The Logic of Language

Language From Within

In this ambitious two-volume work, Pieter Seuren seeks a theoretical unity
that can bridge the chasms of modern linguistics as he sees them, bringing
together the logical, the psychological, and the pragmatic; the empirical and
the theoretical; the formalist and the empiricist; and situating it all in the
context of two and a half millennia of language study.
Volume I: Language in Cognition
Volume II: The Logic of Language
This page intentionally left blank
The Logic of Language

PIETER A. M. SEUREN

1
3
Great Clarendon Street, Oxford OX2 6DP
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Cape Town Dar es Salaam Hong Kong Karachi
Kuala Lumpur Madrid Melbourne Mexico City Nairobi
New Delhi Shanghai Taipei Toronto
With offices in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
# Pieter A. M. Seuren 2010
The moral rights of the author have been asserted
Database right Oxford University Press (maker)
First published 2010
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose the same condition on any acquirer
British Library Cataloguing in Publication Data
Data available
Library of Congress Cataloging in Publication Data
Data available
Typeset by SPI Publisher Services, Pondicherry, India
Printed in Great Britain
on acid-free paper by
CPI Antony Rowe, Chippenham, Wiltshire

ISBN 978–0–19–955948–0

1 3 5 7 9 10 8 6 4 2
to Pim Levelt
for his unfailing support, advice and friendship
This page intentionally left blank
 Contents

Preface xii
Abbreviations and symbols xiv

1 Logic and entailment 1

1.1 What is a logic and why do we need one in the study
of language? 1
1.2 The definition of entailment 7
1.2.1 The general concept of entailment 7
1.2.2 The specific concept of logical entailment 10
1.3 The referential independence of logic: no truth-value gaps 13
1.4 Logical form and L-propositions 16
1.5 The Bivalence Principle, sentence types, and utterance tokens 17
1.6 Some problems with the assignment of truth values 22

2 Logic: a new beginning 27

2.1 Entailment, contrariety and contradiction: the natural triangle 27
2.2 Internal negation and duality: the natural square and the
Boethian square 31
2.3 Logical operators as predicates 37
2.3.1 Meaning postulates 40
2.3.2 Boolean algebra and the operators of
propositional calculus 43
2.3.3 Valuation space modelling: a formal definition 46
2.3.4 Satisfaction conditions of the propositional operators 50
2.3.5 Satisfaction conditions of the quantifiers 52
2.3.5.1 Russellian quantifiers 52
2.3.5.2 Generalized quantifiers 54
2.4 Internal negation, the Conversions and De Morgan’s laws 59
2.4.1 The internal negation again 59
2.4.2 The Conversions and De Morgan’s laws 63

3 Natural set theory and natural logic 67

3.1 Introductory observations 67
3.2 Some set-theoretic principles of natural cognition 71
viii Contents

3.2.1 A résumé of standard set theory 71

3.2.2 The restrictions imposed by NST 74
3.3 Consequences for set-theoretic and (meta)logical relations
and functions 79
3.3.1 Consequences for set-theoretic relations and functions 79
3.3.2 Consequences for (meta)logical relations and functions 84
3.4 The basic-natural systems of logic 88
3.4.1 Basic-natural predicate logic: the necessity of a
cognitive base 89
3.4.2 Hamilton’s predicate logic 103
3.4.3 Basic-natural propositional logic 108
3.5 Neither *nand nor *nall: NST predicts their absence 114
3.5.1 The problem and the solution proposed by pragmaticists 114
3.5.2 Preliminary objections 116
3.5.3 The main objection and a stronger solution 117
3.5.4 Parallel lexical gaps in epistemic-modal and causal logic? 119

4 Logical power, Abelard, and empirical success rates 122

4.1 Aristotelian predicate calculus rescued from undue
existential import 122
4.2 The notion of logical power 127
4.2.1 The logical power of propositional calculus 129
4.2.2 The logical power of Aristotelian-Boethian
predicate calculus 132
4.2.3 The logical power of standard modern predicate calculus 133
4.2.4 The logical power of Aristotelian-Abelardian
predicate calculus 136
4.3 Distributive quantifiers 138
4.4 Predicate logics and intuitions: a scale of empirical success 144

5 Aristotle, the commentators, and Abelard 147

5.1 A recapitulation of ABPC 147
5.2 The not quite Aristotelian roots of ABPC 149
5.2.1 Aristotle’s own predicate logic 149
5.2.2 The ancient commentators 155
5.2.3 The Square representation 156
 Contents ix

5.2.4 An aside on Horn’s and Parsons’ proposal as

regards the O-corner 158
5.2.5 Logic and mysticism: what made logic popular? 170
5.3 Abelard’s remedy 172

6 The functionality of the Square and of BNPC 181

6.1 How to isolate the cases with a null F-class:
the purpose of space 4 181
6.2 Extreme values are uninformative in standard modern
predicate calculus 183
6.3 The functionality of excluding extreme values 184
6.4 The functionality of BNPC 190
6.5 Conclusion 193

7 The context-sensitivity of speech and language 194

7.1 What is context-sensitivity? 194
7.2 Discourse domains 196
7.2.1 The commitment domain and further subdomains 199
7.2.1.1 The notion of subdomain 199
7.2.1.2 Extensional and intensional subdomains 201
7.2.1.3 The epistemic modal subdomains 203
7.2.2 The Principle of Maximal Unity 207
7.2.2.1 Transdominial denotational transparency 207
7.2.2.2 Upward presupposition projection 209
7.2.2.3 Subdomain unification: transdominial consistency 212
7.2.2.4 Minimal D-change 214
7.3 Conditions for text coherence 215
7.3.1 Consistency 215
7.3.2 Informativity 217
7.3.3 Subdomain hierarchies: subsidiary subdomains 219
7.4 Open parameters in lexical meaning 222

8 Discourse incrementation 229

8.1 The incrementation procedure 229
8.1.1 Singular entity addresses and address closure 230
8.1.2 Plurality and quantification 240
8.1.2.1 Plurality and existential quantification 240
8.1.2.2 Discourse-sensitive universal quantification 249
8.1.3 Subordinate subdomains 252
x Contents

8.2 Instructions 254

8.2.1 Conjunction 254
8.2.2 Negation 259
8.2.3 Disjunction 264
8.2.4 Conditionals 270

9 Primary and donkey anaphora 284

9.1 Introduction 284
9.2 Reference by anaphora 288
9.3 Primary anaphora: bound variable or external anaphor? 293
9.4 Donkey sentences 294
9.4.1 The problem 294
9.4.2 The history of the problem 300
9.5 The reference-fixing algorithm 304
9.6 The solution 307
9.6.1 Donkey anaphora under disjunction 307
9.6.2 Donkey anaphora in conditionals 308
9.6.3 Donkey anaphora under universal quantification 309

10 Presupposition and presuppositional logic 311

10.1 Presupposition as an anchoring device 311
10.1.1 Some early history 312
10.1.2 The Russell tradition 317
10.1.3 The Frege-Strawson tradition 321
10.2 The origin and classification of presuppositions 327
10.3 Operational criteria for the detection of presuppositions 331
10.4 Some data that were overlooked 334
10.5 Presupposition projection 342
10.5.1 What is presupposition projection? 342
10.5.2 Projection from lexical subdomains 343
10.5.3 Projection from instructional subdomains 348
10.5.4 Summary of the projection mechanism 351
10.6 The presuppositional logic of the propositional operators 354
10.7 The presuppositional logic of quantification 363
10.7.1 The presuppositional version of the Square and
of SMPC 363
 Contents xi

10.7.2 The presuppositional version of BNPC 368

10.7.3 The victorious Square 370
10.8 The attempt at equating anaphora with presupposition 372

11 Topic–comment modulation 378

11.1 What is topic–comment modulation? 378
11.1.1 The Aristotelian origin of topic–comment modulation 378
11.1.2 The discovery of the problem in the nineteenth century 380
11.1.3 The dynamics of discourse: the question–answer game 386
11.2 Phonological, grammatical, and semantic evidence for TCM 391
11.3 The comment-predicate Bev 395
11.4 Only, even, and Neg-Raising 398
11.5 Why TCM is a semantic phenomenon: the SSV test 406

Bibliography 409
Index 421
 Preface

This is the second and last volume of Language from Within. The first volume
dealt with general methodology in the study of language (which is seen as an
element in and product of human cognition), with the intrinsically inten-
sional ontology that humans operate with when thinking and speaking, with
the socially committing nature of linguistic utterances, with the mechanisms
involved in the production and interpretation of utterances, with the notions
of utterance meaning, sentence meaning, and lexical meaning, and, finally,
with the difficulties encountered when one tries to capture lexical meanings in
definitional terms. The present volume looks more closely at the logic inher-
ent in natural language and at the ways in which utterance interpretation has
to fall back on the context of discourse and on general knowledge. It deals
extensively with the natural semantics of the operators that define human
logic, both in its presumed innate form and in the forms it has taken as a
result of cultural development. And it does so in the context of the history of
logic, as it is assumed that this history mirrors the path followed in Western
culture from ‘primitive’ logical (and mathematical) thinking to the rarified
heights of perfection achieved in these areas of study over the past few
centuries.
The overall and ultimate purpose of the whole work is to lay the founda-
tions for a general theory of language, which integrates language into its
ecological setting of cognition and society, given the physical conditions
of human brain structure and general physiology and the physics of sound
production and perception. This general theory should eventually provide an
overall, maximally motivated, and maximally precise, even formal, interpre-
tative framework for linguistic diversity, thus supporting typological studies
with a more solid theoretical basis. The present work restricts itself
to semantics and, to a lesser extent, also to grammar, which are more directly
dependent on cognition and society, leaving aside phonology, which appears
to find its motivational roots primarily in the physics and the psychology of
sound production and perception, as well as in the input phonological
systems receive from grammar.
The two volumes are not presented as a complete theory but rather as
a prolegomena and, at the same time, as an actual start, in the overall and all-
pervasive perspective of the cognitive and social embedding of language—a
 Preface xiii

perspective that has been hesitantly present in modern language studies but
has not so far been granted the central position it deserves. In this context, it
has proved necessary, first of all, to break open the far too rigid and too
narrow restrictions and dogmas that have dominated the study of language
over the past half-century, which has either put formal completeness above
the constraints imposed by cognition or, by way of contrast, rejected any kind
of formal treatment and has tried to reduce the whole of language to intui-
tion-based folk psychology.
The present, second, volume is, regrettably but unavoidably, much more
technical than the first, owing to the intrinsic formal nature of the topics dealt
with. Avoiding technicalities would have reduced the book either to utter
triviality or to incomprehensibility, but I have done my best to be gentle with
my readers, requiring no more than a basic ability (and willingness) to read
formulaic text and presupposing an elementary knowledge of logic and set
theory.
Again, as in the first volume, I wish to express my gratitude to those who
have helped me along with their encouragement and criticisms. And again, I
must start by mentioning my friend of forty years’ standing Pim Levelt, to
whom I have dedicated both volumes. He made it possible for me to work at
the Max Planck Institute for Psycholinguistics at Nijmegen after my retire-
ment from Nijmegen University and was a constant source of inspiration not
only on account of the thoughts he shared with me but also because of his
moral example. Then I must mention my friend and colleague Dany Jaspers
of the University of Brussels, whose wide knowledge, well-formulated com-
ments, and infectious enthusiasm were a constant source of inspiration.
Ferdinando Cavaliere made many useful suggestions regarding predicate
logic and its history. Finally, I want to thank Kyle Jasmin, whose combined
kindness and computer savviness were indispensable to get the text right. The
many others who have helped me carry on by giving their intellectual, moral,
and personal support are too numerous to be mentioned individually. Yet my
gratitude to them is none the less for that. Some, who will not be named,
inspired me by their fierce opposition, which forced me to be as keen as
they were at finding holes in my armour. I hope I have found and repaired
them all.
P. A. M. S.
Nijmegen, December 2008.
 Abbreviations and symbols

AAPC Aristotelian-Abelardian predicate calculus

ABPC Aristotelian-Boethian predicate calculus (=the Square of Opposition)
BNPC basic-natural predicate calculus
BNST basic-natural set theory
fprop ‘flat’ proposition without TCM
IP incrementation procedure
modprop proposition with TCM
M-partial mutually partial
NPI negative polarity item
NST natural set theory
OSTA Optimization of sense, truth and actuality
PEM Principle of the Excluded Middle
PET Principle of the Excluded Third
PNST principle of natural set theory
PPI positive polarity item
SA semantic analysis
SMPC standard modern predicate calculus
SNPC strict-natural predicate calculus (=ABPC)
SNST strict-natural set theory
SST standard (constructed) set theory
SSV Substitution salva veritate
TCM topic–comment modulation
UEI undue existential import
 Abbreviations and symbols xv

A All F is G ev extreme value (
or OBJ)

I Some F is G iff if and only if
N No F is G LL logical language
A* All F is not-G 8 the universal quantifier
I* Some F is not-G ∃ the existential quantifier
N* No F is not-G N the quantifier NO in BNPC
A! All G is F : standard bivalent negation
I! Some G is F X\Y set-theoretic intersection of X
and Y
P
Q P and Q are equivalent X[Y set-theoretic union of X
and Y
P‘Q P logically entails Q X the complement of set X
in OBJ
P >< Q P and Q are contraries a2X a is an element in the set X
P >< Q P and Q are X3a the set X contains the element
subcontraries a
P#Q P and Q are XY X is properly included in Y
contradictories
U the set of all admissible XY X is included in or equal to Y
situations
OBJ the set of all objects
the null set
VS valuation space [[F]] the extension of predicate F
/P/ the VS of proposition P X [˚ Y X and Y are in full union: X [
in U Y ¼ OBJ
a¼b a is identical with b X OO Y X and Y are mutually
exclusive:
jXj the cardinality of set X X \ Y ¼
; X,Y 6¼
6¼ OBJ
Q >> P Q presupposes P XO
OY X M-partially intersects with
Y: X \ Y 6¼
6¼ X 6¼ Y; X,Y 6¼
OBJ
Q>P Q invites the inference P
æ(a) the reference value of
term a
@ ‘asinus’: donkey pronoun
· basic-natural NEITHER–
NOR
This page intentionally left blank
 1

Logic and entailment

1.1 What is a logic and why do we need one

in the study of language?
The paramount reason why we need logic in the study of language is that logic
is the formal theory of consistency and that consistency is an all-pervasive and
essential semantic aspect of human linguistic interaction. This is true not only
of single sentences but also, and in a much bigger way, of texts and discourses.
And since presuppositions are, if you like, the cement that makes discourses
consistent and since they are induced by the tens of thousands of lexical
predicates in any language, it should be obvious that the logic of presupposi-
tions is a prime necessity for natural language semantics. Yet so as not to
drown the reader in a sea of injudiciously administered complexities, pre-
suppositions (and its logical counterpart, presuppositional trivalent logic) are
kept at bay till Chapter 10. Until then, we stay within the strict limits of
bivalent predicate and propositional logic, though with occasional glances at
multivalence and presuppositions. But the reader will discover that, even
within these limits, there is plenty of room for innovative uncluttering.
A further reason why logic is important for the study of language lies in the
fact that the syntax of the formulae of the various predicate-logic systems
considered is essentially the same as that of the semantic analyses (SAs) that
underlie sentences. And, as was shown in various works belonging to the
tradition of Generative Semantics or Semantic Syntax (e.g. McCawley 1973;
Seuren 1996), the hypothesis that SA-syntax is, in principle, the syntax of
modern logical formulae has proved an exceptional tool for the charting of
striking syntactic generalizations in all natural languages and thus for the
setting up of a general theory of syntax of exceptional explanatory power.
Moreover, a closer investigation of the logic inherent in natural cognition
and natural language will help clarify the hitherto opaque relation between
logic on the one hand and language and cognition on the other. (Ask any
logician what this relation amounts to and you will get a curiously strange
gamut of replies, all of them unsatisfactory.) This is, in itself, surprising
2 The Logic of Language

because language and logic have, from the very Aristotelian beginnings, been
close, though uneasy, bedfellows, never able either to demarcate each other’s
territories or to sort out what unites them. The last century has seen a
tremendous upsurge in both logic and linguistics, but there has not been
a rapprochement worth speaking of. No logic is taught in the vast majority of
linguistics departments or, to my knowledge, in any psychology department,
simply because the relevance of logic for the study of language and mind has
never been made clear.
All in all, therefore, it seems well worth our while to take a fresh look at logic
in the context of the study of language. But, in doing so, we need an open
and flexible mind, because the paradigm of modern logic has come
to suffer from a significant degree of dogmatism, rigidity, and, it has to be
said, intellectual arrogance. Until, say, 1950 it was common for philosophers
and others to play around with logical systems and notations, but this,
perhaps naı̈ve, openness was suppressed by the developments that followed.
The august status conferred upon logic once the period of foundational
research was more or less brought to an end, which was, let us say, around
1950, has not encouraged investigators to deviate from what was, from
then on, considered the norm in logical theory. Yet that norm is based on
mathematics, in particular on standard Boolean set theory, whereas what is
required for a proper understanding of the relations between logic, language,
and thinking is a logic based on natural cognitive and linguistic intuitions.
We are in need of a ‘natural’ logic of language and cognition drawn from
the facts not of mathematics but of language. The first purpose of writing
about logic in this book is, therefore, programmatic: an attempt is made
at loosening up and generalizing the notion of logic and at showing to
linguists, psychologists, semanticists, and pragmaticists why and how logic is
relevant for their enquiries.
An obvious feature of the present book is the attention paid to history. The
history of logic is looked at as much as its present state. This historical
dimension is essential, for at least two reasons. First, there is a general reason,
derived from the fact that the human sciences as well as logic are not
CUMULATIVE the way the natural sciences are taken to be, where new results
simply supersede existing knowledge and insight. In the human sciences and,
as we shall see, also in logic, old insights keep cropping up and new results or
insights all too often prove unacceptably restrictive or even faulty. Since the
human sciences want to emulate the natural sciences, they have adopted the
latter’s convention that all relevant recent literature must be referred to or else
the paper or book is considered lacking in quality. But they have forgotten or
repressed the fact that they are not cumulative: literature and traditions from
 Logic and entailment 3

the more distant past are likely to be as relevant as the most recent literature
and paths that have been followed in recent times may well turn out to be
dead ends so that the steps must be retraced. Recognizing that means recog-
nizing that the history of the subject is indispensable.
The second, more specific, reason is that the history of logic mirrors the
cultural and educational progress that has led Western society from more
‘primitive’ ways of thinking to the unrivalled heights of formal precision
achieved in modern logic and mathematics. This is important because, as
is explained in Chapter 3, it seems that natural logical intuitions have only
gone along so far in this development and have, at a given moment, detached
themselves from the professional mathematical logicians, leaving them
to their own devices. It is surmised in Chapters 3 and 4 that natural logical
intuitions are a mixture of pristine ‘primitive’ intuitions and more sophisti-
cated intuitions integrated into our thinking and our culture since the
Aristotelian beginnings. It is this divide between what has been culturally
integrated and what has been left to the closed chambers of mathematicians
and logicians that has motivated the distinction, made in Chapter 3, between
‘natural’ logical intuitions on the one hand and ‘constructed’, no longer
natural, notions in logic and mathematics on the other.
Historical insight makes us see that linguistic studies have, from the very
start, been divided into two currents, FORMALISM and ECOLOGISM (see, for
example, Seuren 1986a, 1998a: 23–7, 405–10). In present-day semantic studies,
the formalists are represented by formal model-theoretic semantics, while
modern ecologism is dominated by pragmatics. It hardly needs arguing that,
on the one hand, formal semantics, based as it is on standard modern logic,
badly fails to do justice to linguistic reality. Pragmatics, on the other hand,
suffers from the same defect, though for the opposite reason. While formal
semantics exaggerates formalisms and lacks the patience to delay formaliza-
tion till more is known, pragmatics shies away from formal theories and lives
by appeals to intuition. Either way, it seems to me, the actual facts of language
remain unexplained. If this is so, there must be room for a more formal
variety of ecologism, which is precisely what is proposed in the present book.
One condition for achieving such a purpose is the loosening up of logic.
It may seem that logic is a great deal simpler and more straightforward
than human language, being strictly formal by definition and so much more
restricted in scope and coverage, and so much farther removed from the
intricate and often confused realities of daily life that language has to cope
with. Yet logic has its own fascinating depth and beauty, not only when
studied from a strictly mathematical perspective but also, and perhaps even
more so, when seen in the context of human language and cognition. In that
4 The Logic of Language

context, the serene purity created by the mathematics of logic is drawn into
the realm of the complexities of the human mind and the mundane needs
served by human language. But before we embark upon an investigation of
the complexities and the mundane needs, we will look at logic in the pure
light of analytical necessity.
What is meant here by logic, or a logic, does not differ essentially from the
current standard notion, shaped to a large extent by the formal and founda-
tional progress made during the twentieth century. As far as it goes, the
modern notion is clear and unambiguous, but it still lacks clarity with regard
to its semantic basis. In the present chapter the semantic basis is looked at
more closely, in connection with the notion of entailment as analytically
necessary inference—that is, inference based on meanings. This is not in itself
controversial, as few logicians nowadays will deny that logic is based on
analytical necessity, but the full consequences of that fact have not been
drawn (probably owing to the deep semantic neurosis that afflicted the
twentieth century).
During the first half of the twentieth century, most logicians defended
the view that logical derivations should be defined merely on grounds of
the agreed FORMS of the L-propositions or logical formulae,1 consisting of
logical constants and typed variables in given syntactic structures. The deri-
vation of entailments was thus reduced to a formal operation on strings
of symbols, disregarding any semantic criterion. Soon, however, the view
prevailed that the operations on logical form should be seen as driven by
the semantic properties of the logical constants. I concur with this latter view,
mainly because there is nothing analytically necessary in form, but there is
in meaning. This position is supported by the fact that a meaning that is well-
defined for the purpose of logic is itself a formal object, in the sense that it is
representable as a structured object open to a formal interpretation in terms
of a formal calculus such as logical computation.
In earlier centuries, the ideas of what constitutes logic have varied a great
deal. In medieval scholastic philosophy, for example, a distinction was made
between logica maior, or the philosophical critique of knowledge, and logica

1 The notion of L-proposition is defined in Section 3.1.4 of Volume I as ‘a type-level semantically

explicit L-structure, which is transformed by the grammar module into a corresponding type-level
surface structure, which can, in the end, be realized as a token utterance’. L-propositions form the
language of SEMANTIC ANALYSIS (SA), whose expressions (L-propositions) equal logical formulae in some
variety of predicate logic. It is important to note that L-propositions are type-level elements, whereas
propositions are token-level mental occurrences. L-propositions are part of the linguistic machinery
that turns propositional token occurrences into sentence-types of a given lexically and grammatically
defined language system. See also Section 1.4.
 Logic and entailment 5

minor, also called dialectica, which was the critical study and use of the logical
apparatus of the day—that is, Aristotelian-Boethian predicate calculus and
syllogistic. Logica maior is no longer reckoned to be part of logic but, rather,
of general or ‘first’ philosophy. Logica minor corresponds more closely to the
modern notion of logic. During the nineteenth century logic was considered
to be the study of the principles of correct reasoning, as opposed to the
processes actually involved in (good or bad) thinking, which were assigned
to the discipline of psychology. The Oxford philosopher Thomas Fowler, for
example, wrote (1892: 2–6):
The more detailed consideration of [. . .] Thoughts or the results of Thinking
becomes the subject of a science with a distinct name, Logic, which is thus a
subordinate branch of the wider science, Psychology. [. . .] It is the province of
Logic to distinguish correct from incorrect thoughts. [. . .] Logic may therefore be
defined as the science of the conditions on which correct thoughts depend, and the art
of attaining to correct and avoiding incorrect thoughts. [. . .] Logic is concerned with
the products or results rather than with the process of thought, i.e. with thoughts rather
than with thinking.

Similar statements are found in virtually all logic textbooks of that period.
After 1900, however, changes are beginning to occur, slowly at first but
then, especially after the 1920s, much faster, until the nineteenth-century
view of logic fades away entirely during the 1960s, with Copi (1961) as one
rare late representative.
But what do we, following the twentieth-century tradition in this respect,
take logic to be? Since about 1900, logic has increasingly been seen as the study
of consistency through a formal calculus for the derivation of entailments. In this
view, which we adopt in principle, logic amounts to the study of how to derive
L-propositions from other L-propositions salva veritate—that is, preserving
truth. Such derivations must be purely formal and independent of intuition.
According to some logicians, they are based exclusively on the structural
properties of the expressions in the logical language adopted, but others,
perhaps the majority, defend the view that the semantic properties of certain
designated expressions, the LOGICAL CONSTANTS, co-determine logical deriva-
tions, provided these meanings are formally well-defined, which means in
practice that they must be reducible to the operators of Boolean algebra (see
Section 2.3.2 for a precise account). On either view, logic must be a CALCULUS—
that is, a set of formally well-defined operations on strings of terms, driven
only by the well-defined structural properties of the expressions in the logical
language and the well-defined semantic properties of the logical constants.
6 The Logic of Language

When one accepts the dependency on the meanings of the logical constants
involved, one may say that logic is an exercise in analytical necessity.
This basic adherence to the twentieth-century notion of what constitutes
a logic is motivated not only by the fact that it is clear and well-defined but also
by the consideration that it allows us to re-inspect the ‘peasant roots’ of logic,
as found in the works of Aristotle and his ancient successors, from a novel
point of view. Traditional logicians only had natural intuitions of necessary
consequence and consistency to fall back on for the construction of their
logical systems, lacking as they did the sophisticated framework of modern
mathematical set theory. Yet this less sophisticated source of logical inspiration
is precisely what we need for our enterprise, which aims at uncovering the
logic people use in their daily dealings and their ordinary use of natural
language. Pace Russell, we thus revert unashamedly to psychological logic.
Though Aristotle, the originator of logic, did not yet use the term logic, his
writings, in particular On Interpretation and Prior Analytics, show that his
starting point was the discovery that often two sentences are inconsistent
with regard to each other in the sense that they cannot be true simultaneously.
He coined the term CONTRARIES (enántiai) for such pairs (or sets) of sentences.
When two sentences are contraries, the truth of the one entails the falsity of
the other. He then worked out a logical system on the basis of contrariety and
contradictoriness—and thus also of entailment—as systematic consequences
of certain logical constants.
Of course, the question arises of what motivates the particular selection of
the logical constants involved and of the operations they allow for, given their
semantic definition. A good answer is that the choice of the relevant constants
and of the operations on the expressions in which they occur is guided by the
intuitive criterion of consistency of what is said on various occasions. Such
consistency is of prime importance in linguistic interaction, since, as is argued
in Chapter 4 of Volume I, speakers, when asserting a proposition, put
themselves on the line with regard to the truth of what they assert. Inconsis-
tency will thus make their commitment ineffective. When a set of predicates is
seen to allow for a formal calculus of consistency, we have hit on a logical
system, anchored in the syntax of the logical language employed and in the
semantic definitions of the logical constants, whose meanings are specified in
each language’s lexicon. That being so, a not unimportant part of the seman-
ticist’s, more precisely the lexicographer’s, brief consists in finding out how
and to what extent natural language achieves informational consistency
through its logical constants.
Consistency is directly dependent on truth and the preservation of
truth through chains of entailments, also called logical derivations. The
 Logic and entailment 7

operations licensed by the logical constants must ensure that L-propositions,

when interpreted as being true in relation to given states of affairs, yield
L-propositions that are likewise true under the same interpretation. When
they do that, it is said that the logical derivation is VALID. The validity of logical
derivations should depend solely on the MEANING—that is, the SATISFACTION
CONDITIONS—of the logical constants involved and by their syntactic position.
This ensures that the validity of a sound logic is based on analytical necessity.
It does not mean, however, that there can be only one valid logic, a miscon-
ception often found among interested laymen and even among professional
logicians. In principle, there is an infinite array of possible logics, each defined
by the choice of the logical constants and the meanings and syntax defined for
them. But once the constants and their meanings have been fixed, logical
derivations are analytically necessary.
It is now obvious that logic must be closely related to natural language,
since the most obvious class of expressions carrying the property of truth or
falsity are the assertive utterances made by speakers or writers in some natural
language. Of course, one can try and make an artificial language whose
expressions are bearers of truth values, but one way or another such expres-
sions are all calques, sometimes idealized or streamlined, of natural language
expressions.

1.2 The definition of entailment

1.2.1 The general concept of entailment
At this point we need to specify more precisely what is meant by ENTAILMENT.
We begin by giving a definition of entailment in general:
ENTAILMENT
When an L-proposition (or set of L-propositions) P ENTAILS an
L-proposition Q (in formal notation P  Q), then, whenever P is true,
Q must of necessity also be true, on account of the specific linguistic
meaning of P—that is, for analytical reasons.
For example, the L-proposition underlying the sentence Jack has been mur-
dered entails the L-proposition underlying the sentence Jack is dead because it
is in the meaning of the predicate have been murdered that whoever has been
murdered must of necessity be dead.
The entailment relation is, however, subject to an essential proviso: both
the entailing and the entailed (L-)proposition must be identically KEYED to a
chunk of spatio-temporal reality. Definite terms must refer to the same
8 The Logic of Language

objects and the tenses used must have identical or corresponding temporal
values. Thus, in the example given, the proper name Jack must refer to the
same person and the present tense must refer to the same time slice in both
statements. This is the MODULO-KEY CONDITION on the entailment relation. This
condition may seem trivial and is, in most cases, silently understood. In fact,
however, it is far from trivial. It is defined as follows:
THE MODULO-KEY CONDITION
Whenever a (type-level) L-proposition or set of L-propositions P entails
a (type-level) L-proposition Q, the condition holds that all coordinates in
the underlying propositions p and q that link up p and q with elements in
the world take identical or corresponding keying values in the
interpretation of any token occurrences of P and Q, respectively.
The Modulo-Key Condition, however, does not allow one to say that if the
terms Jack and Dr. Smith refer to the same person, (the L-proposition
underlying) the statement Jack has been murdered entails (the L-proposition
underlying) the statement Dr. Smith is dead, and analogously with the names
interchanged. This is so because entailment is a type-level relation and at type-
level it is not given that Jack is the same person as Dr. Smith. To have the
entailment it is, therefore, necessary to insert the intermediate sentence Jack is
Dr. Smith. All the Modulo-Key Condition does is ensure that the term Jack is
keyed to the same person every time it is used.
The Modulo-Key Condition implies a cognitive claim, since keying is the
cognitive function of being intentionally focused on specific objects in the
world in a specific state of affairs. This cognitive claim involves at least the
existence of a system of coordinates for the mental representation of states of
affairs. Whenever the sentences in question are used ‘seriously’, and not as
part of a fictional text presented as such, these coordinates have values that are
located in the actual world. For the entailment relation to be applicable, and
indeed for the construction of any coherent discourse, the participants in the
discourse must share a system of coordinates needed for a well-determined
common intentional focusing on the same objects and the same time. The
mechanism needed for a proper functioning of such a mental system of
coordinates and their values is still largely unknown. We do know, however,
that it is an integral part of and a prerequisite for an overall system of
discourse construction, both in production and in comprehension—the
system that we call anchoring. Since most of this system is still opaque, we
are forced to conclude that what presented itself as a trivial condition of
constancy of keying for entailment relations turns out to open up a vast
 Logic and entailment 9

area of new research, a terra incognita for the study of language, meaning,
and logic.
In addition to entailment, there is also EQUIVALENCE, normally defined as
entailment in both directions: ‘P is equivalent with Q’ is said to mean that
P  Q and Q  P. This will do no harm for the moment, but in Chapter 3 it is
argued that it is probably not a good way of making explicit what (semantic)
equivalence amounts to in natural language and cognition. In natural language
and cognition, equivalence is not so much a (meta)relation, yielding truth or
falsity when applied to any n-tuple of L-propositions, as a cognitive operation
taking two or more L-propositions and turning them into one at a certain level
of representation. As a relation, equivalence makes little cognitive sense, since
when two L-propositions are equivalent at some level of cognitive representa-
tion, they count as one, not as two. As an operation, however, equivalence
makes a great deal of cognitive sense, since what counts as two or more at
some level of representation can be made to count as one at a different level. In
this sense one can say that Jack sold a car to Jim is equivalent with Jim bought a
car from Jack (modulo key). To say that these two sentences are equivalent
then amounts to saying that they are turned into one, or are identified, salva
veritate, at some level, but not necessarily at all levels, of representation.
In the definition of entailment given above we have inserted the condition
‘on account of the specific linguistic meaning of P’. This is, in itself, not
controversial, but the wording implies that necessarily true L-propositions
cannot properly be said to be entailed by any arbitrary L-proposition (the
medieval inference rule ‘verum per se ex quolibet’), and likewise that neces-
sarily false L-propositions cannot properly be said to entail any arbitrary
L-proposition (‘ex falso per se ad quodlibet’). These theorems may be said
to hold in a strictly mathematical sense, yet they fail to satisfy the definition
given, since no specific semantic properties of the entailing L-proposition are
involved. We also consider them to be irrelevant for a proper understanding
of natural language. The entailments that are relevant are subject to the
condition that they derive from the lexically defined meanings of the predi-
cates occurring in the entailing sentence, as it is predicates that produce truth or
falsity when applied to objects of the proper kind, due to their satisfaction
conditions—that is, the conditions that must be satisfied by any object to
‘deserve’ the predicate in question. Since more specific conditions imply less
specific conditions (for example, the condition of being a rose implies the
condition of being a flower), the satisfaction of a more specific predicate by
certain objects implies the satisfaction by the same objects of a predicate
defined by less specific conditions. This is the basis of the entailment relation
we wish to consider. It means that, as long as the objects and the state of affairs
10 The Logic of Language

involved remain the same, the predicates can do their entailment work. We
thus require of the relation of NATURAL ENTAILMENT from P to Q that it be
subject to the condition that the preservation of truth rests on the meaning of the
predicates in the entailing sentence P and on their structural position in P.
Henceforth, unless otherwise specified, when we speak of entailment, what
is intended is natural entailment.

1.2.2 The specific concept of logical entailment

A few of Plato’s students, in particular Aristotle, discovered that some entail-
ments can be formally computed on the basis of certain specific elements
(words) in statements, the so-called LOGICAL CONSTANTS or LOGICAL OPERATORS,
known to medieval philosophers as SYNCATEGOREMATA (Moody 1953: 16–17).
Statements, or, more precisely, the type-level L-propositions underlying them,
allow for a distinction to be made between, on the one hand, logical constants
and, on the other, the nonlogical remainder, which are rendered in logical
analyses by means of symbols called ‘lexical variables’.
Traditionally, the set of logical constants is extremely limited. It consists
of the words representing the notions of ALL and its relatives (the
universal quantifier), SOME and its relatives (the existential quantifier), NOT
(negation), AND (conjunction), OR (disjunction) and, if one wishes, also IF . . .
THEN (material implication). The quantifiers ALL and SOME do service in
PREDICATE CALCULUS, where the lexical variables involved range over predi-
cates. The remaining operators, NOT, AND, and OR (and normally also IF . . . THEN),
serve in PROPOSITIONAL CALCULUS, where the lexical variables involved range
over L-propositions.2 Since propositional calculus can be incorporated
into predicate calculus, the propositional operators can also be put to use in
predicate calculus. The propositional operator NOT, in particular, is indispens-
able not only in propositional calculus but also in any interesting form
of predicate calculus.
Further ‘fields’ of predicate meanings have been discovered, and are per-
haps still to be discovered, that allow for the development of a logical system
in addition to the predicate and propositional logics we already have. Modal
logic, for example, which is as old as predicate logic, has, in addition, the
constants NECESSARY and POSSIBLE, and, in modern times, verbal tenses are
likewise used as logical constants in tense logics. But the thesis we defend in
the present book does not require that we be concerned with logical systems
2 Since Lukasiewicz (1934) it has been known that the logical constants discovered by Aristotle are
ALL, SOME, and NOT. Aristotle's syllogistic shows that he had some awareness of the logical force of AND,
but he never exploited that. Propositional calculus, based on the logical constants NOT, AND, OR, and IF…
THEN is a product of the Stoic philosophers.
 Logic and entailment 11

beyond propositional and predicate logic, though there will be occasional

references to modal logic.
Those entailments that are formally computable with the help of either
predicate calculus or propositional calculus we call LOGICAL ENTAILMENTS. They
form a subcategory of the general category of entailments. Semantically
necessary truths thus comprise logically necessary truths as a subclass.
Logical entailments are computable on account of the formally defined
meanings of the logical constants that occur in sentences. Thus, when we
speak of logical entailment, we mean a necessary consequence resulting from
a calculus built on the meanings of the logical constants in the entailing
sentence:
LOGICAL ENTAILMENT
When an L-proposition P LOGICALLY ENTAILS an L-proposition Q (formal-
ly P ‘ Q), then, whenever P is true, Q must of necessity also be true, on
account of the meaning(s) of the logical constant(s) in P. Logical entail-
ments are by definition formally computable in terms of a logical system.
LOGICAL EQUIVALENCE, moreover, is defined as logical entailment in both direc-
tions: ‘P is logically equivalent with Q’ (or P
Q) means that P ‘ Q
and Q ‘ P.
Logic is thus a formal calculus for the derivation of entailments. A logic is
SOUND when it admits only entailments that are consistent with the entailing
L-propositions. We shall see that natural logic adds the condition that the
entailing L-proposition must itself have the possibility of truth and that the
entailed L-proposition must itself have the possibility of falsity.
All textbooks on logic define the entailment relation without the analyticity
condition—that is, without explicitly stipulating that the entailment must be
due to the MEANING(S) of the logical constant(s) in the entailing L-proposition.
They merely require that when P ‘ Q, truth of P NECESSARILY requires truth
of Q, where necessity is defined negatively as independence of any possible
contingent situation. In the actual practice of modern logic, logical entailments
are taken to follow from the logical system in the logical language used. And
since this logical language consists of logical constants and variables, it is
ultimately the constants that define the entailments. But this is not quite the
same as saying, as we do, that the entailments are due to the MEANING(S) of the
logical constant(s) in the entailing L-proposition. The difference becomes
visible when one considers ‘nonnatural’ or ‘constructed’ theorems or inference
rules, such as the theorem that a necessarily false L-proposition entails any
other L-proposition (the old rule ‘ex falso per se ad quodlibet’ mentioned
earlier), or that a necessarily true L-proposition is entailed by any arbitrary
12 The Logic of Language

L-proposition (‘verum per se ex quolibet’), or the theorem (inference rule),

usually called ‘addition’, which says that, for arbitrary L-propositions P and Q,
P ‘ (P OR Q). Such ‘entailments’ follow from the logical system used,
but they are not based on the meanings of the logical constants in the entailing
L-propositions and are, therefore, not considered to be entailments in the
present context. Although such theorems are mathematically valid, they are
considered counterproductive, even ‘pathological’ (in a nondramatic sense), in
the logic of language.
The formal calculus of logical entailments is made possible on account of
the fact that the meaning definitions of the logical constants contain as central
elements conditions which are defined in terms that allow for formal compu-
tation. Therefore, the assumption that sound logical reasoning in natural
language is possible implies the assumption that natural language contains
expressions that reflect logical constants with meanings admitting of formal
computation, which together form a consistent computational system allow-
ing for the derivation of logically necessary consequences. In this perspective,
it is an obvious thought that it may be worth our while to see if the
logical constants can be legitimately treated as predicates in the logical
language of L-propositions. In Section 2.3 it is shown that the logical con-
stants are indeed naturally treated as (abstract) predicates.
Since there are many different possible consistent systems allowing for the
formal, computational derivation of logical entailments, each system being
defined by the choice and the meanings of the logical constants, the empirical
question arises as to exactly which logic underlies logical reasoning in lan-
guage and, by extension, in thinking. It is often taken for granted, perhaps
more by outsiders than by professional logicians, that the bivalent predicate
and propositional logic which has been considered standard throughout the
twentieth century is the only viable or reasonable logic to operate with. This
view, however, is misguided. Standard modern logic may provide a suitable
metalanguage for the specification of mathematical truths and entailments,
but that does not automatically make it a suitable model for the logic of
cognition and of natural language.
There is, of course, an extensive literature aiming at definitions for, and
logico-mathematical specifications of, the various quantifiers found in natu-
ral language, including and beyond all and some. Yet this literature somehow
seems to miss the point of the present study, which is the logical system
of cognition and of the object language, the system that safeguards consisten-
cy through discourse. Standard logic may provide a suitable descriptive
metalanguage for the meanings of natural language logical operators (as for
well-nigh anything else), but it does not provide a model for the logic of
 Logic and entailment 13

language and cognition. The question of what is the, or a, proper model for
the logic of language has so far received little or no attention. It is this
question that is central to the present study.

1.3 The referential independence of logic: no truth-value gaps

Since logical entailments are necessary consequences which derive their ne-
cessity from the meanings of the logical constants involved, logic is by
definition predicated on the notion of analytical necessity and therefore by
definition independent of any contingent state of affairs. The logical machin-
ery must preserve truth through entailment relations regardless of whatever
specific state of affairs it is applied to, provided the Modulo-Key Condition is
observed. This independence of specific states of affairs we call the referential
independence of logic. It is essential if logic is to be a calculus of entailments.
The referential independence of logic does not preclude the existence of
SPECIFIC LOGICS which look as if they specify entailment relations only for
certain states of affairs, or, as is often said nowadays, for certain knowledge
states. Such logics can be very useful in practice and, caught under the name
of nonmonotonic logics, they abound in the reality of human life. For
example, given the fact that prisoners of war are protected by the Geneva
Convention, one may say that (the L-proposition underlying) a statement like
(1.1a) ‘entails’ (the L-proposition underlying) the statement (1.1b):
(1.1) a. All enemy troops have been taken prisoner.
b. All enemy troops are protected by the Geneva Convention.
But this means no more than that the condition ‘All prisoners of war are
protected by the Geneva Convention’ is silently understood and incorporated
into the entailment relation, so that in reality it is not (1.1a) but (1.2) that
entails (1.1b):
(1.2) All enemy troops have been taken prisoner and all prisoners of war are
protected by the Geneva Convention.
It stands to reason that specific logics are richer than general unrestricted
logics, in the sense that they allow for more conclusions to be drawn. This is
part of their usefulness in ordinary life, which again is why practitioners of
artificial intelligence set great store by developing all kinds of ‘nonmonotonic’
logics.
Yet the notion of ‘specific logic’ is easily misunderstood in that it is believed
that such logics are valid only in certain states of affairs and are thus usable
within the confines of specific restricted knowledge states. In other words,
14 The Logic of Language

they presuppose that certain conditions are fulfilled in the states of affairs the
logic is to be applied to. That having been said, the feeling is that all is well. In
fact, however, such specific logics must be caught under the umbrella of
some sound universally applicable logic, or else there is no specific logic at
all. In the case of examples (1.1) and (1.2), the ‘umbrella’ is completed by the
addition of a silently understood contingent condition, namely that prisoners
of war are protected by the Geneva Convention. In this book, we are not
concerned with ‘specific’ or ‘nonmonotonic’ logics. What we are concerned
with is the more basic, though technologically less challenging, question of
the meanings of the logical constants concerned in the overall, universally
applicable, ‘umbrella’ logic.
Consider the well-known example of traditional Aristotelian-Boethian
predicate calculus. This logic is sound only for states of affairs where the
class of things quantified over is nonnull: it has so-called ‘existential import’,
which makes it nonvalid as a logical system. It is widely believed that this logic
is saved from its undue existential import by the mere stipulation that it
PRESUPPOSES that the class of things quantified over (the F-class) is nonnull.
Once that condition has been stated, so it is thought, the logic is safe. But this
cannot be correct. For either Aristotelian-Boethian predicate calculus is to be
considered a specific logic, in which case it is in need of a general ‘umbrella’
logic, or it is meant to be a general logic, in which case it must specify what
entailments are valid in any arbitrary state of affairs, no matter whether the
class of things quantified over does or does not contain any elements.
Strawson failed to see this when he proposed (1952: 170–6) that (the
L-proposition underlying) a statement like (1.3) lacks a truth value (falls
into a ‘truth-value gap’) because there is in fact not a single Londoner alive
of that age:
(1.3) All 150-year-old Londoners are bald.
If Strawson were right, it would follow that (1.3), which otherwise has
impeccable papers for serving in Aristotelian-Boethian predicate calculus,
falls outside that calculus when used here and now. But as soon as one
Londoner were indeed to reach the age of 150 years, (1.3) suddenly would
have a truth value and would take part in Aristotelian-Boethian predicate
calculus. And then, all of a sudden, it would entail that at least one 150-year-
old Londoner is bald—an entailment defined as valid in that logic. It is easily
seen that this is in stark conflict with the concept of entailment as used in
logic, since it makes entailments dependent on contingencies, on what
 Logic and entailment 15

happens to be or not to be the case, which is precisely what is excluded by the

concept of entailment.
Some authors say that (1.3), as used here and now, is truth-valueless but still
takes part in a logical calculus. Formally, this amounts to saying that the
universal quantifier ALL is a partial, not a total, function, which refuses as
input the otherwise grammatically well-formed expression denoting the
F-class when this F-class contains no members. But either this makes the
logic dependent on contingent conditions, which is inadmissible, or it surrep-
titiously uses the term ‘truth-valueless’ as the name of a truth value and not as a
qualification applied to grammatically well-formed but contingently refused
inputs. It seems that dependence on contingencies is properly avoided by the
introduction of a third truth value in the sense that what is falsity in standard
bivalent systems is split up into minimal and radical falsity. This point of view
is defended and amply discussed and elaborated in Chapter 10.
A similar predicament arises with sentences that contain a definite term
that is either unkeyed (and thus fails to refer), such as The boy laughed, said
without any context, or is well-keyed but fails to refer to an actually existing
entity though the main predicate requires it to do so, such as The present king
of France is bald. Here again, the predicates laugh and bald (and the vast
majority of predicates in any natural language) can be treated as partial
functions refusing unkeyed or not actually existing inputs and thus yielding
the ‘value’ undefined. Yet (radical) negation yields truth in The present king of
France is NOT bald (which makes The present king of France is bald radically
false), whereas the unkeyed sentence The boy did NOT laugh is as devoid
of a truth value as its positive counterpart The boy laughed. It seems prefera-
ble, therefore, to avoid the term and the concept ‘undefined’ altogether
and distinguish, as we do, between the lack of a truth value when a definite
term is unkeyed and radical falsity when a definite term is well-keyed but
fails to refer to an actually existing object whereas its predicate requires it
to do so.
Cases like those exemplified in (1.3) are of great interest to the logic of
language. For Frege and Strawson, the truth value of (1.3) is excluded from the
calculus as (1.3) is taken to lack a truth value, a position which, as we have
seen, is not preferable. In standard modern predicate calculus, (1.3) is consid-
ered true. By contrast, as is shown in Chapters 4 and 5, the twelfth-century
French philosopher Abelard considered (1.3) false in those circumstances, and
the position taken by Aristotle himself implies the same. And to end up with
ourselves, we agree with Abelard (and Aristotle) that (1.3) is false, but we
assign it a special, marked kind of radical falsity, truth-conditionally distinct
from unmarked or minimal falsity. Opinions galore, all of them derivable
16 The Logic of Language

from a semantic definition (either of the quantifier ALL or of the main

predicate of the sentence). This diversity of opinions makes it all the more
mandatory that a decision be reached as regards the real meanings of the
logical constants in natural language.
Meanwhile we see that the very fact that an L-proposition about a given
state of affairs contains a logical constant makes it fall under the logic that
deals with the constant in question. We also see that any L-proposition P
about a given state of affairs must have a truth value: it is either true, false, or
whatever other value has been introduced, no matter whether the objects
referred to are actual or virtual objects. The truth value it has depends on
(a) the meaning of P and the expressions in it and (b) the state of affairs that
P and its referring expressions are about.

1.4 Logical form and L-propositions

It is commonly, though not universally, accepted that if entailments are to be
formally computed it is necessary to reduce the expressions at issue to a
‘regimented’ form, usually called LOGICAL FORM or SEMANTIC ANALYSIS, which is
distinct from the normal or SURFACE FORM in which they occur in actual speech.
Since it is likewise commonly accepted that many, if not most, natural
language sentences contain one or more logical constants, there is, for every
natural language L as a whole, a programme of reduction to the logical
language LL. What we call an L-proposition is the translation of (the propo-
sitional part of) a sentence into any given LL. Since a proper LL is well-
defined, whereas the ‘language’ of pure mental propositions has so far had to
do without any precise definition, we have no choice but to conduct all logical
computations in terms of L-propositions.
Given the distinction between surface structure and L-propositional form
(semantic analysis), some formal procedure for relating the two must be made
available. Until the 1960s, and often still today, the reduction of surface form to
logical or L-propositional form and vice versa was mostly done intuitively, by
rule of thumb. It was not until the 1960s, when formal semantics and formal
linguistics came into being, that this problem was tackled in a systematic, but
far from uniform, way. (For some, including the present author, the
programme of formulating a mapping relation between surface structure
and logical form constitutes the GRAMMAR of L, for others it is part of the
SEMANTICS of L.) We will, however, not now enter into the arena of logical form
reduction. For now, we simply assume that each sentence of a natural language
has a double representation, one at the level of SEMANTIC ANALYSIS (SA) (contain-
ing its logical or L-propositional form) and one at the level of SURFACE FORM.
 Logic and entailment 17

If that is so, there must be a regular mapping system relating the two levels of
representation. And it is up to any one individual whether he or she prefers to
call that mapping system the grammar of L or part of its semantics.
It is customary to say that logical forms or L-propositions are either ATOMIC
or COMPLEX. An atomic L-proposition is seen as consisting of a predicate
F expressing a property and one or more terms used to denote the objects to
which the property expressed by F is attributed. A complex L-proposition
contains at least one propositional operator. We adhere to this distinction,
although it should be understood that it is of a purely logical and not of a
linguistic or grammatical nature. A sentence like:
(1.4) Despite the fact that it had been snowing heavily the whole day, she
decided to drive to the factory, hoping that she would find the answer
there.
is, of course, grammatically complex. Yet it is considered logically atomic, as
it contains no propositional operator. It is up to linguistic analysis to show
that sentence (1.4) is structured in such a way that indeed a property is
assigned to one or more objects. This can only be shown if it is assumed
that some objects are of a kind that allows for linguistic expression by means
of an embedded sentence or S-structure that functions as a term to a predicate
at L-propositional level, so that recursive embeddings of S-structures are
allowed. Embedded S-structures must then be considered to refer to abstract
objects of some kind (see Section 6.2.3 in Volume I). Seuren (1996) gives an
idea of how the grammatical analysis of sentences shows up a hierarchical
predicate–argument structure. The development of a semantics to go with
this type of grammatical analysis is part of a comprehensive research
programme leading to an integrated theory of language.
So far, we have seen that a logic is a calculus of entailments, and that
an entailment of an L-proposition P—which may be a set of L-propositions
conjoined under AND—specifies on analytical, semantic grounds what L-
propositions Q, R, S, . . . (apart from P itself) must likewise be true (modulo
key) when an assertive token utterance of P is true.

1.5 The Bivalence Principle, sentence types,

and utterance tokens
It will be clear that a logic must be geared to the system of truth value
assignments adopted for the (natural or artificial) language in question.
Logic has traditionally followed Aristotle in adopting a strictly bivalent system
18 The Logic of Language

of truth-value assignments, with just two truth values, True and False, in
which all L-propositions with a given key always have a truth value. It was not
until the 1920s that variations on this theme began being proposed, in
particular by the Polish logician Lukasiewicz, but by many others as well
(see Rescher 1969, ch. 1). These variations on the theme of multivalence were
not, on the whole, supported by linguistic intuitions. On the contrary, they
were motivated by a variety of considerations, covering modal logic, future
contingency, mathematical intuitionism, undecidable mathematical state-
ments, and logical paradoxes. It was not until after the 1950s that the notion
of trivalent logic was mooted in connection with natural language, in partic-
ular with presuppositional and vagueness phenomena. Given the great variety
in the motivations for multivalued logics, it is understandable that a certain
amount of confusion ensued, which in turn led to a situation where investiga-
tions into multivalued logics did not achieve a high degree of respectability. In
fact, logicians have, on the whole, been anxious to safeguard logic from any
incursions of multivalence.
Since we, too, are threatening the bivalent shelter of standard logic, it is
important to state as exactly as possible what is meant by the PRINCIPLE OF
BIVALENCE. We define the Bivalence Principle as consisting of two independent
subprinciples:
PRINCIPLE OF BIVALENCE
(i) SUBPRINCIPLE OF COMPLETE VALUATION OF L-PROPOSITIONS:
All well-anchored and well-keyed L-propositions have a truth value.

(ii) SUBPRINCIPLE OF BINARITY:

There are exactly two truth values, True and False; there are no values
between, and no values outside, True and False.

The subprinciple of binarity comprises the (SUBSUB)PRINCIPLE OF THE EXCLUDED

MIDDLE (PEM), which says only that there are no values between True and
False, and says nothing about possible values beyond or outside simply True
and False. The overall principle of bivalence is often confused in the literature
with PEM, mainly because Aristotle wrote only about, or rather against,
possible truth values between true and false, as he wanted to convince
his readers that the Sophists, with what he saw as their wishy-washy relativis-
tic notion of truth, were hopelessly wrong because truth and falsity are
absolute, nongradable opposites. See, for example, Metaphysics, end of
book IV, 1010–12, in particular his statement:
There cannot be an intermediate between contradictories, but of one object we must
either affirm or deny any one predicate.
 Logic and entailment 19

To Aristotle, what for him were equivocations and prevarications produced by

the Sophists on the notions of truth and falsity were more than he could bear.
But he never wrote about, for example, different kinds of falsity, as that
question did not weigh upon his mind.
The other subprinciple, that of COMPLETE VALUATION of L-propositions, holds
trivially for well-anchored and well-keyed L-propositions as defined above.
Any type-level L-proposition that expresses the mental assignment of a
property to one or more objects stands by definition in a relation of confor-
mity or nonconformity with regard to reality. When it fully conforms to
reality as construed by the human intellect, it is true, and when it does not, it
is false. One cannot, mentally or linguistically, assign a property to given
objects without some form of truth or falsity arising. Put differently, when a
listener successfully interprets a well-anchored and well-keyed statement as
produced by a speaker, then the L-proposition underlying that statement
bears a relation of corresondence or noncorrespondence to the state of affairs
at hand. That relation implies a truth value of some kind, no matter whether
the listener or speaker knows the exact value. There are, of course, cases where
truth is not well-defined, as in sentences with vague predicates such as grey,
which has vague borders with white on the one hand and black on the other,
or big, which is not only vague but also dependent on context-bound evalua-
tion. To cater for such cases one may wish to work with a system of truth
values intermediate between True and False, or to set up a grammatical
analysis that accounts for the context-bound evaluative factor, as is done in
Section 9.3 of Volume I. But even then some truth value will be assigned, so
that the subprinciple of complete valuation is not affected.
We are not saying, however, that it is impossible for a sentence to lack a
truth value. On the contrary, sentences that lack a key lack a truth value. For
example, I may, as part of my teaching, write the following sentence on the
board, perhaps to explain the grammatical process of auxiliary inversion in
English:
(1.5) Only then did she post the letter.
I may hold forth about this sentence for some time, explaining to the students
that it is remarkable for certain linguistic reasons. But it would be absurd for
me to ask the students to tell me whether this sentence is true or false, because
I have presented it as a type, not as a token utterance expressing a proposition
about a given state of affairs with a given person and a given letter. Sentence
(1.5), therefore, does indeed lack a truth value when uttered in the teaching
20 The Logic of Language

situation described, precisely because it is not a well-anchored and well-keyed

statement but a mere sentence type.
Of some sentence types, however, it does make sense to ask whether they
are true or false regardless of any specific anchor and key. For example, the
sentence:
(1.6) All humans are mortal.
whether written on a blackboard during a grammar lesson or stuck on
walls and billboards all over town to admonish people of their mortality, or
in whatever other context or situation, does have a truth value. This is so
because it contains no definite terms and no deictic tense, and because ALL
quantifies over the class of humans, which, one may assume, requires no
specific context.3 This makes it a so-called ETERNAL SENTENCE, following the
terminology introduced by the American philosopher Willard Van Orman
Quine in his Word and Object of 1960. As was shown in Section 3.1.4 of
Volume I, Quine opposes eternal sentences to occasion sentences, which
need referential focusing on specific objects in a specific state of affairs
on account of some definite term or deictic tense occurring in them. Eternal
sentences contain only quantified terms, while occasion sentences contain
one or more definite referring terms, that is, terms under a definite determin-
er, such as the definite article the, which needs some form of cognitive
intentional focusing on one or more specific objects for a proper interpreta-
tion. We maintain that it is normal for natural language sentences to be
occasion sentences, eternal sentences representing merely a marginal category
where the dependency on anchoring and keying has been reduced to zero or
near-zero.
The aversion to occasion sentences witnessed in twentieth-century logic
was anticipated by Aristotle, who writes (Prior Analytics 24a17–22):
A proposition is a positive or negative sentence [lógos] that says something of
something. Such a proposition is of three types: universal, particular, or indefinite.
A universal proposition is about all or none; a particular proposition is about some, or
some not, or not all; an indefinite proposition is about something applying or not
applying without any specification as to all or some, as when we say that knowledge of
opposites is the same knowledge, or that pleasure is not the same as the good.

3 We leave aside the many cases where universal quantification does require a specific context, as in
Tout Paris était là (All Paris was present), which, snobbishly, selects only a specific section of Parisian
society.
 Logic and entailment 21

Here we see that Aristotle rules out, perhaps not from his theory of language
but certainly from his logic—the main topic of Prior Analytics—all proposi-
tions that need a specific form of anchoring or keying for the assignment of
a truth value—Quine’s occasion sentences. Aristotle’s entire logic is built on
L-propositions corresponding to Quine’s eternal sentences. And of these
eternal sentences, it is only the universally and existentially quantified ones
that play a role in the logic. ‘Indefinite’, or as we might say, generic, sentences
are given a good deal of attention in On Interpretation, but they play no part
in the logical system. Aristotle would have no truck with occasion sentences,
probably, one surmises, because he saw the problems coming, as one cannot
deal with the logic, or indeed the semantics, of occasion sentences without
taking into account conditions of anchoring and keying, which pose an
immediate threat to the simplicity of the system. His refusal, or perhaps his
inability, to face this threat was canonized during the first half of the twentieth
century and vestiges of that attitude are still found today. This, however, is a
luxury we cannot afford when we investigate the logic and the semantics of
natural language sentences and words.
On the whole, logicians dislike the complications arising out of the condi-
tions of anchoring and keying. What they want is a logic that operates solely
on expressions whose grammatical wellformedness is a sufficient condition
for their having a truth value. They want to read the subprinciple of complete
valuation of properly anchored and keyed L-propositions as the subprinciple
of complete valuation of sentences as grammatical objects. But this is totally
unrealistic with regard to the logic and semantics of natural language. In
natural language, wellformedness of a sentence is condition one for an expres-
sion to have a truth value (though a great deal of implicit correction is allowed
for in practice). Condition two is that it be properly anchored and keyed.
Logicians want the latter condition to be either otiose or nonexistent, a wish
we must reject as being out of touch with the reality of language. We also say
that when an assertive sentence is uttered as a properly anchored and keyed
statement, it necessarily has a truth value, because it is impossible mentally to
assign a property to one or more objects without there being some form of
correspondence or noncorrespondence to what is the case—that is, some
form of truth or lack of truth.
The question now is: how many truth values are there in the natural
language system? Strawson considered it possible for a properly anchored
and keyed statement to have no truth value at all. That must be deemed
inadequate, as was shown in Section 1.3. Strawson’s proposal invites one to
treat his ‘lack of truth value’ as a truth value after all, though inappropriately
named. But if one does that, one needs a logic that takes more than the two
22 The Logic of Language

values True and False. And this is precisely what we find in natural language,
which, in our analysis, operates with the values True, False-1 or ‘minimally
false’, and False-2 or ‘radically false’ (probably with intermediate values
between the three).
So we uphold the subprinciple of complete valuation of properly anchored
and keyed L-propositions as being necessary by definition. But we are
prepared to tamper with the subprinciple of binarity. We feel free to do so
because giving up the subprinciple of binarity enables us to present a more
adequate account of the semantics of natural language, and also because in
logic this (sub)principle seems to be motivated merely by a desire to keep
logic free from the complications arising in connection with anchoring, key-
ing, and gradability. We need to consider the possibility not only of two
different kinds of falsity and thus of three truth values, but also of fuzzy
transitions between truth values. This places the Aristotelian axiom of strict
bivalence in a wider metalogical context, in that standard strictly bivalent
general logic turns out to be the limiting minimal case of an infinite array of
possible, and logically richer, general logics that vary either on an axis of
intermediate truth values or on an axis of semantically defined presupposi-
tional restrictions to certain contexts.

1.6 Some problems with the assignment of truth values

This leaves us with the question of how to assign truth values to L-proposi-
tions. In standard logic it is assumed that truth values are assigned with the
help of model theory, which produces a truth value for any well-formed string
of terms on the basis of (a) a given well-defined state of affairs and (b) given
meaning definitions of the terms and nothing else. In fact, this method of
assigning truth values is sometimes considered to be part of logic, rather than
merely a preliminary to logic. It is thus understandable that standard logic has
a tendency not to take into account the possibility that the machinery that
does the assigning of truth values has to be in a particular state in order to be
able to deliver a truth value. To the extent that standard logic is at all
concerned about the question of how truth values are assigned, it proceeds
on the assumption that the following principle holds:
PRINCIPLE OF COGNITIVE INDEPENDENCE OF TRUTH VALUE ASSIGNMENTS (PCI)
The cognitive machinery assigning truth values does so independently
of any state that the machinery in question happens to be in.
This purified, unworldly view has been of great use to standard mathematics,
which is likewise purely formal, never vague and whose dependency on
 Logic and entailment 23

mental contingencies is easily factorized out. But it has also been applied,
especially since the 1960s, to the study of linguistic meaning, whereby it was
assumed that linguistic meaning, like mathematical meaning, is independent
of mental contingencies and nonvague. This assumption has, however, proved
unwarranted over the past quarter-century. Not only do most predicate
meanings in natural language impose contextual restrictions, called precon-
ditions, which generate presuppositions; they are also often vague and/or they
incorporate all kinds of purely cognitive (often evaluative) conditions, besides
the conditions to be satisfied by the objects themselves to which the predicate
is applied.
We have no choice but to reject PCI as being irreconcilable with natural
language. For if sentences normally require anchoring and keying to have a
truth value, it follows that the machinery that does the anchoring and the
keying—that is, the human mind—must be at some suitable point in the
development of a discourse or context and must be intentionally focused on,
or keyed to, a particular state of affairs for the truth value assignment to take
place successfully.
This is amply borne out by natural language, which violates PCI in a
number of ways. For example, most uttered sentences contain DEFINITE TERMS
referring to specific objects or sets of objects to which, truly or falsely, a
property is assigned. For the reference relation to be successful it is necessary
that the means be available to identify the object or objects in question (Clark
and Wilkes-Gibbs 1990). In most cases, these means can only be provided if
the mind is in a contextually and referentially restricted information state.
Reference clearly requires specific anchoring and keying.
The same is found with regard to type-level lexical meanings. There are
cases where the satisfaction conditions of predicates depend on (contain an
open parameter referring to) what is taken to be normal in any given context.
Consider, for example, the predicate many. If it is normally so that out of an
audience of three hundred taking part in a TV quiz nobody gets the one-
million Euro prize, then, when in one session three participants get it, one can
say in truth that there were many one-million Euro prize winners in that
session. But if only three out of three hundred people voted for me in an
election, then, one fears, it is false to say that many people voted for me.
Or consider cases of what is called DYNAMIC FILTERING in Section 9.6.3 of
Volume I, found all over the lexicons of natural languages. For example, the
conditions for the predicate flat to be satisfied differ considerably when it is
applied to a tyre, a road, or a mountain. And a definite term like the office will
have different interpretations according to whether it stands as a subject term
to the predicate be on fire or, for example, have a day off. In the former case the
24 The Logic of Language

phrase the office is interpreted as referring to a building. In the latter case it is

interpreted as referring to a group of people.
Or take PREDICATES OF POSSESSION, expressed in words such as have, with,
without, or as a genitive or a dative case. As is shown more amply in Section
9.6.3 of Volume I, possession predicates typically require an appeal to world
knowledge for decisions about truth or falsity, as is shown by the following
two examples:
(1.7) a. This hotel room has a bathroom.
b. This student has a supervisor.
Clearly, for (1.7a) to be true, the room in question must have its own
individual bathroom, not shared with other rooms. But for (1.7b) to be true
no one-to-one relation is required, as one supervisor may serve a number
of students. This difference is not linguistic and, therefore, need not be
specified in the lexicon, but is caused by the way the world happens to be,
as far as hotels or university departments are concerned. The lexical
meaning description of possession predicates is thus taken to contain an
open parameter referring the language user to his or her knowledge base.
As is said in Seuren (1998a: 400): ‘If the language user does not know “the
way it is” with this or that (s)he cannot interpret the utterance in question.
For example, an expression like the room’s front wheels requires either a world
picture in which rooms have a known standard relation to front wheels,
or otherwise at least a context where this room’s relation to front wheels has
been explained.’
Cases of what is often called ‘viewpoint’ are likewise telling (see Section 9.7
in Volume I for more discussion). Consider the case of a fountain and a
tree which are located at a distance of, say, twenty yards from each other
(cf. Seuren 1998a: 402–3). Now to decide whether sentences like (1.8a) or
(1.8b) are true it must be known from what angle the fountain and the tree are
being looked at by some ‘homunculus’ or mental viewer in the mind:
(1.8) a. There is a tree behind the fountain.
b. There is a tree to the left of the fountain.
If the mental viewer is standing in the continuation of the line running from
the fountain to the tree, (1.8a) is true and (1.8b) is false. But if the homunculus
stands at right angles with that line, with the fountain to the right, then (1.8a)
is false and (1.8b) is true. Clearly, unless one is describing what is seen on a
postcard, the viewpoint is determined by previous discourse.
 Logic and entailment 25

Viewpoint differences are likewise to be discerned in cases that show an

orientation towards or away from a speaker’s mental position, as with pre-
dicates of the ‘come and go’-type:
(1.9) A few men were waiting to be shown into the office. The door opened
and one of them went in. After two minutes another man went in/
came in.
The difference between the two versions is, though clear, hard to express in
other than metaphorical terms. With went in, the ‘camera’ (or the homuncu-
lus) has, so to speak, remained outside the office. With came in, it has ‘moved
along’ with the first man and is now inside the office. Such phenomena are
clearly semantic, in that they reflect meaning differences, even though they do
not lead to different truth values.
Other lexical restrictions look more like being based on criteria of proto-
typicality (see Section 8.8 in Volume I). Suppose, for example, that John is a
man with no hair on his chest. Then, when one says that his chest is bald, that
would perhaps not be false in a strict sense but it would certainly be inappro-
priate, the reason being, probably, that the predicate bald prototypically but
not truth-conditionally, requires the top of the head when applied to humans.
In similar fashion one observes that what we call ‘red wine’ is called the
equivalent of ‘black wine’ in Modern Greek. We may say in English that a
person, when frightened, turns pale, but not, or rather not, that his or her
shoulders are ‘pale’ when not suntanned. Examples like these, of course, come
in the thousands.
Lexically fixed prototypicality also has a logical counterpart. For
example, when a married man says that he is eating out because all his
wives are ill, then that may be taken to be true in a strict, logical sense,
but not in the ordinary sense which would make one understand that
he has a collection of wives. This example probably hinges on an
element of prototypicality in the linguistic meaning of the logical word all,
combined with the fact that quantification in language is to do primarily
with plurality.
In view of all this it seems that we are forced to conclude that the Principle
of Cognitive Independence lacks empirical backing and seems to have been
added merely to avoid complications in the theory of truth. The fact that it is
widely considered sacrosanct in circles of logicians is part of what one might
call the mythology of logic. Meaning in natural language is a great deal more
complex than what we see of it in standard logic. The standard, logically
coloured concept of meaning is highly idealized and streamlined, which
26 The Logic of Language

creates the impression that one may adequately operate with a strictly bivalent
system of simple truth and simple falsity. Since, however, natural language is
very much more complex and more varied, there is room for the idea that
the logical system of language requires a more finely grained set of truth
parameters than just True and False. This is one of the questions we explore
in this book.
 2

Logic: a new beginning

2.1 Entailment, contrariety and contradiction:

the natural triangle
We start with the notions in terms of which the machinery of any logic is
defined, namely contrariety, entailment, contradictoriness, and subcontrariety.
These are logical relations that may hold between two, or sometimes more,
L-propositional types, depending on their logico-semantic properties, which,
in their turn, are determined by the logical constants or operators occurring
in them. (One remembers that an L-proposition is the type-level linguistic
expression of a proposition in the logico-semantic language LL.) Since
these relations are not satisfied by world situations but by combinations of
L-propositional types, we should really call them ‘metalogical relations’,
as opposed to the object-language logical relations expressed by the truth-
functional operators and quantifiers of LL. In practice, however, we use the
term logical relations in all cases where no confusion is to be expected. In
general, (meta)logical relations hold (are valid) in virtue of the meanings of
the L-propositional arguments. Within a logical system, they hold in virtue of
the meanings of the logical constants in the L-propositional arguments.
(Meta)logical relations should be clearly distinguished from the functors
of the object language LL, also called operators or logical constants, mainly
ALL, SOME, NOT, AND, OR, and IF . . . THEN, which produce truth or falsity depend-
ing on token world situations. In standard logic, the functors of negation
(NOT) and of material implication (IF P THEN Q) have a direct analog to a
(meta)logical relation, as they correspond directly to contradictoriness and
entailment, respectively: contradictoriness holds between any L-proposition
and its negation; entailment holds when the material implication is true in
all admissible situations.
One way of seeing how these (meta)logical relations get off the ground is
the following. Let us take as our point of departure the requirement of
consistency in linguistic interaction. Speakers who commit themselves to the
truth of what they assert or who appeal to listeners to fulfil a request or an
28 The Logic of Language

order must maintain consistency, or else the social fabric will collapse. For
that reason it is of the greatest importance, first to language users and
secondly to language theorists, to spot inconsistencies or, in logical terms,
to spot sets of propositions that cannot simultaneously be true. When such a
set consists of just two propositions, we speak of contrary propositions. This is,
in fact, how Aristotle set up his predicate logic in On Interpretation, which
revolves around the notions of contrariety and contradictoriness. Aristotle
called any two L-propositions which, for semantic reasons, cannot be simul-
taneously true ‘contraries’, and any two L-propositions that can, again for
semantic reasons, be neither simultaneously true nor simultaneously false
‘contradictories’. On the standard (yet oversimplified) assumption that all
natural languages have a logical constant of negation—not in English—which
inverts truth values (modulo key) under the axiom of bivalence, the contra-
dictory of an L-proposition P is its negation NOT(P), or ¬P: the truth of the
one entails the falsity of the other, either way.
To give a trivial example, the two sentences in (2.1a) are contraries, since
anyone who asserts both is guilty of inconsistency (‘><’ stands for contrari-
ety). This means that when (the L-proposition underlying) Joe has been
murdered is true, Joe is not dead must be false and, therefore, Joe is dead
must be true (assuming bivalence). In other words, Joe has been murdered
entails Joe is dead, as stated in (2.1b). (One notes that we have silently passed
from entailment in a general sense to logical entailment. This is legitimate
given the definition of bivalent negation.)
(2.1) a. Joe has been murdered >< Joe is not dead
b. Joe has been murdered  Joe is dead
Entailment can thus be defined in terms of contrariety and the contradiction-
producing negation operator. It is also possible, of course, to define contrari-
ety in terms of entailment and contradictoriness, but that seems less natural,
given the basic requirement of consistency in the use of language. Nor is it
how Aristotle proceeded.
Any relation of contrariety between two L-propositions P and ¬Q, thus
brings along a relation of entailment from P to Q. Figure 2.1 shows the
triangular relation arising from the assumed contrariety of P and ¬Q,
which causes the entailment from P to Q, the contradictory of ¬Q.
This triangle is arguably ‘natural’ in the sense that it may be taken to reflect
natural, as opposed to constructed, set-theoretic structures and relations—a
supposition that is further elaborated in Chapter 3. It forms the natural basis
of the logical system of propositional calculus with the operators AND, OR, and
 Logic: a new beginning 29

P
: entails
CD: contradictories

C C: contraries

CD
Q ¬Q
FIGURE 2.1 The natural logical triangle

NOT, and also of Aristotelian-Boethian predicate calculus (ABPC), which is

based on the quantifiers ALL and SOME and the propositional operator NOT.
But one can go further, still under the bivalence axiom. Since P and ¬Q in
Figure 2.1 are contraries, ¬Q entails the falsity of P, or: ¬Q ‘ ¬P. This is, of
course, nothing but the well-known operation of CONTRAPOSITION: when, for
any L-propositions P and Q, P ‘ Q, then also ¬Q ‘ ¬P. This is easily shown:
suppose Q is false, and therefore ¬Q true, then P cannot be true, for if P were
true, Q would also have to be true in virtue of the entailment P ‘ Q. Since
both P and Q have only two options, True and False, it follows that if P ‘ Q,
then in all cases where ¬Q is true, ¬P is also true, that is, ¬Q ‘ ¬P. This can be
added to the triangle of Figure 2.1, leading to the square of Figure 2.2.
Now that the natural triangle has been extended to a square, a new logical
relation has emerged, that of SUBCONTRARIETY. By definition, two L-proposi-
tions form a pair of subcontraries when they cannot be simultaneously false,
though they may be simultaneously true. In this case, Q and ¬P cannot be
simultaneously false, because, as the square shows, if Q is false, so is P, and
when ¬P is false and thus P is true, Q is true. Clearly, the relation of

P ¬P
CD

C SC SC: subcontraries

CD
Q ¬Q
FIGURE 2.2 The natural logical triangle extended with contraposition
30 The Logic of Language

a. P ¬Q b. P ¬Q
C C

CD CD
CD

SC
Q Q ¬P

FIGURE 2.3 The logical triangle and square in the Boethian arrangement

contradictoriness is the combination of contrariety and subcontrariety—at

least in a bivalent logic with the metalogical relations as defined.
Figure 2.3a shows a different arrangement of the vertices of the natural
logical triangle, extended with contraposition to the square of Figure 2.3b.
This arrangement has a more familiar look as it reminds one of the famous
Square of Opposition in traditional predicate logic, attributed to the Roman-
Christian philosopher and statesman Boethius ( 480–524). For that reason,
we call this the Boethian arrangement. It is shown, however, in Section 2.2,
that the Boethian arrangement is not a very good way of presenting metalo-
gical relations and had better be given up in favour of the arrangement of
Figure 2.2.
Apart from this, it will be clear that the square of Figure 2.2, or Figure 2.3b
for that matter, cannot be the whole story, since, if it were, both the triangle
and the square, in either form, would be dispensable, as they merely reiterate
the entailment relation P ‘ Q (or P  Q) in terms of contrariety, contradic-
toriness and subcontrariety, given the negation as a truth-functional operator
inverting binary truth values. The logic would also be pretty poor, consisting
only of negation (important though that operator is) and a few new relations
that can be defined in terms of each other—hardly sufficient to get a logic off
the ground. What makes these new relations important for the machinery of
logic is the establishing of logical relations between the external negations
occurring in L-propositions of the form ¬P on the one hand, and
corresponding L-propositions with internal negation on the other. What is
meant by ‘internal negation’ is a question to which we turn now.
 Logic: a new beginning 31

2.2 Internal negation and duality: the natural square

and the Boethian square
Just about the most central operator in any logical system is negation,
normally lexicalized as not in English. We concentrate first on ordinary
external negation over L-propositions. This negation is typically a comple-
ment selector. In the simplest case, it selects the complement within a universe
U of admissible situations.
To make this clear, we fall back again on the notion of valuation space,
introduced in Van Fraassen 1971 and briefly discussed in Section 3.3.3 of
Volume I. Leaving the formal definition of the notion of ‘admissible situation’
to Section 2.3.3 below, we define ‘valuation space’ (VS) as follows:
VALUATION SPACE
For any L-proposition P expressing the proposition p, the VALUATION
SPACE (VS) of P, or /P/, is the set of admissible situations that make P
true (modulo key).
Thus, the proposition p (or its L-propositional expression P) is true in
all situations that are elements in the set of situations /P/. Given that
the standard bivalent negation ¬ selects the complement of /P/
within the universe of all admissible situations U, it follows that the proposi-
tion expressed by ¬P is true in all situations that form the com-
plement=P=ð¼ U =P=Þ of /P/ in U. This simplest case is what is found in
standard logic. Since, by definition, an L-proposition P is true just in case the
actual situation sitact is an element of /P/ and false otherwise, it follows
that ¬P is true just in case sitact is an element of /¬P/—that is, =P= —and
false otherwise. In the standard, strictly bivalent, system shown diagrammati-
cally in Figure 2.4 no other possibilities than truth and falsity are left open:

/¬P/

/P/

FIGURE 2.4 VS representation of /P/ and /¬P/ in U

32 The Logic of Language

P ¬P

sitact ∈ /P/ T F

sitact ∈ /¬P/ (= U –/P/) F T

FIGURE 2.5 Truth table of the standard bivalent negation ¬ in terms of VS-modelling

‘not true’ equals ‘false’ and ‘not false’ equals ‘true’. Clearly, if P is necessarily
false, as, for example, the L-proposition underlying the sentence He was
dead for the rest of his life, /P/ ¼
; if P is necessarily true, as in He
was alive for the rest of his life, /P/ ¼ U.
Thus defined, negation can be said to ‘toggle’ between truth and falsity. The
property of the negation simply to invert truth values of propositions is
expressed in the TRUTH TABLE of the standard negation shown in Figure 2.5
(adapted to a valuation space interpretation), where ‘T’ stands for True and
‘F’ for False.
It is possible, however, to vary on this theme. For example, one may define a
negation operator that selects the complement of /P/ within a subset of situa-
tions in U defined on the basis of preceding context or the meaning of the main
predicate in P, in which case room is made for more than one negation. In this
book we argue that this is, in fact, the situation in natural language, where the
main function of the negation NOT is to toggle between the values ‘true’ and
‘minimally false’, minimal falsity being caused by those situations that are
outside /P/ but within P’s subuniverse in U. Further comment on this issue is
provided in Chapter 3, but a full discussion has to wait till Chapter 10.
But what is meant by internal negation? So as not to complicate matters
unduly at this stage, let us say that the internal negation is a negation not over
an L-proposition but over a predicate. This is not an adequate definition, since,
as is shown in Sections 2.3.5.2 and 2.4.1, internal negation is better defined as a
small scope negation over an embedded L-propositional structure, but it will
do for the moment. If it is accepted that a predicate, for example human,
expresses a property possessed by all objects in the world that are indeed
human, then the ‘negation’ of this predicate, not-human, functioning as the
internal negation of an L-proposition, expresses the lack of that property for all
objects that are not human. Call the set of all human objects the EXTENSION of
the predicate human, or [[Human]], in the set of all objects OBJ. Then the set
 Logic: a new beginning 33

of objects that are not human is the extension of the predicate not-human, or
[[NOT-Human]], in OBJ. This makes [[NOT-Human]] the complement of
[[Human]]—that is, OBJ minus [[Human]], or ½½Human—as long as there
are no vague boundaries between what is and what is not human.
The logical interest of internal negation, in the present context, lies in the
relation of DUALITY (Löbner 1990):
DUALITY
Two logical constants X and Y are each other’s duals just in case there is
logical equivalence between X preceded by the external negation and Y
followed by the internal negation and, of course, vice versa.
In standard predicate calculus, for example, the quantifiers ALL and SOME are
each other’s duals, since in that system an L-proposition corresponding to the
form NOT ALL F is G (where F and G are predicates) is equivalent with
an L-proposition of the form SOME F IS NOT-G and, analogously, NOT SOME
F IS G/NO F is G is equivalent with ALL F IS NOT-G. These equivalences are
standardly known as the CONVERSIONS. Henceforth, when dealing with external
versus internal negation, we will use the standard symbol ‘¬’ for external
negation and the symbol ‘*’ for internal negation. That is, ¬P is the external
negation of the L-proposition P and P* is the L-proposition P but with its
main lexical predicate negated by the internal negation. Obviously, in a
strictly bivalent system, double external negation and double internal nega-
tion cancel out: for any L-proposition P, ¬¬P
P and, since for any predicate
C, NOT-NOT-C
C, P**
P.
It must be noted, at this point, that the logical properties of two
L-propositions P and P* are identical for the simple reason that the choice of
lexical predicates is irrelevant for the logic, which is defined by the logical
constants only. We call this the MODULO-*-PRINCIPLE. The relevance of the
internal negation for predicate logic lies in any logical relation of duality
or of a one-way entailment between two logical constants X and Y when
one is preceded by the external negation and the other is followed by the
internal negation. The Modulo-*-principle provides an extra check for
the soundness of a predicate-logic system in that any system that violates the
Modulo-*-principle is by definition unsound.
Logicians have observed that a similar relation of duality exists between the
propositional operators AND and OR, a form of duality known as DE MORGAN’S
LAWS. In standard propositional logic, the operators of negation, conjunction
and disjunction are defined in such a way that for the L-propositions P and Q,
NOT(P AND Q) is equivalent with NOT(P) OR NOT(Q). Analogously, NOT(P OR Q)
34 The Logic of Language

is equivalent with NOT(P) AND NOT(Q), or in standard notation: ¬(P ∧ Q)

¬P ∨ ¬Q and ¬(P ∨ Q)
¬P ∧ ¬Q.
This is, of course, an intriguing parallel with standard predicate calculus,
but if we want to cash in on this parallelism it must first be made clear that
here, too, we have to do with internal negation. To do that, the definition of
internal negation as predicate negation must be generalized so as to cover the
dual quantifiers and the dual propositional operators alike. This is done in
Section 2.3, where all logical operators are reinterpreted as (abstract) predi-
cates. The parallelism between the two calculi is then further elaborated in
Chapter 4. For the moment, we limit ourselves to the dual quantifiers.
Let us revert to the logical triangle of Figures 2.1 and 2.3a, and the logical
squares of Figures 2.2 and 2.3b, respectively, assuming that P and Q are types
of L-proposition defined by logical constants that are each other’s duals. That
is, we assume that:
(2.2) P
¬Q* and consequently ¬P
Q*
Q
¬P* and consequently ¬Q
P*
Now the squares of Figures 2.2 and 2.3b are seen to be more than just
restatements of the entailment relation and to have real logical power. For
now the triangular Figures 2.1 and 2.3a can be expanded to quadrilateral
figures that express the duality relation. This is done in Figure 2.6, where
(a) corresponds with Figure 2.2 and (b) (in either form) with Figure 2.3b.
It now becomes clear why the arrangement of Figure 2.2 (expanded from
the triangle in Figure 2.1) is preferred to the Boethian arrangement of Figure
2.3b (expanded from Figure 2.3a). When the triangles are expanded to squares
in such a way that the dual equivalences are expressed, Figure 2.1 results in
Figure 2.6a, which we call the NATURAL ISOMORPHIC SQUARE (two parallel lines
stand for logical equivalence). But Figure 2.3a turns into Figure 2.6b (in either
form), which we call the IMPROVED BOETHIAN SQUARES. The difference is that the
natural isomorphic square consists of two triangles that are LOGICALLY ISOMOR-
PHIC in that they display the same logical relations of entailment, (sub)-
contrariety and contradiction, the only difference being that the right-hand
side triangle has the internal negation added to P, Q, and ¬Q, giving P*, Q*,
and ¬Q*, respectively—that is, with a negative instead of a positive predicate.
But since the choice of lexical predicates makes no difference to the logic,
which is defined by the logical constants alone (the Modulo-*-principle), the
two triangles making up the natural square are logically identical. It is their
connection through the Conversions that makes a logic spring up.
The interest of their being coupled into the natural square lies in the dual
equivalences. By contrast, the two improved Boethian squares in Figure 2.6b
 Logic: a new beginning 35

P ¬Q* ¬Q
b. C

a. c.
CD CD CD
P ¬Q* Q* P ¬P Q*
CD CD
C
SC C SC
SC
Q Q* ¬P SC
C
improved Boethian 1

P ¬Q P*
CD C CD
Q ¬Q P* Q ¬P* P*
natural isomorphic nonnatural isomorphic
CD CD CD

SC
Q ¬P* Q*
improved Boethian 2

FIGURE 2.6 (a) the natural isomorphic square, (b) the two improved but noniso-
morphic Boethian squares, and (c) the nonnatural isomorphic square

lack this regularity. There the triads <P,Q,Q*> and <P,Q,¬Q> of the left-
hand side triangles correspond to the nonisomorphic triads <¬Q,¬P,¬Q*>
and <P*,Q*,¬P*>, respectively, of the right-hand side triangle. Clearly, the
isomorphic representation displays more regularities in the logical system
than the nonisomorphic representation.
Figure 2.6c shows a second way of making a square out of two isomorphic
triangles with the same logical power.1 Here we have a triangle <P,¬P,Q>,
linked up with the isomorphic triangle <P*,¬P*,Q*>. Now the relations
involved are those of entailment, contradiction, and subcontrariety, with
contrariety thrown in as a bonus due to the linking up through the Conver-
sions. In Chapter 3, however, it is argued that subcontrariety is not a basic-
natural logical relation but a relation that requires a great deal of scholastic
training to be grasped, whereas contrariety is just about maximally natural.
For that reason, the square of Figure 2.6c is called ‘nonnatural isomorphic.’
From a strictly logical point of view, the difference is immaterial, but it is not
when we are in search of natural logic. Since natural logic is what we are after,
we consider Figure 2.6a to be the preferred representation.

