meaning such expressions have in Latin, and
independently of various interpretations they have had since middle ages in
philosophical literature (cf. Hughes and Cresswell, 1968:183ff; Parsons, 1971;
and also Wright and Givn, 1987). I do not intend in this paper to reconcile
the differences between the philosophical and linguistic understanding of the
terms that I am going to propose, nor will I consider the points in which
philosophical and linguistic interpretations may overlap.
Arguments in support of the proposed distinction between the de dicto
and de re domain in language are discussed in two sections. In the first section
I provide an explanation for the development of demonstratives into comple
mentizers. One of the results of this explanation will be an answer to the
question why the forms of the definite article and the complementizer after
verbs of saying are similar in many unrelated languages. Note that the two
morphemes belong to different syntactic categories, one of them, the comple
mentizer, has no independent lexical meaning (at least according to the current
literature) and therefore, a metaphorical extension is not a good explanation
for the two morphemes having the same form. In this section I will also
provide an explanation for the relationship between complementizers and
relative clause markers. In the second section I discuss some of the properties
of the systems of reference with respect to a hypothetical mood. In particular
I will show that in some languages the system of reference with respect to
the de dicto domain has a reduced number of distinctions when compared
with the system of reference to the domain de re.
The following problems will be discussed in the paper: 1. An explanation
of how demonstratives become complementizers; 2. A connection between
the de dicto domain and the 'hypothetical' mood; 3. A speculation on the
THE DE DICTO DOMAIN IN LANGUAGE
221
relative clause markers; 4. Properties of the referential system in the de dicto

domain.
2.
FROM DEMONSTRATIVE TO COMPLEMENTIZER
2.1. What needs to be explained

It is a well-known fact that in many languages complementizers are
identical with or derived from demonstratives (cf. Noonan, 1985). Why a
demonstrative is used in the function of complementizer has not, however,
been adequately explained. Lockwood, 1968:222ff. proposes for Germanic
that the demonstrative was first the object of the verb 'say' in the main clause,
and later became a constituent of the complement clause, becoming thus a
complementizer (conjunction in his terminology). Thus there was an evolution
from: 'He said that: he will come' to 'He said that he will come'. As evidence
for his hypothesis Lockwood provides the following examples from Faroese:
(Traugott (in press) gives similar examples from Old English.)
(1)
eg sigi at hann kemur

'I say that he comes'
(2)
eg sigi ta: hann kemur

'I say that: he comes' (Lockwood, 1968:223)
Sentence (1) contains complementizer at, which derives from demonstra

tive tad, whose full form may be seen in sentence (2). I think that the weakness
of this explanation lies in the unmotivated form of the presumable source for
(1): It is not clear why there should be a demonstrative in the object position
in sentence (2).
2.2. Hypothesis
I would like to propose that the demonstrative is used as a comple
mentizer because it indicates that the following clause belongs to, or should
be interpreted as belonging to, the domain de dicto. The evidence for this
hypothesis will consist of the following:
Showing that the demonstrative has become an anaphora referring
to a proposition, a propositional anaphora, further in the present
paper.
222
ZYGMUNT FRAJZYNGIER
Showing that clauses following the demonstrative belong primarily

(in the historical sense) to the domain de dicto.
Showing how an essentially de dicto complementizer comes to mark
the hypothetical mood.
A case study of complementation in Mupun in which I will show
systematically the development of the function of anaphora into
complementizer.
2.3. Evidence
2.3.1. Demonstrative and propositional anaphora
A propositional anaphora refers to a proposition as a whole rather than
to only one of the components of a proposition. In English this function is
performed by the remote demonstrative 'that', as in the following examples,
discussed from a different point of view in Channon, 1980:
(3)
'We should have champagne and caviar at the party after CLS.
'That's (*It's) a good idea'.
'Fred doesn't want to go, and that's (*it's) the problem.' (Channon,
1980:107)
The pronoun 'it' may be used in de dicto reference as an anaphora

referring to an NP previously mentioned in speech, e.g.:
(4)
John bought a car last year. It proved to be a lemon.
The remote demonstrative 'that' has constraints that set it apart from
the pronoun 'it'. In particular 'that' is constrained in its use in reference to
an NP in the clause, but it is not so constrained in its use as a propositional
anaphora, e.g.:
(5)
John bought a car last year. That proved to be a disaster.

*That proved to be a lemon.
The remote demonstrative may be used in reference to a nearer of two

possible antecedents,1 e.g.:
(6)
John bought a car last year and another car just last week. THAT
proved to be a lemon.
THE DE DICTO D O M A I N IN L A N G U A G E
223
Virtually the same situation obtains in other IE languages, where the

neutral rather than non-neutral pronoun is used as propositional anaphora,
e.g.:
(7)
French:
Russian:
Polish:
F ne veut pas aller, mais a ne me derange pas.

F. ne xoet idti, no to (*tot/ta) menja ne volnuet.
NEG want go but that that(m)/(f) me NEG bother
Fred nie chce i ale to (*ten/*ta) mnie nie martwi.
Fred doesn't want to go, but that doesn't bother me.
2.3.2. Definite article

The purpose of this section is to claim that the definite article has a
function linked with the de dicto domain.
The primary function of the definite article remains somewhat controver
sial. The question is whether the definite article marks an NP as a known or
as mentioned previously in speech. Greenberg, 1985: 282 states: 'In the case
of the [definite] article, besides this anaphoric use, we have rather more
frequently reference to what is known from the situational context or general
knowledge than speech context.' While Greenberg has provided convincing
arguments to show that this claim is true in English, there are equally
convincing arguments to show that in some languages the definite article may
be used only when a noun has been previously mentioned in speech. The
evidence for this claim is provided by the following facts from Mupun. The
definite marker in Mupun is identical with the non-human anaphora. It
may occur after any noun, including proper names and toponyms. If it were
indeed true that the primary function of the definite is to mark a noun that
is known, then we would expect the definite to be used with the first occurrence
of a toponym in the text, at least with those toponyms that are significant
for a given culture, hence assumed to be known to the speakers and hearers
alike. And yet this is not the case. When a toponym occurs for the first time
in a text, it does not have a definite article, e.g.:
(8) yaks n-yit
vom di
del n-siam
then lsG-leave Vom there pass lsG-descend
Then I left Vom and went down to Jos.'
n-jos
PREP-Jos
After the toponym was mentioned in conversation its subsequent mention

in the text may be followed by a definite article, e.g. in a sentence that in the
text is separated from the preceding one by five sentences:
224
ZYGMUNT FRAJZYNGIER
(9)
to,
lokaci dd an n-jos
well t i m e REL 1 SG PREP-JOS DEF
'Well, when I was in Jos.'

Compare also the following example, with the toponym 'Rome', without
the definite article:
(10) to, n-bit (1975) be n-ddm n-rom
well PREP-year
CONS lSG-go PREP-Rome
, in 1975 I went to Rome.'
Four sentences later, the toponym 'Zaria' has a definite article because
it has been previously mentioned in speech:
(11) n-dm -kes
be
wa
ba n-dm di
n-Zaria
lSG-go lsG-finish COMP return go lsG-go there PREP-Zaria DEF

went, finished [the school] and then returned to Zaria.'
Several sentences after somebody was characterized as German, the
following sentence was produced:
(12) yak s dm di n-Germany
then
1PL
go
t h e r e PREP-G.
DEF
'Then we went to Germany.'

Compare also the following sentences. In the first, three toponyms are
introduced into the discourse, and none of them has a definite article. When a
reference is later made to those toponyms they all occur with definite articles:
(13) pak mo dm Dortmund, pak mo dm Essen, pak
some PL go
some PL go
some
mo kuma Wrzburg.
PL also
'Some went to Dortmund, some went to Essen, some went to
Wrzburg.'
(14) to, wen an meme
n-but
pak nen d
well, 1SG 1SG some PREP-inside some people REL 3PL
ddm Wrzburg nd
gO
DEF
'I was among those who went to Wrzburg.'

(15) wur kd to n-Essen n ji
COMPL stay PREP-Essen DEF come
'He came from Essen.'
225
For at least one language, it has been shown that being mentioned
previously rather than previous knowledge is the necessary condition for the
use of the definite article. In this language the article thus marks the noun
that belongs to the de dicto domain. I do not have necessary cross language
information to assert that in all languages that have a definite article, it will
be used to mark a noun mentioned previously in speech, and not necessarily
a noun that is known, but not mentioned in the preceding discourse.
There are languages in which there are two definite articles: One to mark
a noun that is known and the other to mark a noun that has been mentioned
in speech. Thus in Fering ( a North-Frisian language) we have the following
paradigm:
a maan
di maan
'the man'
'the man'
at wuf
di wuf
'the woman'
'the woman'
The article a refers to a generally known referent, and the article di refers
to a referent mentioned in a previous or following context (Karen Ebert, p.c.).
2.3.2. Complements of the verba dicendi
The fact that elements introduced by complementizers after verbs of
saying belong to the domain of speech is true analytically, i.e. it follows from
the fact that the complementizers follow verbs of saying. Additional evidence
is provided by the fact that if a language has a complementizer derived from
a demonstrative, such a complementizer will mark the complements of the
verbs of saying and thinking, while the sentential complements of other verbs
may be marked by other complementizers or by other means. The comple
mentizer derived from demonstratives does not carry a modal function when
occurring after verba dicendi. In Slavic languages the simple form of comple
mentizer (cto in Russian, e in Polish) after verba dicendi does not carry a
deontic modality. If deontic modality is intended after verba dicendi or other
verbs, the complementizer occurs with the particle by, e.g. to-b(y) in Russian
and e- in Polish. Although in many languages the complementizer used
to introduce complements of the verba dicendi is also used to introduce other
complements, there seems to be little doubt that its primary function was,
synchronically and diachronically, to introduce complements of the verba
dicendi.
The third part of the evidence is provided by the many languages in
which complementizers are derived from verba dicendi, such as Hausa cewa
'saying', Yoruba pe, G ake, and other African languages (cf. Lord, 1976),
226
ZYGMUNT FRAJZYNGIER
English-based creoles (cf. Frajzyngier, 1984). The importance of this argument

rests on the fact that it is the verbs of saying rather than some other verbs
that were chosen to serve as complementizers. The following sample of
languages with complementizers derived from verbs of saying is far from
complete, but the fact that the same process has occurred in different families
indicates that it is by no means accidental: Kwa languages (Niger-Congo),
Hausa (Chadic), Amharic (Semitic), Dargua, Tabasara, Xinalug (Deeriev,
1959) (Caucasian).
As admittedly fragmentary data show, complementizers derived from
demonstratives occur primarily after verba dicendi to indicate that what
follows belongs to a de dicto category.
2.4. Complements of verba dicendi as complements of other verbs
It is a fact that demonstratives do not only serve as complementizers
after verbs of saying but also after other verbs. There is, however, a difference
in the function of the complementizer, partially depehding on the verb of the
main clause. When the complementizer derived from demonstratives occurs
after a verb of saying, it does not carry a modal function. When it occurs
after other verbs, it does carry a modal function. Kirsner and Thompson,
1976 noted that after verbs of perception, the complementizer 'that' indicates
an indirect evidence, e.g.:
(16) I saw that he was sleeping
cf. I saw him sleeping
Here is an additional illustration of this fact from Polish:
(17) Powiedziaem
spa
Say-1SG.M.-PAST COMP Sleep-3SG.M.-PAST
said that he slept.'

(18) Widziaem
spa
See-1SG.M.-PAST COMP sleep-3SG.M.-PAST
saw that he slept.'

(19) Widziaem
jak spa
see-lsG.M.-PAST how sleep-3sG.M.-PAST
'I saw him sleeping.'
Although Kirsner and Thompson have observed the function of 'that'
they did not explain why it rather than some other complementizer is used
227
to mark the indirect evidence. I would like to propose that the verba dicendi
complementizer is used to indicate inference because of the strong sense that
information obtained through speech is not as reliable as information ob
tained through direct observation. After verba dicendi the complementizer
does not carry any information about the epistemic value of the complement
clause. The complementizer is associated, however, with verbs of saying,
which intrinsically have epistemic value, such that the information obtained
through hearsay is less reliable than the information obtained through direct
perception (see, hear, feel). The complementizer acquires thus the same episte
mic value as the verb that automatically triggers complementizer's presence,
i.e. the verb 'say'. The complementizer is now available to serve a function
of indicating less than direct evidence with other verbs, including the verbs
of perception.
The association between the domain of speech and the hypothetical
mood is quite widespread across languages. Verbs of saying are used to
indicate less than complete evidence (Frajzyngier, 1985b). In Czech 'doubt in
truth' modality is marked by the morpheme pr which derives historically
from verb prav 'he says'; in Slovak the marker of 'doubt in truth' is vraj,
historically derived from the verb vrav 'he says' (cf. Stieber, 1979: 247ff), e.g.:
(20) On vraj bol v Bratislave
he
was in Bratislava
'Apparently he was in Bratislava.' (Stieber, 1979: 248)
cf.
On bol v Bratislave
'He was in Bratislava.'
Finally, the association between domain of speech and hypothetical is
evinced by the fact that in a number of languages an introduction of a
hypothetical situation is realized periphrastically by preceding the hypotheti
cal proposition with a phrase equivalent to English 'Let's say: ...'. Further
in the paper I will be treating hypothetical mood as an instance of the de
dicto domain with respect to its semantic properties.
2.5. A case study: Complementizer in Mupun
The evidence for the function of demonstrative as a de dicto marker for
NPs and clauses was based so far on fragmentary information taken haphaz
ardly from a variety of languages. The purpose of the present section is to
228
ZYGMUNT FRAJZYNGIER
provide evidence based on a systematic analysis of relevant phenomena in

one language.
2.5.1. The f orm of the complementizer
I will be concerned here only with one complementizer , which is
phonologically identical with the non-human anaphora and with the definite
article discussed earlier. Unlike similar morphemes in other languages,
does not have in contemporary language a deictic function. Thus one cannot
use while pointing at something either close or remote. Its function is thus
limited to an NP or a propositional anaphora. As the anaphora for nonhuman NPs, can function as subject, object, and prepositional phrase
complement, e.g.: (antecedent bold, anaphora underlined)
Subject:
(21) ama yil
dd get -yam

kd
kwance
but world REL PAST PREP-time DEF ANAPHASSOCquiet
rai met
d
n-yak
s
life surpass REL PREP-time DEM
'the world of the past had a much quieter life than the present
world.'
Object and prepositional phrase:
(22) sm bature
mo an mb tu n-ba
name European PL 1SG FUT kill ANAPH 1SG-throw away
nd
PREP ANAPH
'European names, I would eliminate them, throw them away.'

Prepositional phrase:
(23) jir aap mo yak am
mo loom
ji
kd nd
girls PL fetch water 3PL approach come PREP ANAPH
'girls fetch water and they approach with it...'
Recall from a previous discussion that the anaphoric nd also functions
as the definite marker, e.g.:
(24) n-naa jiel wur dafuan nd
lsG-see pity hare DEF
'I pity him, the hare.'
229
2.5.2. Anaphora as complementizer

The anaphora occurs as a complementizer after verba dicendi sat 'to
say', tal 'to ask', and also after verbs denoting mental activitv, equivalents of
'know', 'think', 'remember', 'recall'. The complementizer carries neither
an epistemic nor a deontic function after these verbs. The deontic value of
the complement clause depends to a large degree on the verb of the main
clause, e.g.:
(25) wu sat
-nas
mo
say COMP lsG-beat 3PL
'He said that I beat them.' or

'He said that I should beat them.'
(26) mo sat
de
pa se
3PL say COMP CONS 3F eat
'They said that she should eat it.'

(27) wa daar
de man pa
3F scream COMP 3FL know 3FL
'She screamed that she knows her.'
(28) wu suun di naa pa
dream 3ML see 3FL
'He dreamed that he saw her.'
(29) wu ben
wu pan
an
think COMP remember lSG
'He thinks he remembers me.'
(30) -pan
man wur
lSG-think COMP know
'I think that I know him.'
(31) -man
know
ba
wur baa
n-mapun kas
COMP NEG PERF r e t u r n PREP-NEC
'I know that he didn't return to M.'

(32) -kak
a sm n-Mupun-
lSG-think COMP COP name PREP-M.-Q
'I think it is a name in Mupun, isn't it?'
230
ZYGMUNT FRAJZYNGIER
The fact that has no modal function after verba dicendi provides
evidence that its primary function within a given language is to mark the
complement as belonging to the de dicto domain. The primacy of this function
is further reinforced by the fact that, unlike in Indo-European languages, the
interrogative sentences with the main verb tal 'to ask' have the same comple
mentizer as indicative sentences with the verb sat 'to say'. And the reason
they have the same complementizer is that both are verbs of saying. The
modality of the complement clause after the verb tal is marked by one of
several clause final interrogative markers, but it may also be marked by an
additional modal complementizer kt 'whether' (not illustrated here), e.g.:
(33) mo tal
an
-man
pa-e
3PL ask PREP 1SG COMP lsG-know 3FL-Q
They asked me whether I know her.'

(34) a man ta -kat
me
gwar si
2 know COMP fall lsG-meet QUANT man QUANT PREP

swup as fin-on
wash dog 3M-Q
'Can you imagine that I came across a man washing his dog?'
(35) mo mb tal
mis
niwa siak -
3PL FUT ask PREP man DEF COMP 3PL.L together-Q
'They will ask the husband: are you together?'

Additional evidence for the function of as a marker of a de dicto
complement is provided in the next section.
2.5.3. Complementizer as the sole marker of de dicto

Complementizer , without any modal function, can also occur in
sentences whose main verbs cannot be characterized as verba dicendi. The
importance of this fact for the proposed hypothesis is the following: If it is
indeed true that marks the complement clause as being a de dicto category,
how does one explain sentences with modality neutral complementizer but
without a verb of saying in the main clause. I would like to propose that
such sentences, rather than weakening the hypothesis, provide additional
support for it. Consider the following sentences:
231
(36) n-sin
takarda -ra
a la a tay
PREP-2M COMP 2 take 2 read
lsG-give book
'I gave you a book to read.'
(37) kad krismas
kes e
wu ji
when 1PL make Christmas finish CONS 3M come ASSOC QUANT

jarida -an
n-la
n-ta
paper PREP-1SG COMP lsG-take 1SG read
'When we finished celebrating Christmas he brought me a newspa
per to read.'
These sentences were interpreted as containing, in the embedded clause,
an underlying verb sat 'to say'. In order to check the native speakers' intuition
with respect to these sentences, I asked for and obtained the following
sentence:
(38) wa cet lua -an
-se
3F cook meat PREP-ISG COMP lsG-eat
'She cooked meat for me and told me to eat.'
This sentence cannot be used if there is no actual contact between the
person who cooked the meat and the speaker. Thus it cannot be used felici
tously if the speaker returns home, finds the meat, but the cook is not there.
Sentence (38) is therefore not an equivalent of 'she cooked meat for me to
eat' because the felicity of the English clause does not depend on the partici
pants' being in the same place at the same time. Without the complementizer
, there is no requirement of direct contact between the subject of the main
clause and the subject of the embedded clause, and the following sentence is
grammatical in the situation in which the previous one is ungrammatical:
(39) wacet lua -an
n-se
3F cook meat PREP-ISG lsG-eat
'She cooked meat for me and I ate it.'
Thus the complementizer serves as the only marker of the de dicto
category. Additional evidence for this claim comes from sentences of the
structure: NP S, i.e. sentences without any verb in the main clause. The
'missing' verb is sat 'say'. In elicited sentences the verb sat is most often
present. In conversations, however, the verb sat is most often omitted, e.g.:
232
ZYGMUNT
FRAJZYNGIER
Elicited:
(40) n-sat
lsG-say COMP yes
'I said yes', agreed.'

(41) wu sat
say yes
'He said yes', 'he agreed.'
(42) a sat
n-wur
2M PREP- COMP yes
'Tell him yes.'

Conversations:
(43) wur
'He said yes', 'he agreed.'
(44) npuun fen
father 1SG COMP
'My father said that...'

(45) an
niwa ma can nua ji
-mun
baa
ISG COMP 3PL take hoe 3PL come PREP-1PL come
'I told them to take a hoe and come to join us.'

(46) an
gwar ta me
me mbi n-an
1SG COMP 3ML find QUANT thing
PREP-1SG
'I told him that he should find something for me.'

The verb sat 'to say' may of course appear in natural conversations, e.g.:
(47) wur
sat d du
bal
kawey
PREP say COMP go
3PL.L marry anyhow
'He will tell them to go and marry anyhow.'

From the syntactic point of view, the sentences with the missing main
clause verb are interesting because they cannot be analyzed as cases of simple
deletion of the phonetic form of the verb. The pronominal subject in the
'missing' verb clauses does not occur in the preverbal form but rather in the
independent form, i.e. the form in which it occurs following prepositions,
verbs, and as subject in equational sentences. Sentences without the main
clause verb of saying provide one more piece of evidence that the compie-
233
mentizer is used as an independent marker of the de dicto category. The

presence of the complementizer is not evoked by the main clause verb.
An additional piece of evidence that marks a de dicto category is
provided by the following consideration. The verb dm means 'want' and also
'agree'. This polysemy is known to occur in other African languages (Bernd
Heine, p.c.). When this verb is followed by the complementizer it means
'agree'. When dm is followed by the deontic complementizer the verb
means 'want', e.g.:
(48) -dem
wacet mbise n-darap

lSG-agree COMP 3F cook food PREP-Darap
'I agree that she cook for Darap.'
cf.
(49) -dem
-war cet mbise di

n-darap
lSG-want COMP PREP-3F cook food Anaph PREP-Darap
'I want her to cook for Darap.'
2.5.4. Epistemic modality after verba dicendi
Recall that the complementizer does not have a modal function when
it occurs after verbs of saying. An epistemic modality after verba dicendi is
marked through the use of the marker paa, which indicates 'doubt in truth'.
The marker is derived most probably from the verb paa 'to think'. The modal
paa has all the characteristics of an auxiliary verb: The first and second person
pronouns follow paa and precede other verbs in the clause, while third person
pronoun precedes paa.
The modal paa can not be used with the first person subject of the main
clause because the speaker cannot doubt the truth of his own proposition (cf.
Frajzyngier, 1985b), e.g.:
(50) a sat
wu paa mb yo muan
2 say COMP
FUT go trip
'You said that he will go on a trip (but he may or may not go).'
(51) -sat
wur mba yo muan
lsG-say 2MCOMP
'I told you that he is going on a trip.'

(52) *n-sat wur paa mbd yo muan
for 'I told you that he is going on a trip but I don't think so.'
234
ZYGMUNT FRAJZYNGIER
If the subject of the embedded clause is first person, then the proposition
of the embedded clause marked by the complementizer paa is interpreted as
false, e.g.:
(53) wu sat paa can
say COMP COP
1PL
1 PL cut goat DEF
'He said that it was we who slaughtered the goat (but we didn't).'
cf.
(54) wu sat din paa di kankuri
'He said that he was patient (but I have my doubts).'
(55) wu sat paa yi-can
say COMP COP 2F 2F- CUT goat DEF
'He said that it was you who slaughtered the goat (but I doubt it).'
More evidence that the modal paa indicates doubt in truth is provided
by the fact that it cannot be used after the main clause verb yen 'to think',
e.g.:
(56) *n-yen
wu paa siwa tbaa
lsG-think
drink tobacco
for 'I thought that he smoked tobacco.'
The explanation for the ungrammaticality of the above sentence rests in
the fact that the verb yen 'think' intrinsically indicates doubt in truth. The
construction -yen -S 'I think that S' in Mupun as in English is used as a
device to mark doubt in truth. It is entirely possible, that using two devices
whose primary function is to mark the same semantic category is considered
superfluous and therefore sentences such as (54) are ungrammatical.
2.5.5. De dicto and the hypothetical: Complements of verbs of perception
After verbs of perception, such as na 'see', kl 'hear', the comple
mentizer indicates that the evidence for the event is indirect, that the event
itself has not been observed, e.g.:
(57) n-naa
wu ta n-yil
lsG-see COMP fall PREP-ground
'I saw that he fell down.'
THE DE DICTO DOMAIN IN L A N G U A G E
(58) n-naa wu psiwa

COMP
235
tbaa
PREP smoke tobacco
saw that he smoked tobacco.'

(59) wu kl n-ta
-yil
hear COMP lSG-fall PREP-ground
'He heard that I fell down.'
Similarly to English (cf. Kirsner and Thompson 1976), the difference
between sentences without and sentences with a complementizer is that the
former indicate direct evidence, while for the latter the evidence is indirect.
Thus the verb kl 'hear' indicates only hearsay when followed by the comple
mentizer .
(60) wu kl n-taa n-yil
1 SG-fall
PREP-ground
'He heard me fall down.'
(61) wu kldrj n-taa itn-yil
'He heard that I fell down.'
An indication of the function of after verbs of perception is provided
by spontaneous translation of the verb naa 'to see' evoked by the presence
of the complementizer, viz.:
(62) n-naa a can
lsG-see 2M slaughter goat

'I saw you slaughter a goat.'
(63) nnaa a can
'I thought that you slaughtered a goat.'
An additional piece of evidence for the function of is provided by
essentially negative information: The doubt in truth marker paa cannot be
used as a complementizer with the verbs of perception, hence this functional
gap is filled by the complementizer , e.g.:
(64) *n-naa wu paa ta n-yil
lsG-see
fall PREP-ground
for 'I saw that he fell down.'
(65) *n-naa wu paa siwa tdbaa
PREP
drink tobacco
for 'I saw that he smoked tobacco.'
236
ZYGMUNT FRAJZYNGIER
(66) *wu kl paa n-t -yil

hear lSG-fall PREP-ground
for 'He heard that I fell down.'
(67) *war kl wur paa can
3F
hear
slaughter goat
for 'She heard that he has slaughtered a goat.'
We see therefore a systematic relationship between the verbs of the main
clause, the function of the complementizer, and the modality of the embedded
clause. The complementizer occurring after verba dicendi has no modal
function. The epistemic modality after such verbs is marked by the modal paa.
When the complementizer occurs after other verbs it acquires the function
of epistemic modality marker. Modal paa cannot be used after non verba
dicendi. I believe that a similar development of the function of the comple
mentizer derived from demonstrative occurred in other languages as well.
3.
RELATIVE CLAUSES
There is a partial or complete overlap in the forms that function as

complementizers, definite markers, and relative clause markers. In many
languages the forms of these markers are similar to remote demonstratives
or neutral pronouns, as the following data from a selection of languages from
various families indicate:
Table 1.
Language
Demonstrative
Definite
Complement
Relative
German
Yiddish
English
Ge'ez
Amharic
Beja
Mupun
Ewe
Toba Batak
Yurok
K'ekchi
Drehu
das/die/der
das/die/der
that
za
ya
un/tn
si/sia
na
the < that

0
u/w
/t
a/l
na
ku/k'i
li
la
dass
vos
that
za
ya/ala
0
be/bna
na
das/die/der
vos
that/wh
za
ya
/t
la
s
ku/k'i
li
la-ka
237
It is tempting to use the notion of de dicto domain as a common element

that would explain the formal similarity among the three grammatical cate
gories. What follows is then a speculation on the possible connection between
the relative clause formation in some languages and the demonstrative, defi
nite, and complementizers after verba dicendi.
I would like to consider an analysis whereby, in some languages relative
clauses have been conceptualized as a comment interjected into another
proposition. With respect to the proposition into which it is interjected, the
relative clause is a de dicto category. Hence there is the option of marking
by the same marker that is used as the complementizer or by the same marker
that marks the definite, i.e. by markers that elsewhere mark other de dicto
categories in the language. If we assume that the relative clause is marked as
an interjected proposition with respect to the sentence in which it is embedded,
we can explain why it is that the demonstratives occur as relative clause
markers predominantly in post-nominal relative clauses (cf. Keenan, 1985).
In such sentences, the main proposition has already begun, and it is
interrupted by the relative clause, which is essentially a comment on one of
the components of the main proposition or a modification of one of the
components. In pre-nominal relative clauses, the main proposition will usually
start after the relative clause has ended, hence there is no interjection, and
thus no motivation to treat any element of the utterance as being more
remote, de dicto, than the other. In languages in which both pre-nominal and
post-nominal relative clauses occur, the post-nominal relative clauses must
have the demonstrative preceding them, while the pre-nominal clauses may
or may not have the demonstrative, e.g.:
(68) der Mann, der in seinem Bro arbeitet
ART man who in his
study works
T h e man who is working in his study.'
(69) der in seinem Bro arbeitende Mann
ART
T h e man who is working in his study.' (Keenan, 1985:144)2

(70) ein in seinem Bro arbeitender Mann
ART(INDEF)
'A man who is working in his study.'

(71) poika joka tanssi p o y d l a oli sairas
boy REL dance on-table was sick
T h e boy who danced on the table was sick.'
238
ZYGMUNT FRAJZYNGIER
(72) poydalla tanssinut poika oli sairas

T h e boy who danced on the table was sick.' (Finnish, Karlsson,
1972:106 if.)
Relative clauses in Amharic, which are introduced by the morpheme ya-,
may serve as possible counter-evidence for the lack of demonstratives in pr
nommai relative. The morpheme ya- which at least historically is derived
from a demonstrative precedes the prenominai relative clause:
(73) ya-matta saw
REL came man
'A/the man who came.' (Cohen, 1936:115)
Since this morpheme is also a marker of the genitive construction, the relative
clause in Amharic may be seen as a variant of the genitive construction, with
the relative clause being a syntactic equivalent of the 'possessor' phrase, e.g.:
(74) ya-sum baqlo
chief mule
'A/the mule of the chief.'
(75) ya-baqlo dgdr
foot
T h e foot of the mule.' (Cohen, 1936:78)
Many Tibeto-Burman languages also have genitive constructions used
to form relative clauses (Scott De Lancey, p.c.).
4.
DE DICTO REFERENCE SYSTEMS
4.1. Two properties

The existence of two sets of pronouns, one of which refers only to the
de re objects and the other only to the de dicto objects, is the most obvious
evidence for the validity of the de re-de dicto distinction. The best known
examples of the existence of two sets of pronouns are in languages with
logophoric systems. Logophoric systems, whether identified as such or not,
have been described with various degrees of completeness in Jungraithmayr,
1963/64; Cloarec-Heiss, 1969; Burquest, 1973; Hagge, 1974, who coined the
term; Clements, 1975; Ebert, 1979; Hyman and Comrie, 1981; Stanley, 1982;
239
and Frajzyngier, 1985. As the term was originally used by Hagege, it desig
nated pronouns occurring in the embedded clauses whose referent was the
subject of a verbum dicendi in the main clause, e.g.:
(76) wu sat
di
nas an
say COMP 3MLog beat ISG
'He! said that he! beat me.'
(77) wu sat wu nas an
3M
'he! said that he 2 beat me.' (Mupun)
A less well-known phenomenon is the existence of a formal distinction
between the addressee of a current discourse (a de re addressee) and the
addressee of a reported discourse (a de dicto addressee), e.g.:
(78) ca, peemu ta kayu
laa mu mijiba
say 2P
FUT drive-out man REL stranger
'He said that you (reported speech addressee) will drive the stranger
away.'
(79) ca, ka (2) ta kayu laa mijiba
'He said that you(m.) are going to drive the stranger away.' (Pero,
Frajzyngier, 1985)
Since such systems have been described in the references listed, I will
take their existence as a proof of the distinction between de dicto and de re
reference and not dwell on it anymore.
4.2. Locative anaphora
In some languages there exists a formal distinction between demonstra
tives referring to a de re location and demonstratives referring to a location
mentioned in speech, e.g.:
(80) wu wa
di
come home there (Anaph.)
'He came back from there.'
(81) wu wa
sd
return there (Deictic) (Mupun)
'He came back from there.'
240
ZYGMUNT FRAJZYNGIER
The de dicto demonstratve refers only to a locative argument mentioned

in speech. The deictic demonstrative refers only to a location that can be
pointed to. Note that (81) with a deictic demonstrative is ungrammatical if
referring to a location mentioned in speech. Similarly (80) is ungrammatical
if accompanied by some kind of a gesture pointing to a place (for a description
of this type of anaphora see Frajzyngier, 1989).
In many languages the locative anaphora is derived from a remote
demonstrative:
(82) Put the book here
Put the book there
If you took the book from a box put it back there /*here
(83) Polo etu knigu syuda
put
this book here
'Put this book here.'
(84) Poloi etu knigu tuda
there
Tut this book there.'
(85) Esli ty vzjal knigu iz
jaika tak poloi eo obratno tuda/
if 2 take book from box
then put
it back
there
*sjuda
here (Russian)
'If you took a book from the box put it back there.'
(86) Mets cet livre ici
Mets cet livre l
Si tu as pris le livre d'une boite remettre le l. (French)
The use of the remote rather than the proximate deictic in the anaphoric
function is not accidental, as shown by the sample in Table 2:
The fact that in many languages the anaphoric locative is derived from
a remote deictic can be explained by metaphoric extension, i.e. in the way in
which Greenberg, 1985 explained emergence of the discourse deixis.
4.3. Formal distinctions in the de dicto referential systems
In the discussion to follow I consider that the de dicto category and the
hypothetical mood overlap semantically in that neither of them refers to a
domain of reality. As has been pointed out already in many languages, the
241
Table 2. Locative demonstratives
English
German
French
Spanish
Russian
Polish
Hausa
Proximate
Remote
Anaphoric
here
hier
ici
aqui
zdes'
sjuda
tutaj
nan
there
da
l
alla, alli/ahi
tam
tuda
tam
can
there
da
l
alli/ahi
tarn (Stative)
tuda (Movement)
tam
can
hypothetical mood is encoded as a de dicto category. It appears that the

number of formal distinctions encoded in the derepreference system is different
from the number of formal distinctions encoded in the de dicto reference
system. Typically the de dicto reference system will have fewer distinctions
than the de re reference system. In the following sections I will consider
distinctions made within the systems addressee pronouns, interrogative pro
nouns, and quantifiers.
43.1. Addressees in the de dicto and de re domains
In many languages the gender of the addressee is formally encoded, e.g.:
(87) kogo
widziae
whom-ACC see-PAST-2M
'Who did you (m.) see?'
(88) kogo
widziaa
who-ACC See-PAST-2F
'Whom did you (f.) see?' (Polish)