1
I am indebted to Dany Jaspers (2005: 34–5) for calling this fact to my attention.
36 The Logic of Language

The isomorphy of a representation is important not only because it shows

up more regularities in the system than nonisomorphic representations but
also because it adds a criterion for the consistency of the logic in question.
The criterion can be formulated as follows:
THE ISOMORPHY PRINCIPLE
Given the Modulo-*-principle, the (meta)logical relations defined for
the operators of predicate calculus (the quantifiers and negation) must
be the same no matter whether the main or G-predicate is or is not
characterized by internal negation.
In virtue of the Modulo-*-principle, the identity of the logical relations in the
two triangles <P,Q,¬Q> and <P*,Q*,¬Q*> in Figure 2.6a, or in <P,¬P,Q>
and <P*,¬P*,Q* > in Figure 2.6c, is thus a condition for the soundness of the
logic involved. Since the renderings shown in Figures 2.6a and 2.6c actually
show the isomorphy, whereas those of Figure 2.6b do not, the former are at
least didactically superior to the latter.
The fact that logical systems with X and Y as duals and an entailment
relation from X to Y are reducible to two logically isomorphic natural
triangles, one with and one without the internal negation, connected by the
dual equivalences, has so far, as far as I am aware, not been noted in the
literature. It is nevertheless of major interest, because traditional Aristotelian-
Boethian predicate calculus (ABPC) is built up from precisely the same
elements, with the quantifiers ALL and SOME as each other’s duals and an
entailment from ALL to SOME.
What we call the Boethian arrangement of Figure 2.6b reflects the celebrat-
ed Square of Opposition shown in Figure 2.7a and attributed to Boethius

a. A E b. A ¬I* I*
C CD
C
CD CD SC
C

CD
I ¬I A*
SC
I O A: All F is G
I: Some F is G
E: No F is G / All F is not-G
O: Some F is not-G / Not all F is G

FIGURE 2.7 ABPC represented as the Boethian Square of Opposition and as the natural
square consisting of two isomorphic triangles
 Logic: a new beginning 37

mentioned earlier. The symbol A stands for simple universally quantified

sentences of the type ALL F IS G, the symbol I stands for simple existentially
quantified sentences of the type SOME F IS G. Boethius chose these symbols
because they are the first two vowels of the Latin word affirmo (I affirm). He
added the symbols E and O, the vowels of the Latin word nego (I deny), for
the counterparts of A and I with internal negation—that is, for sentences of
the type ALL F IS NOT-G and SOME F IS NOT-G, respectively.
In spite of its august, time-honoured status, we now see that the Boethian
Square is a defective way of representing traditional predicate calculus, be-
cause it fails to express the function of the internal negation and thus
the generalization that the system is representable as two logically isomorphic
triangles, <A,I,¬I> and <A*,I*,¬I*>, connected by equivalences (Conver-
sions). We, therefore, dispense with the symbols E and O, using instead
A* and I*, respectively, in the natural isomorphic square representation
of Figure 2.7b.2 And we replace the arrangement of Figure 2.7a with that of
Figure 2.7b.

2.3 Logical operators as predicates

So far we have taken a preliminary look at (meta)logical relations. Now we
turn to the internal structure of L-propositions in the object language, in
particular to the logical constants, and in doing so we hold on firmly to a
linguistic point of view. The logical constants are treated as words or lexical
items in a language, with translation equivalents in all natural languages.
Being linguistic expressions, they must have a meaning, just like all other
linguistic expressions. It is our purpose, in this section, to begin exploring
aspects of the linguistic meanings of the logical constants and to show how
their specific logical properties follow from their linguistic meaning.
It is argued that the logical operators are best described as abstract pre-
dicates—that is, predicates in the logical language, LL, whose L-propositions
are taken to be related to the sentences of natural languages through gram-
mar.3 In formal terms, any function associated with a linguistic construct and

2
Jaspers (2005) calls the I-vertex the PIVOT of the triangle, mainly on the grounds that particular
existential knowledge is cognitively prior to general knowledge. The concept of pivot is useful in
natural logic also because the negation of the pivot establishes the relation of contrariety between the
A-vertex and the ¬I-vertex, thus stressing the triangular character of the relations of contrariety,
entailment, and contradictoriness.
3
Interestingly, the view that quantifiers are, logically speaking, predicates is also found in some
nineteenth-century logic textbooks. Thus we find (Sigwart 1895: 160):
38 The Logic of Language

with a truth value as output is seen as a predicate. This is obviously so for

ordinary unary predicates like walk, which deliver a truth value for any given
entity as input. And a binary predicate like love delivers a truth value for any
given pair of entities as input. Likewise, we may consider the quantifier ALL to
be a binary predicate, not over pairs of entities but over pairs of sets of entities
(which makes it, technically speaking, a higher-order predicate), delivering
truth when one set (for example, the set of humans) is included in the other
(for example, the set of mortals), and falsity otherwise. Alternative analyses
are possible (as in the standard Russell notation), but the predicate format is
always useful (and always provides a deeper insight into the matter).
Among the advantages of treating logical operators (constants) as (abstract)
predicates are the following—apart from the general consideration that the
insight into the nature of logic is greatly enhanced. First, treating logical
operators as arguments unifies the format of the semantic definitions of the
logical operators with that of common lexical predicates in that both are statable
as satisfaction conditions. This allows one to see with greater clarity than before
the relation between logic and language. Logic turns out to be a question of
lexical semantics: it is the lexical semantic definitions of the logical constants
that define the logic associated with them (as long as the definitions make for a
consistent system of mathematically definable entailments). It also means that
any lexicological standards and principles taken to be valid for predicates in
general (such as the possibility of inducing presuppositions) will now equally
apply to the logical constants.
Then, assigning predicate status to logical operators facilitates and shows
the way towards the development and foundation of alternative logics, thus
opening up the possibility of evaluating a variety of logics as possible candi-
dates for the status of the logic of language.
The treatment of logical operators as abstract predicates is also linguistical-
ly motivated in that more and more languages are being discovered where
what we see as logical operators actually occur as verbs in surface structure
(see note 4 for some evidence). In general, as is argued in Seuren (1996), the
mechanism of grammar is greatly improved or simplified when all elements
with lexical meaning are treated as predicates at the level of semantic analysis.

Thus, according to its original meaning. ‘All A’s are B’ can only be said in reference to definite particular objects. And
here from a logical point of view the ‘all’ is the predicate (the A’s which are B are all A’s).

That this was not an isolated view is proved by the fact that the linguist Meyer-Lübke held the same
view as regards the existential quantifier TWO (Meyer-Lübke 1899: 352; translation mine):
From the point of view of logic there can be no doubt that in the sentence il arrive deux étrangers [two foreigners
arrive] the subject is il arrive while deux étrangers is the predicate.
 Logic: a new beginning 39

The explicit proposal to treat logical constants formally as predicates is not

new. To my knowledge, it was first made, in the context of the theory of
grammar, by McCawley (1967, 1972: 516–32), who presented it in his usual
visionary way but without much in the way of formal underpinning. Yet, as
regards the quantifiers, the formal underpinning was soon provided in
the theory of generalized quantifiers (Barwise and Cooper 1981), based on
Mostovski (1957). As regards the propositional operators, the basis for their
interpretation as predicates is more or less implicit in the standard machinery,
which takes the extension of an L-proposition to be its truth value. If all
functions to truth values are treated as predicates, the truth functions are
automatically predicates, since they are functions from truth values to truth
values. We, however, develop a different perspective (see Section 6.2 in
Volume I), in which a well-anchored and well-keyed L-proposition is not
taken to stand for its truth value but to be cognitively (but without awareness)
associated with its valuation space—that is, the set of situations in which it is
true. It is shown in a moment that in this latter view the propositional
operators are even more naturally regarded as (abstract) predicates.
In the surface structures of most natural languages, the logical constants
under discussion are, of course, not predicates. They turn up as bound
morphemes, particles, determiners, or what not, but never, or very rarely, as
surface predicates—that is, as verbs or adjectives.4 We do not propose to treat
the logical constants as verbs or adjectives in surface structure, which would
strain even our powers of imagination. What we do propose is to treat them as
predicates in L-propositions—that is, in the logical language used for the
purpose of semantic analysis, along with the common lexical predicates that
may serve as values for the predicate variables.
An interesting question, in this context, is why it should be that the logical
constants, if they are ‘deep’ predicates, tend not to turn up as predicates in

4
In the Finno-Ugric languages, as McCawley observed, the negation operator often turns up as a
surface verb, though usually with a defective paradigm (which, in fact, corresponds closely to the
defective paradigm of the English modal auxiliary verbs such as can, may, must, will). A sentence like
Kevin didn’t laugh comes out as something like ‘Kevin notted (to) laugh’. Similar phenomena are
found in many other languages. For example, in the Amazonian language Dâw (Andrade Martins
2004: 559), the verbal negation suffix -e~h is derived from the negative verb me~h meaning ‘not have’.
Brown and Dryer (2008) describe a language, Walman (a Torricelli language of Papua New Guinea),
where the conjunctor and as a connective linking two or more NPs turns up in surface structure as a
verb. Brown and Dryer (2008: 563) are puzzled by this fact, because, in a predicate-logic analysis, there
seems to be no way in which the NP-conjoining and can be taken to be a verb or a predicate, since a
nominal conjunction can be nothing but an argument to a predicate. Their proposal to view this
verbal and as a serial verb seems plausible for those nominal conjunctions that do not allow for a
reduction to clausal conjunction, as in Vivian and Lesley are a nice couple. As a clausal connector, and is
easily seen as a (truth-functional) n-ary predicate that takes two or more S-structures as arguments.
40 The Logic of Language

surface structure but mostly as adverbials, particles, bound morphemes,

determiners, or phonological modifications. This question is part of the larger
question of why it should be that humans do not talk to each other in the
language of predicate calculus but in surface structures that rather deform the
‘deep’ predicate-calculus structures of the language LL of semantic analysis.
And this again is part of the even more comprehensive question of why
languages should transform SAs into surface structures at all.
It seems that an adequate answer to these questions will at least involve the
consideration that all natural language grammars are characterized by the
property of MATRIX GREED—the fact that the lexical matrix or nucleus of a
sentence structure tends to incorporate all, or almost all, more abstract,
functional elements of the auxiliary system and also, as much as possible,
nominal, verbal, and adverbial elements of embedded clauses (Seuren 2004:
195–8). The lexical matrix is that part of an SA-structure where the world-
related lexical items are concentrated. This makes the lexical matrix the most
‘referential’ part of the SA-structure—the part specifically housing the rela-
tions with objects in the world. It is perhaps not too far-fetched to conjecture
that, both from an evolutionary and from a functional point of view, the
lexical matrix is somehow the ‘core’ of the whole sentence, which cannot be
missed even in deficient or handicapped forms of human communication.
This would provide a basis for a possible answer to the questions just posed.

2.3.1 Meaning postulates

Attempts have been made before to break through the traditional barrier
between, on the one hand, the logical constants, which are standardly taken to
be sui generis and, on the other, the propositional or lexical linguistic material,
which is standardly treated as providing the values for the lexical variables in a
logic. A well-known attempt is Carnap’s proposal to treat lexical meanings as
logical elements allowing for certain inferences in virtue of so-called MEANING
POSTULATES (Carnap 1956: 222–9). For Carnap, the lexical meaning of a predi-
cate like bachelor can be said to legitimize on LOGICAL grounds the inference
that, for example, if anyone is a bachelor, he is not married. The format used
by Carnap for meaning postulates is, exemplified for the item Bachelor:
‘ 8x(Bachelor(x) ! Male (x) ∧ ¬Married(x) ∧. . .∧)
or: ‘it is a theorem that for all x, if x is a bachelor then x is male, not married,
. . . ’. When the semantico-logical entailments specified in this meaning pos-
tulate are incorporated into the logical system, then the sentence John is a
 Logic: a new beginning 41

bachelor not only semantically but also logically entails that John is male, not
married, and so on.
A meaning postulate is thus a stipulated entailment schema or ‘inference
rule’ meant to help define a logical system. It is important to realize that
meaning postulates of the type given above must be read as entailment
schemata that are part of the logical system, and not as formulae within the
system, which would turn them into contingent statements about the world.
For example, suppose it happens to be the case that all houses in the village
have two bathrooms, then it is true to say ‘for all x, if x is a house-in-the-
village, then x has two bathrooms’. But this does not provide a licence to
describe the lexical meaning of the predicate house-in-the-village as implying
the condition ‘having two bathrooms’. Meaning postulates are not meant to
be true about the world but true about the language (which has been
incorporated into the logic in so far as it has been fitted out with meaning
postulates) and hence, once the language has been fixed, necessarily true in
any world. For that reason the entailment sign ‘ is placed in front of the
meaning postulate.
Meaning postulates have had something of a career in formal semantics
and in the philosophy of language, where it has been claimed that they can be
used to define lexical meanings exhaustively, specifying the severally necessary
and jointly sufficient conditions for them to produce truth when applied to
their term referents. The dominant view in formal semantics and in the
philosophy of language has been to accept the principle ‘get your entailments
right and you get your meanings right’.
Although this view has a certain appeal, I believe—and have frequently
argued in the past—that it is basically flawed. I take sides with Frege in that I
take lexical meanings to be defined by satisfaction conditions (Frege’s Sätti-
gungsbedingungen), which have their roots in cognitive criteria. Natural
semantic entailments follow from lexical satisfaction conditions but do not
define them. In my view, moreover, lexical meaning is more than just truth-
conditional satisfaction conditions, in that the satisfaction of lexical semantic
conditions, besides leading to truth, often also leads to other sorts of results,
often of an ill-understood nature. Some aspects of this question are discussed
in Chapters 8 and 9 of Volume I.
The theory of meaning postulates constitutes an attempt at incorporating
natural language into, or superimposing it onto, a logical system, which is
considered to be given a priori and is implicitly taken to be standard modern
logic, the only logical system considered viable. What is tried here is the exact
opposite. For us, any logical system is part of some (natural or artificial)
language. We try to see logical properties as epiphenomena—certainly highly
42 The Logic of Language

interesting and important ones but nevertheless epiphenomena—of the se-

mantics of lexical predicates in natural or artificial languages. Instead of
taking the standard logical system as primitive, we regard the logical system,
any logical system, as derived from such lexically given semantic entailments,
in whatever language, as are seen to allow for a formal truth calculus on the
basis of set-theoretic relations defined in, or following from, them. For us,
therefore, the primitive elements are lexical meanings and the semantic
entailments that follow from them. As regards predicate logic, those lexical
meanings that can be specified in the mathematical terms of set theory are
possible candidates for a logical system.
One reason for taking this approach, apart from any possible empirical or
philosophical arguments, is that it enables us to puncture the myth most
formal semanticists and philosophers of language, and also most pragmati-
cists, implicitly live by, namely that standard modern logic is the only viable
logic and that if that logic turns out not to fit the facts of natural language too
well, the fault lies with natural language, not with the logic. Our approach
creates a space that the theory of meaning postulates denies us, the space
needed to develop sound alternative logics so that we can begin to investigate
which of these logics fits the facts of language best. And this again makes us
put off to a later moment any final answer to the question of the logical
soundness of natural language. For the best part of the twentieth century,
logicians have tried to persuade the world that natural language is vague,
ambiguous, and generally unfit for proper reasoning, with the result that the
new discipline of pragmatics arose, whose practitioners undertake massive
efforts to find functionally motivated excuses for what has been accepted as
being the logical impurity of natural language. We find this premature, since,
in our view, not enough has been done to see if natural language does indeed
suffer from the logical defects imputed to it. And what we have found so far
provides sufficient grounds for assuming that language is not nearly as bad, in
this respect, as it has been made out to be.
It is thus central to our method to regard the traditional logical constants as
the lexical items they are. And since lexical items are, in principle, predicates,
we propose to treat the logical constants as predicates in natural languages.
Not as ordinary ‘descriptive’ predicates, which are mostly defined in terms of
contingent properties that objects and relations in the world must possess for
truth to come about, but as ‘abstract’ predicates of LL, which tend not to
occur as predicates in surface structures and whose satisfaction conditions are
of a mathematical nature.
It follows that we must assume that the human mind is innately equipped
with the machinery required for carrying out such abstract mental
 Logic: a new beginning 43

operations, first at a natural, then at a higher, more constructed, level. And

since there is no way humans can actually report on the operations in
question, it must be assumed that they are wrapped up in, or as part of, a
MODULE, beyond any threshold of awareness and taking place in the hidden,
‘underground’ compartments of the human mind that are screened off from
awareness or conscious access. Much of this book is devoted to an exploration
of the consequences of this point of view.

2.3.2 Boolean algebra and the operators of propositional calculus

But let us revert to the matter at hand. We start with propositional calculus
and look at the logical constants AND, OR, and NOT. IF . . . THEN plays no role in
the present context, first because IF P . . . THEN Q is logically definable as
NOT(P) OR Q, and secondly because it is doubtful whether IF . . . THEN func-
tions as a truth function in natural language (see Kratzer 1979; Evans and Over
2005; see also Section 8.2.4). The three operators AND, OR, and NOT are
standardly considered to be TRUTH FUNCTIONS in the sense that they produce
a new truth value on the basis of the truth value(s) of their component
argument L-proposition(s). The semantic definition of these operators can
be presented in different formats. The standard format (introduced in the
1920s, but based on definitions provided by ancient Stoic philosophers; see
Lukasiewicz 1934) is that of the TRUTH TABLES. The standard truth tables for
these operators are given in Figure 2.8. (‘T’ stands for truth, ‘F’ for falsity; the
variables P and Q range over L-propositions (modulo key); ∧ is the standard
symbol for AND; ∨ stands for OR; ¬ stands for standard bivalent negation.)
Such truth tables may be taken to represent the meanings of the operators
concerned in that they specify the conditions under which (token) L-proposi-
tions composed by them are true or false. These conditions are exhaustively
expressible in terms of the truth values of the component L-propositions,
which makes the operators in question truth functions—that is, functions
from truth values to truth values.
Usually, ∧and ∨ are treated as taking two argument propositions. The
truth-tables of Figure 2.8 also treat ∧ and ∨ as binary operators. In fact,

P ¬P P ∧ Q Q: T F P∨Q Q: T F
———— ————— —————

T F P: T T F P: T T T

F T F F F F T F

FIGURE 2.8 Standard truth tables for ¬, ∧ and ∨

44 The Logic of Language

however, there is no limit to the number of argument L-propositions. The

general rule is that P ∧ Q ∧ R ∧ S ∧ . . . is true only if all argument
L-propositions are true, and false if at least one of them is false (falsity is
‘infectious’). Analogously, P ∨ Q ∨ R ∨ S ∨ . . . is true just in case at
least one of the argument L-propositions is true, and false only if all argument
L-propositions are false (truth is ‘infectious’). The n-ary character of these
two operators is usually expressed by saying that they are ASSOCIATIVE.
In this truth-functional perspective, which derives from the work of
Gottlob Frege (1848–1925), the EXTENSION of an L-propositional token is
taken to be its truth value. Just as the extensions (reference objects) of the
argument terms of a logically atomic L-proposition P determine the exten-
sion (truth value) of P under the lexical predicate of P, in the same way the
extensions (in this case the truth values) of the argument L-propositions of a
logically complex L-proposition determine its truth value under the logical
operator(s) building up the complex L-proposition. Viewed this way, the
logical operators of propositional calculus begin to look like predicates.
Alternatively, as a notational variant, these three propositional operators
can be defined in terms of the operators of Boolean algebra. George Boole
(1815–1864) developed the algebra that carries his name in order to bring logic
and mathematics closer together. This algebra turned out to be directly
applicable to sets as mathematical objects and thus became the central
foundation of set theory.
Boolean algebra has two CONSTANT SYMBOLS, 1 (unity) and 0 (zero), and any
number of VARIABLE SYMBOLS or VARIABLES (x,y,z, . . . ) ranging over 1, 0 and any
elements that may be needed for any given interpretation (such as sets in set
theory). It has, furthermore, two (associative and commutative) binary op-
·
erations, ADDITION (þ) and MULTIPLICATION ( ) defined as follows (‘¼’: mutu-
ally substitutable in all contexts; ‘6¼’: not mutually substitutable in all
contexts):
MULTIPLICATION: (1) x • x ¼ x (3) x • 0 ¼ 0
(2) x • 1 ¼ x
ADDITION: (1) x þ x ¼ x (3) x þ 0 ¼ 0
(2) x þ 1 ¼ 1
_
It also has a unary operation COMPLEMENT x, defined as follows:
COMPLEMENT: (1) 1 ¼ 0 (4) x • x ¼ 0
(2) 0 ¼ 1 (5) x ¼ x
(3) x þ x ¼ 1
 Logic: a new beginning 45

One may add the (nonassociative, noncommutative) operation SUBTRACTION:

SUBTRACTION: (1) x  x ¼ 0 (4) x  0 ¼ x
(2) x  1 ¼ 0 (5) 0  x ¼ 0
(3) 1  x ¼x ·
(6) xy = z iff x z=z and y z=0 ·
Moreover, addition distributes over multiplication and vice versa:
x þ ðy • zÞ¼ðx þ yÞ • ðx þ zÞ and x • ðy þ zÞ¼ðx • yÞ þ ðx • zÞ
Finally, the following laws hold:
(1) x þðx • yÞ ¼ x and x • ðx þ yÞ ¼ x
(2) x • y ¼ z iff x þ z ¼ x and y þ z ¼ y
(3) x þ y ¼ z iff x • z ¼ x and y • z ¼ y

It is easy to see that Boolean algebra formalizes set theory. Let the variables
x, y, z, . . . range over sets, while 0 is the null set
and 1 stands for the
domain of objects OBJ. Interpret multiplication as set-theoretic intersection
(\), addition as set-theoretic union ([), and complement as set-theoretic
complementðxÞ. Boolean algebra now computes all set-theoretic operations.
Boolean algebra also computes the truth functions of propositional calcu-
lus. One way of doing so was developed by Frege. Instead of taking sets as
values of the Boolean symbols, Frege took ‘truth’ as the value of Boolean 1
and ‘falsity’ as the value of Boolean 0 (which is the origin of the widespread
convention, not followed in this book, to use ‘1’ for truth and ‘0’ for falsity).
No other symbols are required. NEGATION (¬) is now interpreted as Boolean
complement, CONJUNCTION (∧) as Boolean multiplication and DISJUNCTION (∨)
as Boolean addition. Let ç(P) stand for the truth value, that is the Fregean
extension, of any atomic or complex L-proposition P. Figure 2.9 shows how
the truth-functional operators of standard propositional calculus are com-
puted as Boolean functions:
All this is generally accepted and part of the standard foundations
of propositional logic. It is, however, also widely known, but less
widely publicized, that this seductive application of Boolean algebra to the

f (P) f (¬P) f (P ∧ Q) f (Q): 1 0 f (P ∨ Q) f (Q): 1 0

–
x ————— • —–——— + —–————
1 0 φ (P): 1 1 0 φ (P): 1 1 1

0 1 0 0 0 0 1 0

FIGURE 2.9 Frege’s application of Boolean algebra to propositional calculus

46 The Logic of Language

truth-functional operators of propositional logic requires an awkward ontol-

ogy, which must attribute real world existence not only to what are perceived
as objects of the kind normally accepted in extensionalist ontologies, but also
to Truth and Falsity (see Section 6.1 in Volume I). Standard model theory, in
particular, faces the uncomfortable consequence that any ‘world’ must be
defined as containing not only what are normally called objects and sets of
(n-tuples of) objects, but also the ‘abstract’ objects Truth and Falsity, a
curious intrusion of extreme, and ultimately unsustainable, Platonism in an
otherwise strictly nominalist universe.

2.3.3 Valuation space modelling: a formal definition

This blemish can be removed by a more direct application of Boolean algebra
to the propositional truth functions, without reading Boolean 1 as ‘truth’ and
Boolean 0 as ‘falsity’, and thus without having to take truth values as exten-
sions of L-propositions. Instead, the machinery of set theory, already seen to
be computable by the Boolean operators, can be put to good use by means of
the method of valuation space (VS) modelling introduced in informal terms
in Sections 3.3 and 2.2.
It is time now to be a little more precise as regards the notion of
VS-modelling. VS-analysis takes a logical language LL whose expressions,
L-propositions, are taken to underlie the sentences of a natural language.
Given one single set of reference assignments Ki (‘key’) for all referring
and indexical terms in the L-propositions of LL, VS-analysis assigns to each
L-proposition P a valuation space /P/, which is the set of admissible situations
under Ki, within the totality U of all admissible situations, in which P is true.
An ADMISSIBLE SITUATION is defined as a set of truth-value assignments under
any given Ki to all L-propositions of LL. If LL contains n logically and
semantically independent L-propositions, each of which allows for truth
(T) or falsity (F) independently of all the others, then, in a standard
bivalent system of truth and falsity, there are, under each Ki, 2n logically and
semantically independent valuations—that is, assignments of Tor F to each of
the L-propositions of the language. Each such set of assignments is called
a ‘valuation’ or ‘admissible situation’. Given a particular L-proposition P, its
VS /P/ is the set of admissible situations (valuations), under any given Ki, in
which P is assigned the value T. An L-proposition P is true just in case the
actual situation sitact is an element in /P/. Clearly, L-propositions formed
with the help of one or more truth-functional propositional operators
have their truth value assigned depending on the truth values of the compo-
nent L-proposition(s) and the way the truth function is defined.
 Logic: a new beginning 47

For those who are inclined to ask philosophical questions about what is
meant by ‘situation’ here, it is important to realize that one can very well
operate theoretically with a formally defined notion of ‘situation’ while
remaining uncommitted as to the precise status of ‘situations’ in the natural
ontology of speakers or, rather, cognizing humans. Although it would be
utterly unrealistic to assume that natural speakers have a fully elaborated
interpretation of set theory onto a system of valuation spaces at their
disposal when interpreting utterances, it is not at all unrealistic to assume
that they have an intuitive and not fully elaborated idea, probably beyond
the threshold of possible awareness and signalled by physical reactions in
the brain, of ‘possible truth in an overall set of situations’, and hence of
notions like (in)compatibility, contradiction, and necessary consequence.
Correspondingly, one may assume that notions like mutual exclusion,
partial intersection and proper inclusion of sets of situations are likely to be
cognitively real. In Chapter 3, a theory is proposed of how naı̈ve, uneducated
humans deal with the plural objects we call sets. In this NATURAL SET
THEORY, mutual exclusion, mutual partial intersection, and proper inclusion
are taken to be the basic-natural set-theoretical notions. If that is so, it
easily follows that their logical counterparts, contrariety, contradiction,
and necessary consequence (entailment), likewise have some form of psycho-
logical reality.
VS-analysis allows for a different interpretation of Boolean algebra onto
propositional logic, in that the set-theoretic functions are used as an interme-
diary. We remember that an L-proposition P, expressing the proposition p,
has associated with it the set of all admissible situations in which p is true—
that is, the valuation space (VS) of P, or /P/. Since set theory is computable by
means of the functions of Boolean algebra, this algebra is applicable to the
truth functions of propositional calculus, with valuation spaces as the sets
involved and the universe of all admissible situations U standing for the
universe of all objects OBJ. So we use the standard interpretation of Boolean
algebra onto set theory, which treats 1 as the universe OBJ of all objects and 0
as the null set
. Since the Boolean variables are taken to range over sets, set-
theoretic COMPLEMENT corresponds to Boolean complement, set-theoretic IN-
TERSECTION to Boolean multiplication, and set-theoretic UNION to Boolean
addition. This translates directly onto the truth functions of propositional
logic, with VSs as sets and the universe U of all admissible situations as OBJ,
in the following way:
48 The Logic of Language

Negation corresponds to set-theoretic complement:

–
for any L-proposition P: /¬P/ ¼ /P/
Conjunction (∧) corresponds to set-theoretic intersection (\):
for any L-propositions P and Q: /P ∧ Q/ ¼ /P/ \ /Q/.
Disjunction (∨) corresponds to set-theoretic union ([):
for any L-propositions P and Q: /P ∨ Q/ ¼ /P/ [ /Q/.
On the basis of the principle that when P is true, the actual situation sitact is
an element in the valuation space of P, or: sitact 2 /P/, the truth-functional
propositional operators ¬, ∧ and ∨ are now set-theoretically definable as
follows (for the sake of simplicity, the operators ∧ and ∨ are treated as binary,
rather than as n-ary, functions):
(2.3) P is true: sitact 2 /P/
–
¬P is true: sitact 2 /¬P/; therefore: /¬P/=/P/
P ∧ Q is true: sitact 2 /P/ \ /Q/; therefore: /P ∧ Q/= /P/ \ /Q/
P ∨ Q is true: sitact 2 /P/ [ /Q/; therefore: /P ∨ Q/= /P/ [ /Q/
Again, when P is necessarily false, /P/ ¼
; when P is necessarily true, /P/ ¼ U.
The method of valuation space modelling subdivides the universe U of all
admissible situations for specific L-propositions or classes of L-propositions.
Thus, in Figure 2.10a, U is subdivided for any arbitrary L-proposition P and
hence also for its negation ¬P, /P/, and=P= being mutually exclusive and
jointly equal to U. Similarly in Figures 2.10b and 2.10c, where U is subdivided
for any arbitrary pair of L-propositions P and Q, which are logically inde-
pendent in the sense that they can be simultaneously singly true, singly false,
both true, and both false. In Figures 2.10b and 2.10c the valuation spaces for P
∧ Q and for P ∨ Q are indicated in terms of the logically independent L-
propositions P and Q. The corresponding VS for each case is marked by
horizontal lines. Figure 2.10 thus shows diagrammatically how the logical
truth functions ¬, ∧ and ∨ are translatable into the set-theoretic functions of
complementðxÞ, intersection (\) and union ([), respectively.
One notes that, just as the set-theoretic functions apply to any arbitrary
set or pair of sets, no matter whether they are null or equal to OBJ, the
propositional truth functions apply to any arbitrary L-proposition P or pair of
L-propositions P and Q, even when they are necessarily true or necessarily
false. If P is necessarily true, so that /P/¼ U, its negation ¬P is necessarily false,
so that=P= ¼
. If either P or Q or both P and Q are necessarily true, then so
is P ∨ Q, since, in that case, /P/ [ /Q/¼ U. And if either P or Q or both P and
Q are necessarily false, then so is P ∧ Q, since, in that case, /P/ \ /Q/¼
.
 Logic: a new beginning 49

a. b. c.
/¬P/

/P/ /P/ /Q/ /P/ /Q/

U U U

/¬P/ /P ∧ Q/ /P ∨ Q/

FIGURE 2.10 Valuation spaces in U for standard ¬P, P ∧ Q, and P ∨ Q

The standard logical relations of equivalence (
), entailment (‘, ), con-
tradiction (#), contrariety (><), and subcontrariety (><) are likewise immedi-
ately expressible in the set-theoretic terms of valuation spaces, as is shown in
(2.4). One notes that contradictoriness (#) combines the conditions of con-
trariety (><) and subcontrariety (><).
(2.4) For all L-propositions P and Q:
a. P
Q iff /P/ ¼ /Q/
b. P ‘ () Q iff /P/  /Q/
–
c. P#Q iff /P/ [ /Q/ ¼ U and /P/ \ /Q/ ¼
(or: /P/ ¼ /Q/)
d. P >< Q iff /P/ \ /Q/ ¼

e. P >< Q iff /P/ [ /Q/ ¼ U
The standard set-theoretic relation of identity (¼) corresponds to the logical
relation of equivalence defined in (2.4a). Inclusion () corresponds to entail-
ment, as shown in (2.4b). Contradictoriness is defined by the set-theoretic
relation of complement, as in (2.4c). Contrariety is defined by the set-
theoretic relation of mutual exclusion, as in (2.4d). And subcontrariety corre-
sponds to a set-theoretic relation that has so far not been honoured with
a name in standard set theory but for which we invent the name of FULL UNION
([˚ ), defined, for sets A and B, as follows:
(2.5) FULL UNION: A [˚ B iff A [ B ¼ OBJ
It follows from (2.5) that, when A [ ˚ B holds, it will be impossible, for any
element o in OBJ, that o 2
= A and o 2
= B: o has to be an element in either A or B,
or in both. The logical counterpart of full union is subcontrariety, since two
L-propositions P and Q are subcontraries just in case it is impossible for both
P and Q to be false simultaneously.
50 The Logic of Language

The notion of LOGICAL INDEPENDENCE of two L-propositions P and Q is of

central importance. VS analysis makes it possible to define this notion in set-
theoretic terms. For pairs of sets of objects, set-theoretic independence means
that an arbitrary object o 2 OBJ can be an element in either set, or in both
sets, or in neither of the two sets involved. For L-propositions P and Q, logical
independence means that the actual situation sitact can be a member of /P/ but
not of /Q/, or of /Q/ but not of /P/, or of both, or of neither. Just as two sets
A and B that partially intersect (A O O B) and therefore do not severally or
jointly equal either OBJ or
are set-theoretically independent, we say that
the L-propositions P and Q are logically independent just in case /P/ and /Q/
partially intersect (/P/ OO /Q/) and therefore do not severally or jointly equal
U or
. Relations between sets have now become metalogical relations and
vice versa.
It is worth observing, however, that there are also differences, due to the
actual needs of the calculi involved. Thus, although ‘partially intersect’ is a
legitimate set-theoretic relation between sets producing truth under certain
conditions and ‘logically independent’ is likewise a legitimate relation be-
tween L-propositions producing truth under certain conditions, yet the
relation ‘logically independent’ plays no role in any calculus of entail-
ments—that is, in any logic. Logical (meta)relations are functional only
to the extent that they impair the mutual independence of the component
L-propositions, so that some form of entailment can be formulated. Similarly,
whereas subcontrariety is a well-known relation in propositional logic, its set-
theoretic counterpart—here called ‘full union’—has not even been given a
name in set theory, probably because it can easily be done without.

2.3.4 Satisfaction conditions of the propositional operators

Now that it has been shown that the propositional operators can be looked
upon as predicates, this perspective can be developed further. Although the
same descriptive format is used for ordinary lexical and for logical predicates,
it is not implied that there are no differences. As has been said, logical
predicates differ from other lexical predicates in that the satisfaction condi-
tions of the former, to the extent that they are part of a logical system, are
expressible in terms that allow for a formal calculus of entailments. In the case
of the standardly known logical constants, it is the purely mathematical terms
of set theory that allow for the formal calculus of entailments, but it is not
unthinkable that similar semantic properties in other sets of predicates allow
for regular logical entailment schemata as well.
 Logic: a new beginning 51

It should be borne in mind also that those predicates that fit into some
logical system may well have other semantic properties that transcend their
strictly logical character and have to do with the often confused or confusing
phenomena of reality and experience. Quantifying predicates like FEW or MANY
clearly have logical properties. But beyond these they also possess nonlogical
semantic properties, such as the implicit appeal to some standard with respect
to which it can be said of a set or collection that it has ‘many’ or ‘few’
members (see Section 1.6). This shows again that it is both useful and natural
to treat logical constants as predicates.
Following this lead we consider ¬, ∧ and ∨ to be predicates. That is, as far
as their role in propositional calculus is concerned, we say that ¬ assigns to a
(well-anchored and well-keyed) L-proposition P the property that sitact is a
member of the COMPLEMENT =P= of the associated valuation space, ∧ assigns to
the (well-anchored and well-keyed) L-propositions P, Q, R, S, . . . the prop-
erty that sitact is in the INTERSECTION of the associated valuation spaces—that is,
of /P/ \ /Q/ \ /R/ \ /S/ \ . . . , and ∨ assigns to the (well-anchored and well-
keyed) L-propositions P, Q, R, S, . . . , the property that sitact is in the UNION of
the corresponding valuation spaces—that is, of /P/ [ /Q/ [ /R/ [ /S/ [ . . . .
Using the notation introduced in Section 3.3 of Volume I for the specification
of the satisfaction conditions of predicates, we now write (2.6a–c), where
‘/P/þ’ stands for any set of two or more valuation spaces of corresponding
L-propositions:
—
(2.6) a. [[¬]] ¼ {P | sitact 2 /P /}
(the extension of ¬ is the set of all L-propositions P such that the
actual situation sitact is a member of the complement of /P/)
\+
b. [[∧]] ¼ {P+ | sitact 2/P/ }
(the extension of ∧ is the set of all sets of two or more L-propositions
P such that sitact is a member of the intersection of all /P/þ)
[+
c. [[∨]] ¼ {P+ | sitact 2/P/ }
(the extension of ∨ is the set of all sets of two or more L-propositions
P such that sitact is a member of the union of all /P/þ)
The propositional constants ¬, ∧ and ∨ are presented as predicates over sets
of (well-anchored and well-keyed) L-propositions, each such L-proposition
having a corresponding valuation space. Technically speaking, the predi-
cates ¬, ∧, and ∨ are thus functions from (sets of) L-propositions to truth
values. This makes them a specific kind of predicate: not first-order predicates
over individual objects but higher-order predicates over (sets of) L-proposi-
tions.
52 The Logic of Language

Pred S+
AND
S S

Pred NP Pred S
sleep Harry NOT
Pred NP
work John

FIGURE 2.11 L-proposition underlying Harry sleeps and John does not work

Assuming (simplistically) that ¬, ∧, and ∨ correspond to NOT, AND, and

OR, respectively, in English we adopt a modified version of the proposal
made in McCawley (1972) to show that these operators also occupy the struc-
tural position of predicates in L-propositional structure. We treat the argument
of a propositional operator as one term—a single S-term for NOT and a complex
Sþ-term for AND and OR—in an L-propositional structure which can be either
verb-initial (VSO) or verb-final (SOV) (the verb-initial order is used through-
out in the present book). Figure 2.11 shows the L-propositional structure
corresponding to the English sentence Harry sleeps and John does not work
(the operators of assertive speech act and present tense have been omitted).
The transformational grammar (see Seuren 1996) then transforms this structure
into the appropriate surface structure.

2.3.5 Satisfaction conditions of the quantifiers

2.3.5.1 Russellian quantifiers So much for the predicate status of the prop-
ositional operators. But how about the quantifiers of predicate calculus? Here
one must realize that the notion of quantifier was subjected to considerable
refinement during the second half of the twentieth century. In the earlier
perspective, quantifiers were functions of a unique kind, introduced by a
special rule and provided with a special, not altogether transparent, model-
theoretic semantics. Sentences like (2.7a,b) were translated into the language
of logic as (2.8a,b), respectively, where ‘8’ stands for all and ‘∃’ for some.5 Let
us call the quantifiers as defined in terms of this system RUSSELLIAN QUANTIFIERS:

5
The symbol ∃, for the existential quantifier (just like the symbols \ and [ for set-theoretic
intersection and union, respectively, and the symbol 2 for class membership, being the first letter of
 Logic: a new beginning 53

(2.7) a. All farmers grumble.

b. Some farmers grumble.
(2.8) a. 8x(Farmer(x) ! Grumble(x))
(for all objects x, if x is a farmer, then x grumbles)
b. ∃x(Farmer(x)∧Grumble(x))
(for at least one object x, x is a farmer and x grumbles)
This is, on the whole, well known and many readers will still have been
brought up in this tradition. What is less well known is the fact that the
Russellian quantifiers are open to an interpretation that treats them as unary
higher-order predicates. To see this, it is important to keep in mind that the
extension of a propositional function F(x), where F is a predicate and x is a
variable ranging over the total set of objects OBJ, is identical to the extension
of the predicate F: [[F(x)]] ¼ [[F]]—that is, the set of objects in OBJ that
satisfy the predicate F.
This is easily seen when one realizes that a variable is a symbol that is to be
rotated over all objects in its range. Thus, the variable x in the L-propositional
function Horse(x) selects each object in the total set of objects OBJ and tests
it for satisfaction of the conditions associated with the predicate Horse.
Those objects that satisfy the conditions of Horse are marked þ, those that
do not are marked . Now the set of all objects marked þ is precisely the set
of horses. For that reason, logicians generally follow Frege who stated that, for
any (unary) predicate F, [[F]] equals [[F(x)]]: the variable has no semantic
(model-theoretic) effect; all it does is indicate that the predicate in question
needs a term in the position of the variable to become an L-proposition.
Moreover, when propositional operators are applied to propositional func-
tions, the result is again a propositional function. Thus, F(x)∧G(x) is a
propositional function and its extension [[F(x) ∧ G(x)]] is the intersection
of [[F(x)]] and [[G(x)]], or [[F(x)]] \ [[G(x)]], or just [[F]] \ [[G]]. And F(x) !
G(x) is again a propositional function whose extension [[F(x) ! G(x)]] is the
——
union of the complement of [[F(x)]] in OBJ and [[G(x)]], or [[F (x)]] [
—
[[G(x)]], or just [[F ]] [ [[G]].

the Greek esti ‘is’), was introduced by Giuseppe Peano in the 1890s when he worked at his ‘Formulario’
project, intended to introduce a unified notational system for the whole of mathematics and logic. The
symbol 8, for the universal quantifier, originates with Gerhard Genzen, who introduced it in his
‘Untersuchungen ueber das logische Schliessen’ (Mathematische Zeitschr. 39 (1934) p. 178), as a proper
counterpart to ∃. An earlier notation for ‘for all x’, used in Whitehead and Russell (1910–1913) and also
in Quine (1952), was ‘(x)’.
54 The Logic of Language

Having said this, we can define the Russellian universal and existential
quantifiers as unary higher-order predicates—that is, as unary predicates over
sets of objects—in the following way:
(2.9) For all sets X in OBJ:
a. [[8]] ¼ { X | X ¼ OBJ}
(the extension of the predicate 8 is the set of all sets that equal OBJ)
b. [[∃]] ¼ { X | X 6¼
}
(the extension of the predicate ∃ is the set of all nonnull sets in OBJ)
Note that the format used for the specification of the logical predicates 8 and
∃ in (2.9a,b), and for the propositional operators in (2.6a–c), is again that
used for ordinary lexical predicates as defined in Section 3.3.2 of Volume I.
As before, the condition specified after the upright bar is the satisfaction
condition of the predicate in question. Technically speaking, therefore, the
Russellian quantifiers are treated as functions from sets of objects to truth
values.
Consider, for example, (2.8a). An application of the truth table of the material
implication shows that when (2.8a) is true, then [[Farmer(x) ! Grumble(x)]] ¼
OBJ, since the propositional function Farmer(x) ! Grumble(x) yields truth
for any arbitrary object in OBJ. (2.8a) is falsified by any object in OBJ that does
satisfy Farmer(x) but not Grumble(x). But if (2.8a) is true, then any object in
OBJ either does not satisfy Farmer(x) or it satisfies both Farmer(x) and
Grumble(x). Analogously for (2.8b), which is true just in case [[Farmer(x) ∧
Grumble(x)]] 6¼
. The predicate status of the Russellian quantifiers is thus
saved by the propositional operators of standard logic.
2.3.5.2 Generalized quantifiers During the 1950s, some logicians discovered
that the Russellian quantifiers are not satisfactory as logical translations of
the natural language quantifying words all and some. Besides the unnatural-
ness of the logical translations of quantified natural language sentences—a
perennial source of bewilderment for beginning students—the main problem
with them is the fact that the method of rendering quantifiers with the help of
propositional operators cannot be extended to other quantifiers such as
MOST or HALF. This blocks a unified analysis of natural language quantifiers
in logical terms, which means an always unwelcome loss of generalization.
Since there is an alternative analysis of the quantifiers which restores the
generalization, it would seem that that analysis is preferable.
The alternative analysis was presented by Barwise and Cooper (1981), who,
falling back on Mostovski (1957), proposed to incorporate the L-propositional
function ‘Farmer(x)’ into the quantifier in the following way:
 Logic: a new beginning 55

(2.10) a. 8x[Farmer(x)](Grumble(x))
(for all objects x such that x is a farmer, x grumbles)
b. ∃x[Farmer(x)](Grumble(x))
(for at least one object x such that x is a farmer, x grumbles)
A semantics can now be provided in set-theoretic terms. As has been shown,
L-propositional functions of the form F(x), where F is a predicate and x
is a variable ranging over OBJ, are interpreted as expressions denoting sets.
The set denoted by the incorporated L-propositional function, in this
case Farmer(x), is called the RESTRICTOR SET and the set denoted by the
remaining propositional function, in this case Grumble(x), is called the
MATRIX SET. The corresponding expressions are called the RESTRICTOR TERM
and the MATRIX TERM. Now the universal quantifier 8 can be said to require
for truth that the restrictor set be a subset of the matrix set, while the
existential quantifier ∃ can be said to require for truth that the restrictor set
and the matrix set share a nonnull intersection. This analysis is known as
the theory of GENERALIZED QUANTIFIERS, as it allows for a generalized treatment
of all natural language quantifiers.
The analysis provided by Barwise and Cooper can be simplified when one
realizes that the quantifiers 8 and ∃ express a relation between two sets, the
restrictor set and the matrix set. Therefore, the quantifiers are properly
regarded as two-place higher-order predicates—that is, as predicates over
pairs of sets, rather than over (pairs of) objects, as is the case with first-
order predicates. The notation of (2.10a,b) is now no longer desirable and
should be replaced by one in which the quantifiers are represented as pre-
dicates over two terms, each term denoting a set. This produces the following
translations in LL for the sentences (2.7a,b):6
(2.11) a. 8x(Grumble(x), Farmer(x))
(the set of farmers is a subset of the set of grumblers)
b. ∃x(Grumble(x), Farmer(x))
(the set of farmers and the set of grumblers intersect)

6
I deviate from the convention to put the restrictor term first and the matrix term second. My
reason for inverting this order is purely linguistic, not logical: it allows for the syntactic rule of OBJECT
INCORPORATION (OI), whereby the second (object) term is united with the predicate to form a complex
predicate as in, for example, 8x[F(x)](G(x)), or ‘for all x that are F: x is G’, corresponding to (2.10a).
The rule OI has a strong position in universal grammar, whereas SUBJECT INCORPORATION is only weakly
supported (Seuren 1996: 300–9). See also note 3 in Chapter 4.
56 The Logic of Language

The quantifiers are indexed for the variable they bind. Thus, (2.12a) translates
as (2.12b):
(2.12) a. All farmers groom some horse.
b. 8x(∃y(Groom(x,y), Horse(y)), Farmer(x))
In general, sentence types like those given in (2.13a–j) are translated as follows:
(2.13) a. ALL F is G 8x(G(x), F(x)) f. SOME F is NOT-G ∃x(¬G(x), F(x))
b. SOME F is G ∃x(G(x), F(x)) g. ALL NON-F is G 8x(G(x), ¬F(x))
c. NOT ALL F is G ¬8 x(G(x), F(x)) h. SOME NON-F is G ∃x(G(x), ¬F(x))
d. NO F is G ¬∃ x(G(x), F(x)) i. ALL NON-F is NOT-G 8x(¬G(x), ¬F(x))
e. ALL F is NOT-G 8x(¬G(x), F(x)) j. SOME NON-F is NOT-G ∃x(¬G(x), ¬F(x))
This opens a new window on internal negation, discussed in Section 2.2. It
is clear from (2.13a–j) that the negation, ¬, can be used externally, to
negate the whole following L-proposition, but also internally, to negate the
propositional function in the subject or matrix term: the propositional
function ¬G(x) denotes the set of all objects that do not satisfy the predicate
G. The language of predicate calculus, and in particular the use of a variable
to help denote the extension of a predicate, thus allows for the (internal)
negation of any propositional function in any position in an L-propositional
structure.
Normally in predicate calculus, however, only the negated matrix term is
used, not the negated restrictor term, although both forms of negation
are allowed in the formal language. This is because the negated restrictor
term is considered logically less interesting, whereas the negated matrix or
subject term plays a major role in predicate logic, owing to the duality of the
two quantifiers, as explained in Section 2.2. In general, internal negation is
defined as the negation of the matrix term.
The universal quantifier 8 can now be defined as assigning the property
that the restrictor set is a subset of the matrix set, and the existential quantifier
∃ as assigning the property that the two sets share a nonnull intersection.
We do not wish to claim that this is an exhaustive description of the linguistic
meaning of these two operators, nor even that this description, as far as
it goes, is a correct rendering of their purely logical meaning in the logic
of language. But it does capture the purely logical meaning of these two
quantifiers in standard modern predicate calculus (SMPC). A formal expres-
sion of these standard meanings in terms of generalized quantifiers is given in
(2.14a,b):
 Logic: a new beginning 57

(2.14) For all sets X and Y:

a. [[8]] ¼ { <Y,X> j X  Y }
(the extension of the predicate 8 is the set of all pairs of sets Y, X,
such that X is a subset of Y )
b. [[∃]] ¼ { <Y,X> j Y \ X 6¼
}
(the extension of the predicate ∃ is the set of all pairs of sets Y, X,
such that the intersection of Y and X is nonnull)
Again, the format used for the semantic specification of the logical predicates
8 and ∃ in (2.14a,b) is identical to that introduced in Section 3.3.2 of Volume I
for nonlogical lexical predicates, with the satisfaction condition specified after
the upright bar. Whereas in (2.9a,b) above the Russellian quantifiers were
treated as UNARY higher-order predicates, the generalized quantifiers of (2.14a,
b) are treated as BINARY higher-order predicates—that is, as functions from
PAIRS OF SETS of objects to truth values.
This makes it possible to analyse the quantifier MOST in a way analogous to
the standard quantifiers 8 and ∃. Sentence (2.15a), translated as the L-propo-
sition (2.15b), contains the binary higher-order quantifier MOST, which is
semantically described as in (2.16):
(2.15) a. Most farmers grumble.
b. MOST x(Grumble(x), Farmer(x))


(2.16) [[MOST]] ¼ < Y; X > jY \ Xj > jXj 2 (NB: jXj: the cardinality of
set X)
(the extension of the predicate MOST is the set of all pairs of sets Y, X,
such that the cardinality of the intersection of Y and X is greater than
the cardinality of X divided by 2)

A similar analysis for HALF is easily given. This means that there now is a
unified category of quantifiers analysed according to one generalized system.
Hence the name.
Figure 2.12 shows the L-propositional tree structure for sentence (2.17):
S1

Pred S2 S3
SOME x
Pred S4 Pred NP
NOT farmer x
Pred NP
grumble x

FIGURE 2.12 L-propositional tree structure of Some farmers do not grumble

58 The Logic of Language

(2.17) Some farmers do not grumble.

This sentence incorporates the unary propositional operator NOT into the
matrix (subject) term S2 of the quantifying predicate SOME (¼∃)—that is, as
an internal negation. The restrictor term S3 of the predicate SOME, takes
the set of objects that satisfy the predicate farmer, i.e. [[Farmer(x)]], as its
extension. (Syntactically, an L-propositional function F(x) requires a dom-
inating S-node, so that the tree structure is of the form S[F(x)]: S is the
dominating node; F is the predicate; x is the (variable) term.) So [[S3]] gives
the required X-set. But how about the Y-set, represented by S2? We have seen
that the propositional operator NOT (¼ ¬) requires the specification of
the VS of its argument-S. We know what is meant by /F(a)/, where a is
a referring term: /F(a)/ is the set of situations that make F(a) true. But
what is /F(x)/?
By definition, since x is a variable ranging over objects, /F(x)/ is a function
from objects to valuation spaces. That is, for each object æ(a) (the reference
value of the term a) in OBJ the L-propositional function F(x) produces
precisely one valuation space /F(a)/. The predicate NOT subsequently
produces truth or falsity for the S2-proposition ¬F(a) according to whether
sitact is or is not an element in /F(a)/. Thus, a truth value is produced for S2
for each single substitution of the variable x with an expression referring to
an object in OBJ, just as the L-propositional function Farmer(x) produces
a truth value for S3 for each single substitution of the variable x with the
name of an object in OBJ. Therefore, S2 gives the Y-set required for (2.17), in
this case the set of all those who do not grumble. The Y-set and the X-set
are now fed into the predicate SOME, so that the proper truth value can be
produced.
All this looks good, much better than the Russellian quantifiers. Other
approaches are possible as well, as shown in Section 4.3, where a modified
system of quantification is presented, the system of distributive quantifiers,
which incorporates into the semantics of the quantifiers the parallelism
between the propositional and the quantifying operators.
More importantly, further research into the discourse-semantic properties of
the quantifiers reveals that, although the logical form as specified for the
quantifier SOME can be used as input both to the grammar of the language in
question and to the discourse-incrementation procedure, the same does not
hold for the quantifier ALL. A sentence like All farmers grumble can, in its
normal interpretation, only be processed if the discourse already contains
the information that there are farmers, to which the new information that
all farmers grumble is then added. In fact, in the normal interpretation,
 Logic: a new beginning 59

a sentence like All farmers grumble is equivalent with All the farmers
grumble, which is also how this sentence is rendered in a large number
of languages other than English, such as French. Somehow or other,
therefore, the universal quantifier ALL must be connected with the definite
determiner the. There are also interpretations where this is not so, as, for
example, with the universal any as in Any doctor will tell you that smoking
is bad. In this case, a formalization with the help of the implication operator
IF . . . THEN seems called for, just as in the Russellian paraphrase of the universal
quantifier.
The discourse-sensitive interpretation of ALL complicates matters consider-
ably. It means, for one thing, that universally quantified sentences require an
underlying form which differs in important respects from the logical form
postulated in our logical system, which does not reckon with discourse
factors. As long as it is pure logic that we are concerned with, this complica-
tion will be ignored and we will simply keep up the fiction that the universal
quantifier runs parallel with its existential counterpart, as is the practice
followed in all logical systems in existence. The difficulties and uncertainties
surrounding the discourse-sensitive quantifier ALL will be discussed in Chap-
ters 8 and 9 of the present volume.
Meanwhile, we pass on to a discussion of the logical systems of proposi-
tional and predicate calculus, looking in greater detail at the logical properties
of external and internal negation in relation to the propositional operators
and the quantifiers.

2.4 Internal negation, the Conversions and De Morgan’s laws

2.4.1 The internal negation again
Logic comes off the ground as soon as internal negation is introduced.
Although the modern notion of internal negation, that is, negation over a
matrix term under a higher operator (small scope negation), was not available
to Aristotle, he still observed that there is a special significance to the
difference between an external and an internal position of the negation in
quantified sentences (L-propositions), and that this significance is lost in
nonquantified L-propositions with only definite terms.
Since this is a central element in logic, some precise comment is in order.
Consider the nonquantified negative sentence (2.18a), rendered in logical
language (LL) as (2.18b):
60 The Logic of Language

(2.18) a. The flag is not green.

b. ¬[Green(the x[Flag(x)])]
(it is not the case that the x such that x is a flag is green)
The definite determiner the x takes the propositional function Flag(x),
denoting the set of all flags, as argument and selects from that set the one
particular flag that the discourse happens to be about. Right from Aristotle’s
day, definite determiners have caused trouble to logicians, the reason being
that the function selecting one particular element from a set is not comput-
able by set-theoretic means (it is noncompositional) but depends on knowl-
edge of the discourse and state of affairs at hand. Russell speaks of ‘phrases
containing the’ as being ‘by far the most interesting and difficult of denoting
phrases’ (1905: 481).
Definite determiners thus make the interpretation of sentences dependent
on cognitive factors, something which twentieth-century logicians have either
denied or tried to ignore. In Russell’s extremely influential article (1905),
which marked the beginning of what Quine called ‘the programme of elimi-
nation of singular terms’, the word the is written out of the logic script and is
reinterpreted as a lexicalization of a complex quantificational construction.
Sentence (2.18a) is analysed there as (2.19a), which, when rewritten in terms of
generalized quantifiers, comes out as (2.19b):
(2.19) a. ∃x[Flag(x) ∧¬[Green(x)] ∧ 8y[Flag(y) ! y ¼ x]]
(there is an x such that x is a flag and x is not green and such that
for all y, if y is a flag, y is identical with x)
b. ∃x(¬[Green(x)], [Flag(x) ∧ 8y(¼(y,x), Flag(y))])
(the set of things that are not green and the set of flags f such that
whatever is a flag is identical with f have a nonnull intersection)
This reads as saying that there is a set of flags consisting of just one member
and that this set has a nonnull intersection with the set of things that are not
green, so that there is exactly one flag and this flag is not green. The sentence
should, therefore, be falsified by as little as the existence of a second flag,
whether green or not green.
This analysis, though logically sound, is an empirical abomination (for
arguments, see Section 10.1.2). It corresponds in no way to the actual use of
the definite article and other definite determiners in natural language. Yet it
has been the dominant, if not standard, analysis in circles of logicians of
language, philosophical logicians, and semanticists throughout the twentieth
century. Short shrift is what this analysis deserves. We keep to the analysis
presented in (2.18b), accepting as a fact of life that definite determiners are
 Logic: a new beginning 61

noncompositional and need cognitive input to do their work. (See Seuren

et al. (2001) for more comment.)
We recall from Section 1.5 that, in Quine’s terminology, sentences contain-
ing at least one referring or indexical term are occasion sentences, while
sentences with only quantified terms are eternal sentences. The point here
is that in negated occasion sentences without any quantifier it makes no
difference whether the external negation (over the whole L-proposition) is
interpreted in the normal way as external negation, or is incorporated into the
predicate, where it is a complement-taking operator within the total set
of objects OBJ. That is, sentence (2.18a) can be analysed not only as (2.18b)
but also as (2.20):
(2.20) [NOT-Green](the x[Flag(x)])
with the predicate [NOT-Green], denoting all those things that are not green.
In other words, the negated predicate is just another predicate, denoting the
complement in OBJ of the set denoted by the non-negated predicate. Now
there is a logical equivalence between ¬[G(a)] (with G as a predicate and a as
a term) on the one hand and [NOT-G](a), where [NOT-G] is the predicate, on
the other. The L-proposition ¬[G(a)] is true just in case sitact 2 /¬[G(a)]/, and
—— ——
therefore sitact 2 /G (a)/. Since the condition for sitact 2 /G (a)/ is that æ(a)—
the reference value of the term a—be an element in [[G]] (the set of objects
——
denoted by the predicate G), the condition for sitact 2 /G (a)/ is that r(a) 2
——
[ G]]. And since[[G]] ¼ [[NOT-G]], the condition that sitact 2 /G (a)/ is that r(a)
2 [[NOT-G]], which makes [NOT-G](a) true. Hence ¬(G(a))
[NOT-G](a) for
sentences without any quantifier.
Aristotle already saw this. Calling an affirmative property assignment
‘unification’ and a negative assignment ‘separation’, he says (On the Soul
430b1–4; see also De Rijk 2002: 253):
Falsehood always involves a unification. For even if you assert that what is in fact pale
is not pale you have included ‘not-pale’ in a unification. But it is equally possible to
call all these cases ‘separation’.

In his more logically oriented On Interpretation (17a32–18a12) he refines the

statement that ‘falsehood ALWAYS involves a unification’. There he shows that
the process of lexical incorporation of the negation into the main lexical
predicate is possible salva veritate only for occasion sentences whose subject
term stands under a definite determiner, such as (2.18a). When the subject
term is quantified, the negation can still be lexically incorporated into the
lexical main predicate, but now no longer salva veritate. In such cases there is
a change of truth conditions and hence of meaning.
62 The Logic of Language

Take, for example, the following sentences (never mind the plural):
(2.21) a. Some flags are green.
b. Some flags are not green.
c. No (¼ NOT-SOME) flags are green.
The negation of (2.21a) is not (2.21b) but (2.21c). This becomes clear when one
considers the logical analysis of the sentences concerned, given in (2.22a–c),
respectively:
(2.22) a. ∃x(Green(x), Flag(x))
b. ∃x(¬[Green(x)], Flag(x))
c. ¬[∃x(Green(x), Flag(x))]
The negation in (2.22c) is the EXTERNAL NEGATION, the negation in (2.22b) is
the INTERNAL NEGATION. (2.22b) translates as (2.23), where the negation may
be seen as having been unified with the predicate Green of the propositional
function it stands over (its scope) into one single complex predicate NOT-
Green:
(2.23) ∃x([NOT-Green](x), Flag(x))
It is widely assumed that, analogously, the external negation of (2.22c) has
been unified with the main predicate of the sentence that forms its scope,
leading to (2.24), with the complex predicate NOT-SOME, lexically realized as no
in English:
(2.24) [NOT-SOME x](Green(x), Flag(x))
Yet although most natural languages have a single lexical item corresponding
to English no and requiring for truth a null intersection of the two sets
involved, closer analysis shows that such quantifiers are probably not in-
stances of lexical unification of NOT with the quantifier SOME but are quanti-
fiers in their own right in a basic-natural system of predicate logic that differs
from the standard system (for a full analysis see Section 3.5).
Occasion sentences with a definite subject term have no equivalent of
the internal negation of (2.21b). This is because a definite term, such as the
x[Flag(x)], does not consist of a propositional function, although it does
contain one. The whole definite term cannot be negated: anything like ¬[the
x[Flag(x)]] is uninterpretable (semantically ill-formed). One may object that
sentences like (2.25a) or (2.25c), which contain the phrase not the flag, are
perfectly possible. Here, however, these phrases are not used as terms but as
predicates of topicalized sentences corresponding to ‘what is green is not the
 Logic: a new beginning 63

flag (but . . . )’. In the logical analysis of such sentences it is useful to extend the
expression the x[Flag(x)] with an identifying copula verb be, as in (2.25d):
(2.25) a. It is not the flag that is green.
b. The FLAG is not green (but . . . )
c. Not the FLAG (but . . . ) is green.
d. ¬[[Be(the x[Flag(x)])](the x[Green(x)])]
In the logical analysis (2.25d) of topicalized sentences like (2.25a–c) the
definite subject term is the x[Green(x)] or ‘the thing that is green’, and the
predicate is [be(the x[Flag(x)])] or ‘be the flag’, unlike its nontopicalized
counterpart (2.18b), where the x[Flag(x)] is the logical subject term and
Green is the logical predicate. In nontopicalized sentences like (2.18a) there
is no way of ‘negating’ the definite term. Anything like Not the flag is green is
interpretable only as a topicalized sentence.
Yet, precisely because a definite term CONTAINS a propositional function,
some form of ‘internal’ negation is possible for the propositional
function inside the definite term. We have, for example, sentences like
(2.26a), analysed as (2.26b), or as (2.26c) with the negation over the proposi-
tional function Catholic(x) incorporated into the predicate (again, never
mind the plural):
(2.26) a. The noncatholics are angry.
b. Angry(the x[¬[Catholic(x)]])
c. Angry(the x[Noncatholic(x)])
Other than in occasion sentences with a definite subject term, the internal
negation typical for quantified sentences does negate a term, namely the
matrix term of the quantifying predicate, as in (2.22b). And this is possible
(semantically well-formed) because the terms of a quantifying predicate are
themselves propositional functions.

2.4.2 The Conversions and De Morgan’s laws

We can now proceed to a renewed discussion of the internal negation, this
time not only in predicate calculus but also in propositional calculus. Let us
keep, for the moment, to predicate calculus. As before, we use the following
abbreviations, where the asterisk signals the internal negation of the matrix
term G(x):
(2.27) A = 8(G(x), F(x)) (ALL F is G) A* = 8(¬[G(x)], F(x)) (ALL F is NOT-G)
I = ∃(G(x), F(x)) (SOME F is G) I* = ∃(¬[G(x)], F(x)) (SOME F IS NOT-G)
64 The Logic of Language

Since ¬[G(x)] is equivalent with [NOT-G](x), A* and A, as well as I* and I,

differ only in that A* and I* have a matrix term with a negative lexical
predicate, whereas A and I have a matrix term with a positive lexical predi-
cate. And since the identity of the lexical predicates is irrelevant to the logical
calculus, which is based exclusively on the logical constants (the Modulo-*-
principle), the logical properties of A* and A must be identical, as must those
of I* and I.
Given the definitions of 8 and ∃ as laid down in (2.14) above for gen-
eralized quantifiers and repeated here for convenience:
(2.28) For all sets X and B:
a. [[8]] ¼ { <Y,X> j X  Y}
b. [[∃]] ¼ { <Y,X> j Y \ X 6¼
}
the following theorems, known as the CONVERSIONS, hold for the generalized
quantifiers 8 and ∃ in standard modern predicate calculus:
(2.29) Conversions:
(a) A
¬ I* 8(G(x),F(x))
¬[∃(¬[G(x)],F(x))]
(ALL F is G
NO F is NOT-G)
(b) I
¬A* ∃(G(x),F(x))
¬[8(¬[G(x)],F(x))]
(SOME F is G
NOT ALL F is NOT-G)
The proofs are simple applications of set theory:
(a) left-to-right:
— —
If [[F]] [[G]], then [[F]] \ [[G]] = Ø. Hence not ([[F]] \ [[G]] ¼
6 Ø),
i.e. ¬[∃(¬[G(x)],F(x))].

right-to-left:
— —
If not ([[F]] \ [[G]]6¼ Ø), then[[F]] \ [[G]] = Ø. Hence [[F]]  [[G]].
i.e. 8(F(x),G(x)).

(b) left-to-right:
— —
If [[F]] \ [[G]] ¼
6 Ø, then not ([[F]]  [[G]]). Hence [[F]]
[[G]],
i.e. ¬[8(¬[G(x)],F(x))].

right-to-left:
—
If [[F]]
[[G]], then not [[F]] \ [[G]] = Ø. Hence [[F]] \ [[G]] ¼
6 Ø,
i.e. ∃ (G(x),F(x)).
As has been said, the relation between the quantifiers 8 and ∃ established by the
Conversions is commonly expressed by saying that they are each other’s duals.
 Logic: a new beginning 65

The Conversions of predicate logic also hold in propositional logic, where

they go under the name of De Morgan’s laws, after the nineteenth-century
London logician Augustus De Morgan. Analogously to (2.27), we define:
(2.30) AND = P ∧ Q ∧ R ∧ S ∧ … AND* = ¬P ∧ ¬Q ∧ ¬R ∧ ¬S ∧ …
OR = P ∨ Q ∨ R ∨ S ∨ … OR* = ¬P ∨ ¬Q ∨ ¬R ∨ ¬S ∨ …
As before, the asterisk stands for the internal negation, in this case distributed
over all members of the set Pþ, which is denoted by the syntactic subject (see
Section 4.3 for some comment). Again, the logical properties of AND and
AND*, as well as those of OR and OR*, are identical, since the identity of the
specific argument L-propositions is irrelevant to the calculus.
Given the definitions of ∧ and ∨ as given in (2.6b) and (2.6c) and repeated
here:
\+
(2.31) a. [[∧]] = {P+ | sitact 2/P/ }
(the extension of ∧ is the set of all sets of two or more propositions
p, such that sitact is a member of the intersection of all /P/þ, each P
being an L-proposition expressing p)
\+
b. [[∨]] = {P+ | sitact 2/P/ }
(the extension of ∨ is the set of all sets of two or more propositions
p, such that sitact is a member of the union of all /P/þ, each P being
an L-proposition expressing p)
the following theorems hold, which makes AND and OR each other’s duals:
(2.32) De Morgan’s Laws:
a. AND
¬OR* or: P ∧ Q ∧ R ∧ S ∧ . . .
¬(¬P ∨ ¬Q ∨ ¬R
∨ ¬S ∨ . . . )
b. OR
¬AND* or: P ∨ Q ∨ R ∨ S ∨ . . .
¬(¬P ∧ ¬Q ∧ ¬R ∧
¬S ∧ . . . )
Again, the proofs are direct copies of theorems in set theory:
(a) left-to-right:
If sitact 2 /P/ \ /Q/ \ /R/, then sitact 2
= =P=[= Q= [ = R/,
i.e. ¬[¬P ∨ ¬Q ∨ ¬R].
right-to-left:
If sitact 2 = =P= [ =Q= [ =R=, then sitact 2 /P/ \ /Q/ \ /R/,
i.e. P ∧ Q ∧ R.
66 The Logic of Language

(b) left-to-right:
If sitact 2 =P= [ =Q= [ =R=; then sit act 2
= =P= \ =Q= \ =R=,
i.e. ¬[¬P ∧ ¬Q ∧ ¬R].
right-to-left:
If sitact 2
= =P= \ =Q= \ =R=, then sitact 2 /P/ [ /Q/ [ /R/,
i.e. P ∨ Q ∨ R.
All this is normal standard logic, though admittedly presented with a slant
towards relativizing standard logic and placing it in a wider cognitive and
linguistic perspective. The question is: why is this of interest? The following
chapters begin to answer that question.
 3

Natural set theory and natural logic

3.1 Introductory observations

NATURAL SET THEORY, or NST, is a hypothesis, open to experimental testing. It is
a novel hypothesis which, therefore, has not, so far, been subjected to any
systematic experimental scrutiny and is based exclusively on what seems to be
plausible at an intuitive or pretheoretical level. As a consequence, the present
chapter can only be of an exploratory nature. Yet the underlying thought (a)
that humans have at their disposal a natural, cognition-based, set theory and a
predicate and propositional logic derived from it, (b) that this set theory and
this logic are likely to differ in important respects from standard set theory
and standard logic, and (c) that these differences may go a long way towards
explaining the differences between natural set-theoretic and logical intuitions
on the one hand and the corresponding standard systems on the other is,
though novel, robust and, it would seem, hard to counter.
NST describes how humans deal cognitively with plural objects. It
appears that NST lives by default or prototype and is thus open to correction
or overriding as a result of more precise thinking shaped, in particular, by
institutional education in more sophisticated societies. NST, like its
corresponding logic, is thus open to bootstrapping into higher levels of
precision and generality. This implies a gradient cline between two extremes
of cognitive achievement, which we name NATURAL SET THEORY and CONSTRUCTED
SET THEORY, respectively. However, since too little is known, as yet, to define
such a cline in a sufficiently precise format, I follow, for the moment, the easier
course of distinguishing different levels of (un)naturalness, in the expectation
that these distinctions, when more fully elaborated, will unfold into a system of
gradual distinctions. For now, I consider natural set theory to be manifest in
two forms: as BASIC-NATURAL SET THEORY (BNST), which stays within the limits
imposed by all defaults, and as STRICT-NATURAL SET THEORY (SNST), which
overrides one or more defaults. CONSTRUCTED SET THEORY overrides all defaults.
NST instantiates the general hypothesis that there are hidden reserves in the
mathematical, or, more generally, the formal powers of human cognition.
68 The Logic of Language

Inquisitive and formally creative humans are capable of cranking up the

current level of mathematical or, more generally, formal performance to a
higher degree of explicitness and formal precision by thinking through the
consequences of their formal operations. Prime examples in the context of
the Western world are, of course, Aristotle, Euclid, Al Huárizmi—who gave
his name to the term algorithm—and the many other creators of the modern
formal sciences known from history. When such efforts lead to new insights,
which are then widely accepted in a given society and integrated into the
educational system, we may consider this a measure of the degree of civiliza-
tion of the society in question.
We thus assume a gradient in the analytical powers of humans ranging
from the ‘rawest’ or most basic level of unsophistication to the most advanced
levels of abstract thinking as found in the centres of science. The degree of
achievement is taken to depend on the cultural, educational and other con-
ditions that trigger the use and development of available cognitive reserves at
the right age interval.
Some extremely interesting and challenging work has been done over the
past few decades regarding the arithmetical capacities of infants, schoolchil-
dren, and members of illiterate cultures.1 Pica et al. (2004) investigated the
arithmetical ability of speakers of Mundurukú, an Amazonian language
spoken by some 7000, mostly illiterate, Brazilian Indians. The language has
no numerals beyond five (the word for ‘five’ being the equivalent of ‘hand’, as
in many South-East Asian languages), reflecting the fact that the speakers
were unable to count beyond five and had great difficulty doing simple
arithmetical sums. Yet after some training, they quickly extended their count-
ing and computing abilities both cognitively and lexically, creating expres-
sions like ‘two hands’, even though their achievements never matched those of
humans born into culturally more developed societies.2
Butterworth states (1999: 7):
What makes human numerical ability unique is the development and transmission of
cultural tools for extending the capability of the Number Module. These tools include
aids to counting, such as number words, finger-counting, and tallying; and also the

1
For example, Ginsburg et al. (1984), Dehaene (1997), Butterworth (1999), Pica et al. (2004).
2
Pierre Pica, p.c. The fact that these speakers created new expressions to name numbers from
existing ones strongly suggests, if not proves, that it was not the availability of the lexical items that
enabled them to ‘think’ further along the number line but that it was in the first place their cognitive
development that required the new lexical expressions, which were then readily composed and which
probably helped them along in a secondary sense, in that the very availability of the expressions
enhanced performance.
 Natural set theory and natural logic 69

accumulated inventions of mathematicians down the centuries—from numerals to

calculating procedures, from counting-boards to theorems and their proofs.

Though Butterworth is no doubt right as regards the role of culture in the

development of human subjects’ arithmetical powers, one must fear that he
errs on the issue of uniqueness. Far from being unique to the human
numerical ability, the power to extend a naturally given ability through
cultural development appears to be much more general. Although the exact
boundaries of this phenomenon are unknown, it does seem that, along with
numeracy and, apparently, also the reading ability, ‘significant evolutionary
precursors [ . . . ] may exist for other currently understudied cultural domains
of human competence such as geometry, algebra, music and art’ (Dehaene
2005: 150–1; see also Dehaene et al. 2006). It thus does not seem too hazardous
to surmise that logic and its underlying set theory are also worthy candidates
to be considered. This is what is investigated in the present chapter.
The method followed here, however, differs from what is found in the
studies mentioned above, which are experimental, performance-driven, and
sometimes supported by neurophysiological evidence. Here, a hypothesis is
proposed and offered for empirical testing, whereby the data are allowed to be
largely intuitive. This method consists in observing how ‘ordinary’ speakers,
whose degree of literacy and cultural sophistication is considered ‘normal’ in
Western society and who distinguish themselves mainly by NOT being aca-
demically trained, interpret and use logical expressions—in particular the
logical constants—in their daily speech. An attempt is then made to reduce
the (well-known) differences with standard modern logic, which is based on
highly constructed mathematical set theory, to naturally given restrictions in
the human way of cognitively dealing with plural objects.
This difference in method is, to a large extent, motivated by the object of
research. In arithmetic or geometry, the technical aspects are relatively simple
and straightforward. As regards arithmetic, it doesn’t take much theory to
define a subject’s restricted arithmetical ability when this subject cannot count
or carry out arithmetical computations beyond five. Against the background
of our superior knowledge of arithmetic, such restrictions are quickly for-
mulated. In fact, we have just done so. It is also easy to understand that early
school training will trigger the growth and development of dormant arith-
metical powers in the brains and minds of young children who grow up in a
culturally developed environment. In fact, a moderately developed arithmeti-
cal ability is simply taken for granted in our society, so much so that we are
surprised to find out that individuals who grow up in a society without any
scholastic tradition are not even able to count beyond the very lowest
70 The Logic of Language

numbers and, therefore, lack words for higher numbers. We also have a
relatively good picture of what it has taken successive Chinese, Indian, Arabic,
and Western civilizations to come to the highly sophisticated science of
arithmetic that we have today. From the point of view of arithmetical theory,
therefore, no great obstacles present themselves in the study of basic-natural
arithmetic.
As regards logic, however, the situation is different. The very task of
formulating restrictions on the logical powers of unsophisticated humans is
the opposite of trivial. It requires considerable technical and theoretical effort
and insight, and standard logical lore fails to provide the tools for doing so.
The foundations and basic notions of logic are far less clear and well under-
stood than those of arithmetic, and the relation with natural cognition and
language is still as problematic as ever. It is hardly surprising, therefore, that
before one can pass on to any experimental work, one is forced to start with
an identification and formulation of the restrictions involved.
The much greater conceptual difficulty of logic compared to arithmetic is
borne out by the fact that what looks like the most firmly embedded logical
intuitions of the human race appear to support a logic that depends for its
application on complete situational knowledge, so that more developed
forms of logic were required for use in situations where knowledge is not
complete—a development that might explain the well-known discrepancies
between natural logical intuitions on the one hand and the concepts and
terms of the first, largely Aristotelian, ‘official’ logic on the other. Such
discrepancies do not occur in the case of arithmetic.
One may also look at our hypothesized basic-natural logic in the light
of the psychological theory of PROTOTYPES (see also Sections 8.6 and 8.8 in
Volume I). In general, prototypes seem to be characterized by the fact that
they maximize common features and thus avoid extremes or limiting cases.
Thus, it is proposed in Section 3.2.2 below that the first principle of BNST
consists in not taking into account the so-called extreme values—that is, the
null set (
), singletons (sets consisting of precisely one element) and the
totality of objects (OBJ). Perhaps one may, therefore, just as well call BNPC by
the name of prototypical logic. The problem is, however, that so little is known
about the conditions that make for prototypes. Frequency won’t do as a
criterion, as is shown in Section 8.8 in Volume I. But what will do is simply
to a large extent still a mystery, to do with hard-to-define notions such as
‘normal’ or ‘obvious’. Standard modern logic, for its part, can then be seen as
the result of the exploration of the extreme cases: when these are taken into
account, the basic-natural notions have to be sharpened. When one looks at
the question from this angle, it becomes clear why one will have to distinguish
 Natural set theory and natural logic 71

degrees of ‘naturalness’. This is precisely because the transition from intuitive

prototypes to well-defined concepts is gradual.
Anthropological linguists and language typologists, driven as they are by a
strong ecologistic motivation, will probably object that it is far from certain
that there IS a naturally given, innate ‘basis’ as part of the genetic endowment
of the human race and hence a basic-natural set theory or logic. It is
uncertain, they will say, whether all natural languages possess (equivalents
of) logical constants, in particular ALL, SOME, NOT, AND, and OR, and, if they do,
whether their meanings are the same in all natural languages.3 All I can say to
that is that there must be some genetically fixed cognitive substrate for dealing
with plurality and for the making of inferences involving plural objects—that
is, a predicate logic—or else the entire international machinery of bringing
education and other forms of development to underdeveloped parts of the
world would be built on quicksand. But, since our knowledge about these
matters is as underdeveloped as those parts of the world we treat to our aid
programmes, there is little else I can do now but make the famous inductive
leap and embrace the simplest and most general hypothesis that, until proven
otherwise, the naturally given lexical meanings of the logical operators under
scrutiny are universally fixed.4

3.2 Some set-theoretic principles of natural cognition

3.2.1 A résumé of standard set theory
It is important, at the outset, to emphasize the distinction between set-
theoretic functions that map n–tuples of sets (a) onto sets and (b) onto
truth values. The former are usually just called set-theoretic functions; the
latter are called relations, denoted by predicates, since predicates typically

3
Steve Levinson tells me that there are languages which use the same word for ‘many’ and ‘all’ and
also languages which use the indefinite article or the numeral meaning ‘one’ for ‘some’, referring to
discussions in Wierzbicka (1996: 74–6, 193–7). What this evidence means remains to be seen. Modern
Greek, for example, has the one word polı́ for both ‘very’ and ‘too’, making a phrase like polı́ megálos
ambiguous between ‘very big’ and ‘too big’ (though megálos by itself can also mean ‘too big’, just as
English late also has the meaning ‘too late’). Yet this does not mean at all that Greek speakers cannot or
do not distinguish between the concepts ‘very’ and ‘too’. They clearly do and, when pressed, they use
parapolı́ for ‘too’, even though parapolı́ still means ‘very much’ or ‘a whole lot’, but it seems to get closer
to ‘too’ than simple polı́.
4
The programme thus outlined in effect amounts to an attempt at replacing current Gricean
explanations for the disparity between logic and language in terms of generalized conversational
implicatures with an explanation based on natural set theory and the cognitive faculty of forming
mental propositions. If the objections raised in Section 1.3.3.2 of Volume I against attempted pragmatic
explanations along Gricean lines have any validity, this seems a worthwhile exercise.
72 The Logic of Language

denote functions from n–tuples of objects (or sets of objects) to truth values.
COMPLEMENT, INTERSECTION, UNION, and SUBTRACTION are set-theoretic functions,
but INCLUSION, for example, is a relation: for any given sets A and B, the binary
relation of INCLUSION, as in A  B, is either true or false. By contrast, the set
functions A  (complement), A \ B (intersection), A [ B (union), or A–B
(subtraction) do not have a truth value. Given the proper number of arbitrary
sets (one for complement; more than one for intersection and union; exactly
two for subtraction) they denote a new set defined by the Boolean functions
complement, intersection, union, and subtraction, respectively. Complement
is unique in that it involves the nonarbitrary set OBJ (the totality of all
objects) as part of its definition. It can be described as a special case of
subtraction, with OBJ–A as output for any set A. In logic, a further use of
the term complement is to denote a relation (‘be in complement with’)
yielding truth for two sets A and B just in case A [ B ¼ OBJ and A \ B ¼
.
The set-theoretic relations have a twofold use in the reduction of logic to set
theory. First, they correspond to (meta)logical relations expressed in terms of
valuation space (VS) analysis. For example, the inclusion relation translates
into a possible metalogical statement that, say, the set of situations /P/ is
included in the set of situations /Q/—that is, P entails Q (P ‘ () Q), which
is true or false depending on the meanings of P and Q, including the mean-
ings of any truth-functional propositional operators they may contain. Sec-
ondly, some set-theoretic relations correspond to quantifiers. For example, ALL
F is G translates, in principle, as saying that the set denoted by F is included
in the set denoted by G.
By contrast, the set-theoretic functions correspond to the propositional
logical constants of the object language LL as realized in any particular natural
language. Just as the set-theoretic functions take sets and deliver sets, the
propositional functions take valuation spaces and deliver valuation spaces.
For example, the operator AND in an L-proposition of the form P AND Q
delivers /P AND Q/—that is, the set of those situations that make P AND Q true,
corresponding to /P/ \ /Q/, the intersection of /P/ and /Q/. Figure 3.1 shows
how set-theoretic relations and functions are interpreted onto metalogical
relations, object-language quantifiers, and propositional operators.
The counterpart in metalogic of the set-theoretic relation MUTUAL PARTIAL
INTERSECTION (henceforth M-PARTIAL INTERSECTION, symbolized as A O O B: the
two sets A and B partially intersect each other and do not severally or jointly
equal either U or
, as in Figure 3.3b) is logico-semantic independence, which
plays no part in the machinery of logic: when /P/ and /Q/ M-partially
 Natural set theory and natural logic 73

metalogical relations: set-theoretic relations: quantifying operators:

contradictoriness complement

equivalence identity

contrariety mutual exclusion No F is G

logical independence partial intersection Some F is G
entailment inclusion All F is G /
Some F is G
subcontrariety full union

set-theoretic functions: propositional operators:

complement negation ("not")
union disjunction ("or")

intersection conjunction ("and"/"or")

subtraction "and/but not"

FIGURE 3.1 The reduction of metalogical relations and object-language operators to set-
theoretic relations and functions

intersect, the L-propositions P and Q are logically (and semantically) inde-

pendent in that the actual situation sitact can be in /P/ but not in /Q/, in /Q/
but not in /P/, in both /P/ and /Q/, or in neither /P/ nor /Q/. That is, no
entailment relation whatsoever holds between logically independent P and Q.
The relations complement, identity and full union lack a quantificational
counterpart in predicate calculus, perhaps because they lack logical interest or
perhaps because the makers of modern logic did not see far enough beyond
natural language. Finally, the set-theoretic function subtraction appears to
lack a single-morpheme propositional counterpart in most languages, but is
expressed as and/but not in English: the valuation space of an L-proposition of
the form P BUT NOT Q is the VS of P minus the VS of Q, or /P/ – /Q/.
Before a start is made with specifying the restrictions we intend to impose
on standard set theory so as to slim it down to natural set theory, taken to
reflect the way human cognition deals with sets, we will, for the sake of
convenience, summarize the standard definitions of the functions and rela-
tions of mathematical set theory. The standard Boolean functions of comple-
ment, union, intersection and subtraction are defined as follows (x ranges
over elements in OBJ):
74 The Logic of Language

(3.1) Standard Boolean functions on sets A and B:

–
a. A =def the set Z such that for all x 2 Z, x 2
= A and for all x 2
= Z, x 2 A
b. A [ B =def the set Z such that for all and only x 2 Z, x 2 A or x 2 B
c. A \ B =def the set Z such that for all and only x 2 Z, x 2 A and x 2 B
d. A–B =def the set Z such that for all and only x 2 Z, x 2 A and x 2 =B
The standard relations are defined in (3.2) (x ranges over elements in OBJ;
FULL UNION has been added for good measure):
(3.2) Standard relations between sets A and B:
– –
a. COMPLEMENT 8x((x 2 A ! x 2= A ) ∧ (x 2
= A ! x 2 A ))
b. IDENTITY 8x((x 2 A ! x 2 B) ∧ (x 2 B ! x 2 A))
c. MUTUAL EXCLUSION 8x(x 2 A ! x 2
= B)
d. M-PARTIAL INTERSECTION ∃x(x 2 A ∧ x 2 B) ∧ ∃x(x 2 A ∧ x 2= B) ∧
∃x(x 2
= A ∧ x 2 B) ∧ ∃x(x 2 = A ∧ x2
= B)
e. INCLUSION of A in B 8x(x 2 A ! x 2 B)
f. FULL UNION 8x(x 2 A ∨ x 2 B)
These relations can also be defined in terms of the standard Boolean
functions:
(3.3) Standard relations between sets A and B:
a. COMPLEMENT iff A [ B = OBJ and A \ B = Ø
b. IDENTITY iff A = B
c. MUTUAL EXCLUSION iff A \ B = Ø
– –
d. M-PARTIAL INTERSEC- iff A \ B 6¼ Ø; A \ B 6¼ Ø; B \ A 6¼ Ø;
TION A [ B 6¼ OBJ;
e. INCLUSION of A in B iff A [ B 6¼ Ø
f. FULL UNION: iff A [ B = OBJ

3.2.2 The restrictions imposed by NST

So much for standard set theory and standard logic. The question is now:
what restrictions are to be imposed on this system so that the discrepancies
between logic and language are correctly predicted? To begin with, let it be
assumed that NST entails that the mind, though naturally capable of proces-
sing and operating with sets or ‘plural objects’, does not naturally represent a
set as either the null set (
) or the universe of all objects (OBJ). It also seems
unnatural, for unsophisticated humans, to regard singletons, or sets consist-
ing of just one element, as ‘sets’, as is borne out by the well-known difficulty of
explaining to beginning students of set theory the difference between, say, the
individual object a and the set {a}. As is shown in Section 8.1.2, for natural
cognition sets are per se plural sets, whose cardinality runs from 2 upward.
 Natural set theory and natural logic 75

This means that, in NST, the null set is not a set at all: the cognitive
counterpart of
is the absence of any set, something which is cognitively
real and may be called ‘null’ but cannot play the role of a set. Whereas ‘null’
still functions cognitively as ‘absence of a set’, the opposite notion of OBJ, as
known in standard modern set theory, is typically the product of advanced
mathematical and/or philosophical thinking and has no place in natural set
theory. It is too nondescript to be cognitively real to formally untrained
minds. What does seem to play a role is the notion of RESTRICTED UNIVERSE OF
R
OBJECTS or OBJ , involving the totality of all objects within a contextually
defined universe of discourse. Therefore, all standard set-theoretic definitions
involving OBJ should, for natural set theory, be redefined as involving the
notion of OBJR, which does count as a natural set. Incidentally, this strategy is
chosen also by many mathematically-minded logicians, especially the earlier
ones. Thus we read (De Morgan 1847: 37–8):
But the contraries of common language usually embrace, not the whole universe, but
some one general idea. Thus, of men, Briton and alien are contraries: every man must
be one of the two, no man can be both. Not-Briton and alien are identical names, and
so are Not-alien and Briton. The same may be said of integer and fraction among
numbers, peer and commoner among subjects of the realm, male and female among
animals, and so on. In order to express this, let us say that the whole idea under
consideration is the universe (meaning merely the whole of which we are considering
parts) and let names which have nothing in common, but which between them
contain the whole idea under consideration, be called contraries in, or with respect
to, that universe.