In a hypothetical discourse, the addressee is always second person mascu
line, as in the following fragment from Szymborska's 'Pisanie yciorysu'
where the poet talks to a hypothetical addressee:
(89) Pisz
tak jakby ze
Write-IMPER-SG SO a s if
sob nigdy nie
With REFL n e v e r NEG.
rozmawia
i
omija
z
daleka
talk-2sG-M-PAST and avoid-2sG-M-PAST from afar
'Write as if you have never talked with yourself, and avoided
yourself from afar.'
242
ZYGMUNT FRAJZYNGIER
The sentence as it stands indicates an addressee that could be either

masculine or feminine. If Szymborska were to use verbs in the feminine rather
than masculine form, the addressee would be interpreted as a concrete woman
whose identity happens to be unknown rather than a hypothetical addressee
that may be a man or a woman. Compare the following sentences built on
the constituents of the line from Szymborska:
(90) z
kim
rozmawiae
a
kogo
omijae
with whom talk-PAST 2M-SG CONJ who-ACC avoid-PAST-2M-SG
z
daleka
from afar
'With whom did you (m.) talk and whom did you (m.) avoid from
afar?' (may be interpreted as having a de dicto or a de re addressee)
(91) z
kim
rozmawiaa
a
kogo
omijaa
with whom talk-PAST 2F-SG CONJ who-ACC avoid-PAST-2F-SG
z
daleka
from afar
'With whom did you (f.) talk and whom did you (f.) avoid from
afar?' (can be interpreted as having only a de re addressee)
Striking evidence for the distinction between a de dicto reference system
and a de re reference system exists in Mupun. The language has a formal
distinction between the second person masculine and second person feminine
singular addressee of an actual speech event, e.g.:
(92)
dem a
mi
2 PREP want COP what
'What do you (m.) want?'
(93)
dem
want COP what
'What do you (f.) want?'
2F PREP
In a hypothetical event this distinction does not exist, and instead there
is only one form for the addressee, second person masculine. In the next four
examples a female speaker talks about obviously female participants in hypo
thetical events, and yet she refers to those participants by second person
masculine a/ka rather than by second person feminine yi, which she would
have used were she to talk in the situation with real, rather than hypothetical,
addressees of the propositions involved:
243
(94) gaskiya, get kadan ka

ba d mo
truly
PAST if
2 with pregnancy NEG PAST 3PL PROGR
bal dik -ka
marry PREP-2M
'Truly, in the past if you were pregnant they wouldn't marry you.'
(95) mo yap wur fua mo we mo sei mo man
3PL check breast 2 PL thing PL till 3PL know COMP
ka ak da s mo bal
n
ka
2 lack, pregnancy before they marry PREP 2M
'Before they marry you, they will check your breasts, etc., till they
are satisfied that you are not pregnant.'
(96) kat naan get sin anjirem
-
mndon dak se i
if God PAST give placenta ANAPH PREP-2M one
only
'If God gave you only one placenta...'
(97) a da sian
sa da
2 go remove PREP DEM like that
'You go and get an abortion just like that.'
These sentences are not grammatical errors, slips of the tongue, or some
other 'exceptional' utterances. Their analyses with other native speakers,
confirmed every single one of them, not only as grammatical, but actually as
the only possible variant in the context of the conversation.
It appears that in a hypothetical event there is also no number distinction.
In the following sentence, the topicalized subject is marked for plural, and it
is inherently feminine. And yet the pronominal form preceding the verb is a,
the second person masculine singular subject pronoun:
(98) jirap d -yak
s
moa vwan sar vwan daa
girls REL PREP-time DEM PL 2 wash hand wash dish
'The girls of nowadays, just wash their hands, wash dishes.' (the
modern girls do not want to do anything dirty)
4.3.2. Interrogative pronouns

The distinction between de re and de dicto reference systems can explain
at least one phenomenon that has not been explained so far. It is well known
that interrogative pronouns in many languages trigger only one verb
244
ZYGMUNT FRAJZYNGIER
'agreement' while non-interrogative pronouns may trigger more than one

agreement, e.g.:
(99) Kto to kupil
who it buy-PAST-3M-SG
'Who bought it?'
cf.
*Kto to kupia
who it buy-PAST-3F-SG (Polish)
The non-interrogative usage of the 'interrogative' pronoun has the same
properties as the interrogative usage, viz., there is only one system of
agreement, e.g.:
(100) Biae plamy jak by kto
wapnem
white stains as if somebody whitewash-INSTR
pochlapatl
splash-PAST-3M-SG
'White stains, as if somebody splashed whitewash all over.' (Polish)
cf.
* Biae plamy jak by kto
wapnem
white stains as if somebody whitewash-iNSTR
pochlapaa
splash-PAST-3F-SG
If one accepts the distinction between de dicto and de re reference, the
explanation of why the interrogative should have no gender distinction
becomes straightforward: kto 'who' and co 'what' are not identified until the
answer is provided. They are thus not de re but rather de dicto elements.
Therefore they have a reduced system of distinctions like other de dicto
elements in Polish. If, however, an interrogative is used with respect to a
specific group of people, i.e. when we know at least one element of the
possible answer, then instead of the general interrogative the so-called relative
pronouns are used. Thus if one were to address a group of boys and suspected
that one of them bought the given object, one would say:
(101) ktry
245
kupi
w h o (-SG.) buy-PAST-M-SG
'Which one bought (it)?'

Addressing a group of girls in similar circumstances one would say:
(102) ktra
kupila
Who (F-SG) buy-PAST-F-SG
'Which one bought (it)?'

The gender distinction is also made in plural pronouns referring to the
actual participants, viz. ktrzy (M.PL) and ktre (F.PL). No number distinc
tion is made in interrogative pronouns when the answer is completely
unknown.
Invoking the notion of de dicto domain we can perhaps explain the often
noted but left unexplained syncretism of the interrogative and indefinite
pronouns, such as in English clauses:
(103) Watch what you eat
Watch who you talk to
This syncretism is known in many Indoeuropean languages (ex. 100
above in Polish) and was already a feature of Proto-Indoeuropean (cf. Meillet,
1937). But it occurs also in other languages, e.g. in Hungarian, Uto-Aztecan,
Chadic, to name just a few of the unrelated families in which the indefinite
and interrogative pronouns have an identical form. The indefinite pronouns
refer to some person, thing, place, etc. whose further identity is unknown,
hence very much an argument belonging to a hypothetical domain, the same
domain to which interrogative pronouns belong. The identical coding of the
two interrogative and the indefinite pronouns in so many languages suggests
that rather than having two different sets, we have only one set, which is used
in different syntactic environments, but always in the same function. This
conclusion is especially evident where specific questions are encoded by other
means along with the interrogative pronouns (cf. Frajzyngier, 1985c).
4.3.3. Plural verb in an interrogative clause

There is a verbal prefix po- in Polish. With many verbs this prefix
indicates plurality of object and plurality of action, e.g.:
246
ZYGMUNT FRAJZYNGIER
(104) Piotr otworzy

okno
Peter --past window
Teter opened the window.'
(105) Piotr pootwiera
okna
Peter PL-open-PAST-3M-SG window-PL
'Peter opened up the windows.'
In interrogative sentences the plural form of the verb is felicitous only
as an echo question. Thus the following sentence would not be felicitous
unless it is preceded by some proposition in which the verbs have the plural
form, such as 'X opened up windows, Y opened up doors':
(106) Co Piotr pootwiera?
what Peter PL-open-3M-SG
'What did Peter open?'
The analogous question with a singular verb, however, is felicitous in
situations in which the plural form is not:
(107) Co Piotr otworzy
Open-PAST-3M-SG
'What did Peter open?'

A possible explanation of the different felicity conditions for the singular
and plural form of the verb in interrogative sentences lies in presuppositions
one has to make before asking the question. The important fact is that for a
speaker to use the plural form of the verb in an interrogative sentence, he
has to assume that the object is plural. The speaker does not have to make
any assumptions about the number of the object in order to use the singular
form of the verb. The link between interrogative and hypothetical (i.e. de
dicto) is that in the interrogative we refer to an object whose characteristics
are not as fully known as in a de re reference.
4.3.4. Quantifiers and de dicto referents

In some languages quantifiers make a distinction between masculine and
feminine forms. Thus in Polish an appropriately coded quantifier has to be
used with masculine and feminine nouns, or when addressing somebody, e.g.:
247
(108) kaid-y
chopiec
every (M) boy
'Every boy.'
kad-a
dziewczyna
every (F) girl
'every girl.'
(109) kaid-y
z
was
i
kad-a z
was
each () PREP you(PL) CONJ each (F) PREP you(PL)
'Every one of you () and every one of you (F).'
In a reference to hypothetical persons only the masculine form is used,
e.g.
(110) dla kadego
co
innego
for everybody-DAT something different
'Something different for everybody.'
A similar situation obtains in German. Ulrike Claudi has drawn my
attention to the following perfectly grammatical sentence once used on a
tampon package. The quantifier in this sentence is in the masculine form:
(111) Die Menstruation ist bei jedem
etwas
anders
DEF menstruation is for everybody (M) a little bit different
'Everybody has a slightly different menstruation.' (Pusch, 1984)
Yet if one were to address a group of girls, one would use the feminine form
of the quantifier, jeder.
4.3.5. Gender in the de dicto domain in English
I would like to claim that the de dicto-de re distinction is/was also
encoded in the reference system of English and that, in particular, for many
speakers of English the system of reference to the de dicto domain is reduced
when compared to the system of reference to the de re domain. Let us take
a typical sentence from contemporary academic English (all underlining
mine).
(112) "The reader will find in this superb volume almost anything he/
she might wish to know about English grammar and its rules."
(Reference Book Review)
248
ZYGMUNT FRAJZYNGIER
And now two typical sentences from academic English of some fifty
years ago:
(113) ". . . and in both cases I have given references, in the Notes and
Bibliography, which will enable the reader to look into things, and,
if he chooses, to arrive at an opinion of his own." (Bloomfield,
1933:VII)
(114) "The ultimate success of the worker in Africa, whether he be trader,
settler, missionary, anthropologist, educator, or administrator,
depends to a large extent upon his language performance." (Ward,
1937:6).
As examples from Bloomfield and Ward indicate, the system operated and
for many speakers still operates in the following way: If a reference is made
to a [ + human] noun from a de dicto domain, and that noun is not inherently
feminine, the form used is the third person pronoun 'he'. It is important that
the reference is made to an element from the de dicto domain rather than to
an element that one could consider to be a 'generic'. In recent times we have
seen a change, at least in some varieties of English, whereby a reference is
made as in the quote from Reference Book Review (ex.112). Whatever was
the social motivation for this change, it reflects an analysis that did not take
into- account the de dicto-de re distinction. The sporadic acceptance of this
change reflects a small triumph of social pressure over language structure.
5.
CONCLUSIONS
I have shown that de dicto-de re distinction may be usefully applied to

explain several apparently unrelated phenomena in language structure. Invok
ing this distinction, one can provide a motivated explanation of why definite
articles, complementizers after verba dicendi, and postnominal relative clauses
have similar markers in a number of unrelated languages. Lack of demonstra
tives as markers of prenominai relative clauses follows naturally from the
proposed hypothesis. I have provided an explanation for the use of demonstra
tive pronouns as complementizers after verba dicendi, and their use as infer
ence-only markers after verbs of perception. The explanation implies that a
rather abstract notion of reference to an element from the domain of speech
emerged independently in a number of unrelated languages. I have shown
that in a number of languages there exist two sets of pronouns, one referring
249
to de re elements and the other to de dicto elements. The set referring to the
de dicto domain does not encode so many distinctions as the set referring to
the domain de re.
ABBREVIATIONS
ANAPH
ART
CONS
F
FL
M
ML
PL.L
anaphora
article
consecutive
feminine
feminine logophoric
masculine
masculine logophoric
plural logophoric
ACKNOWLEDGEMENTS
Work on this paper was supported by the NSF Grant BNS-84 18923. I am grateful
to Dwight Bolinger, Ulrike Claudi, Karen Ebert, Bernd Heine, John Koontz, and
Elizabeth C. Traugott for the close reading of a previous version of this paper and
for their insightful comments and additional examples and references.
NOTES
1.
2.
The following analysis and example is due to Dwight Bolinger in private correspondence.
Although Keenan, op cit. considers sentence (69) to be an instance of the relative clause,
this analysis is not shared by other linguists (Ulrike Claudi and Bernd Heine, p.c.). The
force of my argument here hinges not on the status of this sentence but on the fact that
it represents a pre-nominal rather than a post-nominal modification of a noun by a clause.
REFERENCES
Almkvist, Herman. 1881-1883-1885. Die Bis char i-Sprache. Tu-Bedawie in NordostAfrika. Uppsala: K. Geselschaft der Wissenschaften zu Uppsala.
Anderson, Stephen R., and Edward L. Keenan. 1985. "Deixis." In Shopen 1985,
3:259-308.
Bloomfield, Leonard. 1933. Language. New York: Holt.
250
ZYGMUNT FRAJZYNGIER
Burquest, Donald. 1973. A Grammar of Angas. Ph.D. Dissertation, University of

California at Los Angeles.
Channon, Robert. 1980. "Anaphoric that: A friend in need". In Papers from the
Parasession on Pronouns and Anaphora, Jody Kreiman and Almerindo E. Ojeda
(eds), 98-109. Chicago: Chicago Linguistic Society.
Clements, George Nicholas. 1975. "The logophoric pronoun in Ewe: Its role in
discourse." Journal of West African Languages 10:141-177.
Cloarec-Heiss, F. 1969. Les modalits personelles dans quelques langues oubanguiennes
(Discours direct-Discours indirect). Lige: SELAF.
Cohen, Marcel. 1936. Trait de langue amharique. Paris: Institut d'ethnologie.
Deeriev, Jurij D. 1959. Grammatika xinalugskogo jazyka. Moscow: Akademja Nauk.
Dubois, Jean, Math Giacomo, Lois Guespin, Christian Marcellesi, Jean-Baptiste
Marcellesi and Jean-Pierre Mevel. 1973. Dictionnaire de linguistique. Paris:
Larousse.
Ebert, Karen H. 1979. Sprache und Tradition der Kera (Tschad). Teil III: Grammatik.
Berlin: Reimer.
Frajzyngier, Zygmunt. 1984. "On the origin of say and se as complementizers in Black
English and English-based creoles." American Speech 59(3):207-210.
Frajzyngier, Zygmunt. 1985a. "Logophoric systems in Chadic." Journal of African
Languages and Linguistics 7:23-37.
Frajzyngier, Zygmunt. 1985b. "Truth and the indicative sentence." Studies in Language
9:243-254.
Frajzyngier, Zygmunt. 1985c. "Interrogative sentences in Chadic." Journal of West
African Languages 15(1):57-72.
Frajzyngier, Zygmunt. 1987. "From verb to anaphora." Lingua 72:15-28.
Frajzyngier, Zygmunt. 1989. "Three kinds of anaphora." In Current Progress in
African Linguistics, Isabelle Haik and Laurice Tuller (eds). Amsterdam: Foris.
Greenberg, Joseph. 1985. "Some iconic relationships among place, time, and discourse
deixis." In Haiman 1985:271-287.
Hagge, Claude. 1974. "Les pronoms logophoriques." Bulletin de la Socit de Lin
guistique de Paris 69:287-310.
Haiman, John. 1985. Iconicity in Syntax. Amsterdam: John Benjamins.
Hughes, G.E. and Cresswell, M.J. 1968. An Introduction to Modal Logic. London:
Methuen.
Hyman, Larry M. and Bernard Comrie. 1981. "Logophoric reference in Gokana."
Journal of African Languages and Linguistics 3:19-37.
Karlsson, Fred. 1972. "Relative clauses in Finnish." In The Chicago Which Hunt.
Chicago Linguistic Society, Paul M. Peranteau, Judith N. Levi and Gloria
Phares (eds), 106-114.
Keenan, Edward L. 1985. "Relative clauses." In Shopen 1985:141-170.
Kirsner, Robert, and Sandra A. Thompson. 1976. "The role of pragmatic inference
in semantics: A study of sensory verb complements in English." Glossa
10(2):200-240.
Lockwood, W.B. 1968. Historical German Syntax. Oxford: The Clarendon Press.
Lord, Carol. 1976. "Evidence for syntactic reanalysis: from verb to complementizer
in Kwa." In Papers from the Parasession on Diachronic Syntax. Chicago: Chicago
Linguistic Society, 179-191.
Lowenstamm, Jean. 1977. "Relative clauses in Yiddish: A case for movement."
Linguistic Analysis 3(3):197-216.
THE DE DICTO DOMAIN IN L A N G U A G E
251
Meillet, Antoine. 1937. Introduction l'tude comparative des langues indoeuropennes. Paris: Hachette.
Moyse-Faurie, Claire. 1983. Le drehu. Paris: SELAF.
Noonan, Michael. 1985. "Complementation." In Shopen 2:42-140.
Percival, Keith W. 1981. A Grammar of the Urbanised Toba-Batak of Medan. Pacific
Linguistics, Series B. Nr. 76.
Pusch, L.F. 1984. Das Deutsche als Mnnersprache. Frankfurt.
Shopen, Timothy, (ed.). 1985. Language Typology and Syntactic Description. Cam
Stanley, C. 1982. "Direct and reported speech in Tikar narrative texts." Studies in
African Linguistics 13:31-52.
Stieber, Zdzisaw. 1979. Zarys gramatyki porwnawczej jeykw stlowiaskich.
Warszawa: Pastwowe Wydawnictwo Naukowe.
Szymborska, Wistlawa. 1986. Ludzie na mocie. Warszawa: Czytelnik.
Traugott, Elizabeth Closs. 1980. "Meaning-change in the development of grammatical
markers." Language Sciences 2(1):44-61.
ings: Some semantic-pragmatic aspects of grammaticalization." In Perspectives on
Historical Linguistics, Winfred P. Lehmann and Yakov Malkiel (eds), 245-271.
Traugott, Elizabeth Closs. (In press). "Old English syntax." In The Cambridge history
of English, Vol. 1, Richard Hogg (ed.). Cambridge: Cambridge University Press.
Ward, Ida C. 1937. Practical Suggestions for the Learning of an African Language in
the Field. Oxford: Oxford University Press.
Westermann, Diedrich. 1952. A study of the Ewe language. Oxford: Oxford University
Press.
Wright, S. and T. Givn. 1987. "The pragmatics of indefinite reference: Quantified
text-based studies." Studies in Language 11(1):1-33.
The Grammaticalization of Rhetorical Questions

in Tamil
Susan Herring
California State University, San
1.
Bernardino
INTRODUCTION*
One of the foremost tenets in diachronic grammaticalization theory today

is the notion of umidirectionality, according to which change in meaning from
less to more grammatical is viewed as a linear and irreversible process. Defined
thus in general terms, unidirectionality might appear to be a simple description
of, rather than a claim regarding the nature of, the grammaticalization
process. However inasmuch as the term is frequently extended to characterize
other features of the grammaticalization process which might or might not
hold true for all instances in all languages, the use of the label 'unidirection
ality' makes a claim which must be demonstrated according to its own merits
for each feature proposed.
For example, grammaticalization is widely claimed to be unidirectional
with respect to increasing degree of abstraction; that is, language users tend
to refer to abstract, less accessible concepts in terms of more familiar, concrete
ones, and this tendency is one of the factors which motivates the linguistic
encoding of new concepts in old forms (cf. Heine, Claudi, & Hnnemeyer,
This volume). Such a claim is well-motivated cognitively, and is supported
by synchronic and diachronic evidence from a number of languages.
It has also been claimed that grammaticalization involves semantic
bleaching, or weakening of lexical meaning, and that this process, too, is
unidirectional. The validity of this claim is, however, far less evident than
that for increasing abstraction, since not all meaning change necessarily
involves bleaching, and in numerous instances, as is demonstrated by Traugott
and Knig (This volume), the exact opposite process, that of semantic
strengthening, may also take place. A more comprehensive view would seem
254
SUSAN . H E R R I N G
to be that weakening and strengthening are independent processes, either or

both of which may potentially be a factor in grammaticalization in any given
instance.
Finally, Traugott (1982; 1988) and Traugott & Knig (This volume)
have claimed that grammaticalization is also unidirectional in the sense of
increasing subjectification or 'speaker involvement'; that is, the extension of
meanings encoded by a given lexical or grammatical item is predictably away
from objective, referential meaning towards subjective, "speaker-based" atti
tudes and points of view. Drawing on a model proposed by Halliday and
Hasan (1976), Traugott (1982) identifies three functional-semantic compo
nents which she labels propositional, textual, and expressive. Her prediction
is that when grammaticalization involves a meaning-shift from one functionalsemantic component to the other, it will proceed from the propositional (via,
optionally, the textual) to the expressive level, but not in the reverse direction.
Extensions of this type are accounted for in terms of (among others) the
inclusion within a form's "meaning" of some pragmatic nuance which was
formerly implicated by context alone. An example cited by Traugott is the
diachronic extension of English 'while' from a noun meaning 'period' or
'time' to a temporal connective on the textual level to, ultimately, a concessive
conjunction (in the sense of 'although') which encodes the attitude of the
speaker towards the proposition (Traugott, 1982:254).
What is not clear is that the process of grammaticalization as a whole is
necessarily unidirectional with respect to this characterization, nor indeed,
why we might expect that it should be. To begin, the claim that the historical
trend in language change has been towards increasing grammatical expression
of subjective and pragmatic meanings would seem to imply that at some
remote earlier stage such meanings were not expressed, or were expressed
with a far lower frequency. This would be an awkward position to maintain,
it seems, even aside from the problems inherent in attempting to demonstrate
it empirically. Second, the claim that grammaticalization necessarily has its
roots in the propositional (i.e. local, or lexical) level is at odds with the
findings of a growing body of researchers that point to the discourse-prag
matic origins of a number of grammatical elements. Thus Givn (1979) has
argued in favor of the diachronic discourse basis of morphological relativizers,
causativizers, and other types of clause subordinators in a variety of lan
guages; Hopper (1979, 1982, for Literary Malay) and Herring (1988, for
Tamil) have pointed out the possibility of discourse/pragmatic focus elements
developing into markers of perfective aspect; and DuBois (1987) attributes
THE GRAMMATICALIZATION OF RHETORICAL QUESTIONS IN TAMIL
255
the grammaticalization of ergative case-marking systems to functional con

straints on the rate and nature of information flow in narrative.
The implications of such claims on the unidirectionality hypothesis are
clear: If grammaticalization is defined as the process whereby grammatical
elements (be they morphological or syntactic) come into being, then if there
is any validity at all to the claims of the proponents of grammar ex-discourse,
the grammaticalization process cannot be unidirectional in the sense proposed
by Traugott. Rather the evidence suggests that autonomous grammatical
elements may arise either out of individual lexical or already existing grammat
ical elements on the propositional level, or they may take as their source the
larger discourse/pragmatic context.1 Indeed, it is my assertion that, as in the
case of semantic bleaching mentioned above, the two processes are essentially
independent, and that they are free to interact. Thus, it is theoretically possible
that a single grammatical word or affix may have undergone extensions in
both directions at different times over the course of its development.
If this position is correct, then important questions immediately arise
regarding the nature of the two processes. Do they apply to comparable data
cross-linguistically, or do certain grammatical meanings exhibit an affinity
for one source component or the other? Do individual languages show a
preference for one strategy over the other; that is, are some languages more
'discourse-grammaticalizing', and others more 'lexical-grammaticalizing', or
do both kinds of processes operate in all languages to a similar extent? As
for the specific mechanisms which drive these processes, a number of mecha
nisms which account for meaning change in isolated lexical items have been
discussed in the literature; e.g. analogy, bleaching, strengthening, metaphor,
systems of inference, and so forth. The mechanisms involved in the grammati
calization of discourse functions, however, have yet to be systematically
identified and described. Will we find the same, or similar, motivating factors
at work, or will it be necessary to postulate other mechanisms which derive
not from considerations of meaning but rather from considerations of com
municative function? If the latter turns out to be the case, we must confront
a methodological issue as well: Since "hard" diachronic evidence of communi
cative function especially in the very earliest stages of grammaticaliza
tion is by the very nature of things often excluded, can the process(es)
involved be inferred on the basis of synchronic evidence alone? Or should we
attempt to devise other methods for getting at discourse-functional factors in
language change? I do not propose to address all of these questions, or even
any one of them exhaustively, in the present paper. I do hope, however, to
shed some, light on the nature of the specific mechanisms involved in the
256
grammaticahzation of discourse functions. I propose to do this by examining,

in depth, a particular instance of discourse/pragmatics-based grammaticahza
tion: That of the grammaticahzation of rhetorical questions as markers of
clausal subordination in Tamil.
A word about the organization of the remainder of this paper may be
useful here. In considering the Tamil evidence, I begin by identifying three
rhetorical question types. I then trace what I hypothesize to be the diachronic
evolution of these types, from their pragmatic (interactive and expressive)
origins (Section 3 and Section 4), to their reanalysis as markers of textual
cohesion (Section 4), and ultimately to their grammaticahzation as clausal
morphology (Section 5). This expository sequence is accompanied by a pro
gressive narrowing of focus: Only two out of the original three rhetorical
question types participate in the second and third stages of the above process,
and within those two types, a subset of the actual linguistic forms involved
have attained fully grammaticalized status. By presenting the material in a
sequence iconic with its (hypothesized) historical development, I hope to
illustrate the naturalness of the processes involved, and to hint at the
contextual richness and perhaps too, at a certain random element, leading
to the grammaticahzation of some forms but not of others which shaped
the phenomenon as a whole. For those who prefer the omniscience of hind
sight, however, a more focused reading is also possible. Two forms for which
clear evidence of grammaticahzation will be adduced are the causal conjuction
, and the relativizer -, derived from a WH- word ( 'why') and a tag
question marker, respectively. In the general overview of Tamil rhetorical
questions presented below, the individual stories of these two forms can also
be traced. The final section (Section 6) considers the implications of the Tamil
evidence for the unidirectional hypothesis and grammaticahzation theory.
2.
RHETORICAL QUESTIONS: THE THREE TAMIL TYPES
The data on which this study is based are from Tamil, a Dravidian
language with SOV word order and suffixing, agglutinative morphology.
Except where indicated otherwise, all of the examples presented here are
drawn from a corpus of thirty-five oral narratives which I recorded in Tamil
Nadu, India, in 1986-87. These include personal narratives as well as folk
tales of various types, as related by both professional storytellers and ordinary
adult native speakers. Of special note are two lengthy public performances
257
in the Villu Pttu (lit. 'Bow Song') style, which provide evidence for the
interactive basis of rhetorical questions in traditional narrative.
A striking feature of Tamil oral narration (and to a lesser extent, of
written narration as well) is the frequent use of rhetorical questions. A
rhetorical question is defined here as any utterance which is interrogative in
form, but which as opposed to a genuine, information-seeking question
does not sollicit a response. The three most common rhetorical question types
in Tamil2 are what I refer to as the Classical Rhetorical Question (CRQ), the
Thematicizing Rhetorical Question (TRQ), and the Rhetorical Tag Question
(RTag), illustrated in examples (1)-(3) below:3
2.1.
(1)
"Pl kutikkta pai kta irukkum-?"

milk drink-NEG cat
even be-F3NS-Q
Ts there any cat that doesn't drink milk?!' [CRQ]
(2)
Oru nu purusakra ea cencirukkr;

one day husband
what do-PERF-PR3MS
One day, what did the husband do?' [TRQ]
(3)
Inta pacyattellm vaippka illai, pakkattu rile.

this panjayat&all hold-F3PL TAG next
town-LOC
They hold this panjayat (meeting) and all, right? in the next town.'
[RTag]
CRQs
The first example is typical of the rhetorical question type discussed in

classical rhetoric and speech act theory; I refer to this as the Classical
Rhetorical Question, or CRQ. An oft-described characteristic of this type is
that the surface syntax is conventionally understood to be the opposite of
that of the underlying indicative assertion to which it corresponds. That is,
a positive question implies a negative assertion, and a negative question, a
positive assertion.4 Thus, in uttering example (1), the speaker expresses the
view that there is no such thing as a cat that doesn't drink milk. CRQs may
also be WH- questions; in such cases, the usual sense of the corresponding
assertion is universal (absolute) negation (e.g. ' Where will you find a cat that
doesn't drink milk?!', both in Tamil and in English, normally means 'You
won't find a cat that doesn't drink milk any where!).
258
2.2.
TRQs
The question in example (2) establishes a theme (namely, that the hus
band did something), which the narrator must elaborate upon by answering
his own question in subsequent clauses. The fact that Thematicizing Rhetori
cal Questions (TRQs) require further comment is reflected in Schmidt-Radefeldt's phrase "the rhetorical use of question-answer sequences" (1977:378),
and in Grsillon's (1981) expression question-rponse un seul locuteur, to
refer to essentially the same phenomenon. Theoretically TRQs may be realized
as either Yes-no or WH- questions, although as it turns out, virtually all of
the TRQs in the Tamil data are WH- questions, an interesting fact in its own
right, which can be attributed to pragmatic constraints (cf. Section 4.2 below).
In terms of content, the two most frequently encountered TRQs in the Tamil
corpus are 'And then what did X do?' and 'And then what happened?'.
There is a variant of the Tamil TRQ which involves the addition, in
clause-final position, of the subordinating conjunction (the conditional
form of the quotative verb eN 'to say/ask'); a literal translation of is 'if
(you) say/ask'. 5 Thus, while example (2) is a syntactically independent clause,
TRQs followed by are formally subordinate to some other (finite) clause.
Aside from this formal distinction, the presence or absence of does not
appear to significantly affect the meaning or function of the TRQ, although
from a diachronic perspective, will be seen to play a role in the grammaticalization of TRQs as clausal subordinators (cf. Section 5.1). The TRQ + nn
construction is illustrated in (4) below:6
(4)
Atu vantu varus varus am eppai ,

it TOP year-as year
how ask-coND
'If (you) ask, "How (is it) year after year?'"
could also be added on to example (2) above. Indeed TRQs with and
without appear to be in free variation in oral narration in Modern Tamil.
2.3.
RTags
Example (3) is a tag question, formed by the addition of the invariant

tag particle illaiy (lit. 'is it not?') to an otherwise declarative utterance. In
addition to illaiy (and its phonologically reduced variants, illai, ille, and -le),
there are two other important tag question markers in Tamil: The so-called
"emphatic" clitic -, and the imperative (or, the 2nd person past tense form)
of the verb pr 'to see, look'. These are illustrated in examples (5) and (6):
259
(5)
Inta araca kannai mtikkitfirunt-;

this king eye-ACC close-PROG-p3MS-
T h e king was keeping his eyes closed, right?'
(6)
Ioru- anta mutall kmatu vacciruntr

prunka.
another that first-LOC Kamadenu cow keep-PERF-p3RS see-I
'The other in the beginning, (he) was the one keeping the
Kamadenu cow, right?'
Although there is no syntactic constraint which prohibits the use of any of

the Tamil tags with a negative assertion, such uses are rare in these data;
again, this can be accounted for pragmatically (cf. Section 4.3). Thus the tag
questions considered here overwhelmingly presuppose an affirmative response.
These examples, while representing unrelated types syntactically, are
similar in that all are rhetorical (rather than genuine information-seeking)
questions, which were uttered in the context of narration. The interpretation
of an utterance as a rhetorical rather than a true question is dependent on a
complex of features, including the presence of characteristic or otherwise
incompatible grammatical elements,7 intonation, context, and the shared
knowledge and perspective of speaker and hearer. In my recorded data, it is
generally clear, from the narrators' intonation and the presence or absence
of clause-final pauses in oral delivery, which utterances were intended rhetori
cally and which as true questions (in the case of rhetorical questions, lack of
response from the native speaker listeners further supports my assessments).
Moreover, all question forms used in professional performances must be
construed as serving some other, non-interrogative purpose, since in such
contexts the audience is prohibited by convention from participating verbally
in the discourse.
3.
THE INTERACTIVE BASIS OF RHETORICAL QUESTIONS
Why, then, do narrators and Tamil narrators more than most use
rhetorical questions when telling a story? I will first consider the most straight
forward explanation, which is a pragmatic one: Rhetorical questions, not
unlike genuine questions, are intended to stimulate the involvement of the
listeners in the story by making a direct appeal to their attention and evalua
tive processes. Thus a number of scholars following in the recent German
pragmatic tradition (cf. Glich, 1970; Schmidt-Radefeldt, 1977; and Vandeweghe, 1977) characterize rhetorical questions as "appelative" in function:
260
SUSAN . HERRING
Rhetorische Fragen bei denen der Frage-Charakter mehr oder weniger

deutlich ausgeprgt sein kann lassen sich daher als Appell an die
Aufmerksamkeit des Hrers interpretieren. Der Sprecher, der sie verwen
det, zeigt, dass er sich stndig der Anwesenheit des Hrers bewusst ist.8
(Glich, 1970:229)
Similarly:
Mit seiner Frage macht der Sprechende S einen Appell an den Hrenden
H, und er stellt so eine interpersonale Beziehung her.9 (Vandeweghe,
1977:279)
However, rhetorical questions differ crucially from questions in general
in that not only is no answer expected of the hearer, but the hearer may
actually be prevented from answering, due to "particular situational, social
or institutional conventions" (Schmidt-Radefeldt, 1977:378) which prescribe
to him or her a passive role. One such convention is that which governs oral
narration, which is, except in certain exceptional cases (cf. Section 3.1 below),
essentially monologic. There are a number of reasons why it is to the advan
tage of the narrator to discourage extensivelistener participation. As Popovici
(1981 :i5) observes, le rejet d'un argument ou d'une objection possible implique
un dialogue entrevu et vit ('the rejection of a possible argument or objection
implies a dialogue foreseen and avoided'). Dialogue is to be avoided in that
it is less predictable and less under the control of a single participant than is
a monologue. More serious still, a narrator who gives up the floor to another,
no matter how briefly, risks losing it altogether and never finishing his story.
At the same time, it is the task of the narrator to engage and hold the
attention of his audience, a task all the more critical in oral narration, where
numerous opportunities may arise for the listener to become distracted. The
use of rhetorical questioning strategies allows the narrator to address simulta
neously both of these conflicting concerns. He poses questions as a means of
engaging the attention of the interlocutor. At the same time, rather than risk
interruption or derailment by allowing the interlocutor to answer for himself,
the narrator constructs a hypothetical listener with whom he "interacts", even
speaking at times in this other listener's "voice". Thus a narrator may
"answer" his own CRQ by explicitly stating the assertion it implies:
(7)
Avan peyil poytt itai colratukku viukku

he
fail-ADV go-coND this to-say
house-DAT
varuv-?
Varamtt.
come-F3MS-Q come-NEG-F3MS
Tf (a schoolboy) fails (an exam), will he come (straight) home to
tell about it? He won't.'
261
Or, he may imitate a dialogue by means of a TRQ and its response:

(8)
inta ea
yirutu;
karppam
thus this girl
what become-PFV-PR3NS pregnant
kuu
inta ponnu.
become-PR3NS this girl
'And so what happens to the girl?
She gets pregnant, the girl (does).'
(The implication here is that the listener, unable to contain his or her curiosity,
has posed the question.) 10 This hypothetical listener is an idealized listener,
in that at any given point in the narration, he knows all that he is supposed
to know in order to appreciate the story, and is attentive and eager to receive
whatever information follows.11 Actual listeners are unreliable in these
respects, since they may fail in comprehension and attention for a variety of
reasons.
3.1. Rhetorical questions in traditional performance narrative
Compelling evidence in support of the abstract ideal listener analysis can
be found in the traditional Tamil Villu Pu storytelling genre. 12 In a Villu
Pu performance, a principal narrator (A') is assisted by a secondary
narrator ('B') and several singer/musicians (whose "narrative" duties are
largely restricted to echoing and agreeing with A' and 'B'). Of special interest
is the role of 'B' in the narrative performance. On the one hand, he may
introduce new material into the narrative sequence to a limited extent, either
alone or in conjunction with the main narrator (this constitutes one of the
"exceptions" referred to earlier to narration as monologue). His primary role,
however, is as a physical embodiment of the "ideal listener", whose duty it
is to respond to the main narrator at each appropriate moment in an appropri
ate way. As such, he responds to rhetorical questions by making explicit the
(normally unspoken) responses which they conventionally presuppose. Rhe
torical questions are extremely common in Villu Pu performances. Consider
example (9) below, which contains four CRQs:
(9)
(The wife of the great god Shiva, cursed by her husband, despairs
of accomplishing what he has ordered her to do)
A: "(Nilakalai crtiruttam ceyya namml muiyum-?
: Muiyte!
A: Ptti katta muiyum-?
B: Atuvum muiyte!
262
A:
:
A:
:
Cetikalai natta muiyum-?

Atuvum mutiyte!
Namml ea ceyya muiyum?
Oru kriyamum natakkt!"
A:
B:
A:
B:
A:
B:
A:
B:
"Am I capable of domesticating the land?! [Yes-no]

(I) can't!
Can (I) lay out plots (for cultivation)?! [Yes-no]
(I) can't (do) that, either!
Can (I) plant plants?! [Yes-no]
(I) can't (do) that, either!
What will I be able to do?! [WH-]
Not a single thing will come out right!'"
In this example, Narrator responds negatively to affirmatively-phrased

CRQs, and with a universal negative to a WH- type CRQ, confirming that
this is indeed how the Tamil narrator who makes use of such constructions
intends himself to be understood.
Narrator also responds appropriately to TRQs, not by providing any
information, but by encouraging Narrator A to supply the answer himself,
as in examples (10) and (11):
(10) A: Akk tankai uprum eppai nrtukirr?
: Eppai?
A: [pu] Avar kuluttalavu nalla tanril...
kanniy arkai vantu nrta...
vyalavu taril...
vantu n vilaiyaa...
A: 'How do the seven sisters bathe?
B: How?
A: [sings] Up to their necks in the good water...
the maidens came and bathed...
up to their mouths in the water...
they came and stood playing...'
(11) A: Ank ntu varukinravan yr ?
: Yr?
A: Turiytanan.
A: 'Who is the one who rules there?
B: Who?
A: Duryodhanan.'
263
Observe that in the last two examples, it is not the "listener" (Narrator B)
who introduces the question, as might have been expected on the basis of the
preceding discussion, but rather Narrator A. This is because TRQs also serve
an important organizational function, which will be described later on, and
which is normally reserved for the principal narrator alone. Thus even in his
role as ideal listener, Narrator cannot be relied upon to ask the right
questions at exactly the right time; the wrong question or the wrong timing
could derail the flow of Narrator A's narration.
RTags are also common in the Villu Pu, although they do not elicit
an explicit response as frequently as the other two types. When Narrator
does respond, his responses follow the expected pattern of affirming the
sentential assertion:
(12) A: Namma periyr pasttu irukku prunka, central pasttu.
: Am.
A: Atukku mule pitpr aitile tipeti vittukkitu irunt.
A: 'There's our Periyar Bus Stand, right? The central bus stand.
B: Yes.
A: In front of it, on the platform, he was selling matches.'
(13) (A
A:
:
A:
young god muses aloud)

"Mantirattil cirantatu malaiylam t-ei
Malaiyla ntu t.
Nn malaiylam po t vtum!"
A: '"(The place where they have) the best enchantments (is) Malai
ylam (country), right?
B: (It's) Malaiylam country, indeed.
A: I must go to Malaiylam (country)!'"
The first of these examples contains the polite imperative form prunka; the
second, the bound suffix -. Both function here as tags, as shown by the
English translations.
Note that in none of these examples do the responses provided by
introduce any new information into the discourse. The impression is of a
dialogue, but in fact the range of responses permitted is quite restricted.
The Villu Pu data provide evidence, therefore, of the validity of the "ideal
listener" concept in Tamil oral narration. The value of such a construct is
that it allows the narrator to evoke an interactional dynamic, while at the
same time maintaining ultimate control of the discourse. Hart (1980) claims
that the stylistic device of addressing oneself to some fictional, absent, or
264
inanimate third person audience can be observed, not just in Tamil, but as a
pan-Indian tendency dating back to the earliest recorded literary works. Thus
the high RQ content in contemporary Tamil narrative may well reflect, to
some extent at least, a more traditional interactive narrative strategy.
4.
FROM "EXPRESSIVE" TO "TEXTUAL" FUNCTIONS
Let us now consider uses which extend the RQ phenomenon beyond the
limits of the interactive domain. In this section, I discuss the expressive
pragmatic 'meaning' of each of the basic RQ types, and show how, in the
case of TRQs and RTags, expressivity has been largely replaced by textual
by which I intend discourse-organizational functions. This constitutes a
shift which, it will be argued, paves the way for the eventual grammaticalization of a subset of these elements as clause-linking markers.
4.1.
Expressive functions of CRQs
If rhetorical questions are interactive in the qualified sense just discussed,

they are also expressive, in that each RQ type is associated with a particular
stylistic effect. The classical type, as is often pointed out, is a persuasive device
which characteristically presents the point of view of the speaker as if it were
obvious. From the perspective of the addressee, of course, the point may not
be obvious in the least, but part of the efficacy of a CRQ is that it is more
difficult to refute than an ordinary assertion, in that it presupposes the address
ee's agreement. In the Tamil narrative corpus, CRQs are also used to scoff
('I can't do with my eyes open what you can do with your eyes closed?!'),
boast ('Is there anyone greater than I?!'), express dismay ('What's the use of
being a great king if I can't fulfill my mother's dying wish?!'), show amazement
('Is it possible to behold such a forest?!'), and to convey many other expressive
nuances as well. These examples appear almost exclusively in mimetic, or
quoted sections of text. 13 The use of CRQs in quoted dialogue within narrative
closely parallels their use in actual conversation. As such, the principle func
tion of CRQs in narration is pragmatic; i.e. to represent a conversational
exchange in a more persuasive or expressive fashion.
4.2.
Expressive functions of TRQs
In order to illustrate the expressive value of TRQs, it is necessary to take

into account not only the question clause but the clause or clauses which
265
follow it in sequence, and which serve as its response. Consider, in this

context, example (8) and example (2') (an expanded version of (2) above):
(8)
Aka inta pou ea

yirutu;
karppam
thus this girl what become-PFV-PR3NS pregnant
kutu
inta pou.
become-PR3NS this girl
'And so what happens to the girl? She gets pregnant, the girl (does).'
(2')
Oru nu purusankran ea cencirukkr,

one day husband
what do-PERF-PR3MS
"N vetaikku pokanum"
nu
poyin.
I
hunt-DAT gO-be-neceSSary QUOT gO-PFV-P3MS
' O n e day, what did the husband do?

Saying, ' must go hunting", he went off.'
In her analysis of discourse strategies in Romance, Wehr (1984) charac
terizes question-answer sequences of this type as marked with respect to
simple declarative word order by the addition of the pragmatic feature
[ +SURPRISE]; that is, the information which follows the question is eval
uated, by means of this device, as being in some sense unexpected or reactionworthy. In this schema, the use of a question form functions to create drama
by placing the audience in suspense, arousing their curiosity about what is to
follow.
These observations would seem to apply to English as well, in that the
most straightforward English translations of Tamil TRQ-response sequences
tend to convey the impression that the narrator is attempting to interject a
suspenseful tone, as in the examples above. Here Tamil differs from English,
however, in that (i) the overall frequency of TRQ-response sequences is
notably higher in Tamil; (ii) they may be employed even when the events they
introduce are in no way dramatic or unexpected; and (iii) their use does not
necessarily imply that the speaker evaluates said events as dramatic or unex
pected. That is, while the analysis advanced by Wehr may well have been
true for Tamil TRQs at an earlier stage in their development, the suspenseful
function, possibly through pragmatic unmarking based on frequency of use,
has largely given way to a function on the textual level: That of introducing
new information 14 into the text in a pragmatically "focused" way. WHwords are especially effective in focusing interest on a particular constituent,
a fact which may tie in with the overwhelming predominance of WH- TRQs
(rather than the Yes-no type) in Tamil. Thus a more idiomatic translation of
266
SUSAN . HERRING
(8) above might be, 'So what happens is, the girl gets pregnant', where a
pseudo-cleft construction, rather than a question, translates the Tamil TRQ.
As such, the construction serves as an alternative to syntactically more
complex focus constructions, which in Tamil typically take the form of nominalizations. Viewed in this light, the TRQ and the clause which follows, although
syntactically independent, function in the discourse as a single structural
unity, 15 an observation supported by prosodie and intonational evidence as
well. They resemble topic-comment structures in which the "theme" or
"topic" introduced by the TRQ is commented upon in the response; e.g.
'What happens to the girl is, she gets pregnant'; 'One day what the husband
did was, he said he had to go hunting, and went off'.
The strategy serves a broader organizational function within the story
as a whole by relating entire sequences of short, syntactically independent
clauses to a single focus or theme, thereby creating loose structural unities
reminiscent of paragraphs in written discourse. In the continuation of (4),
given as (4') below, the narrator goes on to add ten more finite clauses, all
of which contribute in some way to the "answer" to the TRQ 'If you ask
"How is it year after year?'":
(4')
Atu vantu arus varusam eppai ,

Kijattuvalle past rampikkum cantai.
Kattuval kiratu oru r.
Kattuvalle t past rampikkum.
Campantillma..pasf cantai anta hariccas pofav mffnka.
Inta tevarunka t pdfuvnka.
Avanka enamvatu colluvnka.
Ivanka ufane poy, avaka itulc acu ru per poy vettituvnka.
Venina ufane, avaka vantu konea pettai veffuvnka.
Ivanka vantu konca konca kufa pettai vettuvnka.
Ippafi tn v arus varusam, april mcam, karektt inta cantai
nafanturum.
'If (you) ask, "How (is it) year after year?" ( = How it is year after
year is,) the fighting starts in Kattuval.
Kattuval's a town.
It starts first in Kattuval.
For no reason.. the Harijans never start fighting first.
It's the Devars who start (it).
They (the Harijans) say something or the other.
267
And then they (the Devars) immediately go over.. five or six people
go and cut (them) up.
As soon as they do that, (the Harijans), they cut a few people.
They only cut a few people.
The fighting takes place like this, year after year, exactly in the
month of April.'
Here the repetition of the adverbial varus varumsam 'year after year' in the
last clause effectively brackets off the TRQ and the ten clauses which follow
as a single cohesive unit within the discourse.
TRQs, as mentioned above, are more frequent in oral than in written
narration, and we are now in a position to account for this distribution. As
a device which may be employed to create structural cohesion while at the
same time preserving a straightforward paratactic "one clause at a time"
(Pawley and Syder, 1977) mode of presentation, TRQ-response sequences are
well suited to the demands of on-line oral narrative production, which tends
to prefer loose sequences of finite clauses to more complex embedded con
structions. Formulaic expressions such as And then what happened?', And
then what did he do?' presumably require a minimum of processing in the
narrator's consciousness, allowing more time in which to organize his thoughts
and plan what he is going to say next (Chafe, 1980). At the same time,
because of their open-endedness as question forms, they help to insure that
the audience will remain attentive until the speaker has succeeded in formulat
ing his next utterances.
The extent to which this strategy is productive can be seen in the following
example, an excerpt from an informal oral retelling of a mythological tale.
Note the narrator's heavy use of TRQs in organizing (and reorganizing) the
linear presentation of his ideas:
(14) anta urile, payakara pacam.
Cappifuratukku kuta oum kiaiyatu makkaukku.
A rjvukku verri vi konfturka.
Kontluratu .. cpptu matilum rjvukku eppai varutu ,
oru itattil iruntu oru muivar vantu cpptu kuluttu viluvru rjvukkka.
Oru arai vay iru cpptu ille; arici ille.
Avvalavu pacam.
E cantai poltatule, irukkira.. panamellm celavaiccu pccu.
Appa muivar kututtuvium potu, iaiyile anta cappata.. inta...
rjvota lunka kottu varraka.
268
SUSAN . HERRING
Varum potu..iaiyile ea parnka tirutaka;

anta cppfellm pituki cppifarnka.
Avaka pituki cppittnka ,
avankalukku cppifaratukku oum ille.
Atale avaka pituki cppifarnka.
'But in that town, (there's) a terrible famine.
There isn't anything at all for the people to eat.
But they're holding a victory celebration for the king.
(How they're) celebrating is... how does a serving of food come
for the king?
A sage comes from someplace and donates food for the king.
(There's) not even half a belly-full of food; (there's) no rice.
(That's) how much famine.
Why? The money they had all got spent in waging war.
Then when that sage donates (the food), in the meantime the food
(was)...
the kings' men bring'it back.
As they're bringing (it back)...in the meantime what do (some)
thieves do?
They snatch up the food and eat it.
Why do they snatch it up and eat it?
They don't have anything to eat.
So they snatch it up and eat it.
The repeated use of rhetorical questions in this section of discourse may
sound odd or even incoherent to native speakers of English, since it is difficult
to ascribe an expressive interpretation to what is not a particularly dramatic
sequence of events (nor is the sequence of central importance for the story
as a whole). If, however, we change the TRQs in the English translation to
pseudo-cleft constructions, as in (14'), it is possible to appreciate something
of the cohesive function which they serve in the original Tamil:
(14') 'But in that town, (there's) a terrible famine.
There isn't anything at all for the people to eat.
But they're holding a victory celebration for the king.
(How they're) celebrating is...
how a serving of food comes for the king is,
a sage comes from someplace and donates food for the king.
(There's) not even half a belly-full of food; (there's) no rice.
269
(That's) how much famine.

(The reason) why is (because) the money they had all got spent in
waging war.
Then when that sage donates (the food), in the meantime the
food...
the kings' men bring it back.
As they're bringing (it back).. .in the meantime what (some) thieves
do is,
they snatch up the food and eat it.
(The reason) why they snatch it up and eat it is,
they don't have anything to eat.
So they snatch it up and eat it.
Clearly we have moved here beyond the realm of expressivity to the
organization and presentation of the discourse itself; or, to employ Traugott's
terminology in a somewhat broader sense, from the "expressive" to the
"textual" level.
4.3. Expressive functions of RTags
In the same way that CRQs are persuasive, and TRQs (in some languages
at least, and probably originally in Tamil as well) suspenseful, a narrator's
use of rhetorical tag questions evokes solidarity with the listener by presuppos
ing the listener's knowledge of the information thus tagged, thereby including
both listener and speaker within the informed sphere of those capable of fully
appreciating the significance of the narrated information. 16 As in the case of
CRQs, the listener need not in actual fact be familiar with the information
thus evaluated, or he may not have the particular information in mind at the
time, such that the RTag serves as well to bring it to the foreground of his
consciousness.
RTags draw on two types of "common" knowledge: That which is
external, and that which is internal to the narrative. The former typically
includes (i) shared cultural knowledge, as reflected in the use of the term
'panjaya (the popular form of village government in India) in the RTag in
example (3); (ii) shared point of view (especially if there is a close social or
interpersonal relationship between speaker and listener); and (iii) knowledge,
both general and specific, that the speaker has reason to believe or chooses
to represent as though he believed is shared by the listener. Narrativeinternal knowledge is information that was either specifically mentioned in,
270
SUSAN . HERRING
or can be inferred from, the previous discourse; this is reflected in examples

(5) and (6).
Beyond this use, which is essentially pragmatic, RTags have a textual
function in Tamil, which is again one of clause-linking. An important charac
teristic of RTags is that they are anaphoric; that is, they refer back, either to
the discourse, or to previous experience of a more general nature. As such,
they are frequently used to specify and retrieve referents which are old
information, i.e. in cases where the narrator wishes to predicate something
new of these referents. Such referents are typically positive and definite; hence
the absence in the data of RTags with negative propositions. The "retrieval
and predication" function of RTags becomes clear once we examine the
examples given in isolation above along with the clauses which follow them
immediately in the discourse:
(3')
Inta pancyattellm vaippnka illai, pakkattu urile.

this panjayat&all hold-F3PL TAG next
town-LOC
Enka app pkma irukkum potu ivan pyituvn.
our father gO-NEG be
time go-PFV-F3MS
They hold this panjayat (meeting) and all, right? In the next town.
When my father couldn't go, this (boy) would go (in his place).'
(5;)
Inta araca kannai mtikkittiruntn-;

this king eye-ACC close-PROG-p3MS-TAG
ivanum kannai tirantu pkkurn.
he-also eye-Acc open see-PR3MS
'The king was keeping his eyes closed, right?
He too opened his eyes and saw (them).'
(6') Ioru- anta mutall kmatnu vacciruntri prunka..

another that first-LOC Kamadenu cow keep-PERF-p3RS see-I:
Oru munivar.
a
sage.
Avari visvmittirarai
viruntukku kppifuvr oru n.
he
Vishvamitra-ACC feast-DAT call-F3R
one day
T h e other- in the beginning [i.e. of the story], (hei ) was the one
keeping the Kamadenu cow, right?' A sagei.
Hei invited Vishvamitra to dinner one day.'
In each of these examples, the situation referred to in the RTag clause is old
or otherwise accessible information; the assertion in the second clause is new
271
information. As in the case of TRQs and the clauses which follow them, it
is possible to speak of a loose structural unity between the RTag and its
following predication, a unity reinforced in many cases by prosody and
intonational contour. Functionally, sequences of RTag + clause may replace
more complex embedded relative clause constructions, especially in the spoken
language. Compare, for example, the loose paratactic version in (3') above
with the version in (15), which contains an embedded (participial) relative
clause construction modifying the nominal head 'panjaya (note: 'FAJP' =
'Future Adjectival Participai'):
(15) Enka app pakkattu urile vaikkum pacayattukku
our father next
town-LOC hold-FAjp panjayat-DAT
pokma irukkum potu, ivan poyituvn.
gO-NEG b e
t i m e
go-PFV-F3MS
'When my father couldn't go to the panjayat (meeting) which (they)

hold in the next town, this (boy) would go (in his place).'
There are a number of advantages which the tag strategy has over the
participial strategy. By preserving the iconic order of the two clauses, the
RTag version not only avoids the necessity of embedding, but it also eliminates
the need to indicate case relations (in the example above, the dative indicating
GOAL) on the nominal head; that is, the paratactic version is non-specific
as to the thematic relationship between the arguments of the two clauses. The
linking of clauses by means of RTags is a strategy, therefore, which facilitates
the presentation of information with a minimum of encoding complexity.
5.
THE GRAMMATICALIZED STATUS OF RQ ELEMENTS
These observations regarding the textual functions of TRQs and RTags

are of interest in their own right, and merit further consideration within the
realm of discourse analysis. For the purposes of the present discussion, what
is to be noted is that the discourse-organizational use of RQs represents a
shift in function away from the basic RQ strategy defined above. We might
hesitate to claim that grammaticalization has taken place, however, since we
are not left with new autonomous grammatical markers, but rather with two
highly specific (albeit extremely productive) construction types. We might
predict, nevertheless, that if the constructions were to grammaticalize further,
it would be as clause-connecting elements of some type, given the evident
trend towards the combining of two finite utterances into one.
272
In fact, this prediction turns out to be correct. The WH- elements

'why', enna 'what', and eppati 'how' (either alone, or in combination with the
conditional conjunction ), show evidence of grammaticalizing away from
TRQ constructions into autonomous conjunctions. As for RTags, while all
three of the Tamil tags presented here can function as informal relativizers,
as examples (3'), (5'), and (6') demonstrate, the particle - has taken on many
of the features of a formal relativizing particle as well. These developments
are considered in greater detail below.
5.1.
From TRQ to conjunction
There is a qualitative shift from the discourse-organizational use of TRQs

described above to the use of the conjunction ( 'why' + ) in the
following sentence:
(16) A van ink illai avan rukku
h
h e r e NEG CONJ
town-DAT
pon.
go-P3MS
'He is not here because he went to his village'.

In purely syntactic terms, this shift involves the displacement of the WHword from its usual clause-second position to the end of the clause, where
it attaches to the subordinating conjunction . (Compare, for example, (16)
above with the TRQ-response focus construction: Ava ink illai (),
avan rukku po 'Why he isn't here is, he went to his village'.) A further
fact regarding the compound conjunction is that it may be analyzed as
belonging exclusively to the second, but not the first clause in the sequence,
as evidenced by the fact that
*Ava ink illai .
*'He isn't there because.'
is not a complete grammatical utterance, while
E ava rukku po.
*'Because he went to his village.'
is fully grammatical in Tamil. In sentence-initial position, functions
analogously with sentence-initial conjunctions such as atonale 'therefore',
'but', and iruntlum 'nevertheless' (the latter two also deriving from condi
tional forms):
Atale ava rukku po.
'Therefore he went to his village.'
273
A ava rukku po.

'But he went to his village.'
Iruntlum ava rukku po.
'Nevertheless, he went to his village.'
It also groups with this class of elements semantically; in particular, 'because'
(REASON) and 'therefore' (CAUSE) express closely related logical concepts.
A further argument for the grammaticalized status of expressions such
as involves intonation and prosody. (In the case of , phonological
reduction is also involved; i.e. a sequence of three alveolar nasals is reduced
to two, in keeping with the phonotactic rules of the language.) WH- words
in Tamil typically exhibit a high, rising intonation with an optional drop in
pitch at the end. The intonation of the subordinator , on the other hand,
falls from the high to the middle range; i.e. to signal a non-final clause
boundary. When, however, the two are combined as in example (16),
tends to be pronounced with a high, level intonation, leading directly into the
following clause. The two strategies can be contrasted graphically as in (16a)
(16a) A van en inke illai a, ava

h
Why
h e r e NEG SUBD
urukku
ponan.
tOWn-DAT go-P3MS
'Why he isn't here is, he went to his village'.

(16b) A van ink illai ava rukku
h
h e r e NEG CONJ
po.
town-DAT go-P3MS
'He's not here because he went to his village'.

The intonational characteristics of are similar to those for other twosyllable sentence-initial conjunctions, such as 'but' and ka 'therefore'.
These same criteria allow us to establish the word n alone as a coordinat
ing conjunction (with the same functional value as ), as in the following
example:
(17) Terle
nau polttu ellm cariy irukk
u,
chariot-LOC nut bolt all okay be-PR3NS-Q QUOT
atutta nl tr
ottaum-.
CONJ next day chariot drive-be-necessary-TAG
ivar tn pkkanum
atellm.
he EMPH look-be-necessary that-a11
'(He checks to see) if the nuts and bolts on the chariot are all okay,
because the next day he has to drive the chariot, right?
He has to take care of all that stuff himself.
274
Thus is, strictly speaking, not an essential ingredient in WH- conjunctions,

although it is more often present than not.
So far, we have been dealing almost exclusively with ; however, a
similar shift can be seen to have taken place with two other WH- elements,
ea 'what' and eppati 'how', which combine with to produce the conjunc
tions ea and eppali. Although they exhibit the same syntactic and
prosodie properties as , it is more difficult to ascribe to them distinct
semantic labels. Indeed, they may sometimes be used interchangeably, or in
place o . This fact suggests that their lexical identity (i.e. as distinct WHwords) may be weakening, even as their grammatical meaning (as coordinat
ing conjunctions) is strengthened.
Finally, it may be mentioned that the written language, which is generally
more conservative than the spoken language, recognizes two of these conjunc
tions, and ea, as unitary grammatical entities. In Written Tamil
they appear as single words, viz. erl, from + enrl (written form of
), and eaverl, from ea + erl, with the glide -v- inserted according
to a regular morphophonemic process. All of this evidence syntactic,
prosodic, and orthographic argues in favor of treating these elements as
fully grammaticalized conjunctions in Modern Tamil.
5.2.
From RTag to relativizer
Turning now to rhetorical tag questions (RTags), recall that a sequence

made up of a clause containing a tag (-, illai(y), or parunko) plus the clause
or clauses which follow may function as an informal relativizing construction
in Tamil. Examples of this are concatenated utterances of the type: T h e king
was standing there keeping his eyes closed, right? He too opened his eyes and
saw them', which may be contrasted with embedded relative clause construc
tions, e.g. 'The king, who was standing there keeping his eyes closed, also
opened his eyes and saw them'. Of the three Tamil tags described here, -
alone has achieved the status of a full-fledged grammatical relativizer. This
status is attested by a number of facts, not the least of which is that it has
been classified as such by at least one native grammarian. Annamalai (1980)
explicitly refers to - marked constructions of the sort we have been discussing
as "tag relative clauses", which he notes are "almost like two consequent
sentences in a discourse" (1980:291). The tag RC is one of three relative
clause types which he identifies for Tamil; the other two are the "participial
RC" and the "pronominal RC". The so-called "pronominal RC" (a caique
from a Sanskrit construction) is not of interest here, but it may be observed
275
that Annamalai's examples of tag and participial RCs, which I reproduce

below, correspond exactly to the distinction made here between informal
paratactic and formal embedded relativizing constructions, as illustrated in
examples (3') and (15) above (note: PAjP = 'Past Adjectival Participle'):
(18a) Nttu
oru payyan vant-
avan
iekkum
yesterday a boy
come-P3MS-TAG he today-also
van tn.
come-p3MS
A boy came yesterday, you know, he came today also'.
(18b) Nettu
vanta
pa yan iekkum vant.
yesterday came-PAjp boy
today-also come-P3MS
T h e boy who came yesterday came today also'.
Further arguments can be made for the grammaticalized status of -.
Syntactically speaking, its role has shifted from that of a clause-final particle
to a particle which may relate an attribute to a nominal head within a clause.
This is illustrated in the following example, taken from a contemporary Tamil
short story: 17
(19) Nn poy aval nirirunt-
anta iattil
I
go she stand-PERF-p3FS-TAG that place-LOC
at
mtiri n katalai
verittu prkkirn.
that-EMPH way stand ocean-ACC stare look-PR 1s
went and stood in the place where she had stood, and stared
at the ocean in the same way'; (lit. 'I went and in the she had
stood- place in the same way stood and stared at the ocean').
Here the finite clause 'she had stood' modifies the noun phrase 'that place',
with the suffix - indicating the subordinate relationship of the former to the
latter; i.e. - translates the English relative pronoun 'where'. We may further
observe that the relativized clause is entirely embedded within, rather than
simply preceding, the matrix clause. Behavior of this sort is associated with
the participial RC type, but not with the tag type, which tends to preserve
the order 'old information'-'new information', and to present information
one clause at a time. This fact alone indicates that - has undergone a
qualitative shift in function in the direction of increased grammatical
autonomy.
Like relative pronouns in English, - may also stand in for the nominal
head, in the sense of 'that which', 'the one who', '(the place) where', etc. Thus
276
SUSAN . HERRING
the heroine in another Tamil short story 18 who, having just been robbed,
responds to her husband's inquiry as to what was stolen with:
(20)
"Cankili, motaram, vte, panam, nnka

kututt-iruntnka-.."
chain
ring
watch money you(resp) give-PERF-p2R-REL
'(Your) chain, ring, watch, money... what you gave (me)!
is clearly not commenting on the husband's past action of giving, but rather
on that which was given. A similar usage is illustrated in the oral example
below:
(21)
"N appo co-

ukalukku purificat?"
I
then say-p1s-REL you-DAT be-understood-p3NS-Q
'Do you understand (now) what I said then?'
Here what is relevant is not the fact that the speaker of the utterance said
something previously, but rather what was said, which had been misunder
stood by the addressee at the time.
As in the case of WH- conjunctions, prosodic cues provide additional
support for the grammaticalized status of -. In example (21) above, there is
no break between co and ukalukku, whereas if we were to literally
interpret the first half as a tag question, we would expect either a pause or
deceleration at that juncture. Moreover, the utterance is characterized by a
single, rather than a two-part, intonation contour. While normal intonation
for the tag - is rising-falling, the intonation for (21) is mid-high and level
throughout, rising only at the end of the sentence to signal the Yes-no
question. On the basis of this and the other types of evidence mentioned, it
is clear that the suffix ~ must be accorded the status of a full-fledged rela
tivizer
6.
FROM MORE OR LESS PRAGMATIC: IMPLICATIONS FOR A

THEORY OF GRAMMATICALIZATION
These observations on the grammaticalization of TRQs and RTags in