Given these assumptions, we now posit the first principle of natural set
theory, PNST–1, which applies to single sets:
PNST–1:
, OBJ, AND SINGLETONS ARE NOT NATURAL SETS
Sets are never cognitively represented as having an EXTREME VALUE—that
is, as the null set (
) or as the totality of objects (OBJ). Nor are they
represented as containing just one element.
Sets that are neither
nor OBJ nor a singleton are called ‘natural sets’. PNST–
1 expresses the fact that NST is a theory of plural objects.
In the absence of any experimental data, and hence of any precise scale
of naturalness, we posit hypothetically that PNST–1 is both basic-natural and
strict-natural in that it strongly resists intellectual construction—perhaps
to different degrees for
, OBJ, and for singletons. Only at a much more
advanced level will the cognitive powers of the human race be able to override
PNST–1.
76 The Logic of Language

A further principle of natural set theory requires the nonidentity, or

distinctness, of any two or more natural set representations. When two or
more sets are cognitively represented, they are naturally taken to be distinct
from each other: any relation of identity is considered to be of a higher
level. This gives the second principle of natural set theory, which eliminates
the relation of identity:
PNST–2: NATURAL SETS ARE DISTINCT
When sets are distinguished cognitively, they are represented as being
extensionally DISTINCT, differing as regards their membership. There is
no basic-natural cognitive relation of identity between sets.
Among other things, PNST–2 accounts for the fact that natural intuition
reads ‘inclusion’ as ‘proper inclusion’ (), and not as ‘included in or identical
with’ ().5 This principle is, again, taken to be valid at both the basic-natural
and the strict-natural level, though it seems to be more easily overridden
in the sphere of object-language predicate logic than in the much more
abstract metatheory of logic where the metalogical relation of entailment
occupies a central place. Thus, whilst it seems relatively easy to gain
the insight that ALL F IS G is true when [[F]] ¼ [[G]] (in a logic where ALL F
IS G is defined as [[F]]  [[G]]), it seems a great deal harder to convince one
that every sentence P entails itself, which amounts to saying that /P/  /P/. It
takes a considerable amount of advanced analytical thinking—more, appar-
ently, than natural cognition and common natural language can bear—for the
latter to be accepted.
In actual fact, PNST–2 is more general. It applies to any kind of object
representation, not only of the plural objects we call sets but also of individual
objects: basic-natural cognition allows no two objects of any kind to stand in a
relation of identity, since that would make them one single object. Yet they do
allow for a mental operation of identification, given distinct levels of represen-
tation. This is what underlies the predicate of identity in natural language:
what was thought to be distinct becomes one (see Section 5.3.2 in Volume I).
Such an operation involves two levels, or stages, of cognitive representation,
one at which the two virtual objects or sets are distinct and one at which they
have merged into one. If truth is claimed for the latter, two virtual objects or
sets have been identified. If truth is claimed for the former, one virtual object
or set has been cognitively split up into two.

5
One notes that this is reflected in the lexicon in that the prefix sub- requires proper inclusion, not
identity: a subcontinent is part of a continent, a subsection is part of a section, and so on.
 Natural set theory and natural logic 77

Moreover, as is pointed out in Section 5.3.2 in Volume I, the identity

predicate, which identifies as one what were thought to be different objects,
is to be distinguished from the value-assigning predicate Bev, which assigns a
value to a parameter, as in My name is Pieter, where my name denotes the
parameter and Pieter denotes the value assigned to the parameter. Obviously,
Bev cannot be missed in a system containing functions. Therefore, Bev is
admitted, allowing for free substitution of either of its terms in any context.
We thus distinguish, wherever it is relevant, between the identity predicate Be
(¼) and the value-assigning predicate Bev written as ¼v (or v¼ when the
value is given first).
We tentatively assume three further principles specifying distinctness con-
ditions, all three less basic than PNST–1 and PNST–2, so that they can be
overridden at the strict-natural level. They apply to the set-theoretic functions
of UNION, SUBTRACTION, and INTERSECTION and to the set-theoretic relation of
INCLUSION:

PNST–3: BASIC-NATURAL UNION REQUIRES TOTAL DISTINCTNESS OF NATURAL SETS

When two (or more) sets A and B undergo UNION, A and B are natural
sets and are, at the level of basic, but not strict, naturalness, TOTALLY
DISTINCT, with no element in common, so that jA [ Bj ¼ jAjþjBj (‘jXj’:
the cardinality of set X).
Union is thus defined, at the level of basic naturalness, only for any two (or
more) totally distinct natural sets. When this condition is not fulfilled, there
is, at that basic level, no union. (As is shown in Section 3.4, PNST–3 accounts
for the exclusive nature of natural language OR. When PNST–3 is overridden,
making it possible for A and B to intersect, exclusive OR is upgraded to
inclusive OR.)
PNST–4 is the obverse of PNST–3, in that it requires total (proper)
inclusion for the basic-natural function of subtraction:
PNST–4: BASIC-NATURAL SUBTRACTION REQUIRES PROPER INCLUSION OF
NATURAL SETS
When A is subtracted from B, A and B are natural sets and, at the
level of basic but not strict naturalness, A is a proper subset of B so that
jB – Aj ¼ jBj–jAj 6¼ 0.
Again, this restriction is overridden at the strict-natural level. Subtraction
is thus taken to be defined, at the level of basic naturalness, for any two
natural sets A and B such that, if A is subtracted from B, the remainder is a
proper nonnull subset of B. When this condition is not fulfilled, there is, at the
78 The Logic of Language

basic-natural level, no subtraction. At the strict-natural level, however, one

may subtract a set of ginger cats from a set of male cats and find that no male
cat remains.
The conditions expressed in the last two subprinciples have been axioma-
tized in standard arithmetic for the calculus of cardinality. They ensure
that, under the principles of basic-natural set theory, Boolean addition and
subtraction agree with their arithmetical namesakes as regards the cardinality
of the sets involved.
By contrast, the set-theoretic function INTERSECTION, like its Boolean coun-
terpart MULTIPLICATION, does not correspond to any arithmetical function.
PNST–5 defines the natural intersection function as excluding both total
distinctness and proper inclusion, leaving only M-partial intersection of
natural sets:
PNST–5: BASIC-NATURAL INTERSECTION REQUIRES M-PARTIAL DISTINCTNESS
OF NATURAL SETS
When two (or more) sets A and B undergo INTERSECTION, A and B are
natural sets and are, at the level of basic, but not strict, naturalness,
M-partially distinct, with some, but not all, elements in common.
Basic-natural intersection is thus defined only for sets A and B (and possibly
more) such that (a) A \ B 6¼
and (b) A \ B 6¼ A 6¼ B 6¼ OBJR. (One
remembers that the symbol OO, or M-partial intersection, is used for the
combination of these two conditions: X OO Y just in case X \ Y 6¼
6¼ X 6¼
Y 6¼ OBJR.) Condition (a) cannot be overridden at any level of set-theoretic
or logical sophistication, but condition (b) is overridden at all levels above
basic. (This easy overriding allows for the upgrading of SOME excluding ALL to
SOME including ALL, as is shown in Section 3.4, and hence for the subaltern
entailment schema from ALL F IS G to SOME F IS G.)
Thus restricted, the natural intersection function ensures set-theoretic inde-
pendence. That is, if A \ B has a value under PNST–5, then A and B are set-
theoretically independent and vice versa. It is then ensured that, for any
element o 2 OBJR, it is possible that o 2 A and o 2 = B, or o 2 B and o 2 = A,
or o 2 A and o 2 B, or o 2 = A and o 2= B.
There is a final principle, relating to the recursive application of the set-
theoretic functions. Whereas in mathematical set theory one can apply the
functions recursively to one’s heart’s content, this is not so in natural set
theory. This applies especially to the function COMPLEMENT, which is taken to
be nonrecursive in basic-natural, and only once-recursive in strict-natural set
theory:
 Natural set theory and natural logic 79

PNST–6: BASIC-NATURAL COMPLEMENT IS NONRECURSIVE AND IS DEFINED AS

THE RESULT OF BASIC-NATURAL SUBTRACTION
At the level of basic naturalness, COMPLEMENT is restricted to one
application. Further, recursive applications are either strict-natural or
–
constructed. Moreover, ‘X ’ is defined as Z – X, where X and Z are
natural sets and Z functions as a provisional, restricted universe of
objects (OBJR).
Psychological experiments may disconfirm all or some of this. If they do not,
one should expect them to provide greater clarity and greater precision.

3.3 Consequences for set-theoretic and (meta)logical

relations and functions
3.3.1 Consequences for set-theoretic relations and functions
Formally, basic NST amounts to an application of Boolean algebra as
defined in Section 2.3.2, whereby the restrictions formulated in (3.4a–f)
hold. In (3.4a–f) a notational distinction is made between the standard
Boolean operators and their basic-natural counterparts, which are provided
with the superscript ‘BN’ (basic-natural). The BN-operators are defined in
terms of the standard operators.
(3.4) PRINCIPLES OF BASIC-NATURAL SET THEORY:
a. PNST–1: 0 and 1 are excluded as values of the Boolean variables.
b. PNST–2: the relations ¼ and 6¼ are eliminated; only the value-
assigning relation ¼v (v¼) and its negative counterpart 6¼v (v6¼) are
admitted.
c. PNST–3: ADDITION (x þBN y) is restricted to x, y such that there is no
·
z such that x y ¼v z.
·
d. PNST–4: SUBTRACTION (x –BN y) is restricted to x, y such that x y ¼v
·
y; x y 6¼v x6¼ v
.
·
e. PNST–5: MULTIPLICATION (x BN y) is restricted to x, y such that x y ·
·
6¼v x 6¼v y; there is a z such that x y ¼v z.
f. PNST–6: COMPLEMENT (x̄ BN) is non-recursive and restricted to an
independently given element y in the range of the variables such
that there is a z such that y –BN x ¼v z.
Thus restricted, Boolean algebra is translatable into basic NST as follows:

·
(3.5) a. BN is interpreted as basic-natural intersection \BN.
b. þBN is interpreted as basic-natural union [BN.
80 The Logic of Language
– –
c. X BN is interpreted as restricted complement X R.
d.  is interpreted as basic-natural subtraction BN.
BN

This, in its turn, is interpretable onto a logical system when the logical
constants are defined in set-theoretic terms. ALL, SOME, and NO are defined in
(3.7a–c), where A stands for ALL F is G, I for SOME F is G (
SOME F is NOT-
G), and N for NO F is G. For internal or ‘subsentential’ negation, as in ALL/
—R
SOME/NO F is NOT-G, [[G]] is to be replaced with [[G]] , the restricted
complement of [[G]] in any OBJR. The external negations of sentences with
ALL, SOME and NO are defined in (3.7d–f). One notes that (3.7c,e) show that ¬I
does not equal N, since the conditions of N exclude [[F]]  [[G]], whereas those
of ¬I allow for [[F]]  [[G]] (namely, when sitact 2 /A/). Basic-natural proper
inclusion () is defined, in terms of NST, as in (3.6).
(3.6) A  B is true iff there is a natural set C such that BBN A ¼v C.
(3.7) a. ALLF is G true [[F]]  [[G]]
(A) iff
b. SOME F is true [[F]] \ [[G]] 6¼v Ø 6¼v [[F]] 6¼v [[G]]; [[F]], [[G]]
G (I) iff 6¼v OBJ or: [[F]] OO [[G]]
c. NO F is G true there is no set H such that Hv= [[F]] \BN [[G]]
(N) iff
–
d. ¬A true sitact 2 /A/R in UR: sitact 2 /I/ or /N/
iff
–
e. ¬I true sitact 2 /I /R in U R: sitact 2 /A/ or /N/
iff
–
f. ¬N true sitact 2 /N /R in U R: sitact 2 /A/ or /I/.
iff
The formal sketch given in (3.4) and (3.5) requires some comment, first as
regards the question of psychological plausibility. Most will agree that it
would be utterly unrealistic to assume that natural speakers have a fully
elaborated interpretation of set theory onto a system of valuation spaces at
their disposal when interpreting utterances. Yet it is not at all unrealistic to
assume that they have a vague, intuitive idea, probably beyond the threshold
of possible awareness, of ‘truth in an overall, possibly infinite, set of situa-
tions’, and hence of notions like necessary consequence, (in)compatibility,
and contradiction. It does not seem to matter, for natural cognition, whether
a set is finite or, technically speaking, infinite, as the notion ‘very large’
appears to cover both ‘infinite’ and ‘very large but finite’. One may perhaps
even speculate that the formal precision of set-theoretic notions in natural
cognition is commensurate with their closeness to the psychological ‘ego’.
 Natural set theory and natural logic 81

Infinite and other very large sets would thus become increasingly ‘misty’ to
the mind as they are considered from a greater distance.
Correspondingly, one may assume that notions like mutual exclusion,
M-partial intersection and proper inclusion of sets of situations are likely to
be psychologically real, these notions being defined without any appeal to the
extreme boundaries of the set-theoretic system, namely the null set
and the
totality of all objects OBJ. Since mutual exclusion, M-partial intersection and
proper inclusion are the key notions in both NST and its application to
natural logic, it seems reasonable to assume psychological reality for both
NST and natural logic.
From a more formal point of view, we start our comment with restricted
complement as a function and as a relation between a set A and its complement
–
A R. Neither the function nor the relation are current in standard set theory, yet
they are of central importance to the study of natural cognition and natural
–
language. The function, written as A R, is defined in (3.8a): it takes OBJR as given
in any situation and any set A as being properly included in OBJR (A  OBJR),
and it delivers OBJR–A. The corresponding relation RC between a set A and its
restricted complement B within OBJR is defined in (3.8b):
–
(3.8) a. A R ¼def the set B such that B v¼ OBJ R BN A.
b. RESTRICTED COMPLEMENT: RC(B,A,OBJ R) iff B v¼ OBJR BN A.
–
Thus, in Figure 3.2, A R equals OBJR –BN A (but remember that multiple
applications of this function are excluded in virtue of PNST–6). The relation
–
between A and A R (horizontal lines) corresponds to natural contradictoriness
–
(that is, within the restricted complement); that between A and A (vertical
lines) to the standard metalogical relation of that name.

A
R
A

OBJR

OBJ

–
FIGURE 3.2 The relation –
between the natural set A, its restricted complement A R, and its
standard complement A
82 The Logic of Language

The set-theoretic relation of identity has been eliminated in virtue of

PNST–2, as has been said. Full union (see Section 2.3.3) is likewise eliminated,
at least for basic naturalness: when A and B intersect it is eliminated in virtue
of PNST–3; when they do not, full union equals complement. More is said
about full union in a moment.
The result is that, when OBJ is replaced with OBJR, there are only four
possible basic-natural relations left between any two natural sets A and B:
mutual exclusion, M-partial intersection, and proper inclusion of A in B or of B
in A. They are defined, in standard terms, in (3.9a–d) and shown in Figure
3.3a,b,c,d, respectively:
(3.9) Basic-natural relations between natural sets A and B within OBJR:
a. MUTUAL EXCLUSION (A OO B) iff A \ B =v Ø
b. M-PARTIAL INTERSECTION (A OO B) iff A \ B 6¼v Ø
–
A \ B R 6¼v Ø
–R
B \ A 6¼v Ø
c. INCLUSION of A in B (A  B) iff B – A 6¼v Ø
–
A \ B R =v Ø
d. INCLUSION of B in A (B  A) iff A – B 6¼v Ø
–
B \ A R =v Ø

a. c.
B
A
A
B

OBJ R OBJ R

b. d.
A

A B B

OBJ R OBJ R

FIGURE 3.3 Mutual exclusion, M-partial intersection, and proper inclusion of A in B

and of B in A as the four basic-natural relations between sets A and B
 Natural set theory and natural logic 83

Inclusion has now been reduced to proper inclusion, as the difference

hinges on the identity of A and B. This is precisely what is needed, since the
inclusion relation strikes nonmathematicians as nonnatural for two identical
sets. Moreover, intersection has been reduced to M-partial intersection
because when, in standard terms, A \ B ¼v
, there is no intersection, and
when A \ B ¼v A or A \ B ¼v B, there is (proper) inclusion but not
intersection, according to PNST–5 (though, as has been said, this latter
condition appears to apply only to basic, not to strict, naturalness).
Mutual exclusion has passed unscathed through the naturalness restric-
–
tions. In particular, they still allow for A and A R, as in Figure 3.2, to be called
mutually exclusive. What makes Figure 3.2 a special case of mutual exclusion
– –
is that the union of A and A R exhausts OBJR, or A [ A R ¼v OBJR.
The same freedom for the union of the two sets involved to exhaust OBJR
is, however, not granted to A and B when they M-partially intersect, as in
Figure 3.4a, or when the one is (properly) included in the other, as in Figure
3.4b. In those cases it may be true in standard terms that A [ B ¼v OBJR, but it
is not true under the restriction imposed by PNST–3, because PNST–3 leaves
basic-natural union undefined for cases where A and B are not totally distinct.
Full union, defined in (2.6) of Chapter 2 and repeated here as (3.10) in a form
adapted to NST, is thus equally undefined when A and B are not totally

distinct, even when OBJ is replaced with OBJR (‘[ R’ stands for full union
within any given OBJR).

(3.10) A [ R B iff A [BN B ¼v OBJR
 –
Whereas it is true, at the level of basic naturalness, that A [ R A R, as in

Figure 3.2, it is false that A [ R B for the sets A and B represented in Figures
3.4a,b, since A and B are not totally distinct. As for Figure 3.4a, the relation
between A and B lacks a name. In standard set theory it lacks one, presumably,
because it is mathematically uninteresting. In basic-natural set theory it lacks
one because no relation is seen there. Only in strict-natural set theory does it
deserve a name (‘full union’), because it corresponds to the metalogical
relation of subcontrariety, which is relevant in predicate calculus. And, as
regards Figure 3.4b, all that is involved is the relation of proper inclusion of A

in B. A [ R B thus delivers truth, in basic-natural set theory, for the natural
–
sets A and B only when B R ¼v A or vice versa—that is, when the one is the
restricted complement of the other, as illustrated in Figure 3.2. And since we
already have a term for both that relation and that function, the notion of
full union appears not to be needed in basic-natural set theory. It begins
to be needed as soon as it is realized that the sets A and B in Figure 3.4a are
84 The Logic of Language

a. b.
B = OBJR
A B
A

OBJR

FIGURE 3.4 Basic naturalness has no full union for not totally distinct A and B

not set-theoretically independent, which makes it desirable, when one de-

scribes the system, to reserve a name for the relation that causes this lack
of independence.

3.3.2 Consequences for (meta)logical relations and functions

The set-theoretic functions and relations can be translated into functions and
relations of logic and metalogic. As one reads from Figure 3.1, the set-theoretic
relations translate into metalogical relations. This is achieved by taking the
valuation spaces of L-propositions P and Q, /P/ and /Q/, respectively, as the
arguments of the relations and by making the universe of all admissible
situations U stand for OBJ. The set-theoretic relations also translate into the
standard operators of quantification in the object-language LL if [[F]] and [[G]]
are the arguments of the relation.
The set-theoretic functions are translated into propositional operators
(truth functions) in the object-language LL again by taking /P/ and /Q/ as
arguments and by making the universe of all admissible situations U stand
for OBJ. At the same time, a VS-model can be set up for the sentence types
ALL F is G, SOME F is G, and NO F is G, showing their metalogical relations,
and likewise for the sentence types of conjunction (P AND Q), disjunction
(P OR Q), and what we may call ‘exjunction’ (NEITHER P NOR Q).
Under the conditions imposed by NST, complement is modified
–
into restricted complement (A R), as defined in (8a), giving rise to a (presup-
position-preserving) restricted negation. PNST–6 excludes double negation.6
Double logically functional negation does occur, but only in culturally

6
Except, of course, when the negation is copied for the functional purpose of reinforcement, as in
the Cockney sentence ´E’s never been no good to no woman, not never.
 Natural set theory and natural logic 85

well-developed speech. Treble logically functional negation is rare and con-

fusing.7 Quadruple logically functional negation is out of the question.
At the level of basic naturalness, disjunction is restricted to nonnull valua-
tion spaces that do not intersect. Conjunction is reduced to M-partial inter-
section and is defined only over actually intersecting sets of situations
(valuation spaces) /P/ and /Q/ such that neither /P/ nor /Q/ equals
or
UR, as in Figure 3.3b. (Subtraction, or /P/ – /Q/, has no counterpart in
propositional logic; if it did it would be restricted to valuation spaces /P/
and /Q/ such that /Q/  /P/.)
As regards the relations, one remembers from (2.4) in Section 2.3.3 that the
standard metalogical relations of equivalence (
), entailment (‘ or ),
contradiction (CD), contrariety (C) and subcontrariety (SC) are expressible
as standard set-theoretic relations in terms of valuation spaces:
(3.11) For all L-propositions P and Q:
a.
(P,Q) iff /P/ ¼ /Q/
b. P ‘()Q iff /P/  /Q/
–
c. CD(P,Q) iff /P/ [ /Q/ ¼v U and /P/ \ /Q/ ¼v
(or: /P/ ¼v Q )
d. C(P,Q) iff /P/ \ /Q/ ¼v

e. SC(P,Q) iff /P/ [ /Q/ ¼v U
Natural set theory now imposes restrictions on these standard metalogical
relations. First, equivalence has to go: it does not exist in natural set theory as
a metalogical relation. Yet, as with the identification of objects in general,
there is a natural cognitive operation of identification, which takes two
sentences that have had different interpretative histories and identifies them
at some level of understanding. For example, one may say that Jack lives in
London is equivalent with Dr. Smith lives in London provided the expressions
Jack and Dr. Smith refer to the same person. Likewise, when we say that, in
basic-natural set theory, SOME F is G is equivalent with SOME F is NOT-G, what
we mean is that they have been identified at some level of theoretical inter-
pretation. Equivalence is, therefore, reinterpreted as identification.
Then, entailment of (3.11b), corresponding to standard inclusion as defined
in (3.9c) and redefined in terms of NST in (3.6), is a basic- and strict-natural
relation, provided it keeps to natural sets and to proper inclusion. The
counterintuitive notion that a necessary falsehood R entails any proposition
(‘ex falso per se ad quodlibet’) has been eliminated, because /R/ ¼v
, and

7
Or, as Larry Horn joked (Horn 1991: 98): ‘If Duplex Negatio Affirmat, we would predict that
Triplex Negatio Negat. […] But […] the geometric effect of the three negations is to motivate all too
often the more appropriate slogan Triplex Negatio Confundit.’
86 The Logic of Language

is not a natural set. Likewise, the equally violently counterintuitive notion that
a necessary truth S is entailed by any proposition (‘verum per se ex quolibet’)
has been eliminated, because /S/ ¼v U, and U is not a natural set. The concept
of naturalness introduced here further restricts the entailment relation in that
identity of /P/ and /Q/ is now also excluded, which rules out the counterin-
tuitive notion of self-entailment.
Entailments following from the theorem (‘inference rule’) of ADDITION have
now also been eliminated. Addition, one recalls from Section 1.2.2, is the
theorem saying that any L-proposition Q can be extended with ‘ ∨ R’ for any
arbitrary R. It seems clear that this theorem, though standard, should be
qualified as nonnatural, since natural speakers will not agree that, for exam-
ple, Joe is dead entails Joe is dead or today is Sunday. NST eliminates this
entailment. In standard terms, /Q ∨ R/ ¼ /Q/ [ /R/ and, because /Q/  (/Q/
[ /R/), one must accept that Q ‘ Q ∨ R for any arbitrary Q and R. NST helps
out, because in cases where Q and R are logically independent, so that /Q/
and /R/ M-partially intersect, basic-natural union excludes by definition
those situations where both Q and R are true. Therefore, the entailment
schema or inference rule of addition breaks down for exclusive OR.
A further ground for the elimination of addition as an inference rule lies in
the definition of entailment given in Section 1.2.1, which requires not only that
truth be preserved but also that this be determined by the specific linguistic
meaning of the entailing L-proposition. This latter condition is not satisfied
in cases of addition, as there is nothing in the meaning of any arbitrary Q that
causes truth to be preserved for Q OR R, R being equally arbitrary. Not so for
the inference rule known as SIMPLIFICATION, which says that P AND Q entails
both P and Q, since here it is the meaning of AND that causes the entailments.
Nothing much thus remains of addition.
All this taken together removes a great deal of counterintuitive excess
baggage and, in fact, restricts the entailment relation to semantically moti-
vated entailment, precisely as is wanted. Since logic has no term for the
entailment relation as restricted by NST and by the stipulation that entailment
is meaning-driven, the term NATURAL ENTAILMENT was suggested in Section 1.2.1.
It would seem that this reduction of entailment to natural entailment—that
is, to the set-theoretic relation of proper inclusion as restricted under NST and
supported by meaning—properly delimits the class of entailments felt to be
natural by native speakers, and hence empirically observable or measurable as
psychologically valid data.
Contradiction has been slimmed down to a contextually restricted UR,
created by presuppositional restrictions on the admissible situations in any
discourse at hand. Contrariety is the only relation that can stand unmodified,
 Natural set theory and natural logic 87

apart from the restriction of the VSs at issue to natural sets of admissible
situations. Finally, subcontrariety has disappeared from the basic-natural
system, as it involves the relation of full union, which has been ruled out.
Yet it reappears in the strict-natural system of metalogical relations, though
not without some considerable cognitive effort.8
It is now clear why in Figure 3.5, adapted from Figure 2.6, (a) is preferred
to (b). (Figure 2.6b, the improved Boethian square in two guises, is no longer
in competition since, in either form, the two component triangles are not
isomorphic.) Figure 3.5a consists of two logically isomorphic triangles with
the strict- (not basic-) natural relations entailment, contrariety, and contra-
dictoriness, subcontrariety being a ‘bonus’ due to the duality of the logical
constants defining P and Q, as defined in (3.12) (¼(2.2) of Section 2.2).
(Properly speaking, we should have a special negation sign for the comple-
ment within a restricted UR, but we leave this detail till Chapter 10, where
presuppositional logic is discussed. Note also that equivalence (
) is allowed
here, owing to the overriding in Aristotelian-Boethian predicate logic of the
basic-natural restriction disallowing identity.)
(3.12) P
¬Q* and consequently ¬P
Q*
Q
¬P* and consequently ¬Q
P*
By contrast, the two triangles of Figure 3.5b, though likewise isomorphic
and made up of three metalogical relations, have the less natural relation of
subcontrariety as a constitutive relation for the two triangles and contrariety
is the ‘bonus’ thrown in owing to duality. Thus, where Figure 3.5a has
contrariety, Figure 3.5b has subcontrariety and vice versa. This is why Figure
3.5a is considered (strict) natural, as against Figure 3.5b which does not fit into
any natural system.

a. P ¬Q* Q* b. P ¬P Q* : equivalents
CD CD
C: contraries
C SC
SC C
CD: contradictories
SC
C
SC: subcontraries

: entails

CD CD
Q ¬Q P* Q ¬P* P*

FIGURE 3.5 (a) the natural and (b) the nonnatural isomorphic square

8
Aristotle, with all his logical acumen, failed to identify it as a logical relation (see Section 5.3): it
was developed by his commentators. And beginning logic students, who still have to rely on their
natural intuitions, tend to find subcontrariety very hard to grasp, as logic teachers know well.
88 The Logic of Language

This leaves the following list of basic-natural and strict-natural metalogical

relations:
(3.13) For all L-propositions P and Q such that /P/ and /Q/ are natural sets:
a. CD(P,Q) iff /P/ [ /Q/ ¼v UR and /P/ \ /Q/ ¼v
(or:
/P/v¼ /Q/)
b. C(P,Q) iff /P/ \ /Q/ ¼ v

c. P ‘ () Q iff /P/  /Q/ (in virtue of the meanings of P and Q)
d. Q ‘ () P iff /Q/  /P/ (in virtue of the meanings of P and Q)
The missing relation, namely where /P/ \ /Q/ 6¼v
while both /P/6/Q/ and
/Q/ 6 /P/, and /P/ 6¼ /Q/ 6¼ U 6¼
has no corresponding name in logic, for
the simple reason that in such a case P and Q are logically independent.

3.4 The basic-natural systems of logic

So far, no great difficulties have been encountered. Problems arise when one
looks at the object-logical operators ALL, SOME, NOT, AND, and OR, and the way
they interact. As regards the operators of quantification, the only natural
relations that can hold between sets A and B, namely mutual exclusion (with
restricted complement as a variant), M-partial intersection, and proper in-
clusion, as shown in Figure 3.3, look, at first sight, as if they are directly
reflected in the quantifiers NO, SOME, and ALL, respectively. Likewise for the
propositional truth-functional operators, which look as if they are the direct
reflections of the corresponding set-theoretic functions as redefined under
NST. Yet when one tries to build a logical system—that is a system which
maintains consistency—on the basis of these parallels between natural sets on
the one hand and logical relations and functions on the other, one finds that
there are complications. The main complication consists in the fact that the
principles of natural set theory set out above require that SOME F is G be true
just in case [[F]] OO [[G]], whereas natural intuitions want it to be true not only
when [[F]] OO [[G]] but also when [[G]]  [[F]]—that is, just in case [[F]] \ [[G]] 
[[F]], as in Some children are orphans.
Before we embark on an analysis of basic-natural predicate logic and its
more advanced cousins defined by Aristotle, Boethius, Abelard, and Russell, a
clarification is needed regarding the supposedly monadic character of these
logics. By this is meant the fact that it looks as if these logics allow only for
quantification over the subject term, while other quantified terms in sentences
(as in All boys admire some football player) remain unaccounted for. These
logics are thus taken to be less expressive than modern predicate logic. It is
 Natural set theory and natural logic 89

generally assumed and widely taught that modern logic overcomes this
restriction and allows for quantification over any argument term in a sen-
tence. It should be noted, however, that this superior expressivity of modern
logic is due not to its logical properties but solely to the formal language in
which its expressions are couched. Since we use, or anyway can use, the same
formal language, with quantifiers, variables and all, for all the different
predicate logics concerned, they all have equal expressive power, as they can
all express quantification over any argument term in a sentence, embedding
one quantifier in the scope of another. Therefore, the fact that we restrict our
analyses to the monadic subject–predicate distinction is immaterial. We do
this only to keep the exposé within reasonable bounds of size and complexity.

3.4.1 Basic-natural predicate logic: the necessity of a cognitive base

Up till now we have assumed that basic-natural predicate calculus (BNPC) is
derivable in toto from basic-natural set theory (NST), just as standard modern
predicate calculus (SMPC) is derivable in toto from standard set theory. We
shall now see that this assumption is unwarranted. BNPC cannot reasonably be
the product of just NST: additional cognitive factors must be invoked. In fact,
we shall see, first, that BNPC cannot be constructed solely on a set-theoretic
base and requires a wider cognitive base in addition to NST. Then it is shown
that SMPC, which finds its apogee in the theory of generalized quantifiers, can
only be constructed in terms of set theory, resisting treatment in terms of wider
cognitive functions. It is in this sense that SMPC has managed to rid itself from
all psychological elements—the ideal that Russell was so keen to achieve. The
question is of special relevance as it shows that the reversal of Russell’s
programme of depsychologization of logic is subject to substantive empirical
constraints, which opens up a new and highly interesting area of research
in both psychology and logic. Finally, we will see that Aristotelian-Abelardian
predicate calculus or AAPC (about which more below), as well as Aristotelian-
Boethian predicate calculus or ABPC (best known as the Square
of Opposition), are ‘mixed’ in that they retain some but discard other non-
set-theoretic features. The question hinges on the semantic definition of the
quantifiers involved.
Obviously, this calls for some explanation. The first stumbling block for a
strictly set-theoretic construction of BNPC is the existential quantifier SOME,
which, for natural intuition, means ‘only some’ or ‘some but not all’. This
means, in set-theoretic terms, that SOME F is G is true just in case [[F]] OO [[G]]
or [[G]]  [[F]]. The condition that [[F]] OO [[G]] poses no problem with regard
to a direct derivation of BNPC from NST in terms of the sets [[F]] and [[G]]: a
90 The Logic of Language

sentence like Some flags are green is true when the set of flags partially
mutually intersects with the set of green things—as is the case in the actual
world. And the relation of M-partial intersection (OO) goes well with the
notion of natural set theory: it is one of the four basic-natural relations
between sets, as shown in Figure 3.3. One would thus expect SOME F is G to
be true just in case [[F]] OO [[G]]. But what we find in natural language is that
SOME F is G is true also when [[G]]  [[F]]. Consider sentences (3.14a–d),
which are obviously true, while [[G]] is properly included in [[F]]:
(3.14) a. Some children are orphans.
b. Some people are Englishmen.
c. Some computers are laptops.
d. Some heavenly bodies are planets.
Moreover, the converses of these sentences raise eyebrows, in that they give
rise to the question of whether one should conclude that some orphans
are not children, or some Englishmen are not people, and likewise for
(3.15c) and (3.15d):
(3.15) a. Some orphans are children.
b. Some Englishmen are people.
c. Some laptops are computers.
d. Some planets are heavenly bodies.
It is thus clear that, for natural speakers, IBN
I*BN, just as in Hamilton’s logic
which is discussed below.9
The problem is thus that, other than in the remaining three systems, the
existential quantifier in BNPC is non-symmetrical, whereas there is no mo-
tivated way in which this lack of symmetry can be said to follow from a
natural set theory. We may, of course, try to bend NST in such a way that the
BNPC existential quantifier follows directly from it, but one has to fear that
this will not bear experimental testing: the chances of such a natural set theory
being empirically adequate must be deemed minimal. The problem is the
more serious because the intuitions that come with the sentences in (3.14) and
(3.15) are robust and beyond reasonable doubt. In fact, the problem requires
an entire rethinking of quantification theory.
To this end, we distinguish between a COGNITIVE and a SET-THEORETIC ap-
proach to quantification. The latter requires just a (basic-)natural set theory.
The former requires a cognitive theory of how cognition deals with plural

9
Blanché (1966) uses the type name Y for IBN
I*BN I do not follow him in this respect, as I do not
want to make the formalism heavier than it need be.
 Natural set theory and natural logic 91

objects in addition to a (basic-)natural set theory. In a set-theoretic analysis,

quantifiers are treated as binary higher-order predicates over pairs of sets, in
the sense discussed in the previous chapter. A cognitive analysis of quantifica-
tion is formulated in terms of propositions in which a property, expressed as a
predicate, is assigned to a plural set (n>1). In this approach, conversion, in
the sense of a swap between the matrix and the restrictor predicate (indicated
by ! in the discussion below), makes no sense.10
In order to make clear what is meant by the cognitive approach, we
must dig a little deeper and look at the genesis of logical operators
generally. In the cognitive approach, we take it that the logical operators
originate as metalinguistic predicates. Thus, the negation operator has as
its cognitive origin the metalinguistic qualification ‘it is false that’ (FALSE).
Likewise, at the cognitive level, an assertion stands under the metalinguistic
operator ‘it is true that’ (TRUE), unexpressed (expressed as a zero element)
in most languages. Such metalinguistic operators can then, by a process that
may be called LOGICAL GRAMMATICALIZATION, be incorporated into the object
language, where they appear as object-language operators (or as zero).
FALSE thus becomes natural language negation. The advantage is greater
flexibility of the object language, in that the negation can then be used
(and become part of a semantic calculus) in any embedded L-proposition
or L-propositional function.11
Pursuing this idea a little further, we can apply it to quantifiers in the
following way. Given a nonquantified L-proposition with a definite plural
subject term, such as The flags are green, we can judge that it is either entirely
true, say VERUM IN TOTO (VIT), or partially true, say VERUM IN PARTE (VIP), or
not at all true, say VERUM IN NULLO (VIN):

10
Interestingly, Aristotle starts in the cognitive mood, as appears from his term en mérei (in part)
(and the Latin translation particularis) for existential quantification, and, as regards universal
quantification, when he says (Int 20a9–15; cf. also Int 17b-12):
For the word every does not make the subject universal but the whole proposition. … So that the words every or no
add nothing else to the meaning than that, whether affirmatively or negatively, the subject is to be taken as a whole.

Yet his syllogistic, set out in the Prior Analytics, is based on the device of letting a predicate (the
Middle Term) occur as a subject. In his syllogistic, therefore, Aristotle follows the set-theoretic
approach, in which both the restrictor and the matrix term represent sets.
11
In fact, it makes sense to regard the radical (presupposition-cancelling) negation, discussed in
Chapter 10, as having grammaticalized only in part. It has to be constructed, in surface structure, with
the finite verb of the main clause and is not allowed in any other, ‘noncanonical’ position. Moreover,
its semantics (echo-effect) shows that it takes a quoted L-proposition as its argument. It would seem
that this perspective on the radical negation deserves further reflection.
92 The Logic of Language

VIT [‘the flags are green’]

VIP [‘the flags are green’]
VIN [‘the flags are green’]

The ‘partial’ in VERUM IN PARTE is to be interpreted as saying that some but not
all flags are green, in accordance with the basic-natural set-theoretic notion
that sees class inclusion as proper class inclusion.
Clearly, this creates room for truth values between just true and false,
because VIP can be specified percentagewise: we can say that it is precisely
70% true that the flags are green, so that, given ten flags, exactly seven flags
must be green for truth to be attained. This aspect of the theory is not
elaborated here, but it is important to realize that the possibility of
an intervalent logic arises inter alia when predicates are assigned to plural
subject terms.
The three meta-operators VIT, VIP, and VIN can be transferred to the
object language, where they can be symbolized as 8BN (corresponding to
the L-propositional type ABN), ∃BN (corresponding to type IBN), and NBN
(corresponding to type NBN), respectively. The truth conditions for the types
ABN, IBN and NBN are as follows:
(a) 8BN [the Fs are G] is true iff all members of [[F]] are G
(b) ∃BN [the Fs are G] is true iff some but not all members of [[F]] are G
(c) NBN [the Fs are G] is true iff no member of [[F]] is G

(There may or may not be Gs outside [[F]])

(1) (2) (3) (4)

[[F]] = Ø

[[F]] [[F]] [[F]]

All F is G (Only) some F is G No F is G

FIGURE 3.6 The four possible situation classes in the cognitive approach
 Natural set theory and natural logic 93

Graphically, the truth conditions can be represented as in Figure 3.6, where

the shading indicates the property expressed by the predicate G. One notes
that these truth conditions are not expressed in terms of two sets but in terms
of one set [[F]], the extension of the plural subject term quantified over. When
[[F]] ¼
, BNPC does not apply, because the null set does not count as a set in
NST. (It is shown below that BNPC remains largely intact when the condition
[[F]] ¼
is taken into account.)
The internal negation, symbolized as *, results from placing the argument
proposition of VIT, VIP, or VIN under the negation operator:
VIT [‘¬[the flags are green]’] ! 8BN[¬[the Fs are G]] type: A*BN
VIP [‘¬[the flags are green]’] ! ∃BN[¬[the Fs are G]] type: I*BN
VIN [‘¬[the flags are green]’] ! NBN[¬[the Fs are G]] type: I*BN
One notes that the argument of VIT, VIP, and VIN is the quoted singular
sentence ‘¬[the flags are green]’, which is equivalent with ‘the flags are not-
green’, because ‘the flags are green’ is a nonquantified, singular sentence, with
a definite term as subject. This explains why Aristotle speaks of ‘negatives’
when what he has in mind is not the external but, in our terms, the internal
negation.
Having thus defined the types ABN, IBN and NBN, and their internally
negative counterparts A*BN , I*BN and N*BN, we can now, with the help of the
external negation based on the metalinguistic operator FALSE, define their
externally negative versions ¬ABN* , ¬IBN
* , ¬NBN
* , :ABN*; :IBN
* and:NBN * . The
valuation spaces, in terms of Figure 3.6, for these twelve sentence types are
now as follows:
/A*BN / = {1} /A*BN / = {3} /¬ABN
* / = {2,3} /¬A*BN / = {1,2}
/I*BN / = {2} /I*BN / = {2} /¬IBN
* / = {1,3} /¬IBN
* //= {1,3}
/N*BN / = {3} /N*BN / = {1} /¬N*BN / = {1,2} /¬N*BN / = {2,3}
Figure 3.7 shows the square representation resulting from the system
described, the VS-model (with the four spaces corresponding to Figure 3.6,
but with space 4 left idling) and the complete dodecagon of all logical
(meta)relations holding in BNPC. This system does not suffer from undue
existential import (UEI), although space 4 is inoperative. This is because A*BN
entails ¬I*BN , and ¬I*BN must be considered true in space 4, regardless of what is
done with regard to A*BN . But we can, if that is wanted, assign truth values to
all twelve types for the situation specified for space 4. When this is done, as in
Figure 3.8, it is not possible to assign truth to the types A and A* in space 4
94 The Logic of Language

without cognitive arbitrariness, as type A expresses the situation class of space

1 and A* the situation class in space 3. There is no good reason why space 4
should be added to these VSs.
One notes that, in BNPC with space 4 operative, I and I* are still equivalent,
but A and N*, and A* and N, no longer are: there is now a one-way
entailment from A to N* and from A* to N. Moreover, in this version of
BNPC, some relations are missing, so that the dodecagon is no longer
complete: it gets a bit undone in the right-hand side area. Yet as a logical
system it is sound. And one may marvel at its power and richness. Its only
drawback, as we see it, is that it does not allow one to say that Some flags are
green until all flags have been checked, because Some flags are green is false in
BNPC when all flags are. This nonlogical but merely functional defect has
been repaired in AAPC and ABPC, and of course also in SMPC, but at the
expense of considerable logical power, as will become clear in Chapter 4.
We can now take a further step forward on the path from basic- to strict-
natural predicate logic and derive Aristotelian-Abelardian predicate calculus
(AAPC). This we do by stipulating that AAPC equals BNPC with space 4
operative, as illustrated in Figure 3.8, except that the expression ‘part of ’ is
now interpreted in the modern standard sense that if set X is part of set Y, then
X  Y, and not X  Y, as in BNPC. Moreover, we let AAA be true also when all
and only all F is G—that is, when the extensions of F and G coincide. In
BNPC, ALL F is G is interpreted as ‘all but not only F is G’; in AAPC the
reading is ‘all and possibly only F is G’. (One already begins to see the need for
an analysis in terms of two sets.)
This gives the following valuation spaces for AAPC:
/AAA/ = {1} /A*AA / = {3} * //= {1,2,4}
/¬AAA/ = {2,3,4} /¬AAA
/IAA/ = {1,2} /I*AA / = {2,3} /¬IAA/ = {3,4} * / = {1,4}
/¬IAA
/NAA/ = {3,4} /N*AA / = {1,4} /¬NAA/ = {1,2} * / = {1,2}
/¬NAA
We now see that ¬IAA
NAA (and, accordingly, :IAA
N*AA
). This enables
us to eliminate the quantifier NAA, and with it the sentence type NAA, from the
system, in favour of ¬IAA and ¬IAA
* . Moreover, the classic subaltern entailments
from A to I and from A* to I* now hold. The Conversions do not hold, but there
is a one-way entailment from AAA to ¬IAA
* and from IAA to ¬AAA * . This gives rise
to the square representation, VS-model, and octagon of Figure 3.9.
Aristotelian-Boethian predicate calculus, or ABPC, best known as the
Square of Opposition, is obtained by illegitimately leaving out space 4
from the system. This is illegitimate because leaving out space 4 from AAPC
(which was not done by Aristotle, as is widely believed, but by his later
 Natural set theory and natural logic 95

a. b. [[F]] = Ø
A N* I*
C
¬A
C C
¬I ¬A N
C C I ¬N
A
¬I ¬N
4 3 2 1
N* ¬I*
C ¬N* ¬A* I*
I N A* ¬N* ¬I*
¬A*

A*
U
A {1}
c.
CD CD
C C
{2,3} ¬A ¬N* {2,3}
C

SC SC CD
SC CD

{2} I* C N* {1}

CD CD

SC C
SC C
{1,3} ¬I* SC ¬N (1,2)
SC SC
C
C SC CD
CD SC CD

{2} I C N {3}
C SC C

CD
CD

SC ¬A* {1,2)
{1,3) ¬I C
CD
A* {3)

FIGURE 3.7 The square, VS-model, and dodecagon of BNPC with space 4 idle
96 The Logic of Language

a. b. [[F]] = Ø
¬A
A N* I*
C ¬A
¬I N
C C ¬I ¬A N
I ¬N
C C A
¬I ¬N
4 3 2 1
N* ¬I*
¬N* ¬A* I*
C ¬N* ¬I*
I N A* ¬A*
N* ¬I*
A*
U
¬A*
A {1}
c.
CD C
C C
{2,3,4} ¬A ¬N* {2,3}
C

SC SC CD
SC

{2} I* C N* {1,4}

CD CD

SC
SC C
{1,3,4} ¬I* SC ¬N (1,2}
SC SC
C
C SC C
CD SC CD

{2} I C N {3,4}
C SC C

SC
CD

SC ¬A* {1,2,4}
{1,3,4} ¬I C
CD
A* {3}

FIGURE 3.8 The square, VS-model, and dodecagon of BNPC with space 4 operative
 Natural set theory and natural logic 97

a. A ¬I* I* c. ¬A {2,3,4}

CD CD SC
{1} A ¬I* {1,4}
C C C C

C CD
CD
I ¬I A* {1,2} SC {2,3}
I C
b. C C
I*
[[F]] = Ø
¬A ¬I* CD

¬A I* SC

¬A I* A* {3}
{3,4} ¬I SC
A ¬I* SC CD
4 3 2 1 2 3 4
I ¬A* ¬A* {1,2,4}
I ¬A*
¬I A*
¬I ¬A*

FIGURE 3.9 The square, the VS-model, and the octagon for AAPC

commentators, as is shown in Chapter 5) leads to UEI.12 Leaving out space 4

yields the classic Square of Opposition, with the VS-model and the octagon of
Figure 3.10, as is now easily checked.
We can carry on in the same vein and define SMPC by stipulating that A
and A* are true when [[F]] ¼
. The result is shown in Figure 3.11. But such a
stipulation is not defensible on cognitive grounds. On cognitive grounds,
such a stipulation is arbitrary, because there is no reason in cognition why the
situations where all the members of [[F]] have a certain property should be
paired, at the level of linguistic expression, with those where [[F]] ¼
. It takes
a good deal of formal training to convince oneself that such a pairing
does make sense in terms of abstract mathematical set theory. We must,
therefore, conclude that SMPC is not constructible in the cognitive mood.
Other than BNPC, AAPC, and ABPC, SMPC requires a construction in pure
set-theoretical terms.

12
Even so, ABPC appears to be the optimal strict-natural system of predicate logic. As is shown in
Chapter 10, ABPC can remain in full force provided it is extended with a presuppositional component
and falsity is split up into presupposition-preserving minimal falsity (F1) and presupposition-
cancelling radical falsity (F2).
98 The Logic of Language

a. A ¬I* I*
CD
c.
C ¬A {2,3}
C
SC CD CD
{1} A ¬I* {1}
CD
C C CD
CD C
I ¬I A* SC
SC
b. I SC I*
{1,2} {2,3}
¬A I*
¬A I* CD CD
C
A ¬I*
3 2 1 2 3
¬A* {3} ¬I A* {3}
I SC
CD CD
I ¬A*
¬I A* ¬A* {1,2}
U

FIGURE 3.10 The square, the VS-model, and the complete octagonal graph of ABPC

One notes that there is no point in maintaining a square notation for

SMPC, as A and I, as well as A* and I*, have now become logically independent
and only the (meta)logical relations of contradictoriness and equivalence, in
so far as they are involved in the Conversions, have been left. This dramatic
decrease in logical power is discussed in Chapter 4.

a. b. ¬A {2,3}
[[F]] = Ø CD
{1,4}
CD {1,4}
A ¬I* ¬I*
A
¬A I*
CD CD
¬A I* {1,2}
A ¬I* I
4 3 2 1 2 3 4
I ¬A* CD
CD I*
I ¬A* {2,3}
¬I A*
A* A*
¬I ¬I {3,4}
{3,4} CD CD
U ¬A* {1,2}

FIGURE 3.11 The VS-model of SMPC and the poor remnants of its octagonal graph
 Natural set theory and natural logic 99

We thus conclude that BNPC can only be constructed in the cognitive

mood and that SMPC can only be constructed in the set-theoretical mood.
AAPC and ABPC are seen to be ‘mixed’ in that they contain elements that
clearly smack of cognition, but also elements that smack of set theory. In other
words, the more ‘primitive’ the logic, the more it depends on cognitive
factors. This makes explicit what precisely is involved in Russell’s programme
of depsychologization of logic.
Summing up, we may enumerate the following features of the logics
concerned:
1. SOME F is G is interpreted as ‘only some F is G0
2. ALL F is G is interpreted as ‘all but not only F is G0
3. ALL F is G is false when [[F]] ¼

4. Space 4 is idle.

These features are distributed as follows over the four predicate logics consid-
ered, whereby one notes that ABPC simply results from AAPC if space 4 is
made inoperative. SMPC lacks all the features:
BNPC 1, 2, 3, 4
AAPC 3
ABPC 4
SMPC --
Does this now mean that the unexpectedly rich, powerful, and sound
predicate logic of BNPC carries the day? Not quite yet, because there still
are a few empirical problems, one of them consisting in the fact that BNPC
fails to account for the strong natural intuition, observed by many authors
(notably Jespersen 1917: 86–91), that makes one feel that ¬A is equivalent with
both I and I*, which means that in the logical system we would like to see an
equivalence relation between ¬A, I and I*. But BNPC fails to oblige as is easily
checked in Figure 3.7b,c. The entailments from I and I* to ¬A hold, but natural
intuition requires a stronger relation. In this respect, ABPC fares somewhat
better, since, in ABPC, ¬A and I* are equivalent, though ¬A and I are mere
subcontraries.
The pragmaticists tackle the problem by an appeal to the Gricean maxims,
reinforced by Horn’s theory of scalarity (Horn 1972, 1989), in virtue of which
the negation, when applied to a quantifiable scale, cuts off only the higher
part of the scale but leaves the remainder intact. Or, as Jespersen put it (1917:
86), ’in negativing an A [ALL F is G; PAMS] it is the absolute element of A that
is negatived’. But this answer, widely subscribed to in pragmatic circles, seems
100 The Logic of Language

curious for a number of reasons. Thus, while it is accepted that A ‘ I (the

positive subaltern of the Square), it is at the same time posited that I | ¬A
(where ‘ | ’ stands for Gricean implicature), and while it is accepted that
¬I ‘ I* (Square), it is posited that, pragmatically, I* | I. In nonscalar terms, this
would amount to saying that a sentence like John has been killed entails John is
dead, while John is dead has the implicature John has not been killed. It strikes
one as very odd that these logical entailments and these pragmatic implica-
tures, allegedly derived from conversational principles, should interact in such
a way. It seems more sensible to posit different logical systems operating at
different levels of cognitive functioning.
Moreover, it is not clear why the psychology of scalarity does not apply to
the analogous propositional operators and and or, which can also be seen
to form a quantifiable scale, in that and involves as many terms as there
are members of the conjunction, while (exclusive) or involves just one.
Yet NOT-AND—that is, NOT(P AND Q AND . . . Z)—is not felt to entail P OR Q
OR . . . OR Z.
We must, therefore, look elsewhere for an explanation. The explanation
proposed here for the problem of ¬A not being equivalent with I and I* in
BNPC falls back on trivalent presuppositional logic (see Chapter 10) and on
the notion of topic–comment structure, also often called ‘information struc-
ture’ (see Chapter 11). Topic–comment structure often comes with negative
sentences. If we take it that topic–comment structure is formalizable as an
underlying cleft structure and that all in a sentence like Ben didn’t eat all of his
meal has comment status, this sentence is then analysed as ‘what Ben ate of his
meal was not all’, which entails presuppositionally that Ben ate some of his
meal, excluding the case that he ate nothing. This sentence thus presupposes,
and hence entails, that Ben ate some of his meal and it asserts that he did not
eat all of it. Similarly for a sentence like Not all flags are green, which, if
analysed as ‘it is not all flags that are green’, excludes the case that there are no
green flags. In sum, given the sentence form It is not all flags that are green, it is
presupposed, and hence entailed, that some flags are green and also that some
flags are not green (given that I
I*). And because, in BNPC, both I and I*
entail ¬A, it follows that ¬A both entails and is entailed by I and I*, which
makes the three equivalent. This solution requires, however, that ¬A-type
sentences be given a topic–comment structure with ‘not all’ as the comment
and that BNPC be extended with a presuppositional component.
A further problem arises in connection with the extension of BNPC to a
full system of predicate logic catering also for cases where [[F]] ¼
, as in
Figure 3.8. The problem is that, in cases where [[F]] ¼
, it seems natural
 Natural set theory and natural logic 101

to say that N is true. Given that this world has no mermaids, it is not hard
to agree that There are no mermaids living in London must be considered
true. But then the corresponding N* sentence There are no mermaids
(that are) not living in London should also be true, because if there are no
mermaids, there will be no mermaid among those entities that are living in
London nor among those that are not. The disturbing fact is, however, that
while There are no mermaids living in London is considered true, There are no
mermaids (that are) not living in London is false for natural intuition and,
in fact, felt to be equivalent with All mermaids live in London. This latter
intuition is accounted for, since, as is shown in Figure 3.7b, both A and N* are
true only in space 1 and false in spaces 2 and 3, which makes them equivalent
within the confines of a model where the condition [[F]] 6¼
is left out of
account.
But in order to see if BNPC is tenable as a logical system, we must go
beyond those confines and take the fourth space into account for cases where
[[F]] ¼
. This has been done in Figure 3.8b, where truth has been assigned to
N in space 4. But then truth must also be assigned to N* on pain of making the
semantics of the operator NO inconsistent. It thus follows that if we
turn BNPC into a fully fledged logical system that also caters for cases
where [[F]] ¼
, as in Figure 3.8, a gross unnaturalness appears, because
truth for N* makes N* clash with natural intuitions. This is, however, not a
problem for the logical system but only for the claim that BNPC reflects
natural logical intuitions. Therefore, if this claim is to be upheld, it is essential
that space 4 for cases where [[F]] ¼
should not be considered to be part of
it—owing to PNST–1, which declares
not to be a set.
Finally, BNPC is still subject to the predicament that it requires
complete knowledge of the verification domain before one is entitled to say
that existentially quantified sentences are true or false. This is illustrated as
follows. Suppose Joe is checking if all the 45 doors in the building are
properly locked. He has come to number 15 and so far all has been well. We
feel that, as soon as he has found that at least one (or two) doors are
properly locked, he ought to be able to say in truth that (at least) some
doors are properly locked. But BNPC does not allow him to do so, because
Some doors are properly locked entails that not all doors are. And Joe cannot
vouch for that entailment. In fact, he must wait till he has checked all
doors before he can say either that some doors are properly locked or
that all are. All he can say after finding that one or more doors are properly
locked is that it is not so that no door is properly locked, or: ‘not (no door is
102 The Logic of Language

properly locked)’, which entails that at least some and perhaps all doors are
properly locked.13
This fact is of great epistemological importance. It is already so that A- and
N-statements require full knowledge of the domain before one is entitled to
claim their truth. These entitlements are thus restricted to finite and practi-
cally surveyable domains. But most domains are infinite or in any case not
practically surveyable and yet we profusely help ourselves to positive and
negative universal statements about them. Strictly speaking, we cannot vouch
for the truth of such statements, yet we venture them, relying on our induc-
tive powers of generalization and thereby taking the risk of falsification. And
this is precisely why they are so useful: almost all of what we consider to be
our knowledge is inductive knowledge, which has come to be established
on the strength of systematic lack of falsification and of the ‘sense’ they
make in terms of larger systems. But this conveys an equal importance to
I- and I*-statements as used in traditional and standard modern logic, because
these statements have the power of falsification. And for this it is needed that
one be entitled to claim their truth without full knowledge of the domain, merely
on the strength of an observation made. If I and I* are added to the list of
sentence types that require full knowledge of the domain before one can
vouch for their truth, this instrument to express a falsification is taken away.
In this sense, BNPC is an obstacle to the expansion of inductive knowledge.
This conclusion is to some extent disconcerting. One might propose that
natural language some is ambiguous between ‘some perhaps all’ (¼ NOT NO)
and ‘some but not all’. But the rules of good methodology make one reluctant
to do that. There is, thus, the basic intuition that some implies or entails ‘not
all’, but further reflection, manifesting itself at the strict-natural level, makes
one see that some ought not to be taken that way. If this were cognitive reality,
it would put a brake on the extension of inductive knowledge and thus on
intellectual development as a whole. Aristotle made the world see that,
on reflection, one must concede that A-sentences entail the corresponding
I-sentences, thereby removing the main blemish in any predicate logic con-
strained by the principles of basic-natural set theory. Needless to say, this

13
Horn (1989: 219) shows that De Morgan was well aware of this epistemological dilemma. Horn
quotes Hamilton (1858: 121):
There are three ways in which one extent may be related to another […]: they are, complete inclusion, partial
inclusion with partial exclusion, and complete exclusion. This trichotomy would have ruled the forms of logic, if
human knowledge had been more definite. […] As it is, we know well the grounds on which predication is not a
trichotomy, but two separate dichotomies. […] Must be, may be, cannot be, are the great distinctions of ontology:
necessity, contingency, impossibility. This was clearly seen by the logicians. But it was not so clearly seen that this
mode of predication tallies, not with the four ordinary forms A, E, I, O, but with the three forms A, (OI), E.
 Natural set theory and natural logic 103

Aristotelian upgrading from basic to strict naturalness conquered the world,

until it was replaced, a century ago, with the modern, highly constructed,
system of predicate logic known as standard modern predicate calculus,
which, ironically, banished again the subaltern entailment from A to I
sentences so as to get rid of undue existential import.
In overall perspective, it seems that we have to conclude that SMPC is not
fully constructible in the cognitive mood, just as BNPC is not fully construct-
ible in the set-theoretical mood. The question hinges on the semantic defini-
tions of the quantifiers. The existential quantifier SOME must be considered
symmetrical for set-theoretic purposes, so that SOME F is G
SOME G is F (I

I!), whereas it must be considered nonsymmetrical for ‘cognitive’ purposes,
so that SOME F is G
SOME F is NOT-G (I
I*). The universal quantifier ALL
must, for full set-theoretic constructibility, be taken to yield the value true in
situations without any F, whereas, for full ‘cognitive’ constructibility, it must
be taken to yield the value false in such situations. The combination of
symmetrical SOME and ‘cognitive’ ALL, as in AAPC and ABPC, creates the
subaltern entailments from A to I and from A* to I*. They are eliminated by
the combination of symmetrical SOME and set-theoretic ALL, as in SMPC.
We have found that all four systems are definable in both the cognitive
and the set-theoretical mood, but that in deriving or constructing the
various systems from underlying cognitive elements, BNPC is maximally
‘cognitive’, whereas AAPC and ABPC are more set-theoretically and less
‘cognitively’ oriented than BNPC. SMPC has shed all ‘cognitive’ elements.
Other than BNPC, AAPC, and ABPC, SMPC requires a construction in
pure set-theoretical terms: it is fully depsychologized. The re-introduction
of the ‘cognitive’ element into logic as presented here is motivated by, hitherto
unacknowledged, empirical considerations regarding human natural logic, in
its various, culture-induced, manifestations.

3.4.2 Hamilton’s predicate logic

The predicate-calculus system at the level of basic naturalness is strongly
reminiscent of, but not identical to, the brand of predicate logic proposed
by the Edinburgh philosopher Sir William Hamilton (1788–1856). The
main publication in this respect is Hamilton’s posthumous (1866).14 Perhaps

14
Part of Hamilton’s claim to fame rests on the drawn-out public polemic between him and the
London logician Augustus De Morgan, which was as famous as it was fierce and even made them take
each other to the courts of justice. The dispute centred upon Hamilton’s predicate logic, which was not
to De Morgan’s liking. See the Appendix in De Morgan (1847), which contains much of their
acrimonious correspondence.
104 The Logic of Language

because Hamilton lost his war with De Morgan, and perhaps also because of
the unique prestige of modern logic, Hamilton’s system of ‘quantification
of the predicate’, as he called it, has largely been forgotten outside circles of
historians of logic.15 Yet, given the undoubted intuitive appeal of at least some
of Hamilton’s logical notions, the Hamiltonian tradition in logic deserves a
closer look. The more so because the Danish linguist Otto Jespersen (1860–
1943), who possessed a finely tuned intuition as regards linguistic matters but
had no logical knowledge (he had probably never heard of Hamilton), came
up with a system of predicate logic that resembles the Hamiltonian system in
every respect except for Hamilton’s ‘quantification of the predicate’ (Jespersen
(1917: 85–92).
A central feature of Hamilton’s logic (see also Cavaliere 2007) is his
insistence on quantification of the predicate. This implies that not only the
subject but also the predicate in the Aristotelian sentence types should be
quantified, despite the grammatical awkwardness of sentences like All men are
some animals.16 He even chides Aristotle (Hamilton 1865: 264–5) for ‘prohibit
[ing] once and again the annexation of the universal predesignation to the
predicate’ (see note 15), continuing ‘Yet this nonsense, (be it spoken with all
reverence for the Stagirite,) has imposed the precept on the systems of Logic
down to the present day’. In this respect, I do not follow Hamilton and take
sides with Aristotle and with standard modern logic, even though one might
perhaps attribute to Hamilton the implicit insight that quantification involves
two sets, the matrix or predicate set and the restrictor set.
Hamilton does not use anything like quantifiers, but prefixes each set
denotation with the operators t (total) or p (partial), using the symmetrical
predicates ‘¼¼’ for ‘coincides with’ and ‘jj’ for ‘excludes’. The following simple
composition rule thus generates all expression types (formulæ) of Hamiltoni-
an predicate logic (‘[Æ /
]’ stands for ‘either Æ or
’):
Formula ¼: [ t / p]X [¼¼ / jj] [ t / p]Y (where X and Y are predicates)

15
Hamilton’s efforts are part of a tradition of predicate logics with ‘quantification of the predicate’
that was rejected by Aristotle (Int 17b13–16) and occasionally discussed during the late Middle Ages,
but which really started in the late eighteenth century and flourished in the nineteenth century (see,
for example, Bochenski 1956, Kneale and Kneale 1962: 349, Cavaliere 2007, Lenzen 2008). A major
factor in the development of these logics was the wish to enrich and strengthen the classical
Aristotelian theory of syllogisms with the help of richer systems of predicate logic.
16
Some languages, including English, marginally allow for sentences like Some humans are all
Englishmen, where all is a so-called ‘floating quantifier’, saying that there is a group of humans who are
all Englishmen. This, however, does not seem to be what Hamilton intended.
 Natural set theory and natural logic 105

Given that predicates may denote complement sets, there is, for every predi-
cate X, with the extension [[X]], a negative counterpart NOT-X, with the
—
extension [[X]] . For the predicates F and G and their negations, there are
thus 32 admissible formulæ.
With some effort, such a system can be made good semantic sense of.
Despite the grammatical illformedness of quasi-sentences like All Englishmen
are some humans, there is an intuitive inkling of what they could possibly
mean. This inkling can be made formally explicit in a variety of ways. An
interpretation that seems to come closest to Hamilton’s intentions is the
following:
Read ‘tX’ as ‘the total set of Xs’ and ‘ pX’ as ‘a nonnull proper subset of
Xs’ (where X ranges over predicates and [[X]] 6¼
6¼ U). Read ‘¼¼ ’ as
‘coincides with’ and ‘jj’ as ‘excludes’.
A quasi-sentence like Some computers are all laptops will then be read, in this
interpretation, as pComputer ¼¼ tLaptop or ‘only a nonnull proper subset
of computers coincides with the total set of laptops’, which is true only if
[[Computer]]
[[Laptop]]. By contrast, Some computers are all non-laptops,
translated as pComputer jj tLaptop or ‘only a nonnull proper subset of
computers excludes the total set of laptops’, is true if either [[Computer]]

[[Laptop]] or [[Computer]] OO [[Laptop]] (‘OO’ stands for mutual partial

intersection). Then, Some computers are some laptops is read as pComputer
¼¼ pLaptop or ‘only a nonnull proper subset of computers coincides with
only a nonnull proper subset of laptops’, true only if [[Computer]] OO [[Lap-
top]]. This is equivalent with Some computers are some non-laptops, which
reads as pComputer jj pLaptop or ‘only a nonnull proper subset of compu-
ters excludes only a nonnull proper subset of laptops’, again true only if
[[Computer]] OO [[Laptop]].
From a purely logical point of view, this system looks as if it has
certain advantages, as the relation of its expressions with set-theoretic con-
stellations is more straightforward than in the logics that quantify only over
the subject (restrictor) term. For example, tF ¼¼ tG is true only if [[F]] ¼
[[G]], while tF ¼¼ pG is true only if [[F]]  [[G]]—a distinction that is not
expressible in the standard modern rendering of the sentence type ALL F is G,
whether in the Peano-Russell notation or in the notation of the generalized
quantifiers.
Yet despite its possible advantages for logic as such, Hamilton’s logic,
whether in his own notation or in any updated version, has an uneasy relation
with natural language. Besides the absence of negation, its translations into a
natural language such as English result either in artificially regimented, and
106 The Logic of Language

even ungrammatical, sentences or in sentences that allow for multiple trans-

lations into the Hamiltonian system. For example, a sentence type like ALL F is
G is true when tF ¼¼ tG or tF ¼¼ pG. And SOME F is G is true when pF ¼¼
tG, pF ¼¼ pG, pF jj tG or pF jj pG. In other words, Hamilton’s insistence on
‘quantification of the predicate’ may have certain advantages for a system of
predicate logic and a concomitant syllogistic, it complicates a translation
mapping onto natural language beyond empirical endurance.
This becomes even more apparent when one considers multiple quantifica-
tion, as in All boys admire some football players, which are not translatable at all
into natural language sentences in terms of quantification of the predicate.
One could go as far as ‘All boys are some (all) [admirers of some football
players]’, but then one still has to introduce predicate quantification for ‘some
football players’ and this is something natural language at least has no
provisions for. Modern logic, of course, has, in principle, provided adequate
logical analyses of multiple-quantification sentences.
If we forget about the quantification of the predicate, we need the three
quantifiers ALL, SOME (but not all), and NO, corresponding with the three
sentence types AHam for ALL F is G, IHam for SOME but not all F is G
(equivalent with I*Ham or SOME but not all F is NOT-G), and NHam for NO F
is G. The three types AHam, IHam, and NHam are mutually exclusive and thus
form three pairs of contraries. They are defined as follows, in purely set-
theoretic terms ([[F]] and [[G]] are natural but not necessarily distinct sets):
AHam (ALL F is G) true iff [[F]]  [[G]]
IHam/ I*Ham (SOME F is (NOT-)G) true iff [[F]]
[[G]] or [[F]] OO [[G]]
NHam (NO F IS G) true iff [[F]] OO [[G]]
If this interpretation of Hamilton’s logic is correct, it only differs from BNPC
in that the condition for A-type sentences is [[F]]  [[G]] in the former but
[[F]]  [[G]] in the latter (as Hamilton allows for identity of two sets).
As we have seen, the condition for I-type sentences is irreconcilable with
any nonarbitrary natural set-theory hypothesis, due to its cognitively unmo-
tivated putting together of the two conditions [[F]]
[[G]] and [[F]]OO[[G]]. A
proper implementation of the set-theoretic restrictions (1) to (6) of Section
3.2.2 yields a logic where I-type sentences are defined as follows ([[F]] and [[G]]
are distinct natural sets):
I (SOME F is G) true iff [[F]] OO [[G]]
That is, SOME F is G is now read as ‘some but not all F is G and some but not
all G is F’. In such a logic, other than in the Hamilton system, I is not
 Natural set theory and natural logic 107

equivalent with I*, because when [[G]]  [[F]], SOME F is G is false (there is
no M-partial intersection) but SOME F is NOT-G is true. By contrast, the
equivalence of SOME F is G with its converse, SOME G is F (I
I!) holds in
such a logic, but not in the Hamiltonian system, again owing to cases where
[[G]]  [[F]]. In such cases, Hamilton makes SOME F is G true but SOME G is
F false, while a logic that straightforwardly translates PNST–5 of Section 3.2.2
into SOME makes both false. A logic where both I
I* and I
I! hold is
inconsistent.
Examples (3.14) and (3.15) quoted above show that natural intuition
supports Hamilton and BNPC and rejects a symmetrical SOME (I
I!). Does
this mean that AAPC and ABPC are already somewhat counterintuitive,
because in these systems the equivalence I
I! holds, owing to the fact that
they count SOME F is G as true when [[F]]  [[G]]? Curiously, and despite cases
like (3.14) and (3.15), symmetrical SOME is likewise supported by intuition.
When I say that some children are male, you will agree that, therefore, some
males are children, and when I say that some males are children, you will agree
that, therefore, some children are male, so that the two-way entailment
relation makes them seem equivalent. Apparently, the intuitive judgements
are influenced by world knowledge. We know that orphans are a proper
subclass of the class of children, which makes it possible for us to say naturally
that some children are orphans. But we also know that the class of children
M-partially intersects with the class of males and also with the class of
females, while it properly includes the class of orphans, as in the diagram
of Figure 3.12.
It thus seems that, given a constellation like that in Figure 3.12, (3.16a,b,c) are
naturally said in truth, but (3.16d) is not:

Children
Orphans
Females Males

Humans

FIGURE 3.12 The set-theoretic relations of children, (human) males, (human) females,
and (human) orphans
108 The Logic of Language
pffi
(3.16) a. pffi Some children are male (female).
b. pffi Some children are orphans.
c. Some males (females) are children (orphans).
d. ! Some orphans are children.
Yet both AAPC and ABPC, in so far as they can lay claim to naturalness, tell us
we should be able to say (3.16d) naturally in truth. Therefore, if we really want
to gauge natural intuitions and express them in a logical system, we should
hold on to BNPC, which does justice to all the intuitions of (3.16). AAPC and
ABPC can thus be said to support natural intuitions less strongly than BNPC,
because they allow for (3.16d), which is excluded by intuition.

3.4.3 Basic-natural propositional logic

Is there an analog in propositional logic to the basic-natural system of
quantification? To a large extent there is, but there are also differences. We
recall from Figure 3.1 that, unlike the quantifiers, which express either rela-
tions between sets of objects or, in the cognitive mood, distributive assign-
ments of properties to members of a set, the propositional operators are
functions mapping n–tuples of VSs on VSs. They become predicates when
predicated of the actual situation sitact, as shown in (2.6a–c) in Section 2.3.4
and in (3.17)–(3.18) below.
NEGATION is not immediately problematic (though, when taken in all its
aspects, it is perhaps the most complex of all the functions). For now, we
adopt the standard definition (2.6a) of Section 2.3.4:
(3.17) [[¬]] ¼ { P j sitact 2 =P=}
(the extension of ¬ is the set of all L-propositions P such that the
actual situation sitact is a member of the complement of /P/)
CONJUNCTION (∧BN) maps onto the BN-intersection of the VSs of the argu-
ment L-propositions, as defined in (3.18a), adapted from (2.6b) in Section
2.3.4 (Pþstands for any number of semantically compatible L-propositions).
As regards DISJUNCTION (vBN), the situation is more complex. We take the
linguistic disjunction operator OR to map onto the BN-union of the VSs of
the argument L-propositions, as defined in (3.18b), adapted from (2.6c) in
Section 2.3.4:
\ BN
(3.18) a. [[^BN]] ¼ { P+ j sitact 2 /P/þ }
(the extension of ∧ is the set of all sets of two or more L-
propositions P, such that sitact is a member of the BN-intersection
of all / P/þ)
 Natural set theory and natural logic 109
[ BN
b. [[∨BN]] ¼ { P+ j sitact 2 /P/ þ }
(the extension of ∨ is the set of all sets of two or more L-propositions
P, such that sitact is a member of the BN-union of all /P/þ)
The difference with (2.6b) in Section 2.3.4 is that the argument L-proposi-
tions cannot be necessarily true or necessarily false (the VSs would equal U and
,
respectively). Moreover, the conjunction as a whole cannot be necessarily true
or necessarily false either. The fulfilment of this latter condition is ensured by
the stipulation (PNST–5) that the VSs of the L-propositions involved must
M-partially intersect. This ensures, first, that there is at least a chance of all the
L-propositions united under conjunction to be true simultaneously. For if the
VSs do not intersect, there is no possible situation inwhich the two L-propositions
are both true, which makes them contraries and their conjunction necessarily
false. Since AND is the standard discourse-increment function, this condition
is, in fact, the same as the condition on any new increment to a discourse
domain, discussed in the Sections 7.3.1 and 8.2.1, that it be compatible with all
earlier increments, to avoid a reduction of the discourse domain’s VS to zero.
Secondly, M-partial intersection ensures that, given the L-propositions P
and Q, /P/ does not include /Q/ or vice versa, which would make the
conjunction necessarily true. The basic-natural semantics of AND thus selects
the M-partial intersection in the basic-natural intersection relation in Figure
3.3b, which is, therefore, defined only for L-propositions that neither include
nor exclude each other in virtue of their meaning.
L-propositions made up with AND are classified as type AND (conjunction);
those with OR as type OR (disjunction); those made up with NEITHER . . . NOR as
type NOR (exjunction). AND thus says that the actual situation sitact is an
element in the intersection of /P/ and /Q/ of Figure 3.13.

NOR

/P/
OR AND OR

/Q/

UR

FIGURE 3.13 VSs for logically independent P and Q and their logical compositions
110 The Logic of Language

As regards the sentence type OR (∨BN), the situation is more complex.

We take the linguistic disjunction operator OR to map onto the BN-union of
the VSs of the argument L-propositions, as defined in (3.18b) above. Under the
restrictions imposed on union by the criteria of basic naturalness, the compo-
nent VSs must not only be natural sets (neither
nor OBJ), but, according to
PNST–3, they must also be totally distinct—that is, without any intersection.
However, natural language OR cannot be restricted to contrary argument
L-propositions: in a disjunction of the form P OR Q, P and Q must be allowed
to be logically independent, so that their VSs, /P/ and /Q/, actually intersect, as
in Figure 3.13. A sentence like (3.19), for example, is perfectly normal even
though it is possible for both disjuncts to be true at the same time:
(3.19) He either eavesdropped or he went through your papers.
To satisfy PNST–3, it is, therefore, necessary to reformulate P OR Q for the
purpose of linguistic interpretation at the level of basic naturalness in such a
way as to ensure that the two or more arguments of the union function are
totally distinct.
This can be done by considering P OR Q to be tacitly understood as:
(3.20) P OR (NOT-P AND Q); formally: P ∨ (¬P ∧ Q)
Sentence (3.19) is then interpreted as ‘He either eavesdropped or he did not
eavesdrop and/but went through your papers’. Now the two disjuncts have
totally distinct VSs, which, therefore, allow for the basic-natural union oper-
ation, even in cases such as (3.19) where the VSs of the disjuncts M-partially
intersect on account of their being semantically independent.
The tacit understanding as formulated in (3.20) has the advantage of
accounting for the unnaturalness of a disjunction where one disjunct entails
the other as in:
(3.21) !The man is dead or he has been killed.
If this is tacitly understood as ‘either the man is dead or the man is not dead
and he has been killed’, then the second disjunct is necessarily false owing to
the contrariety of ‘the man is not dead’ and ‘he has been killed’.
The tacit reading of (3.20) also accounts for the otherwise problematic ana-
phora (donkey anaphora) in sentences like (3.22), again discussed in Section 8.2.2:
(3.22) There is no letterbox in this street or it is well hidden.
According to the book, the anaphoric it in the second disjunct should not be
possible as it appears to lack an antecedent. Yet when read according to
 Natural set theory and natural logic 111

(3.20)—as ‘there is no letterbox in this street or there is one and it is well

hidden’—there is no problem in this regard, because the second disjunct
provides the antecedent, albeit a tacit one.
The disadvantage of (3.20), however, is that it still fails to guarantee that the
two or more arguments of the OR-function are mutually exclusive for seman-
tic—that is, analytical—reasons, since the truth of the one disjunct P is still
sufficient for the truth of the whole disjunction P ∨ (¬P ∧ Q). As a result,
type-AND L-propositions still entail type-OR L-propositions, which must be
avoided in basic-natural propositional logic. One might, therefore, think of a
more drastic reformulation which makes OR truly exclusive with regard to its
two or more disjuncts. This can be done if one regards P OR Q to be tacitly
understood as (3.23), which is not equivalent with standard P ∨ Q :
(3.23) (P AND NOT-Q) OR (NOT-P AND Q); formally: (P ∧ ¬Q) ∨ (¬P ∧ Q)
This, however, has the disadvantage of going against the book, as it seems
to fail to account for (3.22): a reading like ‘there is no letterbox in this street
and it is not well hidden, or there is one and it is well hidden’ appears to
leave the first occurrence of it without a proper antecedent. This obstacle,
however, is removed when the negation in it is not well hidden is treated
as the metalinguistic radical presupposition-cancelling NOT discussed and
defined in Chapter 10. On that reading, the problematic it refers to an
intensional object and thus needs the radical NOT to achieve truth. The
disjunction operator OR now selects the union of /P/ and /Q/ in Figure 3.9,
minus their intersection.
This appears to account for the intuition that natural language OR is
exclusive, excluding cases where both P and Q are true. At the same time,
as is shown in Section 9.2.4, it accounts for a number of clashes between
intuitive judgements regarding conditionals (sentences of the form if P then
Q) and their standard logical analysis in terms of material implication.
But how does this work out for the logic? In the reading (3.23), /AND/ and
/OR/ are mutually exclusive and together form standard /OR/. The only space
not covered in UR is the restricted complement of standard /OR/—that is,
—
/OR/R—normally denoted by the linguistic operator neither a . . . nor b, also
realized as not . . . (either) a or b and here called NOR (exjunction).17 One notes
that NOR cannot be the negation of the exclusive OR of this system, since the

17
In Seuren (1974) it is argued that an L-propositional structure of type AND* (AND [¬P, ¬Q]) is
grammatically transformed into NOR (NEITHER P NOR Q) by the rule of NEGATIVE RAISING. Likewise, ALL F
is NOT G is taken to be transformed into NO F is G. This conforms to the basic-natural propositional
and predicate logic as proposed here and shown in Figure 3.10.
112 The Logic of Language

a.
b. AND NOR* OR*
¬OR ¬AND C
¬OR* ¬NOR*
¬AND ¬AND* C C
¬NOR* ¬NOR
C C
NOR OR
3 AND* OR* 2

AND
NOR* C
¬OR OR NOR AND*
1
¬OR*
¬NOR
¬AND*
UR

FIGURE 3.14 Basic-natural analog for propositional logic with NOR 6¼ ¬OR

—
complement of /OR/, /OR/R, comprises /AND/, and neither P nor Q clearly
means NOT-P AND NOT-Q, excluding P AND Q. Therefore, just as English no
does not correspond to basic-natural NOT-SOME but constitutes a separate
quantifier in BNPC, English nor does not correspond to basic natural NOT-OR
but constitutes a separate sentence-type, which may be called EXJUNCTION, in
the NST-constrained system of propositional logic. Consequently, NOR is
equivalent, in this system, with AND* (within UR), as shown in Figure 3.14.
(One remembers from Section 2.4.2 that the internal negation for the propo-
sitional operators AND and OR distributes over the component L-propositions.
Thus, AND* stands for ¬P ∧ ¬Q ∧ . . . and OR* stands for ¬P ∨ ¬Q ∨ . . . .)
NOR*, ¬NOR, and ¬NOR* may be taken to be undefined at the level of basic
naturalness, due to PNST–6 (the mind baulks at repeated applications of the
complement function). As before, however, they are still taken into account in
Figure 3.14, because we need to know the ultimate logical consequences of the
system.
Empirical evidence for this analysis is derived, for example, from the fact
that a sentence like (3.24a) is naturally and immediately interpreted as (3.24b),
whereas (3.25a) is not at all naturally and immediately interpreted as (3.25b),
even though, from a standard logical point of view, both (3.24) and (3.25)
merely instantiate De Morgan’s laws. This is a fact on which neither the
Gricean maxims, nor indeed the whole of pragmatics, have anything to say:
(3.24) a. He doesn’t like planes or trains.
b. He doesn’t like planes and he doesn’t like trains.
(3.25) a. He doesn’t like planes and trains.
b. He doesn’t like planes or he doesn’t like trains.
 Natural set theory and natural logic 113

Given the analysis presented above, one now sees why this should be so.
Consider not . . . planes or trains in (3.24a) to be an instance of basic-natural
NOR: ‘He neither likes planes nor does he like trains’. Figure 3.14 shows that, in
this reading, NOR is equivalent to AND*, both sharing the same VS. By
contrast, not . . . planes and trains in (3.25a) realizes ¬AND, the restricted
negation of AND: ‘It is not so that he likes planes and he likes trains’. The VS of
¬AND comprises the combined VSs of NOR (
AND*) and OR (
OR*). Since
in basic-natural propositional logic, ¬AND is not equivalent with OR*, the
transition from (3.25a) to (3.25b) is a matter of strict-natural or standard
modern logic and thus requires a more complex computation at a higher level
of achievement.
There is, however, again the problem of consistency through discourse. As
has been widely observed, OR is typically used in situations where the speaker
is uncertain as to which of the disjuncts provides the correct answer to the
question he or she is entertaining but where the speaker has concluded that
either disjunct will do as a good enough answer. This conclusion is naturally
expressed by the auxiliary of epistemic necessity must. A speaker may say
(3.26b), in response to the question (3.26a):
(3.26) a. How did the journalist know?
b. The journalist must have spoken to Ann or to Jeremy.
If it were then found out that the journalist had spoken to both Ann and
Jeremy, it would be incorrect to say that the person who uttered (3.26b) had
been wrong, even if that speaker had failed to think of the possibility that the
journalist had spoken to both Ann and Jeremy.
Therefore, in parallel with the basic-natural system of quantification,
consistency through discourse requires an entailment from AND to OR. This
time it was the Stoic philosophers who, roughly a century after Aristotle,
discovered this fact and upgraded the natural propositional logic of language
and cognition to full consistency through discourse by introducing the

a. A ¬I* I* b. AND ¬OR* OR*

CD CD
C C
SC SC
C C

CD CD
I ¬I A* OR ¬OR AND*

FIGURE 3.15 The strict-natural Squares for predicate and propositional logic
114 The Logic of Language

entailment from AND to OR, thereby creating the inclusive OR of standard

propositional calculus.
The introduction of the parallel entailment schemata A ‘ I and AND ‘ OR
has thus brought about a complete parallelism of the two strict-natural
systems of predicate and propositional logic. The former results in strict-
natural predicate calculus, which is identical with the traditional Square of
Opposition and is shown in Figure 3.15a. The latter results in strict-natural
propositional calculus, which is isomorphic with standard propositional
calculus and is shown in Figure 3.15b.

3.5 Neither *nand nor *nall: NST predicts their absence

3.5.1 The problem and the solution proposed by pragmaticists
We have reached a point where a question can be discussed which has recently
attracted some attention. After it had been observed by Thomas Aquinas (see
Horn 1989: 253; Jaspers 2005: 1), it was again observed more recently (Zwicky
1973; Horn 1972, 1989: 252–67; Levinson 2000: 69–71) that while a large majority
of languages have what is taken to be a single-morpheme expression for NOT-
SOME (such as English no), a single-morpheme expression, say *nall, for the
complex negative predicate NOT-ALL or its presumed equivalent SOME-NOT has
not so far been attested. Analogously, while many languages have a lexeme for
NOT (P OR Q) or equivalently NOT-P AND NOT-Q (English neither . . . nor), a
lexeme corresponding to *nand, for NOT (P AND Q), or equivalently NOT-P OR
NOT-Q, is thought to be either nonexistent or at most extremely rare.
One must, however, be careful. Is it indeed so that *nand and *nall represent
systematic gaps in the lexicons of the world’s languages? It does not look as if
Horn or others who have presented this problem have made a systematic
inquiry in this respect, although there are reasons for doubting the universality
of these lexical gaps. The Tamil specialist Eric Pederson, for example, has
informed me that Tamil does have a single lexeme meaning ‘not and’ or ‘either
one of the two or none’. And Aimable-André Dufatanye, who is a native
speaker of Kenya-Rwanda and a near-native speaker of Kirundi (both African
languages of the Bantu group), tells me that in his language(s) there are fully
lexicalized lexemes for ‘not-all’. Thus, he gave me the forms (subject to the
nominal classification system in the language) sibose (‘not all people’), sihose
(‘not everywhere’), siyose (‘not the whole house’), sizose (‘not all houses’), as
opposed to ntanzu (‘no house’), nta muntu (‘no person’), and so on. It is,
therefore, far from certain that the lexical gaps in question are universal.
Whether they occur in a significant majority of languages and, if so, whether
 Natural set theory and natural logic 115

they are systematic in the languages in which they occur, is yet to be estab-
lished.
If there is indeed a systematic absence of equivalents of *nall and *nand in
the languages of the world, such a gap appears to be matched by similar gaps in
other lexical fields. Thus, it is said, whereas one finds lexicalized equivalents of
epistemic NOT-POSSIBLE, lexicalizations of epistemic NOT-NECESSARY are found
nowhere (lexicalized agentive NOT-NECESSARY, such as English unnecessary is
widely attested). Likewise, NOT-CAUSE is never lexicalized, while NOT-ALLOW is
frequently found in lexicalizations such as disallow, forbid, or prohibit. Typi-
cally, predicates like NECESSARY or CAUSE show semantic characteristics that
may lead one to think that they can or should be classified along with ALL and
AND: they all belong to an ‘all-yes’ section of the lexicon, while POSSIBLE and
ALLOW typically belong to the group of ‘perhaps-yes-perhaps-no’ predicates,
which also comprises the existential quantifier and the propositional connec-
tive OR. The question is: are these similarities reducible to a single principle
and if so, what is it? This question is interesting as it forces one to probe natural
set theory in both the logic and the lexicalization processes of natural language.
Horn, Levinson, and others seek an answer in the pragmatics of language
use. Restricting themselves to *nand and *nall, they argue, in essence, as
follows. Since OR is normally exclusive for pragmatic reasons and pragmati-
cally equivalent to OR-NOT, excluding the case that both argument proposi-
tions are true, there appears to be no need left for an item like *nand which
excludes the simultaneous truth of both argument propositions and may be
taken to imply pragmatically that at least one of the argument propositions is
true. Then, given the pragmatic equivalence of or and *nand, the item without
the incorporated negation—that is, or—would be preferred on grounds of
simplicity in the lexicon, so that *nand is ruled out. Similarly, since SOME,
SOME-NOT, and NOT-ALL are pragmatically equivalent, conveying the intended
meaning ‘some but not all’, it is assumed that there is no need left for a
lexicalized form like *nall meaning ‘not-all’, which is somehow, either logical-
ly or pragmatically, equivalent to SOME-NOT and hence to SOME. Since the
pragmatically equivalent but cognitively and semantically simpler operators
or and some are already available, there is no need for surface lexicalizations
like *nand or *nall (Levinson 2000: 70). Therefore, the lexicalized expressions
all, some, and no will do for the quantifiers, and and, or, and neither . . . nor for
the propositional operators.
This reasoning can be summarized as follows:
(a) The English quantifier word no and its equivalents in other languages
are lexicalizations of underlying NOT-SOME. Analogously, neither . . . nor
stands for NOT-OR.
116 The Logic of Language

(b) Lexicalizations take place at a Gricean-pragmatic level of processing.

(c) At that level, SOME-NOT is equivalent with SOME, both implying NOT-ALL,
and OR-NOT [i.e. NOT-P OR NOT-Q] is equivalent with (exclusive) OR,
both implying NOT-AND.
(d) Since SOME-NOT is also, if not logically in any case pragmatically,
equivalent with NOT-ALL, and OR-NOT with NOT-AND, it follows
that NOT-ALL and NOT-AND are, at the Gricean-pragmatic level of
processing, equivalent with SOME and OR, respectively.
(e) Given the lesser semantic complexity of SOME vis-à-vis SOME-NOT and
NOT-ALL, and given the lesser semantic complexity of OR vis-à-vis OR-
NOT and NOT-AND, it is SOME and OR that are the preferred candidates
for lexicalization.

For the authors who propose this explanation such a system is not a logical
system but represents the way listeners construct a quantified mental model of
a state of affairs described, on the presumption that the speaker has full and
adequate knowledge of that state of affairs and has the intention to be as
informative and helpful as possible—that is, that speakers will commit them-
selves to the maximum of what they know. The main criterion of such a
system is not truth but information value on the presumption of full coop-
erativity and complete knowledge. It is not meant for the computation of
solid entailments grounded in strictly semantic properties, but for practical
inferences. To say that two expressions are pragmatically equivalent then
amounts to saying that they have the same information value on the pre-
sumption specified.

3.5.2 Preliminary objections

The question is, however, whether one may justifiably posit that linguistic
lexicalizations depend on ‘information value on the presumption of full
cooperativity and complete knowledge’. It would seem that such a position
is unwarranted, since the use of language is not restricted to authoritative and
maximally cooperative reporting on states of affairs fully known to the
speaker. For one thing, as has been said, reporting often consists in informing
the listener of what has been found out so far. When inspecting a population
of children, a reporter may say that so far no, or so far some, or so far all,
children have proved to be undernourished, without any commitment as to
what is still to be found out. This point was made by Hoeksema, who first asks
(Hoeksema 1999: 4) ‘If nall is not needed, due to the presence of some, then
why is not all used at all? ’ and then observes (Hoeksema 1999: 5):
 Natural set theory and natural logic 117

In contexts where the speaker has only partial knowledge, there is not even pragmatic
equivalence. If I say that some of my students are gay, one should not infer immedi-
ately that not all my students are gay. Perhaps I am unaware of the sexual preferences
of the remainder. But if I and O are often not even pragmatically equivalent, because
the conditions for the Gricean implicatures are not met, then why should O be
superfluous?

Since any cognitive theory of linguistic interpretation depends primarily on

pre-existing linguistic meanings, rather than on a presumption of full co-
operativity and complete knowledge, it would seem that the pragmatics
answer is not satisfactory on its own terms.
Moreover, as is argued in Chapter 4 of Volume I, speech is not primarily
meant to be a form of reporting or information transfer but consists, in
principle, in taking a socially binding position with respect to a given propo-
sition. Speaking consists primarily in vouching for truth, or in making a
suggestion or a request, issuing an order, asking a question, uttering a wish,
and so on, all with respect to the proposition expressed. The issue, therefore,
is not what factual information is provided by the speaker but what the
speaker commits himself or herself to.

3.5.3 The main objection and a stronger solution

But there is a more decisive argument, related to point (a) in the Horn-
Levinson analysis and, one fears, destroying it. Throughout their analysis,
Horn and Levinson take it for granted that the surface quantifier no stands
for, or is a lexicalization of, NOT-SOME and that neither . . . nor lexicalizes NOT-
OR. But that is far from obvious. As has been shown, BNPC is tenable only if it
is assumed that no does not stand for NOT-SOME but is a quantifier—the
quantifier NO—in its own right. And analogously for neither . . . nor and
NOT-OR. One may, of course, posit a pragmatic equivalence of NO with NOT-
SOME (cutting out the ALL option), and of NOR with NOT-OR (cutting out the
AND option), but it is not clear what pragmatic principle would be strong
enough to underwrite such an equivalence. Of course, the equivalence holds
in standard predicate and propositional logic, which may explain why one is
so easily led into accepting that it also holds at a basic-natural cognitive level,
but it is not to be found at that level, owing to the fact that, at that level, SOME
entails both NOT-ALL and NOT-NO but is not equivalent with either of them.
And analogously for OR, which entails both NOT-AND and NOT-NOR but is not
equivalent with either of them.
We have no problem with the principle that, by and large, lexicalizations
take place at a basic-natural level, be it pragmatic or cognitive in a more
118 The Logic of Language

formal sense—though, as is shown below, lexicalizations involving double

negation also occur, no doubt created at a more advanced level of cultural
development. But the point is that, at the basic-natural level, no is not a
lexicalization of NOT-SOME—to stay with predicate logic. If anything, no
incorporates any, not SOME, but we don’t really know what any amounts
to.18 In many languages, including English, (the equivalent of) no, which
lexicalizes the natural set-theoretic relation of mutual exclusion, clearly does
contain an incorporated negation, but not in combination with a quantifier of
the same logical system. Latin nullus, for example, is a combination of the
negation word ne and unulus, the diminutive of unus (one), like the Dutch
expression niet eentje (‘not a little one’). Greek oudeis (nobody) is literally ‘not
one’ and so on. Such lexicalizations all deny the presence of even the smallest
common element in the two sets involved, [[F]] and [[G]]. It is merely
accidental that this coincides with the negation of existentially quantified L-
propositions—that is, with ¬I—in both traditional and standard modern
predicate calculus. English no and its equivalents in many other languages
thus appear not to lexicalize the meaning NOT-SOME but, rather, the absence of
even the smallest common element in the two sets at issue—both linguistical-
ly and psychologically a plausible form of lexicalization, which, moreover,
lends itself to being incorporated into a variety of sound predicate-logic
systems. An analogous analysis applies to propositional logic.19
Horn (1989) recognizes that his pragmatic system reflects the Hamilton/
Jespersen predicate logic, but fails to mention that in that logic NO does not
stand for NOT-SOME, and analogously for NOR, which does not stand for NOT-
OR, NO and NOR being operators in their own right. Yet if the negation is
considered to be a complement selector, there is no way one can combine the
thesis that SOME excludes ALL with the thesis that NO equals NOT-SOME. And if
the negation is not taken to be a complement selector, it is incumbent on the
pragmaticist authors in question to provide a proper definition of negation,
since in that case it is unclear what negation is or does.
There are thus, at the level of naturalness at which lexicalizations are
deemed to take place, no single-morpheme expressions for NOT-SOME and
NOT-OR. This, in effect, takes the bottom out of the question of why lexicaliza-
tions for NOT-ALL and NOT-AND are systematically absent, as it arose in the first

18
English fuck all stands for ALL-NOT, which is equivalent to NO in BNPC as Figure 3.7 shows.
19
In Seuren (2002) it is observed that both in traditional predicate logic and in propositional logic
no key role is reserved for vertices named ¬A or ¬AND, which is presented as an explanation for the
systematic absence of lexicalizations like *nand or *nall. The present analysis shows that this
conclusion was premature. See Jaspers (2005) for ample comment.
 Natural set theory and natural logic 119

place because NOT-SOME and NOT-OR were thought to lack counterparts for ALL
and AND, respectively. In fact, however, given the lack of single-morpheme
lexicalizations for NOT-SOME and NOT-OR, NST predicts the absence of such
lexicalizations for NOT-ALL and NOT-AND, which in turn, if that absence proves
real, is valuable confirmation for the correctness of our reconstruction of
basic-natural logic.
The real question is not why no and neither . . . nor have no corresponding
counterparts *nall and *nand but, rather, why NOT does not merge with
either SOME or ALL (or with either OR or AND), though it does occasionally
merge with NO, which, with its double negation, is a mild infringement of
point (b) in the Horn-Levinson analysis (lexicalizations of NOT-NOR do not
seem to occur).
The answer may well be found in the consideration that mergers of the
form NOT-SOME or NOT-ALL would unite incompatibles. As has been said, NOT-
SOME would cover both ALL and NO, which will never form a natural cognitive
unit. The same goes for NOT-ALL, which would cover both SOME and NO, again a
very unlikely candidate for lexicalization. Only NOT-NO, which covers
both SOME and ALL, would seem to form an acceptable natural cognitive
unit encompassing the semantic field ‘some, perhaps all’. And indeed, lexica-
lizations of NOT-NO, usually in the stylistic form of an idiomatized understate-
ment (litotes), are, though not frequent, not too hard to come by.
As mentioned by Horn, Latin has a number of instances (see also Jespersen
1917: 90): nonnemo ‘not-nobody ! several persons’, nonnulli ‘not-none !
several’, nonnihil ‘not-nothing ! a considerable amount’, nonnumquam
‘not never ! quite often’, nonnusquam ‘not nowhere ! in several places’.
Dutch has niet-niks ‘not-nothing ! quite something’. Semi-lexicalizations
for ‘not-without ! with a notable amount of ’ are frequently found. A careful
search will no doubt yield a significant number of examples of this nature
in various languages.

3.5.4 Parallel lexical gaps in epistemic-modal and causal logic?

A word must be said here about the presumed analogous gaps in the lexica-
lizations of epistemic modal logic and the logic of causality. As regards
epistemic modal logic, observations have been made to the effect that there
is an epistemic-modal lexicalization impossible but not unnecessary, though
unnecessary is, of course, a current predicate in nonepistemic contexts. Like-
wise, it has been observed that NOT-ALLOW is frequently lexicalized in forms
such as disallow, forbid, keep from, prohibit, whereas NOT-CAUSE does not seem
to be lexicalized ever.
120 The Logic of Language

Given these observations, one may object that the solution proposed for
predicate calculus lexicalizations, which is based on the fact that in basic-
natural logic no does not stand for NOT-SOME but is a quantifier in its own
right, contrarily opposed to SOME and ALL, does not apply in epistemic modal
or in causal logic, because it can hardly be denied that impossible is a lexicali-
zation of NOT-POSSIBLE, or that disallow stands for NOT-ALLOW.
To this objection I reply that although impossible is, of course, a lexicaliza-
tion of NOT-POSSIBLE, the predicate POSSIBLE in question does not belong to
epistemic modal logic. Impossible has incorporated the predicate POSSIBLE in a
variety of its senses, but not in the modal-epistemic sense of ‘it may be
true that . . . ’. The Oxford English Dictionary (OED) gives the following two
main senses for possible, the second of which clearly comprises epistemic
possibility:
1. That may be (i.e. is capable of being); that may or can exist, be done,
or happen (in general, or in given or assumed conditions or
circumstances); that is in one’s power, that one can do, exert, use, etc.
2. That may be (i.e. is not known not to be); that is perhaps true or a
fact, that perhaps exists.) (Expressing contingency, or an idea in the
speaker’s mind, not power or capability of existing, as in 1; hence
sometimes nearly ¼ credible, thinkable.)

But for impossible the OED merely gives the negation of possible 1, not of
possible 2:
1. Not possible; that cannot be done or effected; that cannot exist or come
into being; that cannot be, in existing or specified circumstances.

That is, the OED describes impossible as occurring only as the negative
counterpart of possible in the main sense 1, not in the main sense given
under 2. There is, in other words, no negated counterpart of sense 2 of
possible. Impossible is naturally used in phrases like an impossible task, an
impossible construction, an impossible person, even an impossible truth, and also
in, for example, It is impossible to clear up that mess or It is impossible for the
man to climb the stairs, but never with a that-clause in the sense of ‘it cannot
be true that . . .’. A sentence like It is impossible that he is right strikes one as
deviant.20

20
I hesitate about the litotes not impossible, as in It is not impossible that he is right, which sounds a
great deal better than It is impossible that he is right. Also, as was pointed out by Isidora Stojanovic,
people often react to a piece of new information by saying That’s impossible, meaning ‘That can’t be
true’. It seems obvious that much still remains to be sorted out in this area.
 Natural set theory and natural logic 121

Moreover, whereas the adverbial clausal modifier possibly normally occurs

in an epistemic sense, as in (3.27a), its negative counterpart impossibly cannot
occur in that sense, as one sees from the ungrammatical (3.27b). The proper
negation of (3.27a) is (3.27c), with epistemic can’t as the proper negation of
epistemic may but idiomatically combined, probably for reasons of functional
emphasis, with an otherwise superfluous possibly:
(3.27) a. The man had possibly eaten too much.
b. *The man had impossibly eaten too much.
c. The man can’t possibly have eaten too much.
It seems, therefore, that there is no lexicalization of NOT-POSSIBLE in a
postulated basic-natural epistemic modal logic (‘POSSIBLE’ stands, of course,
for the ‘exclusive’ possibility operator, which excludes NECESSARY). Worse,
there does not seem to be a basic-natural epistemic modal logic, because if
there was one, one would expect to find a lexical analog of basic-natural no,
something like no way, meaning ‘neither POSSIBLE nor NECESSARY’, but no
convincing analog exists, now that impossible has been discarded. Logically
sound modal thinking is probably a product of at least some degree of
civilization and the concomitant cognitive development and would thus not
form a logical system at a basic-natural level, but only at a strict-natural level,
at which NECESSARY(P) entails POSSIBLE(P) and does not exclude it, as it
would do at a basic-natural level. In such a logic, NOT-POSSIBLE will do,
without further lexicalization.
As regards the logic of causality, one observes that the predicate disallow,
unlike forbid, keep from, or prohibit, is the only predicate that is transparently
derived from a negation plus allow. But disallow, together with forbid, denotes
an injunction not to DO something, not a causal blocking that keeps an event
from HAPPENING. Disallow and forbid belong to the semantic paradigm of
permission, not to that of causality. Interestingly, keep from and prohibit
may be used in a strictly causal sense, but they are not demonstrably com-
posed of a negation plus a predicate of causal possibility. This is interesting
because we see a parallel here between causal allow and epistemic possible.
In other words, if there are analogies between the lexicalizations in predi-
cate and propositional logic on the one hand and those in epistemic modal
logic and the logic of causality on the other, they follow the pattern predicted
by the analysis presented here, and not the pattern based on the assumption
that the negative existential quantifier no is to be analysed as NOT-SOME.
 4

Logical power, Abelard, and

empirical success rates

4.1 Aristotelian predicate calculus rescued

from undue existential import
As one can see from Figure 3.15 in the preceding chapter, standard proposi-
tional calculus and traditional Aristotelian-Boethian predicate calculus
(ABPC) run completely parallel and thus have the same logical power. As
long as [[F]] 6¼
, there is a complete parallelism between, on the one hand,
the universal quantifier and conjunction and, on the other, between the
existential quantifier and disjunction. This analogy has not passed unnoticed
in logical circles. It has often been observed that the universal and existential
quantifiers correspond to the conjunction and disjunction operators, respec-
tively: if 8x(G(x),F(x)) is true, then, for all a1, a2, a3, . . . an 2 [[F]], G(a1) ∧
G(a2) ∧ G(a3) ∧ . . . ∧ G(an) is likewise true. Analogously, if ∃x(G(x),F(x))
is true, then G(a1) ∨ G(a2) ∨ G(a3) ∨ . . . ∨ G(an) is also true. In relation to
this, the Conversions of predicate calculus are analogous to De Morgan’s laws
in propositional calculus. Except when [[F]] ¼
, because then there simply is
no conjunction or disjunction. It is for this reason that this analogy has never
been elaborated in the existing formalizations of the quantifiers.1 In Sections
4.2.4 and 4.3, however, it is shown that this analogy can, within the context of
a strictly extensional logic, be saved to a considerable extent, even for cases
where [[F]] ¼
. But then the Conversions of predicate logic must be sacri-
ficed: they must be weakened to e-way entailments.
The entailment schemata of propositional calculus thus correspond to
those of ABPC, but not to those of its twentieth-century successor STANDARD
MODERN PREDICATE CALCULUS (SMPC). This in itself would not be of much

1
Yet the analogy between the universal quantifier and the propositional operator ∧ on the one
hand, and between the existential quantifier and ∨ on the other, was expressed in the symbol ‘’ for
the universal quantifier and the symbol ‘V’ for the existential quantifier in the so-called ‘Californian’
notation, which was much used during the 1960s and 1970s but eventually had to yield to what is now
the standard notation.
 Logical power 123

concern, were it not that ABPC, just like propositional calculus, has maxi-
mal logical power, whereas SMPC, as is shown in Section 4.2.3, sees its
logical power dramatically reduced when compared with ABPC. The loss
of logical power in SMPC is due to the fact that SMPC has given up the
subaltern entailment schema A ‘ I (All F is G entails Some F is G), which is
considered valid in ABPC, just as AND ‘ OR, the analog in propositional
calculus, is valid in propositional calculus. Logicians, who like the idea that
SMPC is the ultimate ne plus ultra of predicate calculus, have never been
very vocal as regards this dramatic loss of logical power, but there is no
denying that it is there. Moreover, as is shown in Chapter 6, the logical
power of ABPC turns out to be highly functional for the transmission of
quantified information. Therefore, if ABPC can be saved for natural lan-
guage, our respect for this logic will be considerably enhanced (many
philosophers of language who dare not doubt the inviolability of SMPC
still look back to ABPC with nostalgia).
The situation as it is creates a dilemma in that SMPC is logically sound
but has bought its health at great cost, whereas avoiding that cost seems
to mean a faulty logic. It is time now to spell out the question in greater
detail.
The reason why ABPC was replaced with SMPC in the wake of scholars like
Frege and Russell about a century ago lies in the fact that ABPC, as it stands,
suffers from what is known as UNDUE EXISTENTIAL IMPORT (UEI). By this is
meant the fact that ABPC only functions when [[F]], the extension of the
predicate F in the standard sentential schema used for predicate logic, is
nonnull—that is, when, in the world as it is, [[F]] contains at least one actually
existing object properly characterized by the predicate F. This is unbearable
to a logician, because logic is meant to be based on meanings only, not
on contingencies in the world. When ABPC is applied to a situation where
[[F]] ¼
, then, given that it is based on a strictly extensional ontology of
actually existing objects only, just as SMPC does, an inconsistency arises
under universal quantification. Consider the sentences (4.1a) and (4.1b),
whose universally quantified subject terms contain a predicate expression
[[F]] with a null extension, either because of the contingent conditions of
the world, as with the predicate be a dodo (the dodo became extinct around
the year 1700), or because of a semantic inconsistency, as with the predicate be
living dead. The question is whether sentences like (4.1a) or (4.1b) are true or
false in the actual world:
(4.1) a. All dodoes are in good health.
b. All living dead are in good health.
124 The Logic of Language

If they are true, it follows in ABPC (by the subaltern entailment schema) that
their existential counterparts should also be true:
(4.2) a. Some dodoes are in good health.
b. Some living dead are in good health.
The sentences of (4.2), however, entail, by the entailment schema of EXISTENTIAL
IMPORT, that there is at least one actually existing dodo and at least one actually
existing entity that is dead while being alive, neither of which is the case.
In ABPC, SMPC and other varieties of predicate logic, existential import is
semantically induced by the existential quantifier (SOME), which requires that
there be a nonnull intersection of [[F]] and [[G]]. In a strictly extensional
ontology, this means the actual existence of at least one element in the
intersection. Since this condition is not fulfilled in (4.1a,b), these sentences
cannot be true. But if they are false, then their negations (4.3a) and (4.3b),
respectively, should be true:
(4.3) a. Not all dodoes are in good health.
b. Not all living dead are in good health.
Now, however, the problem appears again, because, owing to the duality
relation between the universal and the existential quantifiers (the Conver-
sions), (4.3a,b) are equivalent to (4.4a,b), respectively:
(4.4) a. Some dodoes are not in good health.
b. Some living dead are not in good health.
And here, existential import rears its head again, as (4.4a,b) entail again that there
is at least one actually existing dodo and at least one actually existing entity that is
dead while being alive, respectively—quod non. Therefore, (4.1a,b) cannot be false
either. And since ABPC does not allow for any value but true or false and does not
allow for the absence of a truth value, ABPC appears to be in trouble.
More technically minded readers may wish to see a more formal definition
of undue existential import. All right then, here are two possible definitions:
A system of predicate logic suffers from UNDUE EXISTENTIAL IMPORT when
every admissible expression in the system entails a proposition of the
type I or I*—that is, entails the existence of at least one entity of the F-
class quantified over.
Alternatively, a system of predicate logic suffers from UNDUE EXISTENTIAL
IMPORT when there is a proposition or proposition type T such that both
T and ¬T entail a nonnecessary (contingent) proposition of the type I or
I*, which makes I or I*, as the case may be, a necessary truth.
 Logical power 125

Figure 4.5c shows that this is indeed the case in ABPC. To solve this
problem, something must be done. The solution embodied in Russellian
SMPC consists in cutting out the subaltern entailment schema A ‘ I and
declaring (4.1a,b) true in the actual world, in accordance with standard set
theory which says that the null set
is included in all sets. This solution has
the advantage of mirroring mathematical set theory and thus of agreeing with
those forms of mathematics that lend themselves to application to physical
matter. For that reason, Russellian logic has had a great career during the
twentieth century and has acquired unique prestige. Yet it has landed the
study of language in another dilemma, this time not of a logical but of an
empirical nature, because SMPC grossly offends natural linguistic intui-
tions—much more than traditional ABPC. Parsons’ description of the math-
ematical logicians’ typical, not altogether forthcoming, reaction is true to life
(Parsons 2006: 3):
The common defense of this is usually that this is a logical notation devised for
purposes of logic, and it does not claim to capture every nuance of the natural
language forms that the symbols resemble. So perhaps ‘8 x(S x! P x)’ does fail to
do complete justice to ordinary usage of ‘Every S is P ’, but this is not a problem with
the logic. If you think that ‘Every S is P ’ requires for its truth that there be S s, then you
can have that result simply and easily: just represent the recalcitrant uses of ‘Every S is
P’ in symbolic notation by adding an extra conjunct to the symbolization, like this: 8 x
(Sx ! Px) & ∃xSx. This defense leaves logic intact and also meets the objection,
which is not a logical objection, but merely a reservation about the representation of
natural language.

This, however, is mere palliative therapy. It amounts to saying that all one
should do to appease speakers’ logical conscience is define ALL F is G not just as
[[F]] [[G]] but as [[F]] [[G]] and [[F]] 6¼
, adding the clause ‘and [[F]] 6¼
’,
and all is well. But all is not well, because if one does that, De Morgan’s laws
make NOT ALL F is G come out as meaning ‘either [[F]] / [[G]] or [[F]] ¼
’,
which again violates natural intuitions: SOME F is NOT-G clearly implies
intuitively that NOT ALL F is G (in fact, both SMPC and ABPC take the two
to be equivalent). But SOME F is NOT-G stipulates the existence of at least one F
and thus rules out the possibility that [[F]] ¼
, which disqualifies the disjunct
‘or [[F]] ¼
’. The addition of the clause ‘and [[F]] 6¼
’ to the definition of the
universal quantifier thus merely makes the clash of intuitions rear its head
elsewhere. Therefore, whichever way one takes it, SMPC does not sit at all well
with the facts of language. (See Section 5.2.4 for further comment.)
In overall perspective, the following steps are taken. First, in accordance
with the arguments set out in Chapter 2 of Volume I, we give up the
126 The Logic of Language

restriction to a strictly extensional ontology and let in the full realm of

intensional entities. This is necessary because a proposition is by definition
contextually anchored and thus automatically creates the intensional objects
needed for reference when there are no corresponding actually existing
objects, according to the reference hierarchy described in Section 3.5.2 of
Volume I. This makes it intrinsically impossible for [[F]] to be null, since the
very fact that an object, or a class of objects, has been conceived implies the
virtual, or intensional, being of that object or class of objects and thus makes
it possible for that object to be referred to or for that class of objects to be
quantified over. Views of this nature were often aired during the nineteenth
century, but they never caught on, due not only to the unclarity, during that
period, regarding the nature of intensional objects but also to a lack of formal
expertise. When accepted, this view, in effect, removes the obstacle to the
subaltern entailment schema from A-type to I-type L-propositions and allows
for the reinstatement of the subalterns. This means that SMPC can be
discarded for the purposes of language and cognition—though not for the
purposes of the nonmentalistic worlds of physical matter and mathematics, to
which partially defined mind-created intensional objects as described in
Chapter 2 of Volume I are alien.
We thus take it that natural language operates with universal and existential
quantifiers that lack existential import but merely induce an entailment of
being, which can be actual existence or virtual being, depending on whether the
position of the term quantified over with regard to the matrix predicate is or is
not extensional. An extensional term position induces the entailment of exis-
tence; an intensional term position does not. The difference is taken to be
presuppositional: each predicate does or does not presuppositionally require the
actual existence of the referent of any of its terms. In order to uphold both
logical consistency and empirical adequacy, the Aristotelian principle of strict
bivalence is relativized so as to allow for two kinds of falsity, RADICAL FALSITY for
presupposition failure and MINIMAL FALSITY for failure to satisfy the nonpresup-
positional, or update, conditions of the matrix predicate. This makes ALL /SOME
MERMAIDS are IMAGINARY true in the actual world, because the matrix predicate
imaginary is intensional with respect to its subject term. But given that
the predicate have a bank account is extensional with regard to its subject
term, ALL /SOME MERMAIDS have A BANK ACCOUNT is false in the actual world—
in fact, radically false since the falsity rests on presupposition failure. (It can
achieve truth only when it occurs as an instance of implicit intensionalization,
as explained in Chapter 5 of Volume I.) Thus, other than in SMPC, ALL F is G
and SOME F is G are radically false when the [[F]]-class is extensionally null and
the matrix predicate G requires actual existence for the subject-term referent,
 Logical power 127

but they may be true when the matrix predicate does not pose that require-
ment. As far as the language system is concerned, existential import is thus
properly regulated by the lexicon, not by the machinery or the axioms of logic.
This allows language to profit from the maximal logical power of ABPC
within the confines of those situations where all presuppositions are fulfilled
and thus within the confines of the default conditions of linguistic interac-
tion. Chapter 10 contains a detailed description of this logic, which allows for
intensional entities.
Thus protected, ABPC can be saved for language and cognition, though
perhaps not for general mathematics and the physical sciences. The question
is whether the protective measures are justified and how they actually work.
The arguments for intensional objects were given in Chapter 2 of Volume I, so
that we consider that question settled. The presuppositional machinery and
the concomitant logic are defined in Chapter 10. The history and precise
analysis of ABPC is sketched in Chapter 5, while its surprising functionality is
demonstrated in Chapter 6. What remains to be done is show the logical
power of ABPC and of propositional calculus (which, unlike predicate calcu-
lus, has not been superseded by a modern variant). This is what is undertaken
in the following section.

4.2 The notion of logical power

It seems intuitively reasonable to assume that a logical system  is less
powerful to the extent that there are more logically independent pairs of
basic expression types in the logical language defined for . We call basic
expression types the expression types in terms of which the logical system 
has been defined (without vacuous repetitions of negations). We speak of
logical independence of a pair of basic expression types just in case no
(nonnegative) logical relation holds between its members. The term pair is
used here not in the sense of ‘ordered pair’ but merely to refer to a set
consisting of exactly two members, so that, for any entities a and b, <a,b>
and <b,a> are the same pair.
On this basis we may introduce the notion of LOGICAL POWER of a logical
system , based on the number and proportion of logically independent pairs
of basic expression types in : the more logically independent pairs of basic
expression types, the weaker the system. The sum total of the pairs of basic
expressions in  that are not logically independent then constitutes the logical
power of . Other, more refined, metrics are possible, but all we want to
achieve at this moment is to provide some formal support for what is already
clear intuitively.
128 The Logic of Language

Propositional calculus is defined in terms of the following eight basic types,

as specified in Section 2.4.2 (P and Q range over L-propositions):
AND stands for P ∧ Q ∧… ¬AND stands for ¬(P ∧ Q ∧…)
OR stands for P ∨ Q ∨… ¬OR stands for ¬(P ∨ Q ∨…)
AND* stands for ¬P ∧ ¬Q ∧… ¬AND* stands for ¬(¬P ∧ ¬Q ∧…)
OR* stands for ¬P ∨ ¬Q ∨… ¬OR* stands for ¬ (¬P ∨ ¬Q ∨…)
Analogously for the varieties of predicate calculus we are considering, which
are all defined in terms of the following twelve basic types (F and G range over
predicates):
A stands for ALL F is G ¬A stands for ¬(ALL F is G)
I stands for SOME F is G ¬I stands for ¬(SOME F is G)
N stands for NO F is G ¬N stands for ¬(NO F is G)
A* stands for ALL F is NOT-G ¬A* stands for ¬(ALL F is NOT-G)
I* stands for SOME F is NOT-G ¬I* stands for ¬(SOME F is NOT-G)
N* stands for NO F is NOT-G ¬N* stands for ¬(NO F is NOT-G)
(In AAPC, ABPC, and SMPC, N
¬I, which reduces the number basic types
to eight.)
The logical power of the systems under consideration can be visualized in a
simple way by arranging their basic types in a polygon with the corresponding
basic types as vertices and connecting lines between any two vertices symbo-
lizing the logical relations defined for the system. For an octagonal arrange-
ment, there are at most 28 such relations, so that the maximal score for the
systems considered is 28. Anything less means a loss of logical power.
One remembers from Chapter 3 that BNPC has twelve basic expressions,
requiring a dodecagonal representation as in Figure 3.7c or 3.8c in Section
3.4.1. And since the dodecagon that represents BNPC is complete (fully filled
in) (see Figure 3.7c), one must conclude that BNPC is the most powerful
system of predicate logic in existence, more powerful even than ABPC and
vastly more powerful than SMPC, the mathematics-based standard modern
system.
It may seem more paradoxical to some than to others, but it does appear as
if the most primitive natural system of predicate logic is at the same time the
most powerful, and that the power of predicate-logic systems decreases the
more closely their operators reflect the Boolean functions set in mathematical
space. For the time being it is hard to interpret this fact in the context of
philosophical questions relating to the position of mankind in the physical
world. But it does seem certain that it will prove to be of at least some
significance in future discussions of this nature.
 Logical power 129

4.2.1 The logical power of propositional calculus

But before we proceed to computing the values for the systems concerned, let
us revert to the Square notation, which is where we left off in Chapter 2. First
we consider propositional calculus, in particular the entailment schema AND ‘
OR, which gives rise to a natural logical triangle as shown in Figure 2.1, with AND,
OR, and ¬OR for P, Q, and ¬Q, respectively. Given the equivalences resulting
from De Morgan’s laws, a second, isomorphic triangle with the internal negation
can be added, whose vertices are AND*, OR*, and ¬OR*. The two triangles link up
by the De Morgan equivalences at the AND and ¬OR* vertices and again at the
¬OR and AND* vertices.
The logical properties of the two triangles are identical, as they should be,
since whether or not the argument L-propositions are negated is irrelevant for
the logic of AND and OR: all that counts is the logical properties of the truth-
functional operators themselves. The subcontrariety between OR and OR* is an
additional bonus: since ¬OR and ¬OR* are contraries and thus cannot both be
true at the same time, their contradictories OR and OR* cannot both be false at
the same time and thus form a pair of subcontraries. Figure 4.1 shows the
result (repeated from Figure 3.15b).
Yet although Figure 4.1 may contain the whole story, it does not show it.
A minimalist graph consisting of just the equivalences and the contrariety of
AND and ¬OR likewise contains the whole story, but to show it, more is
needed. A more revealing way of displaying the calculus is by means of a
hexagon, as in Figure 4.2, which specifies more logical relations than the square
notation. The two natural triangles, printed in heavy lines, have remained
unchanged, but more logical relations have been added, such as the entailment
from ¬OR to OR*. Yet, compared with Figure 4.1, Figure 4.2 does not add new
information, since those logical relations that are shown in Figure 4.2 and not in

AND ¬OR* OR*

CD

C C
SC

CD
OR ¬OR AND*
FIGURE 4.1 Propositional calculus represented as a natural square formed by two
logically isomorphic natural triangles
130 The Logic of Language

AND ¬OR*

CD

C C CD

C C

OR SC OR*

CD

CD

¬OR AND*
FIGURE 4.2 Hexagonal graph for propositional calculus

Figure 4.1 can easily be deduced from the latter. All Figure 4.2 does is make
explicit some logical relations that are not shown in Figure 4.1. But if that is
what we want, we should present a model that shows all logical relations
between all possible vertices.
A complete representation requires an octagonal model, as in Figure 4.3,
where the natural triangles are again printed in heavy lines. To facilitate the
checking of the logical relations specified in the octagon, the VS of each vertex
is indicated, according to the VS-model of Figure 4.4 and listed in (4.5) below.
This octagon, with its dense network of metalogical relations, may look
forbidding, yet it should be remembered that it simply follows from the
combination of the subaltern entailment (AND ‘ OR) with standard bivalent
negation and the De Morgan equivalences.
In the octagon of Figure 4.3 all edges between vertices represent some
logical relation, giving a complete octagonal graph. No edge between two
vertices would mean that the corresponding L-propositional types are logi-
cally independent: their VSs have a nonnull intersection but neither includes
the other and their union does not equal U. The absence of logically indepen-
dent pairs makes the graph COMPLETE in the standard terminology of graph
theory. This puts the logical power of this system at the maximum of 28,
according to the rough metric introduced earlier.
Figure 4.4 shows propositional calculus in terms of a valuation space (VS)
model. It subdivides U in terms of the eight L-propositional types AND, OR,
AND*, OR*, and their negations. Space 1 is reserved for cases where all com-
ponent L-propositions are true, space 2 for those where one or more are true
and one or more are false, and space 3 for cases where all are false. For example,
 Logical power 131

¬AND {2,3}
CD CD
{1} {1}
AND ¬OR*

CD
C
C CD
C
SC
{1,2} {2,3}
OR SC OR*

CD CD
C

¬OR AND*
SC
{3} CD CD {3}

¬AND* {1,2}
FIGURE 4.3 Octagon as a complete graph for propositional calculus

given the two logically independent sentences The earth is round and Venus is
a planet, U is subdivided for the eight sentences The earth is round AND Venus is a
planet, The earth is round OR Venus is a planet, The earth is NOT round AND Venus
is NOT a planet, The earth is NOT round OR Venus is NOT a planet, and their

¬AND OR*

¬AND OR*

AND ¬OR*

3 2 1 2 3

OR ¬AND*

OR ¬AND*

¬OR AND*
U

FIGURE 4.4 VS-representation of standard propositional calculus

132 The Logic of Language

external negations. When the two (or more) component L-propositions are not
logically independent because, say, there is a contrary pair among them, then
/AND/ ¼
since their conjunction is never true. When the component L-
propositions exhaust U, then /OR/ ¼ U, because their disjunction is always
true. (The reader may try to work out for himself or herself what the VS-
model will look like when the component L-propositions of AND and OR are
equivalent: P ∧ P, P ∨ P, ¬(P ∧ P), ¬(P ∨ P), etc.)
The simple diagrame of Figure 4.4, again, represents the whole of standard
propositional calculus. The VSs of the various L-propositional types are specified
as follows (the numbers stand for the spaces as they are numbered in Figure 4.4):
(4.5) /AND/ = {1} /¬AND/ = {2,3}
/OR/ = {1,2} /¬OR/ = {3}
/AND*/ = {3} /¬AND*/ = {1,2}
/OR*/ = {2,3} /¬OR*/ = {1}

Since {1}  {1,2}, the entailment AND ‘ OR holds. Likewise for the entailment
AND* ‘ OR*, since {3}  {2,3}. Moreover, AND and ¬OR* are equivalent since
their VSs coincide. Likewise for OR and ¬AND*. Then, AND and AND* are
contraries because their VSs do not intersect. And OR and OR* are subcontra-
ries because the union of their VSs taken equals U: {1,2} [ {2,3} ¼ {1,2,3} ¼ U.
It is thus easily seen that the VS representation of Figure 4.4, when written as
an octagon, returns Figure 4.3.

4.2.2 The logical power of Aristotelian-Boethian predicate calculus

Let us now do for ABPC what we did for propositional calculus, disregarding,
for the moment, the blemish of undue existential import. In the light of what
has been done for propositional logic, this exercise is simple, because ABPC is
defined by the same logical relations, including the equivalences, due to the
duality of the universal and existential quantifiers. One only has to replace
AND with A and OR with I. The result is Figure 4.5 (repeated from Figure
3.10), with the VS specification (4.6). Note that space 1 is reserved for cases
where [[F]]  [[G]] and [[F]] 6¼
; space 2 for cases where either [[F]] O O [[G]]
(M-partial intersection: [[F]] \ [[G]] 6¼
6¼ [[F]] 6¼ [[G]]) or [[G]]  [[F]] (so
that [[F]] 6¼
); space 3 for cases where [[F]] OO [[G]] ([[F]] and [[G]] are
mutually exclusive) and [[F]] 6¼
. As with propositional calculus, the system
has maximal logical power, in the terms in which this notion has been
defined, given the subaltern entailment schema from A to I, given standard
bivalent negation and given the Conversions due to duality.
 Logical power 133

a. A ¬I* I*
CD

C c. ¬A {2,3}
C
SC CD CD
{1} A ¬I*{1}
CD
C CD
C
CD C
I ¬I A* SC
SC
b. I SC I*
{1,2}
¬A I* {2,3}

¬A I* CD CD
C
A ¬I*
3 2 1 2 3

I ¬A* {3} ¬I A* {3}

SC
CD CD
I ¬A*
¬I A* ¬A* {1,2}
U

FIGURE 4.5 The Square, the VS-model, and the complete octagonal graph of ABPC

(4.6) /A/ = {1} /¬A/ = {2,3}

/I/ = {1,2} /¬I/ = {3}
/A*/ = {3} /¬A*/ = {1,2}
/I*/ = {2,3} /¬I*/ = {1}

The VS-modelling of ABPC given in Figure 4.5b shows again the fault
of undue existential import: since there is no space where both I-type and
I*-type L-propositions are false, there is no space for those situations where
[[F]] ¼
. Therefore, a fourth space is required containing those situations where
[[F]] ¼
. The logical power of ABPC is again 28, as for propositional calculus.

4.2.3 The logical power of standard modern predicate calculus

We now do the same for SMPC, as before under a strictly extensional
ontology. In SMPC, the generalized universal and existential quantifiers are
defined as in (2.14) of Section 2.3.5.2, repeated here as (4.7):
(4.7) For all sets X and Y:
a. [[8]] ¼ { <Y,X> j X  Y }
(the extension of the predicate 8 is the set of all pairs of sets Y, X,
such that X is a subset of Y)
134 The Logic of Language

b. [[∃]] ¼ { <Y,X> j Y \ X 6¼
}
(the extension of the predicate ∃ is the set of all pairs of sets Y, X,
such that the intersection of Y and X is nonnull)
ALL F is G is considered true just in case [[F]]  [[G]], and SOME F is G is
considered true just in case [[F]] \ [[G]] 6¼
. When [[F]] ¼
, ALL F is G is
automatically true because, in set theory,
is a subset of any set (for all sets X
·
and Y, X  Y iff X Y ¼ X and X þ Y ¼ Y—a condition always fulfilled when
X ¼
).
Figure 4.6 (repeated from Figure 3.11) does for SMPC what Figure 4.5
does for ABPC. In Figure 4.6a, space 1 represents cases where [[F]]  [[G]]
and [[F]] 6¼
; space 2 cases where either [[F]] O O [[G]] or [[G]]  [[F]] and
[[F]] 6¼
; space 3 cases where [[F]] OO [[G]] and [[F]] 6¼
; and space 4 cases
where [[F]] ¼
—the class of situations absent in ABPC. In space 4, A-type
and A*-type L-propositions both count as true and I-type and I*-type
L-propositions as false.
Figure 4.6a parallels Figures 4.5b and 4.4, but whereas Figure 4.4 is in no need
of a fourth space, Figure 4.6a must have one, since, without it, it fails to cater for
situations where [[F]] lacks the required supply of extensional objects. Standard
propositional calculus does not need a counterpart to extensionally null pre-
dicates: a set may be null, but a proposition has nothing to reciprocate with.
Under a strictly extensional ontology, the system of Figure 4.6 is logically
sound: there is no undue existential import and the system is fully consis-
tent—even if it clashes with natural intuitions. But look what has happened to
its logical power, which now amounts to a mere 12. The extra space 4, with
truth for A-type sentences, destroys most of the beautiful, rich logic of Figure
4.5. The logically complete and maximally powerful octagonal graph of Figure

a. b. ¬A {2,3}
[[F]] = Ø CD
{1,4} CD {1,4}
A ¬I* ¬I*
A
¬A I*
CD CD
¬A I* {1,2}
A ¬I* I
4 3 2 1 2 3 4
I ¬A* CD
I*
I ¬A* CD {2,3}
¬I A*
A* A*
¬I ¬I {3,4}
{3,4} CD CD
U ¬A* {1,2}

FIGURE 4.6 The VS-model of SMPC and the poor remnants of its octagonal graph
 Logical power 135

4.5c has been largely dismantled and the traditional Square has vanished
altogether. In effect, the octagon has disappeared and given way to two
isomorphic trapezoids, shown in Figure 4.6b, whose isomorphism reflects
the elementary fact that the internal negation may be added without con-
sequences for the semantics of the vertices (note that A**
A and I**
I).
The entire logical power of the trapezoids derives from the equivalence
¬I
A* (or ¬I*
A)—that is, from the Conversions. Almost all connections
between vertices have been lost, which means that the vertex pairs in question
represent the vacuous relation of logical independence.
The striking news about SMPC is, of course, its heavy loss in logical power
when compared with ABPC, whose 28 logical relations have dwindled to a paltry
twelve: two quadruples <A, I*,¬I*,¬A> and <A*, I,¬I,¬A*>, which are logi-
cally isomorphic owing to the Modulo*-Principle. This loss of logical power is
solely due to the stipulation made in SMPC that A- and A*-sentences are true in
space 4, where [[F]] ¼
. But alas, ABPC is logically faulty, as it fails to cover
situations where G requires nonnull membership of [[F]] yet [[F]] fails to oblige.
To make things worse, SMPC also clashes badly with natural intuitions
about truth and falsity. In fact, for natural language SMPC is an unmitigated
disaster. Consider, for example, sentence (4.8), said by a mechanic to justify
an exorbitant bill for the servicing of a car with a diesel engine, which, as one
knows, has no spark plugs:
(4.8) All spark plugs have been changed.
For SMPC this sentence is true, but any judge presiding over the case brought
by the car owner against the mechanic will consider the latter a liar.
Or take the sentences (4.9a,b), which, in the SMPC book, should both be
counted as true in any situation, common enough in the actual world, where
there are children but no real baby dinosaurs:
(4.9) a. Some children played with all baby dinosaurs.
b. Some children didn’t play with any baby dinosaur.
In fact, given the absence of baby dinosaurs, SMPC makes both sentences
equivalent to the statement that there was at least one child. Yet ordinary
people will consider (4.9a) false in a situation with at least one child and no
baby dinosaurs. And (4.9b) may have to count as true in such a situation, but
only in a trivial and uninformative way.
Pragmatic principles make both sentences equally inappropriate in the
actual world, where everyone knows that there are no living baby dinosaurs.
To utter such sentences will thus violate a number of Gricean maxims. Yet
136 The Logic of Language

these maxims fail to explain why (4.9a) is felt to be false while (4.9b) is
considered true by unsuspecting speakers. More examples of this nature are
easily thought up. In practically all cases the conclusion is that ABPC fits
natural intuitions much better than SMPC, even though the latter reigns
supreme in the world of modern logic and the former is still far from full
empirical adequacy.

4.2.4 The logical power of Aristotelian-Abelardian predicate calculus

Does this mean that SMPC is the best deal possible, within the constraints of a
strictly extensional ontology? Far from it. There is at least one alternative way
to repair ABPC in such a way that the loss of logical power sustained by SMPC
is significantly reduced while natural linguistic intuition is served more
adequately. This remedy was proposed by the medieval philosopher Peter
Abelard (1079–1142) and it is most probably the logic that Aristotle himself
envisaged but failed to elaborate. (Chapter 5 is devoted to a detailed historical
and text-critical analysis of this question.) We call this system ARISTOTELIAN-
ABELARDIAN PREDICATE CALCULUS or AAPC.2
The gist of the Abelardian solution lies in keeping the subalterns but
giving up the Conversions for one-way entailments from A to ¬I* and,
analogously, from A* to ¬I, but not vice versa. This step results in the system
shown in Figure 4.7 (repeated from Figure 3.9). Figure 4.7b does have the
required fourth space for cases where [[F]] ¼
but where, significantly, A-type
L-propositions where [[F]] ¼
are not considered true, as in SMPC, but false,
which agrees much better with natural intuitions. This is the only difference
between SMPC and AAPC, as is easily checked by comparing the VS-models of
Figures 4.6a and 4.7b: in Figure 4.7b, A and A* are false in space 4, while in
Figure 4.6a A and A* are true. It is amazing to see how such a small difference
can provoke such a large difference in logical power.
Undue existential import has been eliminated in AAPC because A is false
when [[F]] ¼
and, in cases where A is false, nothing follows with regard to
any other L-propositional type: as shown in Figure 4.7c, the ¬A-vertex entails
no other vertex and occurs only at the receiving end of some other vertices. As
a reward for the weakening of the Conversions, the octagon has lost less
logical power than the depleted system of SMPC: the metric adopted puts its
logical power at 24, so that it scores better than SMPC, which achieves a mere

2
In Seuren (2002) a solution is proposed for undue existential import that is identical to Abelard’s.
I named it Revised Aristotelian predicate calculus or RAPC. At the time, I was not aware of the fact that
Abelard had proposed an identical solution nine hundred years earlier. To give Abelard his due, I have
renamed the system in question Aristotelian-Abelardian predicate calculus.
 Logical power 137

a. A ¬I* I* c. ¬A {2,3,4}
CD CD SC
{1} A ¬I* {1,4}
C C C C

C CD
CD
I ¬I A* {1,2} SC {2,3}
I C
b. C C
I*
[[F]] = Ø
¬A ¬I* CD

¬A I* SC

¬A I* A* {3}
{3,4}¬I SC
A ¬I* SC CD
4 3 2 1 2 3 4
I ¬A* ¬A* {1,2,4}
I ¬A*
¬I A*
¬I ¬A*

FIGURE 4.7 The square, the VS-model, and the octagon for AAPC

12, but less well than ABPC, which comes to 28. Moreover, intuitions are
better respected. AAPC has thus freed itself from all the restrictions of natural
set theory, whether basic or strict, but has been able to reduce the price for this
freedom considerably by adopting a definition of the universal quantifier 8
that differs from the definition adopted in SMPC.
It thus appears that, from a strictly logical point of view, ABPC represents a
retrograde development with regard to AAPC, which preceded it in time (if
one grants Aristotle the honour of being the originator of AAPC). Not only
did ABPC introduce the logical defect of undue existential import, it also
failed to eliminate the principles of natural set theory. Yet this logical blunder,
if that is what one wishes to call it, was offset by a great advance as regards the
functionality of predicate logic in real life situations. As is shown in Chapter 6,
ABPC has the advantage of cutting out the informational redundancy of
SMPC and it has greater logical power.
It is, therefore, questionable whether SMPC does indeed deserve the exalted
status of inviolate doctrine it enjoys. For nonmentalistic, purely extensional
applications, for which SMPC is fully valid, AAPC is likewise fully valid, but it
appears to be a better, because more powerful, logic. Nowhere do the found-
ing fathers of modern logic or their followers provide reasons for keeping the
138 The Logic of Language

Conversions and giving up the subalterns rather than the opposite. In fact,
they appear to have opted for the most immediately obvious, but not neces-
sarily the most useful, application of Boolean algebra to predicate calculus.
This, however, is a question that falls outside the scope of the present work.
What we can show—and do show in the following section—is that it is not
too hard to provide a simple and logically interesting mathematical charac-
terization of the AAPC quantifiers.
Yet no matter how interesting, perhaps even revolutionary, the Abelardian
solution to the problem of undue existential import may be, it does insuffi-
cient justice to the requirements of cognitive realism (Section 1.3.1 in
Volume I), as it fails to take into account the fact that natural language
quantifies with equal ease over extensional as it does over intensional objects,
as well as the fact that certain predicates yield truth when applied to inten-
sional objects, as is shown in Chapters 2 and 5 of Volume I. For that reason
AAPC cannot be considered a viable candidate for the post of logic of
language, even though it is a highly interesting and intriguing pointer.
Therefore, if it proves possible to resolve or circumvent the problem of
undue existential import, ABPC, also known as the Square of Opposition, will
turn out, after all, to be the most preferable predicate-logic system for natural
language and cognition, pace our mathematical friends.

4.3 Distributive quantifiers

To make AAPC work, the definition of the universal quantifier must be
changed. One way of doing this is by adding the condition X 6¼
to the
condition specified for the generalized universal quantifier in (2.14a) of
Section 2.3.5.2, so that 8 is semantically defined as follows:
(4.10) For all sets X and Y:
[[8]] ¼ { <Y,X> j X  Y and X 6¼
}
(the extension of the predicate 8 is the set of all pairs of sets Y, X, such
that X is a subset of Y and X is nonnull)
This does the job, at least for purely extensional purposes, but not in a very
illuminating way. It merely patches up the standard definition for the purpose
of natural language, thereby destroying the Conversions and, in effect, turning
SMPC into AAPC.
A more principled way is by means of what we call the DISTRIBUTIVE
QUANTIFIERS. This solution can not only be made to encompass intensional
as well as extensional entities, it also shows up the parallelism between, on the
one hand, universal quantification and conjunction and, on the other,
 Logical power 139

existential quantification and disjunction. As was said in Section 4.1, the universal
quantifier yields truth just in case the conjunction G(a1)∧G(a2)∧ . . . ∧G(an) is
true, where a1, a2, . . . , an denote the elements in the extension [[F]] of F.
Analogously, the existential quantifier yields truth just in case the disjunction
G(a1)∨G(a2) ∨. . .∨ G(an) is true, where, again, a1, a2, . . . , an denote the
elements in [[F]]. The reason why this correspondence is not at all popular
in modern standard logic lies in the fact that the correspondence is lost when
[[F]] ¼
, because then there is no conjunction and no disjunction. In the present
perspective, however, this is not a weakness but, rather, an aspect that can be
turned to the advantage of AAPC.
Just like the Russellian quantifiers (see (2.9) in Section 2.3.5.1), the distributive
quantifiers are defined as unary higher-order predicates over sets. Other than
the Russellian quantifiers, however, and more like the generalized quantifiers,
the distributive quantifiers are defined with respect to a designated predicate
(F) and require that the members of [[F]] satisfy the condition of the quantifier
with respect to the extension of the matrix predicate G. Given the predicates
F and G, the condition for 8F is, in simple terms, that for all x 2 [[F]], G(x) is
true, whereas ∃F requires merely that for at least one x 2 [[F]], G(x) is true.
The quantifiers are thus, in fact, predicates over restrictor-predicate exten-
sions with respect to a matrix predicate G. When it is said that All flags are
green, or Some flags are green, then these sentences are interpreted as state-
ments about the set of flags. Provided [[Flag]] 6¼
, the statement is that, for
8Flag, truth is achieved only if Green(a1) ∧ Green(a2) ∧ . . . ∧ Green(an) is
true, and, for ∃Flag, only if Green(a1) ∨ Green(a2) ∨ . . . ∨ Green(an) is
true, where a1, a2, . . . , an denote the elements in [[Flag]].
On the basis of this we say that, for 8F, the set of objects [[F]] must satisfy the
predicate G under conjunction and that, for 8F, the set of objects [[F]] must satisfy
the predicate G under disjunction, for truth to come about. The notion of a set of
objects satisfying a predicate under a propositional operator is defined as follows:
The set of objects [[F]] SATISFIES THE PREDICATE G UNDER CONJUNCTION just in
case either F(a)∧G(a) is true, where a is the only element in the extension
of F (¼ [[F]]), or there is a true conjunction [F(a1) ∧ G(a1)] ∧ . . . ∧
[F(an) ∧ G(an)], where a1, . . . , an form the extension of F.
The set of objects [[F]] SATISFIES THE PREDICATE G UNDER DISJUNCTION just in case
either F(a)∧G(a) is true, where a is the only element in the extension
of F (¼ [[F]]), or there is a true disjunction [F(a1) ∧ G(a1)] ∨ . . . ∨ [F(an)
∧ G(an)], where a1, . . . , an form the extension of F.
140 The Logic of Language

This enables us to define the distributive universal and the existential

quantifier with respect to the restrictor predicate F as follows:
(4.11) For all predicates F and G:
a. [[8F]] ¼ { [[F]] j [[F]] satisfies the predicate G under conjunction }
b. [[∃F]] ¼ { [[F]] j [[F]] satisfies the predicate G under disjunction }
Now, when [[F]] ¼
, falsity results for both the universal and the existential
quantifier. This is so because, in the absence of any x 2 [[F]], F(x)∧G(x) is
false for any element x in U, which makes both ALL F is G and SOME F is G
false. It is easily seen that this predicate logic represents the AAPC variant of
the cognitive approach to predicate logic described in Section 3.4.1.
To facilitate understanding, a toy VS-model can be set up illustrating how
these definitions work. Such a model must be kept extremely small, because
the number of valuations increases exponentially with the number of ele-
ments and/or predicates. Consider a toy model Mtoy with two semantically
independent predicates F and G and three elements a, b, and c. The total
number of valuations (situations) for Mtoy is (23)2 ¼ 64, as shown in Figure
4.8. (With four elements and two predicates, the total number of valuations or
situations would be (24)2 ¼ 256.)
Figure 4.8 shows Mtoy written out as a system of separate valuations, each of
which marks each basic proposition and each conjunction of the form
F(x) ∧ G(x) plus (true) or minus (false), thereby giving a breakdown of U.

Val: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
Fa + - + - + - + - + - + - + - + - + - + - + - + - + - + - + - + -
Fb + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - -
Fc + + + + - - - - + + + + - - - - + + + + - - - - + + + + - - - -
Ga + + + + + + + + - - - - - - - - + + + + + + + + - - - - - - - -
Gb + + + + + + + + + + + + + + + + - - - - - - - - - - - - - - - -
Gc + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +

Fa ∧ Ga + - + - + - + - - - - - + - - - + - + - + - + - - - - - - - - -
Fb ∧ Gb + + - - + + - - + + - - + + - - - - - - - - - - - - - - - - - -
Fc ∧ Gc + + + + - - - - + + + + - - - - + + + + - - - - + + + + - - - -

Val: 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64
Fa + - + - + - + - + - + - + - + - + - + - + - + - + - + - + - + -
Fb + + - - + + - - + + - - + + - - + + - - + + - - + + - - + + - -
Fc + + + + - - - - + + + + - - - - + + + + - - - - + + + + - - - -
Ga + + + + + + + + - - - - - - - - + + + + + + + + - - - - - - - -
Gb + + + + + + + + + + + + + + + + - - - - - - - - - - - - - - - -
Gc - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

Fa ∧ Ga + - + - + - + - - - - - - - - - + - + - + - + - - - - - - - - -
Fb ∧ Gb + + - - + + - - + + - - + + - - - - - - - - - - - - - - - - - -
Fc ∧ Gc - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

FIGURE 4.8 Mtoy for two predicates F and G and three objects called a, b, and c
 Logical power 141

A set of objects [[F]] satisfying the predicate G under conjunction is found in

precisely the following set V8F of situations (valuations):
V8F ¼ f1; 2; 3; 4; 5; 6; 7; 10; 12; 14; 19; 20; 23; 28; 37; 38; 39; 46; 55g:
And the set of valuations V∃F such that for each valuation in V∃F[[F]] satisfies
the predicate G under disjunction is:
V∃F ¼ f1; 2; 3; 4; 5; 6; 7; 9; 10; 11; 12; 13; 14; 17; 18; 19; 20; 21; 23; 25;
:
26; 27; 28; 33; 34; 35; 37; 38; 39; 41; 42; 45; 46; 49; 51; 53; 55g
Since V8F V∃F, it follows that A-type L-propositions entail their I-type
counterparts.
If one reads Flag for F and Green for G, then ALL Flags are Green is true
in all and only the valuations (situations) of V8F because for all and only
these situations is it possible to enumerate the elements in [[Flag]] and
ascertain that they are also marked plus for the predicate Green. Notably, in
those situations where there is no flag—that is, in {8,16,24,32,40,48,56,64}—ALL
Flags are Green is false because there is no true conjunction of the required
sort. SOME Flag(s) is(are) Green is false when there are no flags or there are
flags but none of them is green—that is, in the complement of V∃F, namely
{8,15,16,22,24,29,30,31,32,36,40,43,44,47,48,50,52,54,56,57,58,59,60,61,62,63,64}.
The missing fourth space for situations where [[F]]¼
can thus be
incorporated into the VS-model of the system in Figure 4.7b, but, other
than in SMPC, universally quantified L-propositions are valued ‘false’ in
such situations, just like existentially quantified L-propositions. This amounts
to AAPC, Abelard’s amendment of Aristotle’s predicate logic.
The definitions of (4.11a,b) also shed some light on the fact that in order to
maintain the parallelism between propositional and predicate logic it is
necessary to distribute the internal negation over all arguments of the non-
unary propositional operators. One remembers from (2.30) in Section 2.4.2
that AND* and OR* distribute their internal negation over all their arguments:
AND* ¼ ¬P ∧ ¬Q ∧ ¬R ∧ . . . ∧ ¬Z and OR* ¼ ¬P ∨ ¬Q ∨ ¬R ∨ . . . ∨ ¬Z.
Under the definitions of (4.11a,b), A* and I* distribute their internal negation
over all the instances of the matrix predicate: if the matrix predicate is ¬G,
then the condition is, for A*, that sitact 2 /(F(a1) ∧ ¬G(a1)) ∧ . . . ∧ (F(an) ∧
¬G(an))/ and, for I*, that sitact 2 /(F(a1) ∧ ¬G(a1)) ∨ . . . ∨ (F(an) ∧ ¬G
(an))/, for all ai in [[F]]. If the propositional operators ∧ and ∨ can be
redefined in such a way that they extend over propositional functions under
a quantifier, the generalizing principle may be seen to stem from the quanti-
fiers. This aspect, however, is not elaborated here.
142 The Logic of Language

Given the fact that the quantifiers are now restrictor-bound predicates over
matrix-predicate extensions, their L-propositional syntax had better be
changed as well. Sentences like (4.12a,b) are now better rendered syntactically
as the L-propositions (4.13a,b), where the restrictor predicate is attached to
the quantifier. The syntax of the quantifier now corresponds to (2.10a,b) of
Section 2.3.5.2:3
(4.12) a. All farmers grumble.
b. Some farmers grumble.
(4.13) a. 8x[Farmer(x)](Grumble(x))
(for all objects x such that x is a farmer, x grumbles)
b. ∃x[Farmer(x)](Grumble(x))
(for at least one object x such that x is a farmer, x grumbles)
The corresponding L-propositional tree structure is now different from that
assumed for generalized quantifiers. On this analysis, a sentence like Some
farmers do not grumble is assigned the L-propositional tree structure of Figure
4.9. This configuration makes it possible to use the same variable x for both
the predicate farmer and the predicate grumble. In practice, this means that
the whole structure S1 is true just in case the disjunction ¬Grumble(a1) ∨
¬Grumble(a2) ∨. . . ¬Grumble(an) is true, where a1, a2, . . . , an range over
[[Farmer(x)]], or, in the terminology defined above, the set of farmers satisfies
the predicate not-Grumble under disjunction.

S1

Pred S3

Pred S2 Pred S4
SOME x NOT

Pred NP Pred NP
Farmer x Grumble x
FIGURE 4.9 L-propositional tree structure of Some farmers do not grumble

3
Compared with (2.11a,b) in Chapter 2, Figure 4.9 shows the result of the transformational
syntactic rule of OBJECT INCORPORATION or OI—a rule frequently found in the syntax of natural
languages. Examples are English expressions like take care of or pay attention to, where the object
terms care and attention have been incorporated into the predicate, as is seen from the passives She has
been taken care of or She has been paid attention to. Full lexicalizations where an object term has been
incorporated are, for example, English bearhunting, seafaring, globetrotting, golddigging, brew (‘make
beer’), price (‘put a price on’), pencil (‘use a pencil to write’). Such formations are frequent in many
other languages as well. See also note 6 in Chapter 2.
 Logical power 143

As long as [[F]] 6¼
, the distributive quantifiers 8 and ∃ convert in the

normal way, just like their counterparts in traditional and standard modern
predicate logic. But unlike modern predicate logic, in cases where [[F]] ¼
,
falsity results for both the universal and the existential quantifier, with or
without internal negation.
Of course, we want to know what happens to nonstandard quantifiers like
MOST or HALF when these are defined as distributive quantifiers. Let us look
again at (2.15a,b) of Chapter 2, repeated here in the form adapted as
necessary:
(4.14) a. Most farmers grumble.
b. MOST x[Farmer(x)][Grumble(x)]
The tree structure corresponding to the L-proposition (4.14b) is shown in
Figure 4.10.
The semantic definition of the predicate MOST is given in (4.15):
(4.15) For all predicates F and G:
[[MOSTF]] ¼ {[[F]] j there is a subset Y  [[F]] such that jYj > j½½Fj=2
and Y satisfies G under conjunction}
NB: jXj: the cardinality of the set X)
(the extension of the predicate MOST is the set of all sets [[F]] such that
there is a subset Y of [[F]] whose cardinality is greater than the
cardinality of [[F]] divided by 2 and Y satisfies G under conjunction)
Thus, if most farmers grumble, there is at least one subset Y of the set of
farmers such that Y consists of more than half the number of farmers, and all
members of Y grumble. Clearly, when there are no actually existing farmers, Y
equals
and, therefore, cannot contain more than half the number of farm-
ers. There is then, moreover, no actual situation in which a farmer grumbles.
Therefore, when [[F]] ¼
, sentence (4.14a) is false. The interested reader is
invited to formulate an analogous definition for the quantifier HALF.

S1

Pred S3

Pred S1 Pred NP
MOST x Grumble x
Pred NP
Farmer x
FIGURE 4.10 L-propositional tree structure of Most farmers grumble
144 The Logic of Language

4.4 Predicate logics and intuitions: a scale of empirical success

Let us now take provisional stock and see how the various systems score, at
this stage of the analysis, given their satisfaction conditions for SOME, NO, and
ALL. First we list seven natural logical intuitions. Then we list the satisfaction
conditions for SOME, NO, and ALL in four systems of predicate logic. Finally, we
list the empirical success scores of the four systems considered with regard to
the intuitions listed. (The Hamiltonian system has been left out because it is
identical to BNPC but for the condition that ALL F is G is true in BNPC just
in case [[F]]  [[G]] whereas in the Hamiltonian system it is true just in case
[[F]]  [[G]].) The empirical success scores of the various systems are shown in
terms of the following seven natural logical intuitions (‘‘’: ‘is felt to entail’;
‘
’: ‘is felt to be equivalent’):
Natural logical intuitions:
1. SOME F is G ‘ NOT-ALL F is G I ‘ ¬A
2. SOME F is G
SOME F is NOT-G I
I*
3. SOME F is G
SOME G is F I
I!
4. ALL F is G ‘ SOME G is F A ‘ I!
5. ALL F is G ‘ SOME G is NOT-F A ‘ I!*
6. NO F is NOT-G
ALL F is G N*
A
7. NOT-ALL F is G
SOME F is G
SOME F is NOT-G ¬A
I
I*

Satisfaction conditions for SOME, NO, and ALL in four systems:

SOME: satisfaction condition:
BNPC [[F]] \[[G]]  [[F]] ([[F]], [[G]] are distinct natural sets)
AAPC [[F]] \[[G]] 6¼ Ø
ABPC [[F]] \ [[G]] 6¼ Ø
SMPC [[F]] \ [[G]] 6¼ Ø
NO: satisfaction condition:
BNPC [[F]] OO[[G]] ([[F]], [[G]] are distinct natural sets)
AAPC [[F]] \ [[G]] = Ø
ABPC [[F]] \ [[G]] = Ø
SMPC [[F]] \ [[G]] = Ø
ALL: satisfaction condition:
BNPC [[F]]  [[G]] ([[F]], [[G]] are distinct natural sets)
AAPC [[F]]  [[G]] and [[F]] 6¼ Ø
 Logical power 145

ABPC [[F]]  [[G]] (applicable only if [[F]] 6¼ Ø)

SMPC [[F]]  [[G]]
Empirical success scores for four systems of predicate logic:
BNPC 1, 2, 4, 5, 6
AAPC 3, 4,
ABPC 3, 4, 6
SMPC 3, 6
The first thing that strikes one is that none of the logical systems under
discussion is able to account for intuition 7, which says that NOT-ALL F is G is
felt to be equivalent with both SOME F is G and SOME F is NOT-G. Apparently,
intuition 7 is a special case. In Section 3.4.1 it has been made clear that
intuition 7 is accounted for in BNPC when NOT-ALL F is G is assigned a
topic–comment structure and BNPC has been extended with a presupposi-
tional component, which, incidentally, makes the logic trivalent. Extending
AAPC with a presuppositional component establishes the equivalence of NOT-
ALL F is G and SOME F is NOT-G, lacking in nonpresuppositional AAPC. The
equivalence felt to exist between NOT-ALL F is G and SOME F is G is, of course,
not attainable in either AAPC or ABPC, since these two systems have the
subaltern entailment from A to I.
BNPC scores best, though it lacks an account for intuition 3 which makes
the quantifier SOME symmetrical (I
I!). This question has been commented
on in Sections 3.4.1 and 3.4.2, where it is argued that intuition 3 is deceptive in
that it appears to depend on world knowledge regarding the actual extensions
of the predicates at issue.
AAPC and ABPC (the Square) successfully account for intuitions 3 and 4
because in both systems SOME is symmetrical. Moreover, ALL F is G cannot be
true unless SOME G is F is also true, under the definitions for SOME and ALL in
these systems. One notes that the combination of intuitions 3 and 4 makes for
the intuition that the subaltern entailment (A ‘ I) holds. ABPC, in addition,
accounts for intuition 6, because in ABPC the Conversions hold, so that ¬I*

A. In AAPC, as has been shown, all that can be said is that A ‘ ‘¬ I* but not
vice versa. Finally, SMPC satisfies the intuitions 3 and 6. It satisfies intuition 3
because in SMPC I
I! and it satisfies intuition 6 because in SMPC the
Conversions hold.
It thus results that BNPC (along with the Hamiltonian system) is the most
intuitively satisfying system of predicate logic in existence. It is also internally
consistent and logically faultless. In particular, it does not suffer from undue
existential import. All it suffers from, as has been explained in Section 3.4.1, is
146 The Logic of Language

its inability to warrant the truth of quantified statements under the operators
ALL, SOME, or NO as long as no full situational knowledge has been achieved—a
defect that will hardly have been felt for most of the period in which natural
language came into being and was the vehicle of linguistic interaction
in illiterate, tribal communities. Presumably, BNPC’s failure to be knowl-
edge-independent began to be important only a mere five thousand years
ago when literate civilizations began to spring up all over the world, in South
and Central America, China, Mongolia, Northern India, Mesopotamia, and
Egypt.
 5

Aristotle, the commentators, and

Abelard

5.1 A recapitulation of ABPC

As was shown in the previous chapter, ABPC, for many centuries the one and
only predicate logic in existence, is an exact parallel to standard propositional
logic created by the Stoics. Every student of logic will remember the tradi-
tional Square of Opposition, presented in Figure 5.1a and expressed in terms
of the Boethian symbols A, I, E, and O (see also Section 2.2).
The entailments from A to I and from E to O are known by the traditional
names of positive and negative subalterns, respectively. A and E, moreover,
are known as the universal positive and the universal negative, respectively,
and I and O as the particular positive and the particular negative, respectively.
The relations expressed in the square representation must be extended
by stipulating that I-type L-propositions ‘convert’, in the original
Aristotelian sense of ‘conversion’ (antistrophe), defined in the Prior Analytics
and discussed briefly in Section 3.4.1. ‘Antistrophe’ or conversion is
the process whereby the F- and G-predicates are interchanged. In ABPC, it
can be applied to I-sentences salva veritate: SOME F is G is equivalent to SOME
G is F (I
I!). Since O-sentences are I-sentences with a negated G-predicate,
SOME F is NOT-G is equivalent with SOME NOT-G is F. Moreover, as
E-sentences are, in fact, negated I-sentences, the converse of NOT[SOME F is
G] is NOT[SOME G is F]. Then, F and G are interchangeable in A-sentences
provided both F and G are made negative (contraposition): ALL F IS
G is equivalent with ALL NOT-G IS NOT-F. Moreover, ALL F IS G entails, but is
not equivalent with, SOME G IS F and hence SOME F IS G (the subaltern
entailment).
As has been mentioned several times in the preceding chapters, we consider
the Boethian notation in the form of the classic square unsatisfactory,
even though it was in use for many centuries, mainly because it fails to express
148 The Logic of Language

a. A C
E b. A ¬I* I*
CD
C
CD CD SC
C

CD
SC
I ¬I A*
I O
: equivalents
A: All F is G / No F is not-G C: contraries
I: Some F is G / Not all F is not-G CD: contradictories
E: No F is G / All F is not-G SC: subcontraries
O: Some F is not-G / Not all F is G >: entails

FIGURE 5.1 ABPC represented as the Boethian Square of Opposition and as the strict
natural square consisting of two isomorphic triangles

the roles of external and internal negation and because it fails to express
the Modulo*-Principle. For these reasons, we decided, in Section 2.2, to
replace the symbols E and O with A* and I*, respectively, so as better
to be able to express the fact that E-sentences are in fact A-sentences (that
is, universals) and that O-sentences are in fact I-sentences (particulars), in
both cases with an internal negation on the G-predicate, represented by
the asterisk. We also adopted the sign ¬ for standard (external) negation
so as to express the fact that contradictoriness is systematically caused by
negation.

a. ¬AND b. ¬A
CD CD CD CD
AND ¬OR* A ¬I*
CD CD
CD CD
C C C C
C C

SC SC SC SC

OR SC OR* I SC I*

CD CD CD CD
C C

¬OR AND* ¬I A*
SC SC
CD CD CD CD

¬AND* ¬A*

FIGURE 5.2 Propositional calculus and ABPC as isomorphic complete octagonal graphs
 Aristotle, the commentators, and Abelard 149

It has been shown, in the preceding chapters, that, apart from the notation
used for the vertices, ABPC is better represented not as a square with
two crossing diagonals but as a combination of two triangles connected
at two of their vertices by equivalence relations, as in Figure 5.1b, just as
was done for propositional calculus in Figure 4.1. In Chapter 4, however,
it was shown that the optimal representation for both propositional and
predicate calculus is maximalist: it takes the form of an octagonal graph, as
in Figure 5.2, repeated from Figures 4.3 and 4.5.
Section 4.1 has made it sufficiently clear that, under a strictly extensional
ontology, ABPC suffers from UNDUE EXISTENTIAL IMPORT (UEI). What interests
us here is the historical development of ABPC and of its unknown variant
Aristotelian-Abelardian predicate calculus or AAPC.

5.2 The not quite Aristotelian roots of ABPC

5.2.1 Aristotle’s own predicate logic
What did Aristotle’s own predicate logic, found mainly in his On Interpreta-
tion (Int), look like?1 Specialists say, no doubt correctly, that Aristotle did not
complete his system of predicate logic and elaborated only some aspects. And,
as everyone does who is working at a new formal system, he also dwelled on
aspects of the system that were not, in the end, incorporated into it.
Thus he spends some time on a sentence type called ‘adióristos’ in the first
chapter of On Interpretation. This term means literally ‘indefinite’, but it is
perhaps best rendered as ‘generic’. It is described formally by saying that it
lacks any specific quantifier (ALL or SOME) and semantically, somewhat cryp-
tically, as ‘about a universal class but not universal in character’ (Int 14b7). His
example is Man is white, where ‘man’ denotes a universal class but the
sentence does not imply that all men are white. In fact, Aristotle allows for
both Man is white and Man is not white to be true simultaneously despite the
fact that, grammatically speaking, the latter is the negation of the former
and should therefore be its logical contradictory (Int 17b31–32). He fails,
however, to specify the quantificational truth conditions that would
make this possible. It is easy to see that this sentence type, as long as it
remains as ill-defined as it is, can hardly play a role in any sound predicate

1 The most detailed and authoritative study on Aristotle’s On Interpretation (Perı̀ Hermēneı́as) is
Weidemann (1994), where one finds a translation that is better than most, along with a discussion of
the chronology and authorship of the work, a complete survey of the manuscript tradition, the
tradition of interpretation in Antiquity, the Arab world and the Middle Ages, and the translation
tradition. What it does not have is an analysis of the actual logic involved.
150 The Logic of Language

logic. Such lapses are understandable, given the high degree of difficulty of
predicate calculus and given the fact that Aristotle had to create logic out of
nothing, without any existing terminology and without any formalization
techniques to build on. Many things that were still opaque to him are clearer
to us now.
In the matter of existential import, however, Aristotle appears to have seen
the problem more sharply than most later logicians and many modern
historians of logic, who tend to gloss over it too lightly or even fail to see it.
When one reads Aristotle’s text literally, one sees that he hedges precisely
on the point where existential import becomes relevant. He rejects the
classic Conversions and thus saves his logic from the blemish of undue
existential import, as will be clear in a moment. That Aristotle may well
have seen the danger of undue existential import looming at the horizon is
seldom taken into consideration, presumably, one is inclined to think, be-
cause his commentators, followers, and critics did not, on the whole, discern
it as clearly as they could have. It was not until the late nineteenth century
that the logical problem of existential import began to receive full attention.
Until that time, awareness of this problem seems to have been desultory
and incomplete. Since we have no inclination to underestimate Aristotle,
who is undoubtedly one of the greatest intellectual giants in Western history,
we will have a closer look at the issue.
What we see, when we read the text of On Interpretation closely, is that
Aristotle stops short of stating the Conversions—that is, the equivalence of NO
F is G and ALL F is NOT-G and of SOME F is G and NOT ALL F is NOT-G. The
Kneales noticed this (Kneale and Kneale 1962: 57):
Aristotle [ . . . ] allowed that Every man is not white could be said to entail No man is
white, but rejected the converse entailment.