Tamil lend support to Givn's claim (1979) that "[subordinated] constructions
ar[i]se diachronically, via the process of syntacticization, from looser, con
joined, paratactic constructions" (p.222). The RQ strategies examined here
can be characterized as moving from what Givn calls a 'pragmatic' towards
a more 'syntactic' mode of communication (cf. Table 1).
277
Table 1 The pragmatic vs. the syntactic mode of communication (adapted from Givon
1979:223)
Pragmatic Mode
Syntactic Mode
loose conjunction
.tight
subordination
topic-comment structure
.subject-predicate
structure
slow rate of delivery under
.fast
rate of delivery under a
several intonation contours
single intonation contour
no use of grammatical morphology
elaborate use of grammatical morphology
The 'pragmatic mode' roughly characterizes the discourse-organizational use

of RQs described in Section 4, which relies, as we have seen, on loose topiccomment structures delivered under separate intonation contours and con
joined primarily by virtue of their juxtaposition in the narrative sequence. In
contrast, elements such as and - subordinate one clause to another,
unite pairs of clauses under a single intonation contour, and have themselves
evolved into grammatical morphology in short, display features of the
'syntactic mode'. As Givn himself cautions, these labels represent but two
poles at the extremes of a continuum; there exist numerous intermediate
possibilities (particularly with regard to intonation), and more than one 'stage'
may be attested simultaneously at any given point in the history of a language.
Thus the Tamil data provide evidence for the view that grammatical elements
develop as means of encoding basic discourse-pragmatic functions, such as
focus, topic-comment, anaphora, and attribution. Given the primacy of these
concerns in oral communication, it indeed seems to be the case, as DuBois
(1987) claims, that "grammar codes best what speakers do most".
To summarize, the direction of development which best accounts for the
Tamil data described here takes as its point of departure the interactive use
of rhetorical questions in narration, i.e. to evoke listener involvement in the
events of the story. I have suggested that the narrator's desire to maintain
control of the discourse results in the constraining of the listener's right to
respond, a fact which appears to have contributed critically to the convention
alization of the strategy at the expense of its interactive value. At the same
time, each RQ type has its own expressive character: The persuasive force of
the CRQ, the suspenseful nature of the TRQ, the use of RTags to create an
aura of complicity. As the latter two types begin to develop textual functions
based on their cataphoric and anaphoric natures, respectively, increased usage
inevitably results in pragmatic unmarking, or the gradual diminishing of their
278
SUSAN . HERRING
expressive impact. Via a reanalysis on the textual level, TRQs and RTags
develop as cohesive devices in tandem with the clauses which they introduce.
This, in turn, leads to their ever-increasing specialization as grammatical
markers of conjunction and subordination. By the time they arrive at the
final stage, the question forms have lost their marked, stylistic value; they are
no longer interpreted as questions on any level. The process is gradual,
however, and it is less appropriate to speak of discrete "stages" than it is of
points along a continuum which allow for considerable synchronic overlap
of function.
Thus we may speak of an overall extension of the function of TRQs and
RTags, from the pragmatic (interactive/expressive) level to the textual (organ
izational) level to autonomous clause-level grammar. If this analysis is correct,
it is evidence that the process of grammaticalization, defined in general terms
as the means by which new grammatical elements come into being, is not
'unidirectional' in the sense proposed by Traugott. That is, subjective, prag
matic-based meanings are not always late concomitants of grammaticaliza
tion; rather, as we have seen, they may constitute the very roots of grammar.
In concluding, I return to the questions raised at the outset of this paper.
While a great deal more research is clearly necessary before we can venture
any conclusions as to the relative importance of discourse-grammaticalizing,
as opposed to lexical-grammaticalizing, strategies in the languages of the
world, or even within a particular language, it is likely that discourse-based
grammaticalization will be found to play a more important role than has
been suspected up until now. Given the level of sophistication achieved in
the field of historical semantics as opposed to the relative newness and lack
of a consistent methodology which, unfortunately, has characterized most
studies of discourse-related phenomena to date, it is only natural that we, as
linguists, should have focused most on what we can talk about most easily;
e.g. the study of change in meaning of individual words. Yet the fact that
natural language use is necessarily situated in the context of some larger
discourse means that it is subject to manipulation for discourse-pragmatic
ends. In some cases, such usage may facilitate a shift in meaning and/or
function of the-form or forms involved, as in the example given here of the
grammaticalization of rhetorical questions. Unfortunately, concrete indic
ations of context are typically lacking in the written records which constitute
the basis for historical reconstruction, with the result that factors which might
have been crucially influential may not figure in our analyses at all.
Given this limitation, I believe that we are justified, to a limited extent
at least, in applying diachronic methods to synchronic data. What makes this
279
approach feasible is the fact (supported by a growing body of evidence) that

language change is not discrete, but rather progresses along a continuum,
with old usages co-existing alongside of new ones. It is for this reason that
we are able to observe, or at least infer, the progression of stages from the
evidence in Modern Tamil: RQ forms may still function (to some degree, at
least) interactively, and varying degrees of syntacticization are evidenced in
the narrative data base examined here. An obvious advantage of the method
of "synchronic reconstruction" is that it enables us to study phenomena to
which we might not have access otherwise. As I have noted, there is reason
to believe that most discourse-based grammaticalization falls into this
category.
What of the mechanisms which drive discourse-based grammaticaliza
tion? On the basis of observations made for Tamil, we may cite the following
three processes: Pragmatic unmarking, or the process whereby a stylistically
or expressively-marked usage loses its marked value as a result of frequent
use; reanalysis of function, e.g. from one functional/semantic component to
the other; and, in the sense employed by Givn, syntacticization of loosely
conjoined structures into syntactically unified ones. Although further research
will no doubt expand and refine this list, there is good reason to believe that
each of these processes represents a general, underlying force in language
change.
Processes traditionally associated with semantic change are also involved
in the grammaticalization of Tamil rhetorical questions: Bleaching of lexical
meaning (i.e. of WH- words), and corresponding strengthening of grammatical
meaning, as well as increasing abstraction away from the immediate context
of face-to-face interaction in the direction of grammatical autonomy. Finally,
to return to the original point of controversy, it can be argued that the shift
to rhetorical from interrogative question meaning, which presumably must
have preceded the developments discussed here, is a type of subjectification,
in the sense intended by Traugott. Thus the history of the conjunctions nn,
etc. and the relativizer - may be said to involve both subjectification in
the original extension of function from true to rhetorical questions and
de-subjectification, in the grammaticalization of pragmatic devices as autono
mous clause-linking elements. This supports the hypothesis that subjectifica
tion is a bi-directional process.
Since all known human languages have strategies for forming questions,
it is likely that the rhetorical question is a universal phenomenon as well. I
would not be surprised if a correlation were discovered in other, unrelated
languages between the use of rhetorical questioning strategies and grammati-
280
cal subordination; the evolution of the former into the latter is well-motivated
in terms of basic communicative functions which speakers of all languages
share. Clearly, there is a need to integrate discourse-functional approaches
of this type with the methods and insights of those researchers whose principal
concern has been with lexically-driven meaning change. Yet before this can
be achieved, the existence of functional influences, and their potential impor
tance to the study of grammaticalization, must first gain wider acceptance. I
have presented an analysis which reveals the ways in which functional influ
ences may operate, both synchronically and over time, within a particular
language. At the very least, the evidence is highly suggestive of an alternative
course of development, and as such merits closer consideration within the
domain of grammaticalization studies.
ABBREVIATIONS
F3ns
Future 3e person neuter singular
Pr3ms Present 3e person masculine singular
F3pl
Future 3e person plural
P3ms
Past 3e person masculine singular
P3rs
Past 3e person respectful singular
F3ms Future 3e person masculine singular
Pr3ns Present 3e person neuter singular
F3r
Future 3e person respectful
FAjP
Future adjectival participle
SUBD Subordinator
PAjP
Past adjectival participle
P1s
Past le person singular
P3ns
Past 3e person neuter singular
ACKNOWLEDGEMENTS
*
An earlier version of this paper was presented at the Tenth Annual South Asian
Languages Analysis Roundtable at the University of Washington, July 10-13,
1988. Thanks are due to Suzanne Fleischman, Talmy Givn, Bernd Heine, Eric
Pederson, Harold Schiffman, and Elizabeth Traugott for their helpful comments
on a revised, later version. Any problems that remain, either of fact or of
interpretation, are entirely my own responsibility.
THE G R A M M ATICALIZATION OF RHETORICAL QUESTIONS IN TAMIL
281
NOTES
1.
In contrasting individual lexical elements with the larger discourse/pragmatic context, I

do not mean to suggest that the two are isolatable in actual usage: Natural language use
is always framed in a broader functional context, and discourse strategies (at least, of the
type we are concerned with here) necessarily make use of individual words. The point I
wish to emphasize is that grammaticalization a process traditionally viewed as affecting
the internal semantic structure of individual words may also operate on discoursebased strategies or construction types, in which the individual identities of the words or
morphemes are subordinated to the functional identity of the device as a whole.
2.
This list is not intended to be exhaustive; other rhetorical question types and sub-types
may be identified for Tamil, although I will not undertake to do so here.
3.
The examples in this paper (with the exception of examples (19) and (20), which are
transliterated directly from the written Tamil) are presented in phonemic transcription,
standardized to minimize individual and dialectical variations in pronunciation, while still
preserving characteristic Spoken Tamil forms. The system of transliteration is the same
as that used in the Tamil Lexicon (University of Madras, 1982) and other modern
references. Note should be made of the following diacritics: Underdashes indicate alveolar
phonemes, with the exception of _/, which is a retroflex continuant. Overdashes indicate
length for vowels, and the velar nasal . The other diacritics used a single dot under
retroflex sounds, and for the palatal nasal are standard.
4.
That this need not necessarily be the case, however, is indicated by Pope (1976:61, n. 13)
who cites J. Ross's example: 'Do we need this raise, after all?'. In this utterance both the
form of the question and its corresponding asssertion (e.g. After all, we need this raise')
are positive. A similar example from the Tamil narrative corpus is the following, the
response of some loyal servants to a request from their princess:
NT etu colriyo ceyrm?
'Whatever you say (to do), will we do it?'
( ='Of course. We will do whatever you say.')
The existence of 'double positive' CRQs in no way affects the present analysis.
5.
Readers unfamiliar with Tamil should note that the clausal subordinator , from
Written Tamil enrol ('say + COND'), is not related to the WH- words ea 'what' or
'why'. Nor should the compound conjunction ( + ) 'because' (discussed in
Section 5) be confused with ea 'what'. Aside from being distinguished by vowel length,
which is phonemic in Tamil, the two words have distinct derivational histories, the latter
having existed in its current form since the time of the earliest written records, while the
former is a relatively recent compound derived from a WH- word and an inflected form
of the verb en 'to say'.
6.
Note that in example (4) there is no finite verb in the surface realization of the clause
"embedded" by . Clauses with "deleted" or zero predicates constitute acceptable finite
utterances under certain pragmatic conditions in Tamil (cf. Herring, 1989).
7.
Characteristically rhetorical elements include adverbial phrases such as varu varusam

'year after year' in example (4), which are pragmatically odd in genuine questions, in that
they assert new information. (If such utterances are analyzed as narrative in function,
282
however, then the oddness disappears.) An example of an otherwise incompatible gram
matical element is the adverb kta 'even' in example (1), which normally would not
appear in a genuine question of affirmative structure, in that it presupposes a negative
response. For a discussion of similar elements in other languages, cf. Schmidt-Radefeldt
(1977) for English and German; Pope (1976) for English; and Grsillon (1981) for German
and French.
8.
'Rhetorical questions the interrogative character of which may be more or less clearly
marked may be interpreted as an appeal to the attention of the hearer. The speaker
who uses them demonstrates that he is continuously aware of the hearer's presence'
(Giilich, 1970:229).
9.
'By means of his question, speaker S makes an appeal to hearer H, and thereby establishes
an interpersonal connection' (Vandeweghe, 1977:279).
10.
The implication that the listener has posed the question may also be made explicit. In
place of , which has been effectively bleached of its literal meaning 'if (you) ask', the
expression u co, lit. 'if (you) ask, saying...', appears several times in the corpus, in
contexts where it can only be construed rhetorically, e.g.:
Avar ea pauvr u co,
he what do-F3RS QUOT say-coND
katai ellm euttu vaittiu, oru opatu maikku mle avar varuvr.
wares all
take store
one nine
o'clock after he come-F3RS
'What does he do, (you) ask? Taking up and storing his wares, he comes (to the
temple) after nine o'clock.'
11.
The notion of "ideal listener" evoked here is adapted from Fillmore's (1981) "ideal
reader",
who knows, at each point in the text, everything that the text presupposes at that
point, and who does not know, but is prepared to receive and understand, what the
text introduces at that point (p.253).
12.
For a discussion of the Villu Pttu genre in its cultural context, cf. Blackburn (1988).
13.
In this respect, CRQs differ crucially from TRQs and RTags, which most typically occur
in diegetic, or narrative portions of text. This distribution may well be responsible for the
fact that TRQs and RTags have extended clause-linking functions, while CRQs do not.
That is, unlike CRQs, TRQs and RTags are directly involved in relating the sequential
events of the narrative.
14.
The term 'information' is used here in a broad sense, to include both nominal reference
and verbal assertion (cf. Herring, 1989).
15.
Similarly, Schmidt-Radefeldt, in describing "the rhetorical use of question-answer

sequences" in German and English, comments:
On the strength of textual-pragmatic coherence (that the same speaker asks a question
and answers it all at once) such utterances end up by being one complex declarative
sentence (1977:379).
16.
A similar analysis is developed by stman (1981) for the English tag 'you know'.
17.
From "Enk, Yr, Yrukkkavo..." by Jeyakntan. In Guru Ptam. Madurai: Meenak

shi Puttaka Nilaiyam, 1971.

18.
283
"Amm Mantapam" by Sujt. In Cirukatai Elutuvatu Eppai. Madurai: Meenakshi

Puttaka Nilaiyam, 1975.
REFERENCES
Annamalai, E. 1980. "Some syntactic differences between spoken and written Tamil."
In South Asian Languages: Structure, Convergence and Diglossia, Bh. Krishnamurti
(ed.), 289-93. New Delhi: Motilal Banarsidass.
Blackburn, Stuart H. 1988. Singing of Birth and Death: Texts in Performance. Philadel
phia: University of Pennsylvania Press.
Chafe, Wallace L. 1980. "The deployment of consciousness in the production of a
narrative." In The Pear Stories: Cognitive, Cultural, and Linguistic Aspects of
Narrative Production, W. Chafe (ed.), 9-50. Norwood: Ablex.
DuBois, John W. 1987. "The discourse basis of ergativity." Language 63(4). 805-55.
Fillmore, Charles J. 1981. "Ideal readers and real readers." In Analyzing Discourse:
Text and Talk, D. Tannen (ed.), 248-270. Washington, D.C.: Georgetown Univer
sity Press.
Grsillon, Almuth. 1981. "Interrogation et interlocution." Documentation et Recherche
en Linguistique Allemande Contemporaine, Vincennes (D.R.L.A.V.) 25. 61-75.
Glich, Elisabeth. 1970. Makrosyntax der Gliederungssignale im gesprochenen Franz
sisch. Mnchen: Wilhelm Fink.
Halliday, M.A.K. and Ruqaiya Hasan. 1976. Cohesion in Spoken and Written English.
London: Longman.
Hart, George L. 1980. The Indian Audience: Folklore and Beyond. University of
California, Berkeley, ms.
Herring, Susan 1988. "Aspect as a discourse category in Tamil." Berkeley Linguistic
Society 14. 280-292. University of California, Berkeley.
Herring, Susan 1989. "Accounting for verbless sentences in Tamil narrative: A
discourse-functionalist approach." Paper presented at the 11th South Asian Lan
guages Analysis Roundtable, University of Wisconsin, Madison.
Hopper, Paul J. 1979. "Some observations on the typology of focus and aspect in
narrative language." Studies in Language 3(1): 37-64.
Hopper, Paul J. 1982. "Aspect between discourse and grammar." In Tense-Aspect:
Between Semantics and Pragmatics, P. Hopper (ed.), 3-18. Amsterdam: John
Benjamins.
stman, Jan-Ola. 1981. "You Know": A Discourse-Functional Study. Amsterdam: John
Benjamins.
Pawley, Andrew and Frances Syder. 1977. The One Clause at a Time Hypothesis.
University of Auckland, ms.
Pope, Emily N. 1976. Questions and Answers in English. The Hague Paris: Mouton.
Popovici, Vasile. 1981. "Dialogues rhtoriques." Degrs 28: 11-16.
Schmidt-Radefeldt, Jrgen. 1977. "On so-called 'rhetorical' questions." Journal of
Pragmatics 1(4).
ings: Some semantic-pragmatic aspects of grammaticalization." In Perspectives on
284
SUSAN . HERRING
Historical Linguistics, Winfred P. Lehmann and Yakov Malkiel (eds), Amsterdam:

John Benjamins.
Traugott, Elizabeth Closs. 1988. "Pragmatic strengthening and grammaticalization."
Berkeley Linguistic Society 14. 406-416. University of California, Berkeley.
Vandeweghe, Willy. 1977. "Fragen und ihre Funktionen. Versuch einer Typologie
auf pragmatischer Basis." Semantik und Pragmatik: Akten des II Linguistischen
Kolloquiums, Aachen 1976. 277-286.
Wehr, Barbara. 1984. Diskurs-Strategien im Romanischen. Tbingen: Gunter Narr.
Some Grammaticalization Changes in Estonian

and their Implications
Lyle Campbell
Louisiana State University
1.
INTRODUCTION
In this paper I am concerned with grammaticalizations in two areas of

Estonian grammar, with possible explanations for some of the changes, and
with their implications for theories of grammatical change in general. One is
the rise of a new category of modality in verbs {kaudne kneviis); the other
is the development of question markers. While much work on grammaticaliza
tion is of the top-down variety aimed at the 'big picture' with a broad
brush and bold strokes , such work has been criticized by some for what
they perceive to be lack of rigor. This paper is of the bottom-up type,
beginning with these concrete cases and examining their implications for
theoretical claims, i.e. for what they have to say about some of the broad
strokes. 1
2.
MODUS OBLIQUUS
Balto-Finnic languages have a number of participial constructions for

subordinate clauses of various sorts, and some of these have changed in
Estonian (and Livonian, as well) to create what is traditionally called 'Modus
Obliquus' forms. I begin by giving a thumbnail sketch of the change, followed
by concrete examples and more specific considerations.
The change involves two alternative 'complement' structures with
roughly the same meaning; it deals with the cases of speech-act (SAV) or
mental-state (MSV) main verbs {verba sentiendi et dicendi).
286
LYLE CAMPBELL
In Stage 1, two constructions are available:

a.
Main verb [SAV/MSV] ... et[complement] ... Finite verb
b.
Main verb [SAV/MSV] ... non-finite verb-ACTIVE.PARTCP
In Stage 2, three constructions become available, (a), (b), and:
Main verb [SAV/MSV] et Verb-ACTIVE.PARTCP

At this point the Verb-ACTIVE.PARTICIPLE of (c) was reinterpreted as a
finite verb form, called Modus Obliquus, associated with 'reported' speech,
with an 'evidential' function, where the speaker has not experienced the event
personally or does not wish to take responsibility for the accuracy of the
report called 'indirect' in this paper (called 'inferential' by Comrie,
1981:125).
In Stage 3, the reinterpreted Verb-ACTIVE.PARTICIPLE, now VerbMODUS.OBLIQUUS, came to be employed also in main clauses.
For a more detailed exposition, it will be helpful to begin, not with
Estonian, but with the corresponding constructions in Finnish, a closely
related language which illustrates the state of affairs before the change took
place. Finnish has the two original constructions (a. and b. above) with SAVs
and MSVs, in both the 'present' and 'past', as in 1-4:
(1)
kuul-i-n,
ett hn puhu-u
sii-t
hear-PAST-I that he speak-3.PRES.INDICATIVE it-about
'I heard that he is speaking about it.'
(2)
kuul-i-n
hne-n
puhu-an
sii-t
hear-PAST-I he-SG.GEN speak-PRES.PARTCP it-about
(same meaning as (1))
(3)
poika sano-i, ett is

kv-i
koto-na
boy say-PAST that father visit-PAST home-in
'The boy said that his father visited home.'
(4)
poika sano-i
is-
ky-neen
koto-
boy say-PAST father-SG.GEN visit-PAST.PARTCP home-in
(Same meaning as (3))
The participle constructions are also often employed in clauses which bear
the-sense of 'reported speech' or 'non-commitmen:
SOME GRAMMATICALIZATION CHANGES IN ESTONIAN

(5)
hn nky-y
h
asu-van
287
siell
appear-3.PRES.INDICATIVE live-PRES.PARTCP t h e r e
'It seems that he lives there'/'he seems to live there.'

(6)
hn kuulu-u
h
asu-neen
siell
is.heard-3.PRES.INDICATIVE Hve-PAST.PARTCP t h e r e
'They say that he has lived there'/'it is said that he lived there.'
The Estonian equivalents are:
(7)
(8)
sai kuul-da, et seal ks

mees
ela-b
got hear-iNF that there one.NOM man.NOM live-3.PRES.INDICATIVE
'She came to hear/she heard that a man lives there.'
sai kuul-da seal he
mehe
ela-vat
g o t hear-INF t h e r e One.GEN man.GEN Hve-PRES.PARTCP
'He came to hear / he heard (of) a man's living there.'

The innovative construction, with no Finnish counterpart, is seen in:
(9)
sai kuul-da, (et) seal ks

mees
ela-vat
got hear-iNF that there one.NOM man.NOM live-MODUS.OBLIQUUS
'He came to hear/he heard that (they say) a man lives there.'
(10) isa
tles poja-le, (et) ta sit-vat
father said boy-to (that) he travel-MODus.OBL
homme
linna
tommorrow town.into
'The father told the boy to travel to town tomorrow.'
Later, the Modus Obliquus forms (henceforth labelled 'indirect', INDIR) were
extended, occurring also in main clauses:
(11) ta
tege-vat
t-d
he.NOM do-PRES.INDIR WOrk-PARTV
'They say he is working.'

Forms with the former past participle are more common:
(12) ta
tei-nud
t-d
he.NOM do-PAST.INDIR WOrk-PARTV
'They say he worked'/'he worked, so they say.'
288
LYLE CAMPBELL
(13) naabri
perenaine ole-vat
linna
neighbor.GEN lady.NOM S-PRES.INDIRECT tOWn.intO
sit-nud
travel-PAST-PARTCP
They say the neighbor lady (lady of the house) has travelled to
town.'
Thus, Estonian has created an 'indirect' modality marker employed with
finite verbs from a former participle construction. The 'indirect' (Modus
Obliquus) forms came to be employed also in main clauses.2
3.
EXPLANATIONS
I now turn to proposed explanations of the Estonian changes. While in

recent years historical syntax has received much attention, there remains
considerable controversy with respect to the potential for explanation of
syntactic change, as seen in quotes from a recent book on historical syntax:
Nowadays it is often questioned whether explanation of linguistic change,
and particularly syntactic change, is possible at all. (Gerritsen, 1984:114)
The longer I deal with linguistics the more certain I am that language is
a simple phenomenon and that all true explanations of language phenom
ena are equally simple. (Maczak, 1984:242)
The truth must lie somewhere between these two extremes. Several 'tradi
tional' explanations for the origin of the Estonian Modus Obliquus have been
offered, and it is instructive to consider these they suggest that explanation
in this case is neither simple nor impossible. One is Grnthal's (1941), based
on a presumed ellipsis of the main verb (SAV or MSV), where it is assumed
that (14b), for example, derives from (14a):
(14a) naaber
tle-b ost-vat
kolm hobust
neighbor.NOM says buy-PRES.PARTCP three horses
'The neighbor says he is buying three horses.'
(14b) naaber
ost-vat
kolm hobust
neighbor.NOM buy-INDIR t h r e e h o r s e s
'They say the neighbor is buying three horses.'

Grnthal believed that the main verb, 'say' in this case (as in (14a)), was
understood, self-evident, and could therefore be left out (as in (14b)). The
289
documented stages of the change, however, do not support this interpretation,

and in any case, the semantic route from, for example, "the neighbor says he
..." to "they say/reportedly the neighbor ..." is not at all self-evident.
Kettunen's (1924) explanation was that the Modus Obliquus had devel
oped from the participle in subordinate clauses and then gradually the main
verb was lost. While Kettunen's proposal might seem sensible, the most widely
accepted proposed explanation appears to be Ikola's (1953) of 'contamina
tion', a mixture or confusion of the two constructions, (a) and (b) above, to
produce (c), the 'contamination' (blending) of the et complement and the
participial constructions. In this view, clauses with the et complement conjunc
tion require a finite verb; thus, the participle, when it occurred with et in the
'blended' construction (as in (c) above), was taken to be a finite verb form.
Such a view, however, hardly accounts for the grammatical function or
semantics acquired by this former participle which became a finite verb form,
the new Modus Obliquus ('indirect'). More to the point, since this new con
struction {et ... Verb-vat [Present] or et ... Verb-nud [Past]) occurred only
after speech-act and mental-state verbs, a sense of 'reported speech' or 'noncommitment' was attributed to the participle, even before the change, as is
seen even in the case of the Finnish examples with SAVs and MSVs (cf.
(4)-(6); Hakulinen and Leino, 1987).3
As an alternative to Ikola's widely-accepted 'contamination' explanation,
it is hypothesized that because the participles after SAVs and M SVs already
had an 'indirect' sense, these participles grammaticalized the 'reported speech'
function. This, in turn, permitted the reinterpreted former participles, now
'indirect' modality, to be used in contexts otherwise requiring a finite verb
form, including in contexts as main verbs. This proposal may be similar to
Kettunen's explanation, though with more emphasis on the discourse func
tion. The seeds of this hypothesis are seen in Hakulinen and Leino's (1987)
discourse account of developments in Finnish participial constructions.
Naturally, here, as in the case of most other changes in Balto-Finnic
languages, we are not to be spared proposed external explanations. On the
one hand, some have imagined Baltic (Latvian and Lithuanian) influence.
Latvian and Lithuanian both have a similar phenomenon, the so-called
'relative mood', which bears the suffix -ot and indicates that the speaker
doubts the truth of the statement or does not assume responsibility for its
truth. However, Baltic's Modus Obliquus appears, rather, to be best explained
as the result of Estonian and Livonian influence. (See Thomason and Kauf
man, 1988:243-4.) On the other hand, others see influence from the German
'conjunctive' (subjunctive), used in reported-speech contexts (cf. discussion in
290
LYLE CAMPBELL
Ikola, 1953:43; Hakulinen and Leino, 1987; Laanest, 1982:238-40). It is true

that German has had considerable influence on Estonian and Livonian. The
ultimate explanation, then, may need to be broad enough to include multiple
causation, 'grammaticalization' of an internally motivated, already available,
'reported speech' sense, spurred on by the existence of such a grammatical
category (conjunctive) in German, a language which has had much influence
on Estonian.
The questions are, then: 1. Did the form/structure appear first (through
'contamination') and then later acquire its particular meaning (as in Ikola's
position)? 2. Did the function already exist in the discourse context, permit
ting perhaps urging an independent form for signaling this function to
develop (similar to Hakulinen and Leino's position 1987; cf. Kettunen's
opinion)? 3. Is the Modus Obliquus due to language contact? Or, 4. Is its
origin perhaps due to a combination of these factors, say all of 1-3? It cannot
be solely due to German (or Baltic) influence, since the semantic seeds are
present in the corresponding Finnish constructions, which were not so influ
enced by these languages, and it clearly developed from native morphological
material, not borrowed forms. It cannot be due to the form alone, since the
participial complement is used also with other verbs (non-SAVs and nonMSVs), where it has maintained its original form and meaning. Thus, of the
proposals available, the most promising is the discourse explanation (2), or
multiple causation (4), which also encompasses (2).
4.
QUESTION PARTICLES
The second example deals with a series of changes in question markers,

especially in yes-no questions in Estonian. Old Estonian lost the Proto-BaltoFinnic interrogative suffix *-ko (cf. Finnish: tule-t-ko huomenna? [come-youQ tomorrow?] Are you coming tomorrow?'). This was replaced first with the
question particle es, illustrated in Older Estonian (from 1696 New Testament,
Alvre, 1976:345):
(15) Nd es tee
now
uSSute
Q yOU.PL believe.yOU.PL
'Now do you believe?'

(16) MiSt es minna Sedd pe
tundma
it.from Q I.NOM it
should know.iNF
'How should I know that?'
291
In earlier times, this particle was a clitic/suffix; it developed in the following

way. First, the modern Estonian particles (independent words), such as es
'question particle' and ep 'emphasis', were once bound forms, which became
'decliticized' or 'lexicalized' in Alvre's (1976:345) terms at roughly the
same time, due in part to phonological developments. Final vowels were lost
(apocopated). However, when the clitics *-pa 'emphatic' and *-(ko)-s '(quest
ion-informal speech, [note: -s is a pan-Balto-Finnic 'clitic' used to signal
informal speech, casualness] were attached, the final root vowel was no longer
in word-final position and so was protected from apocope, e.g.: kelt 'from
whom?' > kelt, but keits > keltes; pll 'on (top of)'>pll, but pllp>pll-p. Apocope applied, as expected, to the clitic -p, giving -p; vowel
harmony was lost; and non-initial changed to e, ultimately yielding peallep.
This vowel-loss rule left many apocopated stems, with less common alternant
forms ending in a -V when the clitics were attached. Given the frequency and
salience of the apocopated stems, these forms with attached clitics were ripe
for reinterpretation. The morpheme boundary was reinterpreted; the V was
considered to be part of the clitic: kelte-s > kelt-es; peale-p>peal-. With
vowel harmony lost, there was no longer evidence that -es and -ep were
phonologically dependent upon other words, rather than independent (though
unstressed) particles. They were reinterpreted, lexicalized as specific lexemes,
as independent words. (Cf. Ariste, 1973; Alvre, 1976.)
The 'decliticization' which created the new 'affirmative adverb' ep as an
independent word, like the new question word es, also goes against the
expected unidirectional change from independent word to clitic/affix (cf.
Ariste, 1973; Alvre, 1976 for details), discussed below. Therefore, it is worth
citing Ariste's (1973:37) abstract with regard to the creation of this indepen
dent word:
Estonian has an affirmative adverb ep 'yes, indeed, just so, then', which
is unknown in the other Balto-Finnic languages. This adverb ... has
developed from the unproductive affirmative suffix -p, -ap, -p, Finnish
-pa, -p. Such words as sestap, sestep 'therefore indeed', miksep 'why
then', siisap, siisep 'right/just (so)', seep or seep {see on) 'that is it indeed',
the modern bearers of the Estonian language have unlinked [zergliedert]
metanalytically, that is, falsely: sest ep, miks ep, siis ep, see ep. After this
suffix was lexicalized, the word ep could change its syntactic position and
precede the affirmed words: see ep ep see. (My translation, LC.)
The original value of the clitic -s was to signal 'informal' speech, and was
with great frequency attached to the Balto-Finnic question suffix. The vowel
was optionally lost in certain Balto-Finnic languages, as in the informal
292
LYLE CAMPBELL
question suffix, -ks, in colloquial Finnish, which itself became optional in

Estonian, leaving the alternation of -ko-s, -ks and -s as signals of questions
(particularly for informal questions). Alvre (1981:30) explains the remaining
residue of -ks in Estonian as follows:
The remnants of the Estonian ks-type are not especially numerous. The
spread of the question word kas has had a disintegrating effect on this
category. The suffix *-kos, which extends back to Proto-Balto-Finnic, has
shortened to ks in the separate development of Estonian. The same path
of evolution is presently current in the Finnish language, where in the
colloquial language vielks 'still?', onks 'is [it]?', oliks 'was [it]?', etc.
compete with the standard language shapes vielk (s), onko(s), oliko (s),
etc. (My translation, LC.)
Thus, this question particle (es) in Estonian ultimately derives from a suffix
signaling informal speech (-s), which became independent and came to be
used in conformity to Wackernagel's principle of clitics in sentence second
position. (See Alvre, 1976, 1981; Nevis, 1984, 1985a, 1985b:lll-3, 1988, In
press.)
Later, the Estonian es 'question particle' (now archaic and dialectal) was
again replaced, this time by kas, illustrated in:
(17) kas ta tuli?
Q he came
'Did he come?'
(18) kas ta ei tulnud?
Q he NEG came
'Didn't he come?'
It has been documented through the careful investigation of Estonian histori
cal documents that this particle developed from what was originally a postpo
sition ka(:)s 'with' (e.g. mehe ka(:)s [man.GEN with] 'with the man'), which
later gave rise to adverbs meaning 'also, and', e.g.: Are you going with>are
you going also > and/also are you going > Q are you going? It is from this
sense that this question particle eventually arose, e.g. from sentences such as,
'and [kas], are you going', the kas acquired a sense and function of introducing
questions (cf. Alvre, 1976, 1983; Nevis, 1988, In press for details).
Initially, questions introduced with kas could appear only in the affir
mative; there were no examples with negation. Later, this restriction disap
peared and kas came to be employed regularly also with negative verbs, as
for example in (18). (For details, see Alvre, 1983; Nevis, 1984, 1985a, In
press.)
5.
293
IMPLICATIONS
Whatever the ultimate explanation for these grammatical changes, they

are clear and well-documented, with implications for a number of general
claims about grammatical change. It is to these that I now turn. While some
of these are of much less interest than others, they illustrate the approach
being advocated here, to consider what these concrete examples have to say
about various claims concerning grammatical change.
1. Syntactic change affects main clauses before subordinate clauses; the
converse does not occur. This claim, or close paraphrases thereof, has been
made by several scholars, starting with Biener, 1922a, 1922b (compare Givn,
1971, 1984; Lightfoot, 1981:228, etc.).4 The general idea involved is the belief
that change starts "in main clauses and may or may not ultimately come to
affect subordinate clauses, but that it never (or usually does not, in a weaker
version) begin in subordinate clauses, later reaching main clauses. In its strong
version, this claim is clearly falsified by the Estonian Modus Obliquus example,
where the change from participle to 'indirect' finite verb clearly began first in
lower verbs and later appears in main clauses.
2. Related to the first claim, but not stated so generally, is: "The more
dependent the SUB-clause is semantically/pragmatically on the MAIN-clause,
the less likely are independently-expressed [tense-aspect-modality]
markers to appear in the SUB-clause. The uncoded categories of the
SUB-clause may then be inferred, in some systematic fashion, from semantic
or pragmatic information given in the main clause or in the discourse context"
(Givn, 1984:315). Again, the Estonian Modus Obliquus ('indirect') goes
against this claim, in that category developed became expressed
independently in subordinate clauses, in spite of (or perhaps because it
was aided by) its predictability in this context.
Nevertheless, there appears to be value to both these claims. That is, in
general, subordinate clauses do contain fewer morpho-syntactic contrasts
than main clauses, suggesting that on the whole fewer changes would begin
in subordinate clauses, since main clauses would normally also already have
most of the essential morpho-syntactic trappings of the subordinates, plus
additional things not found in lower clauses. With respect to the second claim,
the foreground-background distinction (while somewhat imprecise) plays an
important role. Subordinate clauses and irrealis categories tend to
correlate in discourse with 'backgrounded' material (cf. Givn, 1984:288).
This being the case, it is not so surprising that the Estonian 'indirect' (an
irrealis form) should have developed in subordinate clauses.
294
LYLE CAMPBELL
3. Changes in structure may affect the syntax of grammatical relations

before the morphology that encodes them, with the result that morphology
may reflect a previous syntactic situation (cf. Anderson, 1980; Comrie, 1980;
Givn, 1971, 1984; Heine et al., This volume, etc.), or in Givn's (1984:34)
words, "diachronic change in syntax often removes a particular structure
from its original functional domain while at least initially leaving the
structure itself relatively intact." Heine et al. (This volume) express it as,
"Since conceptual shift precedes morphosyntactic and phonological shift, the
result is asymmetry between meaning and form ... languages ... show exam
ples of morphemes or constructions which have acquired a new meaning or
function although they still retain the old morphosyntax." This claim appears
to hold for our Estonian Modus Obliquus example. While the morphological
form, the participle (-vat or -nud), remains unchanged, its grammatical status
is completely reinterpreted in these constructions. However, the development
of the question particle es from a bound 'clitic' -s would seem to go against
the thrust of this claim; surely the 'decliticization' change in morphological
status took place before the form was reinterpreted as a question marker
to be employed in second position. 5
4. The extension of the grammatical function of a morphological marker
proceeds by the removal of conditions on the rules that assign the marker
(Harris, 1985:382-4; cf. also Plank, 1980). This claim would appear to be
related to the last one; the Estonian changes appear to conform. The extension
of the -vat and -nud participles to Modus Obliquus status is reflected in removal
of the constraint first that they should not appear after the et complement
conjunction, then later, that they should not appear in main clauses. Similarly,
the earlier constraint against kas questions in negative clauses was later
removed.
5. The earlier restriction against kas in negative sentences conforms to
Givn's (1975:94) and Disterhefs (1980:114) claims that syntactic change
may be actualized in affirmative clauses before negative ones; the converse
does not occur.
6. A weaker sort of claim is that grammatical morphemes usually
(always?) eventually arise out of lexical words through semantic bleaching
and phonological reduction (cf. Bybee and Pagliuca, 1987; Givn, 1984:48;
Langacker, 1977; etc.) While in general there can be no quarrel with this
claim, it is nevertheless the case that the examples considered here do not
come from lexical items, but rather from the (re)grammaticalization of partici
ples, bound clitics, and postpositionals. Perhaps more room should be allowed
295
for the origin of new grammatical morphemes from other, not-so-lexical

resources, as well.
7. There is a very widely accepted claim with something like the following
form: In language change, independent words tend to lose their boundaries,
cliticize, and/or become bound morphemes; the reverse does not happen (or is
very rare indeed) bound morphemes do not (do only rarely) split and
become independent elements (cf. Comrie, 1980; Givn, 1971, 1984; Langacker, 1977, etc.). Again, few would want to quibble with this claim, at least
not in its weaker form. Nevertheless, in the development of the question parti
cle es, Estonian has done exactly what is not supposed to be done (or
done only very infrequently), reanalyzing a suffix and separating it off as an
independent form. Since this is an important point, perhaps surprising to some
supporters of the claim, it is worth pointing out that other examples of 'decliticization' (or 'upgrading' in Nevis' terminology), while not frequent, are reason
ably well-known (cf. Janda, 1981; Nevis, 1984, 1985a, 1985b, 1988, In press).
8. Langacker (1977:66) claims that there is a general tendency in lan
guage, revealed in grammatical change, to bring morpheme boundaries into
line with syllable boundaries. While at some level there may be something to
this claim, Balto-Finnic languages abound in morphological changes which
establish morpheme boundaries at odds with syllable boundaries. The origin
of the Estonian question particle es is a case in point. From forms such as
p-ll-s 'on top o f [syllables pl.ls], the development of an independent
es is hardly to be expected from Estonian syllable structure, though morpheme
structure may well hold clues in this case.
9. A claim that one can infer from some writings goes something like
this: When a language changes from postposed to preposed grammatical
elements, the preposed elements are usually drawn from pre-exisitng elements
which had different functions; when, on the other hand, a language changes
from preposed to postposed, it often uses the same elements in the same
functions (compare M. Harris 1978, A. Harris 1985). Should this claim have
any measure of validity, the development of the Estonian sentence-initial
question particle kas from the postposition *kans[sJaX 'with' might prove to
be an example.
6.
CONCLUSIONS
I hope that the phenomena considered in this paper, together with their
implications for claims about grammatical change, has illustrated the value
of the bottom-up approach.
296
LYLE CAMPBELL
ACKNOWLEDGEMENTS
This material is based u p o n work supported by the N a t i o n a l Science F o u n d a t i o n
under G r a n t N o . BNS8712240; any opinions, findings, and conclusions or recommen
dations expressed in this publication are those of the author and d o not necessarily
reflect the views of the National Science F o u n d a t i o n .
NOTES
1.
While there are a variety of definitions for grammaticalization about, I intend my use of
the term to include not only change from lexical status to more grammatical, but also
the creation of new grammatical elements from already existing ones (cf. Kuryowicz,
1965:52).
2.
Auli Hakulinen (personal communication) has expressed reservations concerning the

discussion of 'main' and 'subordinate' clauses in treatments of Balto-Finnic non-finite
participial forms, since on the surface the participle appears as a constituent of the main
(and only) clause, and not obviously as a subordinate clause. While this is important to
keep in mind, apparently the theories considered here (both formal and functional) do
consider non-finite participials to be non-main clauses (cf. particularly their treatments
of preposed relative clauses without finite verb forms), at least with respect to their claims
about the nature and directions of grammatical change, which is my main concern with
these examples in this paper.
3.
Some of these examples, though clearly not all, illustrate the kind of "complex clauses"
which are "obligatorily irrelis in their modality" (Givn, 1984:286), discussed later in
this paper.
4.
In this context Lightfoot (1981:227), discussing Givn's ideas and relating them to
Emonds' notion of structure-preserving root transformations, concludes essentially the
same thing:
It will follow ... that non-structure-preserving innovations will enter the language first as
root transformations, affecting just root sentences, and only later percolating through the
grammar to affect the phrase structure rules and thus the structures in embedded clauses.
5.
Balto-Finnic offers a more telling exception; clearly changes in the form can precede and
trigger changes in the syntax, as can be seen in the well-known and related example of
Finnish participial embedded clauses (cf. Anttila, 1972; Breckenridge and Hakulinen,
1976; Hakulinen and Leino, 1987; Ikola, 1959; Svensson, 1983; Timberlake, 1977). Briefly,
historically Finnish had constructions equivalent to the following examples, which are
presented in their modern form for ease of understanding except for the crucial informa
tion about case endings, which later changed:
(1)
nin poja-m
juokse-va-m
saw.I boy-SG.ACC run-PRES.PARTCP-SG.ACC
'I saw that the boy runs/I saw the boy running.'
(2)
nin
poja-i
juokse-va-t
saw.I boy-PL.ACC run-PRES.PARTCP-PL.ACC
'I saw that the boys run/I saw the boys running.'
297
However, with the sound change of *-m>-n, the 'singular accusative' and the 'singular
genitive' became homophonous, both -n. The construction was reinterpreted from a main
verb with its direct object and a participial attribute, both in the accusative case, to a
verb with a participial complement, with the noun as a genitive subject of the participial
complement, i.e.:
(3)
nin poja-
juokse-van
S a w . I boy-SG.GEN run-PRES.PARTCP