Yet these authors failed to see the relevance of this rejection. Further down in
their book, when discussing Abelard’s rejection of the Conversions and his
appeal to Aristotle who, like Abelard, considered NOT EVERY human is white
and not SOME human is NOT-white, to be the contradictory of EVERY human
is white, their comment is (Kneale and Kneale 1962: 210):
It is true, of course, that Aristotle wrote Greek words corresponding to Non omnis
homo est albus [Not every human is white], but it seems clear that he did not intend to
convey by these words anything different from the doctrine later attributed to him by
Boethius.

This not only contradicts what they wrote on p. 57, but it is also, one must
fear, just wrong. Why should Aristotle not be taken at his actual words?
 Aristotle, the commentators, and Abelard 151

If that is done, what results is a sound, though incomplete, system of predicate

logic.
The actual passages in Aristotle are Int 17b16–26 (translations mine):
I use the term ‘contradictorily opposed’ for positive and negative statements, the one
attributing a property universally and the other not universally to the same objects,
such as Every human is white versus Not every human is white, or No human is white
versus Some human is white. The term ‘contrarily opposed’ is used for the pair
universal positive and universal negative, such as Every human is white versus No
human is white, or Every human is just versus No human is just. It is impossible for
these latter two to be simultaneously true, but their contradictories can sometimes be
simultaneously true of the same class, as in Not every human is white versus Some
human is white.

and Int 20a16–23:

Since the contrary of All animals are just is the statement meaning No animal is just, it
is clear that these two will never be true at the same time nor be predicated of the same
thing. But their contradictories, that is Not all animals are just and Some animal is just,
will sometimes be simultaneously true. Then we have the following entailments:
All men are not-just entails No man is just, and Some man is just entails the former’s
contradictory Not all men are not-just. For in that case there must be at least one
just man.2

These passages are crucial. For if, as Aristotle actually does here, the
Conversions are given up for one-way entailments (A* ‘ ¬I, and therefore
also A ‘ ¬I*, because if A* ‘ ¬I then also A** ‘ ¬I*, and A**
A) but not
vice versa, the logic is sound: it then no longer suffers from its central logical
defect, undue existential import. This fact is too important for it to pass
unnoticed and it seems less than fair to Aristotle to ascribe this crucial
avoidance of a basic logical error to mere good luck on his part.
Yet it is true that Aristotle fails to be explicit on several points where one
would have liked him to be a little less sparing of his words. For example, he
gives no evidence of an explicit awareness of the subaltern entailments. Yet they
follow directly from what he does present explicitly. He does say clearly and
repeatedly that the truth of ALL F is G requires the falsity of its contrary NO F
is G. He also says explicitly and repeatedly that the contradictory of NO F is G
is its nonnegative counterpart SOME F is G. Given this, it is hard to imagine

2 This last sentence is problematic owing to the extreme density of Aristotle’s style at this point. The
literal translation of the Greek Anángkē gàr eı̃naı́ tina is ‘For there must be some’. Most existing
translations leave the opacity of this sentence unclarified. I have followed Weidemann’s translation
(Weidemann 1994: 20): ‘denn notwendigerweise ist denn ja irgendeiner gerecht’, which makes perfect
sense: if there is one just man, then it cannot be the case that all men are unjust.
152 The Logic of Language

that he failed to see that, therefore, the truth of ALL F is G requires the truth of
SOME F is G—that is, the positive subaltern entailment. And analogously for
NO F is G, which, by contraposition, entails NOT ALL F is G—that is, the
negative subaltern entailment for the negation of ALL F is G (though not
necessarily for its supposed equivalent SOME F is NOT-G). Moreover, he implies
the validity of the subaltern entailments at Prior Analytics (25a8–14):
The positive sentence does convert [in the Aristotelian sense; PAMS], though not as a
universal but as a particular, for example, if every pleasure is a good, then some good
must be a pleasure. Of particular sentences the positive does convert (since if some
pleasure is good, then some good will also be a pleasure), but the negative particular
does not convert, because if some animal is not human, it does not follow that some
human is not an animal.

This says that an A-sentence entails the converse (antistrophe) of the

corresponding I-sentence and that an I-sentence and its converse are equiva-
lent. It follows immediately, of course, that an A-sentence entails the
corresponding I-sentence. Any impartial reader will agree that, therefore,
Aristotle does have the subaltern entailments. He should, in any case, be
given the benefit of the doubt in this respect.3
In similar manner we notice that Aristotle fails to posit the relation of
subcontrariety explicitly. He certainly has no term for it. While his term for
contraries was enantı́ai and for contradictories he used the terms antı́phasis
and antiphatiko^s antikeı́menos (‘contradictorily opposed’), as at Int 17b16–17,
or sometimes simply antikeı́menos (‘opposed’), as at Int 17b24 (see also De
Rijk 2002: 103), there is no term for subcontraries. The first known occurrence
of the concept of subcontrariety is in the logical treatise Perı́ Hermēneı́as,
written in the second century CE and reliably attributed to the Latin author
Apuleius (best known for his Metamorphoses or The Golden Ass). In this
treatise, the Latin term subpares (‘nearly equal’) is used for subcontraries,
incongruae for contraries and alterutrae for contradictories (Sullivan 1967: 65;
Londey and Johanson 1987: 56, 88–89, 111). (Apuleius has no term for the
subalterns (Londey and Johanson 1987: 109) and is a little reticent on the
subject of the Conversions (Sullivan 1967: 71), though he does use the term
æquipollens for them.)
The first known occurrence of the Greek term hypenantı́ai, which underlies
Latin subcontrariae, is in the fifth-century Greek commentary on Aristotle’s
On Interpretation by Ammonius, who writes: ‘The particulars are called sub-
contraries (hypenantı́ai), because they are placed below the contraries [in the

3 See also Kneale and Kneale (1962: 58), where these authors express essentially the same conclusion.
 Aristotle, the commentators, and Abelard 153

Square; PAMS] and follow from them.’ (Busse 1897: 92). (One may surmise
that diagrams were commonly used in classroom teaching, or else it is hard to
understand how Ammonius could use the expression ‘below the contraries’;
see also Section 5.2.3.)
At Int 17b25 and 20a17–18 (quoted above) Aristotle says that the contradicto-
ries of contraries may be simultaneously true. But he does not, or not
explicitly, draw the further consequence that the falsity of the one excludes
the falsity of the other: they cannot be simultaneously false. Had he done
so, he would have established the relation of subcontrariety between I-type
and ¬A-type sentences. We shall be generous and assume that Aristotle did
know about subcontrariety, even though he did not give it a name. Aristotle
may not have seen all the details, but to deny him the insight that if
two sentences cannot be simultaneously true, their contradictories cannot be
simultaneously false, would seem to do him an injustice. In this we are
supported by the Kneales, who take it that Aristotle was, in fact, aware of
the relation of subcontrariety even though he had no term for it (Kneale and
Kneale 1962: 56):
The two particular statements [i.e. I and I*; PAMS] have been said by later logicians to
be subaltern to the universal statements under which they occur in the figure and sub-
contrary to each other. Although he does not use these expressions, Aristotle is
interested in the relations so described, and assumes that sub-contraries cannot
both be false though they may both be true. This is shown by his description of
them as contradictories of contraries.

Yet while one appreciates the Kneales’s generosity vis-à-vis Aristotle as

regards subcontrariety, one must note their apparent confusion on the subject
of the Conversions. For in Aristotle’s system of predicate logic (AAPC),
the pair < I,I*> is not part of the relation of subcontrariety, whereas the
pair < I,¬A> is, as are the pairs < I*,¬A*>, <¬A,¬I*>, and <¬I,¬A*>, as one
can read from Figures 5.3 and 5.6a. Aristotle merely accepted the one-way
entailments from A to ¬I* and from A* to ¬I and therefore from I* to ¬A
and from I to ¬A*, which makes A and I*, and A* and I, contraries, but not
¬I and ¬I*, so that I and I* cannot be subcontraries.
What appears to be the authentic Aristotelian system as it can be culled
from his texts is based on the two independent stipulations (i) and (ii),
from which further theorems follow (‘><’ stands for contrariety; ‘’ for
subcontrariety):
(i) A ><¬I (contraries) (ii) A* ‘ ¬I (entailment)
From this follow the theorems: From this follow the theorems:
154 The Logic of Language

¬A  I subcontraries) A* >< I (contraries)

A ‘ I (positive subaltern) ¬A*  ¬I (subcontraries)
¬I ‘ ¬A (contraposition) I ‘ ¬A* (contraposition)
A* >< ¬I* (Modulo*-Principle) A >< I* (Modulo*-Principle)
¬A*  I* (subcontraries) ¬A  ¬I* (subcontraries)
A* ‘ I* (negative subaltern) A ‘ ¬I* (Modulo*-Principle)
¬I* ‘ ¬A* (contraposition) I* ‘ ¬A (contraposition)
This system, incomplete as it is, is presented in Figure 5.3, which differs
from Kneale and Kneale (1962: 55) and Parsons (2006: 4), but is in agreement
with Thompson (1953: 259). The relation of contradictoriness of any pair < T,
¬T> (T stands for any of the sentence types defined) has been added because
it is implicit in Aristotle’s definition of sentential negation. This system is,
though not fully elaborated, logically sound, other than ABPC, which is not.
As has been shown, Aristotle rejected the Conversions, cutting them down
to the one-way entailments specified above. That this is sufficient to avoid
undue existential import is not apparent from the incomplete Figure 5.3, but
it is from Figure 5.7a, where all the missing logical relations have been filled in.
There one sees that the types ¬A, ¬I, ¬A*, and ¬I* do not entail any I-type or
I*-type sentence, which absolves the system of the charge of undue existential
import. In Figure 5.2b, however, which represents ABPC, one sees that all

¬A
CD SC
A ¬I* CD: contradictories
C: contraries
CD SC: subcontraries
C >: entailment

SC
C
I I*
C

C
CD

¬I A*
SC
SC CD
¬A*
FIGURE 5.3 Aristotle’s predicate logic as presented in his texts
 Aristotle, the commentators, and Abelard 155

sentence types as represented by the vertices entail an I-type or an I*-type

sentence, which restricts the system to situations where [[F]] 6¼
.

5.2.2 The ancient commentators

There is a remarkable contrast between, on the one hand, the very long
history of predicate logic and, on the other, the amount of persistent misun-
derstanding and misattribution.4 There is no doubt that the Master laid the
foundations for predicate logic, and in better-informed circles (e.g. Kneale
and Kneale 1962; De Rijk 2002) it is also known that his own system of
predicate logic differs in certain important respects from traditional ABPC
as developed by his commentators, especially the Latin author Apuleius
(125–180 CE), the Greek Ammonius (440–520 CE), and his younger
Roman contemporary Boethius (480–524 CE)—that is, five to eight cen-
turies after Aristotle.
One should realize that formal predicate logic, as presented in Aristotle’s
texts, was not at all a popular subject in ancient times. There was a rich
tradition of writing commentaries on Aristotle’s works, in the context of
higher-education teaching of Aristotelian philosophy, but the interest was
almost entirely focused on questions of the compatibility or incompatibility
of Platonic and Aristotelian doctrines and, after the third century CE, also the
more mystically oriented philosophy introduced by Plotinus known as Neo-
Platonism. There was a secondary focus on Aristotle’s theory of the syllogism,
but clearly no focus at all on predicate calculus (for documentation, see
Sorabji 1990, 2004).
Until the advent of the Middle Ages, formal predicate logic was but a
sideshow of a sideshow. Compared with the large number of known ancient
commentators on Aristotelian philosophy in its each and every aspect (see
Sorabji 2004), the number of commentaries written on Aristotelian predicate
logic is extremely small.5 Apuleius, Ammonius, and Boethius are, to my

4 The same goes for propositional calculus, where it took Lukasiewicz (1934) to show that this
calculus did not originate with Aristotle, as had been widely thought till then, but with the Stoics.
Significantly, Lukasiewicz wrote (Borkowski 1970: 198): ‘The history of logic must be written anew, and
by an historian who has fully mastered mathematical logic.’ Admittedly, the situation has changed to
some extent during the seventy-odd years that have passed since, but not enough to make
Lukasiewicz’s words irrelevant. It is still so that historians of logic can do with some additional
formal training. Just as technical translators need knowledge not only of the languages concerned
but also of the subject matter of the text, historians of a technical subject need not only historical,
linguistic, and textual knowledge but also the expertise required for a proper grasp of the subject at
hand. Unfortunately, this latter requirement is too often not met.
5 With the exception of the three authors mentioned, all the ancient authors known to have written
commentaries on Aristotelian logic restricted themselves either to basic semantic notions (such as the
156 The Logic of Language

knowledge, the only three who commented on the machinery of Aristotelian

predicate logic and whose works have survived. And for them, too, formal
predicate logic was but a minor sideline.
Apuleius was a North-African intellectual with a wide range of interests,
including the writing of literature, as has been said above. Ammonius was a
professor of philosophy in Alexandria. As regards his interest in logic, we may
quote Ebbesen (1990b: 445):
[T]hough Ammonius taught lots of logic classes in his time, he never managed to
become really interested in the discipline. Like most of his contemporaries, he
considered acquaintance with the Organon as only a stepping-stone on the way to
real philosophy.

Boethius was a Christian-Roman aristocrat who not only became prominent

in political life but was also a writer and intellectual of note. Despite his short
life (he fell into disgrace with the king, Theodoric the Great, and was executed
at the age of 44), he left an impressive uvre, of which his work on Aristote-
lian predicate logic was only a small part. Since the formal-logical interest of
all three authors was merely marginal, one may confidently surmise that their
insight into the nature and the technical details of predicate logic was limited,
to say the least.
Yet it was the ancient commentators, in particular Ammonius and
Boethius, who shaped predicate logic into a form that was passed on to the
Middle Ages and reigned supreme till the beginning of the twentieth century.

5.2.3 The Square representation

The classic Square of Opposition was only partially Boethius’ own
invention—if at all. The historical roots of the tradition to represent Aristo-
telian predicate calculus in the form of a rectangular figure are not entirely
clear. Aristotle himself appears to imply some sort of geometrical figure for
expository purposes, but according to Londey and Johanson (1987), the actual
square notation goes back to the treatise Perı̀ Hermēneı́as mentioned in
Section 5.2.1 and reliably attributed to the Latin author Apuleius (125–180
CE). The text of Apuleius’ treatise (which is written in Latin despite its Greek

relation between thought, word, and thing) or to the theory of the syllogism. This goes, for example,
for Alexander of Aphrodisias (±205 CE) and also for the Neo-Platonist Porphyry (±234–±305 CE)
mentioned by Ebbesen (1990a). The immensely influential ancient authority on medical science, Galen
of Pergamon (±129–199 CE), is thought to have written a commentary on Aristotelian predicate logic.
Unfortunately, however, the relevant parts of this commentary have not, as far as is known, survived
the wear of time. Sorabji (2004), a source book of Aristotelian commentaries on logic and
metaphysics, has nothing at all on predicate logic.
 Aristotle, the commentators, and Abelard 157

title) makes it clear that Apuleius saw that a rectangular arrangement would
capture Aristotelian predicate logic as set out in his On Interpretation and
modified through a few centuries of teaching practice.
Apuleius describes in detail what the ‘quadrata formula’ (Sullivan 1967: 64)
should look like. It is made up of the four vertices A (‘omnis voluptas bonum
est’ or ‘every pleasure is a good’), A* (‘omnis voluptas bonum non est’
or ‘every pleasure is not a good’), I (‘quaedam voluptas bonum est’ or
‘some pleasure is a good’) and I* (‘quaedam voluptas bonum non est’
or ‘some pleasure is not a good’)—though, of course, Apuleius did not use
these symbols. The positive and negative universals (A and A*) are to be
placed ‘in superiore linea’ and the positive and negative particulars (I and I*)
‘in inferiore linea’ (Sullivan 1967: 110). This terminology strongly suggests a
teaching practice where it was customary to actually draw the diagram.
One may assume that something like a square representation, whether or
not actually drawn, was used in whatever teaching in predicate logic took
place during subsequent centuries. It was left to the Roman Boethius, who
probably fell back on his slightly older contemporary the Greek Ammonius,
to complete Aristotle’s logic and cast it into the mould of the classic Square of
Opposition, which subsequently became part of the stock-in-trade of predi-
cate logic.

Adfirmatio universalis Negatio universalis

Omnis homo iustus est CONTRARIAE Nullus homo iustus est

Universale Universale
universaliter universaliter

S C E
O S
V N A V
T I
B R R B
A O
A T A
L C L
T I T
D D
E A I E
R C
R T TO R
N N R N
O IA
A C A
E E E
Adfirmatio particularis Negatio particularis
Quidam homo Quidam homo iustus
SVBCONTRARIAE
iustus est non est
Universale Universale
particulaliter particulaliter

FIGURE 5.4 Boethius’ own diagram of the Square of Opposition

Source : Meiser 1880: 152
158 The Logic of Language

For Boethius’ text in this regard I consulted Meiser (1880). There one finds
that Boethius did actually draw the Square diagram, prefacing it by the words
(Meiser 1880: 152):
Superioris autem disputationis integrum descriptionis subdidimus exemplar, quatenus
quod animo cogitationique conceptum est oculis expositum memoriae tenacius
infigatur.
[We have provided an integral descriptive diagram of the above discussion, since what
is understood by the mind and by thought is more enduringly fixed in memory when
shown to the eye.]

Then follows the diagram, of which Figure 5.4 is an exact replica.

5.2.4 An aside on Horn’s and Parsons’ proposal as regards the O-corner

An aside must be inserted here regarding a view found in Horn (1989, 1997)
and followed up in Parsons (2006, 2008). The view reduces to two claims. The
first, which one finds discussed in one or two late medieval texts but which
was abandoned by the end of the Middle Ages and revived in the 1950s by a
few American historians of logic, in particular Moody (1953), amounts
to saying that it is not the universals (A and A*/¬I) that generate
existential entailment but the positives (A and I), as opposed to the negatives
(A* and I*/¬A), which do not. The view is summarized by the formula that it
is not ‘quantity’ but ‘quality’ that is responsible for existential import: the
affirmative categorical types are taken to have existential import but the
corresponding negatives (that is, with internal negation) can be true when
[[F]] ¼
. In more direct parlance, it is said that the I* and A*-corners should
be allowed to ‘leak’. We shall see that this claim is correct in so far as the logic
based on the quantifiers thus defined does not suffer from undue existential
import (UEI). The second claim is that this has in fact been the way the
O-corner has been interpreted throughout the logical tradition. We will see
that the second is incorrect and fails to do justice to historical truth.
If, like Horn, one belongs to the pragmatics camp, such a view appears, at
first sight, to make good sense. For if the Square can be saved from UEI and
thus be shown to be logically sound after all, the gap between pragmatics and
logic is narrowed considerably, since the Square stays much closer to natural
intuitions than SMPC does. But clearly, for that remedy to work, the Square
must not have to be re-interpreted in a way that makes the clash with the same
natural intuitions worse than it was before. Unfortunately, it is to be feared
that this condition is not fulfilled. On the new interpretation, the clash with
natural intuitions is even worse than it was with respect to SMPC.
 Aristotle, the commentators, and Abelard 159

Logically speaking, one might be tempted to think that letting the I* and
A*-corners ‘leak’ has a great deal going for it because it looks as if the classic
Square is preserved in full glory when A* and I* no longer require that the
F-class should be nonnull, as becomes clear when one sets up a VS-model and
an octagonal representation along these lines, as in Figure 5.5 below. In virtue
of considering space 4 a truth-maker for A* and I* but not for A and I, the
resulting octagon turns out identical to that of the Square (Figure 5.2b). This
may well give rise to the idea that the Square can be saved in full if the truth
assignments are made in this way. But, as is shown below, this advantage is
bought at the price of having to admit an ambiguity in the semantic definition
of the existential quantifier.
Horn (1989: 23–30) is at pains to convince his readers of the historical
legitimacy of the view that ‘existential import is determined by the quality of
the proposition; affirmative (A and I) propositions entail existence, while
negative ones (E and O) do not’ (Horn 1989: 24). He assigns an impressive
pedigree to this view, tracing it back to Aristotle in so far as Aristotle denies
the truth of a singular, nonquantified sentence like Socrates’ son is ill when
Socrates has no son. Just as Socrates’ son is not ill is true when Socrates has no
son, Not all Socrates’ children are ill should likewise be considered true, and
thus All Socrates’ children are ill should be considered false, when the good
man has no children. Yet, although this parallel is sometimes drawn (e.g.
Thompson 1953: 257), it is false, since Aristotle clearly distinguishes between
‘singular sentences’ (‘occasion sentences’ in Quine’s terminology) and quan-
tified sentences or categoricals (Quine’s ‘eternal sentences’). Moreover, nei-
ther Aristotle nor any other ancient author has been found prepared to
maintain that Some of Socrates’ children are not ill is true when Socrates is
childless.
Moody (1953) proposes a more modest pedigree. Trying to defend the
logical soundness of the Square, he claims (1953: 50–3), followed by Klima
(1988: 18–19), that the medievals generally held that only the affirmatives in
the Square have existential import while the negatives do not. Buckner,
however, clearly showed in his conference paper (2007) that this is at best
tendentious and at worst just wrong, since this interpretation of the intended
meaning of I*-type propositions is overwhelmingly belied by the medieval
philosophical literature. It appears to be closer to the truth to say that during
the fourteenth century and only then this view was discussed but far from
generally accepted.
In Moody’s defence it must be said that he recognizes the fact that, even
though withholding existential import from the O-corner saves the Square
from UEI, it clashes with natural intuitions (Moody 1953: 51–2):
160 The Logic of Language

The particular negative must, as negation of the universal affirmative, be analyzed as a

disjunction of the negations of the two parts of the universal affirmative.
Consequently the formula of the particular negative is not properly represented by
the word formula ‘Some F is not a G’, but only by the formula, ‘Not every F is a G’,
which is satisfied either because nothing is an F, or because something is an F which is
not a G. (italics mine)

The same observation is made by Thompson, who agrees with Moody’s

analysis (Thompson 1953: 253):
Even if we agree with the new defenders of Aristotle that the decision which leads to
the modern analysis is repugnant to ordinary speech, we can still argue that this is
more desirable than a decision repugnant to logical analysis itself.

Our perspective, however, is the opposite of Thompson’s: we are primarily

interested in natural human logic, and an O-corner as envisaged by Moody
and Thompson is, therefore, of no interest to us, even if it were tenable from
an abstract, and inevitably artificial, logical point of view.
Horn, who closely follows Moody, agrees (Horn 1989: 27):
. . . an O-type . . . statement corresponds more exactly to the nonentailing Not every
F is a G than to the entailing Some F is not a G. The former is automatically true but
the latter false in a state of F-lessness.

Yet he fails to draw the conclusion that, therefore, this interpretation of the
Square is artificial and unhelpful for an understanding of the natural logic of
human language and cognition.
In Horn (1997) one finds a shift towards what Thompson, in the quote
given above, calls ‘logical analysis itself ’ and away from natural usage. Here,
Horn defends the view that only affirmative propositions in the Square of
Opposition have existential import, while the negatives do not, so that Some F
is not a G should be taken to be automatically true ‘in a state of F-lessness’.
This view, which is, again, called ‘traditional’ (Horn 1997: 157), is attributed to
Aristotle’s commentators Apuleius and Boethius, to the twelfth-century
French philosopher Abelard, and to a few modern authors, in particular
Carroll (1896), Strawson (1952), and Kneale and Kneale (1962).
These references, however, do not stand up to scrutiny. As has been shown
in Section 5.2.2, the commentators certainly did not subscribe to the analysis
proposed by Horn and Parsons. Abelard did not even have an O-corner, as is
shown in Section 5.3, because, like Aristotle, he distinguished I*, which does
have existential import, from ¬A, which does not. The eccentric but well-
informed Oxford mathematician-logician-writer-photographer Charles
Dodgson, better known under his nom de plume Lewis Carroll, wrote a
 Aristotle, the commentators, and Abelard 161

logic textbook for children (Carroll 1896). This author clearly assigns existen-
tial import to I*, as is shown by his reduction of Some apples are not ripe to:
Some j existing Things j are j not-ripe apples (Carroll 1896: 77).

Horn’s statement (Horn 1997: 157) that Carroll defended the position that the
O-corner lacks existential import is thus clearly erroneous.
The same applies to his reference to Strawson’s logic textbook. There we
read, for example (Strawson 1952: 166):
It is already agreed that the I and O forms are to be regarded as having existential
import.

And as regards the Kneales’s famous book on the history of logic, we read
(Kneale and Kneale 1962: 58):
. . . the assertion of the existence of a man who is white, or not-white, as the case may
be, already involves an assertion of the existence of a man.

Parsons (2006), which, mainly through its publication on the internet,6 has
had a considerable influence, repeats Horn’s claim to an ancient pedigree.
Apparently, Parsons considers this historical claim important, as it occurs in
the opening paragraph of his article (Parsons 2006: 1):
For most of this history [of the Square; PAMS], logicians assumed that negative
particular propositions (‘Some S is not P’) are vacuously true if their subjects are
empty.

Yet no evidence is provided for this statement, which, to the best of my

knowledge, does not hold water. What one does find in the thirteenth and
fourteenth centuries is a debate on whether Every man is an animal is true
with no men existing.7 In this case, the example itself is important because
animal is an essential, not an accidental, property of man. The debate ran
mainly along party-philosophical lines. The nominalists (Ockham, Buridan)
argued that, with no men existing, Every man is an animal is false. They
concluded that, therefore, Some man is not an animal is true, not realizing
that this conclusion is false, because when ¬A is considered false with no
F existing, there is an entailment from I* to ¬A but not vice versa, as is shown
in Figures 4.7c and 5.7a. But the realists (Scotus, Aquinas) ruled this out on
grounds of philosophical a priori: if it is part of the essence of man to be an
animal, then falsity for Every man is an animal is ruled out as a matter of

6 Parsons (1997) is a larger version of Parsons (2006), published in the normal, old-fashioned way.
7 I owe this information to Edward Buckner (email correspondence).
162 The Logic of Language

principle. Since the Church backed the realists (Ockham was excommuni-
cated in 1328), their view prevailed and the nominalist view, which errone-
ously implied that the O-corner has no existential import, was marginalized.
But one should note that this debate was not triggered by the question of UEI,
which, apparently, did not figure at all prominently in the minds of the
philosophers involved, but by the great metaphysical debate on universals.
It might well have been superfluous if both parties had taken the trouble to
have a closer look at the Aristotelian-Abelardian version of predicate logic.
Moody states (1953: 51):
Since existential import was considered to belong only to affirmative sentences, it is
sufficient, for the falsity of an affirmative and hence for the truth of the contradictory
negative, that one of the terms stands for nothing.

This statement, however, is not backed up with any crucial quotation. The
closest he comes is a quote from Buridan’s Sophismata, Ch. 2, Concl. 14
(Moody 1953: 51):
Omnis particularis negativa vera, ex eo est vera ex quo universalis affirmativa sibi
contradictoria est falsa.
[Every true particular negative is true on the grounds that the universal affirmative
which is its contradictory is false.]

But Buridan does not go so far as to state that particular negatives are
vacuously true when the subject class is empty. Moody’s statement that
‘existential import was considered to belong only to affirmative sentences’
is, therefore, based on his interpretation, not on textual evidence. One may
wonder why the medieval authors did not put into words what Moody takes
them to imply. Did they baulk at the idea of having to call a sentence like Some
unicorns are not animals true despite the absence of unicorns in this world or
is this simply not what they implied?
Abelard, as shown in Section 5.3, solved the predicament, in the early
twelfth century, by giving up the Conversions and letting the O-corner keep
its existential import. In this he was followed by Walter Burleigh in his De
Puritate Artis Logicae, written around 1328 in reply to Ockham’s Summa
Logicae, which was written around 1323. For Burleigh, as is shown in Section
9.4.2, an A-type sentence like Every man who has a son loves him and its
corresponding I*-type Some man who has a son does not love him are not
contradictories but only contraries.
In any case, Moody’s skewed interpretation of medieval logic is now widely
taken to reflect historical reality, especially in North America. Many or most
American authors now take it for granted that it was ‘the general medieval
 Aristotle, the commentators, and Abelard 163

view that affirmative sentences are false if their subjects are empty, whereas
negative sentences are true if their subjects are empty’ (King 2005: 266). Yet
never is one presented with an actual reference to that effect: it very much
looks as if Moody has been believed on the strength of his authority, not on
the strength of actual evidence.
Nor, it seems, on the strength of actual insight. When providing the truth
conditions for the four Aristotelian sentence types, King, in his otherwise
excellent article, unwittingly assigns the same conditions to ¬I as he does to
I*, which suggests a less than full grasp of the issue. Yet the issue seems to
weigh on his mind, or else it is hard to see why he should start his enumera-
tion of truth conditions for the four types with an I*-sentence whose subject
term is empty (King 2005: 266):
Truth conditions for assertoric present-tense categorical sentences are straight-
forward. For instance, the particular negative sentence ‘Some vampires are-not
friendly’ is true just in case what ‘friendly’ personally supposits for, namely people
who are friendly, does not include anything—note the negative copula—for which
‘vampire’ personally supposits. Universal affirmatives (‘Every S is P’) are true when
everything their subjects supposit for their predicates also supposit for; particular
affirmatives (‘Some S is P’) when their predicates supposit for at least one thing their
subjects supposit for; universal negatives (‘No S is P’) when the predicate does not
supposit for anything the subject supposits for.

Parsons (2008: 5) does come up with something that looks like evidence,
quoting a passage from Ockham, who lived in the fourteenth century. This
passage, however, occurs in a wider context which I quote here more fully,
both in Latin and in my English translation (Ockham, Summa Logicae II.3).
The text opens Chapter II.3 of Ockham’s Summa Logicae, entitled ‘What is
required for the truth of propositions that are both indefinite and particular’
(I have italicized the parts that are quoted by Parsons):8
Viso quid sufficit ad veritatem propositionis singularis, videndum est quid requiritur
ad veritatem propositionis indefinitae et particularis.
Et est primo sciendum quod si non vocetur propositio indefinita nec particularis nisi
quando terminus subiectus supponit personaliter, tunc semper indefinita et particularis
convertuntur, sicut istae convertuntur ‘Homo currit’, ‘Aliquis homo currit’; ‘Animal est
homo’, ‘Aliquod animal est homo’; ‘Animal non est homo’, ‘Aliquod animal non est
homo’. Et ad veritatem talium sufficit quod subiectum et praedicatum supponant pro aliquo
eodem, si sit propositio affirmativa et non addatur signum universale a parte praedicati;

8 The quote is taken from: http://individual.utoronto.ca/pking/resources/Ockham/Summa_logicae.

txt
164 The Logic of Language

quod dico propter tales ‘Aliquod animal est omnis homo’, ‘Aliquis angelus est omnis
angelus’. Sed si talis sit negativa, requiritur quod subiectum et praedicatum non
supponant pro omni eodem, immo requiritur quod subiectum pro nullo supponat, uel
quod supponat pro aliquo pro quo praedicatum non supponit. Et hoc quia ad veritatem
talium sufficit veritas cuiuscumque singularis. Sicut ad veritatem istius ‘Aliquod animal
est homo’ sufficit veritas istius ‘Hoc animal est homo’ vel ‘Illud animal est homo’;
similiter ad veritatem istius ‘Animal non est homo’ sufficit veritas istius ‘Hoc animal
non est homo’, quocumque demonstrato. [ . . . ] Et ideo si nullus homo nec aliquod animal
sit nisi asinus, haec consequentia non valet ‘Homo non est asinus, igitur aliquod animal non
est asinus’. Similiter non sequitur ‘Homo albus non est animal, igitur homo non est
animal’ nisi ista propositio sit vera ‘Homo albus non est homo’. Tamen affirmative bene
sequitur [ . . . ] quia semper, sive homo sit animal sive non, bene sequitur ‘Homo currit,
igitur animal currit’, similiter bene sequitur ‘Homo albus est animal, igitur homo est
animal’, sive homo sit albus sive non. Sic igitur patet quomodo indefinita vel particularis
est vera si subiectum supponat pro aliquo pro quo non supponit praedicatum. Hoc
tamen non semper requiritur, sed quandoque sufficit quod subiectum indefinitae vel
particularis negativae pro nullo supponat. Sicut si nullus homo sit albus, haec est vera
‘Homo albus non est homo’, et tamen subiectum pro nullo supponit quia nec pro
substantia nec pro accidente.
[Now that we have seen what suffices for the truth of singular propositions, let us see
what is required for the truth of propositions that are both indefinite and particular.
First we note that, if the label ‘indefinite’ or ‘particular’ is assigned to a proposition
only when the subject term has individual reference, then indefinites and particulars
are always interchangeable, as in ‘A man runs’ and ‘Some man runs’; ‘An animal is a
man’, ‘Some animal is a man’; ‘An animal is not a man’, ‘Some animal is not a man’.
And it is sufficient for the truth of such propositions that the subject term and the
predicate refer to some same thing, if the proposition is affirmative and no universal
sign is added to the predicate; I am saying this because of examples like ‘Some animal
is every man’, ‘Some angel is every angel’. But if such a proposition is negative, it is
required <for its truth> that the subject and the predicate do not refer to all
identicals—that is, it is required either that the subject refers to nothing or that it
refers to something to which the predicate does not refer. And this is because it suffices
for the truth of such propositions that at least one, no matter which, singular
proposition is true. Just as it suffices for the truth of ‘Some animal is a man’ that
‘This animal is a man’ or ‘That animal is a man’ is true, in like manner it suffices for
the truth of ‘An animal is not a man’ that there be some, no matter which, animal that
can be actually pointed at, of which ‘This animal is not a man’ be true. [ . . . ] And
therefore, if there are no men and no animals except a donkey, the following argument is
not valid: ‘A man is not a donkey; therefore, some animal is not a donkey’. In the same
way, the following is not valid ‘A white man is not an animal; therefore, a man is not an
animal’, unless it is also true that ‘A white man is a man’. Yet these arguments are valid
in the affirmative form [ . . . ] because, whether a man is an animal or not, the
 Aristotle, the commentators, and Abelard 165

argument ‘A man runs; therefore, an animal runs’ is always valid. Or, to take another
example, ‘A white man is an animal; therefore, a man is an animal’ is a valid argument,
whether a man is white or not. Thus it is clear why an indefinite or particular
proposition is true in case the subject refers to something to which the predicate
does not. But this is not always required, for sometimes it suffices for the truth of a
negative indefinite or particular proposition that the subject refers to nothing. For
example, when there are no white men, ‘A white man is not a man’ is true even though
the subject refers to nothing because it refers neither to a substance nor to an
accident.]

I apologize for this very long quote, but it is necessary because it places the
quote given by Parsons in an entirely different light. In this passage, Ockham
appears to be making an attempt at incorporating a logic of indefinites into
that of the Square, as one sees from the opening sentence, where he does not
speak simply of particulars but of propositions that are both indefinite—that
is, ‘generic’, without a quantifier—and particular. Ockham systematically
distinguishes indefinites (generics) from particulars in that he uses the bare
noun (adorned with the indefinite article in the translation) in indefinites but
the noun preceded by aliquod or aliquis (‘some’) in particulars. In other
words, Parsons misquotes Ockham when he translates ‘ad veritatem talium
sufficit’ as ‘it is sufficient for the truth of [a particular] proposition’ because
talium (such) does not simply stand for particular propositions but for
propositions that are both indefinite and particular.
We remember from Section 5.2.1 that Aristotle spends some time over
indefinites (adióristoi) in the first chapter of On Interpretation, without
coming anywhere near a logic for them. Apparently, Ockham is now at
pains to achieve what Aristotle had failed to do. It is clear from the text that
Ockham withholds existential import from internally negative indefinites, as
he considers A man is not a donkey true in cases where there are no men but
there is a donkey, which in itself is quite reasonable (A unicorn is not a donkey
is reasonably called true in the actual world). In fact, all his examples where
internal negation yields truth when the subject class is empty are examples of
indefinites, not of particulars.
But it is also clear that he is pushing for an identification of indefinites and
particulars: ‘indefinites and particulars are always interchangeable’. And the
last-but-one sentence of the quote reads: ‘But this is not always required, for
sometimes it suffices for the truth of a negative indefinite or particular
proposition that the subject refers to nothing.’ Apparently, Ockham has
some qualms about this, because he continues: ‘For example, when there
are no white men, “A white man is not a man” is true even though the subject
refers to nothing’, giving again an example of an indefinite, and not a
166 The Logic of Language

particular, negative. The overall impression is that he is unhappy with the

unnaturalness of this truth assignment to internally negative particulars,
more so than to internally negative indefinites.
In any case, the passage shows that Ockham is experimenting here, in very
much the same way as so many other logicians (such as Hamilton, Moody,
Thompson, or Parsons) have done, with remedies for the Square.9 But this is a
far cry from saying that it was traditionally or generally accepted in medieval
philosophy that the O-corner lacks existential import, or that existential
import is restricted to the affirmatives while the negatives are without it.
Parsons’ generalizing statement that this is how ‘late medieval logicians
understood’ the Square (2008: 5) is, therefore, unsupported by sufficient
evidence, or, to put it bluntly, just a myth.
As hardly anything of note was presented on this issue between the Middle
Ages and the nineteenth century, I will now turn to that century and we will
see that there is hardly a trace of the thesis that the O-corner lacks existential
import. Fowler (1892), for example, implicitly assumes existential import for
the O-corner without mentioning the problem of the null F-class or the
distinction between extensional and intensional being (Fowler 1892: 36, 79).
Schiller (1912), though apparently unaware of the LOGICAL problem of existen-
tial import in connection with the Square, is acutely aware of the ONTOLOGICAL
distinction between extensional and intensional being (Schiller 1912: 107–8).
He also assumes existential import for the O-corner without further com-
ment (Schiller 1912: 154–5).
The German logician Sigwart mentions the logical problems caused by
existential import (Sigwart 1895: 172):
The negation of ‘all A’s are B’ means ‘the A’s which are B are not all A’s’; and the way
in which we understand the negation must depend upon whether the judgement
was intended as empirically or as unconditionally universal. [ . . . ] Aristotle taught
[ . . . ] that universal affirmative [¼ A] and particular negative [¼ I*] judgements, and
universal negative [¼ A*] and particular affirmative [¼ I] judgements are
contradictorily opposed. This doctrine leads to false conclusions unless attention is
paid to the difference between empirically valid and [unconditionally] valid
[universal] judgements.10

9 King (2005: 243) writes ‘Ockham’s Summa Logicae (The Logic Handbook), written ca. 1323, is a
manifesto masquerading as a textbook’, implying that the book was meant to promote Ockham’s
grand nominalist philosophical conception. Be that as it may, it is clear that the Summa Logicae does
not represent standard views that were taught in the Arts faculties of medieval universities.
10 The text, which is itself a translation from the German original, reads ‘empirically valid and
universally valid judgements’, which makes no sense; I have restored the text according to the only
possible intended meaning.
 Aristotle, the commentators, and Abelard 167

He thus seeks a solution in a distinction between empirically universal and

unconditionally universal ‘judgements’, the former with, the latter without,
existential import (Sigwart 1895: 163):
If a judgement concerning ‘all’ is unconditionally universal, then it is clear that no
direct statement is made about the actual existence of the subjects, though this is most
certainly presupposed by empirical judgements if they refer to actual things at all.

The distinction is then reduced to one of necessity versus possibility—a

solution that smacks of medieval metaphysics and would not satisfy nowa-
days. Interestingly, Sigwart noted (Sigwart 1895: 173) that Aristotle avoided the
equivalence of ¬A and I*, quoting exactly the passage from Int 17b16 given in
Section 5.2.1, and thereby belying his statement, made one page earlier, that
‘Aristotle taught . . . that universal affirmative and particular negative judge-
ments . . . are contradictorily opposed’. In this respect, therefore, Sigwart and
the Kneales are of a kind. Be that as it may, Sigwart’s text would make no sense
if the I*-form were to lack existential import.
The not so traditional nineteenth-century logician De Morgan (followed in
this respect by Carroll 1896: 70) uses his notion of restricted universe of
discourse (see Section 3.2.2) to circumvent the problem of existential import,
though he does not discuss the issue explicitly. This allows him to accept the
traditional Square, which he does, complete with existential import for the
whole Square and hence also for the I*-form (De Morgan 1847: 6).
In Volume II of his Collected Works, entitled Elements of Logic, Peirce writes
(Peirce 1974: 223, paragr. 2.376):
. . . the distinction between affirmative and negative propositions is purely relative to
the particular predicate. No doubt many logicians have assumed that negative
propositions are distinguished from ordinary affirmative propositions in not
implying the reality of the subject. But what, then, does ‘Some patriarch does not
die’ mean? Besides, all admit that propositions per se primo modo do not imply the
existence of the subject, although they be affirmative. At any rate, the resulting
syllogistic, if consistent, is very objectionable.

Again, one sees that this philosopher rejects the Horn–Parsons analysis,
because it clashes not only with ordinary usage but also with syllogistic
reasoning.
This, more or less random, selection of texts rather bodes ill for Parsons’
(and Horn’s) contention that ‘For most of this history, logicians assumed that
negative particular propositions (“Some S is not P ”) are vacuously true if
their subjects are empty’.
168 The Logic of Language

But, history aside, does it make semantic or logical sense to deny the O-
corner existential import? At first sight one might think it does, because if the
A-corner has existential import, one would expect it to follow that ¬A, and
thus its alleged equivalent I*, do not. Yet a little reflection will show that this
makes the semantics of the natural-language existential and universal quan-
tifiers SOME and ALL inconsistent in a strictly extensional system. According to
Horn and Parsons, existential import is induced by the universal and existen-
tial quantifiers only when they combine with a nonnegative predicate G but
not when they combine with a negative predicate not-G. This would make
their satisfaction conditions dependent on the lexical choice of the main
predicate. But in extensional predicate logic the main predicate is represented
by a lexical variable—in this case, G or not-G—and should, therefore, have
no bearing on the logical properties of the operators that define the logic. The
Horn–Parsons position has the extraordinary implication that, for example,
Some men are bachelors has existential import, but Some men are not married
hasn’t, or that All John’s views are erroneous has existential import, but All
John’s views are not right hasn’t!
Logically speaking, Parsons’ proposal amounts to a VS-model and an
octagon as shown in Figure 5.5. Here one sees that, in a state of F-lessness,
A and I are taken to be false but A* and I* true. This has the surprising effect
that the corresponding octagon remains identical with that holding for
traditional ABPC, as is easily checked when one compares Figure 5.5b with
Figure 5.2b. It may well have been this fact that has motivated Parsons and

a. b. ¬A {2,3,4}
[[F]] = Ø
CD CD
¬A I* {1} A ¬I* {1}
¬A I*
CD
CD
¬A I* C C
C
A ¬I* SC
4 3 2 1 2 3 4 SC
I ¬A* I SC I*
{1,2}
I ¬A* {2,3,4}

¬I A* CD CD
A* C
¬I

U {3,4} ¬I SC
A* {3,4}
CD CD

¬A* {1,2}
FIGURE 5.5 The predicate logic proposed by Parsons
 Aristotle, the commentators, and Abelard 169

those he follows to propose the view that the Square can be saved by assuming
that the affirmatives have existential import while the negatives do not. Yet
there is a big problem, in that this interpretation of ABPC makes it impossible
to provide consistent definitions for the two quantifiers 8 and ∃, or, in
traditional terms, for the words ALL and SOME.
This is easily shown. If, as Parsons does, one takes A, or ALL F is G, to be

true just in case [[F]] 6¼
and [[F]] \ [[G]] ¼
, then A*, or ALL F is NOT-G,
must be taken to be true just in case [[F]]6¼
and [[F]] \ [[G]] ¼
, which
gives both A and A* existential import. Likewise, if I, or SOME F is G, is taken
to be true just in case [[F]] \ [[G]] 6¼
, then I*, or SOME F is NOT-G, must be

taken to be true just in case [[F]] \ [[G]] 6¼
which gives both I and I*
existential import. The reason for this is that 8 (ALL) and ∃ (SOME) are binary
higher-order predicates expressing relations between two sets X and Y, no
matter which. Therefore, it makes no difference whether X or Y is character-
ized by positive or by negative satisfaction conditions. Yet Parsons defines A

as requiring that [[F]] 6¼
and [[F]] \ [[G]] ¼
, as above, but A* as requiring
that [[F]] ¼
or [[F]] \ [[G]] ¼
, which precludes one single definition for 8
(ALL). Likewise, Parsons defines I as requiring that [[F]] \ [[G]] ¼
, as above,

but I* as requiring that [[F]] ¼
or [[F]] \ [[G]] 6¼
, which again precludes
one single definition for ∃ (SOME). In other words, ALL and SOME have become
ambiguous between those cases where the G-predicate is positive and those
where it is negative. In the latter case it (arbitrarily) loses its natural existential
import. Parsons and company thus buy the logic of the Square of Opposition
but have to live with ambiguous quantifiers.
This forces Parsons to give up the classic notion of quantifier and thus to
reject both the traditional Aristotelian syntactic template (NOT) ALL/SOME F is
(NOT-)G and the modern notation, whether in the Russellian or in the
generalized-quantifier format. And this is exactly what he does. Parsons
(2006, 2008) falls back on the device of ‘quantification of the predicate’
discussed in Section 3.4.2 in connection with Hamilton’s predicate logic.
(This device was also known to Ockham, as one sees from the long quote
given above from Ockham’s Summa Logicae.) Like Moody and Thompson,
however, he dismisses naturalness as a criterion (Parsons 2008: 5):
What is important is that the logical notation be coherent and useful. If it does not
perfectly match the usage of ordinary language, that is not on its own important for a
system of logic. Indeed, if you are sure that ordinary language universal affirmatives
should be false when their subject term is empty, then you may represent that fact by
translating them into modern logical notation adding a conjunct. Instead of
symbolizing ‘Every A is B’ by ‘8x(Ax ! Bx)’, symbolize it as ‘∃xAx & (Ax ! Bx)’.
170 The Logic of Language

Yet he does not mention the disastrous consequences this has for the natural-
ness of the logic when this proposal is combined with the requirement that
the Conversions stay intact. Horn and like-minded pragmaticists should
realize that this does not help them at all.
In this context, Parsons devises an entirely new logical language which is
severely at odds with natural language, both syntactically and semantically,
and where SOME and ALL are defined not as words but syncategorematically,
and in terms not just of what may be the case ‘in the model’ but also of ‘truth
under an assignment . Following Klima (1988), Parsons then introduces an
artificial ‘zero element’ which, arbitrarily, makes ALL F is G and SOME F is G
false but ALL F is NOT-G and SOME F is NOT-G true. I will not here comment on
the logical merits of this proposal, since it is irrelevant for the natural logic of
language and cognition. All I wish to note here is that proposals to the effect
that only the Aristotelian affirmatives have existential import while the
negatives do not are doomed to take leave from the world of natural intui-
tions and to disappear into abstract logical space.

5.2.5 Logic and mysticism: what made logic popular?

But let us revert to our original topic. As we know, Boethius determined the
form of predicate logic until the twentieth century. It is usually said that this is
in virtue of the fact that he was a Christian and that he wrote in Latin, Greek
being almost universally unknown till the thirteenth century. In fact, most of
the standard Latin terminology is due to him (though many specialists believe
that in presenting his version of Aristotelian predicate logic he heavily
depended on Ammonius). This, however, is only part of the story. It seems
safe to surmise that the primary cause of Boethius’ enormous influence on the
history of logic lies in the general human penchant to look for mystic
messages in formal systems.
In this respect, the medieval literati had a field day. The seeds of mysticism
that had been sown during late Antiquity now had the chance to come to full
fruition, no longer suppressed by a pagan imperial power that had, on the
whole, connived at the spread of mystical currents, including Christianity
with its Eastern roots, among the slaves and the labourers but had stifled such
tendencies among the military, the educated, and the ruling classes. No
sooner had Christianity been given full rein than the hidden reserves of
esoteric mysticism and occultism exploded, despite the frantic efforts of the
official Church to contain the explosion. The subsequent rise of Islam only
added fuel to the flames.
 Aristotle, the commentators, and Abelard 171

During the Middle Ages, one witnesses not only a genuine professional
interest in logic, both syllogistic and predicate calculus (as shaped by
Boethius), but also a dramatic increase in mystic, kabbalistic and other occult
exercises, all of a well-defined formal nature, connected with letters of the
alphabet, geometrical figures, esoteric symbols, heavenly bodies, and num-
bers. One thinks, for example, of the mysticism woven around the Fibonacci
numbers and the golden ratio, also known as the golden section or the divine
proportion—a mysticism that has been of all times but was particularly
strong during the Middle Ages.
Syllogistic and predicate logic, with their nice geometrical designs, likewise
offered a wonderful opportunity for mystics and kabbalists to look for hidden
messages sent down from the clouds of eternity. A good example is the
Catalan Ramon Llull (1232–1316), who developed a system based on a
nonagon (reminiscent of Aristotle’s nine categories) listing the attributes of
God: goodness, greatness, eternity, power, wisdom, will, virtue, truth, and
glory. From these he meant to derive deductively all important eternal truths,
especially the Christian dogmas. The nonagon was realized in four different
‘figures’, the first of which was designed to generate the Aristotelian syllo-
gisms. The other three had different functions, all of a ‘cosmic’ nature (Eco
1995: 56–64; Wikipedia s.v. Ramon Llull).
Tellingly, the Arabic word for logic is mant. iq, from the Greek word mantike
(prophecy, fortune-telling). For a long time, logic and kabbalistic mysticism
were closely linked—a fact which, paradoxically, strongly contributed to the
increased popularity and prestige of logic. Once the political shackles of the
Roman Empire had been shed and Christianity had gained the ascendancy,
esoteric formal symbolisms began to cast their spell not only on the common
folk, as in late Antiquity, but also on the educated classes and the lay rulers.
And most of these were fascinated more by the mystical depths attributed to
the formal systems than by their purely intellectual content (to the extent that
there was any).
In modern times, we have, by and large, managed to banish occultism from
mathematics, logic, and science and to a large extent also from the prevailing
social norm system, if not quite from actual practice.11 But the irony is that

11 One should realize that even reputable academics have, at times, indulged in kabbalistic and
similar occult exercises. Mark Alford writes the following about Isaac Newton (Alford 1995):
In fact, Newton was deeply opposed to the mechanistic conception of the world. A secretive alchemist and heretical
theologian, he performed countless experiments with crucibles and furnaces in his Cambridge chambers, analyzing
the results in unmistakably alchemical terms. His written work on the subject ran to more than a million words, far
more than he ever produced on calculus or mechanics. Obsessively religious, he spent years correlating biblical
172 The Logic of Language

the ‘scientific attitude’, which began to be manifest during the late Middle
Ages and has dominated Western culture ever since, may well owe its growth
and its enormous influence in large part to the very craving for hidden truths
and for contact with the supernatural that it so successfully managed to get
rid of.

5.3 Abelard’s remedy

Petrus Abaelardus, known in the English world as Abelard (1079–1142), was
the opposite of a mysticist but he was destroyed, one may say, by one who was,
St. Bernard of Clairvaux. Abelard is, of course, best known on account of his
passionate but deeply unhappy relationship with Héloı̈se, tragically scarred by
his violent castration at the behest of Héloı̈se’s ‘uncle’ (more likely her natural
father), canon Fulbert of Notre Dame in Paris. Although Abelard and Héloı̈se
found themselves forced by the powers that were to lead separate monastic
lives, leaving their son Astralabius in the care of Abelard’s sister, they re-
mained devoted to each other for the rest of their unhappy days. All this is
well known. Less known is the fact that this brilliant, flamboyant, and
amazingly independent thinker, teacher, writer, and composer of religious
hymns and (now lost) popular love songs was excommunicated twice by the
Holy Church and died a broken man, having been the victim of a persistent
and extremely vicious defamation campaign led by envious countrymen,
especially the obscurantist St. Bernard of Clairvaux. The Church then ensured
his lasting vilification through its centres of education. He was the first to
apply the term theology to the rational investigation of the nature of God—an
activity considered blasphemous by Bernard and by the Pope (as well as by the
Islam religious authorities of the period). In this context, he wrote extensively
prophecy with historical events. He became deeply convinced that Christian doctrine had been deliberately
corrupted by the false notion of the trinity, and developed a vicious contempt for conventional (trinitarian)
Christianity and for Roman Catholicism in particular. Newton’s religious and alchemical interests were not tidily
separated from his scientific ones. He believed that God mediated the gravitational force, and opposed any attempt
to give a mechanistic explanation of chemistry or gravity, since that would diminish the role of God.
The Dutch mathematician Luitzen E. J. Brouwer (1881–1966), founder of Intuitionism in
mathematics, had strong mystical tendencies. In 1905, Brouwer published a pamphlet entitled ‘Leven,
Kunst en Mystiek’ (‘Life, Art and Mysticism’ see Brouwer 1966), a text drenched in romanticism, in
which he attempted to integrate mathematics and mysticism (see also Van Dalen 1999). Wittgenstein’s
‘mathematical’ work was likewise not without a fair shot of mysticism, and his Tractatus Logico-
Philosophicus has been much more influential among speculative thinkers and in the artistic
literature than among more sober-minded philosophers. One wonders if it would have had the same
appeal if it had lacked its quasi-formal style and structure.
Another example is Ferdinand de Saussure, one of the founding fathers of modern linguistics, who,
for the last twenty years of his life, indulged in ‘occult speculations about hidden messages in Greek
and Latin poetry, based on occurrences or repetitions of syllables and sounds’ (Seuren 1998a: 146; see
also Starobinski 1979). Many more such examples could be mentioned.
 Aristotle, the commentators, and Abelard 173

on the mystery of the Trinity. A century after his death, under the influence of
Thomas Aquinas, the Church reversed its position on the legitimacy of
theology and in particular the study of the Trinity, but Abelard was never
duly acknowledged, let alone rehabilitated.12
Apart from all this, however, Abelard was also a consummate logician, who
was probably the first, after Aristotle, to be aware of the problem of undue
existential import in ABPC. He proposed a solution which is not only as
sound as it is simple but also reflects Aristotle’s original intention: dissolve the
Conversions into one-way entailments from A to ¬I* and from A* to ¬I. Yet,
perhaps because the Church was keen to erase Abelard’s heritage from history,
this solution never came to the surface. It has played virtually no role in the
tradition of logic, where, to the extent that its existence was acknowledged, it
was not understood and hence misrepresented.13
One sometimes finds, as, for example, in Horn (1997) and Parsons (2006)
discussed in Section 5.2.4, Abelard discredited with the view that I*-sentences
have no existential import, which would mean that the existential quantifier is
taken to induce existential import when followed by a positive, but not when
followed by a negative matrix predicate. If Abelard had indeed held this view,
he would be subject to the same criticism as has been voiced in Section 5.2.4
with regard to Horn’s and Parsons’ description of ABPC. For Abelard, how-
ever, what is at issue is not whether I*-sentences have existential import
(which he was sure they have) but whether ¬A-sentences have it. One should
realize that in SMPC ¬A-sentences do have existential import, while in ABPC
there is a clash of truth values when ¬A-sentences are subjected to this
question. It is the denial of existential import to ¬A-sentences, and the
attribution of existential import to I*-sentences, that makes Aristotle’s and,
after him, Abelard’s logic worthwhile and saves it from inconsistency. It has,
apparently, been hard for logicians to see that the Conversions are far from
unassailable.
Kneale and Kneale (1962) also mention Abelard’s solution and, apparently,
likewise fail to see his point. They do, however, acknowledge Abelard’s
importance for the history of logic, as appears from the following passage
(Kneale and Kneale 1962: 204; translation between square brackets mine):

12 One may consult McLeod 1971, Gilson 1978, or Clanchy 1997 for authoritative and highly readable
accounts of this dramatic episode in human history.
13 In Seuren (2002) a solution is proposed for undue existential import that is identical to
Abelard’s. I named it Revised Aristotelian predicate calculus or RAPC. At the time, I was not aware of
the fact that Abelard had proposed an identical solution nine hundred years earlier. To give Abelard his
due, I have renamed the system in question Aristotelian-Abelardian predicate calculus.
174 The Logic of Language

Abelard’s mind was the keenest (though not in all respects the most admirable)14 that
had been devoted to the subject for more than a thousand years, and he approached
his task with the belief that it was still possible to make discoveries: ‘Non enim tanta
fuit antiquorum scriptorum perfectio ut non et nostro doctrina indigeat studio, nec
tantum in nobis mortalibus scientia potest crescere ut non ultra possit augmentum
recipere’ (De Rijk 1956: 535). [For the perfection of the ancient writers was not such
that their doctrine could not profit from our investigations, nor is it possible for
science to grow to such an extent in us mortals that it can no longer be improved.]

What Abelard proposed was that, for cases where [[F]]¼
, A- and A*-
sentences, as well as I- and I*-sentences, should be considered false, while
their negations should be considered true (Kneale and Kneale 1962: 210–11):
while he [¼Abelard] admits that Nullus homo est albus [No human is white] can be
regarded as the contradictory of Quidam homo est albus [Some human is white] because
nullus is merely an abbreviation of non ullus, he now refuses to allow that Quidam homo
non est albus [Some human is not white] is the contradictory of Omnis homo est albus
[All humans are white], as Boethius had maintained, and says that Aristotle dealt with
the question more subtly when he offered Non omnis homo est albus [Not all humans are
white] as the contradictory. It is true, of course, that Aristotle wrote Greek words
corresponding to Non omnis homo est albus, but it seems clear that he did not intend
to convey by these words anything different from the doctrine later attributed to him by
Boethius. Abelard, on the other hand, thinks that Non omnis homo est albus is
something distinct in meaning from the particular negative proposition Quidam
homo non est albus, and therefore outside the usual scheme of four categorical forms.
His reason for introducing this complication is that he assumes existential import for
Omnis homo est albus, though apparently not for Nullus homo est albus. The assumption
seems curious after his explicit statement that the word est occurring as pure copula
involves no assertion of existence; but there can be no doubt of his doctrine on this
point, since he insists that even the seeming tautology Omnis homo est homo would be
false if there were no men: ‘Cum autem Quidam homo non est homo semper falsa sit
atque Omnis homo est homo homine non existente, patet simul easdem falsas esse: unde
nec recte dividentes dici poterunt’ (De Rijk 1956: 176) [But since Some human is not
human is always false and All humans are humans is false when no humans exist, it
follows that these two can be false at the same time, so that it must be incorrect to call
them contradictory opposites]. We must therefore suppose that in his view it is the
word omnis which introduces existential import.15

14 One wonders about the relevance, or indeed the stringency, of the Kneales’s moral reservations.
This remark about Abelard’s allegedly not so admirable ‘mind’ is best taken as a late reflection of the
biographical and other historical facts mentioned at the outset of the present section. The Kneales
ought to have known better.
15 On Abelard’s doctrine of the existential value of the copula verb esse, see Rosier-Catach (2003 a).
Since Rosier-Catach’s meticulous analysis shows that Kneale and Kneale appear to be right on this
 Aristotle, the commentators, and Abelard 175

This is a curiously interesting passage, since it shows not only that Kneale
and Kneale, with all their barely hidden irony, failed to see Abelard’s point—a
failure they share with the entire logical tradition—but also, more important-
ly, that Abelard’s analysis is perfectly coherent and, as was shown in the
preceding section, in accordance with what can be gathered from Aristotle’s
own text, despite Kneale and Kneale’s assurance to the contrary. Since Abe-
lard’s proposal is in full agreement with Aristotle’s text, it is legitimate to
speak of ARISTOTELIAN-ABELARDIAN PREDICATE CALCULUS, or AAPC. But let us turn
to Abelard’s own texts so as to unravel what he himself actually proposed.
In his Dialectica, his main work on logic (full edition De Rijk 1956), Abelard
is at pains to distinguish as clearly as possible between the external (‘pre-
posed’) negation, which always creates contradiction, and the internal (‘inter-
posed’) negation, which creates contrariety under the universal quantifier. We
read, for example (De Rijk 1956: 177; translation mine):
The preposed negation thus has a different logical power from the interposed negation. A
sentence that says Every human is not white is not equivalent with Not every human is white,
and Some human is not white says something different from Not any human is white.16

In this context Abelard argues, in accordance with the Stoics, that the only
real guarantee for pure contradiction is the preposing of the negation and
not the insertion of an internal negation combined with a change from universal
to existential quantifier or vice versa (De Rijk 1956: 176; translation mine):
Similarly for categorical propositions, where the only real truth-value-inverting
(dividens) contradiction of any arbitrary positive proposition appears to be the one

complex issue, we subscribe to their conclusion that, for Abelard, ‘it is the word omnis [and not the
suppletive copula verb esse ; PAMS] which introduces existential import’. One will note that, according
to the analysis presented in the present book, existential import is not induced by any quantifier either
but by the extensionality of argument-term positions under any given predicate (see Section 10.7 and
Section 3.5.1 in Volume I).
16 It is not clear to what extent Abelard was aware of the distinction between sentences and their
underlying L-propositions (logical form), though valuable details for a reconstruction are presented in
Rosier-Catach (1999, 2003b, 2003c). Had he used a modern European language, he would have noticed
that in sentences with a definite, non-quantified subject term the external (sentential) negation is not
‘preposed’ but woven into the surface sentence (in many languages in construction with the finite verb
form). But he used Latin, where the logically external negation can always be literally preposed, that is,
placed at the beginning of the surface sentence, although most of the times it does not actually occupy
that position. His insistence on the syntactic distinction between preposed (external) and interposed
(internal) negation suggests that he was thinking in terms of a logical language with both
propositional and predicate variables, combining Aristotelian and Stoic logical analyses.
176 The Logic of Language

that has the negation preposed to it so that all its entailments are lost (totam eius
sententiam destruit). For example, the contradictory of Every human is human is Not
every human is human, and not Some human is not human, since there are situations
where the first and the third are simultaneously false. For when not a single human exists,
both of these two propositions are false: Every human is human and Some human is not
human. [ . . . ] The proposition Some human is not human [ . . . ] is always false. For what
it says is totally impossible: it cannot be the case and nature can offer no instance of it.
[ . . . ] In no situation can the same thing be both human and not human at the same
time. For it is a well-nigh eternal law that what is not included under negation is excluded
under it. [ . . . ] But since Some human is not human is always false and All humans are
humans is false when no humans exist, it follows that these two can be false at the same
time, so that it must be incorrect to call them contradictory opposites.

The quintessence of Abelard’s proposal is that the Conversions should be

given up in favour of the one-way entailments from I* to ¬A and from I to ¬A*
(or, equivalently, from A to ¬I* and from A* to ¬I), resulting in a system where
the subalterns are preserved as entailments from universal to existential quan-
tification (and thus not from ¬I* to I or from ¬I to I*) and where A-type and
A*-type sentences are false in cases where [[F]]¼
. The method of VS-model-
ling enables us to complete the system without effort and to show immediately
and precisely what the logical consequences are of this move.
AAPC as shown in Figure 5.6b differs from SMPC of Figure 5.6a only in
that space 4 (the outer ring), which represents the situations where [[F]]¼
,
makes A-type and A*-type sentences false instead of true. In fact, all four

a. [[F]] = Ø b. [[F]] = Ø

A ¬I* ¬A ¬I*
¬A I* ¬A I*
¬A I* ¬A I*
A ¬I* A ¬I*
4 3 2 1 2 3 4 4 3 2 1 2 3 4
I ¬A* I ¬A*
I ¬A* I ¬A*
¬I A* ¬I A*
¬I A* ¬I ¬A*

U U

FIGURE 5.6 VS representation of (a) SMPC and (b) AAPC

 Aristotle, the commentators, and Abelard 177

¬A {2,3,4}
a.
CD SC
{1} A ¬I* {1,4}
C
b. A ¬I* I*
CD
C
CD

{1,2} SC SC
{2,3}
C C C C
I I*
C

C
CD
CD
I ¬I A*

{3,4} ¬I A* {3}
SC
SC CD

¬A* {1,2,4}
FIGURE 5.7 The octagonal (a) and the square (b) representation for AAPC

Aristotelian sentence types are made false in AAPC by the situations in space
4. (One remembers from Section 4.2.4 that space 1 covers the situations where
O [[G]] or [[G]]  [[F]], and space
[[F]] [[G]], space 2 covers those where [[F]] O
3 those where [[F]] OO [[G]].)
Given the logical relations as defined in (4) of Section 2.3.3, one reads
from Figure 5.6b that the Conversions have been replaced with one-way
entailments from A to ¬I* (or from I* to ¬A) and from A* to ¬I (or from I
to ¬A*), since /A/ /¬I*/ and /A*/ /¬I/, but not vice versa.17 Then, the
subaltern entailments from A and A* have been preserved, since /A/  / I/
and /A*/  /I*/.18 The subcontrariety between I and I* has been lost, but I and
¬A are still subcontraries, since /I/ [ /¬A/¼U (and, of course, analogously
for I* and ¬A*).19 In addition, the number of contrary pairs has increased.

17 /A/= {1} and /¬I*/= {1,4}; therefore /A/  /¬I*/; therefore A ‘ ¬I* but not vice versa. Then, /A*/
= {3} and /¬I/= {3,4}; therefore /A*//¬I/; therefore A * ‘¬I but not vice versa.
18 /A/= {1} and /I/= {1,2}; therefore /A/  /I/; therefore A ‘ I but not vice versa. Then, /A*/= {3} and
/I*/= {2,3}; therefore /A*//I*/; therefore A* ‘ I* but not vice versa.
19 /I/= {1,2} and /¬A/= {2,3,4}; so /I/[/¬A/= U; so I and ¬A are subcontraries. Then, /I*/= {2,3}
and /¬A*/= {1,2,4}; so /I*/[/¬A*/= U; so I* and ¬A* are subcontraries. Note that I and I* are not
subcontraries: {1,2} [ {2,3} 6¼ U. This means that I and I* are logically independent in this system: they
can be both true, both false, singly true, and singly false.
178 The Logic of Language

b. ¬A {2}
a.
CD
[[F]] = Ø {1}A C ¬I* {1,2}
¬A ¬I* C
CD CD
A ¬I* CD
C
2 1 2 CD

{1} I C I* Ø
I ¬A*
[[F]] ° Ø ¬A* C
C
¬I CD C
U
C
{2}¬I A* Ø
CD
CD
¬A* {1,2}
FIGURE 5.8 VS-model and octagon for ALL F IS F and so on

The corresponding octagonal graph is shown in Figure 5.7a, where the valua-
tion spaces of Figure 5.6b are given for each vertex. A comparison with the
graph-theoretically complete octagonal graph for ABPC of Figure 5.2b above
shows that AAPC is less powerful than ABPC but a great deal more powerful
than SMPC, whose poverty is shown in Figure 4.6b.
Abelard devotes some space to sentences where the F- and G-predicates are
identical, as in his example EVERY human is human, SOME human is NOT-
human, and so on. In such sentences, AAPC remains intact, as it should, but
acquires a number of extra entailment relations, many of which are counter-
intuitive. Since the extensions of the F- and G-predicates coincide, the spaces
1 and 2 of Figure 5.6b collapse into one space. Space 3 disappears, as there can
be no difference between two identical predicate extensions, and space 4
remains intact, leaving just two spaces, one for [[F]] 6¼
and one for [[F]] ¼

, as shown in Figure 5.8a, which returns the octagon of Figure 5.7b. In Figure
5.8a, A* and I* are not represented, since in AAPC sentences of the form ALL F
is NOT-F and SOME F is NOT-F are necessarily false, as noticed by Abelard. (It is
an interesting exercise to read Abelard’s text quoted above with the Figures 5.7
and 5.8 at hand.)
The octagonal graph of Figure 5.8b is graph-theoretically complete, in
that there is at least one, sometimes two, logical relations between every two
vertices, but the important point is that all logical relations of AAPC, as
 Aristotle, the commentators, and Abelard 179

shown in Figures 5.5b and 5.6a, are preserved and in some cases reinforced.
Figure 5.8b just has a much richer supply of logical relations than ordinary
AAPC, though the relations it has in excess of those of ordinary AAPC are
largely counterintuitive, owing to the fact that in AAPC with two identical
predicates sentences of the types A* and I* are necessarily false and thus
have a null VS. That being so they mathematically entail any arbitrary
sentence—a form of entailment that counts as nonnatural and counterintu-
itive, as is argued in Chapter 3. At the same time, they are contrary with
regard to any arbitrary sentence, since they will never be true together with
any sentence, including themselves. The combination of entailment and
contrariety is, of course, highly counterintuitive. But then, predicate calcu-
lus with two identical predicates is something of a logician’s prank and is not
part of ordinary language or cognition. One remembers that one of the
principles of natural set theory proposed in Chapter 3 specifically rules out
the identity of sets that have been introduced as sets in their own right and
hence the identity of predicate extensions that have been given different
names.
What results from all this is that giving up the Conversions for one-way
entailments and declaring A-type and A*-type sentences false for situations
where [[F]] ¼
results in a better logical and linguistic deal than keeping the
Conversions and declaring A-type and A*-type sentences true in such situa-
tions, as SMPC does.
Why Abelard’s solution has never been incorporated into the logical tradi-
tion is hard to say. In the absence of a specialized study on the Abelard
tradition, one can only guess at the reasons for this historical anomaly.20 In
the context of the present work, Abelard’s contribution is highly significant, as
it brings us one step closer to the isolation of the situations where [[F]] ¼
by
declaring all four Aristotelian sentence types false in those situations.
Yet, even with Abelard’s solution, we are still far removed from an adequate
treatment of the logic of quantification in natural language. Somehow, space 4
must be ‘put on hold’ and treated as being hors concours, but this cannot be
done with the means at our disposal at this point in the exposition. It is done
in Chapter 10, where the principle of strict bivalence is sacrificed in the

20 One can think of a few possible reasons, given his superior intellect, combined with his gift for
debating and ridicule, his insistence on rational as opposed to mystic thinking, his frequent clashes
with official theology, and his defiant attitude with regard to the stifling moral prescriptions imposed
and enforced by the Church, which twice condemned him for heresy.
180 The Logic of Language

context of the theory of presupposition and contextual anchoring. What

can be shown, however, within the confines of strict bivalence and a strictly
extensional ontology, is that the setting apart of the situations where [[F]] ¼

as a separate, marked class is functional with regard to linguistic interaction.
This is done in the following chapter.
 6

The functionality of the Square

and of BNPC

6.1 How to isolate the cases with a null F-class:

the purpose of space 4
The previous chapters have shown that when no account is taken of situations
where [[F]] ¼
(space 4 in most of the VS-models considered), SMPC
is isomorphic with propositional calculus and identical with both
AAPC and ABPC. Prima facie, therefore, it would be very helpful if situations
where [[F]] ¼
could be dispensed with in predicate logic, since that would
leave only BNPC to cope with as a possible alternative. In fact, one of the final
conclusions of this book, reached in Chapter 10, is that the logic of language
has indeed proved capable of doing away with situations where [[F]] ¼
,
which are, in the end, reserved for cases of presupposition failure. Language
and logic have thus, in a way, ‘screened off ’ presupposition failure from the
default, presupposition-preserving systems of both propositional and predi-
cate logic—and with good reason.
In the present chapter we will see that this ‘default turn’ taken by natural
language and cognition is, perhaps unexpectedly, highly functional. The
avoidance not only of the null set but also of the total set of all objects OBJ,
as required in our natural-set-theory hypothesis (Section 3.2), automatically
reduces SMPC not only to ABPC but, with the BNPC quantifiers as defined in
Section 3.4.1, also to BNPC.
We are not proposing, of course, to dispense with situations where [[F]] ¼

just like that. Quantifiers express relations between sets and the null set plays
its part not only in the set-theoretic game but also in real-life situations. Actual
situations where [[F]] ¼
abound in any world. A sentence like:
(6.1) All dodoes are in good health.
is both perfectly grammatical and perfectly interpretable. It may even be a
normal sentence in some fictional story, or in some story about the island
182 The Logic of Language

Mauritius as it was before the year 1700, when there were still dodoes alive. It
must, therefore, be considered to be a rightful member of the class of A-
sentences even if the F-class, the class of dodoes, happens not to be instantiated
in the actual world as it is now. No logic can afford to leave such sentences out
of account: U is the set of ALL possible situations, not just of those that happen
to have a nonnull F-class. Nor can the predicate variable F be restricted to
those predicates that happen to have a nonnull extension. Space 4, therefore,
reserved for situations where [[F]] ¼
, cannot be eliminated from the system.
It can, however, be isolated and ‘put on hold’. For one thing, space 4 is special in
that it is only needed for the purpose of catering for situations where [[F]] ¼
.
This is shown as follows. For a situation sit to belong to space 1 or 2, it is necessary
that [[F]] 6¼
and [[G]] ¼ 6
, since both spaces make I-sentences true, which
require a nonnull intersection of [[F]] and [[G]]. For sit to belong to space 3, which
houses the situations where SOME F is NOT-G is true, [[F]] and ½½G must be
nonnull, but [[G]] is free: either [[G]] ¼
or [[G]] 6¼
, since all that is required
is a nonnull intersection of [[F]] and the complement of [[G]]. Only in space 4 is
it allowed, and necessary, that [[F]] ¼
, while [[G]] may or may not be null.
This insight is not revolutionary. It is widely known that SMPC is the result of
adding situations where [[F]] ¼
to the set of situations in which ABPC is valid.
Yet it is useful to look at this fact a little more closely. Let us speak of FACT 1:
FACT 1
If the set of situations where [[F]] ¼
is disregarded, SMPC is
transformed into traditional Aristotelian-Boethian predicate calculus
(ABPC or the Square) and becomes logically isomorphic with
standard propositional calculus. This provides a wealth of logical
relations that are absent in SMPC.
FACT 1 is important because there is more. It is not hard to see that, if the set
of situations where [[F]] ¼
can somehow be put on hold, it is only the
VS-extension of A-type or A*-type sentences that is affected. The reason is
simple: space 4 in the VS-model of SMPC (Figure 3.11), which is reserved for the
situations where [[F]] ¼
, only makes sentences of type A and A* true; the I-type
or I*-type sentences still remain restricted to spaces 1, 2, and 3. This gives us FACT 2,
which is likewise not revolutionary, yet much less present in the minds of the
professionals than FACT 1, despite its relevance in the context of the logic of
language:
FACT 2
The reduction of SMPC to ABPC only affects A-type or A*-type
sentences and has no effect on the VS-extension of I-type or I*-type
 The functionality of the Square and of BNPC 183

sentences. Under a strictly extensional ontology, the difference between

the two calculi must therefore lie in the semantics of the universal
quantifier 8.

6.2 Extreme values are uninformative in standard modern

predicate calculus
Now we come to a point which has so far not played any role at all in the
literature even though it appears to be of great importance for an adequate
insight into the way logic is incarnated in human language and cognition. Let
us define the notion of extreme value for a set variable as follows:
When we say that, for a predicate X, its extension [[X]] has an EXTREME
we mean to say that either [[X]] ¼
or [[X]] ¼ OBJ.
VALUE,

The point we wish to make is that when [[F]] or [[G]] or both have an
extreme value, then the set-theoretic relation between the two sets involved is
fully determined. Hence, the truth value, in SMPC, of quantified L-proposi-
tions in the language of predicate calculus (LPredC) involving the predicates
F and G, with or without the external or internal negation, follows automati-
cally: their truth or falsity is determined by Boolean computation alone,
no further inspection of the world being needed. But when the situation
is such that neither [[F]] nor [[G]] has an extreme value, the truth value of
an L-proposition describing that situation depends on the contingent set-
theoretic relation between [[F]] and [[G]]. The class of situations characterized
by the condition that neither [[F]] nor [[G]] has an extreme value we call the
CLASS OF CONTINGENT (OR MUNDANE) SITUATIONS. The sentences that describe a
contingent situation we call CONTINGENT (OR MUNDANE) SENTENCES.
Some comment is in order. What is said is that in all cases where [[F]] ¼

or [[G]] ¼
or [[F]] ¼ OBJ or [[G]] ¼ OBJ, the truth value of any L-
proposition in LPredC, involving the predicates F and G is fully determined
by Boolean computation. This is so because:
(a) The extreme values
and OBJ are instances of Boolean 0 and 1,
respectively.
(b) The quantifiers in SMPC are exclusively defined in terms of the set-
theoretical functions \, [, and Complement.
(c) The set-theoretical functions \, [, and Complement correspond to
the Boolean functions of multiplication, addition, and complement,
respectively.
184 The Logic of Language

(d) These Boolean functions are defined in terms of the constants 1, 0 and
at most one variable symbol (see Section 2.3.2).
As regards (d), one should note that, as soon as more than one variable
symbol is involved whose value is not Boolean 1 or 0, the values of the
Boolean functions are no longer determined a priori but depend on any
values the variable symbols may have in any application of Boolean algebra
to some domain.
From (a)–(d) it follows that, in all cases where [[F]] or [[G]] or both have an
extreme value, the truth value of any quantified L-proposition in LPredC
involving the predicates F and G is computable from these values, while this
is not so when both [[F]] and [[G]] are natural sets in the sense of Chapter 3 and
thus avoid extreme values. The inverse does not hold. For example, when 8x
(Gx,Fx) is true (in SMPC), both [[F]] and [[G]] may be null or nonnull or equal
to OBJ, though not in every combination. The existential quantifier is more
restrictive: when ∃x[Gx,Fx] is true, neither [[F]] nor [[G]] may be null, though
either may be equal to OBJ (provided OBJ 6¼
). We thus formulate FACT 3:
FACT 3
When [[F]] ¼
or [[G]] ¼
or [[F]] ¼ OBJ or [[G]] ¼ OBJ, the truth
value of any L-proposition in LPredC involving the predicates F and G is
fully determined by Boolean computation.
The reason this is relevant is that the use of an L-proposition of LPredC in
natural language is now seen to be informative only for situations belonging
to the class of contingent situations, that is, for situations where [[F]] 6¼
6¼
OBJ and [[G]] 6¼
6¼ OBJ. In the other cases, where the truth value simply
follows from an extreme value, it is more informative to specify the status of
[[F]] and [[G]], which gives the set-theoretic relation between them as an
automatic consequence. The relevance of FACT 3, therefore, lies not so much
in the logic itself as in more pragmatic considerations regarding the function-
ality of human language.

6.3 The functionality of excluding extreme values

The next step in the argument is easy to guess: it is shown that logic itself
ferrets out redundancy. In doing so, it provides the instruments for an
optimal application to the practical needs of language users—a point of
view not so far developed in pragmatics or semantics. The argument is
that, if the evolution of human language has favoured informativeness in
sentences and if it turns out that those sentence types that are by definition
 The functionality of the Square and of BNPC 185

informative happen to fit into a logical system whose application is restricted

but which is (much) more powerful than SMPC (or even AAPC), then one
will not be surprised to find that language has made the more powerful
but restricted logical system its own and has sought remedies for the parts
that had to be cut off.
In other words, human language will then be seen to favour the more
powerful predicate logic that comes with the contingent class of situations
where [[F]] 6¼
6¼ OBJ and [[G]] 6¼
6¼ OBJ, and to have done so without
compromising the soundness of the system. How this tour de force has been
achieved is described in global terms in Section 4.1, and in more precise terms
in Chapter 10. For the moment we ask the reader to bear with us while we
unravel the standard system of SMPC.
The clearest way of showing what is at issue is to present a graphic
breakdown. Consider Figure 6.1, which presents a matrix for the various
possible situations that make L-propositions in LPredC with the predicates F
and G true, without the restriction that neither [[F]] nor [[G]] may be null or
equal to OBJ. The status of [[F]] and [[G]] is specified for each of the ten
columns: they are either null or nonnull or equal or not equal to OBJ, which is
itself again nonnull or null.
In Figure 6.1, the following notational conventions are used.
(a) ‘[[X]]’ is replaced with ‘X’ for any predicate X.
(b) ‘O
’ stands for NULL OBJ (therefore O
¼ O
).
(c) ‘Oþ’ stands for NONNULL OBJ (therefore Oþ ¼
).
(d) ‘X 6¼ ev’ (‘ev’: extreme value) stands for ‘[[X]] 6¼
6¼ OBJ’
for any predicate X (in terms of Chapter 3 we say: X is a natural set).
(e) ‘X OO Y’ stands for X \ Y 6¼
; X 6¼ Y 6¼ ev; X 6 Y; Y6 X (mutual-
partial intersection of two natural sets).
Each possible combination of [[F]] and [[G]] status with each of the eight
set-theoretic relations that are possible between [[F]] and [[G]] has a slot in the
matrix. There are 80 such slots for the combined [[F]] and [[G]] status and the
set-theoretic relations specified. The marking ‘–’ for a slot means that the
values specified for the column entail that the set-theoretic relation specified
for the row cannot hold. Such slots, therefore, represent impossible situations.
The marking ‘þ’ means that the values specified for the column entail that the
set-theoretic relation specified for the row must hold. The marking ‘?’ means
that the values specified for the column are insufficient to determine whether
the set-theoretic relation specified for the row does or does not hold.
Take, for example, slot 3, the top slot in column iii: when [[F]] ¼ Oþ and
[[G]] 6¼ ev—that is, [[F]] equals a nonnull total set of objects OBJ and [[G]] is a
186 The Logic of Language

i ii iii iv v vi vii viii ix x

F ≠ ev F = Ø F = O+ F ≠ ev F = Ø F = O+ F ≠ ev F = Ø F = O+ F = Oø
G ≠ ev G ≠ ev G ≠ ev G = Ø G = Ø G = Ø G = O+ G = O+ G = O+ G = Oø

1 2 3 4 5 6 7 8 9 10
1 F⊂ G ? + – – – – + + – –
11 12 13 14 15 16 17 18 19 20
2 G⊂F ? – + + – + – – – –
21 22 23 24 25 26 27 28 29 30
3 F=G ? – – – + – – – + +
31 32 33 34 35 36 37 38 39 40
4 FO
OG ? – – – – – – – – –
41 42 43 44 45 46 47 48 49 50
5 F∩G = Ø ? + – + + + – + – +
51 52 53 54 55 56 57 58 59 60
6 F⊂ G ? – + – – – + – + –
61 62 63 64 65 66 67 68 69 70
7 F=G ? – – – – + – + – +
71 72 73 74 75 76 77 78 79 80
8 F⊂ G ? + – + + – – – – –

FIGURE 6.1 Situation table for LPredC

natural set of objects and thus a proper part of OBJ—the relation of proper
inclusion of [[F]] in [[G]] (F  G) cannot possibly hold, so that slot 3 must be
marked ‘–’ and thus represents an impossible situation. Or take slot 7, the top
slot in column vii. Here it is specified that [[F]] is a natural set (‘F 6¼ ev’) and
[[G]] equals nonnull OBJ (‘G ¼ Oþ’). That being so it simply follows that [[F]]
is properly included in [[G]]. One notes that the condition of row 4 ([[F]] and
[[G]] are both natural sets and intersect partially) is compatible only with the
condition of column i and incompatible with those of all other columns.
What does Figure 6.1 tell us? First, it is clear that column i represents the
class of contingent situations as defined above, since it is only in column i that
both [[F]] and [[G]] avoid the extreme values
and OBJ. This illustrates the
fact that when either [[F]] or [[G]] has an extreme value, the truth value of a
set-relational statement is predicted on a priori grounds. This is perhaps
undramatic, but it does no harm to see it visually illustrated.
The next point has a little more drama to it. The reduction of the predicate-
logical constants to set theory allows us to set up valuation spaces for the
various L-proposition types of LPredC. The procedure specified in (6.2) allows
us to list the slots that yield truth for any L-proposition of SMPC and thus
compose a VS-model:
 The functionality of the Square and of BNPC 187

(6.2) Check for each slot n whether formula is true given R and C, where:
formula stands for any predicate-logic L-proposition,
R for the set-theoretic relation specified for the row and
C for the status of [[F]] and [[G]] specified for the column.
When formula is true, add the slot number n to the list.
For example, given the predicate-logical L-proposition ALL F is G or A,
formally written 8x[Gx,Fx], we list, for row 1, the following slots: 1, 2, 7, and 8.
Row 2 yields no slot, since 8x[Gx,Fx] cannot be true when [[G]]  [[F]]. Row 3
yields the slots 21, 25, 29, and 30, because 8x[Gx,Fx] is always true when [[F]]
¼ [[G]]. Row 4 again yields no slot because 8x[Gx,Fx] is false when [[F]] O O
[[G]] and all other slots in row 4 represent impossible situations. It is a bit
tedious to do this for all eight L-proposition types of SMPC and for all rows
and columns, but the result is given in (6.5).
One should note that the procedure defined in (6.2) automatically selects
only those matrix slots that are marked ‘þ’. This is because the slots marked ‘–’
represent impossible situations, given their combined values in R and C. These
slots, therefore, cannot play a part in a valuation-space model listing all
possible situations. It is useful to have the universe U of all possible situations
at hand:

(6.3) The universe U of all possible situations (all slots marked ‘?’ or ‘þ’):
{ 1,2,7,8,11,13,14,16,21,25,29,30,31,41,42,44,45,46,48,50,51,53,57,59,61,66,
68,70,71,72,74,75}
Some situations have been printed in boldface and larger font. This has
been done to make them stand out as the members of the class of contingent
situations, listed separately in (6.4). These, one remembers, are the situations
where, given the status of [[F]] and [[G]], their set-theoretic relation does not
follow automatically.
(6.4) {1,11,21,31,41,51,61,71}
In all other situations the set-theoretic relation stated for [[F]] and [[G]] in a
row follows from their status specified for them in their column. Take, for
example, situation (slot) 2: given the column condition [[F]] ¼
, [[G]] 6¼ ev,
it is necessarily so that [[F]]  [[F]]. Or take situation 44 with the column
condition [[F]] 6¼ ev and [[G]] ¼
. Here it is necessarily so that [[F]] \ [[G]] ¼

. Analogously for all situations whose numbers are not printed in bold.
Here, then, is the valuation-space table for all eight basic expressions in
SMPC:
188 The Logic of Language

(6.5) a. /8x[Gx,Fx]/ (/A/) ¼ {1,2,7,8,21,25,29,30,42,45,48,50,57,59,68,70,72,75}

b. /¬8x[Gx,Fx]/ (/¬A/) ¼ {11,13,14,16,31,41,44,46,51,53,61,66, 71, 74}
c. /8x[¬Gx,Fx]/ (/A*/) ¼ {2,8,25,30,42,45,46,48,50,66,68,70, 71,72,74,75}
d. /¬8x[¬Gx,Fx]/ (/¬A*/) ¼ {1,7,11,13,14,16,21,29,31,41,44,51,53,57,59,61}
e. /∃x[Gx,Fx]/
/ ∃x(Fx,Gx)/ (/I/) ¼ {1,7,11,13,14,16,21,29,31,41,44,51,
53,57,59,61}
f. /¬∃x[Gx,Fx]/
/¬∃ x(Fx,Gx)/ (/¬ I/) ¼ {2,8,25,30,42,45,46,48,50,66,68,70,
71,72,74,75}
g. /∃x[¬Gx,Fx]/ (/I*/) ¼ {11,13,14,16,31,41,44,46,51,53,61,66,71,74}
h. /¬∃x[¬Gx,Fx]/ (/¬I*/) ¼ {1,2,7,8,21,25,29,30,42,45,48,50,57,59,
68,70,72,75}
Note that Figure 6.1, the procedure (6.2), the universe U of all possible
situations (6.3), the class of contingent situations (6.4) and the valuation-
space table (6.5) are all based on SMPC, where the subalterns do not hold but
the Conversions do.
All this illustrates the point, made above in connection with FACT 3, that
quantified sentences (which describe set-theoretic relations between two sets)
are maximally informative when describing contingent situations—that is,
those listed in (6.4) and in column i of Figure 6.2. This is so because when
they describe a noncontingent situation, it is superfluous to express the
relation that holds between the sets involved, since it is implicit in the values
specified for them. In such cases, therefore, it is more economical merely to
specify these values—something which is usually implicit in the current
context or situation.
We are now coming closer to the conclusion we have been after for the
past few pages. Suppose it were possible to construct a partial logic just for
the class of contingent situations specified in (6.4) and therefore involving
only natural sets, then the resulting predicate logic would be identical
to ABPC, described by a complete octagonal graph as in Figures
4.5c or 5.2b. This is easily shown by simply singling out the bold numbers
from (6.5):
(6.6) a. /8x[Gx,Fx]/ (/A/) = {1, 21}
b. /¬8x[Gx,Fx]/ (/¬A/) = {11, 31, 41, 51, 61, 71}
c. /8x[¬Gx,Fx]/ (/A*/) = {71}
d. /¬8x[¬Gx,Fx]/ (/¬A*/) = {1, 11, 21, 31, 41, 51, 61}
e. /∃x[Gx,Fx]/ (/I/) = {1, 11, 21, 31, 41, 51, 61}
 The functionality of the Square and of BNPC 189

f. /¬∃x[Gx,Fx]/ (/¬I/) = {71}

g. /∃x[¬Gx,Fx]/ (/I*/) = {11, 31, 41, 51, 61, 71}
h. /¬∃x[¬Gx,Fx]/ (/¬I*/) = {1, 21}
Now A ‘ I, since {1, 21}  {1, 11, 21, 31, 41, 51, 61} and likewise A* ‘ I*, since
{71}  {11, 31, 41, 51, 61, 71}, two entailments that do not hold for (6.5).
Moreover, the Conversions hold since /A/ ¼ /¬I*/, /A*/ ¼ /¬I/, /I*/ ¼ /¬A/
and /I/ ¼ /¬A*/. It thus seems that a concentration on column i of Figure 6.2
brings along a very significant enrichment of the logical power of the system
involved.
This was to be expected, since it is well known that ABPC is valid when
[[F]] 6¼
and [[G]] 6¼
, conditions that are satisfied by column i. But the
validity of ABPC can be extended well beyond column i, provided, of
course, the logic can be kept sound despite the cancelling of certain classes
of situations. In fact, in order to save ABPC all that has to be got rid of is
columns ii, v, viii, and x, where [[F]] ¼
. The universe of situations U of
SMPC is then reduced to UR, which comprises only the following 20
situation types:
(6.7) UR ¼ {1,7,11,13,14,16,21,29,31,41,44,46,51,53,57,59,61,66,71,74}
The valuation spaces thus restricted then look as follows:
(6.8) a. /A/ = {1,7,21,29,57,59}
b. /¬A/ = {11,13,14,16,31,41,44,46,51,53,61,66,71,74}
c. /A*/ = {46,66,71,74}
d. /¬A*/ = {1,7,11,13,14,16,21,29,31,41,44,51,53,57,59,61}
e. / I/ = {1,7,11,13,14,16,21,29,31,41,44,51,53,57,59,61}
f. /¬I/ = {46,66,71,74}
g. / I*/ = {11,13,14,16,31,41,44,46,51,53,61,66,71,74}
h. /¬I*/ = {1,7,21,29,57,59}
Now ABPC still holds: A ‘ I, since {1,7,21,29,57,59}  {1,7,11,13,14,16,21,29,
31,41,44,51,53,57,59,61} and A* ‘ I*, since {46,66,71,74}  {11,13,14,16,31,
41,44,46,51,53,61,66,71,74}. The Conversions also still hold, since, again, /A/
¼ /¬I*/, /A*/ ¼ /¬I/, /I*/ ¼ /¬A/ and /I/ ¼ /¬A*/.
Incidentally, AAPC equals SMPC but for the fact that 2, 8, 25, 30, 42, 45, 48,
50, 68, 70, 72, and 75 are moved from (6.5a) to (6.5b) and from (6.5c) to (6.5d),
because A and A* are judged false in AAPC but true in SMPC when [[F]] ¼
.
190 The Logic of Language

This yields AAPC. AAPC also holds for the mundane situations plus those
where [[F]] 6¼
, but it stops being valid when the situations where [[F]] ¼

are taken into account. Within these constraints, however, AAPC is not
maximally powerful: ABPC, which operates within the same constraints, is
more powerful. But AAPC has the advantage of lacking UEI.
We thus see in clear detail what we had suspected all along: if the elimina-
tion of the class of situations where [[F]] ¼
is feasible, that will be sufficient
to boost the predicate logic from the impoverished SMPC to the maximally
powerful ABPC. What has also been shown is that the elimination of situa-
tions with a null F-class does not affect the valuation spaces of I-type and
I*-type sentences: their valuation spaces in (6.8) are identical to those in (6.5).
It is therefore only the A-type and A*-type sentences and, of course, the
complements of the I-type and I*-type sentences, that are affected by the
shrinking from U to UR. This illustrates what was stated above in FACT 2.

6.4 The functionality of BNPC

What we have not investigated so far is what happens to SMPC when all
situation classes (slots) that are incompatible with the restrictions valid for
BNPC are removed from Figure 6.1 on account of these slots not being
operative in the cognitive machinery at basic-natural level. In order to carry
out such an investigation, we must specify the semantics of the quantifiers
8, ∃ and N (NO for sentence type N) in BNPC. This is done in the following
manner. In BNPC, the quantifier 8 is defined as requiring that [[F]]  [[G]]
([[F]] is a subset of [[G]]) and [[F]], [[G]] 6¼ ev. The quantifier ∃ is defined as
requiring that [[F]] O O [[G]] ([[F]] and [[G]] M-partially intersect) or [[G]] 
[[F]] (that is, just in case [[F]] \ [[G]] 6¼ [[F]]). The quantifier N is defined as
requiring that [[F]] O O [[G]] (that is, [[F]] \ [[G]] ¼
, while [[F]], [[G]] 6¼ ev).
This eliminates columns ii–x as well as row 3, reserved for the nonnatural
identity relation of the natural sets [[F]] and [[G]]. (External and internal
negation are standard, but for Restriction 6 of Section 3.2.2.)
All that remains of Figure 6.1 when cut down in the manner indicated is
column 1, listing the mundane situation classes, minus slot 21. The valuation
spaces specified for the eight basic sentence types in SMPC were given in (6.5)
above. The reduction of (6.5) by the elimination of columns ii–x and of
row 3 in Figure 6.2 gives (6.9a–h). The new BNPC-quantifier N as defined
above, giving rise to the basic sentence types N, N*, ¬N and ¬N*, makes for
(6.9i–l):
 The functionality of the Square and of BNPC 191

(6.9) a. /A/ = {1 }
b. /¬A/ = {11, 31, 41, 51, 61, 71}
c. /A*/ = {41, 61, 71}
d. /¬A*/ = {1, 11, 31, 51}
e. /I/ = {11, 31, 51}
f. /¬I/ = {1, 41, 61, 71}
g. /I*/ = {11, 31, 51}
h. /¬I*/ = {1, 41, 61, 71}
i. /N/ = {41, 61, 71}
j. /¬N/ = {1, 11, 31, 51}
k. /N*/ = {1}
l. /¬N*/ = {11, 31, 41, 51, 61, 71}
The following logical relations are now seen to hold, precisely as specified
in Figure 3.7 of Chapter 3:
(6.10) A
N* ¬A
¬N*
A*
N ¬A*
¬N
I
I* ¬I
¬I*
A ‘ ¬A* N* ‘ ¬A*
A ‘ ¬I / ¬I* N* ‘ ¬I / ¬I*
A ‘ ¬N N* ‘ ¬N
A* ‘ ¬A / ¬N* N ‘ ¬A / ¬N*
A* ‘ ¬I / ¬I* N ‘ ¬I / ¬I*
I ‘ ¬A / ¬N* I* ‘ ¬A / ¬N*
I ‘ ¬A* / ¬N I* ‘ ¬A* / ¬N
We may also keep the BNPC-definitions of the three quantifiers 8, ∃ and
N but disregard the basic-natural restrictions of Section 3.2.2, thereby
reading BNPC as an unrestricted logic. When we do that, BNPC is extended
to cover all situation classes of Figure 6.2. The VSs thus specified are listed
in (6.11):
(6.11) The VSs of the basic sentence types of BNPC as an unrestricted logic
a. /A/ ¼ {1}
b. /¬A/ ¼ {2,7,8,11,13,14,16,21,25,29,30,31,41,42,44,45,46,48,50,
51,53,57,59,61,66,68,70,71,72,74,75}
c. /A*/ ¼ {41,61,71}
192 The Logic of Language

d. /¬A*/ ¼ {1,2,7,8,11,13,14,16,21,25,29,30,31,42,44,45,46,48,
50,51,53,57,59,66,68,70,72,74,75}
e. /I/ ¼ {11,31,51}
f. /¬I/ ¼ {1,2,7,8,13,14,16,21,25,29,30,41,42,44,45,46,48,
50,53,57,59,61,66,68,70,71,72,74,75}
g. /I*/ ¼ {11,31,51}
h. /¬I*/ ¼ {1,2,7,8,13,14,16,21,25,29,30,41,42,44,45,46,48,50,
53,57,59, 61,66,68,70,71,72,74,75}
i. /N/ ¼ {41,61,71}
j. /¬N/ ¼ {1,2,7,8,11,13,14,16,21,25,29,30,31,42,44,45,46,48,50,
51,53,57,59,66,68,70,72,74,75}
k. /N*/ ¼ {1}
l. /¬N*/ ¼ {2,7,8,11,13,14,16,21,25,29,30,31,41,42,44,45,46,48,50,
51,53,57,59,61,66,68,70,71,72,74,75}
Here one sees that even when extended to cover all situations, the six
nonnegated basic sentence types of BNPC, A, A*, I, I*, N, and N*, are still
restricted to the mundane situations represented in column i. This remark-
able property is not shared by ABPC, as is shown by (6.8) above.
It thus appears that BNPC, though invariably dismissed as unimportant by
professional logicians, is not only a superbly powerful logic but is also highly
functional in that it automatically focuses its nonnegated basic expressions on
the mundane situations. Yet it has the fatal drawback of not allowing for
existential statements in the absence of complete knowledge of the domain.
BNPC also has a second, less serious drawback, to do with expressive power.
As a result of the incorporation of PNST–2 (‘natural sets are distinct’) into the
semantics of the BNPC quantifiers, BNPC is unable to produce a true
quantified sentence when [[F]] ¼ [[G]]: the slots 21, 25, 29 and 30, filling the
row characterized by the relation [[F]] ¼ [[G]], do not occur in the VS of any
of the six nonnegated basic sentence types of BNPC.1 The same does not hold
for the slots in row 7, defined by the relation ½½F ¼ [[G]] (or, equivalently,
[[F]] ¼ ½½G): slot 61 figures in both /A*/ and /N/ of (6.11), which, as one can
see, are identical. This is because in row 7 the sets [[F]] and [[G]] are both
natural sets and distinct from each other. Hamilton and Jespersen can thus
rest in peace: their logic has finally found the recognition it deserves, even
though it has its limitations.

1
This lack of expressive power does not occur in Hamilton’s notation as presented in Hamilton
(1866). As is shown in Section 3.4.2, Hamilton writes tF = tG when [[F]] = [[G]].
 The functionality of the Square and of BNPC 193

6.5 Conclusion
It is time to summarize our conclusions. The first conclusion is that our
analysis of predicate calculus has shown that if there is a way to restrict
predicate calculus to those situations where [[F]] 6¼
, predicate calculus
will have the same maximal logical power as propositional calculus, as both
will then conform to the logical entailment system described by the complete
octagonal graphs of Figure 4.5c.
The second conclusion concerns a fact that has so far either been unknown
or been allowed to lie unexploited. We have been able to single out a core class
of contingent situations characterized by the condition that both [[F]]
and [[G]] avoid extreme values, that is, the condition that [[F]] 6¼
6¼
OBJ and [[G]] 6¼
6¼ OBJ. This class is significant (a) because the quantifying
L-propositions of LPredC are maximally informative when they describe such
situations (in all other cases the truth or falsity of an LPredC L-proposition
follows from the status of [[F]] or [[G]]) and (b) because a predicate logic
restricted to this class of contingent situations is maximally powerful (given
its eight basic expressions), while the general system of SMPC has only weak
logical power. Therefore, functionality will be boosted if a system of predicate
logic can be developed that sustains the restriction to the class of contingent
situations, so that quantifying sentences describing these situations can
benefit from a maximally powerful logic. In addition, we have found that
such a system will apply not only to the situations of the contingent class but
also to those other noncontingent situations where the F-class is nonnull. The
loss to the general SMPC system will thus remain limited, while the gain will
be maximal.
Finally, we have seen that BNPC is not only an extremely powerful predi-
cate calculus but is also an extremely functional one, in that, when all
restrictions are lifted, its nonnegated basic expressions automatically restrict
their valuation spaces to the mundane situations of column i.
The question we are facing now is the following: how will the predicate
logic of language cater for the situations that are missing in the restricted UR?
This question is answered in the following chapters.
 7

The context-sensitivity of speech

and language

7.1 What is context-sensitivity?

The context-sensitivity of natural language has been a spanner in the works of
all formal theories of language and its logic. Formal logicians, starting with
Aristotle, have tended to flee into the safety of eternal sentences, either leaving
occasion sentences out of their account or trying to reduce them to eternal
sentences. In linguistic theory, context-sensitivity has cropped up only re-
cently and during that short period it has mostly been treated at an informal
(pragmatic) rather than at a formal level. In this and the following chapters
we try to gain an overall view of all aspects involved in what we consider
context-sensitivity to be, and in doing so we attempt to push up the level of
formality, and hence of clarity and precision, of the analysis and description
of the phenomena concerned.
In succinct terms, context-sensitivity is the fact that, as a matter of princi-
ple, natural language utterances need an appeal to both encyclopedic and
contextual knowledge for their proper interpretation. Apart from (ad hoc or
conventionalized) nonlinguistic cues, such as gestures, that normally come
with spoken utterances, sentences (types) underlying utterances (tokens)
contain systematic lexical and grammatical devices that refer the listener in
specific ways to available encyclopedic and contextual knowledge. And speak-
ers must build into their sentences the appropriate context-related devices
needed for a proper interpretation.
Context-sensitivity, in other words, is the fact that listeners must
integrate their comprehension of the utterance as an expression of an underly-
ing type-level L-proposition or semantic analysis (SA) into the proper
overall interpretation in the token situation. We use the term comprehension
for the identification of the type-level, schematic, linguistically encoded
message, so that it can be repeated, written down, or transcribed, or said
 The context-sensitivity of speech and language 195

to be ambiguous or uninterpretable. Interpretation, by contrast, is taken

to be the full reconstruction of the hic et nunc intent—that is, speech
act-cum-proposition—conceived and expressed by the speaker, or, if one
wishes, the application of the schematically organized linguistic expression
to a unique situation complete with the multitude of facets and angles from
which it can be considered.
In order to be properly interpreted, utterances need context, in the sense
of cognitive backing shared between speaker and listener. In fact, this depen-
dency on shared knowledge is so deeply ingrained in the semantic system of
human language that one is entitled to say that, across the board and as a
matter of principle, utterance interpretation needs a shared-knowledge base
for it to work properly. The few utterance types that appear to be interpretable
without shared knowledge may be considered cases whose context-sensitivity,
and thus their dependency on shared knowledge, has been reduced to zero,
but their occurrence is far rarer than logicians and some philosophers of
language would have us believe. This is the well-known SHARED-KNOWLEDGE
THESIS, defended in particular by Clark (1992) as an indispensable prerequisite
for the proper interpretation of utterances.
For the shared-knowledge thesis to work, the schematic and incomplete
information carried by the incoming utterance has to be integrated somehow
with what speaker and listener possess in the way of accessible knowledge
(memory). This integration requires, inter alia, a mechanism which we call
ANCHORING, described in Chapter 3 of Volume I and appealed to repeatedly in
subsequent chapters. Anchoring is the process of linking up utterances with a
specific memory called DISCOURSE DOMAIN (D), which accumulates the infor-
mation contributed by successive utterances into a cognitively coherent
complex. The information carried by a single utterance consists, in principle,
in the INTENT associated with the utterance in question, as defined in Sections
3.1.1 and 4.3 of Volume I—that is, the total package of the committal or
speech-act force and the propositional content. This intent is stored in, or,
in the technical jargon, incremented to, the D at hand. Once a well-anchored
interpretation is achieved, the result is again fed and integrated into cognition
in specific ways. The fact that, as a matter of principle, utterances require
anchoring is constitutive for natural human language. Natural language
production and comprehension are, by constitution, incremental.
Context-sensitivity is nature’s way of ensuring that the time and energy-
consuming effort of producing physical utterances is minimized. This is
achieved by maximizing the dependency on what speaker and listener assume
196 The Logic of Language

to be knowledge shared between them: what is already known need not be

said, provided the speech material that is made physically available contains
the necessary cues as to what elements in the common knowledge are called
upon. Sometimes it is elements in speaker–hearer’s general encyclopedic or
world knowledge, sometimes it is the knowledge of features in the speech
situation, and in all but the rare instances of eternal sentences it is knowledge
of what has been built up between speaker and listener in the way of a specific
discourse domain: anchoring to an available D is an essential prerequisite for
successful interpretation, the result of which is then, in its turn, incremented
to the D at hand and subsequently to cognition as a whole.
Sentence-types are tailored to the anchoring requirements of the utterances
realizing them. They contain elements, call them ‘anchor hooks’ if you like,
that make their utterance tokens fit for interpretation only when the anchor
hooks are properly attached to the available D. This, of course, also poses
conditions on any D for the acceptance and integration of newly uttered
sentences: sentence-types are made, semantically, and also grammatically, in
such a way that they fit only into certain classes of D and not into others.
Sentence types pose conditions on Ds for proper anchoring and thus for
proper interpretation.
As far as is known today, there are at least four specific devices ensuring that
the anchoring conditions of sentences are satisfied: external anaphora, pre-
supposition, topic–comment structure, and open parameters in lexical mean-
ings, also called free variables. The latter are discussed in Section 7.4 (having
been the topic of discussion in Sections 1.3.3.2, 8.2.1, 9.6.3, and, especially, 9.7
of Volume I). The former three form the contents of Chapters 9, 10, and 11,
respectively. But before we take a closer look at these devices, we turn our
attention to the notion of discourse domain.

7.2 Discourse domains

It is probably premature to try to give a formal definition of the notion
‘discourse domain’ at the present primitive state of our knowledge about
them. The best we can do is build up the notion gradually, starting from
what seems to be required intuitively. With this proviso, we can say that a
discourse domain D is a form of memory storing both the speech-act quality and
the fully interpreted L-propositional content of the utterances produced by the
participants in a linguistic exchange, possibly a monologue, in their order of
production.
 The context-sensitivity of speech and language 197

A discourse domain is thus assumed to store token-level, fully con-

textually integrated interpretations, not just type-level meanings. This
assumption seems to be justified by the consideration that there would be
little point in having a memory for unfinished products, while great advan-
tages accrue from having a memory for products that have just functioned in a
specific context. It is thus, so to speak, at the gate of a discourse domain that
open access is required to all available conscious knowledge of speaker and
hearer, who can thus channel any appeal to the shared-knowledge
base required for proper interpretation. Having said this, one must admit
to the humbling fact that the mechanisms for linking up discourse domains
with the whole of conscious knowledge have so far remained almost totally
undisclosed.
Discourse domains must, to some degree at least, be specific for the use of
natural language utterances, since they show properties and mechanisms
that are not found in other forms of memory. Yet they also have properties
and mechanisms in common with other forms of memory, in particular with
general episodic memory. For example, a discourse domain shares with
general episodic memory the property that its degree of detail corresponds,
on the whole, with the recency of the token event: the older the event or
utterance, the more global and the less detail-specific the storage—apart
from token occurrences that stand out for some reason, often because
of some emotionally salient association. Yet discourse domains are also
subject to laws and principles that are not found in general episodic memory.
It is these that we focus on in the present context, taking for granted
that memory, with all its different compartments, its intricate storing
and retrieval systems, its distortions over time, and its subconscious caves,
is still largely a mystery.
It is important, right at the outset, to clear up a misunderstanding
which is as widespread as it is confusing. It is often thought, especially
among philosophers, that assertions are defined by the speaker’s intention
to say only what is true or at least what the speaker believes to be true and
by the listener’s presumption that that is what the speaker does. Take,
for example, Dummett’s view, discussed earlier in Section 4.2 of Volume I,
according to which assertions differ from other speech acts in that they do
not establish a socially binding commitment on the part of the speaker
but are merely meant to make the listener believe, on the presumption of
the speaker’s honesty, that their propositional content is true (Dummett
1973: 301–2):
198 The Logic of Language

A command has definite consequences: disobedience to a command given by a person

in authority confers on that person a right to punish, or at least reproach, the person
commanded. . . . An assertion has no such definite consequences. [ . . . ] Assertions
take place against the background of a custom of uttering them with the intention of
saying something true.

Another example, likewise discussed in Section 4.2 of Volume I, is found in

Fodor (1983: 132):
Strictly speaking, I suppose, a convention must be something one can adhere to if
one chooses; so perhaps the principle at issue is not ‘Say only what is true’ but
rather ‘Say only what you believe.’ General adherence to the latter injunction will
license inferences from utterances to how the world is, given the assumption (which is,
anyhow, in all sorts of ways epistemologically indispensable) that much of what people
believe is true.

Though plausible enough as a description of what often happens in

verbal interaction, an appeal to the principle of ‘saying only what you believe
to be true’ and to the fact that ‘much of what people believe is true’ utterly
misses the essential function of assertions—not to mention other kinds of
speech act. As has been emphasized repeatedly, the essential function of
an assertion is to enable the speaker to enter a commitment as to the truth
of the proposition expressed. It is this function that determines the incre-
mentation of assertions to discourse domains. Incrementations of either
assertions or other kinds of speech act have nothing to do with belief
and only a little with actual truth. They merely register what has been said
and under what social commitments, no matter whether anyone believes the
propositional content stored to be true or false—as long as there is something
to be taken to be true or false. It follows from the speech-act analysis
presented in Chapter 4 of Volume I that the primary operator over any D
is not belief but some form of commitment, appeal, or allowance. All com-
partments or subdomains of a D are linked up to an overarching D which
either commits the speaker to its truth, or constitutes the speaker’s overall
appeal to the listener, or sets rules of behaviour.1 Let us call this overarching
domain the COMMITMENT DOMAIN (even though it is a partial misnomer), as
distinct from any of its subdomains.2

1 Appellations (name-callings) may seem to form a special case. They hardly ever extend over a
whole discourse domain, though marvellous extended invectives are known in world literature.
2 The notion of commitment domain was first introduced in Hamblin (1970), who speaks of
commitment background.
 The context-sensitivity of speech and language 199

7.2.1 The commitment domain and further subdomains

7.2.1.1 The notion of subdomain The internal structure of a discourse do-
main has thus far been explored only to a very limited extent. For one thing, as
has been said, the fact that the speech-act quality of utterances must be
recorded has so far been largely neglected in both the more formally oriented
and the more impressionistic literature. Nor is there an even remotely com-
plete account, formal or informal, of the fact that the contributions made by
different speakers must be stored in different subdomains, labelled for each
speaker. How these different speaker domains are interrelated is, though
perhaps intuitively clear in individual cases when one sets oneself to think
about them, a question that still lacks a precise general answer. Some limited
amount of work has been done, in the context of model-theoretic semantics,
on intensional subdomains (treated as sets of possible worlds) and their
relation to the commitment domain, but close to nothing is known about
the interrelations between various intensional subdomains, nor is anything
known about the nonintensional subdomains created by negation, disjunc-
tion, implication, and conjunction. And the reason for this remarkable state of
affairs is to be sought in the fact that, since the 1960s, logic-inspired model-
theoretic semantics, which held total sway over theoretical semantic studies,
has been more interested in keeping up the paradigm as defined by its
originators than in discovering the facts of language. Semantics, in other
words, has been allowed to swerve far too much over to formalism, which
has disturbed the delicate balance that is to be observed between formalism
and ecologism (see Section 1.3 in Volume I).
There being thus not much to go by, it seems sensible to make a fresh start
and develop an uncluttered view of what the notion of subdomain can
sensibly be taken to involve. In defining, or describing, what is meant by
subdomain, we disregard for the moment the distinction between noninten-
sional or extensional subdomains on the one hand and intensional subdo-
mains on the other (the former having been left largely uninvestigated in the
formal-semantic literature). One remembers from Section 6.1 in Volume I
that the difference resides in the fact that free substitution of co-referring
terms salva veritate is blocked in intensional subdomains while it is not in
their extensional counterparts.
To define the notion of subdomain, we start with defining embedded
sentential terms (S-terms) in L-propositional structure (see also Section
6.2.3 in Volume I):3

3 For an elaborated grammatical theory of S-term embedding, see Seuren (1996).

200 The Logic of Language

An EMBEDDED S-TERM is an argument term to an L-propositional

predicate (which may be a logical operator, possibly ending up
in surface structure as an adverbial or morphological element)
structurally defined as an S-structure which can function both as a
full L-proposition with a truth value and as an L-propositional
function whose variable is bound by a higher quantifier (see
Section 2.3.5.1).
S-structures embedded under quantifiers are thus not embedded
S-terms as defined here, because these S-structures are per se propositional
functions containing at least one bound variable and can thus never
function as a full L-proposition. Analogously for relative clauses,
which again obligatorily contain a variable linking them up with the an-
tecedent.
The most straightforward cases of embedded Ss are, of course, direct-
object complement clauses under verbs like know, believe, hope, see, etc.,
no matter whether these clauses occur as fully tensed finite clauses under a
complementizer (that, if ) or whether they are reduced to an infinitival
or a participial. But subordinate clauses under a conjunction (because, while,
etc.) are likewise embedded Ss, since they can be analysed as sentential
(phrasal) argument terms to the conjunction, which is then analysed
as a predicate. A sentence like She left because she felt unwell is then
analysed, at SA-level, as ‘Because (S1 ½she left, S2 ½she felt unwell}’, where
S2 is the direct-object clause and S1 the subject matrix-S. The
grammar makes the predicate because incorporate the object clause
S2 ½she felt unwell, giving the complex predicate PRED[PRED[Because]S [she felt
unwell]], which is then, as a whole, lowered into the subject matrix-S to a
peripheral position—that is, either to the far left or to the far right.
These syntactic details are of lesser interest in the present context (see Seuren
1996 for a fully elaborate theory). What counts here is the general notion of
embedded S-term.
The incrementation of an embedded L-propositional S-term either creates
or augments a SUBDOMAIN.
A SUBDOMAIN within the overall commitment domain D is a subunit
of incrementation in which, subject to the kind of subdomain, either
an entire intent or just the L-propositional content is stored by
incrementation.
 The context-sensitivity of speech and language 201

7.2.1.2 Extensional and intensional subdomains Some subdomain-creating

predicates (operators) are such that an increment in their subdomain has
direct and unambiguous consequences for the commitment domain, or,
more precisely, for the immediately superordinate domain. Negation, for
example, is taken to increment the L-propositions in its scope in a separate
subdomain. But any L-proposition forming the scope of negation is immedi-
ately marked as being banned from the immediately superordinate domain,
on pain of inconsistency. Similarly for the disjunction operator OR: its two
or more component L-propositions are lodged in separate (mutually exclusive;
see Sections 3.3.2 and 8.2.3) subdomains, with the condition attached that one
of them is to be incremented to the immediately superordinate domain,
though it is as yet uncertain which. IF also creates its own subdomain, with
the condition attached that as soon as the antecedent L-proposition
is incremented to the immediately superordinate domain, so must the conse-
quent L-proposition be. And the conjunction operator AND may be seen as
creating a subdomain storing the component L-propositions in their order of
occurrence and requiring that the corresponding increments are immediately
added to the superordinate domain at hand.
In similar fashion, increments to the subdomain created by the object
S-term of the predicate cause and its cognates are immediately incremented
to the superordinate domain (see Section 6.2.3.1 in Volume I). When I say, for
example:
(7.1) The storm made the roof collapse.
the information that the roof collapsed is immediately understood as having
been added to my commitment domain as I spoke, if it had not been added
earlier (when ‘the storm’ is comment and ‘what made the roof collapse’ is
topic).
Many other subdomain-creating predicates, however, are not so simple.
It is well-known that FACTIVE PREDICATES induce the presupposition that
their factive S-term has been incremented to the superordinate domain
before the utterance sporting the factive predicate as its main predicate
can be incremented. As a rule, factive predicates create or continue INTENSION-
AL SUBDOMAINS, in which SSV is disallowed, as is seen from the sentences
(2a,b), repeated from Section 6.2.3.1 in Volume I, which are semantically
independent in that they may both be true, or both false, or singly true,
or singly false, even though the terms morning star and evening star denote the
same actual object:
202 The Logic of Language

(7.2) a. John realizes that the morning star is uninhabited.

b. John realizes that the evening star is uninhabited.
Not all factive S-terms, however, are intensional in this standard sense.
Consider the factive subject clause in (7.3a). Replacement of the term Jackson
by the term the butler does not affect the truth value of the utterances in
question as long as Jackson and the butler are the same person, even if this fact
is unknown to speaker and/or listener. If (7.3a) is true, so is (7.3b), and if
(7.3a) is false, so is (7.3b):
(7.3) a. That Jackson’s fingerprints are on the glass proves that he is the
murderer.
b. That the butler’s fingerprints are on the glass proves that he is the
murderer.
In general, an intensional subdomain that blocks SSV stands under an
operator (predicate) requiring reference to someone’s THOUGHTS for truth to
arise. This was Frege’s position and, as was pointed out in Section 6.1 in
Volume I, it seems to be correct in principle, the principle being that the terms
occurring in thoughts mentally denote cognitive representational packages
or ‘addresses’ representing actually existing or merely thought-up entities.
Addresses thus may or may not be instantiated by actual entities and they may
have to be identified or disidentified for truth to arise. How intensional and
nonintensional (extensional) subdomains are incremented is the subject of
discussion in Sections 8.1.3 and 8.2.
The intensional subdomains most commonly dealt with are created (or
continued), on the one hand, by the so-called PREDICATES OF PROPOSITIONAL
ATTITUDE, such as believe, hope, expect, wish, and many others, all of which
contain an appeal to the subject’s world of thought and, on the other, by the
MODAL PREDICATES, although, as is argued in the following section, the inten-
sional status of modal subdomains is highly questionable.
In general, domains under a propositional attitude predicate are not
restricted by the consistency criterion, since one may well truthfully attribute
to a human being inconsistent beliefs, hopes, wishes, and so on.4 Nor is SSV
freely allowed in these subdomains, since thoughts have wings of their own
and may fly off to virtual realms whose relation to the world construed as
being actual may become as tenuous as in Alice’s Wonderland, or more.

4 For example (with thanks to Barbara Partee), a person may very well believe in evolution theory
and at the same time believe that of necessity every human individual must have had two human
parents.
 The context-sensitivity of speech and language 203

7.2.1.3 The epistemic modal subdomains By contrast, domains under an

epistemic modal predicate are fully subject to the consistency criterion.
Whether they are also intensional in the sense of blocking free SSV is
a moot point. Epistemic modals are complex and hard to understand. In the
formal-semantics literature they take pride of place, next to predicates of
propositional attitude, as prototypical creators (or continuators) of intension-
al contexts. Yet a closer analysis reveals that it is at least not obvious that they
block free SSV in their embedded S-terms. They are thus not, or at least
not clearly, intensional in the intended sense. We will now have a somewhat
closer look at the epistemic modal predicates. Yet, since a full treatment
of natural language modals would require a monograph in its own right if
not more, we must content ourselves here with a summary discussion.
Let it be observed first that natural language modal predicates are
either EPISTEMIC or AGENTIVE, but never metaphysical. Metaphysical modalities
are philosophical constructs which vary with the ontology adopted.
Natural languages are, on the whole, unaffected by such philosophical lucu-
brations. What one finds in the lexicons of natural languages is, to begin
with, a basic distinction between epistemic and agentive modalities. Within
these two main divisions, modalities occur in many guises which, however,
seem to gravitate around the two poles of NECESSITY and POSSIBILITY. In
this respect they resemble the quantifiers, which also occur in many different
guises but gravitate around a universal and an existential pole. It has often
been pointed out that universal quantification corresponds with necessity
in modal logic, while existential quantifiers are on a par with operators of
possibility.
What epistemic possibility operators express is partly to do with the
compatibility of the proposition expressed in the argument S-term
with speaker’s knowledge at the time of speaking. In other words, it is implied
that, as far as speaker’s knowledge goes, nothing excludes the actuality of
the virtual situation referred to by the argument S-term. Analogously,
part of what epistemic necessity operators say is that, as far as speaker’s
knowledge goes, the actuality of the virtual situation referred to by the
argument S-term is a necessary consequence, on grounds of deduction,
induction, or knowledge of causal laws. Thus, when I say (7.4a), I imply
that it is compatible with my knowledge at the time of speaking that the
fire was caused by witchcraft. And when I say (7.4b), I imply that it follows
from what I know at the time of speaking that the fire was caused by
witchcraft.
204 The Logic of Language

(7.4) a. The fire may have been caused by witchcraft.

b. The fire must have been caused by witchcraft.
c. The fire cannot have been caused by witchcraft.
Yet epistemic modal statements are not simply statements about compati-
bility with, or necessary consequence of, speaker’s knowledge at the time of
speaking. If that were so, such statements would be true or false depending
merely on the relation of compatibility or necessary consequence invoked for
the modalized proposition with regard to what the speaker’s knowledge
amounts to at the time of speaking. But that is not what epistemic modal
statements are. One cannot say, for example, that (7.4a) is false and hence
(7.4c) is true, merely because the speaker’s knowledge state, whatever it
amounts to, leaves no room for the possibility that the fire was caused by
witchcraft. (7.4a) may also be false, and (7.4c) true, because the speaker’s
knowledge state, in so far as it pertains to the cause of the fire, is factually
incorrect. There are thus two conditions that must be fulfilled for (7.4a) to be
true, one of compatibility with the speaker’s knowledge base and one of
factual correctness of that knowledge base. So perhaps (7.4a) should be read
as saying what is said in (7.5a), and (7.4b) as (7.5b):
(7.5) a. My relevant knowledge state K is correct and it is compatible with K
that the fire was caused by witchcraft.
b. My relevant knowledge state K is correct and it follows from K that
the fire was caused by witchcraft.
Interestingly, it is not evident that this blocks SSV in the embedded
S-terms. Consider (7.6a,b), with the corresponding paraphrases (7.7a,b):
(7.6) a. The morning star may be inhabited.
b. The morning star must be inhabited.
(7.7) a. My relevant knowledge state K is correct and it is compatible with K
that the morning star is inhabited.
b. My relevant knowledge state K is correct and it follows from K that
the morning star is inhabited.
Is the term morning star, in (7.7a,b), freely interchangeable salva veritate with
the coreferential term evening star? This is a difficult question.
Let us consider first the case that (7.7a) or (7.7b) is true. Then, on the
present analysis, speaker’s knowledge state K, in so far as it pertains to the
possible inhabitation of the morning star, is correct. But does this mean that
the speaker also knows that the terms morning star and evening star refer to
 The context-sensitivity of speech and language 205

the same object? If so, SSV is ensured and the operator may does not create an
intensional subdomain. But perhaps the speaker does not know that the
morning star and the evening star are the same planet and perhaps his
evidence shows that the evening star cannot be inhabited, though the morn-
ing star just may be, given the conditions under which the planet has been
observed. In that case, SSV is effectively blocked, which would make at
least the operator may intensional. But then again, an interlocutor might
say that if (7.6a) is true, then so must (7.8a) be, because morning star and
evening star are just different names for the same object. And the speaker
who produced (7.6a) will then have no choice but to agree. And analogously
for (7.6b) and (7.8b):
(7.8) a. The evening star may be inhabited.
b. The evening star must be inhabited.
The issue seems to turn on the meaning of the word relevant: what does one
take the ‘relevant knowledge state K’ to amount to? It doesn’t look as if this
question can be resolved here and now. We will, therefore, allow it to rest until
new insights arise. It may be added that the question of free SSV, even though
it triggered essential developments in semantics during the twentieth century,
is less important for an adequate understanding of natural language than it is
for an adequate philosophical notion of what ‘truth’ amounts to.
By contrast, if (7.7a) or (7.7b) is false, SSV appears to be freely applicable.
For in that case either K, in so far as it is relevant, is incorrect, in which case
substitution of the one term for the other does not affect the truth value, or K,
in so far as it is relevant, is correct but there is no compatibility or necessary
consequence, in which case SSV again fails to affect the truth value.
In actual fact, the widespread belief that epistemic modal subdomains are
intensional in the sense of disallowing SSV simply is a consequence of the
wildly unrealistic philosophical construct which reduces the extension of
S-terms embedded under modal predicates to sets of possible worlds. Since
the set of possible worlds in which the morning star is inhabited is different
from the set of possible worlds in which the evening star is inhabited, SSV
must, according to this analysis, be blocked. But this analysis is typically
generated by philosophical and formal a prioris. It lacks any psychological
plausibility, let alone any empirical support.5

5 The tangle was confounded by W. V. O. Quine (1953: 143–4), who, in an effort to show that SSV is
blocked in modal contexts, confused the value-assigning predicate bev with the predicate be of
identification (see Section 5.3.2 in Volume I). Quine wrongly regards a sentence like (i) as an
identity statement and not as a statement assigning a value to the parameter ‘the number of
planets’, which is what it is. Following up on this mistake, he shows that a replacement of the term
206 The Logic of Language

As regards agentive modals, in particular those of permission and obligation,

SSV seems to be fully warranted. As a matter of intuitive observation, (7.9a)
and (7.9b) are seen fully to allow for the substitution of Mount Everest for
Chomolungma and vice versa (one remembers from Section 3.1.1 of Volume I
that the two are different names for the same mountain). No-one can say that
the addressee has not fulfilled the obligation imposed by (7.9b) because (s)he
has climbed Mount Everest and not Chomolungma; and analogously for (7.9a):
(7.9) a. You may now climb Mount Everest.
b. You must now climb Chomolungma.
The agentive modal may may be seen as saying that the set of rules and norms
of behaviour actually in force is compatible with the event mentioned in the
embedded S-term to be realized. Analogously, agentive must may be seen as
saying that the existing set of rules and norms of behaviour requires the
realization of the event mentioned in the embedded S-term.6 Clearly, the
realization of an event is insensitive to the terms used to refer to the entities
mentioned, which means that SSV is freely allowed.
Apart from this factual, truth-conditional use, however, agentive modals are
also open to performative use. (7.9a,b), for example, can be read not only as
statements of fact about the existing set of rules of behaviour, but also, in a
performative sense, as speech acts whereby the speaker uses his or her authori-
ty to bring about a change in the existing set of rules of behaviour by either
allowing or ordering the addressee to carry out a specific act.7

nine in (ii) with the number of planets leads to the obvious falsity (iii), which is why he holds that the
modal necessity operator does not allow for free SSV and is, therefore, intensional:
(i) The number of planets is nine.
(ii) Nine is necessarily greater than seven.
(iii) The number of planets is necessarily greater than seven.
Needless to say, this argument, influential though it may have been, comes to nothing.
6 The L-propositional form of epistemic and agentive modal statements is taken to be identical.
In many languages, including most languages of Europe, the modal predicate forms part of the
Auxiliary System and takes an embedded subject-S-term. The modal predicate is lowered into the
subject-S-term, where it may end up as a finite verb form (as in English), or as a morphological
element (as, for example, in Turkish). See Seuren (1996: 79–84, 111–16, 159–60, 221–2) for extensive
discussions and analyses in different languages.
7 Dutch and Low-German allow for an agentive prepositional phrase with agentive modals, under
the preposition van (von). Sentences like (i) are normal standard Dutch:
(i) Hij moet van zijn leraar de sommen afmaken.
he must of his teacher the sums finish
His teacher has told him to finish his sums.
Dutch van was, until a few centuries ago, the standard preposition for passive agent phrases (standard
modern Dutch has door); German von still has that function. This suggests that, at least for these
languages, the agentive modals may represent underlying passive predicates approximately of the
form ‘it has been made possible/necessary (by A) for B to do C’. We will, however, not pursue this
issue here.
 The context-sensitivity of speech and language 207

7.2.2 The Principle of Maximal Unity

There is a principle at work in the overall construction of domains which
ensures the maximal unity of the entire commitment domain including its
subdomains. Let us speak of the PRINCIPLE OF MAXIMAL UNITY or PMU. PMU is
akin to the OSTA Principle discussed in Section 3.4.2 of Volume I, which
optimizes sense, truth, and actuality, both principles being of a strictly func-
tional nature. PMU ensures that increments are, by default, managed in such a
way that only minimal changes are brought about in the existing D. The
principle can be overridden, but such overridings of the default process always
involve a post hoc correction and hence a metalinguistic statement.
As far as can be seen at present, PMU manifests itself mainly in four
ways, which we call transdominial referential transparency, upward presup-
position projection, subdomain unification, and minimal D-change,
respectively.

7.2.2.1 Transdominial denotational transparency By transdominial denota-

tional transparency is meant the phenomenon that addresses occurring in any
domain or subdomain are retrievable from any other domain or subdomain.
(The term denote is used for the connection made between a definite
noun phrase in an uttered sentence and the corresponding discourse address.
The term refer is reserved for the connection made between a definite noun
phrase (or the corresponding address) and the actual or virtual object repre-
sented by the address in question. Reference is thus mediated by a discourse
address.)
The denotation function, which thus takes a definite noun phrase
and delivers it at a discourse address, is free to roam through the speaker/
listener’s entire D, as indeed through the speaker/listener’s entire knowledge
base, to establish denotation, and hence reference. An apparent exception, as
is explained in Sections 7.3.3, 8.2.2, and 10.8 seems to be the use of sentential
negation, which may seem to ban the incrementation of the negated proposi-
tion to the domain or subdomain that is operative at the time the negation is
used and thus to make impossible any denotation of an address thus created
(see Section 10.8 for further comment).
Disregarding negation for the time being, we see, for example, that, given
an address for the Eiffel Tower in the commitment domain at hand, this address
is immediately retrievable for the purposes of any subdomain. Consider
sentence (7.10):
208 The Logic of Language

(7.10) Ann fears that the Eiffel Tower has been hit.
It is not necessary for the subdomain under fear to be construed as requiring
the existential introduction of an address for the Eiffel Tower into the sub-
domain itself before (7.10) can be processed, as in (7.11), even though such an
interpretation is not excluded:
(7.11) Ann fears that there is a thing called ‘Eiffel Tower’ and that it has been hit.
The normal reading of (7.10) is such that the Eiffel Tower is taken to be a really
existing object, which can be given an address in the commitment domain.
Yet the actual existence of the Eiffel Tower is not entailed by (7.10), as can be
seen from (7.12), whose first conjunct, Ann is under the illusion that there is a
thing called ‘Eiffel Tower’, entails that there is no such thing as the Eiffel Tower.
If the second conjunct, she fears that it has been hit, entailed that the Eiffel
Tower exists, the conjunction as a whole would be incoherent or inconsistent,
but it is not: (7.12) is a fully coherent piece of discourse:
(7.12) Ann is under the illusion that there is a thing called ‘Eiffel Tower’ and
she fears that it has been hit.
Transdominial referential transparency likewise holds between subdo-
mains. Consider the following sequence of sentences:
(7.13) Roy is thought to have a sister. One hopes that she is more honest
than him.
In (7.13), she stands for the intensional object ‘Roy’s reputed sister’, who may
not exist at all. This object is represented in the intensional subdomain of what
people think. Yet it recurs in the intensional subdomain of what people hope.
This is made possible by the principle of transdominial referential transparency.
When an inconsistency arises between any domains or subdomains where
mutual consistency is required, this inconsistency is not due to transdominial
reference but to a conflict between actual and virtual being (see Section 10.8).
Thus, sentence (7.14) does allow for transdominial reference, but suffers from
inconsistency between the subdomains of knowledge and hope:
(7.14) !! I know Roy has no sister, but I hope that she is more honest than
him.
The formal system, or, if one wishes, the logic of transdominial consistency
has so far not been given any attention in the literature. No attempt is
made here to develop such a logic. Further research will have to bring greater
clarity.
 The context-sensitivity of speech and language 209

7.2.2.2 Upward presupposition projection Transdominial denotational trans-

parency by itself is, however, insufficient to explain the fact that (7.10) is
normally interpreted as implying the actual existence of the Eiffel Tower.
All it does is make it possible for (7.10) to be interpreted in such a way that
the term the Eiffel Tower refers to an actually existing object, represented
in the commitment domain. But that does not explain the preference for
such an interpretation. To account for that, we have to assume that sentence
(7.10), by default, induces the invited inference that the Eiffel Tower has
actual existence.
Unlike entailments, which cannot be overridden or neutralized, invited
inferences can, as is shown in (7.11) and (7.12). Thus, while (7.15) entails
that the Eiffel Tower actually exists, owing to the fact that the predicate
hit is extensional with regard to its object term (Section 3.5 in Volume I),
(7.10) merely invites the inference that it actually exists:
(7.15) The Eiffel Tower has been hit.
This weakening of entailment to invited inference is based on a default
mechanism which we call ‘upward presupposition projection’ (following the
introduction of the term presupposition projection in Langendoen and Savin
(1971) in the context of the projection of presuppositions from subdomains
into higher domains). Presuppositions have a tendency to percolate upwards
and they do so under a variety of conditions and in different guises.
The notion of upward presupposition projection, or just presupposition
projection, is defined as follows:
(UPWARD) PRESUPPOSITION PROJECTION:
(Upward) presupposition projection is the incrementation of
presuppositions generated within an embedded S-term Sn, or of
presuppositions or invited inferences that originated as presuppositions
and were projected into Sn, to the higher domain of the embedding
L-proposition, either as full presuppositions or as invited inferences.
The PROJECTION PROBLEM consists in defining the conditions under which
presuppositions or invited inferences of embedded S-terms do in fact
project—that is, are incremented to the immediately higher domain and
any higher domain over it prior to the incrementation of the whole sen-
tence—and in what form. This problem dominated the presuppositional
literature during the 1970s, at the expense, one must say, of more basic
research into the question of what presuppositions are and how they are
generated.
210 The Logic of Language

Mostly, when presuppositions project upwards, they are weakened to

invited inferences, as is demonstrated in the following sentences:
(7.16) a. Maggie hopes that her boyfriend has come back.
b. Maggie’s boyfriend has come back.
c. Maggie’s boyfriend has been away.
d. Maggie has a boyfriend.
Sentence (7.16b) expresses the L-propositional object term of (7.16a), (7.16c) is
a presupposition of (7.16b), and (7.16d) of (7.16c). The question is: are (7.16d)
and (7.16c) also presuppositions of (7.16a)? That is, does (7.16a) also, like
(7.16b), presuppose that Maggie has a boyfriend and that her boyfriend has
been away? The answer is that the presuppositions (7.16d) and (7.16c) do
make it to the superordinate domain of (7.16a), but only in the weakened
form of an invited inference. Full presuppositions form a subclass of entail-
ments, but (7.16a) does not entail that Maggie has a boyfriend and that the
young man has been away. It only invites that inference and that invited
inference can be overridden, as is shown by (7.17), which entails that Maggie
does not have a boyfriend, so that he cannot have been away:
(7.17) Maggie has deluded herself into believing that she has a boyfriend
and she hopes that he (her boyfriend) has come back.
Here, the presupposition (7.16d), carried by (7.16c), is restricted to the sub-
domain under the predicate believe.
In the terminology that was introduced in Karttunen (1973: 178) and which
was much en vogue during the 1970s, predicates like hope or believe are FILTERS in
that they let presuppositions through in the weakened form of default invited
inferences which are open to cancellation. By contrast, predicates introducing
or continuing a subdomain that need not be consistent, either internally or with
respect to the superordinate D, in particular all metalinguistic predicates,
including predicates of verbal reporting like say, ask, or mention, are PLUGS, in
that they block the upward projection of presuppositions. A final category of
predicates, in Karttunen’s taxonomy, is formed by the so-called HOLES, which let
presuppositions through as full presuppositions, no matter when and where.
Factive predicates are clear holes, in that they induce the presupposition that
their factive embedded S-term is true, which requires in turn the truth of the
presuppositions entailed by the embedded S-term in question.
The empirical question of the precise conditions for presupposition pro-
jection is discussed in detail in Section 10.5, where the hole-filter-plug termi-
nology is given up and where it is argued that PMU is indeed the main
 The context-sensitivity of speech and language 211

guiding principle, given the lexical meanings of the predicates concerned.

Here we only wish to point out that presuppositions of embedded clauses are
under pressure to project upwards, so that they have a maximally restrictive
effect on the overall total D in question and thus make it maximally informa-
tive. How and when this tendency is thwarted is a matter for discussion in
Section 10.5. The conclusion there is that the reasons for the thwarting of this
tendency are, on the whole, grounded in PMU, though there is a remainder of
cases where PMU appears to be restricted by linguistic form.
A case in point is natural language negation. It has been, and still is, a moot
question whether natural language negation (not in English) is a hole, a filter,
or a plug. According to standard propositional logic, it must be a plug, since
the standard logical negation blocks any entailment of its argument proposi-
tion, except necessary truths, which follow from any proposition. According
to accepted lore in pragmatics, it must be a filter turning presuppositions of
the embedded proposition into invited inferences, which can be overridden
on pragmatic grounds. For those authors who support this view, in particular
Horn (1985, 1989), pragmatics should also be able to account for metalinguis-
tic uses of negation, where the negation signals the speaker’s dissatisfaction
with the choice of words in a previous utterance.
In Seuren (1980, 1985, 1988, 2000) the position is defended that normal
unmarked, minimal negation is a hole, letting the presuppositions of its
argument-S through in full force. But next to minimal negation, a strongly
marked metalinguistic radical negation is posited, which cancels all presup-
positions and requires special accent to do so. This position is based on the
fact that the possible occurrence of both the minimal and the radical negation
is universally constrained by certain grammatical and lexical environments,
which are unrelated to any pragmatic factors (for a detailed discussion see
Section 10.4).
Moreover, the radical negation clearly has a metalinguistic character, but, as
argued in Seuren (1988), distinct from other metalinguistic uses of negation.
Prima facie at least, these facts appear to support the assumption of a true
ambiguity between minimal and radical NOT, the latter being obligatorily
metalinguistic and each being excluded in certain linguistic contexts and
uniquely possible in others. Yet, if we have to do with an ambiguity of
the negation operator, it is not an idiosyncratic ambiguity such as are
found all over the lexicon. On the contrary, since the semantic behaviour of
negation is most probably subject to systematic grammatical and lexical
restrictions that apply across the languages of the world, any ambiguity of
the negation operator should be the result of yet unknown universal seman-
tico-grammatical principles. Moreover, as indicated in note 12 of Chapter 3,
212 The Logic of Language

the metalinguistic character of the radical negation may well be a not fully
grammaticalized remnant of an original truly metalinguistic operator ‘Not
true (the utterance u)’.

7.2.2.3 Subdomain unification: transdominial consistency A further manifes-

tation of PMU consists in subdomain unification. By this is meant the fact that
(sub)domains maximally pool their resources under the condition of seman-
tic consistency and according to available world or situational knowledge.
Consider the following sentences:
(7.18) a. Paul may be at home and he may be having his breakfast.
b. Paul may be at home and he may be in hospital.
The first conjunct of (7.18a), Paul may be at home, introduces a subdomain of
possibility. The second conjunct, he may be having his breakfast, continues
that same subdomain, since, by default, one’s home is where one has one’s
breakfast. Typically, in cases of domain unification or continuation,
subsequent additions restrict earlier additions. Thus, in the possibility sub-
domain of (7.18a), having one’s breakfast restricts the being at home, in that
one can do a variety of things when one is at home: cook, read, watch TV,
entertain guests, or indeed have one’s breakfast.
By contrast, the second conjunct of (7.18b), he may be in hospital, does not
continue the subdomain of possibility introduced by the first conjunct. Here
we have two subdomains of possibility, one in which Paul is at home and one
in which he is in hospital. The reason is, obviously, that being in hospital is, in
all but highly exceptional cases, incompatible with being at home. Moreover,
being in hospital does not restrict being at home, as being in hospital is not
something one normally does at home. Consequently, (7.18b) results in two
distinct possibility subdomains, one for Paul’s being at home and one for his
being in hospital.
In cases of domain-splitting, as in (7.18b), one often finds the connective or
instead of the perhaps more regular and. This is no doubt due to the fact that
(7.18b) forces a choice between two subordinate domains, only one of which
can be true. This is remarkable because normally OR leaves a choice not
between two parallel operators over subdomains but between two parallel
domains, as in:
(7.19) Paul is at home or he is in hospital.
The use of or in cases like (7.18b) is semantically anomalous since the
juxtaposition of two possibilities calls for and:
 The context-sensitivity of speech and language 213

(7.20) Possible[Paul is at home] and Possible[Paul is in hospital]

If, as some say it is, this use of OR is natural for speakers of natural languages,
an adequate semantic analysis should show that this use of OR is a regular
consequence of the semantics of OR. This, however, is more than we can
achieve here.
A further consequence of domain splitting is the blocking of CONJUNCTION
REDUCTION. Whilst (7.18a) can be reduced to (7.21a), (7.18b) cannot, salva
veritate, be reduced to (7.21b):
(7.21) a. Paul may be at home and be having his breakfast.
b. !! Paul may be at home and in hospital.
This means that the syntactic process of CONJUNCTION REDUCTION applied
to subdomains must be taken to be conditional upon domain unification.
This may be taken as an argument against the thesis that grammatical
processes are modular—that is, operating on mere formal input and
algorithmically delivering a mere formal output, without any external inter-
ference (see Section 7.2.2 in Volume I). It follows that if both the modularity
thesis for grammars and the syntactic rule of CONJUNCTION REDUCTION
are to be upheld, the input structures to a grammar must, where
relevant, carry a formal mark indicating whether the second domain
is or is not unified with the first: domain split/unification must then be
encoded in the syntax. We must, unfortunately, leave this question unresolved
here.
Inverting the order of the conjuncts in (7.18a), as in (7.22a), leads to a
different interpretation. But in (7.18b), where domain unification is blocked,
the difference is at most stylistic:
(7.22) a. Paul may be having his breakfast and/or he may be at home.
b. Paul may be in hospital and/or he may be at home.
Normal world knowledge makes (7.22a) anomalous. Normally, or
by default, the second conjunct restricts the first and not vice versa.
Since, in (7.22a), the first conjunct restricts the second, an anomalous
interpretation imposes itself. As a result, (7.22a) forces a scenario in which
Paul’s having his breakfast does not take place at home but elsewhere, in
a hotel, for example. Such an interpretation blocks domain unification
and hence conjunction reduction.
214 The Logic of Language

7.2.2.4 Minimal D-change If we extend the PRINCIPLE OF MAXIMAL UNITY through

time, we get the PRINCIPLE OF MINIMAL CHANGE, which implies that listeners appeal
to all their cognitive resources in order to keep their current discourse
domain from being unnecessarily crowded. On the one hand, this is an applica-
tion of the general cognitive principle in virtue of which the mind strives
for maximal efficiency. It seems an inborn feature of the mind to keep natural
ontologies minimal and subject to functional classification. Babies quickly
learn to identify different occurrences of the same person or object as being
precisely that. They also quickly acquire the ability to identify different individ-
ual objects or persons as belonging to one type or class—Leibniz’s ‘identity
of indiscernibles’. This activity of creating maximal order in the mind with a
minimum of expenditure and effort is naturally extended to the construction
of discourse domains. On the other hand, however, it is also a matter of
the listener’s ‘aligning’ his or her current discourse domain with that of the
speaker, on the overall assumption that speakers likewise minimize their models
of the situation under discussion—that is, their discourse domains (Pickering
and Garrod 2004).
One manifestation of this principle was discussed in Section 7.2.2.3, where
the case was discussed, in connection with example (7.18), of two modal
subdomains being automatically united by default into one domain when
world knowledge supports such unification. More examples are provided by
the phenomena of transdominial referential transparency and of presupposi-
tion projection discussed in Sections 7.2.2.1 and 7.2.2.2, respectively.
A further illustration of this principle is the practice of especially journalists
and reporters to enrich anaphoric reference with new information by repla-
cing a simple pronoun with a full lexical NP, as in (7.23):
(7.23) Yesterday evening a Swiss banker was arrested at Heathrow Airport.
The fifty-year-old bachelor declared that he had come to Britain to
kidnap the queen.
Here the NP the fifty-year-old bachelor adds new information about the
banker who is said to have been arrested at Heathrow Airport, not about a
hitherto unmentioned individual introduced into the discourse domain by
means of the device of post hoc suppletion (accommodation). In other words,
post hoc suppletion is activated only as a last resort, as when I start a story in
something like the following way:
(7.24) The fifty-year-old bachelor was at the end of his tether. No matter
whom he approached in the street, they all shied away as if he were a
leper.
 The context-sensitivity of speech and language 215

Now there is nothing for the NP the fifty-year-old bachelor to link up with, so
that the listener or reader is left with no other option than to apply post hoc
suppletion of the presupposition carried by (7.24), namely that there appar-
ently was a fifty-year-old bachelor approaching people in the street. But no
such measure is needed for (7.23), because world knowledge easily allows for
an identification of the Swiss banker and the fifty-year-old bachelor men-
tioned in the news item. The principle of Minimal D-change now says that
because such an identification is possible, it is mandatory, unless blocked by
specific information provided in the discourse or in the situation at hand. The
principle of minimal D-change thus amounts to a restriction imposed on any
new incrementation process to minimize the number of subdomains and the
number of addresses for individual objects or sets of objects.

7.3 Conditions for text coherence

The main principle underlying discourse incrementation processes is the
closest possible approximation of the nature and the location of what is
meant to be the TARGET SITUATION, which is the situation the speaker aims at
describing. Or, in less ponderous terms, discourse incrementation is the
search for the target situation. This search must safeguard consistency, since
inconsistency of a discourse leaves no possible situation in which it can be
true. It must be informative in the sense that every new increment helps to
home in on the target situation, restricting the number of possible situations
in which D can be true (unless a new increment recapitulates what has been
achieved so far). And it must keep subdomains apart whenever that is
necessary. These three conditions are discussed, in that order, in the following
subsections.

7.3.1 Consistency
Any D must be internally consistent for the simple reason that an inconsistent
set of statements cannot be true for any situation at all. Speakers and listeners
have a profound awareness of this condition, as any discourse comes to a halt
when an inconsistency is detected.
Although it does not matter much whether a D in course is actually true or
not, it does matter whether it can be true. There is a CONSTRAINT OF POSSIBLE
TRUTH with regard to the propositional content of any commitment domain.
A commitment domain must, as a whole, be consistent, so that there is
something to the truth of which the speaker is committed, or whose truth
the speaker wants to be realized through the listener, who is requested,
216 The Logic of Language

allowed, or obliged to bring it about. (One remembers from the opening

sentences of Chapter 1 that it is because of the importance of textual consis-
tency that so much attention is paid to logic in the present book.)
A commitment domain is like a novel: no matter how strange or outlandish
the story, complete with the linguistic utterances of each character, it must be
at least consistent or else the novel will fail to interest readers and it might
even simply break down. The moment an inconsistency is detected by a hearer
in a commitment domain, it gives rise to puzzlement and possibly even to
incomprehensibility. Inconsistency makes it impossible for a commitment
domain to be true and thus makes it unfit for any truth commitment or
for any appeal to the listener to make it true, or for any rule-setting to be
followed properly. Even so, however, inconsistencies may occur in certain
kinds of subdomain. Obviously, a subdomain under the operator INCONSIS-
TENT is expected to contain inconsistent information. Or the inconsistency
may be attributed to a given speaker, who will then be understood as making
inconsistent utterances. But the overall commitment domain, the overarching
tale, appeal, or rule-setting must be consistent, or rather, must not contain
any detected inconsistencies.
Interestingly, the consistency condition of commitment domains may be
seen as resulting from the principle of natural set theory PNST–5, presented in
Section 3.2.2 and stating that, at a basic-natural cognitive level, the intersec-
tion of two sets A and B must be M(utual)-PARTIAL intersection (A O O B, or:
A \ B 6¼v
6¼v A 6¼v B). Given that this intersection function translates into
basic-natural propositional logic as the functor AND, it follows that the
conjunction of two L-propositions P, Q must leave a nonnull valuation
space, so that there is at least one possible situation in which P AND Q is
true (modulo key). Since AND is also the prototypical incrementation function
of utterances, PNST–5 formally ensures logico-semantic consistency of com-
mitment domains as well as of any subdomain that requires consistency.
In actual fact, the condition of consistency should be carried further and
extended to the speech-act force of utterances. Just as propositional content is
subject to truth conditions, the speech-act character of utterances, as specified
in Section 4.3 of Volume I, is subject to liability conditions, which have to do
more with interpersonal relations than with truth. (There may even be a legal
aspect to speech-act commitments or appeals, as interpersonal relations very
much involve giving and taking, as well as pleasing and offending.) It follows
that the notion of consistency applies not only to propositional content but
also, at a higher level, to speech-act commitments, appeals, or rule-settings,
which have their own consistency criteria.
 The context-sensitivity of speech and language 217

Something was said about this latter aspect in Section 4.2 of Volume I, with
regard to a few examples from the literature showing a clash between speaker’s
overtly expressed commitments or appeals on the one hand and his or her
loyalty to the commitment or appeal on the other. The examples discussed
were Austin’s quote from Euripides’ Hippolytus: ‘my tongue took an oath, but
my mind remained unsworn’, Hamblin’s ‘I am obliged to order you to do D,
and I hereby do so; but my private advice to you is not to’, as well as the
famous Moore paradox ‘the cat is on the mat, but I don’t believe it’. And the
conclusion was that the existing literature is largely unclear with regard to
speech-act consistency. In general, the speech-act force of utterances has been
neglected in linguistic, semantic, philosophical, and pragmatic studies, where
all attention has been focused on propositional content. To remedy that
situation and bring greater clarity to the issues concerned is, however, a
research programme of such magnitude that we cannot possibly hope to do
much about it within the confines of the present study.

7.3.2 Informativity
A new increment to any D must be informative in the sense that it narrows
down the set U of possible situations in the direction of the target situation,
or else it must recapitulate what has been achieved so far and draw an
inference that has not so far been made explicit, as is typically the case in
sentences starting with therefore. In general, we speak of the PRINCIPLE OF
INFORMATIVITY or PI. The underlying rationale of PI seems to be a basic need
in linguistic interaction to home in on the target situation, which is to be
described up to the degree of precision needed for the purpose at hand. Given
our system of valuation-space modelling, this requirement can be cast into a
formal mould.
Formally speaking (and without taking into account the class of inference-
drawing increments), the principle of informativity (PI) is defined as follows:
PRINCIPLE OF INFORMATIVITY (PI)
Each successive increment in a discourse domain D must constitute a
further restriction of /D/, provided the restricted /D/ 6¼
.
Noncompliance with PI results in either an erratic or an incomprehensible
discourse (the latter in particular when /D/ is reduced to
).
Consider, in an abstract formal sense, the universe of all possible situations
U to be the disjunction (in the standard sense) of all possible propositions,
and hence the union of the VSs of all infinitely many possible L-propositions:
/P1/ [ /P2/ [ /P3/ [ . . . .As a general principle we say that before any discourse
218 The Logic of Language

has started /D/ ¼ U. Although this makes the initial /D/ (¼ U) a nondescript
entity, and hence unfit for natural cognition, it helps to see how a discourse
can start with the implicit question, anticipated by the speaker: ‘Exactly what
proposition do you, speaker, want to increment for the purpose of the present
interaction?’ Any first L-proposition P presented for incrementation can be
regarded as an answer to that question. P then restricts the initial /D/ (¼ U)
to /P/, in the sense that the target situation is an element in the new, more
restricted, /D/.
We will speak of any given (old) D as Do, and of the resulting new D as Dn.
It is not required that, for an increment of P, /P/ be included in /Do/, which
would make P entail Do. But it is not forbidden either. When /Do/ contains
/Dan is human/ and /P/ is /Dan is a student/, or, presuppositionally, when
/Do/ contains /Dan was married before/ and /P/ is /Dan is divorced/, then P
entails Do and the discourse is still coherent and informative. In most cases,
however, P and Do will be semantically independent with regard to each
other, so that /Dn/ will consist of the nonnull intersection of /P/ with /Do/.
This process of incrementing P to Do is shown graphically in Figure 7.1.
Anticipating the analysis of presuppositional phenomena in Chapter 10, we
posit that, for normal or default negation, /NOT(P)/ is the complement of /P/
within /Do/, since an incrementation of NOT(P) is an answer to the (implicit
or explicit) question ‘P?’ within the interpretative limits of Do. In Figure 7.1,
/Do/ is marked by horizontal lines, and /Dn/ by vertical lines. Since by
definition /Dn/ is included in /Do/, /Dn/ is, in fact, marked by both horizon-
tal and vertical lines. This makes the standard incrementation procedure an
instance of AND-conjunction.

/not(P) /

/Do/ :

/Dn/ :

/P/
/Dn/ in /Do/ :
U

/Dn/ = /P/ ∩ /Do/

FIGURE 7.1 /Do/ narrowed down to /Dn/ after incrementation of P
 The context-sensitivity of speech and language 219

7.3.3 Subdomain hierarchies: subsidiary subdomains

Although there is, as yet, no fully elaborated logic encoding the conditions of
transdominial consistency, it may be observed that there are hierarchies of
subdomain-creating predicates (see Seuren 1985: 417–22). Consider the fol-
lowing sentences:
(7.25) a. Dan knows that Philip has a sister who is a divorcee.
b. Dan believes that Philip’s divorced sister has a daughter.
c. Dan hopes that Philip’s sister’s daughter is still a child.
When these sentences are ordered serially, with normal anaphoric provisions
and conjunctive concatenation, the following coherent text comes about:
(7.26) Dan knows that Philip has a sister who is a divorcee. He believes that
she has a daughter and he hopes that this daughter is still a child.
From a presuppositional point of view, these sentences may be taken to have
the following structure (where YX stands for ‘Y presupposing X’, X, Y, . . .
ranging over L-propositions):
(7.27) a. Dan knows that P.
b. Dan believes that QP.
c. Dan hopes that RQ.
The point is that the text of (7.26) becomes incoherent when the main
predicates are exchanged, as, for example, in:
(7.28) ! Dan hopes that Philip has a sister who is a divorcee. He believes that
she has a daughter and he knows that this daughter is still a child.
The first sentence in (7.28) is not incoherent. But the second sentence strikes
one as both an incoherent continuation of the first and incoherent in itself.
(7.28) can be improved by the insertion of conditionals, as in:
(7.29) Dan hopes that Philip has a sister who is a divorcee. He believes that,
if Philip has a divorced sister, she has a daughter and he knows that, if
she has a daughter, this daughter is still a child.
This sentence is, though a little odd for pragmatic reasons, fully coherent in
itself. To remove the pragmatic oddity one must invent a proper context, no
matter how far-fetched. For example, one may assume that Dan has been told
about Philip’s sister and her daughter from a marriage that has been dissolved.
He doesn’t know if this is true, but he hopes that it is at least true that Philip
has a divorced sister, because he wants to marry into Philip’s family, which is
220 The Logic of Language

quite rich. Since this sister has always been mentioned in connection with a
daughter, Dan believes that, if Philip has a sister, this sister has a daughter.
Dan knows that Philip’s parents are in their mid-forties, which makes it
practically impossible for them to have a grown-up granddaughter. Therefore,
Dan knows that, if Philip’s sister has a daughter, this daughter must still be a
child. Perhaps a starting point for a suspense thriller!
Subdomain hierarchies do not infringe upon the principle of transdominial
denotational transparency discussed in Section 7.2.2.1, as one might be in-
clined to think. This is because that principle is about interpretational acces-
sibility of addresses in the commitment domain or any subdomains, whereas
subdomain hierarchies are about coherence. Although a sequence of sentences
like (7.28) may be deemed incoherent on account of its transgressing the
boundaries set by subdomain hierarchies, there is no problem as regards the
denotations of the various definite noun phrases: she in the second sentence
clearly denotes the sister-address set up in the first sentence, even though a
subdomain of belief does not fit well into a subdomain of hope.
An attempt at setting up a few subdomain hierarchies was made in Seuren
(1985: 417–22). One such scale, intended to apply to descending epistemic
strength, is shown in Figure 7.2. Such scales imply that the discourse may
proceed from the column marked 0 to any column marked by a higher
number, while predicates in the same column are not subject to a sequenti-
ality constraint.
The predicates of column 1 are all factive predicates, which means that the
truth of the object that-clause is presupposed. This makes it possible, though
perhaps only marginally so, to formulate a coherent sequence of conjuncts as
exemplified in (7.30):
(7.30) Molly regrets that her brother is in jail but (it is true that) the man is
dangerous.
This is not a counterexample to the epistemic-strength scale of Figure 7.2,
because factive predicates induce a presupposition of truth for their that-
clauses, so that whatever is said in the that-clause is retrievable from the
higher truth domain.

0 → 1 → 2 → 3
true know believe hope
realize think wish
regret … want
… try
…
FIGURE 7.2 Intensional scale for epistemic strength
 The context-sensitivity of speech and language 221

0 → 1 → 2 → 3
true must probable may
necessary likely possible
… … …
FIGURE 7.3 Intensional scale for epistemic inference

A further intensional scale may be set up for predicates of epistemic

modality, discussed in Section 7.2.1.3. Such a scale is presented in Figure 7.3
and more such scales can be constructed.
A theory of subdomain hierarchies is important not only for a proper
understanding of how subdomains are interrelated, and thus for an adequate
insight into the criteria for textual coherence and consistency, but also for
another, equally important and more specific, reason, to do with presupposi-
tion projection (see Section 10.5). Suppose the sentence (7.25c) occurs in a
context where it is given that Philip has no sister. Now the presuppositions
‘Philip has a sister’ and ‘Philip’s sister has a daughter’ of the embedded clause
that Philip’s sister’s daughter is still a child cannot be projected into the
commitment domain D and must remain restricted to a subdomain. The
question is: which subdomain? It cannot be the subdomain of what Dan
hopes, because then the interpretation would be ‘Dan hopes that Philip has a
sister and that this sister has a daughter and that this daughter is still a child’,
which is not the way (7.25c) is interpreted. On the contrary, Dan may well
hope that Philip does not have a sister and that, if Philip does have a sister,
that this sister does not have a daughter. But, if these hopes are dashed by the
way things are, he may hope that at least this daughter is still a child. The
question is, therefore, if these presuppositions are prevented from rising into
a higher subdomain and also from staying within the subdomain of what Dan
hopes, where do they go? The answer must be that they go into the sub-
domain of what Dan believes to be the case, even if Dan’s beliefs have not been
the topic of discussion in the current D.
When a subdomain requires a different subdomain for the storage of
nonprojected presuppositions, we call it a SUBSIDIARY SUBDOMAIN. Subsidiary
subdomains have to refer to a RECIPIENT SUBDOMAIN for the storage of non-
projected presuppositions of embedded clauses. In all cases, whether the
presuppositions of embedded clauses are projected or not, they are ENTAILED
in the associated recipient subdomain. Thus, in all cases where it is true to say
that Sandra hopes that her son has returned, it follows by way of semantic
entailment that Sandra believes that she has a son and that this son has been
away. They are actually projected as default inferences as long as the current
222 The Logic of Language

discourse domain is compatible with Sandra having a son and this son having
been away.
Subdomains created by the predicate hope are thus subsidiary subdomains
and they have to fall back on a recipient subdomain of what the subject-term
referent of hope believes, to cater for nonprojected presuppositions.
To the best of my knowledge, this particular phenomenon has not so far
been discussed in the literature. If this is so, it is surprising because the
phenomenon is of considerable importance for a proper understanding of
the processes of utterance interpretation. It seems, incidentally, that the class
of predicates that create subsidiary subdomains is identical with the class of
emotive factive predicates that do not allow for substitition salva veritate of
topic–comment modulation—a phenomenon hitherto unknown and dis-
cussed in Sections 3.2 and 6.2.3.2 of Volume I. We revert to this important
topic in Section 10.5.2 and in Chapter 11.
Right now, we leave these and related questions open and merely point out
that these aspects of discourse incrementation have been neglected in the
literature, despite their obvious relevance.