This grammatical change is seen clearly with plurals, when one contrasts (2) with (4):
(4)
nin poik-i-en juokse-van

S a w . I b o y - P L - G E N run-PRES.PARTCP

Thus, the morphological change both preceded and triggered the syntactic change in this
case, showing that the claim cannot be framed in the strongest possible form.
REFERENCES
Alvre, Paul. 1976. "Vana kirjakeele ksisnu" [Old standard Estonian question words].
Keel ja Kirjandus 19:343-50.
Alvre, Paul. 1981. " Veelaks-tuun kirjakeeles ja murdeis" [The Veelaks 'still?' type in
the standard language and in the dialects]. Keel ja Kirjandus 24:24-30.
Alvre, Paul. 1983. "Zur Herkunft der Wrter kas und teps in der estnischen Sprache."
Soviet Finno-Ugric Studies 19(2):81-89.
Anderson, Stephen R. 1980. "On the development of morphology from syntax." In
Historical Morphology, Jacek Fisiak (ed.), 51-70. Mouton: The Hague.
Anttila, Raimo. 1972. An Introduction to Historical and Comparative Linguistics. New
York: Macmillan.
Ariste, Paul. 1973. "Eesti rhumrsna 3p" [The affirmative adverb ep in Estonian].
Journal de la Socit Finno-ougrienne 72:33-7. Helsinki: Suomalais-Ugrilainen
Seura.
Biener, Clemens. 1922a. "Zur Methode der Untersuchungen ber deutsche Wortstel
lung." Zeitschrift fr deutsches Altertum und deutsche Literatur 59:127-44.
Biener, Clemens. 1922b. "Wie ist die neuhochdeutsche Regel ber die Stellung des
Verbums entstanden?" Zeitschrift fr deutsches Altertum und deutsche Literatur
59:165-79.
Breckenridge, Janet and Auli Hakulinen. 1976. "Cycle and after." In Papers from the
Parasession on Diachronie Syntax, Sanford B. Steever, Carol A. Walker, and
Salikoko S. Mufwene (eds), 50-68. Chicago: Chicago Linguistic Society.
Bybee, Joan L. and William Pagliuca. 1987. "The evolution of future meaning." In
Papers from the 7th International Conference on Historical Linguistics, Anna G.
Ramat, Onofriu Carruba and Giuliano Bernini (eds), 109-22. Amsterdam: John
Benjamins.
Comrie, Bernard. 1980. "Morphology and word order reconstruction: Problems and
Prospects." Historical Morphology, Jacek Fisiak (ed.), 83-96. The Hague: Mouton.
298
LYLE CAMPBELL
Comrie, Bernard. 1981. The Languages of the Soviet Union. Cambridge: Cambridge
University Press.
Disterheft, Dorothy. 1980. The Syntactic Development of the Infinitive in Indo-Euro
pean. Columbus, Ohio: Slavica.
Gerritsen, Marinei. 1984. "Divergent word order developments in Germanic lan
guages: A description and a tentative explanation." Historical syntax, Jacek Fisiak
(ed.), 107-36. Berlin: Mouton.
Givn, Talmy. 1971. "Historical syntax and synchronic morphology: An archaeolo
gist's field trip." Chicago Linguistic Society 7:394-415.
Givn, Talmy. 1984. Syntax: A Functional-typological Introduction, Vol. I Amsterdam:
John Benjamins.
Greenberg, Joseph H. 1963. "Some universais of grammar with particular reference
to the order of meaningful elements." In Universais of Language, Joseph H.
Greenberg (ed.), 73-113. Cambridge, Mass: MIT Press.
Grnthal, W. 1941. Itmerensuomalaisten kielten yksikn nominatiivi objektin edusta
jana aktiivin yhteydess. [The nominative singular of Balto-Finnic languages as
representative of the object in active contexts.] Helsinki.
Hakulinen, Auli and Pentti Leino. 1987. "Finnish participial construction from a
discourse point of view." Ural-Altaische Jahrbcher 59:35-43.
Hale, Kenneth. 1973. "Deep-surface canonical disparities in relation to analysis &
change: An Australian example." In Current Trends in Linguistics, Thomas Sebeok
(ed.), 11:401-458. The Hague: Mouton.
Harris, Alice C. 1985. Diachronie Syntax: The Kartvelian Case. New York: Academic
Press [Syntax and Semantics 18].
Harris, Martin B. 1978. The Evolution of French Syntax: A Comparative Approach.
London: Longman.
Heine, Bernd, Ulrike Claudi and Friederike Hnnemeyer. This volume. "From cogni
tion to grammar: Evidence from African languages."
Ikola, Osmo. 1953. "Viron ja Liivin modus obliquuksen historiaa." [On the history
of the oblique mode of Estonian and Livonian.] Suomi 106(4). Helsinki: Suoma
laisen Kirjallisuuden Seura.
Ikola, Osmo. 1959. "Erist suomen syntaktisista siirtymist." [On certain Finnish
syntactic changes.] Suomen Kielen Seuran Vuosikirja 1:39-60. Helsinki.
Janda, Richard. 1981. "A case of liberation from morphology to syntax: The fate of
the English genitive-marker (e)s." Syntactic change. Natural Language Studies
25:59-114.
Kettunen, Lauri. 1924. Lauseliikmed eesti keeles. [Constitutent Structure in Estonian.]
Tartu.
Kuryowicz, Jerzy. 1965. "The evolution of grammatical categories." Esquisses Lin
guistiques II, J. Kuryowicz, 38-54. Munich: Fink.
Laanest, Arvo. 1982. Einfhrung in die ostseefinnischen Sprachen. Hamburg: Buske
Verlag.
Langacker, Ronald W. 1977. "Syntactic reanalysis." In Mechanisms of Syntactic
Change, Charles N. Li (ed.), 59-139. Austin: University of Texas Press.
Lightfoot, David. 1981. "Explaining syntactic change." In Explanation in Linguistics,
N. Hornstein and D. Lightfoot (eds). London: Longman.
Maczak, Witold. 1984. "If I was instead of if I were." In Historical Syntax, Jacek
Fisiak (ed.), 237-46. Berlin: Mouton.
Nevis, Joel A. 1984. "A non-endoclitic in Estonian." Lingua 64:209-24.
299
Nevis, Joel A. 1985a. "Language-external evidence for clitics as words: Lappish

particle clitics." Chicago Linguistic Society 21:289-305.
Nevis, Joel A. 1985b. Finnish Particle Clitics and General Clitic Theory. Ann Arbor:
University Microfilms International. Ohio State University Ph.D. dissertation.
Nevis, Joel A. 1988. "On the development of the clitic postposition category in
Estonian. Finnish-Ugrische Forschungen 68:171-97.
Nevis, Joel A. In press. "Dediticization in Old Estonian." Ohio State University
Working Papers in Linguistics.
Plank, Frans. 1980. "Encoding grammatical relations: Acceptable and unacceptable
non-distinctness." In Historical Morphology, Jacek Fisiak (ed.), 289-326. The
Hague: Mouton.
Svensson, Pirkko Forsman. 1983. Satsmotsvarigheter i Finsk prosa under 1600-talet:
Participialkonstruktionen och drmed synonyma icke-finita uttryck i jmfrelse med
sprkbruket fre och efter 1600-talet. Helsinki: Suomalaisen Kirjallisuuden Seura.
Thomason, Sarah G. and Terrence Kaufman. 1988. Language Contact, Creolization,
and Genetic Linguistics. Berkeley: University of California Press.
Timberlake, Alan. 1977. "Reanalysis and actualization in syntactic change." In Mecha
nisms of Syntactic Change, Charles N. Li (ed.), 141-77. Austin: University of
Texas press.
The Last Stages of Grammatical Elements:

Contractive and Expansive Desemanticization
Joseph H. Greenberg
Stanford
University
The present study resumes and expands some of the topics discussed in
Greenberg (1978) on the stages of the definite article and Greenberg (1981)
which is devoted to two additional instances of Stage III articles, NiloSaharan k- and Penutian -s.1
I will be concerned with the final stages of these and certain other
grammaticalized elements. Here, as in the earlier stages one finds concomitant
phonological, semantic and distributional changes. Phonologically there is a
tendency to reduction. Semantically, investigators have often mentioned
desemanticization or loss of meaning. This is a complex phenomenon which
cannot be separated from what are here called distributional characteristics,
that is, the range of environments in which the item occurs. One tendency is
for contraction to the point at which it is found in fossilized form in only a
few lexical forms. We can then say that it is lexicalized in the sense that
synchronically it is a part of the host morpheme. We may also say that it is
desemanticized in the sense that it can no longer be assigned a meaning.
Another alternative is for it to expand its distribution, initially by semantically
motivated extensions, but in an increasingly arbitrary way so that its meaning
becomes highly disjunctive and even a prototypical definition cannot be
readily formulated. In a quite different way, then, than with contractive
lexicalization, increasing extension leads to zero intension, so that the item
has become desemanticized. There is, however, with expansive lexicalization
a further possibility, namely its reinterpretation in a new function. In all cases
known to me the new meaning is grammatical. We may call this process
regrammaticalization.
More concretely I propose to study, even if only in a preliminary fashion,
some phenomena whose consideration derives quite naturally from earlier
work on the development of the definite article. This usually originates from
302
JOSEPH H. GREENBERG
a demonstrative, and the later stages of such an article as found in a number

of languages I have called the Stage II and Stage III article.
One topic I will discuss concerns further developments of the anaphoric
pronoun. As is well known the anaphoric pronoun in almost all instances
derives from a demonstrative just as does the article. In the instance of the
Romance languages the anaphoric pronoun and the article both derive from
the same distance demonstrative ille.2 I shall outline some instances in which
the anaphoric pronoun goes through stages reasonably comparable to that
of the definite article although this is of less frequent occurrence than in the
case of the article. Moreover, in its final stages it may undergo significant
reinterpretation, which I believe goes beyond what is usually called reanalysis
in discussions of grammaticalization.
Along with this we find a typical sort of variation which for want of a
better term I shall call random variability and which is once more quite
comparable to that of the Stage III article, for example, Nilo-Saharan k-.
The development of third stage articles from the beginnings of a definite
article in a demonstrative takes a long period of time. For example, in the
historically attested case of Aramaic we can trace the development of a
suffixed -a through the later stages of the article over a period of approxi
mately 3,000 years. Yet its origin in a demonstrative -<-h, accepted by
Semitists on comparative grounds, precedes our earliest written records of
the language. During such a long period most languages will break up into
separate dialects and languages several times. The phenomenon here called
random distribution is encountered when such languages or dialects are
compared.
A well attested instance is Nilo-Saharan k- which is found almost exclu
sively in nouns. 3 What I here call random variability takes the following
forms. If we call the form with prefixed the articulated form and the one
without it the unarticulated, then we may summarize the sort of phenomena
encountered as follows:
1.
2.
3.
The same dialect or language sometimes shows the articulated and

non-articulated form of the same stem in free variation. 4
Even among closely related languages of the same subgroup we
frequently find the articulated form in one and the non-articulated
in the other.
More broadly within the same linguistic stock there is hardly a
form which does not show some variation among the languages in
this regard so that comparativists reconstruct both forms as original
THE LAST STAGES OF GRAMMATICAL ELEMENTS
4.
303
and as variants in the ancestral language. The contribution of

diachronic typology is to show how these variants arise in the
process of grammaticalization particularly in its later stages.
Finally one may note that certain languages seem in general to
prefer either the articulated or non-articulated form. Thus in the
Nilotic languages, a subbranch of Nilo-Saharan, the Western
Nilotic languages such as Dinka and Shilluk show initial k- much
less frequently than Eastern Nilotic languages. For example in the
word for 'cow' Shilluk (West Nilotic) has singular dhean whereas
Masai (East Nilotic) has (en)-ki-terj in which en- is a new second
stage article.
It will be shown that this sort of random variability is found in later

stages of other grammatical forms. Moreover, basically similar alternations
are found in variants of phonological origin, e.g. in the breakdown of vowel
harmonic systems. Whereas for grammatical formatives the variation will be
that between presence and absence of the element, in phonological cases it
will rather be the choice between two phonological alternants which are no
longer phonologically conditioned in any clear way. In at least one well
known instance, so called moveable s- of Indo-European, which appears
initially in a random fashion on roots beginning with an unvoiced stop, liquid
or nasal, opinions are divided as to whether the origin is phonological, as the
outcome of word-sandhi, or whether it is morphological. As with grammatical
items, there is the possibility of interpretation of phonological items as having
a grammatical significance, e.g. umlauting in German. This might be called
grammaticalization from below. It is, compared to the grammaticalization of
lexical items, relatively infrequent and often a subsidiary method which
accompanies others of the more common type. Nevertheless, its properties as
compared to that of grammaticalization of the more usual type are I believe
well worth exploring. This is not attempted here.
In general I think it will be fruitful to consider within the same basic
frame of grammaticalization, processes of development of grammatical ele
ments from all sources. These will include the origin of grammatical elements
from morphemes of more concrete meaning by semantic change, the grammat
icalization of variants of phonological origin and the third major source,
reanalysis with morpheme boundary shift. An example of the latter is the
German neuter plural in -er in which the r<*z, a variant of s arising by the
operation of Verner's law, was formerly the formative of s-stems which was
304
JOSEPH H. G R E E N B E R G
reinterpreted as a plural marker after the loss of final vowels had produced
an unmarked plural.
Finally a somewhat speculative attempt will be made to help explain the
phenomenon of random variation whether in phonology or grammar which
I believe are similar in their sociolinguistic complications regarding language
variation.
As a point of departure, I will briefly recapitulate the particular points
in Greenberg (1978) which have a bearing on the present argument. A definite
article which almost always derives from a distance or unmarked demonstra
tive may go through two further stages which I have called the Stage II and
Stage III article. A Stage II article arises from the usual definite article by
adding, roughly speaking, the uses of the indefinite article thus combining
the uses of both. 5 Along with this there is incorporation of the article in the
noun giving rise to a prefixed or suffixed article. Because of the high frequency
of combined definite and indefinite uses it becomes the "normal" form of the
noun and the one that is usually elicited when an investigator asks for the
lexical equivalent of a noun in his or her own language. However at this
stage, unarticulated forms still survive in functional, or sometimes convention
alized alternations resulting from the analogical spread of originally purely
functional distinctions.
On the one hand, the non-articulated form typically survives in forms
which are automatically definite and therefore did not require articles in the
first stage. Among these are common nouns, which when used as proper
names of either persons or place will frequently show the unarticulated form,
vocatives, and nouns with demonstrative modifiers. On the other hand, we
find generic uses such as the incorporated noun object, nominal predication,
dependent genitives in compounds and adverbial, particularly locative, uses
e.g. 'at home', 'on foot'.
Stage III is defined by the absence of functional contrasts such as those
still found in Stage II. At this point the articulated form will spread to
virtually all nouns, or less commonly, the articulated variant will be ousted
everywhere by the non-articulated variant. Thus, in regard to the Bantu preprefixes, which are to be interpreted as Stage II articles, and which are
common in East Africa, there are instances in which in genetically closely
related languages one has retained the pre-prefix and the other has lost it.
Thus Bisa, a Bantu language of Zambia, has retained it, while Ila, a member
of the same Bantu subgroup and also spoken in Zambia, has lost it, e.g. Bisa
u-mwezi, Ila mweze 'moon'.
305
It should be emphasized that this whole sequence of events is a gradual

process, and that each stage presents a cluster of characteristics which are to
be viewed as prototypes.
There are two basic classes of instances. If the language to begin with
had noun classes and the demonstratives, which become articles, are also
classifying elements that belong to the same system, then the process becomes
one of renewal of a class system the inherited markers of which had become
phonetically eroded.6
Where there was no system of nominal classifications and the new
markers do not introduce one, we simply find the articulated as the normal
form of a considerable number or of virtually all nouns. There is in this case
the possiblity of reeinterpretation of the articulated form and the attribution
to it of new functions.
The most common and grammatically significant new function is that
since, in fact it exists on virtually all nouns, it becomes a sign of nominality
and becomes a productive morpheme which derives nouns from verbs and
from adjectives in languages in which adjectives when predicated are to be
considered a special subclass of stative verbs.
A far less frequent and occasional function is that of distinguishing
homonyms through the use of the articulated versus the non-articulated
forms. Another sporadic use is that of distinguishing singularity from plurality
or, more likely, singulative from collective. In the case of Nilo-Saharan, this
occurs in individual instances in some languages, e.g. Kanuri k-am 'person',
am 'people'. There are some languages, however, in the Eastern subgroup of
Nilotic in which this is a fairly common formation, e.g. Karimojong (e)kilen 'knife'; (ni)lenia. I believe that a likely basis for this usage, which is also
found in sporadic instances with Penutian -s, is that the collective is very
much like the generic in that it is viewed as a unity and often requires
additional grammatical mechanisms to distinguish units as against the undi
vided whole.7
We now turn to the question of possible further stages of the anaphoric
pronoun which might parallel that of the article. I have by now found several
instances in Amerind languages as well as in Northeastern Asia of a very
similar sequence of events in the development of the anaphoric pronoun
which show a fairly obvious similarity to later stages of the article.
I shall treat briefly one of these examples and only as it appears in
Chibchan-Paezan where it develops from /, the common third person singular
anaphoric, generally as a bound form, prefixed as a possessive on the noun
and the subject or object on the verb.8 We will be concerned here with
306
JOSEPH H. GREENBERG
developments as verb object and as possessive. As a verb object it sometimes

comes to be used as indefinite specific in the following way. It will appear on
transitive verbs where there is no overt object but will be absent if the object
is present. Thus in a narrative we might have something like 'He came, he
ate and he went away'. In a language of this sort, a pronominal object which
might be glossed as 'something' is expressed by a third person marker. If, on
the contrary, the narrative account proceeded as follows: 'He came, he ate
beans and went away', then no pronominal object would appear on the verb
'to eat'.
In Itonama, a Chibchan language of Bolivia, this stage is clearly
illustrated in Camp and Liccardi (1967) in which z'- is described as an unspeci
fied object. There are in this language, in addition to ordinary objects, a
whole series of incorporated objects, mostly body parts. When none of these
is present then z- is compulsory, e.g. mas-t?it?ye 'cut the arm' (literally armcut), chas-dotye 'grain-pound' versus i-t?it?ye 'to cut' (unspecified object) and
i-dotye 'to pound' (unspecified object). For Bribri, a language of the Talamanca branch of Chibchan spoken in Costa Rica, Lehmann (1920:285)
describes a similar situation. "When no object is mentioned the verb has z'-."
There are still other Chibchan languages for which this usage holds, which
are not discussed here. The next development is that, since z'- can only be
used with transitive verbs, it becomes a derivational element which transitivizes intransitive verbs. The mechanism involved here is essentially similar to
the reinterpretation of the Stage III article as a nominalizer. In Warrau, a
Paezan language of Venezuela, the z'- prefix functions not only as a transitivizer
but also as a causative (Barrai, 1957:92), e.g. miki 'see', imiki 'show'. This is,
of course, a common situation with regard to transitivizers. In English 'hang'
(tr.) may be paraphrased as 'cause to hang (intr.)'.
The same marker i- prefixed to nouns, originally as third person inalien
able possessor, goes through a different though analogous course of change.
After his description of Bribri z'- as an indicator of indefinite object, Lehmann
goes on to cite, for the same language, its development, generally in body
part nouns, to a generalized meaning. He cites a similar development of z'- in
Rama in which, as he states it, "z'- before the noun indicates a general form
i-nut his, her face = face in general" (ibid.). Here we have a change to general
indefinite rather than specific indefinite, i.e. 'anybody's face' rather than
'someone's face.'
Whereas I have not noted any instances in regard to z'- in the verb of
random variation across or within dialects or languages, this does happen in
regard to the noun. 9 For example, in Chibchan there are a number of instances
307
in which forms with the z' prefix are reinterpreted as belonging to the stem so
that i-ba simply means 'blood' and i-ta 'hand'. For definite third person
possessor one rather finds a-, e.g. -ba 'his/her blood' with a marker a- which
is widespread in Chibchan-Paezan and always marks third person definite
possessor (Greenberg, 1987:286). On comparative grounds it must be more
recent than z'-. We frequently find cognates across languages in which some
languages have z'- and others do not. For example, corresponding to Chibchan
iba 'blood' is Murire (Guaymi subgroup) bea, Estrella (Talamanca group,
Costa Rica) pe and Mura, a Paezan language of Brazil be all with the meaning
'blood'. For Timucuan, a Paezan outlier in Florida, Gatschet (1880:485) gives
the variant forms chini and ichini for 'nose'. This is a widespread ChibchanPaezan root which occurs elsewhere without z'-, e.g. Gayon (Jirajara subgroup
of Paezan) kin 'nose'.
In two different word lists of Boruca, a Chibchan language of the
Talamanca subgroup in Costa Rica reported in Lehmann (1920:346-8), those
of Thiel and Valentini, one finds instances in which body part forms have
been recorded with z'- in one writer and without z'- in the other, e.g. Valentini
ikasa, Thiel kasa 'tooth', but Valentini uran, Thiel iuran 'flesh'. Although such
material should obviously be treated with caution, it is nevertheless striking
that what is historically the third person possessive should have been elicited
in a sporadic way as the equivalent of Spanish terms such as diente, carne
whereas where possessive prefixes are found it is usually the first person
singular which occurs.
The fossilization of a third person which has become indefinite in mean
ing might also be expected in kin-terms which are the other large class of
inalienably possessed nouns. Many American Indian languages treat them as
verbs with relational meaning. However in Timucua a whole series of kinterms have their general form in z'- to which may be added possessive suffixes,
clearly an innovation though derived from inherited pronominais, e.g. iti
'father'; iti- 'my father'; isa 'mother'; itora 'grandfather'; itele 'father's
brother'; -isale 'mother's sister'; yame 'brother-in-law' (Gatschet 1880:485).
For Rama, Lehmann cites i-tuun with the meaning 'someone's father, a father
in general' (1920:422).
A typologically similar development to that of Chibchan z'- is found in
Algic and Salish m-. Here it seems from the start to have been an indefinite
pronoun. With regard to the verb, Boas in his discussion of Thompson River
Salish (1898:30) describes a stage which is parallel to that already described
in regard to Chibchan i'- In Thompson River from xwe- to look', for example,
we have, xwe-em meaning 'to look at an unspecified object'. This contrasts
308
JOSEPH H. GREENBERG
with xwe-es 'to look at it' i.e. a definite object. It reflects the ambiguous
position of the specific indefinite that whereas it becomes a transitivizer in
Chibchan-Paezan, in many Salishan languages -m rather indicates the intransi
tive. We can see that the contrast with the definite object in Salish easily leads
to generalization of the indefinite specific object form to an intransitivizer.
As a possessive prefixed to the noun, m- in Algic and Salish gives rise to
developments which are extremely close to that of Chibchan-Paezan /-. In all
the Algic languages m- appears as a marker of impersonal possessive on
nouns often in fossilized form so that it becomes part of the stem. When thus
desemanticized such a form as 'somebody's hand' comes to mean 'a human
hand' or 'hand' in the abstract. There is random variation across languages
between forms with and without m-. Michelson (1938:103) states that "it is
a well-known fact that in Proto-Algonquian times stems with initial *m- have
non-initial cognates lacking this ra-." In her comparisons within Algic (Algonkian, Wiyot, Yurok) Haas (1958) freely allows forms with and without m- in
body part terms in her comparison.
In Salishan languages m- appears in fossilized form on a number of
body-parts. Kuipers (1967:116) notes in his grammar of Squamish "m- a
prefix on body parts." In instances like Upper Chehalis ma-qs-n 'nose' mu
is clearly a prefix since -qs- here and elsewhere in Salish is a suffix meaning
'nose' and -n is found in body parts. All this is quite like Algonkian but in
Salish m- survives in only a small number of lexical items and always in
fossilized form. 10
If we consider the foregoing examples, we can see that there are two
extremes to what has simply been called desemantization in regard to meaning,
and fossilization, that is, absorption into other historically distinct mor
phemes, from the morphosyntactic point of view. At one end we have what
might be called shrinkage or contractive desemanticization and at the other
end expansive desemanticization.
In instances like the m- in body part terms in Salishan or the / of kinterms
in Timucua, only a few examples are left and we have as it were the "accident"
that a disproportionate set of body parts or kinterms have the item in
question. In both instances a non-historically oriented treatment, that of
Kuipers in Squamish and Gatschet in Timucua, cannot help but notice an
element which is certainly non-productive and marginal synchronically but is
susceptible to historical-comparative explanation. At the other extreme are
instances like k- on nouns in some Nilo-Saharan languages in which, if one
were to attempt a definition of it synchronically one would have a highly
disjunctive definition which would cover a very large number of instances,
309
not unlike an attempt to define the meaning of gender classes in languages

like Sanskrit and Latin. This latter type is ripe for regrammaticalization
though of course it need not occur.
The other characteristic that all of the examples cited have in common
is what has been called here random distribution within and across related
languages and dialects.
An interesting case is that of Chibchan-Paezan *-kwa. In Greenberg
(1987:298-299), there is a brief discussion of which the following is a some
what expanded and emended version. A full study could easily be of mono
graphic proportion. It is well-known from grammaticalization studies that an
element that becomes grammaticalized can also continue in ungrammaticalized form, e.g. Latin habere 'to have', which continues as a verb in Romance
languages (French avoir; Italian avere etc.), and as future suffix in grammati
calized form (e.g. French aimer-, Italian amer- 'he/she will love').
In the case of Chibchan-Paezan *-kwa, it appears that a noun which
designated round objects and whose original meaning was an independent
noun, can still be found in Terraba (Talamanca group) gwa 'egg' and Cuna
kwa-kwa 'nut' and perhaps in other Chibchan languages. In the Cuna form
-kwa has been suffixed as it is to a large number of nouns in that language.
We may distinguish two lines of development. One is that, in accordance
with a common Chibchan-Paezan pattern, it participates in superordinate
compounds, just as for example the word for 'water', li, ri, ni is in many
Chibchan languages suffixed to words indicating liquids. The other type of
development is as a numeral classifier. In the first of these functions it
undergoes expansive desemanticization and in some instances regrammaticali
zation; in the latter instance, as a numeral classifier, after becoming the
general classifier, with the loss of the classifier system, it survives in some
languages in a restrictively desemanticized fashion, fossilized on certain
numerals.
These two types of development are treated here successively. In some
languages like Chibchan proper it is largely confined to certain objects per
ceived as round, and usually small. Examples are up-kwa 'eye'; pihi-gwa 'hole';
fa-gua 'star'. However even here it shows a tendency to spread to other nouns,
e.g. kip-kwa 'place'. We find once more random variation cross-linguistically,
e.g. corresponding to up-kwa 'eye' in Rama we find up 'eye, fruit, round
thing'. In some languages it has spread very far indeed. For example, Holmer
(1952:54) says regarding -kwa "It is the most common suffix in Cuna and at
the same time typical of a whole group of languages including Chibcha." The
same writer states:
310
-kwa has actually such a wide variety of meanings that we consider it a

general formation of concrete nouns. Originally it was (as in the related
languages of the Talamanca group) restricted to names of a certain class
of small objects which are found to have related or identical names in
the different languages of the Indians such as stone, egg, nut, bud, flower,
star... (Holmer, 1946:188)
Already in Cuna we find signs of semantic reinterpretation, one being to form
singulatives from collectives, e.g. wini 'beads', win-kwa 'a single bead' and a
diminutive function, e.g. machi-kwa 'boy'. A more productive reinterpretation
is as that of agent noun, e.g. ope-kwa 'a bather' (Holmer, 1952:55) and
abstract quali tites from adjectival roots, e.g. sipu-kwa 'white, whiteness'.
In certain Chibchan languages we find the reinterpretation as a verbal
noun dominant, e.g. Millcayac, a Paezan language of Argentina, in which
gue forms verbal nouns, e.g. cheri 'to give', cherigwe 'gif, pi- 'to die' pigue
'death' (Valdivia, 1943).
As a numeral classifier it appears in some languages with round objects
but often in more extended uses as a more general classifier. For example in
Cuna it can also appear with persons as in tule pakke-kwa 'person fourclassifier'. Its general significance is also detectible in the fairly widespread
interrogative 'how many', Cuna pi-kwa, Chibcha fi-gwa, Binticua bi-ga, Borunca bi-k (Chibchan subgroup of Chibchan-Paezan only).
In some languages in which the numeral classifier system no longer
functions it survives only in fossilized form with certain numerals, e.g. Kagaba
(Aruaco subgroup of Chibchan in Columbia) mai-gwa 'three'; ku-gwa 'seven';
abi-gwa 'eight'; aita-gwa 'nine' and u-gua 'ten'. With the word for 'three' cf.
Margua meia and, with -gwa fossilized as -g, Paya mai-g; with 'seven' compare
Margua kuki, and with 'eight' Margua avi.
Numeral classifiers, articles and indefinite objects are of course by no
means the only grammatical elements which undergo desemanticization.
Whether all do is a question I am not prepared to answer. Other examples
of desemanticization are diminutives (e.g. in Romance) and honorifics. To
illustrate the former there are numerous instances in Romance languages in
which the Latin diminutive in -/- ~ -cul- (the former alternant with first and
second declension nouns and the latter with the third, fourth and fifth declen
sion) survives in one Romance language in the diminutive forms and in
another not with diminutive meaning, e.g. French soleil <soliculus vs. Italian
sole, both meaning 'sun' as against French frre and Italian fratello 'brother'.
As an example of an honorific we may take Japanese o- prefixed to
nouns. While still having a distinct honorific meaning it has in a number of
311
nouns become desemanticized so that ocha is now the ordinary word for 'tea'
and for 'rice'.
The case of the Latin diminutive is more complex than set forth above
and I believe these complications are typical of derivational elements. The
diminutive mentioned above was already fossilized in some instances in Latin
and sometimes with meanings differing from that of diminutive, e.g. puella
'girl' <puer-la 'puer'='boy'; stella 'star' (cf. Greek aster) and osculum 'kiss'
cf. os 'mouth'.
In the case of derivational elements the framework of desemanticization
is generally more complex and would be a subject for special treatment. The
process here affects not the element as a whole but rather in each individual
occurrence with a stem there are found separate and idiosyncratic semantic
changes. Where, as with the Semitic derived forms of the verb, forms are held
together in the form of a distinct paradigm, the class may survive indefinitely
as a more and more conventionalized morphological subclass. On the other
hand it is in danger of simply disappearing as has happened in all forms of
colloquial Arabic with the fourth, originally causative, conjugation or of
merging with others via analogical processes, e.g. with the Semitic 'intensive'
(Arabic class II) in Amharic where it now coincides with the original underived
form of the verb while inheriting some morphological features of the original
'intensive' conjugation.
Another possibility for derivational elements is incorporation into the
inflectional system. Here we have a parallel to lexicalization but the morpheme
into which the derivational item is incorporated is inflectional. Thus the Latin
inceptive suffix sc- after desemanticization became a part of the inflectional
endings of certain classes of verbs in Romance languages, e.g. Italian finisco
'I finish', but fini-amo 'we finish'. The details differ for the various Romance
languages. For, example in French finissons it is incorporated whereas for
Italian finiamo it is not. As with other derivational elements we find that this
is an extended process. Thus in Latin it is already fossilized in crescere 'to
grow', Italian crescere, and French crotre in which it has been lexicalized
rather than incorporated into the inflectional system, a process that begins in
Vulgar Latin. In the initial portion of this paper it was stated that in the
breakdown of vowel harmony systems, one finds distributions across lan
guages and dialects similar to that of morphological elements. Vowel harmony
systems break down either by merger (e.g. Aliutor, a dialect of Koryak) or
simply by disregard of the rules, often through external influence (e.g. Iranicized Uzbek).
An example of the latter type is the group of East Mongolian languages.
312
In the western group which includes Classical Mongolian as well as Buryat,

Khalka and others, there is a system of back-front vowel harmony which is
stem-driven. By this is meant that the stem has constant vowels but the affixes
have back and front variants which harmonize with the vowel or vowels of
the stem. One such alternating pair is a (back harmony) versus e (front
harmony). The results for a series of affixes with former a~e alternation is
shown in the following table:
Table 1. Inflectional affixes in four East Mongolian languages
Plural
Ablative
Locative
Instrum.
Comitat.
Causative
Past Part.
Baoan
Dagur
Dunsian
Monguor
le
se
ra
gale
ge
sang
se
aare
gala
gaa, gee
sen
la
se
le
ga
sen
dza
re
la
ga, ge
dzan
We have seen that random distribution as defined here is basically similar

in morphology and phonology. How are we to envision this process from the
sociolinguistic point of view? I think that we should consider that for every
individual instance variants, idiosyncratically distributed across small groups,
or even individuals, existed in the ancestral speech community.
Within each group that later became a distinct language or subgroup
there was a particular distribution, of course, subject to later analogic changes
and sometimes with an inherited preference for one form or the other. We
may compare these with 'founder' groups of the population geneticists.
The ensuing results simply do not fit either the conventional family tree
or any modification of it based on areal considerations. However, it only
affects a very small part of the total linguistic structures involved.
The topic being considered here is, of course, a very large one and my
conclusions are at best tentative. However, if it will direct the interest of
linguists interested in grammaticalization to a more detailed study and analy
sis of the last stages of the process, it will have accomplished its purpose.
NOTES
1.
I am indebted to Bernd Heine and Elizabeth Traugott for valuable criticisms and sugges
tions regarding the present paper.
313
2.
However in Sardinian both the article and the anaphoric pronoun developed from Latin
ipse 'self.
3.
For details see Greenberg (1981). The k- under discussion here is always prefixed, is
followed by a vowel before consonant stems and is distinct from the widely attested
plural which often contrasts with a t singular. By a separate development from its original
demonstrative meaning, k- also occurs in some of the languages prefixed to the verb to
indicate third person.
4.
I use the term free variation when a particular source gives both forms as variants. No
doubt sociolinguistic investigation would show that the distribution of variants involve
individual preferences, and generational and social class differences, but such information
is not provided in the sources.
5.
I have hypothesized that the starting point of the Stage II article is the indefinite specific
use later expanded to the indefinite non-specific. This is because the indefinite specific
allows for an ambiguous interpretation which could lead to the usage of the Stage II
article which includes all the basic uses of the indefinite article in a language like English.
In at least one language, Bemba, there is an article with both definite and specific indefinite
uses but which is not employed for the non-specific indefinite. I have noted that in English
there are occasions on which the speaker uses the definite article on the assumption that
the hearer shares the identification with him but that the assumption turns out to be
incorrect leading to a specifying (which) question by the hearer. This problem requires
further research. It is noteworthy that Stage II articles only occur in languages which
have no indefinite article and are particularly frequent in Africa, an area in which indefinite
articles are uncommon.
6.
It is possible for languages with noun classes, of which gender classes can be considered
a subtype, to develop a second or third stage article which does not distinguish class, e.g.
Aramaic -a, or for a language without noun classes to develop second or third stage
articles which distinguish noun classes, e.g. Maasai, a Nilo-Saharan language which
developed a set of Stage II articles which indicate sex gender.
7.
I am indebted to Keith Denning for this suggestion.
8.
This section as well as those on Algonkian and Salish m and Chibchan-Paezan kwa are
expanded versions of the discussions in Greenberg (1987:56-7, 280-1, 284, 298-9).
9.
However, there may be examples of the random survival of i- on verbs both transitive
and intransitive in Timucuan. Gatschet (1880:485) describes i- as a verbal prefix. Among
the examples are iparu 'swallow'; ikiti 'insult'; ichuki 'throw away, spill' but also intransi
tives such as ikwaso 'to scream'. Very striking in the present connection is the example
iriboso 'to flood something'.
10.
Comparison with the process discussed in this paper by which a third person pronoun
ultimately becomes a transitivizer has an obvious bearing on the much discussed question
of Tok Pisin and other Oceanic creoles that use -im as a transitivizer. I believe it strengthens
the case for substrate influence since the process itself takes a long time and is most likely
to have derived from the influence of languages which had already gone through it. I
have not myself investigated the historical origin of markers of transitivity in Melanesian
languages.
314
REFERENCES
Barrai, Basilio de. 1957. Diccionario Guar ao-Espaol Espaol-Guarao. Caracas: Sucre.
Boas, Franz. 1898. "NtlakyapamuQ. Twelfth report on the northwestern tribes of
Canada." In Annual Reports of the British Association for the Advancement of
Science, 27-36.
Camp, Elizabeth and Millicent Liccardi. 1967. Vocabularios Bolivianos 6. Riberalta,
Beni, Bolivia: Summer Institute of Linguistics.
Gatschet, Albert A. 1880. "The Timucua language." Proceedings of the American
Philosophical Society 18:465-502.
Greenberg, Joseph H. 1978. "How does a language acquire gender markers?" In
Universais of Human Language III, Joseph H. Greenberg, Charles A. Ferguson,
and Edith Moravcsik (eds), 47-82. Stanford: Stanford University Press.
Greenberg, Joseph H. 1981. "Nilo-Saharan movable -k as a stage III article (with a
Penutian parallel)." Journal of African Languages and Linguistics 3:105-112.
Greenberg, Joseph H. 1987. Language in the Americas. Stanford: Stanford University
Press.
Haas, Mary. 1958. "Algonkian-Ritwan: The end of a controversy." International
Journal of American Linguistics 24:159-73.
Holmer, Nils Magnus. 1946. "Outline of a Cuna grammar." International Journal of
American Linguistics 12:185-97.
Holmer, Nils Magnus. 1952. Ethno-linguistic Cuna Dictionary with Indices and Refer
ences to a Critical and Comparative Cuna Grammar and the Grammatical Sketch
in Cuna Chrestomathy. Goteborg: Ethnografiska Museet.
Kuipers, Aert H. 1967. The Squamish Language. The Hague: Mouton.
Lehmann, Walter. 1920. Zentralamerika. Berlin: Reimer.
Michelson, Truman. 1938. "Algonquian notes." International Journal of American
Valdivia, Luis de. 1943. Los Textos Millcayac del P. Luis de Valdivia con Vocabulario
Espaol-Allentiac-Millcay por Fernando Marquez Miranda. La Plata: Universidad
de La Plata.
Substrates, Calquing and Grammaticalization in

Melanesian Pidgin
Roger M. Keesing
Department of Anthropology
The Australian National University
1.
INTRODUCTION1
Melanesian Pidgin has been widely recognized as unusual, in the

worldwide spectrum of pidgins and creoles, in the degree of syntactic expan
sion and stabilization attained despite the fact that, until recently, it remained
primarily a plantation lingua franca acquired by adults. Melanesian Pidgin
incorporates grammatical devices such as transitivity-markers and mecha
nisms for embedding relative clauses not found in "true creoles" (Bickerton,
1981,1984) not to mention the contact jargons antecedent to them. While
this has been taken by such scholars as Bickerton and Mufwene (1986) to
imply that Melanesian Pidgin has a history different from those of wellknown northern hemisphere pidgins and creoles, the special nature of that
history, and its linguistic consequences, remain poorly understood.
In a recent book (Keesing, 1988), I argue that the obscurities that shroud
this sociolinguistic and linguistic history are partly a result of the enshrining
of the New Guinea dialect, Tok Pisin, as the canonical form of Melanesian
Pidgin, and the consequence that the Vanuatu (New Hebridean) and Solo
mons dialects of pidgin remain poorly documented and little known to the
international linguistic community. It is these dialects, I argue, that encode
and preserve the most critical linguistic evidence of how pidgin in the Pacific
acquired its distinctive degree of expansion and stabilization. Moreover, in
the Solomons and Vanuatu, the relationship between pidgin and the languages
natively spoken by those who have historically been its primary speakers
remains in view. Tok Pisin obscures these developmental processes, because:
316
1.
2.
3.
ROGER M. KEESING
A century ago, the plantation system of German New Guinea was cut
off (by the imposition of political barriers) from a vast regional pidgin
speech-community that incorporated much of the central and southwest
ern Pacific and Queensland.2
Thus cut off from the regional pidgin speech-community, Tok Pisin was
extensively relexified (from indigenous Bismarck Archipelago languages,
and to a minimal degree, from German); the original regional pidgin
lexicon was derived almost entirely from English and from nautical
pidgin(s) of northern hemisphere origin (save, pikariini, etc.).
Tok Pisin, in being transplanted from the area where Pacific pidgin
developed, acquired a set of substrate languages very different from those
whose speakers had been the primary contributors to that initial devel
opment.
The question of substrate languages is crucial. Although Pacific pidgin

was used as a regional lingua franca across vast stretches of ocean and it
was for thirty crucial formative years primarily a lingua franca of shipboard
communication, not plantation life almost all of the Islanders who parti
cipated in its development were speakers of Oceanic Austronesian languages.
Most were speakers of languages that apparently fall into a single subgroup
of Oceanic (which I will call Eastern Oceanic, or EO), marked by conservatism
in preserving Proto-Oceanic grammatical patterns. 3
Pidgin in the Pacific acquired a remarkable degree of grammatical elabo
ration partly because, from the 1840s onward, Islanders took the lexical
resources of English and nautical jargon and progressively hammered them
into grammatical designs common, at an abstract level, to their native lan
guages. In doing so and in keeping an evolving code accessible to superstrate speakers they drew massively on universal-based faculties of
simplification, and followed paths of minimal markedness/universality.
Hence, incorporating substrate patterns into an emerging pidgin is, as Good
man (1985), Mufwene (1986), Mhlhusler (1987) and others have noted, in
no way incompatible with universalist theory or the impress of superstrate
patterns: It is the interplay between the three that generates the kinds of
phenomena I will examine.
What makes Solomons Pijin and Vanuatu (New Hebrides) Bislama so
much more revealing of this dialectical process than Tok Pisin is, firstly, that
they clearly incorporate quite specific grammatical patterns common to East
ern Oceanic Austronesian substrate languages; and secondly, that the bending
processes that historically shaped Pacific pidgin are still going on. The Solo-
SUBSTRATES, CALQUING AND GRAMMATICALIZATION
317
mons and Vanuatu dialects are still connected to particular subgroups of EO

languages.4 Out of the resources available in the pidgin of the 1890s by
which time Melanesian Pidgin had acquired most of its present grammatical
resources patterns congruent with particular sets of substrate languages
have been selected. We find evidence of this process in the divergence between
the Solomons and New Hebridean dialects since the end of the Labor Trade
separated the two speech communities (ca. 1905), and in regional differences
in the pidgin of older speakers within each country. Speakers who acquired
Pidgin as young adults demonstrably calque Pidgin onto their native lan
guages, lexically and grammatically.
The unusual degree of grammatical elaboration in Melanesian Pidgin, in
contrast to other documented pidgins, is primarily the product of a process
whereby lexical elements from the superstrate language and nautical jargon
were equated with grammatical elements in Oceanic Austronesian languages:
A process underway for almost 150 years. Evidence of this process is encoded
in a grammatical statigraphy of Melanesian Pidgin dialects, a stratigraphy
we can establish on both distributional and textual grounds. Patterns common
to all three major dialects of Pidgin and attested in the nineteenth-century
texts must have been incorporated a century ago, when Tok Pisin was sepa
rated from the other two dialects; patterns common to Solomons Pijin and
Bislama must have been incorporated by the turn of the century, when the
Labor Trade ended.
Let me be more explicit at the outset about the relevance of Melanesian
Pidgin to a developing understanding of the grammaticalization process.
Melanesian Pidgin has undergone processes of grammaticalization fundamen
tally similar to those being documented for natural languages (as Sankoff,
Mhlhusler and others have shown for Tok Pisin). But this, I argue, is not
the main mechanism whereby Melanesian Pidgin has acquired a syntactic
complexity far beyond that documented for other pidgins. Where the substrate
languages whose speakers play the main part in the development of a pidgin
are sufficiently closely related to have a common basic grammatical pattern
(as is the case with the Eastern Oceanic Austronesian languages), lexical items
(or stylistic variants, such as resumptive pronouns) in the lexifying superstrate
language can be equated directly with grammatical elements common to
substrate languages. It is this process, parallel and complementary to gram
maticalization in a strict sense, that I will illustrate in the sections to follow.
318
2.
R O G E R M. KEESING
TRANSITIVE MARKING
The transitive suffix -im in Melanesian Pidgin has been widely recognized
as a grammatical pattern anomalous in the spectrum of pidgins/creoles. The
"-urn" form sporadically used in northern hemisphere jargons was almost
certainly introduced to the Pacific, through the usages of nautical jargon and
perhaps a developing New South Wales pidgin used in white-Aboriginal
communication. What is interesting is what happened to this form when it
reached the Pacific, and why. In almost all EO languages of the central Pacific,
the semantic role relationship of agent to patient (and secondarily, to goal
or stimulus) is marked with a transitive suffix (reconstructed for as *-i,
and manifest in most EO languages as -Ci-) attached to verb roots (Pawley,
1973, Pawley and Reid, 1979). Once available to Pacific Islanders, "-urn" was
apparently quickly equated with the transitive marker in substrate languages,
hence became a grammatical element by a direct borrowing and calquing.
Although there is no phonological counterpart to the nasal element in "-urn"
in Oceanic transitive suffixes, the form otherwise fits perfectly into an Oceanic
pattern. 5 Hawaiians encountered by Dana (1840) in San Diego in 1835 are
recorded as using lock him up chest. By 1844, me like urn man belongen noder
place ( like foreigners') is recorded in Fiji, from a Fijian who had travelled
widely on trading ships (Erskine, 1853:461). In 1852, Oh, by, by Riko catch
im putty 'im in calaboose ('Riko will capture him and put him in jail') is
recorded in the Cook Islands (Lamont, 1867:98).7 From Tanna (southern
New Hebrides) in 1869, we have two separate texts in which "make him"
(mek-im) is recorded (Keesing 1988, Chapter 4).
Textual and distributional evidence indicates that the lexicon of the
Melanesian Pidgin of the Labor Trade (mid 1880s) already included a substan
tial set of verbs, marked as transitive, which represent significant innovations,
as reanalyses of or extrapolations from the English source words. The innova
tions are both semantic and morphological; they occur in all three daughter
dialects (I here use Solomons Pijin orthographic renderings):
kas-im
lukaot-im
kapsaet-im
lus-im
win-im
'to
'to
'to
'to
'to
reach (a place), to obtain (s.t.)' (from English "catch").

find (s.t.) by searching'
pour out contents of (s.t.)' (from English "capsize").
leave (a place)' (from English "lose").
exceed (s.t.), to beat (s.o.)' (from English "win").
The transitive suffix in Melanesian Pidgin operates in exactly the same

way as the transitive suffix in EO languages: It marks the relationship to a
319
direct object NP (canonically in the semantic role of patient); and like the
transitive suffix in many of the EO languages, it incorporates a zero-marked
third-person reference, as in:
mi no luk-im
'I didn't see it/him'(SIP),
except where it is followed by a pronoun, as in:
mi no luk-im olketa
'I didn't see them'.
3.
PREPOSITIONAL VERBS
A further pattern not only underlines the direct correspondence between

Pidgin -im and the Oceanic transitive suffix, but makes strikingly clear that
it was Islanders, not English-speakers, who were the primary agents in the
incorporation of transitive-marking into Pacific Pidgin. This pattern occurs
both in Solomons Pijin and, to somewhat lesser degree, in Bislama.8 A
widespread pattern in Oceanic languages, clearly preserved in the EO lan
guages of Vanuatu and the southeastern Solomons, is the use of what have
been called "prepositional verbs." What had at one time been second verbs
in compound verbal constructions have become detached, such that they
function semantically and syntactically as prepositions, but morphologically
retain the structure of transitive verbs (in being marked with transitive-suffixes
and/or bound clitic object pronouns). Kwaio, a Southeast Solomonic language
of Malaita on which I have worked extensively (Keesing, 1975, 1985), will
serve to illustrate. In Kwaio, transitivity is marked either by the transitive
suffix --a (where -a is a clitic 'it' pronoun) or by the pronominal element
(-a) alone.
fe'e-ni-a
fono-si-a
folo-si-a
fa'a-si-a
suri-
'ani-
'usi-
dari-
'with (s.o.)'
'against (s.o., s.t.)'
'across (s.t.)'
'away from (s.o. or a place)'
'around, beside, on top o f (and metaphorically, 'because of,
about')
'with (an instrument)'
'attaining a destination (s.o., a place)'
320
ROGER M. KEESING
In some cases, the roots are still used as verbs, stative or active {fono is
'be blocked', folo is 'be across', dari- is 'find (s.t.)'). I n others, the verbal use
has disappeared. A quite similar set of prepositional verbs in the To'aba'ita
language of northern Malaita is discussed by Lichtenberk (This volume).
Lichtenberk also discusses the way some of these prepositional verbs (in
Kwaio, suri- and fa'a-si-a) serve as complementizers in Malaita languages.
In Solomons Pijin, a set of prepositional verbs corresponds directly, in
morphology, syntactic function, and semantics, to the prepositional verbs
pervasive in substrate languages of the southeastern Solomons, and particu
larly, of Malaita: 9
wet-em
agens-em
abaot-em
koros-im
antap-em
raon-em
kas-em
'with (s.o.)' [often weit-em or wit-im]

'about, around (s.t.)'
'across (s.t.)'
'on top of (s.o., s.t.),
'around'
'attaining a destination (s.o., a place)'
In Bislama, only wet-im 'with (s.o.)' is common, but agens-em and kas-em
have been recorded as prepositional verbs. 10
The occurrence of these prepositional verbs in the two dialects of Melanesian Pidgin that most directly preserve the lexical and syntactic structures of
the pidgin of the Labor Trade is striking evidence both of the nineteenthcentury origin of these patterns and of the close correspondence between
Melanesian Pidgin and the grammars of EO languages. The prepositional
verbs in the two closely-related Pidgin dialects transparently represent a
grammatical pattern based directly on constructions of substrate languages,
in the creation of which Europeans could have played no direct part even
though the lexical elements came ultimately from English.
4.
THE CALQUING PROCESS: THE CASE OF KAM AND GO
The process whereby lexical borrowings from English acquire grammati

cal functions corresponding to those in substrate languages through calquing
can be illustrated by some of the uses to which the English verbs "come" and
"go" have been put in Melanesian Pidgin dialects.
As basic verbs that would have been part of a core English vocabulary
used in giving orders and expressing intentions to "the natives," "come" and
321
"go," not surprisingly, turn up in Pacific jargon from the texts of the 1840s
onward. In 1852, for example, recorded in the Cook Islands (Lamont,
1867:96), we find me likey go Carifona; ship he no go.
"Come" and "go" also serve as auxiliary verbs in English, at least in
colloquial registers: Thus, "come and do it" or "go and fetch it" become
"come do it" and "go fetch it." We get a collusion here between superstrate
patterns and common substrate patterns in Oceanic languages, where such
actions are commonly rendered with serial verb constructions. 11 We find even
in the early texts evidence of both "come" and "go" (kam and go) being used
by Islanders in serial constructions, in ways transparent to both substrate
and superstrate speakers.
Thus we find, in 1844 in Fiji, You come see me (Erskine, 1853:461). In
1850, in the Loyalties, we find:
Great fool Uea man, steal little thing he no want, big ship come and kill
him (Erskine 1853:347) ('Uea men are great fools in stealing things they
don't want; a big ship will come and kill them'; note the relative clause
'that he doesn't want'.)
What is recorded as "come and kill him" was probably kam kil-im. In 1854,
on Kosrae in the eastern Carolines, we have every kanaka: big island, small
island, all go and kill every man board ship (Hezel, 1983:114). In 1869, in the
southern New Hebrides, we find Plenty man come hear you make him bokis
sing(Paton, 1894:77).12
Such constructions, using kam and go as first verbs in serial verb construc
tions, are common in all three contemporary dialects of Melanesian Pidgin.
Some examples from a text in Solomons Pijin recorded from an elderly
speaker of Maringe (Santa Isabel) 13 will serve to illustrate:
olketa kam faet aotsaet long ples blong mifala long Baolo
LOC Baolo
they come fight outside LOC place POSS us
'They came and fought off (the shore) at our place at Baolo.'
mifala kam luk-im olketa pulande
nao
we
come see-TRS them be plentiful PRF
'We came and saw that there were lots of them.'
olketa Diapane go nao, olketa go luk-im
PL
Japanese go PRF they go see-TRS
'The Japanese went, went and saw
322
ROGER M. KEESING
tu-fala endi long rot

two-ADJ end LOC road DEM
'the two ends of the road.'
olketa kam dorop-em baum, baum, baum
they come drop-TRS bomb bomb bomb
They came and dropped bomb after bomb.'
We could go on to ask whether a Maringe speaker or a Kwaio speaker
of Solomons Pijin could go further, in equating go and kam in such serial
constructions with verbal auxiliaries or other verbal particles in their native
languages thereby possibly giving them a grammatical force and/or a
specific semantic import not transparent to English speakers (or, perhaps, to
one another). Could kam be equated by a Maringe speaker with one kind of
grammatical element and by a Kwaio speaker with a different grammatical
element, on the basis of calquing? I will return to these questions.
Another pattern found in both Bislama and Tok Pisin is the use of kam
and go in serial-clausal constructions that express deictic-directional mean
ings. Thus, in Tok Pisin (Givn 1987):
i wokabaut go 'he went away (from a reference point)'
em karim i kam 'she brings it (toward a reference point)'.
The rendering of such constructions in Solomons Pijin points particularly
clearly to the impress of EO substrate languages. In Oceanic languages
generally, and pervasively in the EO languages of the zone where Pacific
pidgin underwent its major development, two directionals, used postverbally,
serve to express the meanings "hither" and "thither": What are reconstructed
for PEO as *mai and *watu (these are apparently derived from lexical verbs
'come' and 'go', hence historically represent grammaticalizations in Oceanic).
Thus, in Kwaio we find ngari-a mai 'bring it' and ngari-a kau 'take it (away)',
lit. "convey it hither" and "convey it thither." In Solomons Pijin, kam is
systematically used where mai is used in substrate languages, and go is
systematically used where reflexes of *watu, such as Kwaio kau, are used:
tek-em kam 'bring it'
tek-em go 'take it (away)'
Older speakers use these particles in ways that correspond exactly to patterns
in their native languages. Thus for Kwaio speakers
323
ori-si-a
mai
ask-TRS-it hither
'ask about it'
is rendered in Pidgin as:
ask-em kam.
Two fragments from my texts from Kwaio speakers who learned Pijin in pre
war plantation contexts will serve to illustrate how such forms are used in
discourse:14
hem-i stop-em go long sam-fala pepa
he
plug-TRS DEI with some-ADJ paper
'He plugged it up with some paper.'
mifala seke nao ma 'efofilen mekenois nao
we
check PRF and airplane it buzz
DEI PRF
'We checked, and an airplane was buzzing toward us.'
A further attestation of the impress of the EO substrate on Melanesian
Pidgin in the incorporation of go is a pattern where the verb go, reduplicated,
serves to indicate the passage of time in a narrative, a pattern widespread in
the EO languages of Vanuatu and the Solomons. We find it in Kwaio leeleka
(where leka is 'go'). In Kwaio as in other Oceanic languages, reduplication
of a verb usually indicates repetition or continuation of the act described by
the verb. An extract from a text will illustrate leka 'go' used in both these
ways:
leeleka leeleka leeleka ma gai- ma
go-go go-go go-go and mother-my and
'And as time went on, my mother and
ma a a-gu
'agaaa gala taataungai ma
father poss-me them2 15 they2 work(RED) and
my father worked and
gala leeleka
naa fonunga ma gala leeleka naa
they2 go(RED) LOC feast
and they2 go(RED) LOC
they'd go to feasts and they'd go to
boni- makete ma ogulaa
ma na ku naanai 'ifi
event-it market and gathering and me I stay
home
markets and gatherings, and I'd stay at home.'
324
ROGER M. KEESING
Melanesian Pidgin dialects all use this device of go-go to indicate the
passage of time, as in a fragment from a text from an old Lau (Malaita)
speaker of Solomons Pijin:
olketa lukat-em
mifala, go-go olketa kas-em
sam-fala ...
they search-TRS us
go-go they catch-TRS some-ADJ
'They searched for us, and eventually caught some (people) ...'
In Lau, lea is 'go' and lelea indicates the passage of time: lelea ka sui is 'when
it is finished' (Fox, 1974:116).
Another text fragment from the same Lau speaker shows go (as well as
kam) in serial verb constructions and go-go as marker of time elapsing:
oraete olketa nao olketa ofen-em olketa bokis
OK
they come PRF they come open-TRS PLU box
'Then they came and they opened all the boxes,
infikis,
tekeaot evri kaleko ia, goo-go i
finisi
investigate take out each garment DEM go-go it be finished
investigated, took out every piece of cloth, until it was finished,
olketa go torou-em
they
long faea
gO thrOW-TRS LOC
fire
they went and threw (them) on the fire.'

Many northern and central Vanuatu languages use exactly the same
device. Thus in , ba is 'go' and bababa... indicates the passage of time in
a narrative (Darrell Try on, personal communication). Not surprisingly, Vanu
atu Bislama uses go-go-go in exactly the same fashion as Solomons Pijin.
Note that there are two processes involved in the way kam and go have
been incorporated into Pijin, one of which involves lexical equivalences and
the other of which involves grammaticalization, at least to some degree. In
fitting the lexical verbs derived from English "come" and "go" into serial
verb constructions corresponding to those in substrate languages, Islanders
have needed to do no reanalysis. The verbs 'come' and 'go' are fitted into the
same slots in Pidgin that they occupy in Island speakers' first languages. The
same is true for the use of go-go-go to mark the passage of time in narrative,
even though the semantic connection between physical movement and the
passage of time in the substrate languages is metaphorically based. Once go
has been equated with a lexical verb 'go', no further semantic reanalysis is
needed: The semantic connection already exists.
325
In the case of the postverbal directionals, however, the EO 'hither' and

'thither' forms are not lexical verbs for 'come' and 'go'. Hence, a metaphorbased extension of the semantics of kam and go was required in order for
Solomons Pijin to adopt these as equivalents to *mai and *watu forms. The
further exemplifications I will give of the calquing process include both direct
transfers based on lexical equivalence and more indirect ones involving
reanalyses or extensions that parallel (although short-cut) grammaticalization
processes in the development of natural languages.
5.
PREVERBAL PARTICLES MARKING ASPECT/MOOD/

MODALITY
We find further compelling evidence of the more direct paths toward

lexicalization and grammaticalization if we examine the aspect-markers of
Solomons Pijin and Bislama in relation to those of substrate languages. I will
look most closely at Solomons Pijin, for which my data are richest, and in
particular at the relationship between the substrate languages of Malaita and
Pijin.
In EO languages, tense and aspect are not extensively marked. However,
verb phrases in EO languages characteristically have preverbal and postverbal
slots filled by particles marking aspect, mood, and modality. These particles
may be free or bound to subject-marking pronouns or verbs. Documentary
and distributional evidence indicates that lexical elements drawn from Eng
lish "by and by," "just," "might," "now," etc. were by the 1880s being
grammaticalized in Melanesian Pidgin as equivalent to Oceanic markers of
aspect/modality/mood. In some cases, the transfer of lexical material into
Oceanic grammatical slots required greater reanalysis by Islanders: In all
three Melanesian Pidgin dialects, ating (derived from English "I think," but
analyzed by Islanders as monomorphemic) is used as a preverbal particle
indicating 'maybe, probably' (often used interchangeably with maet
("might").
Baabae (or its short form bae) is characteristically not a future tense
marker, but rather a marker of irrealis or non-accomplished mode, corre
sponding in function to a particle reconstructed for PEO (Pawley, 1972).
Whereas English speakers have consistently used it in clause-initial position
as a temporal adverb, there has been a pull, from the 1880s onward, to bring
bae into the verb phrase as a grammatical marker, and to use to it express
irrealis meanings as well as establish a future time-frame. In many EO
326
ROGER M. KEESING
languages, markers of mode (dubitive, irrealis, etc.) that canonically fit within
the verb phrase can be fronted for emphasis; this is an option used frequently
for both bae and maet in Bislama and Solomons Pijin, one which allows
superstrate and substrate speakers to meet in the middle, as it were, while
using different grammatical analyses. 16
The Pidgin of the Queensland Labor Trade apparently incorporated a
range of elements available as verbal particles. Which ones were used, and
the grammatical force assigned to them, depended in part on the use of
particles in the speakers' particular substrate languages. A fully adequate
degree of mutual intelligibility could apparently be achieved through speakers
of particular substrate languages using these markers of aspect, modality and
mood in ways corresponding to those of their native languages: The indige
nous languages are close enough in this regard that it seems that little semantic
content would be lost in the process.
The process can be illustrated with the particle des (or tes or das), from
English "just." The des form is manifest both in texts from the late nineteenthcentury Labor Trade (recorded, of course, as "just") and in contemporary
texts from older speakers of Vanuatu Bislama and Solomons Pijin. For
Bislama, Charpentier has pointed out close correspondences with substrate
languages. Thus, to take Charpentier's (1979: 353) example from Vetmbao
(South Malekula) and Bislama:
naji nga-mandrxa
him he-ASP
drink 17
'He just drank.'
em tes trink
him he ASP drink 18
A Solomons example will show the same pattern of correspondence,
where the form taken from English "just" has acquired grammatical force as
an aspect marker. In Kwaio, the particle bi'i, fitting between the subjectmarking pronoun 19 and the verb, marks the action of the verb that follows
as having taken place in the immediate past.
ngai e bli aga-si-
him he just see-TRS-it
'He just saw it.'
Both Kwaio speakers from whom I elicited Pijin equivalents of some 400
Kwaio sentences20 consistently gave das or tes as the Pijin equivalent. Thus:
327
hem tes luk-im

him he just see-TRS
'He just saw it.'
Given that English "just" fits into a similar preverbal position and that
such markers are common in Atlantic creoles, is there any evidence that
Kwaio speakers are indeed using tes21 as equivalent to bi'i? Here we need to
look first at the marking of irrealis/non-accomplished mode in Kwaio.
The Kwaio future marker is ta- (with allomorphs te- and to-), prefixed
to the subject-marking pronouns.
gila ta-la
leka
them FUT-they go
They will go.'
The Solomons Pijin equivalent I elicited was
olketa bae
go
them FUT 3PRS go
They will go.'
This pattern of equating ta- with bae/baebae is consistent throughout a very
large corpus, to mark both a probable future event and the irrealis status of
the event described. A second Kwaio speaker gave exactly the same equiva
lences throughout.
In Kwaio, the irrealis/future marking particle can be conjoined with bi'i,
with a semantic import not predictable from the two forms used separately.
It indicates that the action of the verb will take place in a short time. Both
Kwaio speakers from whom I elicited Pijin equivalents used baebae and das/
tes together to express these meanings:
ta-goru
bi'i aga-si-a
FUT-WEINCL just see-TRANS-it
'We'll see it in a while.'
bae iumi
das luk-im
FUT WEINCL just see-TRANS
Such detailed correspondences make it inescapably clear that Kwaio
speakers use these preverbal markers of aspect/mood/modality in ways that
directly follow the grammar of their native language, using in effect
formulas of equivalence between the two languages. Yet at the same time,
the resources for doing so constitute part of the pidgin linguistic system dating
328
ROGER M. KEESING
back to the nineteenth century. The evidence that Kwaio speakers are calquing
on their native language in using verbal particles in Solomons Pijin goes
further. In my large corpus, 22 the entire set of Kwaio particles marking the
time-frame of the verb, some of which are preverbal and some postverbal,
corresponds in their Solomons Pijin usage to a set of particles derived from
English but carrying exactly the same import as the Kwaio particles, and
placed in exactly the same slots. I will shortly examine another appropriation
of kam and go by Kwaio speakers of Solomons Pijin that further illustrates
how English-derived lexical forms can be given grammatical force through
direct calquing onto grammatical elements in substrate languages.
The lexical elements of the pidgin of the late nineteenth century Labor
Trade were fitted into the aspect-marking system of New Hebridean languages
in a somewhat different way. In EO languages of the northern and central
New Hebrides, whose speakers were numerically preponderant in the main
period of the Labor Trade, a durative aspect marker indicating that an act
is in progress is common. The Bislama of Vanuatu has incorporated stap,
which in other contexts is a verb meaning 'stay, be present, exist', as a durative
aspect marker:
iej i-ndang-lualua na-ndam (Nasiariana, Malekula)
he he-DUR-dig
ART-yam
'He's in the process of digging yams.'
naji ng-u-xoel dram (Vetmbao, Malekula)
he he-DUR-dig yam
'He's in the process of digging yams.'
em i stp -im yam
him he DUR dig-TRS yam
'He's in the process of digging yams.' (Charpentier, 1979:351)
la lo retired
na lolo vanua (Tangoa)
they DUR talk (RED)
LOC inside village
'They are having a discussion in the village.'
oli stap toktok
inside long velej
they DUR tok(RED) inside LOC village
'They are having a discussion in the village.' (Camden, 1979:100-101)
In Southeastern Solomonic languages such marking of durative aspect
is not generally found. In Solomons Pijin stap is common as a verb but does
not occur as a durative aspect marker. 23 In the Malaita languages, to indicate
329
that an action is underway, one can either use a temporal or use a perfectmarking particle after the verb a marker that indicates either that action
of the verb has commenced or that it has been completed. It is to this particle
in the Malaita languages, and the corresponding form incorporated into
Solomons Pijin, that we now turn.
6.
POSTVERBAL PERFECT-MARKING PARTICLES
Textual evidence and contemporary usages in Bislama indicate that nao

(derived from English "now") was being used in the pidgin of the Labor
Trade as a temporal adverb. In Bislama, it continues to function in this way:
mi mi slip nao''maintenant, moi je dors' (Charpentier, 1979:360).
In Solomons Pijin, nao has come to serve as an aspect marker, corre
sponding to forms universal in Malaita languages forms which characteris
tically resemble nao phonologically, in having nVV (or nVCV) shapes (see
Simons 1986). These perfect markers, occurring in a postverbal slot within
the VP, serve to articulate a state at a reference time (the time of the speech
event) to an earlier state or event, to indicate that the two are essentially and
inseparably connected, and to focus attention on the present state. In addition
to indicating that an act has been completed or (with stative verbs) that a
state has come into being, they serve to mark prospective perfects (i.e., that
the action is underway), and hence are used where northern and central
Vanuatu languages would use durative aspect marking. Some examples from
Malaita languages will serve to illustrate:
'Are'are (Geerts, 1970:75):
e
siko
no 'o
it b e
finished
PRF
'It is finished.'
e tara'ae no'
he start PRF
'He has started.' 24
Kwara'ae (Deck, 1934:38):
ka sui
'
it be finished PRF
'It is finished.'
330
ROGER M. KEESING
nia leka'
he go PRF
'He has gone.'
Lau (Ivens, 1934:73):
e
sui
na
it be finished PRF
'It is finished.'
e
lea na
h go PRF
'He has gone.'