7.4 Open parameters in lexical meaning

In the present section, we take some time to look at the lexical aspects of
context-sensitivity, keeping in mind what was said at the outset of the present
chapter, namely that sentences underlying utterances contain systematic
lexical and grammatical devices that refer the listener in specific ways to
available encyclopedic and contextual knowledge.
It is often thought or implicitly assumed that predicate meanings, as
codified in their satisfaction conditions, are lexically fixed in such a way
that they automatically produce truth or falsity when applied to appropriate
reference objects. This assumption, however, though not unreasonable
in itself, is unwarranted because it fails to take into account the important
fact that in many, perhaps most, cases, the satisfaction conditions imply an
appeal to nonlinguistic world or situational knowledge, not codified in
the language system acquired during childhood. The truth and falsity of
assertive utterances are thus not just the product of linguistic compositional
computation, but are co-determined by nonlinguistic knowledge, either of
a general encyclopedic or of a context-bound, situational nature. Formal
semanticists should be worried by this, since such appeals to general or
 The context-sensitivity of speech and language 223

situational knowledge prove that they have been wrong in insisting that
the truth conditions of sentences are compositionally derivable from the
satisfaction conditions of the predicates occurring in them and the structural
positions they occupy. Let us consider a few examples, some of which have
been discussed earlier, especially in Sections 9.6 and 9.7 of Volume I, though
in a slightly different context.
Gradable adjectives provide a prime example. These are adjectives that
allow for grade modifiers such as rather, very, or a little. They also allow
for comparatives and superlatives.8 Typical examples are expensive, old,
large, wide, smart, popular, rich, safe, fast, and many others, as opposed to,
for example, closed, empty, rectangular, frontal, dead, postprandial, which
are, in principle, nongradable. The applicability of gradable adjectives
(when used absolutely, that is, not in a construction that implies a form
of comparison) depends on, usually socially recognized, standards, such
as standards of cost, age, size, monetary value, etc., for the objects denoted
by their subject terms. The description of such standards is not part of
the description of the language concerned but of (socially shared) knowledge.
Thus, when I say:9
(7.31) He is an old man.
the truth of what I say depends on socially acknowledged norms for calling a
man old. How the norm is selected is still largely unknown. It is unclear, for
example, what norm is to be applied in a case like Apes are intelligent. Are they
meant to be intelligent with regard to humans, or compared to other animals?
There is a large amount of literature dealing with gradable adjectives, and
many issues have so far remained unresolved. But it is clear, across the board,
that no solution will be found unless cognitive factors are fully integrated into
the semantics of gradable adjectives. The point here is that the criteria for
truth or falsehood are not given in the linguistic description of the meanings
of these adjectives but in (socially shared) knowledge. Such adjectives thus
need an open parameter (sometimes also called ‘free variable’) in their

8 Interestingly, some adjectives are nongradable in literal use but become gradable when used
metaphorically. For example, the adjective self-contained is nongradable when applied to an apartment
but gradable when applied to a person’s character. Likewise for square, round, full, angular, pedestrian,
human, savage, and many other adjectives, which are nongradable when used literally (especially in a
technical context), but gradable when used nonliterally or less strictly.
9 One recalls from Section 9.3 in Volume I the proposal that a sentence like (7.31) is to be analysed as
‘He mans oldly’ (or ‘He olds his being a man’), in analogy with He is a good teacher and He teaches well.
224 The Logic of Language

semantic description referring the speaker/hearer to the relevant elements in

general knowledge.10
Further examples of cognitive dependency are possession predicates, like
English have, lack, with, without, and whatever lexical specification is needed
for genitives, datives, and possessive pronouns. These clearly require general
encyclopedic knowledge, and often also contextual knowledge, for their
proper interpretation and thus for the assignment of truth values. Consider
the following examples (repeated from (7.17a,b) in Section 9.6.3 of Volume I,
where they are discussed in a different context):
(7.32) a. This hotel room has a bathroom.
b. This student has a supervisor.
For (7.32a) to be true it is necessary that there be one unique bathroom
directly connected with the room in question, whose use is reserved for the
occupants of that room. When the room has a note stuck to the door saying
that its bathroom is at the end of the corridor to the right, while the same
bathroom serves all the other rooms in the corridor, (7.32a) is false—not just
misleading but false, as any judge presiding over a court case brought by a
disgruntled hotel guest will agree. But for (7.32b) to be true no such unique-
ness relation is required, as one supervisor may, and usually does, have many
students to look after. The predicate have does not determine the precise
nature of the relation between the referents of the subject and object terms.
What is needed for full interpretation is, for (7.32a), knowledge of the world of
hotels and, for (7.32b), knowledge of the world of universities or similar
institutions of higher education. The same goes for the parallel sentences:
(7.33) a. This is a hotel room with a bathroom.
b. This is a student with a supervisor.
Possession predicates, therefore, must be specified in the lexicon as involv-
ing an appeal to what is normally the case, or has been specified to be the case,
regarding their term referents. Linguistically, they merely express a known
relation of appurtenance between the kind of object referred to in subject
position and the kind of object referred to in object position (see Janssen
1976). The semantic description (satisfaction condition) of have and other

10 Typically, in the case of gradable adjectives, the boundary between truth and falsehood forms
what is often called a ‘grey area’, in which gradable statements are neither clearly true nor clearly false,
letting in a ‘fuzzy’ logic with transitional truth values. Gradability thus goes hand in hand with fuzzy
truth values.
 The context-sensitivity of speech and language 225

possessive predicates is thus taken to contain a parameter for ‘what is known’,

making the interpretation of this predicate in each token occurrence truth-
conditionally dependent on situational or world knowledge.
Possessive pronouns appear to allow for a larger range of ‘known relations
of appurtenance’ than, for example, have. Sentence (7.34) may be uttered by a
gardener who has no proprietary rights to ‘his’ flower beds other than his
duty to tend them:
(7.34) Please don’t mess up my flower beds.
To say ‘I have flower beds’ would be inappropriate in the circumstances.
Many such examples can be given. Consider the predicate flat said of a
road, a tyre, a mountain, a face, or the world. There is an overall element
‘spread out, preferably horizontally, without too much in the way of protru-
sions or elevations’, but that in itself is insufficient to determine what ‘being
flat’ amounts to in these cases. The full meaning comes across only if it is
known what roads, tyres, mountains, faces, and the world are normally
thought to be like. Dictionaries, even the best ones, limit themselves to giving
examples, hoping that the user will get the hint.
Another example is the predicate fond of, as in the following sentences
(copied from Section 9.6.3 in Volume I):
(7.35) a. John is fond of his dog.
b. John is fond of cherries.
c. John is fond of mice.
Clearly, there are different, incompatible, kinds of fondness, depending on
how socially shared world knowledge tells us to practise it. Sentence (7.35c) is
ambiguous in this respect, as John’s fondness may be of the kind expressed in
(7.35a) or of the kind expressed in (7.35b). The common element in the status
assigned to the object-term referents is something like ‘being the object of
one’s affection or of one’s pleasure’, but such a condition is insufficient to
determine full interpretation. It is no doubt for that reason that a sentence like
(7.36) strikes one as somehow infelicitous:
(7.36) John is fond of his dog and of cherries.
Cognitive dependency is an essential aspect in the description of
predicate meanings. The fact that many predicate meanings contain a param-
eter referring to an available nonlinguistic, language-independent knowledge
base means that neither utterance-token interpretation nor sentence-
type meaning are compositional in the accepted sense of being derivable by
226 The Logic of Language

(model-theoretic) computation from the linguistic elements alone. As

regards utterance-token interpretation, this is already widely accepted,
owing to the deconstructivist forces at work in pragmatics. The noncompo-
sitionality of sentence-type meaning, defined at the level of language descrip-
tion, is now likewise beginning to be accepted by theorists of natural
language. This type-level noncompositionality, however, does not force
the conclusion that the specification of the satisfaction conditions of predi-
cates is not truth-conditional, only that standards embodied in socially
accepted knowledge and information provided by context may become part
of the truth conditions of sentences in which the predicate occurs.
As was said in Section 9.6 of Volume I, the term polysemy is often used for
phenomena such as those presented above. At the same time, however, it is
widely recognized that this is, in fact, little more than a label used to give the
problem a name. The problem itself lies in the psychology of concepts. One
may assume that there are socially shared concepts like ‘possession’, ‘flatness’,
‘fondness’, but it is not known in what terms such concepts are to be defined.
In a general sense, Fodor (1975, 1998) is probably right in insisting that lexical
meanings are direct reflexes of concepts that have their abode in cognition but
outside language. Yet both the nonlinguistic concepts and the corresponding
lexical meanings must be defined one way or another. And the question is
whether this can be done in terms of the famous necessary and sufficient
conditions. If so, then, according to Fodor, the language in which such
conditions are to be formulated cannot be any form of natural human
language but must be a ‘language of thought’, which is categorially different
from any natural language and whose terms and combinatorial properties will
have to be established as a result of psychological theorizing. We are still very
much in the dark as regards such questions. It is clear, in any case, that
phenomena like those shown in (7.31)–(7.35) pose a serious threat to any
attempt at setting up a model-theoretic theory of lexical meaning, such as
Dowty (1979). The neglect of the cognitive factor quickly becomes fatal in
lexical semantics.
As is explained in Section 9.7 of Volume I, context-bound or
situational knowledge plays a role in the interpretation of predicates
that involve a VIEWPOINT or PERSPECTIVE, such as the pair come and go, or
predicates like to the right/left of, in front of, behind. Consider, for example,
the sentences:
(7.37) a. John looked around. The box was to his left.
b. I looked around. The box was to John’s left.
 The context-sensitivity of speech and language 227

In (7.37a), the viewpoint is taken by John, and the box must be to his left as he
sees it. In (7.37b), this is not necessary: the box must be to John’s left as
I (speaker) see it, while for John it may be anywhere around him. That
predicates like left, right, in front of, behind, and so on, and also pairs of the
type come and go, are sensitive to viewpoint is no doubt due to the fact that
they involve ego-related localizations. (One thinks of the quasi-problem of
why mirrors invert left and right but not up and down.)
Moreover, FUNCTION is known to be a determining factor in lexical mean-
ings, in particular in the domain of artefacts. What makes a coat a coat is not
its size, shape, material, or what not, but its intended function—a criterion to
be satisfied not by the object itself but by the use to which it can be put
according to whatever, possibly very creative, cognitive criteria.
A further source of cognitive dependency lies in a semantic component of
evaluation. As was already pointed out by the Greek Sophists, the truth of a
sentence like There is a pleasant breeze depends primarily on what humans
perceive as ‘pleasant’, under varying conditions, and only in a secondary sense
on the physical properties of the object so predicated. This point has great
philosophical importance, as philosophers argue about the question of
whether predicates like good and just (the central concepts in ethics), and
beautiful (central in aesthetics), are to be defined in terms of world properties
alone, or in terms that co-involve personal evaluation. As is shown in Section
3.4 of Volume I, this question applies likewise to the predicate true.
One further source of vagueness lies in the fact that satisfaction conditions
of predicates often centre around prototypical ‘ideals’ (Rosch 1975). Some
objects are closer to the intended prototype than others. A sparrow, for
example, is closer to the prototype of ‘bird’ than an ostrich or a penguin.
The notion of prototypicality plays a role in lexical semantics, in that pre-
conditions (the class of satisfaction conditions that give rise to presupposi-
tions) often select prototypical circumstances. The preconditions of the
German predicate kahl (bald, bare), for example, include the condition that
the subject-term referent is prototypically a human being or his/her head,
prototypically covered with hair on the top of the head. The prototypicality
appears from the fact that subjects, when asked what they think of first on
hearing the word kahl, almost invariably answer that they think of a human
head. Yet the subject-term referent may also, nonprototypically, be another
kind of object, normally covered with other growth, such as feathers or leaves,
or with decorative artifacts.
The update condition (giving rise to standard entailments) is simply that
the growth or decoration which is normally there, is not there. This allows for
phrases like der kahle Kopf (the bald head), der kahle Mann (the bald man),
228 The Logic of Language

der kahle Baum (the bare tree), der kahle Vogel (the bald bird), die kahle
Landschaft (the bare landscape), die kahle Wand (the bare wall). One notes,
incidentally, that English has two predicates to cover this semantic field: bald
and bare. Yet, when asked what the English equivalent is of German kahl, most
people will reply bald, not bare. This is because of the prototype of kahl, which
centres on hair on the human head, so that the more marginal cases slide out
of focus. Prototypicality is thus an autonomous cognitive parameter that
plays a role in the satisfaction conditions of many predicates.
This is as far as we can go in the present context. But even this cursory
discussion shows that lexicographers are not all that wrong when they view
theoretical semantics with a fair amount of scepticism.
 8

Discourse incrementation

8.1 The incrementation procedure

So far, we have spoken only informally about what we call addresses—that is,
cognitive representations of either actually existing or somehow thought-up
entities or sets of entities. We shall now try to be a little more precise.
Addresses have their place in the general knowledge base of any individual
human being (and also of many nonhuman animals). A large proportion of
these addresses is intentionally related to really existing entities, but some are
lodged in some subdomain of the general knowledge base capable of repre-
senting virtual entities. For example, you and I have an address for Sherlock
Holmes, but we both know that Sherlock Holmes never existed but was
thought up by the great nineteenth-century writer of detective stories Arthur
Conan Doyle. The address for Sherlock Holmes is, therefore, lodged in a
separate general-knowledge subdomain labelled ‘In the stories by Arthur
Conan Doyle’. And that subdomain is itself related to an actually existing
set of books and stories, whose existence is partially grounded in actual books,
consisting of actual paper (or in electronic form), and partially also in what
we have called social reality in Section 2.1.2 of Volume I.
The general knowledge base may thus be seen as containing at least two
kinds of unit: (a) ENTITY ADDRESSES known to represent actually existing entities
or sets of entities and (b) DOMAIN ADDRESSES each again encapsulating entity
and/or domain addresses, and so on. Entity addresses that are thus
encapsulated in a domain address are, for the most part, intensional ad-
dresses. An intensional address is an entity representation, but the
corresponding entity does not necessarily have to have actual existence. It
may be a virtual entity in the sense explained in Chapters 2 and 5 of Volume I.
It often happens that the title-holder of the knowledge base in question is
uncertain as to the ontological status of an entity. In such cases, the address is
lodged in a subdomain of uncertainty or doubt. If this were not so, hypotheses
and theories would not be possible.
If this is acceptable in principle, we may posit that each commitment
domain D is itself a, possibly ephemeral, intensional subdomain in the domain
230 The Logic of Language

of general knowledge, indexed for the linguistic exchange at hand. We further

posit that, at each mention of an entity by the speaker, the listener activates the
corresponding address in his or her general knowledge base.
It follows that even in cases where reference is made in direct speech to an
actually existing entity, that reference is intrinsically intensional, as it is
processed through the intensional commitment domain at hand. It is for
that reason that we spoke of the intensionalization of extensions in the last
sentence of Chapter 2 in Volume I. All semantics is intensional. Necessarily,
the commitment domain D is itself a mental construct, just like the mental
constructs that are considered intensional (and hence block substitutivity).
The commitments laid down in the commitment domain often apply to
imagined situations whose relation to what is considered the real world is
codified by the governing predicate. Thus, when I say that John thinks that the
moon is made of green cheese, I have made myself responsible for the truth of
the statement that John thinks that the moon is made of green cheese, not for
the truth of the statement that the moon is made of green cheese. This latter
putative fact is stored in a subdomain indexed for what John thinks. There are
thus intensional subdomains for what the speaker, or a person mentioned in
the discourse, hopes, realizes, has forgotten, and so on, or for what he or she
considers possible, probable, fortunate, and so on.
The speaker may, of course, wish to introduce a new entity into the
discourse, knowing or anticipating that the listener does not yet have a
corresponding entity address in his or her general knowledge base. The proper
thing for the speaker to do, in such cases, is to use an existentially quantified
statement. In practice, however, speakers, and certainly fiction writers, are not
so accommodating and throw the listener (reader) so to speak in medias res by
using a definite description for the entity referred to, leaving it to the listener
(reader) to infer that an address for such an entity has to be set up. This is
what we have called post hoc suppletion or accommodation, exemplified in
example (7.24) of Section 7.2.2.4.

8.1.1 Singular entity addresses and address closure

It is useful to have a formal notation for the processes mentioned, not only
because formal notations add to the clarity of the theory but also because they
often reveal complications that had hitherto remained unobserved. In this
section, some elementary symbolism is introduced for the representation of
the process of incrementation to a given discourse domain D, leaving out the
more complex cases involving plurality and quantification, and also incre-
mentation to subdomains of D. These more complex procedures, along with a
more complex notation, are proposed in the following two sections.
 Discourse incrementation 231

Let us assume that a speaker wishes to introduce a new address by uttering

the deceptively simple sentence There was a cat. The example is, of course, a
little stilted, and it hardly represents a normal speech event, but that is how
formalisms have to be built up. We use the following notation for the new
address just introduced:
(8.1) d–1 [a j Cat(a)]
(8.1) is to be read as follows. ‘d–1’ is the address label facilitating address
retrieval (the number 1 is arbitrary). The part between the square brackets
reads as ‘There is an a such that a is a cat’, or ‘There is an a such that a 2
[[Cat]]’. Tense is disregarded for the time being; tense domains are not
discussed here. The symbol ‘a’ thus acts as a unary (Russellian) existential
quantifier requiring that [[Cat]] be nonnull. If the sentence had been There
was a cat that ran away or Some cat ran away, the address would be as in (8.2):
(8.2) d–1 [a j Cat(a), Run away(a)]
requiring for truth that the intersection of [[Cat]] and [[Run away]] M-par-
tially intersect (or, for BNPC, that [[Run away]] may be properly included in
[[Cat]]; see Section 3.4.1). If spoken in direct discourse, (8.1) and (8.2) have a
truth value which depends on the state of affairs (situation) referred to on the
one hand and the satisfaction condition defined for the existential quantifier
a and for the predicates Cat and Run away on the other.
Now let the sentence There was a cat be followed by The cat ran away. The
definite description the cat needs an existing address to ‘land at’, or denote.
The question is: what happens to the address d–1 when the information
carried by the sentence The cat ran away is incremented to it? One might
think that the information ‘the cat ran away’ is simply added to the address
(8.1), as in (8.2), which is read as ‘There was a cat that ran away’, with both
occurrences of the variable a bound by the quantifier a. But this cannot be
correct, even though it (or its counterpart in predicate logic) is widely
assumed to be correct by logicians and formal semanticists.
A quick way to see why this is so is the following. Compare the following
two sentences (see also (8.25a,b) below):
(8.3) a. John has a girlfriend and this girlfriend is Australian.1
b. John has a girlfriend who is Australian/an Australian girlfriend.

1
(8.3a) may be rephrased as John has one girlfriend,—who is Australian, with a nonrestrictive
relative clause. Nonrestrictive relative clauses have the force of a subsequent conjunct, as appears from
the fact that they may be followed by a polar tag with speech-act force, as in: John has one girlfriend,—
who is Australian, isn’t she?
232 The Logic of Language

These two sentences differ radically in what they say. Sentence (8.3a) implies
that John has a girlfriend, who is Australian. (8.3b), by contrast, is compatible
with a sequel like and many others who are not. (8.3a) is a conjunction of two
sentences; (8.3b) is not. It is clear that this girlfriend in (8.3a) cannot represent
a variable bound by the existential quantifier, represented by the indefinite
article a, if only because one can say I believe that John has a girlfriend and
I hope that this girlfriend is Australian, where binding of this girlfriend under
one single quantifier is impossible. The principle of generality dictates that the
same must then apply to this girlfriend in (8.3a).
Nothing changes for (8.3a) when the definite description this girlfriend is
replaced with the anaphoric pronoun she which, again, cannot represent a
variable bound by the indefinite article (quantifier) a. But what does it
represent? We say that it represents an instance of PRIMARY ANAPHORA—that
is, anaphora where the antecedent is not itself a referring expression but an
address that has been set up existentially just before and is ‘visited’ by a
subsequent anaphoric expression for the first time.
The problem for (8.3a), with she for this girlfriend, is—and this is the basic
problem of primary anaphora, further discussed in Chapter 9—how to
account for the status of the anaphoric pronoun she as a referring pronoun
not bound by the existential quantifier represented by a, but somehow
recovering its antecedent from the preceding existentially quantified sentence.
In anticipation of the treatment of this problem in Chapter 9, we now
introduce the technique of ADDRESS CLOSURE, which ensures that a pronominal
reference used in primary anaphora is not bound by the preceding quantifier
but is represented as a definite term in D.
First look at (8.4), where address closure has taken place. We say that an
address that has not been closed is an OPEN ADDRESS. Thus, d–1 in (8.1) is an
open address, where a is an existential quantifier. An open address is closed
the moment a definite term retrieves its denotation from a preceding existen-
tially quantified sentence and thus becomes a referring term. Address closure
is represented as ‘//’. The incrementation of The cat ran away, uttered right
after There was a cat, thus looks like (8.4):
(8.4) d–1 [a j Cat(a) // Ran away(the a(Cat(a)))]
Address closure is needed to establish reference. It changes a from being a
quantifier—that is, a function from sets to truth values—to being a reference
function selecting an object (the referent) from a set of objects. Reference is a
function that takes a set and delivers a specific element in that set. The
reference function has been a source of discomfort to modern semantics
 Discourse incrementation 233

because it cannot be defined within the confines of standard compositional

model theory: there is no way of selecting one specific individual from a plural
set by mathematical means alone. For it to work, an external input from
cognition is needed, in particular from knowledge of and about the context
and situation of the utterance and the (restricted) universe of objects OBJR
that the speech is about. This is a further indication that compositionality is
not a hard-and-fast principle of the interpretation of natural language utter-
ances.
The effect of closure is thus that a no longer functions as an existential
quantifier, as in (8.1) or (8.2), but as a definite determiner over the proposi-
tional function Cat(a), or, in other words, as a function from the predicate
extension [[Cat]] to a particular object in [[Cat]], its reference or æ-value.
Before address closure, a is an existential quantifier, but, after address closure,
a has become a definite determiner and the whole address d–1 has become a
cognitive representation of an entity.
As in the case of (8.3a), it is natural to replace the phrase the cat with the
anaphoric pronoun it, which is then an instance of primary anaphora. The
incrementation of There was a cat. It ran away will then be something like:
d-1 [a j Cat(a) // Run away (1)]
The symbol ‘1’, acting as the subject term to the predicate Run away and
corresponding with the address label d–1, is not a bound variable but a
resumptive pronominal expression resuming ‘the a such that a was a cat’.2
The address d–1 can thus be read as ‘The a such that a was a cat,—it ran away’,
where ‘a j Cat(a)’ represents the referential part of the address and where the
propositional content is located in the part that follows address closure.
It is by means of the device of address closure that the difference between
(8.3a) and (8.3b) is accounted for. (8.3a) is assigned a discourse representation
with address closure: ‘There is an a such that a is a girlfriend of John’s, and
this a is Australian’ or ‘John has a girlfriend,—who is Australian’. But (8.3b) is
assigned one without address closure, read as ‘There is an a such that a is a
girlfriend of John’s and such that a is Australian’ or ‘John has a girlfriend who
is Australian’. It is, if you wish, the difference between a nonrestrictive and a
restrictive relative clause.
Now we extend the mechanism so as to show how addresses are generated
from the semantic analyses (SAs) of sentences—that is, from L-propositions.
Consider again the sentence There was a cat, as in (8.5a), with its (matrix-S)

2
One might think of using the notation ‘Run away(a)’ for the part after closure. This, however,
would make it impossible to make cross-references, as in (8.8c,d) below.
234 The Logic of Language

SA (8.5b), represented as an L-propositional (SA) tree structure in (8.5c), and

resulting in the D-address (8.5d):
(8.5) a. There was a cat.
b. ANx[Obj(x), Cat(x)]
c.
So

Pred NP1 NP2

ANx
^x S1 ^x S2

Pred NP Pred NP
Obj x Cat x

d. d–1 [a j Cat(a)]
In (8.5b,c), AN is the existential quantifier, also known as ∃, but AN is
reserved for a single entity, as opposed to SOME, which is reserved for the
plural existential. AN is treated as a binary higher-order predicate over pairs of
sets, with, in this case, the subject term Obj(x) (see Section 5.6 in Volume I)
and the object term Cat(x). The index x in ANx binds the variables x in Obj(x)
and Cat(x). The operator AN requires for truth that there be at least one
element common to the sets denoted by the two terms.
There is a problem here regarding actual and virtual being. Following
the argument developed in Section 5.6 of Volume I, (8.5a) should not entail
the actual existence of a cat, since the predicate Obj(x) is intensional,
which would allow it to intensionalize the set of cats in virtue of Virtual
Object Attraction discussed in Section 5.4 of Volume I. That (8.5a) is felt to
entail actual existence might be attributed to the fact that a sentence like (8.5a)
is normally used under an operator of place, as in There were cats in the cellar,
where the operator in the cellar ensures an extensional interpretation. Since,
in natural speech, the verification domain is normally restricted to a given
situation known and accepted by speaker and listener to be actual and not
just virtual, the default interpretation of (8.5a) would then implicitly impose
that situation as a local restrictor turning the sentence into an extensional
statement. Yet intuitively, (8.5a) does entail actual existence. For example,
when asked Are there unicorns? I can reply in truth No, only in stories. We
revert to this question below.
The SA-structure (8.5c) is the tree-structure counterpart of (8.5b). NP1
is the subject term, NP2 the object term, and the caret symbol ‘^’ is a
set-denoting operator: ‘the set of things x such that . . . ’. The grammatical
 Discourse incrementation 235

process transforming (8.5c) into (8.5a) is not at issue here. (Roughly,

NP2 is incorporated into PRED[AN], forming the complex predicate PRED[AN -^x
[Cat(x)]], which is then lowered into the position of the subject term x
of Obj, which is lexicalized as Be there. For details see Seuren (1996: 300–9).)
The present concern is the incrementation procedure IP turning (8.5c) into
the D-address (8.5d). IP scans the highest SA-predicate first. In the case at
hand, the highest predicate is AN, which is an instruction to create a new
singular address (a first-order address over individuals). An ADDRESS LABEL d–1
is set up (the number 1 is arbitrary), identifying the address for later reference.
The contents of the address is given between square brackets. Here a is the
ADDRESS HEAD, representing AN and binding the variables. It stands for the
Russellian first-order existential quantifier—a function from sets of indivi-
duals to truth values, typed ((e,t),t).3
For (existential) quantifiers, the object term NP2, denoting the restrictor
set, is incremented first for at least two reasons. The first reason is that this
procedure guarantees the proper scope order, as is demonstrated in (8.10)
below. The second reason is that it is by the restrictor set that the address is
identified and selected for closure as a result of a following definite NP (such
as the cat).
This gives the new address d–1 [a j Cat(a)]. Normally, the subject term is
then added, but not in the case of Obj(x), it being axiomatically understood
that there are (actual or virtual) things. Therefore, only NP2 is incremented
after the upright bar indicating the scope of a. (One notes the analogy of d–1
with the Russellian formula ∃x[Cat(x)].) The address d–1 has a truth value
and is read as ‘there is/was a cat’. Existentially quantified addresses are open
addresses.
Now we turn for a moment to the question of existential import. In
standard logic, the existential quantifier posits actual existence of the common
element. But that will not do for language, given sentences like (8.6a) which
do not entail that the cat in question actually existed:

3
Typing of terms and predicates is a commonly used device in formal semantics, introduced by the
Polish logician Ajdukiewicz during the 1930s. It is based on the typing of entities as e (entity) and truth
values as t (the symbols are due to Montague). It enables one to follow a compositional function
calculus from entities to truth values, as done in categorial grammar, where the finally resulting value
must be typed t. A set of entities, denoted by what is known as a first-order predicate, is typed (e,t),
that is, a function from individual entities to truth values. A Russellian quantifier is a unary second-
order predicate, typed as ((e,t),t), that is, a function from sets of individual entities to truth values. The
existential quantifier, for example, takes a set of individuals and assigns it the value TRUE just in case the
set is nonnull and otherwise the value FALSE.
236 The Logic of Language

(8.6) a. A cat was worshipped there.

b. ANx[Be worshipped(x), #Cat(x)]
c. d–2 [a j #Cat(a), Be worshipped(a)]
(8.7) a. A child laughed.
b. ANx [Laugh(x), Child(x)]
c. d–3 [a j Child(a), Laugh(a)]
For the semantics of language it is stipulated that when the matrix term of the
existential quantifier is an intensional predicate denoting a cognitive process and
capable of yielding truth for both actual and thought-up entities (Frege’s
‘thought predicates’), while the restrictor term is extensional, yielding truth
only for actually existing objects, the extensional term is intensionalized, so that
it applies to thought-up entities as well. This is the rule of VIRTUAL OBJECT
ATTRACTION (VOA) presented in Section 5.4 of Volume I. Since in (8.7b) both
terms are extensional by nature, as only really existing individuals can truthfully
be said to laugh or to be a child, (8.7a) is rendered as (8.7b). But in (8.6a) Be
worshipped is an intensional predicate as it may yield truth also for fictitious
objects. Therefore, Cat(x) is intensionalized to #Cat(x), where ‘#’ indicates that
the set denoted by #Cat(x) may also contain virtual or thought-up cats as a
result of VOA. This allows true existential quantification over virtual objects.
Obj(x) in (8.5) is taken to be intensional by nature denoting the axiomati-
cally given nonnull set of actual and virtual objects that ‘are there’. But since
Obj(x) is not a thought predicate, it does not intensionalize the other term
under AN but is itself automatically extensionalized when the other term is
extensional, since an extensional predicate F(x) denotes a purely extensional
set [[F(x)]], which makes the intersection of [[F(x)]] and [[Obj(x)]] again purely
extensional. This ensures an entailment of actual existence for extensional
terms, but also the absence of such an entailment for intensional terms, as in
There was an imaginary cat or There are nonexistent unicorns.
Now back to address closure. As we have seen, an open address is closed
when denoted in a subsequent clause by a definite term. Address closure,
symbolized by a double oblique stroke, changes the address head from being a
function to truth values to being a reference function selecting an object (the
referent) from a set of objects. Reference functions over individuals are
typed ((e,t),e) (taking a set and delivering an individual), and are thus
type-reducing. It has been said a few times already that this function is not
compositional, not even given a model.
Now let D contain the open address d–1 as in (8.5d) above and also an open
address d–2 [a j Mouse(a)]. Then a sentence like (8.8a), with the SA (8.8b),
results in the two parallel increments (8.8c) and (8.8d):
 Discourse incrementation 237

(8.8) a. The cat caught the mouse.

b.
So

Pred NP1 NP2

Catch
Det S1 Det S2
the x the x
Pred NP Pred NP
Cat x Mouse x

c. d–1 [a j Cat(a) // Catch(1,2)]

d. d–2 [a j Mouse(a) // Catch(1,2)]
NP1 in (8.8b) reads ‘the x such that x is a cat’, and analogously for NP2. In (8.8c)
and (8.8d), the head a has been retyped from ((e,t),t) to ((e,t),e); the proposi-
tional function preceding closure—Cat(a) or Mouse(a)—denotes the input set
typed (e,t). Catch(1, 2) is the proposition saying that the individual selected by
the reference function a in d–1—the cat—caught the individual selected in d–
2—the mouse. The incrementation procedure IP first scans the predicate of So.
Catch being a binary lexical verb (and not a quantifier), IP is put to work on the
definite NP1 first, to be followed by the definite NP2.
The DENOTATION PROCEDURE d for definite NPs is as follows:
For any NPi under a definite-NP operator, say, the:
(a) the takes the predicate of the S under NPi and selects the matching
address d–n. There must, in principle, be only one such address in D (but
see Section 9.5.1 on primary anaphora and the reference-assignment
procedure).
(b) d–n is closed (if still open), and the SA-tree is added to the closed
address, with the number n of d–n for the NPi-constituent.
Thus, d(NP1) in (8.8b) selects d–1 [a j Cat(a)], and NP1 is replaced with 1;
d–1 is closed and the SA-tree, with 1 in place, is added to the now closed d–1:

d–1 [a Cat(a) // So ]
Pred 1 NP2
Catch
Det S2
the x
Pred NP
Mouse x

The procedure is repeated for NP2 in d–1, yielding the two parallel
increments:
238 The Logic of Language

d–1 [a Cat(a) // So ]
Pred 1 2
Catch

d–2 [a Mouse(a) // So ]
Pred 1 2
Catch

For practical reasons trees are written as bracketed strings, giving (8.8c) and
(8.8d), respectively.
A sentence like (8.9a), with SA (8.9b) is incremented as follows, with D
containing d–1 [a j Cat(a)]:
(8.9) a. The cat caught a mouse.
b.
So

Pred NP1 NP2

ANx
^x S1 ^x S2

Pred NP Pred NP
Catch NP3 x Mouse x
Det S3
the y
Pred NP
Cat y

c. d–2 [a j Mouse(a), Catch(1,a)]

d. d–1 [a j Cat(a) // [b j Mouse(b), Catch(1,b)]]
The new d–2, created in virtue of PRED[AN], is fitted out with S2 and S1, in that
order (with a for x):

d–2 [ a S2 , S1 ]
Pred NP Pred NP
Mouse a Catch NP3 a

Det S3
the y
Pred NP
Cat y

d–1 in (8.9d) contains a subordinate open address. An open address can be

stored under another address d–n provided it contains either the variable
 Discourse incrementation 239

bound by d–n (for open addresses) or the definite term n (for addresses after
closure).
Double existential quantification is treated analogously. (8.10a), with SA
(8.10b), yields the open address (8.10c):
(8.10) a. A cat caught a mouse.
b.
So

Pred NP1 NP2

AN x
^x S1 ^x S2

Pred NP3 NP4 Pred NP

AN y Cat x
^y S3 ^y S4

Pred NP NP Pred NP
Catch x y Mouse y

c. d–1 [a j Cat(a), [b j Mouse(b), Catch(a,b)]]

IP causes d–1 to be set up in such a way that the cat-address takes scope
over the mouse-address: ‘there is a cat a such that there is a mouse b such
that a caught b’. Since, however, it is possible to refer subsequently to the
mouse caught by the cat, as when one says The mouse escaped, an
independent open address for the mouse in question is also required. To
that end, an address of the form (8.11) is set up in virtue of a process of
inferential bridging:
(8.11) d–2 [a j Mouse(a), [b j Cat(b), Catch(b,a)]]
Here the mouse-address takes scope over the cat-address, but the difference
is irrelevant, as scope differences do not matter for two successive existential
quantifiers. The discourse may now continue with definite terms like the
cat that caught a mouse, closing d–1, or the mouse that was caught by a cat,
closing d–2.
The difference is not irrelevant, however, in more complex cases, such as:
(8.12) John claims that he owns a Ferrari, but I have never seen it.
Here, it stands for ‘the Ferrari John claims he owns’ and thus needs an address
for that Ferrari to land at and close if necessary. It would appear that such an
address, if still open, must look like
d–3 [a j Ferrari(a), Claim(John, Own(John,a))]
240 The Logic of Language

This, however, requires for truth an actual specific Ferrari, claimed by John to
be owned by him. But the Ferrari in question may well be, and probably is,
a virtual Ferrari because John may well be bluffing about his racing monster.
So there we are: the Ferrari claimed by John to be his property may well
be a virtual vehicle but to refer to it it looks as if we need an open address
that asserts its actual existence.
The only solution available, given the machinery as developed so far, is to
apply the intensionalization operator #, as in (8.7) above, and establish an
address like (8.13) for ‘There is a Ferrari John claims he owns’:
(8.13) d–3 [a j #Ferrari(a), Claim(John, Own(John,a))]
This correctly speaks of a specific Ferrari, namely the one John claims he owns,
but, owing to the fact that the existential quantifier no longer entails actual
existence by itself, this specific Ferrari need not actually exist. The assignment
of the intensionalization operator # to the predicate [[Ferrari(a)]] is driven by
the fact that the open address originates in the intensional context created by
the predicate Claim, in whose scope the open address for the Ferrari was set
up. The still open address d–3 can now be selected by the phrase the Ferrari
John claims he owns to land at and close.

8.1.2 Plurality and quantification

8.1.2.1 Plurality and existential quantification Now we turn to plurality. The

semantics of plurality is among the most difficult topics in the study of
natural language and, perhaps for that reason, among the least studied,
despite plurality being a basic category in the grammars of languages.
There is a certain amount of daring and incisive literature on plurality,
especially from the point of view of quantification, in the tradition of
formal model-theoretic semantics, but even though this literature has clearly
advanced the frontiers of our knowledge, the new ground it has conquered
is far from fully explored. What I can offer in this respect is, therefore,
tentative, fragmentary, and suggestive, even more so than in most other
parts of this book.
One thing is clear. Since in most natural languages plurality starts with
cardinality 2, plurality, in these languages, requires the notion of PLURAL POWER
SET (Ppl), which differs from what is standardly called the ‘power set’ operator,
symbolized as P. The standard power set of a set X, P(X), is the set of all
subsets of X, including the null set
and all sets consisting of just one
element, the so-called singletons. The plural power set of a set X differs in
 Discourse incrementation 241

that, for any set X, Ppl(X) is P(X) minus
and all singletons.4 We might also
say that Ppl(X) is the set of all natural subsets of X plus X itself, following the
definition of natural set given in Chapter 3.5 The very fact that plurality as a
linguistic category requires the notion of plural power set as just defined, as
opposed to the standard notion of power set, demonstrates, if not the validity,
certainly the reasonableness of the natural set theory hypothesis put forward in
Chapter 3.
To express this distinction formally, we can usefully employ the type-
raising distributive operator ‘::’, defined over predicates, for the language of
SAs and discourse addresses. Let [[P(x)]] denote, as before, the extension
of the predicate P(x)—the set of individuals x such that x satisfies P—then
[[::P(x)]] is defined as follows (x ranges over sets of individuals):
 
½½:: PðxÞ¼Def Ppl ½½PðxÞ

The extension of ::P(x) is thus the set of sets of at least two individuals x
such that each x satisfies P. The expression ::Happy(the children) reads as
‘the set of children in question is an element in Ppl([[Happy(x)]]), the plural
power set of Happy’. In other words, the sentence The children are happy is
true just in case the set of children in question is a natural subset of the total
set of those individuals that are happy. This requires that more than one child
is happy, because singletons are excluded from plural power sets. When P is
transitive and both of its terms are definite and plural, :: distributes indis-
criminately over the subject and the object term referents.
An open plural address is normally of the form (8.14c), representing (8.14a)
with SA (8.14b):
(8.14) a. There were (some) cats.
b.
So

Pred NP1 NP2

SOME
x̄ S2
^x¯ S1 ^x¯

Pred NP Pred NP
:: Obj x :: Cat x

 j ::Cat(ā)]
c. d–4 [a
4
For languages, such as classical Arabic and Ancient Greek, with a morphological category ‘dual’, in
addition to ‘singular’ and ‘plural’, special provisions must be made for sentences quantifying over two
elements. It is not clear, at this stage of the enquiry, whether an underlying numeral two will suffice for
the purpose.
5
Standardly, if X has cardinality n, P(X) has cardinality 2n. For plural power sets, as one will easily
figure out for oneself, if X has cardinality n, Ppl(X) has cardinality 2n–(nþ1).
242 The Logic of Language

Assuming ABPC to be the logic in charge (but SMPC will do as well in this
case), the plural existential quantifier SOME yields truth just in case there is a
nonnull intersection of at least one plural set of individuals of the two term
extensions concerned, which now are sets of sets of individuals. SOME is again
an instruction to set up a new address of the right type. In (8.14c), a 
represents plural SOME and binds the variable. (8.14c) thus requires that
there be at least one set of at least two actually existing cats.
The distributive operator :: makes it possible to account for a sentence
like (8.15a), rendered by the SA (8.15b) and the corresponding open address
(8.15c):6
(8.15) a. There were cats (that were) running away.
b.
S

Pred
SOME x̄ NP1 NP2

^x¯ S1 ^x¯
S2 S3
Pred NP
:: Obj x¯ Pred NP Pred NP
:: Cat x¯ :: Run away x̄

 j ::Cat(ā), ::Run away(ā)]

c. d–5 [a
When the predicate is second order by nature and thus typed ((e,t),t), such
as the predicates disperse, congregate, disband, or numeral predicates such as
three, or nominal predicates such as team or platoon, (army unit consisting of
men), the distributive operator :: is not needed. Thus, a sentence like There is
a platoon results in the singular address (8.16a), with the second-order
predicate Platoon over the plural variable a . The two forms can be com-
bined, as in (8.16b,c) representing, respectively, Some cats dispersed and There
were three cats:
 j Platoon(ā)]
(8.16) a. d–6 [a

b. d–7 [a j ::Cat(ā), Disperse(ā)]
 j ::Cat(ā), 3(ā)]
c. d–8 [a

6
Sentence (8.15a) is meant to be taken in the purely existential way, and not as Some of the cats were
running away, which is perhaps better treated in a format analogous to that of All (of) the cats were
running away, as in (8.30) below. Perhaps all quantifiers should be open to a double treatment, one in
which they do and one in which they do not involve a definite restrictor term.
 Discourse incrementation 243

Mathematically, an address can be of any order, yet, apart from exceptional

cases, language stops at second-order predicates requiring third-order ad-
dresses for their plurals, as in (8.17), which reads There are platoons. Any
higher-order nouns are treated as second-order.
(8.17) j ::Platoon(a)]
d–9 [a
Plurals are difficult, and only some of the problems can be dealt with here.
One such problem is the distinction between distributive and collective read-
ings. In the collective reading, The men carried the bags reads as saying that the
men as a group carried the bags as a group (while, say, the women carried the
pots), leaving it open whether there were subgroups of men carrying sub-
groups of bags. The distributive reading says that each of the men carried one
or more bags and each of the bags was carried by one or more men. The
linguistic expression thus underdetermines the actual state of affairs.
Collective readings are incremented analogously to singular increments.
Thus, given an open plural cat-address, such as d–4 in (8.14c), the addition
They ran away is incremented either as in (8.18a), where the cats are said to
have run away as a group, or as (8.18b), where the cats are said to have run
away individually. Closure has turned a  into a second-order determiner or
reference function (((e,t),t),(e,t)), selecting a set from a set of sets:
(8.18)  j ::Cat(ā) // Run away(10)]
a. d–10 [a
 j ::Cat(ā) // ::Run away(10)]
b. d–10 [a
To compute the truth value of Run away(10) in (8.18a), it is necessary to
type-raise the predicate Run away from (e,t) to ((e,t),t), so that it can process
an input of type (e,t), that is, a set of objects. The type-raising is triggered by
the fact that the address head is a not a. Type-raising of a predicate P implies
that, despite the transition from individuals to groups, the satisfaction con-
ditions of P remain unchanged.
This condition cannot be fulfilled by all predicates. Most nominal predi-
cates, such as cat, dog, tree, house, being reserved for individual predication,
disallow type-raising, because a group of cats cannot itself again be a cat, and
likewise for the predicates dog, tree, house and whatever object-naming pre-
dicates one has in mind. These predicates must be marked in the lexicon as
blocking type-raising, as opposed to, for example, run away or be happy or
carry, because a group can run away as well as an individual can, and likewise
for being happy and carry.
 j ::Cat(ā)]
(8.18a) is thus read intuitively as ‘the set of cats referred to by [a
is a set of at least two individuals running away as a group’. The distributive
244 The Logic of Language

reading of They ran away, incremented as in (8.18b), lets the cats in question
run away individually.
Typical collective readings are found in sentences like:
(8.19) a. The mice have been at the cheese.
b. The Americans were the first to land on the moon.
In their common reading, these do not imply that all the mice have been at
the cheese, or that all Americans were the first to land on the moon, as they
are about the mice, or the Americans, as a group. (Embarrassingly, these
sentences are true even if there was one single mouse at the cheese or
one single American on the moon—a fact that our formalism is as yet unable
to account for.)
Now consider the at least three-way ambiguous (8.20a) with the SAs
(8.20b) (in two versions) and (8.20c):
(8.20) a. The men carried a bag.
b.
So

Pred NP1 NP2

AN x
^x S1 ^x S2

Pred NP3 NP Pred NP

(::) Carry x Bag x
Det_ S3
the y
Pred NP
_
:: Man y

c.
So

Pred NP1

::S1 Det_ S2
the x
Pred NP2 NP3 Pred NP
_
ANy :: Man x
^y S3 ^y S4

Pred NP
_ NP Pred NP
:: Carry x y Bag y

d. d–11 [bj Bag(b), (::)Carry(12,b)]

j ::Man(ā) // [b Bag(b), (::)Carry(12,b)]]
e. d–12 [a

f. d–12 [aj ::Man(ā) // ::lx [b j Bag(b), ::Carry(x,b)](12)]
 Discourse incrementation 245

The SA (8.20b) has two guises, one with and one without the distributive
operator :: over the predicate Carry. With ::, (8.20b) says that there was a bag
that the men carried individually, so that the same bag was carried as many
times as there were men. Without ::, we have the group reading, saying that
the men combined forces to carry one single bag. IP produces d–11 in (8.20d)
as the result of (8.20b), with or without the distributive operator :: over
Carry. In either reading there is just one single bag. Supposing D already
contains the open address
 j :: Man(a)]
d–12[a
(‘there were men’), this address is now closed, analogously to d–1 in (8.11d),
resulting in (8.20e), again with or without the distributive operator :: over
Carry.
(8.20c), however, does not speak of one single bag but says that the set of
men in question is one of those sets of individuals such that each individual
had a bag to carry, so that there were at most as many bags as there were
individuals, and perhaps less, if two or more of the men carried a single bag.
This latter reading is incremented as (8.20f): the set of men referred to by d–12
was such that for each man there was a bag carried by him (alone or with one
or more others).
The main predicate of (8.20c) is the propositional function (¼predicate)
S1. S1 is a tree-structure version of what is known in logic as a LAMBDA
PREDICATE. The lambda operator º creates predicates, enabling one to incorpo-
rate quantificational and other operators into a predicate. In this case, the
lambda predicate denotes the set of those sets of at least two individuals who
have a bag to carry, individually or collectively. A variable is needed to ensure
that this lambda predicate is a propositional function rather than an open
address with a truth value. This variable is of a different register from those
used so far in the address notation. For that reason we revert to the end of the
alphabet and use x, here type-raised to x because the carriers are groups of at
least two individuals. The lambda predicate is incorporated as such into the
address notation.
With this lambda predicate, the sentence says that the set of men in
question is one of those sets of at least two individuals who have a bag to
carry, individually or collectively. The collective or group reading need not be
represented, as it is already given in (8.20b) without :: over Carry, which may
be seen as an instance of lambda reduction. Therefore, (8.20c) only gives the
distributive reading, which is not captured by (8.20b). This reading requires
two occurrences of the distributive operator ::, one for the predicate ::ºx̄ [b j
246 The Logic of Language

Bag(b), ::Carry(x̄,b)] and one for the embedded predicate ::Carry(x̄,b). It

is needed for the former so as to get as many bags as there are carriers; it is
needed for the latter to ensure that the bags can be distributed over individual
carriers. (The first occurrence of :: in (8.20f), just before º, ensures that (8.20f)
is not merely a lambda-extracted version of (8.20d) with ::).
How does IP work for (8.20) in its various versions? Let D contain an
open address d–12 [a  j ::Man(ā)]. Then for (8.20b) IP creates a new open
address d–13 [b j Bag(b), (::)Carry(12,b)], saying that there is/was a bag
which the men of d–12 carried collectively (without ::) or individually
(with ::). d–12 is now closed, analogously to d–1 in (8.11d), representing the
two readings of (8.20b). This gives (8.20d) in its two versions.
As regards (8.20c), its group reading selects the open address d–12 [a j
::Man(ā)], closes it and adds the lambda predicate without the two occur-
rences of ::. Lambda reduction reduces the result to d–12 in (8.20e). The
distributive reading of (8.20c) again selects the open address d–12 [a  j ::Man
(ā)], but places the distributive operator :: both before the main (lambda-)
predicate and before the embedded predicate Carry, as explained above.
This reading requires the setting up, by inferential bridging, of an open
plural bag-address creating room for subsequent reference to the bags
distributed over the men carrying them. The formal elaboration of this
form of inferential bridging requires an open plural address saying something
like ‘There is a plural set of bags such that each bag belonging to it was carried
by one (or more) of the men referred to by d–12 of (8.20e)’. This address can
then be closed to allow reference to ‘the bags that the men carried’ in the
context of interpretation (8.20c) of the sentence The men carried a bag—that
is, given (8.20f). The technical and notational elaboration of such an address
is something I happily leave to others. For the moment we will let such
technicalities rest.
We now turn to the quantifiers MANY and FEW in sentences of the type
(8.22a)—that is, without any implication that the cats were numerous within
a given group, as in Many of the cats were asleep. Sentences that contain
existential quantification of the sort ‘some of the X’ or ‘few/many of the X’
are perhaps better treated in a format analogous to that of (8.30) (see note 6).
That there is a clear semantic difference is shown by examples like (8.21a,b).
(It is worth noting that this difference has not been accounted for in any
semantic theory in existence.)
(8.21) a. Six elephants were travelling with the circus.
b. Six of the elephants were travelling with the circus.
 Discourse incrementation 247

In the strictly existential sense, the quantifiers MANY and FEW are treated as
variants of the neutral existential quantifier SOME, regarded as a binary
second-order predicate over sets. Thus, a sentence like (8.22a), with the SA
(8.22b), comes out as (8.22c) (leaving aside, of course, the question of what
gradable MANY and FEW imply in any given context):
(8.22) a. Many cats were asleep.
b.
S

Pred NP1 NP2

MANY x¯
^x¯ S1 S2
^x¯
Pred NP Pred NP
:: Asleep x¯ :: Cat x¯

 j ::Cat(ā), Asleep(ā)]
c. d–14 [MANY a
Like SOME, the quantifying predicates MANY and FEW are higher-order by
nature, requiring terms denoting sets of sets. (8.22b) says, in effect, there is a
nonnull intersection between the set of plural cat sets and the set of plural
object sets of beings that were asleep, while the intersection of the set of cats
and the set of beings that are asleep has a high cardinality.
Formally, we define the semantics of the quantifiers MANY and FEW in the
following way (cf. (2.14b) in Section 2.3.5.2):
(8.23) For all sets X and Y:
a. [[MANY]] ¼ { <Y,X> j Ppl(Y) \ Ppl(X) 6¼
, jY \ X j is high }
(the extension of the predicate MANY is the set of all pairs of sets
Y, X, such that the intersection of Ppl(Y) and Ppl(X) is nonnull
and the cardinality of the intersection of Y and X is high)
b. [[FEW]] ¼ { <Y,X> j Ppl(Y) \ Ppl(X) 6¼
, jY \ X j is low }
(the extension of the predicate FEW is the set of all pairs of sets Y,
X, such that the intersection of Ppl(Y) and Ppl(X) is nonnull and
the cardinality of the intersection of Y and X is low)
One notes that this definition makes a sentence like Few cats were asleep false
when there was only one sleeping cat, because in such a case Ppl([[Asleep]]) \
Ppl([[Cat]]) ¼
, singletons being excluded from plural power sets. If one
finds that unsatisfactory, one may reduce the satisfaction conditions for
MANY and FEW to simply ‘jY \ X j is high’ and ‘jY \ X j is low’, respectively,
248 The Logic of Language

leaving out the condition Ppl(Y) \ Ppl(X) 6¼
. It then follows automatically

that Y \ X 6¼
.
One is reminded of what is said in Section 3.2 of Volume I regarding
example (3.12), repeated here as (8.24a), and in Section 9.3 of Volume I
regarding example (9.5), repeated here as (8.24b):
(8.24) a. Coffee does not grow in Africa.
b. John does not get up at five in the morning.
c. John did not catch many fish.
The point is that the negation applies specifically to the predicate of
the highest S-term under the negation operator, as it denies the assignment
of the property expressed by that predicate to its argument term(s). The
more deeply embedded S-terms thus remain unaffected. The same is
found in (8.24c). Just as, in a normal interpretation, (8.24a) does not deny
that coffee grows or that there is coffee but that it is in Africa that
coffee grows, and just as (8.24b) does not deny that John gets up but that
it is at five in the morning that he does, in the same way (8.24c) does not
deny that John caught fish but, rather, that the fish he caught were plentiful.
This confirms the analysis presented in (8.22b), where MANY is the highest
predicate.
For operators like MANY or FEW, the closure operation is important for
empirical reasons, as appears, for example, from (8.25a,b), whose obvious
semantic difference is unaccounted for in most semantic theories:
(8.25) a. Nob had few students who failed.
b. Nob had few students, and they failed.
The difference corresponds with the closure operation. Let D already
contain an address d–15 [a j ‘Nob’(a)], standing for ‘There is an a
called “Nob”’. Both (8.25a) and (8.25b) set up a new address d–17 for
the students Nob had. For (8.25a) d–16 is left open, but for (8.25b) d–16 is
closed:
 j ::Student(ā), ::Have(15,ā), ::Fail(ā)]
(8.26) a. d–16 [FEW a
 j ::Student(ā), ::Have(15,ā) // ::Fail(16)]
b. d–16 [FEW a
Now consider (8.27), which has both a group and a distributive reading.
The existential predicates FEW and MANY say that the intersecting set required
by the existential quantifier is small or large, respectively:
 Discourse incrementation 249

(8.27) a. Few cats caught many mice.

b.
So

Pred NP1 NP2

FEW –x
^ x¯ S1 ^ ¯x S2

Pred NP3 Pred NP

NP4 –
MANY – :: Cat x
y
^–y S3 ^–y S4

Pred NP NP Pred NP
–
(::) Catch –x
–
y :: Mouse y
-
 j ::Cat(ā), [MANY
c. d-17 [FEW a bj ::Mouse(b-), (::)Catch(ā,b-)]]
The group reading says ‘a small group of cats caught a large group of mice’. In
this reading subsequent definite reference can be made to the large group of
mice, as in These mice had escaped from a laboratory, which requires the
inferentially added address:
(8.28) d-18 [MANY a  j ::Mouse(ā), [FEW -bj ::Cat(b),
 Catch(b,ā)]]


But in the distributive reading, with ::Catch(b,ā), subsequent definite
reference is not possible, which means that inferential bridging of the kind
at issue must be blocked. The passive of (8.27a):
(8.29) Many mice were caught by few cats.
is equivalent to (8.27a) only in the group reading. In the distributive reading
scope differences destroy the equivalence.

8.1.2.2 Discourse-sensitive universal quantification We now pass on to the

universal quantifier ALL. Here we must refer to what was said at the end of
Section 2.3.5.2 regarding the discourse-sensitivity of the universal quantifier
ALL in its normal interpretation. A great deal was said in the chapters on
logic about the logical properties of this quantifier and its existential counter-
part without taking into account any complications that might arise in the
context of discourse semantics. It is now time, however, to be serious about
the context-sensitive aspects of the universal quantifier.
Since the only way of introducing entity representations into a D is by
means of existential quantification, it follows that universal quantification
cannot open a discourse and needs previous existential quantification to fall
back on. In other words, universally quantified sentences (type A) are per se
occasion sentences, not eternal sentences. If standard modern predicate
250 The Logic of Language

calculus (SMPC) treats A-type sentences as eternal sentences, it can only do

so by assuming a nonnull universe OBJ of extensional objects (an implicit
assumption seldom present, nowadays, in the minds of logicians) and by
letting the variables range over OBJ.7 In principle, therefore, SMPC pushes the
context-sensitivity of the universal quantifier to the very limit of the universe
of objects as a whole. This fact is recognized by a number of formal semanti-
cists, who have proposed various strategies for restricting the domain of
universal quantification (e.g. Stanley and Szabó 2000; Peters and Westerståhl
2007). I will not discuss these proposals here, as they fit into a very different
theoretical frame, which I reject for general reasons. The point here is that
presupposition, and in particular the accommodation mechanism, must be
taken to play a central part in any such strategy. For the rest, the problem
seems reducible to the general problem of fixing definite reference.
In any case, some restriction on the domain of the universal quantifier is
required if justice is to be done to natural language. A sentence like All farmers
grumble, is only interpretable if D already contains an (open or closed)
address for a set of at least two farmers. And likewise for variants of all,
such as every or each. This is, it would seem, the origin of the existential
import assigned to ALL in Aristotelian-Abelardian and Aristotelian-Boethian
predicate calculus.
This perspective requires a rethinking of universal quantification as a
whole—an enterprise which is not feasible within the constraints of the
present text. What can be presented here is, therefore, again, tentative and
provisional. I propose to treat the quantifier ALL from now on as a binary
higher-order predicate expressing a relation between a definite restrictor set
and a set of elements satisfying the matrix predicate. The truth condition does
not change. An A-type sentence is still true just in case the (definite) restrictor
set is an element in the plural power set of the matrix set: ALL F is G is true just
in case [[F]] 2 Ppl([[G]]).8

7
It is possible to hide the discourse-dependency of the universal quantifier in a conditional (see
Section 8.2.4), as when one says If there is a set of farmers, then all farmers grumble, which seems to be a
sentence type involving the use of any as in Any doctor will tell you that smoking is bad. But here again,
the set of farmers has to be introduced first, albeit under the conjunction if. And, of course, saying If
there is a set of farmers, then all farmers grumble is not the same as saying All farmers grumble.
8
One notes that the condition [[F]] 2 Ppl([[G]]) is equivalent with the condition [[F]] [[G]], with
[[F]] and [[G]] as natural sets, just as [[F]] 2 P([[G]]) is equivalent with [[F]]  [[G]]. The advantage of
the formulation [[F]] 2 Ppl([[G]]) is that it provides a unified solution to the type problem caused by
standard analyses for sentences like All the farmers dispersed. The analysis 8x(Farmer(x) ! Disperse
(x)) or, in terms of generalized quantification, 8x(Disperse(x),Farmer(x)) will not do because
Disperse(x) contains a type error. Interestingly, the standard analysis may be taken to explain why
the tenseless ‘eternal’ sentence All farmers disperse is infelicitous, but it fails to account for the
felicitousness of an occasion sentence like All the farmers dispersed. The analysis given here provides
 Discourse incrementation 251

The definite restrictor set is thus commanded, in SA-structure, by the

definite determiner the, as shown in (8.30b). This analysis establishes a link
between ALL and the definite determiner the, which, in fact, optionally occurs
with all in ordinary English sentences such as All the farmers grumble, or All of
the farmers grumble, or, in the ‘floating quantifier’ version, The farmers all
grumble, where all occupies the place of an adverbial expression. In French
and Italian, the definite article is even obligatory: Tous les paysans grognent or
Tutti i contadini brontolano.
For sentence (8.30a), this gives the SA presented in (8.30b) and the resulting
incrementation shown in (8.30d), based on a previously existing (open)
address (8.30c):
(8.30) a. All (the) farmers grumble.
b.
So

Pred NP1 NP2

ALL –
x
^–x S1 Det S2
the –
x
Pred NP Pred NP
– –
x
:: Grumble x ::Farmer

 j ::Farmer(ā)]
c. d–19 [a
 j ::Farmer(ā) // ALL[::Grumble(19)]]
d. d–19 [a
In this version of ALL, (8.30b) is read as ‘the individual grumbling of each
farmer was total as regards the set of farmers’. The quantifier ALL thus functions
as an adverbial modifier of the L-proposition Grumble(the farmers).9
For (8.30a) to be incremented, D must already contain d–19 [ā j ::Farmer
(ā)] (‘there are farmers’), thus ensuring that the class of farmers is nonnull.
The address d–19, if still open, is closed, following the primary definite
reference to this particular set of farmers. [[::Grumble(x)]] denotes the set

the answer. Even so, ‘occasion’ ALL is not suitable for just any higher-order predicate. A sentence like All
farmers are numerous is incoherent even though numerous is an intrinsically higher-order predicate.
The reason seems to be that the natural language semantics of ALL requires that no member of the
restrictor set (the set of farmers, in this case) be left out—a condition that makes sense for predicates
like disperse or sit in a circle but not for a predicate like numerous.
9
This might help explain the hitherto unexplained fact that all is allowed to ‘float’, occupying an
adverbial modifier position, as in The farmers all grumble, whereas such ‘floating’ is impossible, or
anyway much less current, for existential quantifiers. Note that ‘floating’ quantifiers typically occupy
adverbial positions in the sentence, not only in English but in all languages I have so far checked.
252 The Logic of Language

of sets of at least two individually grumbling individuals. ALL (::Grumble(19))

says that the grumbling of the farmers in question applies to the full set of
farmers at issue. This makes, strictly speaking, ALL(::Grumble(19)) equivalent
with ::Grumble(19), though it emphasizes that the predicate ::Grumble
applies to all the farmers without exception. The difference between ALL and
plural the is thus taken to be that while they both select the set of farmers
defined by the address head of d-19 after closure, and let the sentence say that
this set of farmers is an element in the set of sets defined by the type-raised
predicate ::Grumble, ALL specifically requires, redundantly, one would say,
that no member of the set of farmers be left out.
We are now in a position to say that, whereas the existential quantifier is
an instruction to set up a new open address, the universal quantifier is an
instruction licensing the closure, if necessary, of any still open address for any
representative of the restrictor set and the addition of the predication of the
matrix set.
Thus, suppose D contains an address d-20 [a j Farmer(a), See(I,a)]
saying ‘I see/saw a farmer’, the universal quantifier as used in (8.30a) licenses
the closure of d-20 and the addition of the predication Grumble(20), result-
ing in:
(8.31) d–20 [a j Farmer(a), See(I,a) // Grumble(20)]]
It seems that EVERY distinguishes itself from ‘occasion’ ALL mainly in that
EVERY excludes group readings. The specifics of EACH are not touched upon
here. Nor will I attempt to disentangle the complexities of universal or
existential ANY, which has so far eluded all researchers.10

8.1.3 Subordinate subdomains

As has been said, a D may contain subdomains. These are either instructional
or alternative or subordinate. INSTRUCTIONAL SUBDOMAINS are set up in virtue of
some increment instruction, as with the universal quantifier discussed in the
previous section, or the negation, discussed in Section 8.2.2. ALTERNATIVE

10
What has remained unexplained in the account given above is the possibility of the negative
polarity item any occurring in restrictive relative clauses attached to the restrictor term of sentences
quantified with all and every, but not, at least according to many speakers, with each:
(i) Every student who had done any work passed.
(ii) All students who had done any work passed.
(iii) *Each student who had done any work passed.
Nor do I have an explanation for the fact that each and all are allowed to ‘float’, as in The students
each (all) went home, while this possibility does not exist for every: *The student every went home.
 Discourse incrementation 253

SUBDOMAINS are created by OR (disjunction) and IF (implication), to be dis-

cussed in Sections 8.2.3 and 8.2.4, respectively. SUBORDINATE SUBDOMAINS are the
topic of the present section. They are a special kind of address. Like ordinary
addresses, they can be open or closed. They also have D-properties, in that
they may contain their own addresses, increments and instructions. They are
represented both as a special kind of address, labelled D–n, and as an indexed
domain.
Let us consider an example, by way of succinct explanation. A sentence like
(8.32a) is incremented as (8.32b), where the variable D ranges over virtual
facts (see Section 6.2.3.1 in Volume I). (8.32b1) reads ‘there is a possible fact D’.
Possible carries an instruction to set up a subdomain specifying the virtual
fact in question. This subdomain is represented in (8.32b2) read as ‘There is a
planet called “Minerva” and it is inhabited’:
(8.32) a. There may be a planet Minerva and it may be inhabited.
b1. D–1 [D j Possible(D)]
b2. ¿ D–1: d–1 [a j Planet(a), ‘Minerva’(a) // Inhabited(1)]
‘D–1’ in (8.32b1) is the label of the subdomain introduced as a result of the
predicate Possible. The sign ‘¿’ specifies the contents of the new D–1.
It in (8.32a) refers opaquely, as it finds its antecedent within D–1. Transpar-
ent reference, with open d–21 given in the superordinate D, is shown in (8.33).
In (8.33c) d–21 is closed and the predication D–1[Inhabited(21)] is
added, saying that 21 is represented as being inhabited in the possibility
subdomain D–1:
(8.33) a. There is a planet Minerva and it may be inhabited.
b. D–1 [D j Possible(D)]
¿ D–1: Inhabited(21)
c. d–21 [a j ‘Minerva’(a), Planet(a) // D–1[Inhabited(21)]]
A few general principles hold for subdomains. First, addresses from
the commitment domain ‘percolate downward’ into subdomains as
shown in (8.33), where d–21 in D–1 is taken from D. This downward
percolation is stopped only if the subdomain in question explicitly
blocks the address in question. Then, presuppositions of clauses incre-
mented in subdomains ‘percolate upward’ into higher domains, includ-
ing D, unless blocked either by their explicit negation or by lack of
cognitive backing. This process is called PROJECTION. Both processes
follow from the Principle of Maximal Unity, in particular its subprin-
ciple of Minimal D-change, discussed in Section 7.2.2.4, both of which
254 The Logic of Language

serve the functional purpose of ensuring maximal unity and coherence

in the overall D-structure.
Anaphora may delve into subdomains under intensional predicates. In
(8.34a), for example, the brother Marion believes she has is anaphorically
referred to by he under the intensional predicate be the talk of the town. In
(8.34b), the movie Geert says he has made (but may well not exist at all) is
referred to by the anaphoric pronoun it in the second conjunct standing
under the intensional predicate be all over the news:
(8.34) a. Marion believes that she has a brother, and he is the talk of the
town.
b. Geert says he has made an anti-Islam movie and it’s already all over
the news.
The machinery of the incremental construction of discourse domains and
subdomains is the main explanatory factor for the lack of substitutivity salva
veritate in intensional contexts, which has been the dominant driving force in
theoretical semantics during the twentieth century.

8.2 Instructions
The incrementation procedure IP is also able to follow INSTRUCTIONS con-
straining the further development of any given D. All standard operators of
propositional logic are, from a discourse-semantic point of view, instructions.
The logic of the propositional operators is seen as an emergent property of
basic-natural set theory combined with the discourse-semantic incrementa-
tion instructions. In the following subsections it is shown how this ‘emer-
gence’ can be traced in detail. This way, our combined theory of basic-natural
set theory and discourse-semantic incrementation is meant to provide an
alternative to the currently dominant pragmatic accounts.

8.2.1 Conjunction
The sentential functor AND is, in principle, nothing but an instruction to
increment the conjuncts in the order given. It is the basic discourse-incre-
mentation functor. In many cases, IP is iconic in that it follows the temporal,
causal, or motivational order of the events or situations described, as appears
from the difference between (8.35a) and (8.35b):
(8.35) a. She went to Spain and married. (A Spaniard?)
b. She married and went to Spain. (Alone or with her husband?)
 Discourse incrementation 255

This is not always so, as will become clear in a moment. When the incremen-
tation is iconic, we speak of an ORDERED INTERPRETATION.
Whether the difference between (8.35a) and (8.35b) is truth-conditional or
not is hard to say.11 If I hear (8.35a) and am then told that, in fact, she married
first and then went to Spain, I think I would feel cheated and I might be
prepared to say that what I was told was false. Standard logic is unable to
account for this, but it is consistent with a dynamic logic as proposed in
Groenendijk and Stokhof (1991).
But this is by no means always so. In sentence (8.36), the order of the
conjuncts is definitely relevant but not, it seems, truth-conditionally so:
(8.36) It’s raining and we’re out of booze.12
Here the ordered interpretation is motivated by the speaker’s wish to paint a
picture of utter misery. It wouldn’t be so bad if it were raining and speaker and
company still had a sufficient supply of spiritual refreshments, but being out
of booze while it’s raining is, one gathers, the ultimate agony for speaker and
company.
As is said in Section 7.2.2.3, the second conjunct normally restricts the
first and not vice versa, precisely because, as is said in Section 7.3, the
normal function of every new increment consists in homing in on the target
situation, restricting the number of possible situations in which D can be
true. This is, however, not an absolute rule, because sometimes the
new increment recapitulates what has been said or draws a conclusion, as
in (8.37), where (in terms of valuation spaces) the fact of John’s being in
his office is restricted by the fact that his light is on. Yet in (8.37) John’s light
is on precedes he is in his office, which is made possible by therefore, which
draws a conclusion:
(8.37) John’s light is on and, therefore, he is in his office.
The standard function of every new increment to restrict the set of possible
situations in which D is true is naturally expressed in terms of VS-modelling
(see also Figure 7.1 in Section 7.3.2). Take the two logically and semantically
independent sentences (where she is keyed to the same person):

11
According to Cohen (1971) the difference is truth-conditional, but according to Gazdar (1979:
69–71) it is a Gricean implicature, defeasible by, for example, the addition of in reverse order to a
sentence like (8.35a). Gazdar’s argument, however, fails to convince, because the sentence with that
addition is not the sentence without it: the one may have an entailment which the other lacks.
12
See Blakemore and Carston (2005) for related and highly interesting cases of and-conjunction
such as, for example, Paul can’t spell and he is a linguist.
256 The Logic of Language

a. b.

/39/ /39/

/38/ /38/
/38/

/38 and 39/ /39 and 38/

U U

FIGURE 8.1 The order of incrementation reflected in ‘superimposition’

(8.38) She went to Spain.

(8.39) She married.
Their valuation spaces are rendered in Figure 8.1 as /38/ and /39/, respectively.
The VSs /38/ and /39/ have an M-partial intersection precisely because (8.38)
and (8.39) are logico-semantically independent, so that it is possible for both
to be true, both to be false or for the one to be true and the other to be false. If
D consists of the combined increments of (8.38) and (8.39), /D/ is the
intersection of /8.38/ and /8.39/, as shown in Figure 8.1, no matter whether
one takes Figure 8.1a or Figure 8.1b. The order of incrementation is reflected
in the suggestion of superimposition. Figure 8.1a shows the result of incre-
menting (8.38) first, followed by (8.39), as in (8.35a). Figure 8.1b represents the
inverse order, corresponding to (8.35b). In Figure 8.1a her marrying took place
against the background of her already being in Spain. In Figure 8.1b her going
to Spain took place in the context of her already being married. A three-
dimensional representation would perhaps do a better job, but the reader no
doubt gets the idea.
Yet, as linguists know, the word and, or its equivalents in the languages of
the world, is used in several distinct ways that give the impression of multiple
ambiguity. In English one finds full speech-act conjunction, as in, for exam-
ple, Go home and nobody will know what has happened. This form of conjunc-
tion has received relatively little attention in the linguistic literature on
conjunction. L-propositional conjunction, as in John lives in Paris and
Harry lives in London, is the form studied by logicians, who also look at the
conjunction of L-propositional functions, typically under quantifiers, as in
 Discourse incrementation 257

Some politicians only do politics and forget about policy—a conjunction type
that linguists tend to stay away from.13
It won’t do to speak of multiply ambiguous and unless one is forced to. The
ideal solution would be to reduce all existing varieties of conjunction to
speech-act or propositional conjunction, but the literature is far from unani-
mous on the success of that enterprise.
The reduction of natural language and to full L-propositional conjunction
is complicated by the fact that the conjunctor AND does not conjoin full
L-propositions in surface structure but induces the grammatical process of
CONJUNCTION REDUCTION (CR). The grammars of all languages allow for
a variety of ways to shorten full L-propositional conjunction in the
corresponding surface structures, which then require a reconstruction of the
full L-propositional form by way of a syntactic parsing procedure.14 In this
regard, some notable success has been booked in linguistics, except for so-
called PHRASAL CONJUNCTION, as in (8.40), which is, of course, not reducible to
‘John is a nice couple and Rose is a nice couple’:
(8.40) John and Rose are a nice couple.
Phrasal conjunction clearly requires a separate analysis, which may result in
the conclusion that here, too, one has to do with L-propositional conjunction,
albeit in an encapsulated form. But no well-motivated answer to this question
has so far come to light.15
CR is already visible in (8.35a,b), where the common subject of both L-
propositional conjuncts is eclipsed in the second conjunct. More drastic cases
of CR are shown in (8.41) (the commonly used labels for the type of CR have
been added:
(8.41) a. (Both) Bert and Alex bought a new car. (Left CR)
b. Bert bought a car for Mary and a horse for Sue. (Right CR)
c. Bert bought a car and Alex b/o/u/g/h//t a horse. (Gapping)
d. Bert likes and Alex hates the play. (Right-Node Raising)
What we now see is that some of these reduced forms are and some are not
open to an ordered interpretation. To begin with, the form of CR known as

13
But see Seuren (1996: 325–6) for the scope-sensitivity of AND with regard to quantifiers and the
ensuing restrictions on the syntactic processes of CONJUNCTION REDUCTION.
14
For more detailed discussions, see, for example, Van Oirsouw (1987), Seuren (1996: 323–38).
15
Nor do I have an account of the idiosyncratic use of and in texts like:
(i) Moving now to sports—and in London, Arsenal have qualified for the Europe cup.
As far as I am aware, this use of and is restricted to BBC radio English.
258 The Logic of Language

‘Left CR’ appears not to be sensitive to the order of incrementation. Consider

the following examples (where his widow is to be read as ‘the woman Bert was
married to before he died’ and not as, for example, ‘the widowed woman he
was dating’):
(8.42) a. Bert and his widow each died in a car crash.
b. His widow and Bert each died in a car crash.
c. The same car was rented by Bert and by his widow.
d. The same car was rented by Bert’s widow and by him.
Clearly, Bert’s widow can’t be said to have died in a car crash unless Bert
died first, when his wife was not yet a widow. Yet (8.42b) is not anomalous,
which suggests that Left CR is indifferent to ordered interpretation.
The situation is more complex for cases of Right CR, as in (8.42c,d)
or (8.43) and (8.44). (8.43a) and (8.44a) seem all right but the corresponding
b-sentences do not, presumably because an ordered interpretation forces itself
on the sentences involved. Yet both (8.42c) and (8.42d) also seem both all
right, despite the fact that they involve Right CR. I know of no literature about
such questions, but prima facie it would seem that perhaps the crucial factor is
whether or not the underlying L-propositions have a common main verb, in
which case there is no order-sensitivity, or different main verbs, in which case
the interpretation is order-sensitive.
(8.43) a. Alex killed Bert and married his widow.
b. !Alex married Bert’s widow and killed him.
(8.44) a. Alex emptied the bottle and threw it away.
b. !Alex threw the bottle away and emptied it.
Gapping again makes for conjunctive sentences that are not sensitive to an
ordered interpretation, as one can see from (8.45a,b). Although, normally
speaking, one has to die before one can be buried, sentence (8.45b) does not
strike one as anomalous:
(8.45) a. Bert filmed the woman’s death and Alex /f/i/lm/e/d/ her burial.
b. Alex filmed the woman’s burial and Bert /f/i/lm/e/d/ her death.
Right-Node Raising again requires an ordered interpretation, since (8.46a)
tells a different story from (8.46b). Here, the difference comes again close to
being truth-conditional, especially when the word again is placed in final
position. (8.46c) is anomalous as again lacks an antecedent:
 Discourse incrementation 259

(8.46) a. Bert closed and Alex opened the door (again).

b. Alex opened and Bert closed the door (again).
c. !Alex opened again and Bert closed the door.
It thus seems that Left CR and Gapping result in sentences that are not
sensitive to an ordered interpretation, whereas Right CR may and Right-Node
Raising does result in sentences that are. As far as can be seen, the difference
resides in the fact that the L-propositional forms underlying cases of Left CR
and Gapping have the same main lexical predicate in both conjuncts,
whereas those underlying cases of Right CR and Right-Node Raising may
have different main lexical predicates. This, together with an ‘event’ interpre-
tation of the L-propositions concerned (at least for temporal ordering),
may be the necessary and sufficient conditions for successively ordered
interpretation, but there may well still be further factors involved. The matter
must remain undecided for the moment.

8.2.2 Negation
Negation looks simple but that appearance is deceptive. It has all the com-
plexities one should expect of natural language phenomena. Like AND, it
primarily takes propositions and propositional functions in its scope, but,
unlike AND, it cannot stand over a speech-act operator. It may restrict itself
to predicates, as in nonconformist, non-Catholic, but such cases can, in princi-
ple, be treated as negation over a propositional function because that is
what first-order predicates are, propositional functions that take one or
more individuals and deliver a truth value—that is, of type (e,t). The same
applies to the internal negation of predicate calculus, as in Some fishermen do
not swim. Here again, the negation is construed as applying to a propositional
function, in this case the function ¬ ::Swim(x).
Lexically incorporated negation, discussed in Section 8.6.4 of Volume I,
presents many problems, as already noted by Aristotle. The words polite and
impolite denote not contradictory but only contrary properties. Neither
immoral nor amoral is the contradictory of moral, though both are contrary
with it. This, however, is a matter of lexical semantics, not relevant in the
present context.
What is relevant is the phenomenon of metalinguistic usage, as in:
(8.47) a. Not Liz, you twit, but Queen Elizabeth has just been on TV.
b. The man isn’t intelligent, he’s a whopping genius.
c. In this house, we don’t eat grub, we eat food.
d. I did NOT only see Act One, Two, and Three, I only saw Act One.
260 The Logic of Language

Cases like these are discussed in Horn (1985, 1989) and, in their wake, in a
flurry of publications during and after the 1990s. The main trend, in this
literature, is to treat these cases as instances of a ‘pragmatic’ transfer from
the standard ‘descriptive’ negation to a metalinguistic, more or less ‘meta-
phorical’, negation, but how exactly this metamorphosis takes place, has never
been made clear in a proper falsifiable way.
In Seuren (1988, 2000) I argue that if such pragmatic accounts predict
anything, it is this: when someone suggests that, say, some politician is a crook
by quoting the famous ironical line And Brutus is an honourable man from
Shakespeare’s Julius Caesar, a second person, who strongly believes that the
politician in question is entirely blameless, should, if the pragmatic account is
correct, be able to rebut this suggestion by saying And Brutus is NOT an
honourable man, which, of course, does not work. All such a person can
do is say something like To hell with your Brutus! It is also argued there that,
while cases like (8.47a,b,c) are instances of lexical-choice correction, cases
like (8.47d), though likewise of a ‘metalinguistic’ nature, belong to a separate
category, namely the category of presupposition denials. This latter point
is taken up in Section 10.4, and we will not touch on it here. Cases of lexical-
choice correction are discussed below.
Leaving aside, for the moment, distracting phenomena such as those
mentioned above, we start with ordinary sentential negation. For IP,
sentential negation, as in NOT-S, is an instruction banning the incrementa-
tion of S from D. The banning order is symbolized in D-representations as
‘ * ’. Ordinary (default) negation is presupposition-preserving, which means
that the non-negated proposition must be normally incrementable, or, if
you like, must ‘have the right papers’, for the D at hand. IP takes the
subject-S (the scope) of NOT in the SA-tree and processes it first without
negation (if it hadn’t been processed already, for it to be negated subse-
quently by a second speaker). Subsequently, NOT places an asterisk before
the increment.
Double negation bans the banning, and thus establishes incrementation in
a roundabout way, based on inferential bridging. Treble negation, as is easily
seen, makes one dizzy and confused, in accordance with the principle of
natural set theory PNST–6, discussed in Section 3.2.2, which says that the
function COMPLEMENT is nonrecursive in basic-natural, and only once-recur-
sive in strict-natural set theory. Quadruple negation is, though formally well
defined, entirely unrealistic as a cognitive process—unless, of course, the
later negations are not logically functional but serve only to lend emphasis
to the first, original, negation, as in the New York Black-English sentence
(8.48a), taken from Labov (1972: 130), or the London Cockney (8.48b), taken
 Discourse incrementation 261

from the 1960s BBC TV play Cathy Come Home (see also note 23 in Chapter
8 of Volume I):
(8.48) a. It ain’t no cat can’t get in no coop.
b. ‘E’s an odd fella. ‘E ain’t never been no good to no woman, not
never.
Grammatically, both (8.48a) and (8.48b) have four negations (if we
disregard the rhetorical addition not never in (8.48b)) but only the first is
logically functional; the other three result from the well-known grammatical
process of NEGATION COPYING, found in a vast number of languages in some
form or other, and putting a copy of the ‘original’ negation on some or all
subsequent existential operators (see Seuren 1996: 269). (8.48a), where the
copying of the negation even penetrates into a relative clause, is equivalent
to standard English There isn’t any cat that can get into any coop. (8.48b)
reads as the standard English He is an odd fellow. He hasn’t ever been any
good to any woman, which, somehow, makes less of an impression than
(8.48b).
Following the argument in Chapter 2, we treat negation as an abstract
predicate in SA-structure, like the other propositional operators and the
quantifiers. Its subject-S is what is normally called its scope. Thus, (8.49a)
has the SA (8.49b). (8.49c) is the incremental result on the basis of the pre-
existing and therefore pre-recorded open address d–2 [a jMouse(a),[b j Cat
(b), Catch(b,a)]] (‘A cat caught a mouse’) given above in (8.11).
(8.49) a. The mouse did not escape.
b.
S

Pred S
NOT
Pred NP
Escape
Det S
the x
Pred NP
Mouse x

c. d–2 [a j Mouse(a), [b j Cat(b), Catch(b,a)] // *Escape(2)]

Sentential negation, as exemplified in John has no car, thus amounts to an
instruction for the D at hand to refuse the existential introduction of
an address for a car of John’s and the addition, after closure, of the predication
[b j Car(b), Have(22,b)] to any, possibly open, address d–22 [a j ‘John’(a)]
for someone called ‘John’. Thus, given the, possibly open, address d–22 [a j
262 The Logic of Language

‘John’(a)], its extension to (8.50a) is banned, just as the introduction of an

address of the form (8.50b):
(8.50) a. d–22 [a j ‘John’(a) // *[b j Car(b), Have(22,b)]]
b. *d–23 [a j Car(a), Have(22,a)]
Negated open addresses cannot be closed, as there is nothing to close. This
explains the incoherence of, for example, (8.51a), where it calls upon an
address that has just been banned:
(8.51) a. !John has no car. It is in the garage.
b. John has no car. So it can’t be in the garage.
c. John has no car. His car (?It) is a figment of your mind.
In (8.51b), we detect an echo effect, in that it fits naturally into a context
in which someone has just said that, or has put up the question of
whether John’s car is in the garage. The second sentence of (8.51b) is thus
to be read as ‘the car just said (suggested) to be John’s car cannot be in the
garage (because John has no car)’. Analogously for (8.51c), which seems to
fall back, again, on an assertion (or question) uttered earlier by someone
asserting that (or questioning whether) John has a car. The second sentence
declares that object to be a figment of the previous speaker’s mind. Why it
should be that, in cases of this nature, the full NP (his car) goes down much
better than the anaphoric pronoun it, is, admittedly, a question to which
I have no answer. (Incidentally, such examples show again the necessity of
accepting that speakers naturally make reference to nonexistent or virtual
objects; see Section 10.8 for more detailed discussion of anaphoric reference to
objects said not to exist.)
A sentence like (8.52a) is thus incremented as (8.52b):
(8.52) a. No mouse was caught.
b. *d–24 [a j Mouse(a), Be caught(a)]
But (8.52a) is not equivalent with (8.53a,b) (in the sense in which ALL takes
scope over NOT), though (8.53a,b) is equivalent with (8.54a,b):
(8.53) a. All the mice were not caught.
b. d–25 [a j ::Mouse(ā) // ALL*(::Be caught(25))]
(8.54) a. None of the mice were caught.
 j ::Mouse(ā) // *SOME(Be caught(26))]
b. d–26 [a
Now back to (8.47a–c), repeated here for convenience:
 Discourse incrementation 263

(8.47) a. Not Liz, you twit, but Queen Elizabeth has just been on TV.
b. The man isn’t intelligent, he’s a whopping genius.
c. In this house, we don’t eat grub, we eat food.
These are cases of lexical-choice correction and, therefore, of a meta-
linguistic use of the negation. No grammar and no proper semantics has
been set up, as yet, for cases of this nature, probably because the dominant
attitude has been, over the past twenty years, to relegate such phenomena
to pragmatics.
It cannot be our purpose here to present a full formal account of such
sentences, as that would require a separate monograph. What we can do is
point out that sentences of this type have a common structure representable
as something like (8.55a), where the quotation marks signal the metalinguistic
nature of the sentences at issue. A more formal (SA) rendering of (8.55a)
would be (8.55b), where the caret quotes secure reference to the phonological
(or, if appropriate, the phonetic) form of the words in the range of the
variables a and b and where Bev is the value-assigning predicate discussed
in Section 3.2:
(8.55) a. The proper expression for x in ‘—x—’ is not ‘a’ but ‘b’.
b. NOT [Bev ^a^ (the proper expression for x in ^—x—^)];
Bev ^b^ (the proper expression for x in ^—x—^)
(8.47a) is then read as ‘The proper expression for who has just been on TV is
not ^Liz^ but ^Queen Elizabeth^’. (8.47b) is read as ‘The proper expression
for what this man is is not ^intelligent^ but ^a whopping genius^’, while
(8.47c) reads as ‘The proper expression for what we eat in this house is not
^grub^ but ^food^’.
In the analysis given in (8.55b), this type of sentence is a metalinguistic variety
of the cleft construction, expressing a topic–comment-modulated proposition.
This is confirmed by the fact that the isolation of x from ‘—x—’ is subject to
the normal isolation—that is, extraction or insertion—constraints that have
found to be valid in grammars, as appears from, for example, (8.56a), where
Liz’s cannot be isolated from the position occupied by Queen Elizabeth’s
because the position involved is a modifier genitive position and modifiers
cannot be isolated from S-structures (we can say John whose wife died but
not John’s who wife died). Not so in (8.56b), where the position involved is
that of a full NP, not of a modifier, or in (8.56c,d), where no isolation has taken
place:
264 The Logic of Language

(8.56) a. *It isn’t Liz’s, you twit, that he praised hat but Queen Elizabeth’s.
b.
It isn’t Liz, you twit, that he praised the hat of but Queen
Elizabeth.
c.
He praised not Liz’s, you twit, but Queen Elizabeth’s hat.
d.
He praised the hat not of Liz, you twit, but of Queen Elizabeth.
Similar observations can be made regarding (8.47b) and (8.47c):
(8.57) a. *It isn’t an intelligent man’s that he is son but a whopping
genius’s.
b.
It isn’t an intelligent man that he’s the son of but a
whopping genius.
c.
He is not an intelligent man’s but a whopping genius’s son.
d.
He is the son not of an intelligent man but of a whopping genius.
(8.58) a. *In this place, it isn’t grub we have standards but food.
b.
In this place, it isn’t grub we have standards for but food.
c.
In this place, we don’t have grub but food standards.
d.
In this place, we have standards not for grub but for food.
How exactly the grammar turns a (8.55b)-type structure into a
corresponding surface structure, or what parameters are required in a seman-
tic theory for a proper formal interpretation of cases like (8.47a–c) are, as has
been said, questions that have so far received no attention at all in the
linguistic literature. Given this unsatisfactory state of affairs, it would be futile
even to try to present an incrementation procedure for such cases.

8.2.3 Disjunction
The disjunctive operator OR is far from analogous to its conjunctive counter-
part AND. While AND can be used to conjoin speech acts, as in the sentence Go
home and nobody will know what has happened quoted above, it is not at all
clear that OR can be used in an analogous way. One might think of an example
like (8.59a), but it is unclear how it should be read in the systematic terms of
speech act operators and propositions. We revert to this issue below, when
discussing the tacit expansion of disjunctions to the form specified in (8.66).
(8.59) a. Don’t try, or you’ll get caught.
b. Do you want coffee or do you want tea?
(8.59b) is likewise not without problems. One may presume that it reads as
something like ‘I am asking which: what you want is coffee or what you want
is tea’. If that is correct, OR is used in combination with the specific-question
 Discourse incrementation 265

operator WHICH asking for a choice to be made among given alternatives. Little
is known about how such a combination works, owing to the primitive state
of research on the grammar of speech acts. One notes, incidentally, that
(8.59b) is unambiguous as regards the scope of OR, unlike (8.60), which is
ambiguous (unless intonation is taken into account):
(8.60) Do you want coffee or tea?16
Like NOT, OR can be used to correct lexical choices, as in:
(8.61) He was a sad, or rather pathetic, man.
AND can only be used this way when followed by NOT, as in (8.62a), and in such
cases it can be left out. One notes that the use of but instead of and, as
in (8.62b), takes away the metalinguistic reading, leaving only the object-
language reading:
(8.62) a. He was a sad, (and) not a pathetic, man.
b. He was a sad, but not a pathetic, man.
Like AND, OR has a form of phrasal coupling, as in the example John and Rose
are a nice couple quoted above as an instance of phrasal conjunction. The
disjunctive version would be something like:
(8.63) Aberdeen or Inverness is an impossible choice.
Yet phrasal disjunction, as in (8.63), has a metalinguistic flavour which
phrasal conjunction lacks. (8.63) reads as ‘“Aberdeen or Inverness” is an
impossible choice’.
One remembers from Section 7.2.2.3 the curious reduction of (8.64a) to
(8.64b):
(8.64) a. Paul may be at home and he may be in hospital.
b. Paul may be at home !and/
or in hospital.
On the whole, the relation between natural language AND and OR is far from
clear. It is certainly not exhausted by the purely logical account of these two
operators.