Kwaio (see Keesing, 1985):
(ngai) e leka no'
him
he go PRF
'He has gone.'
gala age-a no'
they2 do-it PRF
'They have done it./They are doing it now.'

e 'akwa
no 'o
he run away PRF
'She has run away.'
e
mou
no'o
it be broken PRF
'It's broken.'
For such sentences, speakers of Solomons Pijin use:
hem-i go nao
hem-i finis nao
hem-i ranawe nao
tufala-i du-im nao
hem-i birek nao
'He has gone.'

'It is finished.'
'She has run away.'
'They have done it/are doing it now.'
'It's broken.'
The direct correspondence between nao and the postverbal perfect-marking

particles in the Malaita languages has been noted and documented by Simons
(1986). I have argued that with this standardization of a perfect marker in
Solomons Pijin has come the virtual disappearance of bin regularly used
331
as a marker of completed action in regional dialects of Bislama as well as

Tok Pisin, and attested widely in nineteenth-century texts.
Simons also notes that in the Malaita languages, the perfect marker or
an allomorph of it is used after a noun or focal pronoun to topicalize or
foreground it. Thus, in Kwaio:
gila no'o la age-a
them TOP they do-it
They're the ones who did it.'
ngai ne-e
aga-si-a
him - see-TRS-it
'He's the one who saw it. (Here ne- is an allomorph of no'o)'
Simons notes that nao is used in Solomons Pijin in an identical manner:
hem nao i save
him TOP he know
'He's the one who knows.'
olketa waeteman nao i save wak-em
PLU European TOP he able do-TRS
'It's the Europeans who know how to do it.'
This example is striking in two respects. Presumably, the use of n W as a
topicalizing particle in Malaita languages derives historically from its use as
a perfect marker, through the kind of grammaticalization processes explored
in these volumes. 25 Hence, the use of nao in Pijin to correspond to nVV in
both its grammatical slots as perfect-marker and as topicalizer represents
the kind of shortcut to grammaticalization through direct calquing with which
I am concerned in a clear and striking way. Secondly, the path to this calquing
was facilitated by a phonological correspondence between a lexical form
available in the superstrate language, "now," and the particle in substrate
languages with which it was equated.
I have suggested that speakers of particular Solomons languages can, as
it were, go beyond the limits of a common code in calquing available Pijin
forms onto particular grammatically-salient forms in their native languages
a process I illustrated with the use of bae ... tes constructions by Kwaio
speakers. We need to look more closely at this phenomenon, since it implies
that like speakers of Chinook Jargon as documented by Silverstein
(1972) speakers of a common code may be using different grammars. In
doing so, we shall return to kam and go, in their role as preverbal auxiliaries.
332
7.
ROGER M. KEESING
KAM AND GO AS AUXILIARIES
I shall look first at two Kwaio forms which function as auxiliaries within
the verb phrase (neither one is a verb in its own right: They fit into the slot
between SRP and verb characteristically occupied by aspect markers). In
Kwaio, kee usually serves in narrative to indicate the sequential relationship
between acts described in successive clauses. Kee before an active verb indi
cates that the act described in the previous clause preceded the act marked
with the auxiliary. The two Kwaio speakers in their sixties from whom I
elicited Pijrn equivalents of Kwaio sentences used go (or a reduplicated gogo)
as equivalent to kee, with absolute regularity.
Kwaio:
e kee nigi lolo'o me-'e bi'i leka folo-si-a
he AUX arrive on top and-he just go across-TRS-it
Then he arrived up there and went across (it).'
Pijin:
hern-i gogokas-im
antafu ana hem-i tas go koros-im
AUX reach-TRS on top and he
just go across-TRS
he
Note the pervasive pattern of calquing here, in which virtually all the mor
phemes in the two sentences directly correspond (including the final preposi
tional verb; note also the calquing of tas on bi).26
Kwaio:
ta'a
la kee kw ai-ri lau a-i
people they AUX report more Loc-it
'Then the people reported about that as well.'
Pijin:
olketa pipol
PLU
olketa-i go
people they
taleaotlong hem moa
AUX r e p o r t LOC
it
more
Then the people reported about that as well'

Kwaio:
alata gila leka mola27 gila kee a'a-ri-a
time they go just
they AUX carry-TRS-it taro-shoots
'When they set off, they were carrying taro shoots.'
Pijin:
taem olketa go nao olketa go kar-em
long taro
time they go PRF they AUX carry-TRS shoot GEN taro
333
The Kwaio form me'e, preceding an active verb, also operates as an

auxiliary in narrative, to convey meanings that in English would be expressed
by "went ahead and ...." The two Kwaio speakers used kam as an auxiliary
corresponding to me'e, with complete regularity.
Kwaio:
gila me'e fane
naa 'ue la'akau
they AUX ascend LOC vine DEM
'And then they went ahead and climbed that rattan vine.'
Pijin:
anaa olketa kam koafu long des-fala robu
and they AUX ascend LOC DEM-ADJ vine
'And then they went ahead and climbed that vine.'
Kwaio:
rua solodia gala me'e ula i
suri-a
two police they2 AUX stand LOC beside-him
'Two policemen would then stand flanking him.'
Pijin:
tu-fala solodia tufala kam sten weit-em
two-ADJ police they2 AUX stand with-TRS
Two policemen would then stand flanking him.'
The striking correspondences shown in these paired sentences, where in
most cases there is a morpheme-by-morpheme correspondence between the
Kwaio and Pijin, is manifest throughout the corpus: I have examined these
parallels elsewhere in some detail (Keesing, 1986, 1987, 1988, Chapter 13).
What is important here is that although calquing by Kwaio speakers
gives their uses of kam and go a measure of grammatical as well as lexical
force, these forms need not have precisely the same force for speakers of
other Solomons languages who use these forms. Let me turn to a Solomons
Austronesian language very distantly related to Kwaio, 29 Maringe, spoken
on Santa Isabel. I have illustrated above Pijin sentences produced by a
Maringe speaker which use kam and go as auxiliaries in serial verb con
structions:
kam faet
kam luk-im
go luk-im
kam dorop-im baum
334
ROGER M. KEESING
For Maringe, our only grammatical data come from a forthcoming sketch
grammar by Geoffrey White, the introduction to his dictionary (White, Forth
coming). White gives no information about serial verb constructions, but he
gives two exemplifying sentences in which the verb tei 'go' appears as the first
element in compound verbs:
na tei a noda ...?
FUT go bathe we ...
'Shall we go bathe ...?'
tei rughe ba-go
belo gno
go beat PSBL-you drum DEM
'Why don't you go beat the drum over there?'
Tei here is the common verb for 'go'. Although there are no examples given
by White in which the common verb mei 'come' is used in serial constructions,
we may guess that it is used in the same way as tei.
When Maringe speakers use kam and go in Solomons Pijin, they are
presumably using a grammatical system in which these forms function as the
initial verbs in serial verb sequences. When Kwaio speakers use the same
forms preceding other verbs, they have both a different semantic import and
a greater grammatical significance: They appear to be homonymous with the
verbs 'come' and 'go' (as, too, are the directionals kam and go, used postverbally as equivalent to mai and kau in Kwaio). The connections between these
forms that led to their incorporation into different grammatical categories
are, of course, exactly the sorts of metaphoric connections whose role in
grammaticalization is explored in a number of the papers in these volumes.
The point of all this is that a Melanesian Pidgin grammatical element
derived from an English lexical form may have acquired this grammatical
force not through the usual chain of grammaticalization but through a direct
calquing onto a substrate pattern. The semantic bleaching (if there is any)
takes place, as it were, in the very process of borrowing. The point can be
clarified with reference to nao as perfect-marker and topicalizer. The gradual
grammaticalization process whereby, in the history of Cristobal-Malaita lan
guages, a perfect-marking particle became used as topicalizing particle did
not have to be paralleled by a similar metaphor-based grammaticalization
process in Pijin. Once nao in the pidgin of the Labor Trade was equated with
n W as perfect marker, which in itself required little bending, it could be
transferred directly into the topicalization slot.
335
Another example from Solomons Pijin may further clarify this process.
In Malaitan and other Southeast Solomonic languages, relationships in space
are expressed through inalienably possessed nouns, treated grammatically as
if they were body-parts (or other parts-to-wholes): "The underneath of the
house," "(at) the back-part of the tree" (to express 'behind the tree'), etc.
Not surprisingly, relationships in time are treated in the same way: Metaphori
cally, being in front of or behind is used to express 'before' or 'after'. Thus
in Kwaio, na'o-na 'ifi 'in front of the house', and na'o-na omea 'before the
mortuary feast'; and buri-na 'ai 'behind the tree' and buri-na afolonga 'after
the wedding feast'. The incorporation of "behind" into a developing Pacific
Pidgin followed the same ("natural") metaphoric path. Thus, in the Solomons
Pijin of older Malaita speakers, bihaen long haos 'behind the house' and
bihaen long faet 'after the fight'. In nineteenth-century Melanesian Pidgin
fastaem was adopted from English "first time" (in such constructions as mi
go fastaem, yu go bihaen). Solomons Pijin adopted fastaem for 'before'
(fastaem long faet 'before the fight'). Strikingly, this chain of connection
being set into motion, Solomons Pijin (at least as used by older Malaita
speakers) adopted fastaem as a spatial marker: fastaem long haos 'the front
of the house'. Here we get a reversal of the normal sequence whereby spatial
meanings become metaphorically extended to temporal meanings, a reversal
made possible by the already-existing connections in substrate languages.
What matters is clearly not the direction of the original grammaticalization
in substrate languages, but the existence of a semantic bridge across which a
borrowed form can move through the calquing process. 30
Further illumination can come from the prepositional verb 'with'. In
adopting the lexical form "with" from English, but adding the transitive suffix
-im and using it as a prepositional verb, New Hebrideans and Solomon
Islanders in Queensland were in a sense grammaticalizing a lexical form. But
it was a very different grammaticalization process, and a less circuitous one,
than the process whereby in Oceanic languages prepositional verbs evolved
from second verbs in serial verb constructions. Similarly, fitting English "go"
into a series of slots as marker of passing-time and a directional as well
as a lexical verb Solomon Islanders were short-cutting rather than recapitu
lating the grammaticalization process whereby the lexical verb "go" spread
through these slots in the evolution of Oceanic Austronesian languages.
The way Kwaio speakers use go as equivalent to their auxiliary kee, and
use kam as equivalent to their auxiliary me'e, serves further to illustrate the
point. I have been unable to find, in other Southeast Solomonic languages,
clear evidence of the grammaticalization process whereby these Kwaio auxilia-
336
ROGER M. KEESING
ries developed. It is possible that they, or at least me'e, developed from lexical
verbs which have subsequently lost their verbal senses. Perhaps kee, indicating
that the events described in the clause followed those recounted in the preced
ing clause, is derived from a future/irrealis marker kajke common in Southeast
Solomon languages. Whatever the historical derivation of these auxiliaries,
the calquing whereby go was equated with kee and kam was equated with
me'e did not require the same grammaticalization sequence although as I
have noted, it did require a metaphoric extension of the lexical senses of go
and kam.
It should come as no surprise, since Silverstein's important work on
Chinook Jargon (1972), that throughout this process, speakers of English
(and other European languages) have brought to their encounter with Pacific
pidgin relatively different grammatical systems, and applied to pidgin quite
different grammatical analyses.
Once the core structures of Oceanic syntax had been incorporated into
the developing pidgin of the Pacific, a process which (judging by distributional
evidence and textual records) had effectively taken place by a century ago,
the lexical resources of the superstrate language provided the means for
further local elaborations. When the original vast and dispersed Pacific Pidgin
speech community began to separate into regional speech communities iso
lated from one another by colonial politics, forms such as "come" and "go"
and "stop" 3 1 could develop in different ways, being mapped onto grammatical
patterns broadly common to particular subgroups of substrate languages.
Thus, after the separation of New Hebridean and Solomons pidgin speech
communities at the beginning of this century and the establishment of internal
plantation systems, we find the standardization of stap as a durative aspectmarker in Bislama and nao as a perfect marker in Solomons Pijin. As recent
work by Darrell Tryon and Jean-Michel Charpentier in Vanuatu and my own
work in the Solomons shows, even within the now-separated national speech
communities, there are striking regional variations, some of which seem to
go back to the late nineteenth century. The pidgins of the two countries, still
very similar, so closely follow the grammatical and semantic patterns of
substrate languages, often on a morpheme-to-morpheme basis, that speakers
of pidgin can calque directly onto their native languages. As I have illustrated
here (and in Keesing, 1988 and Forthcoming), this calquing process then
allows a further fine-tuning of pidgin grammar to the particular syntactic
patterns and discourse strategies of speakers' first languages.
8.
337
CONCLUSIONS
The syntactic expansion, stabilization and regularization of Melanesian

Pidgin dialects has undoubtedly entailed grammaticalization, in a fairly nar
row sense. Lexical forms in an early jargon have been transformed into
grammatical elements in Pidgin dialects. The scenarios I proposed differ from
those advanced by Sankoff (1980)32 and Mhlhusler (1980) partly in placing
most of these expansions squarely in the nineteenth century rather than the
twentieth: Most of them are reflected in all three dialects of Melanesian
Pidgin, hence must have already have taken place by the time the Tok Pisin
lineage was separated by German political intervention a century ago (see
Keesing, 1988, Chap. 5); most appear in nineteenth-century texts. The scenar
ios I propose differ as well in seeing these expansions as directly connected
to the grammatical patterns broadly common to the Oceanic Austronesian
languages of the zone where nineteenth-century Pacific pidgin evolved. These
connections have been obscured from view in New Guinea, where Tok Pisin
has acquired a new and very different array of substrate languages.
The grammaticalization of preverbal markers of tense, mode, aspect and
mood out of lexical forms such as "by and by" and "might" and "I think,"
discussed above, will serve to illustrate. Pulling these forms into verb phrases
as grammatical markers clearly required some of the processes of grammati
calization on which other papers in these volumes are focused. However, the
existence of corresponding grammatical forms and patterns in substrate lan
guages ones that have often not been apparent to Tok Pisin specialists
would seem to have led to some shortcuts in the grammaticalization process
that would not have been possible if a simple jargon had progressively
developed into a creole without mapping onto a relatively uniform substrate
pattern. To take another example, discussed by Mhlhusler (1980), using
the transitive suffix to generate causative forms (hot-im 'heat it', etc.) in
Melanesian Pidgin dialects required some grammatical generalization; but
contra Mhlhusler, this pattern of forming causatives was already extensively
developed in EO languages. 33 It did not have to be invented, drawing on
universal faculties of language expansion, as Mhlhusler infers.
I have sought to show that where speakers of substrate languages, in a
situation of sustained interlingual communication such as that which prevailed
in the Pacific, are analyzing the lexical resources of a superstrate language in
terms of their own grammars, the processes of grammaticalization that oper
ate in natural languages can be complemented by still more direct pathways
to grammatical forms. For Malaita speakers, once nao had been calqued onto
338
ROGER M. KEESING
the perfect marker, it could be used as a topicalization marker with virtually

no reanalysis, semantic or syntactic.
Melanesian Pidgin, as Bickerton and many others have realized, consti
tutes a rare and special case, but one deeply revealing of the faculties enabling
language-learning, language simplification, language elaboration, and grammaticalization processes, both indirect and direct.
ABBREVIATIONS
ADJ
ART
DEIC
PRF
PSBL
TRS
adjective
article
deictic
perfective
possibility
transitive suffix
NOTES
1.
I am indebted to Elizabeth Traugott and Bernd Heine for particularly valuable suggestions
for revision of the original conference paper, and to fellow participants in the conference,
particularly Zygmunt Frajzyngier, Talmy Givn, Charles Li, Frank Lichtenberk, Ger
Reesink, and Gillian Sankoff, and to Christine Jourdan, for helpful discussion.
2.
Contra Mhlhusler (1976, 1978; Mosel and Mhlhusler, 1982; cf. Salisbury, 1967), I
argue (1988, Chapter 5) that the pidgin used in Samoa and parts of the Bismarcks until
the late 1880s was essentially the same as the pidgin being used at the same time in the
recruiting areas of the New Hebrides and the Solomons, and in Queensland.
3.
And innovations apparently assignable to Proto-Eastern Oceanic. The problems of Oce

anic subgrouping, and the tentative nature of the Eastern Oceanic subgroup in relation
to Southeast Solomonic languages, are discussed in Keesing, 1988, Chapter 6. In relation
to the arguments of my book, and this paper, the question on which this subgrouping
problem mainly rests whether the patterns that seemingly define the putative EO
subgroup may be shared retentions from or some interstage language is irrelevant.
It is the grammatical commonalities, not their time depth, that matter for these purposes.
On Oceanic grammar and EO languages, see Pawley, 1972 and 1973.
4.
By this I mean that Pidgin is still a second language for most speakers, and that their
first languages are the same EO languages that historically shaped the development of
southwestern Pacific pidgin; in contrast, Tok Pisin historically acquired a new set of
substratum languages, beginning at the turn of this century.
5.
A nice example of this occurs in Pohnpeian (in the eastern Carolines in Micronesia, an
area where Pacific pidgin disappeared over a century ago) where -im in the form kik-im
('kick s.o.') is treated as an allomorph of the Oceanic transitive suffix.
6.
Modern Solomons Pijin would use mi laek-em man blong nara ples.
339
7.
In the Solomons Pijin of older speakers, ou, baebae Riko kas-em, put -urn long kalabus
would be quite acceptable. The form derived from "calaboose" is one of a very substantial
number of lexical items widespread in Pacific Pidgin by the 1870s, and documented in all
the regional dialects of Melanesian Pidgin.
8.
But not in Tok Pisin, a distributional pattern that suggests a post-1890 incorporation into
the pidgin of the Labor Trade. However, it is possible that prepositional verbs were in
use a century ago but disappeared in New Guinea Pidgin because the support from
substrate languages had been lost in the process of transplantation.
9.
As I show in Keesing (1988, Chap. 11), speakers of Malaita languages have been the
dominant component of the Solomons labor force for well over a century, on the
plantations of Queensland, Fiji, New Caledonia and Samoa, and later, in internal Solo
mons plantations.
10.
Terry Crowley, personal communication, Jan. 1988.
11.
See e.g. Crowley 1984; examples from Maringe are given in a later section.
12.
Mhlhusler (1980) claims that until 1880, a developing Melanesian Pidgin had such a
simple syntax that formation of periphrastic causatives was impossible. This 1869 text
shows such a causative construction more than a decade earlier than Mhlhusler's claim
would allow.
13.
Provided to me by Geoffrey White.
14.
Substantial extracts from these texts have been published in Keesing (1986, 1987, and
1988).
15.
I have analyzed elsewhere (Keesing, 1986, 1988, Forthcoming) in detail the close corre
spondence between Melanesian Pidgin pronominal systems and those of substrate (and
particularly EO) languages. Here, I give only glosses of pronouns, to avoid the need for
lengthy explanation.
16.
See Silverstein (1972) for an analysis of this process in Chinook Jargon.
17.
FP here designates a focal pronoun, fitting into the subject NP slot (and usually syntacti
cally optional); SRP designates a subject-referencing pronoun, an obligatory subjectpronominal element in the verb phrase, marked for person and number.
18.
See Keesing (1988, Chaps. 9 and 10) for a sustained argument that the so-called "predicatemarker" in Melanesian Pidgin dialects is a pronoun, and corresponds directly to the
subject-marking pronominal elements in EO languages obligatory subject constituents
of verbal clauses represented in Vetmbao by a clitic prefixed to the verb and in the
examples from Malaita languages that follow, by such independent pronominal forms as
e and gala.
19.
See Keesing (1985) and, for Oceanic, Keesing (1988, Chap. 6).
20.
The sentences were the exemplifying ones given in my Kwaio grammar (Keesing, 1985).
See f.n. 22.
21.
Or des or das.
22.
More than four hundred Kwaio sentences, from my Kwaio Grammar (1985), almost all
of them originally taken from texts, for which equivalents in Pijin were elicited from
middle-aged Kwaio men, independently (and a year apart).
23.
In Tok Pisin, stap used as a durative marker occurs as a second verb in clausal serial
340
ROGER M. KEESING
constructions a further alternative analysis of the lexical resources of the Melanesian
Pidgin of the 1880s. Thus em bruk-im stap 'He keeps breaking it' (Givn, 1987).
24.
In the Oceanic pattern, no distinction is made between 'he', 'she' or 'it', but I have made
such distinctions according to context in glossing the morphemes.
25.
It it not strictly relevant for present purposes to propose an analysis of this development
in Malaita languages. However, Heine (personal communication, 1 January 1989) com
ments that "in some Niger-Congo languages of Nigeria (Nupe, Gwari), a verb 'take' has
been grammaticalized as a kind of perfective marker on the one hand and a completive
focus ("foregrounding") marker on the other. This appears to be similar to the situation
in Kwaio: it would seem that the topicalizer has sort of a focus/cleft function."
26.
The only difference between the two sentences is the use of kas-im 'reach' as a transitive
verb in Solomons Pijin, where Kwaio uses an intransitive 'arrive'.
27.
The Kwaio postverbal mola, which in many contexts is best glossed as 'just' or 'only',
sometimes serves in narrative as a perfect-marking particle, as it does here.
28.
The use of the Solomons Pijin prepositional verb weit-em 'with (him)' as corresponding
to the Kwaio prepositional verb suri- 'alongside (him)' again is striking, and was used
by both speakers.
29.
See Ross (1986).
30.
Heine (personal communication) suggests that "the development of first time (>fastaem)
from a temporal to a spatial marker among older Malaita speakers might form one of
the rare counter-examples to the unidirectionality principle of grammaticalization." The
point bears pondering. The case suggests to me that it is precisely where such calquing
is operating that we might expect such "counter-examples."
31.
Commonly used in working class British (and Australian) English to mean 'stay, be at a
place'.
32.
Which includes papers by Sankoff (1977), Sankoff and Brown (1976) and Sankoff and
Laberge (1973).
33.
See Keesing (1988, pp. 124-27).
REFERENCES
Bickerton, D . 1981. Roots of Language. A n n Arbor, M I : K a r o m a Publishers.
Bickerton, D . 1984. " T h e language bioprogram hypothesis." Brain and Behavior
Sciences 7:173-221.
Camden, W. 1979. "Parallels in structure of lexicon and syntax between N e w Hebrides
Bislama and the South Santo language as spoken at T a n g o a . " In Papers in Pidgin
and Creole Linguistics 2. Canberra: Australian National University [Pacific Linguis
tics A - 5 7 ] .
Charpentier, J.-M. 1979. Le Bislama(n): Pidgin des Nouvelles-Hbrides. Paris: S E L A F .
Clark, R. 1979. " I n search of Beach-la-Mar: Towards a history of Pacific Pidgin
English." Te Reo 22/23:3-66.
Crowley, T. 1984. "Serial verbs in Paamese." Paper presented at the F o u r t h Interna
tional Conference on Austronesian Linguistics, Suva.
341
Dana, R.H. 1840. Two Years Before the Mast. New York: Harper.
Deck, N. 1934. Grammar of the Language Spoken by the Kwara'ae People of Mala,
British Solomon Islands. Journal of the Polynesian Society reprint No.5, from Vols.
42-43.
Erskine, J.E. 1853. Journal of a Cruise Among the Islands of the Western Pacific.
London: J.Murray.
Fox, C.E. 1974. Lau Dictionary. Canberra: Pacific Linguistics [Pacific Linguistics
C-25].
Geerts, P. 1970. 'Are'are Dictionary. Canberra: Pacific Linguistics [Pacific Linguistics
C-14].
Givn, T. 1987. Serial Verbs and the Mental Reality of 'Event'. Final Project Report
to the National Endowment for the Humanities.
Goodman, M. 1985. Review of D. Bickerton, Roots of Language, International Journal
of American Linguistics 51(1):109-137.
Hezel, F.. 1983. The First Taint of Civilization: A History of the Caroline and Marshall
Islands in Pre-Colonial Days, 1521-1885. Honolulu: University of Hawaii Center
for Pacific and Asian Studies, University of Hawaii [Pacific Islands Monograph
Series, No.l].
Ivens, W.G. 1934. A Vocabulary of the Lau Language, Big Mala, Solomon Islands.
Journal of the Polynesian Society Supplement, Memoir No.l 1. Wellington: Polyne
sian Society.
Keesing, R. 1975. Kwaio Dictionary. Canberra: Australian National University [Pacific
Linguistics C-35].
Keesing, R. 1985. Kwaio Grammar. Canberra: Australian National University [Pacific
Linguistics B-88].
Keesing, R. 1986. "Subject pronouns and tense-marking in southeast Solomonic
languages and Solomons Pijin: Grounds for substratomania?" In Papers in Pidgin
and Creole Linguistics 4, S. Romaine et al. (eds), 97-132. Canberra: Australian
National University [Pacific Linguistics A-72].
Keesing, R. 1987. "Pijin calquing on Kwaio: A test case." In A World of Language:
Festschrift for Prof. S. A Wurm, D. Laycock and W. Winter (eds), 335-360.
Canberra: Australian National University [Pacific Linguistics C-100].
Keesing, R. 1988. Melanesian Pidgin and the Oceanic Substrate. Stanford, CA: Stan
ford University Press.
Keesing, R. Forthcoming. "Melanesian Pidgin and Oceanic syntax: Further evidence
from Solomons languages." To appear in Proceedings of Fifth International Con
gress on Austronesian Linguistics, Te Reo.
Lamont, E.H. 1867. Wild Life Among the Pacific Islanders. London: Hurst and
Beckett.
Lichtenberk, F. Volume I. "On the gradualness of grammaticalization."
Mosel, U. 1980. Tolai and Tok Pisin: The Influence of the Substratum on the Develop
ment of New Guinea Pidgin. Canberra: The Australian National University [Pacific
Linguistics B-73].
Mosel, U. and P. Mhlhusler. 1982. "New evidence for a Samoan origin of Tok
Pisin." Journal of Pacific History 17(3-4):166-175.
Mufwene, S.S. 1986. "The universalist and substrate hypotheses complement one
another." In Substrata Versus Universals in Creole Genesis, P. Muysken and N.
Smith (eds). Amsterdam/Philadelphia: John Benjamins.
342
ROGER M. KEESING
Mhlhusler, P. 1976. "Samoan plantation Pidgin English and the origin of New
Guinea Pidgin: An introduction." Journal of Pacific History 11(2):122-125.
Mhlhusler, P. 1978. "Samoan Plantation English and the origin of New Guinea
Pidgin." Papers in Pidgin and Creole Linguistics 1. Canberra: Australian National
University [Pacific Linguistics A-54].
Mhlhusler, P. 1980. "Structural expansion and the process of creolization." In
Theoretical Orientations in Creole Studies, A. Valdman and A. Highfield (eds).
New York: Academic Press.
Mhlhusler, P. 1981. "The development of the category of number in Tok Pisin."
In Generative Studies on Creole Languages, P. Muysken (ed.). Dordrecht: Foris
Publications.
Mhlhusler, P. 1985a. "External development of Tok Pisin." In Handbook of Tok
Pisin (New Guinea Pidgin), S.A. Wurm and P. Mhlhusler (eds). Canberra:
Australian National University [Pacific Linguistics C-70].
Mhlhusler, P. 1985b. "Internal development of Tok Pisin." In Handbook of Tok
Pisin (New Guinea Pidgin), S.A. Wurm and P. Mhlhusler (eds). Canberra:
Australian National University [Pacific Linguistics C-70].
Mhlhusler, P. 1987. Pidgin and Creole Languages. London: Basil Blackwell.
Paton, G. 1895. Letters and Sketches form the New Hebrides. London: Hodder and
Stoughton.
Pawley, A. 1972. "On the internal relationships of Eastern Oceanic languages." In
Studies in Oceanic Culture History, Vol. 3, R.C. Green and M. Kelly (eds). Pacific
Anthropological Records 13:1-142. Honolulu: Bernice P. Bishop Museum.
Pawley, A. 1973. "Some problems in proto-Oceanie grammar." Oceanic Linguistics
12:103-188.
Pawley, A. and L.A. Reid. 1979. "The evolution of transitive constructions in
Austronesian." In Austronesian Studies: Papers from the Second Eastern Conference
on Austronesian Languages, P.B. Naylor (ed.), 103-130. Ann Arbor, MI: University
of Michigan.
Ross, M. 1986. "A genetic grouping of Oceanic languages in Bougainville and the
Western Solomons." In Proceedings of the Fourth International Conference on
Austronesian Linguistics, P. Geraghty, L. Carrington, and S.A. Wurm (eds). Can
berra: Australian National University [Pacific Linguistics C-94].
Salisbury, R. 1967. "Pidgin's respectable past." New Guinea 2(2):44-48.
Sankoff, G. 1977. "Variability and explanation in language and culture: Cliticization
in New Guinea Tok Pisin." In Linguistics and Anthropology, M. Saville-Troike
(ed.). Washington D.C.: Georgetown University Press (Reprinted in Sankoff, 1980).
Sankoff, G. 1980. The Social Life of Language. Philadelphia: University of Pennsylva
nia Press.
Sankoff, G. and P. Brown. 1976. "The origins of syntax in discourse: A case study
of Tok Pisin relatives." Language 52:631-666 (Reprinted in Sankoff, 1980).
Sankoff, G. and S. Laberge. 1973. "On the acquisition of native speakers by a
language." Kivung 6:32-47 (Reprinted in Sankoff, 1980).
Silverstein, M. 1972. "Chinook jargon: Language contact and the problem of multi
level generative systems." Language 48:378-406;596-625.
Simons, L. 1986. "Malaitan influence on two grammatical particles in Solomon Islands
Pijin." In Papers in Pidgin and Creole Linguistics, No.4, S. Romaine et al. (eds).
Canberra: Australian National University [Pacific Linguistics A-72].
White, G. Forthcoming. Dictionary of Cheke Holo (Maringe). Canberra: Australian
National University [Pacific Linguistics, In press].
Volume I
Index of languages
A
Alamblak 86, 124
Algic 307-308
Algonkian 308, 313
Aliutor 331
Altaic 61
Amharic 226, 236, 238, 311
Annamalai 274
Arabic 311
Aramaic 302, 313
'Are' are 46, 47, 53,54, 65,329
Arosi 46, 47, 53,65,70, 78
Ashante, see Twi
Atlantic creoles 327
Austro-Tai 121
Baoan 312
Beja 236
Binticua 310
Bislama 316-317, 318, 320, 322, 324,
325, 326, 328, 329-330, 336
Boruca, Borunca 307, 310
Bribri 306
Buryat 331
Cayuga 12
Chehalis, Upper 308
Chibcha, Chibchan 306-307, 309, 310
Chinese 121
Mandarin 82, 99
Chinook Jargon 331, 336, 339
Chuave 124
Creole, Caribbean, 121
Cuna 309, 310
Cushitic languages 154
Czech 227
D
Dagur 312
Danish 199
Dargua 226
Dinka 303
Drehu 236
Dusnian 312
Dutch 195
E
Estonian 195, 197, 285-297
Estrella 307
Ewe (Gbe) 154, 161-164, 172-176, 182,
183, 184, 236
F
Faroese 221
Finnish 197, 204, 238, 286, 289, 290,
291, 292, 296
Formosan 46
French 18, 24, 25, 27, 31, 195, 197, 198,
199, 210, 223, 240, 241, 280, 282, 289,
290, 291, 292, 309, 310, 311
Old 26
G
Ga 28, 225
344
INDEX OF LANGUAGES
Gayon 307
Gbe, see Ewe
German 12, 192, 195, 199, 201, 204,
212, 236, 237, 241, 247, 280, 282, 289,
290, 303, 316
Ge'ez 236
Gorokan group 107
Greek 311
Guadalcanal-Nggelic 46
Gujarati (Gujerati) 27
Gwari 340
H
Hausa 225, 226, 241
Hebrew, Biblical 12
Hindi 27
Hungarian 245
I
Iroquois 83
Italian 309, 310, 311
Itonama 306
J
Japanese 12, 310
Kadam 182
Kagaba 310
Kalam 85, 86, 89, 93, 96-107, 110, 111,
114, 115, 116, 117, 118, 119, 120, 124
Kanuri 305
Karimojong 305
Kashmiri 27
Khalka 312
Klamath 12
Koryak 311
Kuliak languages 155
Kwaio 46, 53, 54, 70, 71, 72, 73, 74, 78,
319-320, 322, 323, 326, 327, 328, 330,
331, 332, 333, 334, 335, 339, 340
Kwara'ae 46, 47, 53, 54, 70, 71, 73, 78,
329
K'ekchi 326
L
Latin 18, 25, 195, 196, 197, 198, 204,
206, 210, 213, 220, 309, 310, 311, 313
Vulgar 311
Latvian 289
Lau 46, 47, 53, 54, 70, 71, 73, 78, 324,
330
Lithuanian 289
Livonian 285, 289
M
Maasai 303, 313
Malaita 324, 325, 329, 330, 337, 339,
340
Malay, Literary 254
Malayo-Polynesian 46
Mandarin see Chinese
Marathi 27
Margua 310
Maringe 321-322, 333, 334
Masai see Maasai
Melanesian Pidgin 7, 78, 315-340
Mon Khmer 121
Mongolian 311-312
Monguor 312
Mupun 222-223, 227-236, 239, 242
Mura 307
Murire 307
N
Newari 12
Niger-Congo 121, 226, 340
Nilo-Saharan 301, 302, 303, 305, 308,
313
Oceanic 46, 323

Oromo 154
P
Pacific jargon 321
Pacific pidgin 316, 322, 336, 337, 338,
339
Paya 310
INDEX OF LANGUAGES
Penutian 301, 305
Pero 239
Pohnpeian 338
Polish 223, 225, 226, 241, 244, 245
Proto-Algonquian 308
Proto-Balto-Finnic 290
Proto-Indo-European 245
R
Rama 6, 12, 306, 307, 309
Rumanian 195, 199, 204
Russian 223, 225, 241
S
Sa'a 46-47, 53, 54, 64, 65,78
Sahaptn 12
Salish 307, 308, 313
Sanskrit 274, 308
Sardinian 313
Selayarese 12
Shilluk 303
Slavic, Old Church 137, 138, 145
Slovak 227
So 155, 177-179, 182
Solomon Pidgin, see Solomons Pijin
Solomons Pijin 316-326, 328-336, 339,
340
Spanish 195, 204, 241, 307
Squamish 308
Swahili 87
Swedish 195
T
Tabasara 226
Tairora 89, 107-114, 115, 117, 118-119,
124
345
Tamil 253-283
Tepes 182
Terraba 309
Teso 169, 184
Thompson River Salish 307
Tibeto-Burman 83, 121, 238
Timucua, Timucuan 307, 308, 313
Toba Batak 238
Tok Pisin 83, 87-96, 100, 102, 110, 111,
114, 115, 117, 119, 120, 124, 313, 315,
337, 339
To'aba'ita 38-78, 320
Turkish 199
U
Uzbek 311
V
Vanuatu 315, 316, 336
Vetmbao 326, 339
W
Waata 154
Warrau 306
Wiyot 308
X
Xinalug 226
Y
Yiddish 236, 446
Yoruba 225
Yurok 236, 308
Volume I
Index of names
A
Abraham, W. 2, 12, 195, 198
Allan, . 152, 160, 181
Alvre, P. 290, 291,292
Andersen, H. 143, 144, 193
Anderson, L.B. 220, 294
Anderson, S.R. 220, 294
Annamalai, E. 274
Anttila, R. 210,211,296
Ariste, P. 291
Aristotle 124
Aronson, D. 86
Atlas, J.D. 191, 194
Awolaye, Y. 152
Awobuluyi, O. 84, 121
Bamgbose, A. 84, 121

Barrai, . de 306
Bauerova, M. 146
Becker, P. 124
Benveniste, E. 209
Berlin, B. 156
Berman, R. 124
Bickerton, D. 8, 10, 315, 338
Biener, . 293
Blakemore, D. 193, 198
Bloomfield, L. 248
Blust, R. 46
Boas, F. 149, 249
Bopp, F. 2, 149
Bradshaw, J. 82, 121
Braunmller, K. 195
Breckenridge, J. 296
Breedlove, D.E. 156
Brown, P. 340
Bruce, L. 109
Brugman, C. 80
Burquest, D. 28
Bush, U. 145
Bybee, J.H. 3, 9, 11, 19, 29, 123, 149,
151, 156, 181, 189, 190, 207, 208, 209,
294
Camden, W. 328
Camp, E. 206
Campbell, L. 2, 6, 11
Capell, A. 78
Carlin, E. 181
Carlson, R. 2, 12
Carpenter, P.A. 86
Chafe, W. 86, 88, 267
Channon, R. 222
Charpentier, J.-M. 326, 328, 336
Chomsky, N. 124
Chung, S. 37
Claudi, U. 11, 123, 149, 152, 157, 168,
183, 184, 207, 247, 249, 253
Clements, G.N. 238
Cloarec-Heiss, F. 238
Cohen, M. 238
Comrie, B. 238, 286, 294, 295
Craig, C. 2, 5, 12
Cresswell, 220
Croft, W. 6
Crowley, T. 82, 83, 95, 121, 321, 339
348
INDEX OF NAMES
D
Dana, R. 318, 340
Deck, N.C. 71,72,73,78,329
DeLancey, S. 83, 238
Denning, . 313
Deeriev, J. 226
Dietrich, 204
Dirven, R. 210
Disterheft, D. 294
DuBois, J. 3, 76, 254, 277
Durie, M. 82, 121
E
Ebert, K. 225, 238, 249
Edmiston, P. & M. 124
Eisler-Goldman, F. 86
Erskine, J.E. 318, 321
Essien, O.E. 152
F
Ferres, S. 86
Fillmore, C.J. 282
Fleischman, S. 157, 207, 280
Foley, W. 85, 121
Fox, C E . 78, 324
Fraenkel, E. 146
Frajzyngier, Z. 3, 11, 214, 226, 227,
233, 238, 239, 240, 245, 338
G
Gabelentz, G. von der 2, 149
Gamillscheg, E. 26
Gatschet, A.A. 307, 308, 313
Geerts, P. 78, 329
Geis, M.L. 194, 195
Genetti, 6, 12
Gerritsen, M. 37, 288
Givn, T. 2, 3, 5, 8, 11, 12, 18, 23, 37,
57, 82, 83, 86, 121,122, 149,153, 155,
157, 160, 169, 171, 181, 184, 189, 190,
193, 207, 220, 254, 276, 277, 279, 280,
293, 294, 295, 296, 322, 338, 340
Goodman, M. 316
Goral, D. 82, 121
Grace, G.W. 46, 85, 124
Greenberg, J.H. 2, 5, 6, 7, 8, 11, 169,
183, 219, 223, 240, 301, 304, 307, 309,

313
Grsillon, A. 258, 282
Grice, .P. 193
Grnthal, W. 288
Gudschinsky, S.C. 152
Guillaume, G. 149, 156
Guimier, . 149, 156
Glich, E. 259, 260
H
Haas, M. 308
Hagge, . 238
Haiman, J. 12
Hakulinen, . 289, 290, 296
Halle, M. 210
Halliday, M. 183, 254
Harris, A.C. 294, 295
Hart, G.L. 263
Hasan, R. 254
Healey, 214
Heine, . 1, 2, 4, 6, 7, 11, 18, 24, 33,
123, 149, 152, 153, 156, 157, 162, 166,
167, 168, 169, 171, 178, 182, 183, 184,
190, 198, 207-208, 212, 213-214, 233,
249, 296, 312, 340
Herring, S.C. 3,5, 11,254
Hezel, F. 321
Hilders, J.H. 184
Hobbs, J.R. 193
Holmer, N.M. 309, 310
Hook, P. 3, 9, 11, 27, 213
Hooper, J.B. see Bybee, J.H.
Hopper, P. 1, 3, 4, 5, 8, 11, 12, 19, 30,
37, 65, 75, 78, 157,175,176,184, 203,
254
Horn, L.R. 194, 214
Hughes, 220
Humboldt, W. von 2, 149
Hnnemeyer, F. 11, 33-34, 122, 253
Huttar, G. 121
Hyman, L. 82, 84, 121, 122, 238
I
Ikola, O. 289, 290, 296
Ivens, W.G. 64, 78, 330
INDEX OF NAMES
J
Jakobson,R. 210
Jespersen, . 33
Johnson, M. 151, 181, 183, 207
Jourdan, C. 338
Jungraithmayr, H. 238
Just, M.A. 86
Kaplan, . 150
Karlsson, F. 238
Katz, J.J. 84, 124
Kaufman, T. 289
Keenan, E.L. 95, 104, 220, 237
Keesing, R. 2, 7, 11, 54, 71, 78, 315,
318-319, 330, 333, 336-337, 338, 330,
340
Kemmer, S. 6, 213
Kettunen, L. 289, 290
Kirsner, R. 226, 235
Koontz, J. 249
Knig, E. 2, 4, 5, 7, 11, 151, 165, 166,
181, 199, 213
Kronasser, 210
Kuipers, A.E. 308
Kumpf, L. 86
Kuryowicz, J. 2, 149, 172, 182, 296
L
Laanest, . 290
Laberge, S. 340
Labov, W. 4
Lakoff, G. 151, 181, 183, 207
Lamont, E.H. 318, 321
Langacker, R. 7, 9, 149, 167, 184, 294,
295
Lawrence, J.C.D. 184
Lehmann, C. 1, 2, 4, 6, 7, 17, 20, 21, 25,
28, 149, 156, 171-172, 182, 183, 184,
189, 190, 191
Lehmann, W. 306, 307
Leino, P. 289, 290, 296
Leumann, M. 211
Levinson, S.C. 190, 191, 194, 213
Li, . , 37, 75, 84, 121, 338
349
Liccardi, M. 306
Lichtenberk, F. 3, 4, 11, 12, 42, 46, 59,
78, 320, 338
Lightfoot, D. 76, 293, 296
Lockwood, W.B. 221
Lord, . 28, 37, 82, 121, 122, 123, 155,
167, 225
Lhr, R. 23
M
Martin, J. 37
Matisoff, J.A. 2, 7, 12, 82, 121
Matsumoto, Y. 6
Maczak, W. 288
Mauthner, F. 181
Meillet, . 2, 17, 18, 33, 34, 149, 190
Michelson, T. 308
Mitchell, . 195
Mithun, M. 3, 9, 12, 83
Mosel, U. 338
Mufwene, S.S. 315, 316
Mhlhusler, P. 316-317, 337, 338-339
Mulac, . , 12, 20, 83
Napoli, D.J. 204

Nevis, J.A. 292, 295
Nichols, J. 3,4, 11, 145, 146
Noonan, M. 221
Nurse, D. 152, 181
Olson, M. 85, 121

stman, J.-O. 283
P
Pagliuca, W. 9, 11, 29, 149, 151, 156,
190, 207, 208, 209, 294
Patokova, .V. 145, 146
Paton, G. 321
Pawley, A.K. 45, 46, 82, 85, 96, 118,
119, 124, 267, 318, 325, 338
Peirce, C.S. 193
Perkins, R.D. 11
Plank, F. 294
350
INDEX OF NAMES
Pope, E.N. 281

Popovid, V. 260
Potebnja, A.A. 146
Pusch, L.F. 247
Q
Quirk, R. 23
R
Raven, P.H. 156
Reesink, G. 338
Reid, L.A. 318
Reh, M. 1, 6, 18, 24, 123, 152, 156, 162,
166, 167-168, 169, 171, 178, 182, 183184, 190
Rickford, J. 10
Romaine, S. 12,37,75, 192
Ross, J.R. 122, 281
Ross, M. 340
Rude, N. 2, 12
S
Salisbury, R. 338
Saltarelli, M. 37
Sankoff, G. 10, 317, 337, 338, 340
Sapir, E. 120, 207, 210
Schachter, P. 84, 121
Schaller, H. 145
Schiefelin, E. & B . 124
Schmidt-Radefeldt, J. 258, 259, 260,
282
Schneider, W. 157
Scholz, L. 124
Schottelndreyer, . 250
Schtz, . 121
Serzisko, F. 181
Shepherd, S.C. 209
Shibatani, M. 3, 12
Silva-Corvaln, C. 37
Silverstein, M. 331, 336, 339
Simons, L. 329, 330, 331
Slobin, D.I. 12, 120-121, 124
Sperber, D. 165, 190, 192, 214
Stahlke, H. 82, 84, 121
Stanley, C. 238
Stern, G. 210-211
Stieber, Z. 227
Svensson, P.F. 296
Svorou, S. 182
Swadesh, M. 152
Sweetser, E.E. 8, 157, 158, 190, 198,
207, 213
Syder, F. 267
Szymborska, W. 241, 242
T
Talmy, L. 8, 83, 158, 191
Tesnire, L. 171
Thiel, 307
Thomason, S. 289, 368
Thompson, S. 3, 5,12, 20, 30, 33-34, 82,
83-84, 99, 121, 157, 175, 176, 184,
204, 226, 235
Thurman, R. & R. 124
Timberlake, A. 3, 4, 37, 131, 296
Tomiin, R. 86
Tossou, K. 182
Traugott, E.C. 1, 4-5, 7, 11, 19, 31, 33,
77, 151, 156, 158, 165, 166, 181, 182,
184, 189, 191, 207, 208, 209, 213, 219,
253-255, 269, 278, 279, 281, 312, 338
Tryon,D. 324, 336
U
Ullmann, S. 211
V
Vaillant, A. 146
Valdivia, L. de 310
Valentini, 307
Van Schooneveld, C H . 144
Vandeweghe, W. 259, 260
Venezky, 214
Vincent, A. 107, 124
W
Wackernagel, 292
Watson-Gegeo, . 54, 78
Wehr, . 256
Werner, H. 150
INDEX OF NAMES
White, G. 334, 339
Whorf, B.L. 84, 120
Wiegan, N. 5, 37
Wilkins, D. 211
Wilson, D. 165, 190, 192, 214
Wright, S. 220
Z
irmunskij, V.M. 156
Zwicky, A.M. 150, 194, 195
351
Volume I
Index of subjects
A
abduction 143, 193
ablative 40, 44, 47, 48, 50, 51, 52, 66,
67, 68, 71, 73, 177, 312
ablaut 24
absolut(iv)e case 177
abstraction 155, 156, 253, 279
generalizing 156, 157
isolating 156, 157
metaphorical 157, 158
abstractness 4, 172
addressee
de dicto 239
de re 239
adjective 305
adjustment 123, 178
adverb
durative 133
future 197
phrasal 203
preference 203
temporal 203
adverbial, time 132
affixation 23, 24
agent 87, 318
agent noun see noun
agreement 96, 98, 107, 108, 190, 244
alienable (category) 172, 184
allative 44, 47, 48, 52, 66, 67
allomorphy 338
ambiguity 59, 146, 180
semantic 162, 178
analogical extension, see extension
analogy 7, 17, 123, 143, 211, 255

anaphora 219, 221, 223, 228, 229, 277
locative 239, 240
propositional 222, 228
andative 78
anaphoric marker 189
andative 117, 118
animacy 42, 44, 45, 63, 64
aorist 134, 136
apocope 291
appositive 137
article (definite) 20, 30, 219, 223-225,
228, 301, 302, 304, 310, 313
indefinite 24
aspectogenesis 11
aspectuality 130, 132, 134, 135, 139,
141, 142, 143
asymmetry 172, 174, 178
autonomy
discourse 31
grammatical 275
auxiliary (verb) 23, 24, 25, 27, 29, 144,
189, 209, 321, 322, 332-337
modal 214
backgrounding 41, 42, 293

basic vocabulary see vocabulary
benefactive (beneficiary) 48, 58, 59, 82,
87
bleaching, (semantic) 4, 5, 149, 155,
156, 190, 253, 255, 279, 294, 334
body parts 306, 307
354
INDEX OF SUBJECTS
bondedness 5, 189
boundary
constituent 168
creation 184
loss 184
morpheme 291, 295
syllable 295
calquing 315-340
case
inflection 9
marker 6, 28, 30, 38, 40, 59, 61, 67,
118
marking 82, 102, 104, 121, 178
relation 271
cataphora 96, 107, 108
categoriality 34, 176, 189
degree of 30
principle 30, 31
category, grammatical 11
causation 194
causative 81, 87, 90, 92, 95, 98, 100,
105, 196, 308, 312, 337
cause 166, 273
center, deictic 155
chain, grammaticalization 171-174, 178,
179, 180, 181, 184, 334
chaining 119, 160, 173, 181
change
grammatical 297
language 3, 4, 7, 21, 198
lexical 207
linguistic 21, 192
meaning 5
metaphorical 4, 212
metonymic 212
morphological 297
sound 296
semantic 19, 207, 208, 210, 212, 311
syntactic 37, 297
channel, grammaticalization 9, 171
classification, nominal 305
classifier 30
numeral see numeral classifier
clause
adverbial 81, 122
complement 229, 230
complex 296
conjoined 193
coordinate 169
embedded 234, 236, 239, 296
main 6, 12, 169-170, 201, 229, 230,
231, 232-234, 236, 239, 286-287,
293, 296
purpose 72, 73, 81
relative see relative clause
subordinate-6, 169-170, 201, 285,
289, 293-294, 296
type 115, 116
clause-chaining language 96, 107, 199
cline 5
clitic 24, 189, 192, 291, 294, 319, 339
coalescence 21
co-grammaticalization 93, 95, 102, 106,
111, 112, 117, 118
co-lexicalization 82, 93, 95, 102, 106,
111, 112, 117, 118
coding, explicit 57
coding efficiency see efficiency
cognate 23, 47, 53, 54, 64, 70, 74, 76,
78,308
collective 23, 47, 53, 54, 64, 70, 74, 76,
78, 308
collective 305, 310
comitative 44, 47, 48, 49, 51, 52, 54, 55,
56, 57, 58, 61, 63, 64, 65, 66, 75, 312
comment 237
participial 297
sentential 225
complementizer 38, 40, 66, 67-74, 77,
78, 220-236, 248
completive (aspect) 90, 99, 118
complexity
semantic 182
syntactic 317
concatenation, verbal 12
concept
abstract 9
possessive 159
INDEX OF SUBJECTS
concessive 5, 190-193, 199, 201, 202,
203, 209
condensation 21, 149
condition, phonological 56
conditional 93
confective 44, 47, 48, 52, 56
conjugation 311
conjunction 38, 60, 61, 62, 63, 64, 66,
75, 77, 90, 91, 92, 94, 115, 122, 166,
221, 272, 273, 274, 276, 278, 289
concessive 254
coordinating 60, 75, 273
subordinating 258, 272
connective 189-191, 193, 198, 199, 203,
208, 209
constituent see dependent constituent
contamination 289, 290
context 4, 8, 62, 122, 136, 139, 142, 143,
158, 164, 181, 193, 196-198, 200, 205,
211, 212, 214, 289, 290, 293
discourse 11, 255
pragmatic 255
semantic 9
context-induced reinterpretation 164167, 171, 181
continuum 10, 122, 163, 164, 179, 180,
279
of bondedness 189
contrast 32
controller 130
conventionalization 11, 130, 144
coordination 60, 61, 62, 63, 64, 65, 77
copular construction 138
creativity 10
creole 315, 318, 337
D
de dicto domain 219-249
de re domain 220, 238, 241, 244, 246,
247-249
de-categorialization 30-33, 176, 184
decliticization 291, 294
deduction 143
definite see article
definiteness 6
355
degrammaticalization 150, 182

deixis see center
deixis, discourse 240
demonstrative 10, 152, 169, 171, 172,
220, 221-236, 238, 239, 240, 302, 304,
305
locative 241
deontic modality 208, 209, 229
dependency 170
dependent constituent 154
depletion, semantic 156
derivation 10
desemanticization 4, 156, 301-313
contractive 308
expansive 308-309
deviation 139
diachrony3, 11,77, 136, 172
dialect continuum see continuum
diminutive 310
directional 47, 322, 324-325
discourse
autonomy 30, 31
world of 158
function 31
grammaticalizing 255
perspective 2
role 174-176
strategy 3
distance, cognitive 175
divergence 24-25, 31, 32
domain
cognitive 161
conceptual 157
functional 22, 24
of reality 219, 240
of speech 219, 227
drift, language-internal 10
dual 43, 45,61
E
efficiency, coding 57, 71
elaboration 338
ellipsis 44, 57, 288
embedding 191
entropy, semiotic 129, 143
356
INDEX OF SUBJECTS
epistemic causal 198

modality 229, 233-234
phrase 12
marker 209
event 30, 81-125, 134, 195, 201, 205, 234
speech 242
evidential function 286
evidentiality marker 83
evolution 6, 27, 182
expansion 6, 315, 337
explanation 74, 77, 220, 244, 246, 248,
259, 288-290, 293, 308
teleological 9
expressive function 254, 264
expressivity 269
extension 123, 138, 139, 140, 325
diachronic 254
gradual 164
metaphorical 214, 220, 240, 336
extrapolation 21, 130, 143
F
feminine (gender) 33, 242, 246, 247, 248
finiteness, degree of 102
fixation 21
focus (marking) 139, 254, 265
foregrounding 43, 331, 340
form, life 156
fossilization 307, 398
frequency
discourse 26
of use 40, 51-56, 66, 67, 68, 69, 70,
75, 291, 304
function see ideational, interpersonal,
textual
deontic 229
epistemic 229
pragmatic 191
future marking 11
G
gender 241, 247, 308,313
marker 29
sex 169
generalization 7, 308
genesis 11
genitive 172, 178, 238, 297
given information see information
goal (function) 271, 318
grammar, emergent 78
grammaticalization
definition of, 1-3, 38, 149, 189, 296
degrees of 23, 24
chain see chain
instantaneous 123
principles of 17-33
grammaticization see
grammaticalization
grammatization see grammaticalization
H
hierarchy
implicational 5
personal 160
role 160
social status 160
homonymy 175, 180, 305
homphony 180
honorific 310
hybrid, linguistic 5, 174
hypothetical mood see mood
I
iconicity 71, 74, 76, 85, 86,168, 178,180
ideational function 158, 183
idiom, frozen 8
idiomatization 9, 129, 143, 144
illocutionary force see force
image-schema 191, 213
implicature, conversational 190, 192,
194, 197, 211
inalienable (category) 184, 335
inceptive 23, 136, 196
indefinite 306, 307, 310
indexing 42, 44, 47
inchoative 140
inferential (report) 286
inflection 189
informal speech 291
information
INDEX OF SUBJECTS
given/old 270
new 270
information processing 11
informativeness
principle 191, 192
strengthening of 207, 210
ingressive 154
innovation 11, 130, 143, 146, 307
intentional 153, 154
interpersonal function 158, 183
interrogative pronoun see pronoun
intonation 192, 259, 266, 273, 277
intransitive 89, 313
intransitivizer 308
irrealis 96, 97, 108, 293, 296, 325, 327,
335
iterative (aspect) 23
kin-term 307, 308

L
language
discourse-grammaticalizing 255
lexical-grammaticalizing 255
OV119
SVO 87, 93, 107, 119
langue 176, 189
layer 32
layering 22, 32
lexical-grammaticalizing 255, 278
lexicalization 7, 8, 301, 311, 325
lexicostatistics 152
lingua franca 315, 316
location 20
logophoric system 238
M
manipulability 157, 174, 176
manner 158
masculine (gender) 241, 242, 246, 247
meaning, pragmatic 5
meaning component see component
merger 311
metaphor see extension; vehicle
357
metaphor 11, 157-160, 165-168, 176,

181, 183, 190, 207, 208, 210, 212, 255,
325, 334-336
categorial 157, 161, 181, 183
conceptual 183, 207
expressive 183
taboo 183
metathesis 78
metonymy 4, 11, 164, 165-167, 210-212
modal particle see particle
mode
pragmatic 277
syntactic 277
modus obliquus 285-289, 293-294
mood, hypothetical 220, 227, 240
morphologization 4, 21, 28
motivation, cognitive 10
motivation for grammaticalization 9
N
narrative 97, 104, 130, 135, 139, 143,
196, 255, 264, 324, 332, 333, 340, 306
negation 257, 292
negator 26, 27
new information see information
nominality 305
nominalization 182, 266
nominalizer 47, 306
noun
agent 310
body part 6, 306, 308
class 305, 313
class system 10
concrete 310
descriptive 131, 132, 140, 141
function 132, 140, 141, 142
identificatory 131, 132, 141
predicative 129, 130
verbal 310
nouniness 122
number 129
numeral classifier 309, 310
object, obligatorification 21, 25, 27
358
INDEX OF SUBJECTS
optionality 30
order see word order
orientation, spatial 152, 182
origin of grammar 18
lexical 10
verbal 124
ossification 129, 143
overlapping 162, 163, 173, 177, 180
packaging
cognitive 81-125
temporal 11, 86
panchronic grammar 130
model 135, 143, 144
panchrony 3, 11, 172
paradigm 203, 311
grammatical 9
morphological 10
paradigmatization 21, 28
parole 76, 189
part-whole proposition 170
participant 30, 33, 176
speech 157
participial 30
participle 137, 144, 196, 275, 285, 286290, 293, 294, 296, 297
nominal 12
verbal 12
past tense 96, 326
pause 86, 88, 92, 93, 94, 95, 100, 102,
105, 110, 111, 112, 113, 114, 115, 119
periphrasis 23, 24
perspective, diachronic 258
phonologization 4
pidgin 315-340
pidginization 55
plural 42, 43, 45, 57, 60, 71, 78, 243,
245, 246, 304, 312, 313
plurality 245, 305
polygrammaticalization 12
polyseme 180
polysemy 161, 175, 180, 193, 195, 198
possession 159
inalienable 184, 307
verbal 154
possessor 60, 183,238,306
inalienable 307
postposition 9, 292, 295
pragmatic feature 265
predicate-marker 90
predicative 129, 130
predictability 143
prepositional verb, see verb
pressure, discourse-pragmatic 9
preterite 141
principle 11, 21, 32
problem-solving strategy 150, 160, 181,
212
process 6, 122, 157, 159, 207
coginitive 192
cyclical 142
progressive (aspect) 153, 182, 184, 207
prolative 44, 47, 48, 49, 52, 66, 67
pronoun
anaphoric 302, 305, 313
interrogative 159, 241, 243-245, 310
relative 244, 275
reflexive 152
resumptive 317
proper name 223, 304
preposition 85, 124, 198, 209, 221, 234,
237, 242, 246, 254
action 153
hypothetical 227
interjected 237
locational 153, 182
motion 153
part-whole 153, 181
source 153, 181
propositional
function 254
meaning 31
prosody 273
proto-language 77
prototype 305
pseudo-cleft construction 266, 268, 280
purpose clause see clause
push chain 74
INDEX OF SUBJECTS
359
R
real world 124
reanalysis 7, 58, 75, 123, 143, 149, 167170, 178, 184, 302, 303, 324-325
constituent-internal 167, 184
reason 67, 68, 70, 74, 273
re-categorialization 184
reduction 7, 149
phonological 40, 51, 56, 59, 294
in scope 6
reduplication 323
reference point 89
referentiality 157, 174
reformulation 184
register 39
regrammaticalization 294, 301-313
regularity 76
regularization 337
reinterpretation 7, 291, 305, 306, 310
relation
scalar 115
part-whole 211
relative mood 289
relativist position 84
relativization 42, 44, 57, 220
relativizer 276
reformulation 184
reported speech 286, 289
resegmentation 184
re-semanticization 29
retextualization 129-146
rule, phrase structure 296
serial verb 3, 81-125

construction 321-322, 333, 335, 339
sex, see gender
shift, see topic
boundary 184, 303
conceptual 166, 180
functional 74
functional-semantic 30
simplification 316, 338
singular 60, 243,246, 313
singularity 305
singulative 305, 310
space 157, 159-165, 173, 175
topological 8
specialization 25-28, 32
specific, indefinite 305
speech-processing 120
split 24, 162, 166
stabilization 315, 337
status, categorial 175
stem, co-lexical 121
strategy, communicative 8
strengthening 253, 255, 279
stress 56
subduction 149
subjectification 254, 279
subordination 278, 280
subordinator 6
clausal 12, 256, 258
substance, phonetic 182
substrate 7, 11,313,315-340
superstrate 11, 316, 317, 321, 326, 331,
336, 337
switch-reference 96, 214
syllable 56, 57
syncretism 220, 245
synecdoche 211
syntacticization 3, 18, 149, 276, 277
syntax, historical 288
S
sandhi 303
scale, grammaticalization 184
selection restriction see restriction
semantics, historical 131
T
tag question 258
tag relative clause 274
teleology 9
text see narrative text
Q
quality 157, 159-165, 173, 176, 183, 184
quasi-auxiliary (verb) 27
question particle 285, 292, 294
question, rhetorical 253-283
quotative 70
360
INDEX OF SUBJECTS
U
unidirectionality 4, 7, 11, 150, 167, 193,
199, 253, 255
universalist position 84
unmarking, pragmatic 278, 279
random 302
typological 120
variation
contextual 136
dialect 281
free 37, 302, 313
random 308
vector verb 27
vehicle, metaphorical 157
verb
active 320, 333
copular 20
effective 28
manipulative 81
mental-state 196, 285, 289
modality 81
prepositional 319-320, 335, 339
serial see serial verb
speech-act 285, 289
stative 320, 329, 305
strong 23
vector see vector
of motion 20, 26-28, 137, 218
of perception 226, 234, 235, 248
of saying 225, 226, 230, 232, 233
verbhood 120
vocabulary
basic 152, 153
core 320
vocative 304
vowel harmony 291, 303, 311-312
V
variability
coding 120
W
word order 12, 18, 37, 169, 170, 192,
255, 256, 265
text 323
textual
function 158, 183, 254, 264, 270,
278
meaning 31
textuality 130, 131, 133, 134, 136, 139,
141
theme 258, 266
token 39, 52, 54, 55, 62, 63, 67, 69
topic 12
topic-comment structure 266, 277
topicalization 331
toponym 223
transfer 161, 168, 171, 178, 325
transfer
metaphorical 7, 161, 164, 190, 213
metonymic 7
transformation, root 296
transition 23
transitive marker 318
transitivizer 306, 308, 313
translation, cross-cultural 84
typology
diachronic 303
language 7

Aproaches or Gramamticalization PDF

Uploaded by

Document Informationclick to expand document information

Document Informationclick to expand document information

Copyright:

Available Formats

Aproaches or Gramamticalization 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

Aproaches or Gramamticalization PDF

Uploaded by

Copyright:

Available Formats

APPROACHES TO GRAMMATICALIZATION I

TYPOLOGICAL STUDIES IN L A N G U A G E (TSL)

Charles Li (Santa Barbara)

Volumes in this series will be functionally and typologically oriented, cove

Elizabeth Closs Traugott and Bernd Heine (eds)

ELIZABETH CLOSS TRAUGOTT

JOHN BENJAMINS PUBLISHING COMPANY

Library of Congress Cataloging-in-Publication Data

Section A: General Method

On some principles of grammaticization

On the gradualness of grammaticalization

Serial verbs and the mental reality of 'event': Grammatical vs.

From cognition to grammar: Evidence from African languages

The semantics-pragmatics of grammaticalization revisited

The de dicto domain in language

The grammaticalization of rhetorical questions in Tamil

Some grammaticalization changes in Estonian and their implications

Section A: Verbal structure

Back to the future

The emergence of perfective aspect in Indo-Aryan languages

Section B: Argument structure

Grammaticization of topic into subject

From V/2 to subject clitics: evidence from Northern Italian

From postposition to subordinator in Newari

The evolution of dependent clause morpho-syntax in Biblical

A quantitative perspective on the gramaticization of epistemic

The grammaticization of the German modal particles

Section E: Spectra of grammaticalization patterns

Areal and universal dimensions of grammatization in Lahu

Ways to go in Rama: a case study in polygrammaticahzation

Grammaticalization and related changes in contemporary German

ELIZABETH CLOSS TRAUGOTT AND BERND HEINE

THE LEXICAL ITEM> MORPHEME MODEL

Early work on grammaticalization viewed it as a diachronic phenomenon.

THE DISCOURSE > MORPHOSYNTAX PERSPECTIVE

fixing of discourse strategies. For example, Givn (1979:209) characterizes

THE GRAMMATICAL CODING PERSPECTIVE ON

So far we have reviewed issues of grammaticalization that are one way

THE RELATION OF DIACHRONIC GRAMMATICALIZATION

For those who see grammaticalization as a diachronic process, the ques

ELIZABETH CLOSS TRAUGOTT AND BERND HEINE

to zero, i.e. increased morphologization, and phonologization. Like other

THE UNIDIRECTIONALITY OF GRAMMATICALIZATION

Firmly entrenched in diachronic perspectives on grammaticalization is

of grammaticahzation in 2. above). Traugott and Knig challenge this concept

ELIZABETH CLOSS TRAUGOTT AND BERND HEINE

The discussion so far has focused on unidirectionality, and what kinds

"lexicalization" is also a counterexample, for example when phonological

MECHANISMS OF LANGUAGE CHANGE RELEVANT TO

A number of mechanisms of language change have already been alluded

CONDITIONS LICENSING GRAMMATICALIZATION

A further set of questions for the diachronic perspective on grammaticali

ELIZABETH CLOSS T R A U G O T T A N D BERND H E I N E

likely to become grammaticaUzed. For example, it seems unlikely that wall

is, for example, a well established channel of grammaticalization leading from

10. SOME QUESTIONS FOR THE FURTHER STUDY OF

ELIZABETH CLOSS TRAUGOTT AND BERND HEINE