16
There is an anecdote about Bertrand Russell, said to have replied ‘Yes’ to a stewardess asking him
(8.60). According to the anecdote, Russell wanted to teach the poor lady that or is inclusive, not
exclusive, but, if that is so, he would have had to ignore the stewardess’s intonation, which, in all
likelihood, had a rising tone on coffee and a falling tone on tea.
266 The Logic of Language

The most common use of OR, however, is its use as a propositional

connector, as in (8.65a), which can be reduced by grammatical reduction
processes to (8.65b):
(8.65) a. This house has no basement or this house has no garden.
b. This house has no basement or no garden.
A variant of full propositional OR is OR connecting propositional functions,
as in:
(8.66) Most of our students are Portuguese or Brazilian.
As regards full propositional OR, we hark back to Section 3.4.3. There it is
proposed that, at a basic-natural level, OR is best accounted for by assuming that
sentences of the type P OR Q (limiting ourselves to binary disjunction) should be
analysed, at some underlying level, as (8.67), repeated from (3.23) in Section 3.4.3:
(8.67) (P AND NOT-Q) OR (NOT-P AND Q); formally: (P ∧ ¬Q) ∨ (¬P ∧ Q)
The reason for this assumption is, as one remembers, that it is only in this way
that the basic-natural requirement that the arguments of the OR-function are
mutually exclusive can be met, even though P and Q are themselves logically and
semantically independent, meaning that /P/ OO /Q/. In order to ensure that the
two disjuncts are mutually exclusive even though P and Q are logically and
semantically independent, we expand P to P AND NOT-Q and Q to NOT-P AND Q.
The tacit exclusion (negation) of the second disjunct in the first and vice versa
rests on the TRUE ALTERNATIVES CONDITION for Ds, requiring that the alternative
increments under disjunction be truly distinct. This explains the much-
debated ‘exclusive’ character of OR as resulting from principles of coherent
discourse construction.
When we say that P OR Q is to be ‘tacitly’ read as (8.67), what we mean is
not that the underlying SA, which is the input to the grammar module,
should be of the form (8.67). This would go against all principles of sound
grammatical theorizing. What is meant is that the expansion from P OR Q to
(8.67) should be specified (a) as part of the (basic-natural) lexical semantics of
OR and (b) as part of the IP-instruction associated with OR. This way the
expansion to (8.67) is kept out of the syntax, as it should be.
We posit that the IP for normal propositional OR is an instruction, in
the sense defined above, of the following form, which ensures that the
‘tacit’ reading is materialized in the IP-result, not in the grammar. Each
subdomain now corresponds automatically to a valuation space that excludes
the others:
 Discourse incrementation 267

INSTRUCTION FOR L-PROPOSITIONAL OR

For a propositional disjunction DIS of the form P1 OR P2 OR . . . Pn,

expand the disjunct P1 of DIS to P1 AND NOT-P2 . . . NOT-Pn and analo-
gously for all Pm (1 < m  n). Then set up as many separate subdomains
as there are expanded disjuncts, whereby each subdomain must be
properly anchored to the given D.
It must be stressed that the instruction to expand disjunctions to the format
of (8.67) is taken to hold only at a basic-natural level. At stricter natural levels,
the instruction must be taken to start with the second sentence: ‘Set up as
many separate subdomains as there are expanded disjuncts.’ Yet the basic-
natural interpretation stays with us all the time, even when we operate at a
stricter level of naturalness, or else the type of donkey anaphora shown in
(8.68) would not be a problem for formal semanticists.
The subdomains set up as a result of the OR-instruction naturally give rise
to the explicit or (mostly) implicit question of which of the subdomains
should be elected to continue the D at hand. This is not only one of the
sources of topic–comment modulation, it also drives the further development
of D in that the participants in the discourse feel urged to look for arguments
that will decide in favour of one of the alternatives.
The IP for OR accounts for the possibility of ‘donkey’ anaphora, as in (8.68),
where the anaphoric pronoun it appears to lack an antecedent:
(8.68) Either Socrates does not own a donkey or he feeds it.
Given the condition in the OR-instruction saying that the two expanded
disjuncts must be well-anchored to the current D, one sees that both ‘Socrates
does not own a donkey and (therefore) he does NOT feed it’ (with presupposi-
tion-cancelling radical NOT) and ‘Socrates owns a donkey and he feeds it’ are
proper increments to one and the same given D.
There is a problem with OR as a connector of propositional functions, as
in (8.66) above. Sentence (8.66) appears to lack the ‘exclusive’ character
generally adhering to OR, as it seems readily and naturally to allow for
truth when some of the students in question are both Portuguese and
Brazilian, having dual nationality. The sentence may then be taken to want
to say that most of the students speak some variety of Portuguese. And
indeed, donkey anaphora seems to lose its facilitation in existentially
quantified sentences like (8.69a). This would then be explained by the
assumption that (8.69a) is not subject to the expansion procedure specified
268 The Logic of Language

above. But this does not apply to (8.69b), the propositional-function

equivalent of (8.66) with MOST, where donkey anaphora appears fully
acceptable.
(8.69) a. ??Some farmers do not own a donkey or they feed it.
b.
Most (of the) farmers do not own a donkey or they feed it.
We must leave this problem open and hope that future research will produce
an answer.
The expansion of P OR Q to (8.67) has a wider application than to just
propositional OR. It applies, in an analogous but far from fully understood
way, also to the quasi-speech-act OR of (8.59a) and (8.59b). The first disjunct
of (8.59a) is, perhaps, expandable to ‘I urge you not to try and I predict that
you will not get caught’, but it is unclear what is to be done with the second
conjunct or you’ll get caught, which is not readily expandable to ‘I urge you to
try and I predict that you’ll get caught’. As regards (8.59b), if the reading
proposed above holds, it expands to ‘I am asking which: what you want is
coffee and not tea, or what you want is tea and not coffee’. Likewise for phrasal
disjunction, as in (8.63), which expands naturally to ‘Aberdeen and not
Inverness or Inverness and not Aberdeen is an impossible choice’. Where the
expansion does not seem to apply in any way at all is in cases of metalinguistic
lexical-choice correction, as in (8.61) above.
Yet another of the many problems connected with OR (and IF) that there is
no space to explore more fully is the following. Given the expansion of
disjunctions according to the format of (8.67), the logic of OR need not
account for its exclusive character: OR itself can remain inclusive, as logic
prefers it to be. But do we also follow the logicians in claiming that the
subdomains set up as a result of the OR-instruction are fully extensional? That
is, do these subdomains freely allow for substitution salva veritate (SSV) of
co-referential terms? The question is of no direct relevance to the IP, but it is
directly relevant to the semantics of discourse domains. Consider the follow-
ing sentence:
(8.70) Either the morning star is the evening star, or there are ten planets.
According to standard propositional calculus, SSV is guaranteed in these
cases. If, for example, the term evening star is replaced with morning star, as
in (8.71), the result may be a little quaint but in the actual world, where the
two names refer to the same entity—the planet Venus—the truth value does
not change:
(8.71) Either the morning star is the morning star, or there are ten planets.
 Discourse incrementation 269

It is clear that (8.70) and (8.71) express quite different speaker commitments,
just as Frege’s original The morning star is the evening star expresses a quite
different speaker commitment from The morning star is the morning star. But
this is due to the different truth conditions. The actual identity of the
morning star and the evening star ensures that, despite the different truth
conditions, the truth value stays unchanged.17
As regards D-structures, it seems to be generally so that subdomains
created in virtue of an instruction associated with a propositional operator
are always extensional in the sense defined. This holds for conjunction,
negation, disjunction, and also, as will become clear in a moment, for condi-
tionals and it seems to be due to the fact that the increments stored in these
subdomains must be well-anchored—that is, ready for direct attachment
to the current D (they must have ‘the right papers’ for the current D).
Being a subdomain set up under such an instruction is thus a sufficient
condition for the subdomain to be extensional, but it is not a necessary
condition. For there are extensional subdomains that are not the result of
such an instruction but are embedded under a lexical predicate, such as
the subdomain embedded in the object position of the verb Cause, as in
(8.72a), or the factive subdomain embedded in subject position under verbs
like Prove or Suggest, as in (8.72b):
(8.72) a. The arrival of the police caused the butler to flee.
b. That the butler had fled suggested that he was guilty.
Suppose the butler in question is identical with the person referred to by
the phrase the man with the white gloves, any truth value of (8.72a) or
(8.72b) remains unchanged when the man with the white gloves is substituted
for the butler.
Semantically, it is clear why this should be so. Just as Frege said (see Section
7.2.1.2), a subdomain is intensional when it reflects the thought-contents of a
thought-predicate, usually in object position. Since Cause is not a thought-
predicate, it is listed in the lexicon as extensional with regard not only to its
subject but also to its (possibly sentential) object term, while Suggest is
lexically marked as extensional with regard to its subject term, whether
sentential or nominal, but intensional with regard to its object term because
Suggest is a thought-predicate and its object term contains the contents of
what is suggested.

17
I have wavered on this issue in the past but, on reflection, I think I must come down in favour of
the standard position and retract what I argued for in Seuren 1985: 396.
270 The Logic of Language

It would seem that this difference must be formally marked in D-structures,

as it has not so far been shown to follow from any structural or otherwise
formally recognizable feature in D-structures. The simplest way of doing so is
to borrow the marking used for the lexicon (in our notation, the asterisk; see
Section 5.3 in Volume I). But it would be preferable if the intensional nature of
any given subdomain were seen to follow from independent factors formally
rendered in D-structure representations.

8.2.4 Conditionals
Let us now have a look at conditionals—that is, sentences of the type if P then
Q. Ever since the Stoic philosophers set up their propositional logic during the
third century BCE, natural language IF has been widely regarded, especially by
philosophers and logicians, as representing the truth-functional operator of
material implication—a view that immediately gave rise to public controversy,
as appears from a little epigram written by the Alexandrian librarian and poet
Callimachus in the third century BCE, saying ‘Even the crows on the rooftops
caw about the nature of conditionals’ (Kneale and Kneale 1962: 128). The
question is: were the Stoics, and with them the logical tradition till the present
day, right or wrong in considering IF a truth-functional propositional operator?
The answer to that question is not simple. Perhaps one should say that to a
considerable extent they were right, but to some extent they were not. Such an
answer, of course, calls for some comment.
Most semanticists assume that the natural language subordinating conjunc-
tion if represents the well-known truth-functional operator ‘!’ of propositional
logic, also known as ‘material implication’, which yields falsity only when the
antecedent clause (or if-clause, also called by the ancient Greek name prótasis) is
true and the consequent clause (also known as apódosis) is false, and truth in all
other cases. Yet the same semanticists accept, or admit, that this assumption is
tenable only with a heavy supply of pragmatic support. It seems incontrovertible
that the equation of natural language if with material implication falls far short
of accounting for the way if is naturally used by speakers, which gives rise to the
question of whether the deficit is to be made up for by pragmatics or by a
different kind of logic, for example basic-natural logic, combined with the
mechanics of discourse incrementation. The latter is preferable on methodolog-
ical grounds, as it makes for greater precision.18 It is also the view defended here.

18
Johnson-Laird (1986) agrees that context plays an important role in the interpretation of
conditionals, but he takes a dim view of the powers of logic to account for their vagaries. He writes
(1986: 73):
 Discourse incrementation 271

In searching for a logical, or in any case formal, basis we are helped a great
deal by what we have found regarding the semantics of OR. In standard logic,
of course, P!Q is equivalent with ¬P ∨ Q, and it is interesting to see to what
extent this parallelism can be sustained—an issue to which we return below.
Meanwhile we can say that the parallelism fails in some respects. For example,
it fails with regard to the number of propositional terms taken by IF: whereas
OR can take any number of arguments IF is restricted to two. There is a parallel
in that speech-act conditionals are as unclear as regards their analysis as are
speech-act disjunctions. A sentence like (8.73), though perfectly normal from
the point of view of English usage, has not so far received a proper analysis:
(8.73) Shoot me, if you dare.
NP-conditionals, as a parallel to NP-disjunctions and NP-conjunctions, are
questionable. (8.74a) is quaint, though better with also; (8.74b) is a great deal
better:
(8.74) a. ?If Aberdeen then (also) Inverness is an impossible condition.
b. If not Aberdeen then Inverness is an impossible dilemma.
Finally, conditionals seem to be subject to different conditions of contex-
tual fit from disjunctions. For example, if someone asks ‘Why does Philip pay
so much income tax?’, (8.75a) may be an appropriate answer but (8.75b) can
hardly be:
(8.75) a. Either he doesn’t know the law or he is a bachelor.
b. If he knows the law, he is a bachelor.
There are, therefore, considerable differences between conditionals and dis-
junctions, despite the fact that they are logically interchangeable.

How do people interpret conditionals? They set up a mental model based on the meaning of the antecedent, and on
their beliefs and knowledge of the context. They then determine the nature and the degree of the relation between
antecedent and consequent. This process may lead to a recursive revision in the antecedent model. Finally, if need be,
they set up a scenario relating the model of the consequent to the antecedent model. The relation may be merely that
the consequent state of affairs is relevant to a protagonist in the antecedent model, or it may be a logical, temporal,
causal or deontic relation between the two models.
What are the logical properties of conditionals? They are many and various. Conditionals are not creatures of a
constant hue. Like chameleons, as I once put it, they take on the colour suggested by their surroundings. Their logical
properties depend on the relation between antecedent and consequent, and that in turn depends on beliefs.

Although I can agree with Johnson-Laird in certain respects, I do find him more than a little failing in
his duty to be precise. In particular, to say that the logic of conditionals is like a chameleon, taking its
properties from ‘the relation between antecedent and consequent’ is a little too vague to my taste.
Greater precision is possible and should, therefore, be striven for.
272 The Logic of Language

But let us now look at the similarities. The schema IF P then Q does
not oblige the listener to add either P or Q to D. What it says is that
there are two possibilities: either P is to be incremented, but then also Q, or
else NOT-P is incremented. The question is: what happens when NOT-P is
incremented? Strict-natural and constructed standard logic say that no com-
mitment is entered with regard to Q in case NOT-P is incremented: Q may be
true (incremented) or false (not incremented). But this is not what basic-
natural logic says. Basic-natural propositional logic follows the principle of
natural set theory PNST–3, formulated in Section 3.2.2, and repeated here:
When two (or more) sets A and B undergo union, A and B are natural sets and are, at
the level of basic, but not strict, naturalness, totally distinct, with no element in
common, so that jA [ Bj ¼ jAj þ jBj.

In standard logic, as one knows, IF P then Q is equivalent with IF P then

(P AND Q), just as NOT-P OR Q is equivalent with NOT-P OR (P AND Q).
Intuitively, this makes sense in that, according to the instruction for condi-
tionals given below, IF P then Q tells the listener that the incrementation of
P licenses the incrementation of Q, salva veritate. The normal choice
between incrementing or not incrementing P is thus enriched with the
information that the former option carries Q as an extra bonus—a central
element in inductive thinking.
Basic-natural logic saves the equivalence of IF P then (P AND Q) and
NOT-P OR (P AND Q), but now under the conditions that hold for basic-
natural OR, making it ‘exclusive’. That is, NOT-P OR (P AND) Q is
rephrased as (8.76), according to the expansion scheme of (8.67). This
is so because if the two possibilities mentioned in (P ∧ Q) ∨ ¬P are to be
mutually exclusive for the antecedent clause ¬P and the consequent
clause Q, the possibility of both being true must be excluded. Thus,
exactly as in the case of disjunction discussed in Section 3.4.2, condi-
tionals must be expanded according to the schema formulated in (8.67)
above, that is as (8.76), thereby eliminating the possibility of NOT-P and
Q being true simultaneously. Exclusive OR and bi-implicational IF are
thus of a piece:
(8.76) (P AND Q) OR (NOT-P AND NOT-Q); formally: (P ∧ Q) ∨ (¬P ∧ ¬Q)
This, in effect, makes IF P then Q equivalent with standard logical P $ Q,
just as basic-natural intuition wishes it to be.
 Discourse incrementation 273

Given this, let us now formulate what we see as the incrementation

instruction for the conditional IF-operator at a basic-natural level of opera-
tion. It seems that the following will do:19
INSTRUCTION FOR ‘IF P then q’
For a conditional COND of the form IF P then Q, expand COND to IF P
then Q AND IF Q then P. Then set up two subdomains, one for P AND Q
and one for NOT-P AND NOT-Q. In case P is incremented to D, there is a
licence, ceteris paribus, for Q to be incremented subsequently, and in
case ~P is incremented to D, there is a licence, ceteris paribus, for ~Q to
be incremented subsequently, both salva veritate, whereby each subdo-
main must be properly anchored to the given D.
At a strict-natural level of operation the expansion is limited to IF P then P
AND Q, which is equivalent with standard logical P ! Q and amounts to
saying that once P is incremented there is a licence, ceteris paribus, for Q to be
incremented as well, without further ado.
It is thus clear that at both the basic-natural and the strict-natural level
donkey anaphora is warranted, as in a sentence like:
(8.77) If Nancy has a husband, he is Norwegian.
This so because once ‘Nancy has a husband’ has been incremented, the new
husband-address provides an anchor for the pronoun he.
The incrementation instruction for conditionals implies that it is uncertain
whether P can or must be incremented to D. It thus gives rise to the explicit or
(mostly) implicit question of whether P can be incremented. As such, con-
ditionals are, like disjunctions, a major source for topic–comment modula-
tion discussed in Chapter 11. And, again as with disjunctions, participants in
the linguistic interaction taking place will feel urged to look for arguments
that will decide whether or not P can be incremented to D.
It also implies that both P and NOT-P must be properly anchored to, or, as
has been said, must have the ‘right papers’ for, the D of the moment, while the
‘right papers’ for Q (or, as the case may be, NOT-Q) are predicated on the D of
the moment enriched with P (or, as the case may be, NOT-P). This is
important for a correct assessment of the conditions for presupposition
projection in conditionals, as is explained in Section 10.5.

19
In anticipation of Chapter 10, the negation operator is considered to be the presupposition-
preserving minimal negation and not standard bivalent¬.
274 The Logic of Language

As long as P has not been incremented, the if-clause is held in abeyance, in a

separate nonintensional subdomain. In this respect, the antecedent P may be
regarded as a yes/no-question. Instead of saying If you are ready, come with me,
one may also say Are you ready? (If so/Then) come with me. In many languages, in
fact, conditionals have a grammatical form that equals, or is strongly reminis-
cent of, yes/no-questions.20
Yet what we call if-clauses may also be expressed in a number of different
grammaticalized expression modes. Thus, one sometimes finds and-conjunc-
tions for the expression of conditionals (‘You are right and we must sell
the business’). In the West-European languages this form is mostly
reserved for cases where the antecedent clause is an imperative, as in Do
that and I’ll lock you up. In Turkish, this form is appropriately used, next to
the canonical form of a conditional clause with the suffix -se/-sa, to emphasize
that the antecedent cannot possibly be true, as in Sen yapmadın da kim
yaptı? (You didn’t do it and who did?) (Lewis 1984: 268). Prepositional
(postpositional) phrases under a pre(post)position meaning ‘in the
event of ’, in construction with a nominalized clause, are also found (‘In
the event of your being right, we must sell the business’). This latter mode
of expression for conditionals is found in a fair number of languages, includ-
ing some Papuan languages (Wegener 2008: 267) and Turkish (with the
postposition takdirde; Lewis 1984: 186). Sometimes, the antecedent is in the
imperative, as in Turkish: Uzatma, bırakır giderim (Don’t carry on, I’ll
stop and go) (Lewis 1984: 268). And this list surely does not exhaust the
possibilities available to the human race for the expression of conditional
propositions.
Clearly, a conditional sentence leads to incoherence, or at least to vacuity,
when the current discourse D already contains the information that the
antecedent clause is false. All one can do, in such a case, is use a counterfactual

20
Comrie (1986: 87) mentions examples (i) and (ii), in German and English, respectively, where the
antecedent has the grammatical form of a question. Dutch has the same phenomenon, as shown
in (iii):
(i) Hätte er das getan, wäre ich glücklich gewesen.
(ii) Had he done that, I would have been happy.
(iii) Had hij dat gedaan, was ik gelukkig geweest.
Harris (1986: 276–7) points at the frequently found parallel between conditional clauses and
embedded or indirect questions: ‘We must now look briefly at one quite separate use of SI/si/se in
Romance, namely as the complementizer required when the embedded sentence was originally a
polar question.’ The same phenomenon is found in English, which uses if for both conditional
antecedents and embedded polar questions.
Turkish regularly uses questions, next to other constructions, to express conditionals: O geldi mi,
ben burada durmam (Did he come? I won’t stop here) (Lewis 1984: 267).
 Discourse incrementation 275

construction of the type If you were clever, you wouldn’t buy shares, counter-
factuals being analysed as IF P then Q, with the presupposition that P is false.
But apart from these discourse-related aspects of conditionals, there is a
great deal of explanatory profit to be gained from expanding, in the sense of
(8.76), IF P then Q to IF P then Q AND IF Q then P—that is, from reading the
material implication as the material bi-implication. It is widely known
among formal semanticists (though much less outside these circles) that
the standard logical material implication leads to strongly counterintuitive
results, often called ‘paradoxes’ when applied to natural language condi-
tionals.21 The following analysis shows that most of the ‘paradoxes’ resulting
from the application of the material implication to the semantics of natural
language conditionals simply vanish as soon as natural language if is read as
if and only if.
Just for the sake of clarity, let us set up a truth table for the propositions P,
Q, and R and four relevant compositions in the following way:
TABLE 8.1. Truth table for P, Q, and R and four compositions
valuations: 1 2 3 4 5 6 7 8
P T F T F T F T F VS: {1,3,5,7}
Q T T F F T T F F VS: {1,2,5,6}
R T T T T F F F F VS: {1,2,3,4}
a. (¬P ∧ ¬Q) ∨ (P ∧ Q) or: P$Q T F F T T F F T VS: {1,4,5,8}
b. (¬R ∧ ¬P) ∨ (R ∧ P) or: R$P T F T F F T F T VS: {1,3,6,8}
c. (¬R ∧ ¬Q) ∨ (R ∧ Q) or: R$Q T T F F F F T T VS: {1,2,7,8}
d. (¬(P ∧ R) ∧ ¬Q) ∨ ((P ∧ R) ∧ Q) T F F T F F T T VS: {1,4,7,8}
or: (P ∧ R)$Q

The translation of Table 8.1 into a standard VS-model for the logically and
semantically independent L-propositions P, Q, and R is shown in Figure 8.2.
For the three sentences P, Q, and R, there are eight valuations such that all
possible combinations of truth (T) and falsity (F) are represented. Given the
truth values for P, Q, and R in each column (valuation), the truth values of
their logical compositions follow automatically. I have selected four specific
compositions for reasons that will become clear in a moment.
Composition a: (¬P ∧ ¬Q) ∨ (P ∧ Q) or: P $ Q
representing NOT-P OR Q (or: IF P, then Q, or: IF Q, then P) as construed on
the principles of basic-natural logic, is true in the valuations 1, 4, 5, and 8,

21
An incisive study in this regard is Veltman (1985), to which we refer below.
276 The Logic of Language

/P/

7
3
8 /R/
5 1 4

6 2

/Q/
U

FIGURE 8.2 The valuation spaces of Table 8.1 in a VS-model

giving the valuation space {1,4,5,8}: P and Q are either jointly true or jointly
false.
Composition b: (¬R ∧ ¬P) ∨ (R ∧ P) or: R $ P
representing basic-natural NOT-R OR P (or: IF R, then P, or: IF P, then R), is
true in the valuations 1, 3, 6, and 8, giving the valuation space {1,3,6,8}: R and
P are either jointly true or jointly false.
Composition c: (¬R ∧ ¬Q) ∨ (R ∧ Q) or: R $ Q
representing basic-natural NOT-R OR Q (or: IF R, then Q, or: IF Q, then R), is
true in the valuations 1, 2, 7, and 8, giving the valuation space {1,2,7,8}: R and
Q are either jointly true or jointly false.
Composition d: (¬(P ∧ R) ∧ ¬Q) ∨ ((P ∧ R) ∧ Q) or: (P ∧ R) $ Q
representing basic-natural IF P and R then Q (or: IF Q then P and R), is true
in the valuations 1, 4, 7, and 8, giving the valuation space {1,4,7,8}: either P, Q,
and R are jointly true or P and R are jointly false or Q is false.
It is now easy to read off what entails what. Thus, a ∧ b ‘c (implication is a
transitive relation), because {1,4,5,8} \ {1,3,6,8} ¼ {1,8} and {1,8}  {1,2,7,8}.
But a =‘ d, because {1,4,5,8}  = {1,4,7,8}, which means that the embarrassing
standard-logic entailment from IF P then Q to IF P and R then Q (antecedent
strengthening) no longer holds. We will come back to this in a moment, when
we discuss example (8.84).
 Discourse incrementation 277

This analysis provides a number of solutions to old problems. As has been

noted by many authors, treating the IF-operator as truth-functional material
implication leads to serious clashes with intuition, often referred to by the
embellishing term ‘paradoxes’. These come in two classes. Some are solved
with the help of basic-natural logic, so that implication is read as bi-implica-
tion, but others withstand such a solution. For the latter, a remedy is to be
sought in the discourse-semantic aspects of conditionals. We first consider a
few cases that are solved in terms of the basic-natural logical analysis of
conditionals as bi-implication.
First and obviously, this analysis accounts directly for the intuition that IF P
then Q is naturally, but not in an analytically compelling way, understood as
implying the converse IF Q then P, or, equivalently, as IF NOT-P then NOT-Q,
turning the antecedent clause from a merely sufficient into a sufficient and
necessary condition (the reading P $ Q (composition a) for IF P then Q is
throughout symmetrical for P and Q, and thus also allows for the expression
IF Q then P). This intuition does not reflect analytical necessity, but it clearly
has the weaker status of an invited or default inference, which can be over-
ridden by a modicum of careful thought. This phenomenon is an obvious
target both for pragmatics and for our basic-logic hypothesis. It has given rise
to a rather large body of, mostly pragmaticist, literature over the past thirty or
so years, so much so that it has acquired a special name: since Geis and Zwicky
(1971), this phenomenon has been known as CONDITIONAL PERFECTION, or, in the
words of Horn (2000), the move from IF to IFF.22 It has been shown above that,
like exclusive OR, the much discussed phenomenon of conditional perfection
follows from the principles of basic-natural set theory.
A further ‘paradox’, likewise solved by an appeal to the basic-natural logical
analysis of implication as bi-implication, is the following. Consider the
following pair of sentences:
(8.78) a. If Amsterdam is in Belgium, the present pope is German.
b. If Amsterdam is in Belgium, I am the pope.
Elementary world knowledge tells us that the antecedent clause of (8.78a) is
false but the consequent clause true. This would make the whole of (8.78a)
true, but I am confident that any systematic testing of speakers’ judgements
will reveal that this is not what speakers feel (8.78a) to be. The truth of
(8.78b), where both the antecedent and the consequent (at least if uttered by
someone other than the pope) are false, is easier to accept. (8.78b) can be

22
For a survey of the literature in this respect, see Van Der Auwera (1998) and also Declerck and
Reed (2001). See also Johnson-Laird (1986: 59–60) for an illuminating discussion.
278 The Logic of Language

used to lend strong emphasis to the speaker’s assertion that Amsterdam is in

Belgium is false, because if you accept one obvious falsity, you may as well
accept any.
This discrepancy is explained by the natural tendency to interpret condi-
tionals as biconditionals (‘conditional perfection’). For if (8.78a) were intui-
tively felt to be true, given world knowledge, then (8.79a,b) would also have to
be felt to be true because basic-natural logic allows both for obversion
(contraposition) of a conditional, as in (8.79a), and for the negation of both
clauses, as in (8.79b), as is easily read off from Table 8.1. Both obversion and
negating both clauses leave the truth conditions of IF P then Q unchanged
because all positions of P and Q in composition a are symmetrical and adding
a negation throughout to each elementary sentence likewise changes nothing.
(8.79) a. If the present pope is German, Amsterdam is in Belgium.
b. If Amsterdam is not in Belgium, the present pope is not German.
But both (8.79a) and (8.79b) are clearly false, and felt to be false, in the present
world. Therefore, the alleged truth of (8.78a) clashes with the intuition that
conditionals are ‘really’ biconditionals, since both the negating and the
exchanging of the antecedent and consequent clauses lead to clear-cut falsity.
But when the same test is applied to (8.78b), we see that the strengthening to a
biconditional makes no difference for its truth value as felt intuitively: both
(8.80a) and (8.80b) remain true, also for natural intuition:
(8.80) a. If I am the pope, Amsterdam is in Belgium.
b. If Amsterdam is not in Belgium, I am not the pope.
A further, well-known, ‘paradox’ of this nature is exemplified in (8.81). In
the standard analysis, where IF P then Q is considered false only when P is
true and Q false, and true in all other cases, a sentence of the form NOT(IF P
then Q) amounts to the assertion that P is true and Q false, because when
NOT(IF P then Q) is true, P is true and Q is false. But this is highly
counterintuitive. It would turn a demonstration of the truth of (8.81) into a
proof for the nonexistence of God:
(8.81) It is not so that if God does not exist, there are no moral rules.
Sentence (8.81) sounds quite reasonable, and is, in fact, a central tenet in
humanist moral philosophy. Yet if we apply a standard-logic truth-table test
to (8.81), the result is clearly incompatible with the way it is naturally
understood. Let ‘P’ stand for ‘God exists’ and ‘Q’ for ‘there are moral rules’,
then the logical structure of (8.81) is as shown in (8.82), which shows
 Discourse incrementation 279

that truth is obtained for this formula by assigning falsity (F) to P and truth
(T) to Q.
(8.82)
¬ (¬ P ® ¬ Q)
F T
T ® F
F
T

According to the standard truth table for the material implication, the only
way for ¬P ! ¬Q to be false is for ¬P to be true and for ¬Q to be false, and
hence for P to be false and Q to be true, as in (8.82). Only then can the falsity
of ¬P ! ¬Q be turned into truth by the wide-scope negation. All other truth-
value assignments to P and Q will lead to falsity for the whole formula.
On this, standard, interpretation, (8.81), in fact, expresses the assertion that
God does not exist and that there are moral rules. But this is not the way (8.81)
is understood by normal speakers. No (normal) humanist philosopher will
accept that the truth of (8.81) amounts to saying that God does not exist but
moral rules do. What (8.81) says is, rather, that the nonexistence of God does
not license the claim that there are no moral rules.
Again, an appeal to a basic-natural-logic interpretation helps out. When
translated in terms of exclusive OR, (8.81) reads as follows:
(8.83) It is not so that either God exists and there are moral rules,
or God does not exist and there are no moral rules.
In formal notation, this amounts to: ¬((P ∧ Q) ∨ (¬P ∧ ¬Q)). Now it is no
longer so that the only way for (8.83) to be true is for P to be false and for Q to
be true, as in (8.82). To see this, just take composition a of Table 8.1,
representing IF P, then Q, and invert the truth values. This gives the valuation
space {2,3,6,7}, the complement of {1,4,5,8}. In the valuations 2, 3, 6, and 7, it is
not so that P is always false and Q is always true, as is the case for ¬(¬P ∨ Q).
On the contrary, in the valuations 2, 3, 6, and 7, P and Q never have the same
truth value: if the one is true, the other is false and vice versa. Therefore, all
that is needed for the truth of (8.83), and thus for the basic-natural truth of
(8.81), is that P and Q have different truth values. The ‘paradox’ of (8.81) thus
evaporates under the analysis in terms of basic-natural logic.
A further interesting example, provided by Veltman (1985: 194), is the
following:
280 The Logic of Language

(8.84) a. If you add sugar to your coffee, it tastes good.

b. Ergo: if you add sugar and petrol to your coffee, it tastes good.
In standard logic, a true conditional remains true when the antecedent is
strengthened with further conditions (‘antecedent strengthening’), because the
antecedent merely states a sufficient condition for the truth of the consequent,
not a necessary one. Antecedent strengthening seems to follow from ordinary
reasoning, yet it leads to strongly counterintuitive results. One is inclined to
reason that if it is true that adding sugar to my coffee makes it taste good, then, if
I add both sugar and petrol, I do add sugar, which should thus still make my
coffee taste good. But it doesn’t make my coffee taste good at all.
One immediate answer is that, apparently, IF is not truth-functional but
merely provides a licence to add the consequent clause ‘ceteris paribus’—that
is, without any relevant additional changes in the situation. This is why the
Instruction for IF P then Q, as formulated above on page 273, is predicated on
the ceteris paribus condition.
Apart from this, however, basic-natural logic helps out. Let P be You add
sugar to your coffee, Q The coffee tastes good and R You add petrol to your coffee.
The basic-natural analysis of (8.84a) is then (¬P ∧ ¬Q) ∨ (P ∧ Q), or com-
position a of Table 8.1, with the valuation space {1,4,5,8}. The basic-natural
analysis of (8.84b) is (¬(P ∧ R) ∧ ¬Q) ∨ ((P ∧ R) ∧ Q), or composition d of
Table 8.1, with the valuation space {1,4,7,8}. Clearly, since {1,4,5,8}  = {1,4,7,8},
(8.84a) does not entail (8.84b). Therefore, the annoying problem of anteced-
ent strengthening does not arise in basic-natural logic.
The basic-natural analysis is even better than that. Add to (8.84a) the
statement:
(8.84) c. If you add petrol to your coffee, it does not taste good.
Antecedent strengthening now gives:
(8.84) d. If you add petrol and sugar to your coffee, it does not taste good.
Yet at the same time antecedent strengthening based on (8.84a) gives (8.84b), so
that one must conclude that if one adds both petrol and sugar to one’s coffee it
does and does not taste good at the same time—as long as one adheres to the
standard analysis of IF and thus to antecedent strengthening.
The standard analysis of conditionals fails to give an answer: (8.84a) entails
(8.84b) and (8.84c) entails (8.84d), but nothing follows when (8.84b) and
(8.84d) are asserted simultaneously, even though our intuition warns us that
(8.84b) and (8.84d) are incompatible. In the basic-natural analysis, however,
(8.84b) and (8.84d) turn out to be contradictories. This is quickly shown in
terms of VS-analysis:
 Discourse incrementation 281

(8.84a) /If P then Q/ ¼ {1,4,5,8} (see Table 8.1, comp. a)

(8.84b) /If P and R, then Q/ ¼ {1,4,7,8} (see Table 8.1, comp. d)
(8.84c) /If R then not-Q/ ¼ {3,4,5,6} (check in Table 8.1)
(8.84d) /If P and R, then not-Q/ ¼ {2,3,5,6} (check in Table 8.1)
There are no entailment relations between (8.84a) and (8.84b) or between
(8.84c) and (8.84d)—that is, there is no antecedent strengthening—but
(8.84b) and (8.84d) are both mutually exclusive (contraries) and in full
union (subcontraries), hence contradictories. This is obviously in much
better agreement with natural intuitions than what results from the standard
analysis.
So far, we have solved a number of ‘paradoxes’ by means of the bi-
implicational analysis corresponding with basic-natural logic. There are,
however, other ‘paradoxes’ that resist such a solution. Veltman (1985: 194–5)
provides the following example:
(8.85) If Jones wins the election, Smith will retire to private life.
If Smith dies before the election, Jones will win it.
Ergo: If Smith dies before the election, he will retire to private life.
In standard terms, the logic of this argument is sound, since the relation of
material implication is transitive: if P ! Q and Q ! R, then P ! R. But,
other than in the preceding cases, if the material implication is replaced with
the biconditional, the argument remains logically sound: if P $ Q and Q $
R, then P $ R. Yet the argument is, in fact, absurd. Why should this be so?
Here we need stronger measures than just basic-natural logic. As was said at
the very outset of Section 8.2, the logic is an EMERGENT PROPERTY of the
discourse-semantic incrementation instructions.23 In fact, IF becomes a logi-
cal operator only when discourse factors are left out of account. This implies
that the incrementation instructions are richer than the logic emerging from
them. It is this extra, nonlogical but purely discourse-semantic resource
afforded by the incrementation instructions that we must fall back on now.
The discourse-semantic account of conditionals proposed here implies that
incrementation of Jones wins the election will only license the incrementation
of Smith will retire to private life if the D at hand is not changed in any
significant way by subsequent incrementations, and analogously for the
second premiss of the argument If Smith dies before the election, Jones will
win it. The argument as a whole is valid only if the combined incrementations

23
Cf. Johnson-Laird (1986: 55): ‘A corollary of this theory is that the logical properties of
conditionals derive from their interpretation and not from any formal rules associated with them.’
282 The Logic of Language

of Jones wins the election and Smith dies before the election still licenses the
incrementation of Smith will retire to private life. And this is, of course, not so.
This, by the way, is also the conclusion reached in Veltman (1985). Veltman
argues that the logic of conditionals, to the extent that it can be a natural logic,
is predicated upon a given ‘data base’. In our terms, this means that the logic
of conditionals must be restricted by a Modulo D or ceteris paribus condition.
It would seem that such a condition hardly lends itself to full formalization,
though attempts are made in artificial intelligence to solve the formalization
problem by means of nonmonotonic logics.
Further ‘paradoxes’ resisting a solution in purely logical terms arise in the
context of what is known as CONTRAPOSITION. In standard logic the theorem
holds that whenever P ! Q is true, so is ¬Q ! ¬P and vice versa. In other
words, P ! Q
¬Q ! ¬P. The same theorem holds in basic-natural logic,
because the full expansions of P ! Q and of ¬Q ! ¬P are identical: namely
(¬P ∧ ¬Q) ∨ (P ∧ Q). Yet, as is observed by McCawley (1981: 50), there is a
clear difference between (8.86a) and (8.86b):
(8.86) a. If you don’t have somebody to take my place, I won’t leave.
b. If I leave, you have somebody to take my place.
The first puts a condition on my leaving, while the second expresses a conclusion
that can be drawn post hoc in the event that I leave. McCawley does not say so,
but the difference seems to be connected with the fact, discussed in Section 8.2.1,
that, as a matter of default, discourse incrementation follows the temporal,
causal, or motivational order of the events or situations described. Given the
IF-instruction, this means that the antecedent P is literally an ‘antecedent’: the if-
clause is incremented before the consequent clause. Sentence (8.86a) respects this
principle, since the lack of a replacement is the reason for my not leaving. (8.86b)
does too, but in a different way. What (8.86b) says is that in case I leave one may
conclude that a replacement has been available. Here my leaving precedes the
drawing of the conclusion.24

24
Comrie (1986: 83–4) implicitly confirms the principle that, normally, the if-clause is incremented
before the consequent clause:
Greenberg (1963: 84–5) states the following Universal of Word Order 14 concerning the linear order of the two
clauses:
In conditional statements, the conditional clause [¼protasis, BC] precedes the conclusion [¼apodosis, BC] as the
normal order in all languages.
Work leading up to the present paper has uncovered no counterexamples to this generalization. Although many
languages allow both orders, protasis–apodosis and apodosis–protasis, many grammars note explicitly that the usual
order is for the protasis to precede, and presumably the same will hold for many languages where the grammars are
silent on this point. In some languages the protasis must precede the apodosis, in particular in languages with a rigid
rule requiring the finite verb of the main clause to stand sentence-finally (e.g. Turkish). Since the positioning of
 Discourse incrementation 283

In conclusion, we may say that some headway has been made towards the
realization of the programme of reducing the deviations from the standard
logical analysis of conditionals observed in ordinary language usage to the
differences between standard and basic-natural logic and to the design prop-
erties of discourse domains, including the instruction for conditionals. To the
extent that this programme is successful, an appeal to mostly not very well
defined pragmatic principles becomes unnecessary.

protases in such languages can be viewed as just a special case of the general rule whereby subordinate clauses must
precede main clauses, this does not necessarily say anything specific about conditional constructions. However, this
same restriction to protasis-apodosis order is also found in some languages which do not have a strict subordinate–
main clause order restriction, suggesting that there is indeed something special about conditional clauses in this
respect, i.e. the preponderance of the protasis–apodosis order in languages with free clause order is not ‘just
statistical’, but does reflect something significant about language.
 9

Primary and donkey anaphora

9.1 Introduction
Apart from the continuous appeal to shared knowledge, there are in principle
four main devices in natural language serving the purpose of linking up
utterances with the current discourse in such a way that coherence is safe-
guarded: anaphora, presupposition, topic–comment modulation, and open
parameters in lexical meanings. The latter have been discussed in Section 7.4
and we will not deal with them again. The remaining three devices are
relevant in this and the following two chapters because they show up the
inadequacy of the Russellian-Quinean-Montagovian paradigm of natural
language semantics, with standard modern predicate calculus (SMPC) at its
centre. Even though the standard paradigm is meant to account for just the
truth-conditional properties of natural language sentences, leaving out the
clutter due to the exigencies of communicative usage, it breaks down on a
particular form of external anaphora, commonly called donkey anaphora,
discussed in the present chapter. It also breaks down on presupposition, as
is shown in Chapter 10. And one draws a total blank, not only in semantics
but also in the theory of grammar, when one looks for any account at all of
topic–comment modulation as distinct from predicate–argument structure.
Topic–comment modulation, already amply commented upon in Section 3.2
of Volume I, is taken up again in Chapter 11, where its role in the cementing of
discourse coherence is further elaborated and situated in a more general
cognitive and linguistic context.
The last three chapters of this book are thus a three-pronged attack on the
standard paradigm. But the standard paradigm has to cope with other threats
as well, less to do with the context-sensitivity of natural language and the
criteria for textual coherence. The main threat, apart from the three discussed
in Chapters 9, 10, and 11, is the problem of propositional attitudes, discussed
earlier in Sections 2.1.1 and 6.1 of Volume I. The inability of the standard
paradigm to account for this problem has proved to be due to the very basic
tenets of this paradigm, in particular its strictly extensional house ontology,
 Primary and donkey anaphora 285

which is incapable of dealing with intensional phenomena. Here, too, the

assumption of a cognitive discourse domain, storing all information of any
discourse at hand regarding actual or virtual entities and facts, provides the
theoretical space required for an adequate solution.
This critique of the standard paradigm does not come out of the blue. Over
the past twenty or thirty years there has been a growing awareness of its
insufficiency, the emphasis being variably on one or more of the issues
mentioned above. One may say that a movement has been gaining ground
that no longer supports the rarified, logic-inspired view of natural language
and its semantics but aims at developing a more ecological perspective, hard
though that proves to be.
During the 1980s, a spate of publications saw the light, all trying to develop
a context-sensitive semantics for natural language sentences by means of the
introduction of a cognitive intermediary station—called data base, discourse
domain, discourse representation, mental space, whatever—housing mental
representations of anything spoken about, whether actually existing or only
virtual. In these theories, reference is no longer seen as some sort of extrasen-
sory, ectoplasmic link, established by some Kripkean flight of metaphysics
between definite terms occurring in utterances on the one hand and world
objects on the other. Instead, the somewhat more reasonable view is taken
according to which definite expressions establish reference via mental repre-
sentations of entities of whatever kind, our ‘addresses’, set up earlier in the
discourse by means of existential quantification. What mechanism accounts
for the intentional relation between the addresses and the objects focused on
is still largely a mystery, but at least it is a general mystery—the mystery of
what is known as intentionality—affecting all existing theories of the mind.
One important consequence of this new perspective is that definite terms in
sentences are now seen as being directed to discourse addresses before any
possible relation to really existing world entities can be established. In practi-
cal terms this means that the Russell-Quine programme of elimination of
singular terms has been given up: referring expressions have been reinstated,
after almost a century of exile. The king of France has again become the king
of France, even though he does not exist. The referential problem created by
his nonexistence is taken to be solvable, in principle, by the introduction of
the intermediary station of a discourse domain.
Seuren (1972, 1975), Isard (1975), and Stalnaker (1978) were early
harbingers, setting out the perspective of a store accumulating the information
carried by successive utterances in a discourse and thus being added to by
every new utterance. The first more formally precise approach was DISCOURSE
REPRESENTATION THEORY (DRT), introduced in Kamp (1981) and further
286 The Logic of Language

elaborated in Kamp and Reyle (1993). The roots of this theory lie in the donkey-
anaphora problem, discussed in Section 9.2.2, which is correctly diagnosed as
potentially fatal for the established paradigm of possible-world formal seman-
tics. In this theory, the mechanism of reference is mediated by a cognitive
system of mental representations, whose relation to any actual world is a matter
of independent concern. The discourse representations envisaged by Kamp and
Reyle not only contain entity representations but also store any propositional
information about the intended referents provided by prior linguistic input.
This halfway station of mental representations creates the extra room needed
for a semantic account of donkey anaphora. DRT is not a logical theory but a
formal theory of utterance incrementation, even though the format in which
newly incremented information is represented looks very much like the well-
known structure of SMPC expressions. The actual corresponding logic has been
investigated by Groenendijk and Stokhof in a number of joint publications,
notably their 1991 paper on ‘Dynamic Predicate Logic’. The main difficulty with
DRT is that its focus is too much on the donkey-anaphora problem, leaving
entirely out of account the notion of presupposition, let alone that of topic–
comment modulation, indispensable though these are for any adequate theory
of discourse incrementation. Nor does it offer a principled solution to the
problem of reference to nonexisting entities: there is no theory of virtual or
intensional entities. In general, one may say that DRT is a typical example of a
theory that has been fully formalized before the object of the theory has been
looked at from all angles so that one’s familiarity with the object gives one an
adequate idea of what is to be formalized. Such premature formalization may
impress the world for some time, but it is detrimental to the advancement of
lasting knowledge.
Practitioners of DRT have not been insensitive to the criticism that
their theory fails to account for presuppositions. Van der Sandt (1992) made
an attempt at incorporating presupposition theory into the anaphora-based
framework of DRT by equating presupposition with anaphora. This made it
look as if the presupposition deficiency of DRT could be remedied in one
swoop by declaring anaphora and presupposition one—a point of view
readily taken over by DRT-practitioners who were all too eager to dispose
of presuppositions, which they had all along regarded with diffidence
and suspicion and which they were only too glad to get rid of. In fact, in
the short section on presupposition in the article on DRT in the Stanford
Encyclopedia of Philosophy (Geurts and Beaver 2007), the authors take it for
granted that, indeed, presupposition and anaphora are of a piece. Since this
misunderstanding has meanwhile gained some currency, a separate section
(Section 10.8) has been added to Chapter 10, dedicated to its refutation. There
 Primary and donkey anaphora 287

it is shown that this prima facie absurd view is sustainable only if the notions
of presupposition and of anaphora are kept fuzzy and essential facts are
ignored.
The SITUATIONAL SEMANTICS of Barwise and Perry (1983) is another represen-
tative of this class of theories. This theory sprouts mainly from the authors’
dissatisfaction with two aspects of the standard paradigm, namely its inability
to cope with propositional attitudes (see Sections 2.1.1 and 6.1 in Volume I)
and the lack of a proper demarcation of the contextually and situationally
restricted universe of discourse in terms of which utterance interpretation
takes place. No attention is paid to the donkey-anaphora problem, nor is
there an account of reference to nonexisting entities. Presuppositions likewise
stay out of the picture, even though their function in the delimitation of
restricted universes of discourse should have been obvious. The question of
topic–comment modulation remains untouched.
In Fauconnier (1985) a system of ‘mental spaces’, complete with ‘subspaces’,
is proposed in order to account for what is called ‘transdominial denotational
transparency’ in Section 7.2.2.1. This highly readable but entirely informal
little book concentrates on examples like The girl with brown eyes has blue eyes
(see also Section 5.3 in Volume I). Contrary to what one might expect, this
sentence is not internally inconsistent. Suppose there is a picture containing
the portraits of a number of girls, one of whom is portrayed as having brown
eyes, even though in reality she has blue eyes. In such a situation, the sentence
is true. The expression the girl with brown eyes then refers to the girl as
portrayed in the picture, which is mentally represented as a special subspace.
The predication has blue eyes takes the listener back to the real world, mentally
represented as the overall commitment domain. But the same sentence may
also be taken as saying that the girl who in reality has brown eyes has blue eyes
in the picture. Which of the two readings applies, depends on the context in
which the sentence is uttered. This essay is illuminating in many ways but its
scope is too restricted, and its elaboration too informal, for it to qualify as a
theory of context-sensitive utterance interpretation.
All three theories or approaches mentioned thus lack the generality that is
required for a proper theory of context-sensitive utterance interpretation.
I myself started out on this road as early as 1972, with my lengthy paper (1972)
in Leuvense Bijdragen, which contains a fairly elaborate section on discourse
incrementation and presupposition as a condition on incrementability.
I believe this was the first publication, even preceding Isard (1975), where
the notion of discourse incrementation was mooted (apart from Stout 1896,
mentioned and quoted extensively in Section 3.2 in Volume I). Yet it was
never acknowledged, then or later, though many protagonists of the new
288 The Logic of Language

contextually oriented semantics were Dutch and thus had no difficulty

reading it. This was followed by Seuren (1975), which was again written in
Dutch, again contained lengthy sections on discourse semantics and was
again left unacknowledged by the Dutch establishment, though not by the
establishments in neighbouring countries. In 1979, I presented a paper at the
Mannheim Institut für deutsche Sprache on the three-valued logic which I take
to be required for presupposition theory, published as Seuren (1980). Hans
Kamp, whose seminal article appeared in 1981, had been invited to reply to
this paper. Then came my 1985 book Discourse Semantics, where a fairly
elaborate theory of discourse incrementation is presented, complete with
the notion of address closure discussed in Section 8.1.1. In that book, the
basis of the theory of discourse semantics is widened to include not only its
original starting point of presupposition theory but also donkey anaphora
and topic–comment modulation, as well as the question of propositional
attitudes. The present book may be seen in part as a sequel to my Discourse
Semantics of 1985, in that it attempts to lay as broad a basis as possible for the
new theory.

9.2 Reference by anaphora

Let us start with the problem of donkey anaphora, which, however, needs
some introduction. The word anaphora is Greek and it means literally
‘pick-up’ or ‘uptake’. What we mean by anaphora in linguistics is, roughly,
the phenomenon that certain nominal expressions exercise their referential
power via another expression or element in the discourse domain, called
the antecedent, which is either a directly referring nominal expression or
a quantifier or, in cases of what we call ‘primary anaphora’, the open address
established by a previous existentially quantified sentence. Anaphoric expres-
sions are usually third person personal or possessive pronouns, but they
may also be full lexical noun phrases, such as the English phrases the former,
the latter, or the so-called ‘epithet’ pronouns, about which more in a
moment, or the indirectly referring noun phrases serving as ‘enriched’
pronouns illustrated in the Swiss banker example (7.23) of Section 7.2.2.4.
Many languages, such as the languages of Europe, have number and gender
distinctions for their anaphoric pronouns—whereby gender corresponds
variably to either grammatical or natural gender—to facilitate anaphoric
uptake, but not all languages do. Creole languages, for example, tend to
do without gender distinctions and Malay even does without number
distinctions.
 Primary and donkey anaphora 289

There is a vast, but not entirely conclusive, literature that attempts to

establish rules and principles for the uptake function of anaphoric
expressions, called anaphora resolution. The commonly accepted view is
that a primary distinction must be made between sentence-internal and
sentence-external anaphora, depending on whether the antecedent is to
be located within the same sentence or outside it. This view is largely,
but not entirely, correct. It is true that sentence-internal anaphora is
subject to structural conditions in a way that external anaphora is not—a
difference that is reflected in the fact that it is mostly grammarians who
write on internal anaphora resolution, whereas the literature on external
anaphora is largely psycholinguistic and/or pragmatic.1 But the two
cannot be separated too strictly, as the following will make clear.
Let us, for a moment, look at sentence-internal anaphora, which is subject
to sentence-internal structurally definable restrictions. First consider clause-
internal anaphora, as in (9.1a–d). One notes that in (9.1a,b) the possessive
pronoun his is ambiguous between a reading in which the office belongs to
John (the internal reading) and one in which it belongs to some other person
who is not mentioned (the external reading). But in (9.1c,d), the anaphoric
pronoun he must refer to an external person and cannot take the noun phrase
John as its antecedent:
(9.1) a. In his office John reads detective stories.
b. John reads detective stories in his office.
c. In John’s office he reads detective stories.
d. He reads detective stories in John’s office.
The least one can deduce from this is that there are structural restrictions
on the choice of antecedents for anaphoric uptake. That these are not trivial
transpires when one looks at clause-external but still sentence-internal anaph-
ora, as in (9.2a–d). Here, both the internal and the external reading are
possible for (9.2a,b,c) but not for (9.2d), which only allows for the external
reading:
(9.2) a. While his office was being cleaned, John stood on the balcony.
b. John stood on the balcony while his office was being cleaned.
c. While John’s office was being cleaned, he stood on the balcony.
d. He stood on the balcony while John’s office was being cleaned.

1 For detailed discussions of sentence-internal anaphora, see Reinhart (1983), Seuren (1985: 346–86;
1986b) and Weijters (1989).
290 The Logic of Language

The odd one out is the c-variant in (9.1) and (9.2): (9.1c) only allows for the
external reading, but (9.2c), like (9.2a,b), allows for both the internal and the
external reading.
The matter becomes even more intriguing when EPITHET PRONOUNS are taken
into account (see also the examples in (4.41) of Chapter 4 in Volume I). These
are, grammatically speaking, not really pronouns but full lexical noun
phrases, though with an anaphoric function. They are always unstressed
and usually express an evaluation of some kind. Examples are the following
(the epithet pronouns are in italics):
(9.3) a. Where is John? I just saw the great genius leaving the building.
b. As John entered the room, the maniac saw it was empty.
Now consider again (9.1a–d), but with the epithet pronoun the fool instead
of the neutral his or he:
(9.4) a. In the fool’s office John reads detective stories.
b. John reads detective stories in the fool’s office.
c. In John’s office the fool reads detective stories.
d. The fool reads detective stories in John’s office.
One sees that the internal reading has disappeared altogether: all four sen-
tences only allow for the external reading. However, the same does not hold
for the epithet analogs of (9.2):
(9.5) a. While the fool’s office was being cleaned, John stood on the balcony.
b. John stood on the balcony while the fool’s office was being cleaned.
c. While John’s office was being cleaned, the fool stood on the balcony.
d. The fool stood on the balcony while John’s office was being cleaned.
Here, as in (9.2), the internal reading is allowed for (9.5a,b,c) but not for
(9.5d).
The epithet-substitution test is a useful but neglected diagnostic in anaph-
ora theory. It shows, for example, that there must be some hidden difference
between the status of his in (9.1a,b) as opposed to (9.2a,b). This difference is
expressed formally in some languages, like Latin or Swedish, which both
distinguish between a reflexive and a nonreflexive third person possessive
pronoun. In Latin, the reflexive possessive pronoun is the adjectivally de-
clined suus, while the nonreflexive variant is expressed in the singular by
the genitive eius (of him/her/it) of the third person personal pronoun is
(this, he) and in the plural by eorum (of them). Analogously, Swedish
distinguishes between the adjectival sin and the personal pronoun genitive
hans, the former being the reflexive, the latter the nonreflexive his. In both
 Primary and donkey anaphora 291

Latin and Swedish, the translations of (9.1a,b) differentiate between

the internal reading, which has suus or sin, and the external reading, which
has eius or hans, respectively. By contrast, (9.2a,b) are ungrammatical in
Latin and Swedish with suus or sin, but grammatical (and ambiguous) with
eius or hans, respectively.
A similar phenomenon crops up in that-clauses and dependent questions:
(9.6) a. John told me that he was unable to attend the meeting.
b. John told me that the great hero was unable to attend the meeting.
c. John asked whether it was all right for him to leave.
d. John asked whether it was all right for the poor sod to leave.
Again one sees that insertion of an epithet pronoun blocks the internal
reading.
Pronouns occurring in a position where they are bound by a quantifier
never allow for substitution salva resolutione by an epithet pronoun:
(9.7) a. Somebody must have been thinking that he would win.
b. Somebody must have been thinking that the brute would win.
c. In those days, every husband expected his wife to obey him.
d. In those days, every husband expected his wife to obey the brute.
Clearly, the expression the brute in (9.7b) and (9.7d) blocks the bound-
variable reading of him in (9.7a) and (9.7c), respectively, thus leaving the
external anaphoric reading as the only possible one.
Sentence-external anaphora is, on the whole, less subject to structural
restrictions than to considerations of distance and pragmatic plausibility.
It is also the form of anaphora that plays a major role in the machinery
that links up utterances with the discourse and the situation at hand.
Inevitably, therefore, sentence-external anaphora is more relevant, for
the present purpose, than sentence-internal anaphora. Yet sentence-
internal anaphora cannot be ignored, because there is an issue in the litera-
ture as to the status of certain pronouns that cannot be rendered in
SMPC, the so-called donkey pronouns. According to some, these should
be treated as sentence-internal pronouns, but it is argued here that they
represent a particular form of sentence-external anaphora called primary
anaphora.
In order to discuss these issues properly, we need to be able to fall back on
at least a summary survey of the restrictions valid for sentence-internal
anaphora. For that reason, we posit the following tentative and incomplete
list of anaphor categories and their resolution conditions:
292 The Logic of Language

I Clause-internal anaphora.
a. Reflexive anaphora: takes subject (sometimes indirect object) as ante-
cedent; not always formally marked.
b. C-command anaphora: anaphor must be C-commanded by anteced-
ent.2
c. Non-C-command anaphora: antecedent must precede anaphor.
II Clause-external but sentence-internal anaphora.
a. Indirect reflexive anaphora: occurs only in complement clauses and
anaphor takes subject, direct or indirect object of commanding clause
as antecedent; rarely formally marked.
b. Nonreflexive anaphora: occurs in clauses of any rank and anaphor takes
any NP in any other clause as antecedent, but if antecedent is in a lower
clause, it must precede the anaphor.3
III Bound-variable anaphora. The anaphor stands for a bound variable in the
semantic analysis (SA) of the sentence in question and is, therefore,
subject to the structural conditions of variable binding in LL, not in
surface structure.
IV Sentence-external anaphora. The antecedent is any overt or implicitly
understood NP in preceding text or in the situation given. Anaphora
resolution is subject to gradable criteria of closeness, pragmatic
probability, and syntactic function, besides, of course, to restrictions
imposed by gender and number, if any.4,5

2 C-command (Reinhart 1983) is a tree-structural notion defined as follows:

A node A C-commands a node B just in case the first node up from A dominates B and A does not
dominate B and B does not dominate A.
There is C-command anaphora in (i) but not in (ii); (iii) is probably a case of formally unmarked
reflexive anaphora (‘*’ stands for ‘epithet substitution not permitted’:
(i) She told John about his (*the bugger’s) parents.
(ii) She told John’s daughter about his (the bugger’s) parents.
(iii) His (*the bugger’s) mother told John to stop laughing.
3 Hence the difference between (9.2a,b,c) on the one hand and (9.2d) on the other. In (9.2a,b,c),
John can be the antecedent of his, but in (9.2d) it cannot, because John occurs in a lower clause but does
not precede his.
4 It is pointed out in Tasmowski-de Rijck and Verluyten (1982) that external anaphora to objects
given in the situation but not mentioned in previous text is subject to gender restrictions that follow
the grammatical gender of the noun that would have been used had the object actually been named, as
is shown by the following examples, both meaning ‘you will never get it through the door’:
(i) Tu ne le feras jamais passer par la porte. (said of a desk, French: le bureau)
(ii) Tu ne la feras jamais passer par la porte. (said of a table, French: la table)
5 An extremely interesting category of external primary anaphora was discovered by Moxey and
Sanford (1986/7) (see also Sanford et al. 2007). They found that a large majority of test subjects, who
 Primary and donkey anaphora 293

We stipulate, furthermore, that epithet anaphora is possible, salva resolutione,

only for the categories Ic, IIb, and IV. These can be grouped together as
unbound anaphora; the remaining categories can be grouped as bound anaph-
ora. Whereas bound anaphora is subject to a variety of structurally defined
forms of binding, the main, and perhaps even necessary, condition on un-
bound anaphora is antecedent precedence. Cases of a sentence-external ‘an-
tecedent’ following later in the text are extremely rare, as opposed to bound
anaphora, where this phenomenon is sometimes called by the name of
cataphora.
Having said this, we now pass on to primary anaphora, which has been
our target all along: all that has been said about anaphora so far was only said
because we needed to get at primary anaphora, a central but hitherto unsolved
problem in the semantics and the logic of natural language.

9.3 Primary anaphora: bound variable or external anaphor?

Primary anaphora is a particular form of sentence-external anaphora, namely
that form of anaphora where the antecedent is not a referring expression or a
universal quantifier but an existentially introduced address. A primary ana-
phor usually occurs right after the existential introduction of its antecedent,
which, again, is not a linguistic element in preceding text but the newly
introduced address. Primary anaphora has caused a great deal of confusion
owing to the wish of certain prominent investigators to treat primary
anaphors as bound-variable pronouns. It is argued here that this is misguided.
Primary anaphora is a category in its own right.
A typical example of primary anaphora is given in sentence (9.8), where the
pronoun serving as primary anaphor has been printed in italics:
(9.8) Socrates owns a donkey and he feeds it.
I have deliberately used a ‘donkey’ example, to make it clear that cases of
primary anaphora are of a piece with the well-known cases of donkey

were given the task of completing sentences like (i) interpreted the pronoun they as referring to the
members who were not at the meeting—a form of anaphora for which they invented the term
complement anaphora:
(i) Few students were at the meeting. They …
The subjects would, for example, complete the sentence as They had gone out with their girl-friends.
The only way to explain this seems to be the assumption of a discourse address established for the
students who were not at the meeting, the complement of the set delimited by FEW. If anything, this
shows the necessity to fall back on cognition-driven discourse incrementation processes for the
explanation of reference fixing.
294 The Logic of Language

anaphora. But I might as well have adopted the example There was a cat; it ran
away discussed in Section 8.1.1 in connection with the procedure of address
closure, which, as we shall see, is essential in the present context.
Why is primary anaphora important? It is important in its own right
because it is a central instrument for maintaining coherence in discourse,
but it is also important because it constitutes a problem for the logical analysis
of natural language sentences that has hitherto not found a final solution. It
turns out that established logic lacks the means to account for primary
anaphora and that this failure is entirely due to its decision to keep occasion
sentences away from the analysis. In other words, standard modern logic trips
over primary anaphora and it does so precisely because it fails to make room
for context-sensitivity.

9.4 Donkey sentences

The problem of donkey anaphora is extremely well known and has been so
since the publication of Peter Geach’s 1962 book Reference and Generality.
Owing to this book, the problem has acquired its curious name ‘donkey-
sentences problem’, since the relevant examples in Geach (1962) all involve
mention of donkeys. We shall see in a moment that Geach was less than frank
about his sources, but before that, let us see what the problem actually
amounts to.

9.4.1 The problem

In terms of modern logic, the problem is that the status of it in a
sentence like (9.8) is unclear. In (first-order) SMPC, a term in a propo-
sition that has a truth value must either be an expression referring to an
actually existing entity or set of entities, or be a variable bound by a
quantifier and ranging over such entities. In fact, SMPC went further and
adopted Quine’s ‘programme of elimination of singular terms’, trying to
avoid the problem of reference to nonexisting entities by reducing all
definite descriptions to quantified terms. A systematic application of this
programme should, in Quine’s eyes, provide answers to all the many
problems connected with reference. In his Word and Object (1960: 181–6),
Quine supported and generalized Russell’s analysis of definite terms,
proposing that the regimented ‘canonical’ form of sentences, which dis-
plays just their logical properties without the clutter accrued from the
impure conditions of usage, should contain no referring expressions at
all. All statements about the world could then be expressed with the help
 Primary and donkey anaphora 295

of the two standard quantifiers and propositional functions containing

the variables bound by them. In principle, this leaves SMPC sentences
with only one kind of argument term: variables that are bound by a
quantifier. Hence Quine’s slogan: ‘To be is to be the value of a variable’
(Quine 1953: 13, 15). The problem is that the so-called donkey anaphors
do not fit into this theory.
Consider it in sentence (9.8). If this it is a referring expression and,
therefore, (9.8) has the logical form ‘P ∧ Q’, there are two problems. First,
in case P is false, the it in Q fails to refer, so that Q remains without a truth
value, which is against the rules. Secondly, suppose Socrates has two donkeys,
one of which he feeds while the other he does not, then (9.8) is true, but so is
(9.9a), analysed as (9.9b):
(9.9) a. Socrates owns a donkey and he does not feed it.
b. P ∧ ¬Q
And this is intolerable, as propositional logic tells us that P ∧ Q and P ∧ ¬Q
are contraries. It thus looks as if the it of sentence (9.8), or, for that matter, of
sentence (9.9a), cannot be a referring term and must, therefore, be a bound
variable.
This is, in fact, what is proposed in Geach (1972: 115–27). There, Geach
analyses (9.8) and (9.9a) as (9.10a) and (9.10b), respectively (using the Rus-
sellian and not the generalized quantifiers):
(9.10) a. ∃x[Donkey(x) ∧ Own(Socrates,x) ∧ Feed(Socrates,x)]
b. ∃x[Donkey(x) ∧ Own(Socrates,x) ∧ ¬Feed(Socrates,x)]
Yet, as was pointed out in Seuren (1977, 2001: 316–18), Geach failed to consider
sentences like (9.11a,b), with each conjunct under an independent higher
operator:
(9.11) a. Socrates must own a donkey and he may feed it.
b. I believe that Socrates owns a donkey and I hope that he feeds it.
No binding of it under a quantifier, as in (9.10a,b), is possible here. For either
the quantifier must have large, overall scope for a donkey, turning (9.11a,b)
into sentences about a specific donkey that Socrates must own and may feed,
or that I believe Socrates owns and I hope he feeds. But this is not what these
sentences mean. Or else, to ensure proper binding, the second operator (may,
hope) must be placed in the scope of the first (must, believe), in something like
the following way:
296 The Logic of Language

(9.12) a. MUST[∃x[Donkey(x) ∧ Own(Socrates,x) ∧ MAY[Feed

(Socrates,x)]]]
b. Believe(I,∃x[Donkey(x) ∧ Own(Socrates,x) ∧ Hope(I,
Feed(Socrates,x))])
Or, in the language of decent citizens:
(9.13) a. It is necessary that Socrates owns a donkey and that he may feed it.
b. I believe that Socrates owns a donkey and that I hope that he
feeds it.
It is obvious that (9.13a,b) do not mean what (9.11a,b) do. In fact, what
(9.13a,b) mean is a little quaint, to say the least. What one would like to see
is an analysis of, indeed, the form P ∧ Q, but with a stranded variable in Q for
the pronoun it. But stranded variables are anathema in modern predicate
logic. It thus appears that Geach’s analysis comes to nothing and, worse, that
the it in question constitutes an embarrassment for SMPC.
The cases with intervening operators, such as (9.11a,b), have already been
taken care of in Section 8.1.1, where the sentence, here repeated as (9.14), is
discussed:
(9.14) John claims that he owns a Ferrari, but I have never seen it.
The mechanism set out in Section 8.1.1 sets up an open intensional address for
the specific but not necessarily actually existing Ferrari John claims he owns.
Owing to the fact that the existential quantifier has been robbed of existential
import, this address does not entail that it is an actually existing Ferrari that is
at issue. After all, the Ferrari may well be, and probably is, nothing more than
a virtual object. This address can then be fallen back on for closure and
definite reference. The donkey pronouns it in (9.11a,b) or (9.14), and all
other donkey pronouns, are, in my theory, to be analysed as external definite
anaphoric pronouns that take an earlier established address as their anteced-
ent. According to the Principle of Maximal Unity of discourse domains
(Section 7.2.2), there are no domain restrictions on pronominal anaphora:
all (sub)domains are anaphorically accessible from all (sub)domains.
A second reason why Geach’s attempt at solving this problem does not
work, besides the intervening operators as in (9.11), is provided by the epithet-
substitution test. In the following sentences, the anaphora resolution remains
unchanged, which shows that the anaphoric expressions cannot stand for
bound variables but must be cases of external anaphora and thus belong to
category IV:
 Primary and donkey anaphora 297

(9.15) a. Socrates owns a donkey and he feeds the animal.

b. Socrates must own a donkey, and he may feed the animal.
c. I believe that Socrates owns a donkey, and I hope that he feeds the
animal.
But this is not all. In formal semantics, the problem of donkey sentences
started out with sentence types different from the simple conjunctions of
(9.8) or (9.9a). The standard types of donkey sentence are exemplified in
(9.16a,b,c):
(9.16) a. Either Socrates does not own a donkey or he feeds it.
b. If Socrates owns a donkey, he feeds it.
c. Every farmer who owns a donkey feeds it.
Here, the same problem crops up. In standard model-theoretic semantics,
based on SMPC and standard propositional logic, these sentences must be
considered true in worlds not containing any donkey at all. As regards (9.16a),
Socrates does not own a donkey is true in such a world, which should make
(9.16a) true were it not for the fact that the pronoun it in the second disjunct
he feeds it cannot be subjected to the Russell treatment for definite descrip-
tions because it isn’t a definite description but a pronoun. It must, therefore,
represent a bound variable.
Likewise for (9.16b). In a world without any donkeys, Socrates owns a
donkey is false, which should make (9.16b) true in such a world. But again,
the pronoun it in the consequent clause resists Russell’s treatment, so that it
must represent a bound variable. Similarly again for (9.16c). In a world
without donkeys there can be no donkey-owners, which should make
(9.16c) true in such a world. Again, however, a logical translation of (9.16c)
requires that the pronoun it be taken to represent a bound variable. However,
as is shown in (9.17a–c), the normal and most straightforward translation
method of natural language sentences into SMPC leads to trouble because it
refuses to be bound (the ‘asinus’ symbol @ stands for the term that causes the
binding trouble):
(9.17) a. ¬∃x[Donkey(x) ∧ Own(Socrates,x)] ∨ Feed(Socrates, @)
b. ∃x[Donkey(x) ∧ Own(Socrates,x)] ! Feed(Socrates, @)
c. 8y[Farmer(y) ∧ ∃x[Donkey(x) ∧ Own(y,x)] ! Feed (y, @)]
The problem is, obviously, that the variable x, which one would like to see in
the position of the asinus @ in (9.17a–c), is not allowed to occur there because
it falls outside the scope of any of the quantifiers involved.
298 The Logic of Language

And, again, the analyses given in (9.17a–c) fail to satisfy the epithet-
substitution test, which shows that the it-pronouns in question cannot be
bound-variable pronouns but must be instances of either clause-external but
sentence-internal anaphora (category IIb), or sentence-external, in particular
primary, anaphora (category IV):
(9.18) a. Either Socrates does not own a donkey or he feeds the animal.
b. If Socrates owns a donkey, he feeds the animal.
c. Every farmer who owns a donkey feeds the animal.
Quine (1960) shows no awareness of the donkey-anaphora problem. He
does, however, deal with a similar problem posed by sentences of the type
(1960: 138):
(9.19) If any member contributes, he gets a poppy.
If the word any is taken to represent the existential quantifier, the pronoun he
is left stranded, or, as Quine says (1960: 139), ‘left high and dry’:
(9.20) ∃x[Member(x) ∧ Contribute(x)] ! ∃y[Poppy(y) ∧ Get(@,y)]
Quine then proposes not to use the existential quantifier and get the universal
quantifier to do all the work, stipulating that ‘by a simple and irreducible trait
of English usage’, every always takes the smallest and any the largest possible
scope. (9.19) would then translate into SMPC as (21):
(9.21) 8x[[Member(x) ∧ Contribute(x)] ! ∃y[Poppy(y) ∧ Get(x,y)]]
This proposal was used later in an attempt to solve the problem of the
stranded variables in cases like (9.17a–c). The idea was to translate (9.16a–c)
with the help of the universal quantifier only:
(9.22) a. 8x[Donkey(x) ! [¬Own(Socrates,x) ∨ Feed(Socrates,x)]]
b. 8x[Donkey(x) ! [Own(Socrates,x) ! Feed(Socrates,x)]]
c. 8x8y[[Donkey(x) ∧ Farmer(y) ∧ Own(y,x)] ! Feed (y,x)]
This does indeed eliminate the scope problem raised by (9.17a–c). Yet the
medicine has proved worse than the ailment. First, one wonders why natural
language chooses to use, clearly without any problem for natural interpreta-
tion processes, surface-structure representatives of the existential quantifier (a
donkey, no donkey), allowing unbound variables to dangle, instead of the
perfectly available surface-structure representatives of the universal quantifier,
if that is the quantifier used in the underlying logico-semantic structure. In
other words, one wonders what could justify the sudden change in the
translation or mapping relation between the logico-semantic and the
 Primary and donkey anaphora 299

grammatical form of the sentences involved. A Quinean appeal to ‘simple and

irreducible traits of English usage’ obviously won’t do.
Secondly, Quine failed to mention that in a sentence like If some member
contributes, he gets a poppy the same scope problem occurs for it as in If any
member contributes, he gets a poppy. One wonders if he would have shrunk
from putting forward the daring proposal to treat the word some as a
representative of the universal quantifier.
Thirdly, the epithet-substitution test shows that he in (9.19) cannot repre-
sent a bound variable. And finally, and most seriously, Quine’s universal
quantifier ploy falls again foul of possible intervening operators, as is shown
by the following sentences:
(9.23) a. Either Socrates no longer owns a donkey or he still feeds it.
b. If it’s bad that Socrates owns a donkey, it’s good that he feeds it.
c. Every farmer who is thought to own a donkey is expected to feed it.
The only way of binding it under the spurious universal quantifier is to place
the quantifier over the whole structure, as in (9.24a,b,c), respectively.
(9.24) a. 8x[Donkey(x) ! [NO LONGER[Own(Socrates,x)] ∨ STILL
[Feed(Socrates,x)]]]
b. 8x[Donkey(x) ! [BAD[Own(Socrates,x)] ! GOOD[Feed
(Socrates,x)]]]
c. 8x8y[[Donkey(x) ∧ Farmer(y) ∧ Thought[Own(y,x)]] !
Expected[Feed (y,x)]]
Yet (9.24a,b,c) are incompatible with what the sentences in question mean.
(9.24a) is true, in SMPC, when there are no donkeys at all, and also, if there
are donkeys, when Socrates still feeds them all even if he has never owned any
of them. But (9.23a) cannot be considered true in such a case.
Likewise for (9.24b), which fails because it is gratuitously true (in SMPC) if
there are no donkeys, whereas (9.23b) cannot be true in such a case. (Since
both operators it’s bad and it’s good are factive, (9.23b) presupposes that
Socrates owns and feeds a donkey, and presupposition failure cannot lead
to truth.) Moreover, (9.23b) is not a statement about all donkeys, but rather
says that if it’s bad that Socrates is a donkey-owner, it’s good that he feeds the
donkey he owns.
Likewise again for (9.24c), which is true in all cases where there is no
specific donkey thought to be owned by any specific farmer. In that case, no
donkey and no farmer will satisfy Thought[Own (x,y)], which makes (9.24c)
300 The Logic of Language

gratuitously true. But (9.23c) can be false in such a case, namely when there is
at least one farmer thought to be a donkey-owner (though no-one has an idea
about which donkey he owns) but not expected to feed the animal he is
thought to own. For (9.23c) is not about all donkeys but about farmers who
are thought to be donkey-owners.
And again, the epithet-substitution test shows that the its in (9.23a,b,c)
cannot be bound-variable pronouns but must be external anaphors. One is,
therefore, forced to conclude that SMPC is unable to account for them in a
way that bears normal scientific generalization and avoids ad hoc solutions.
The same conclusion holds for sentences of the types illustrated in (9.8),
(9.11), (9.16), (9.19) and (9.23).
The upshot is, therefore, that there is a hard core of sentences, those
containing donkey anaphora, which resist translation into SMPC. With or
without Russell’s analysis of definite descriptions, or Quine’s programme of
elimination of singular terms, SMPC is intrinsically unable to account for
donkey anaphora. The donkey sentences contain definite expressions, pro-
nouns, or pronominal epithets, which are neither directly referring expres-
sions nor bound variables but indirectly referring expressions whose
antecedent is hidden in a preceding existentially quantified sentence—the
category of primary anaphora, which is not catered for in SMPC. One must
conclude that, for such sentence types, SMPC is, by the terms of its own
charter, unable to provide an empirically adequate and methodologically
sound logico-semantic analysis.

9.4.2 The history of the problem

Something like the donkey-anaphora problem was already known in the
late Middle Ages. As has been said, the currency of the term ‘donkey
sentences’ in modern formal semantics originates with the 1962 book
Reference and Generality by the British philosopher Peter Geach, whose
discussion of certain sentences, all about donkeys, awakened the interest of
modern logicians. But Geach did not mention—apart from token refer-
ences on page 116 to ‘another sort of medieval example’ and on page 118 to
‘medieval logicians’—that he took his cue from the British nominalist
philosopher Walter Burleigh (1275–after 1344), who introduced donkey
sentences in the context of supposition theory, the medieval equivalent of
reference theory.6 In Burleigh (1988: 92), written around 1328, one finds
examples like:

6 I owe this information to Joachim Ballweg of the Institut für deutsche Sprache in Mannheim.
 Primary and donkey anaphora 301

(9.25) Omnis homo habens asinum videt illum.

Every man owning a donkey sees it.
Burleigh’s problem has nothing to do with the empirical inadequacy of SMPC
as an analytical-descriptive language for natural language meanings. That
problem would have been an anachronism in his day, even though the truth
conditions for quantified sentences were the subject of widespread debate. His
problem was of a different nature. Although SMPC was not yet about, it was
known, in Burleigh’s day, that there are what we now call bound-variable
pronouns, as in:
(9.26) All boys feared that the dogs would bite them.
Burleigh investigates the question of anaphora resolution for bound-variable
pronouns and puts forward the principle that these may never take as
antecedent a constituent of the clause containing the quantifying expression.
In this context, he presents (9.25) as an apparent counterexample, since the
pronoun illum takes as antecedent asinum, which stands under the same verb
(videt) and is thus in the same clause. His answer is that one should not look
at the surface structure but at the underlying logical structure, where asinum,
the antecedent of illum, is not a constituent of the same clause but of the
subordinate relative clause qui habet asinum (who owns a donkey), which
saves his principle. Presumably, Burleigh did not realize that in his analysis
illum is not a bound-variable pronoun but an instance of external anaphora,
in fact, a special case of primary anaphora.
Geach (1962) discusses the same problem and in the same context: how to
account for the antecedent relation when the antecedent occurs in a relative
clause under a quantified term. His treatment of (9.25) and similar cases in his
(1962) is, on the whole, rather circumstantial and not always entirely perspic-
uous, due, it seems, to his double perspective of medieval and modern
theories of reference and quantification. But it does transpire that he disagrees
with Burleigh, though without ever mentioning him. Other than Burleigh,
and in accordance with the twentieth-century trend to disenfranchise definite
terms, he insists on treating illum as representing a bound variable. He must,
therefore, look for some way of binding illum in the way of standard modern
quantification theory or SMPC.
It would stand to reason, he says (1962: 117), to treat man who owns a donkey
in the sentences (9.27a,b), which he considers to be contradictories, as a
complex predicate ‘replaceable by the single word “donkey-owner”’. But if
we do that, (9.27a,b) ‘become unintelligible [ . . . ] because “it” is deprived of
an antecedent’:
302 The Logic of Language

(9.27) a. Any man who owns a donkey feeds it.

b. Some man who owns a donkey does not feed it.
A solution could conceivably be found in rewording these sentences as
(9.28a,b) (1962: 117):
(9.28) a. Any man who owns a donkey, owns a donkey and feeds it.
b. Some man who owns a donkey owns a donkey and does not feed it.
Yet, he says, (9.27) and (9.28) are not equivalent: whereas (9.27a) and (9.27b)
are contradictories, in accordance with the logic of both the classic Square of
Opposition and SMPC, (9.28a) and (9.28b) are not (1962: 118):7
[F]or both [(9.28a) and (9.28b)] would be true if each donkey-owner had two donkeys
and beat only one of them. Medieval logicians would apparently have accepted the
alleged equivalences; for they argued that a pair such as [(9.27a)] and [(9.27b)] could
both be true [ . . . ] and were therefore not contradictories. But plainly [(9.27a)] and
[(9.27b)], as they would normally be understood, are in fact contradictories; in the
case supposed, [(9.27b)] would be true and [(9.27a)] false.

The ‘medieval logicians’ Geach argues against are in fact none other than
the never-mentioned Walter Burleigh, who adds the following comment to
his discussion of (9.25), thereby denying that (9.27a) and (9.27b) are contra-
dictories (1988: 92–3; translation mine):
It follows that the following are compatible: ‘Every man owning a donkey sees it’ and
‘Some man owning a donkey does not see it’. For assuming that every man owns two
donkeys, one of which he sees and one of which he does not see, then it is not only
true to say ‘Every man owning a donkey sees it’, but also to say ‘Some man owning a
donkey does not see it’. In the same way, suppose that every man who has a son also
has two sons, and that he loves the one but hates the other, then both the following are
true: ‘Every man who has a son loves him’ and ‘Some man who has a son does not love
him’.

Burleigh and Geach are thus seen to disagree on account of the truth
conditions of sentences like (9.27a,b). For Burleigh, these two sentences are
compatible and not contradictories. For Geach, however, they are contradic-
tories.

7 Geach has his farmers beat their donkeys. As this would offend the feelings of many readers with
more developed notions of animal rights, I use feed, rather than beat, in my examples.
 Primary and donkey anaphora 303

Geach’s solution is to analyse a relative clause under a quantified term as an

implication under universal and a conjunction under existential quantifica-
tion:
(9.29) a. Any man, if he owns a donkey, feeds it.
b. Some man owns a donkey and he does not feed it.
This, he says, ‘is quite unforced and does give us a pair of contradictories, as it
ought’ (Geach 1962: 118). He does not say what precise SMPC renderings he
has in mind for (9.29a,b), but we can give him the benefit of the doubt and
translate them as (9.30a,b), respectively:
(9.30) a. 8x[Man(x) ! 8y[[Donkey(y) ∧ Own(x,y)] ! Feed(x,y)]]
b. ∃x∃y[Man(x) ∧ Donkey(y) ∧ Own(x,y) ∧ ¬Feed(x,y)]
Whether this is indeed ‘quite unforced’ is a matter of taste, but it is true that
all variables are properly bound and that (9.30a,b) are indeed contradictories
in the terms of SMPC. Yet, as before, the analysis founders on intervening
independent operators, as in:
(9.31) a. Any man, if he wants to own a donkey, must promise to feed it.
b. Some man wants to own a donkey and he does not promise to feed it.
It is thus clear that Geach’s analysis fails on all counts.
As regards the truth value of sentences like (9.27a,b) or (9.16c) in cases
where someone (some farmer) owns two donkeys, one of which he feeds
while he lets the other starve, for all predicate logics discussed, whether
(9.27a) and (9.27b) are taken to be contradictories, as in SMPC and ABPC,
or contraries, as in BNPC and AAPC, when (9.27b) is true, (9.27a) is false.
This means that a single man who owns a donkey he does not feed is a
counterexample to (9.27a), which is exactly what intuition tells us should be
the case.8
It also means that it is no longer sufficient for the verification of universal
statements, such as (9.27a), to check for each individual member m of
the class of donkey-owners quantified over whether the statement ‘m owns
a donkey and (s)he feeds it’ is true. The OSTA principle, discussed in Section

8 Isidora Stojanovic pointed out that a sentence like Every time Mary goes out with a Frenchman, he
pays for her drinks does not seem to be falsified by some occasion where Mary goes out with two
Frenchmen, only one of whom pays for her drinks, although then there is a Frenchman she goes out
with who does not pay for her drinks. I consider this a valid objection and all I can say is that,
depending on contextual or situational factors, the phrase a Frenchman, or, as in (9.27a), a donkey, is
apparently interpreted as ‘one or more Frenchmen’ or ‘one or more donkeys’, respectively. The
sentences in question would then read as ‘Any man who owns one or more donkeys feeds them’ and
304 The Logic of Language

3.4.2 of Volume I, will automatically assign truth to such a statement as soon

as m owns one donkey (s)he feeds, thereby opening the possibility that
donkeys owned but not fed by m are left out of account so that many possible
counterexamples are missed out on. What is needed for the verification of,
say, (9.27a) is the absence of any true statement of the form ‘ m owns a donkey
and (s)he does not feed it’. This may well have consequences for the philoso-
phy of knowledge or for theories of scientific methodology, but I can hope-
fully be forgiven for not elaborating this aspect any further.

9.5 The reference-fixing algorithm

It has become clear from what has been said so far that donkey anaphors cannot
be treated adequately in terms of variable binding. The solution proposed by
Geach has been shown to be unworkable. Donkey anaphors must be treated as
definite unquantified pronouns that take as their antecedents not a definite or
indefinite noun phrase in a preceding clause or sentence but an open discourse
address set up in the current D as a result of a preceding existential quantifica-
tion. In other words, donkey anaphora is primary anaphora. This statement
is important because it shows that discourse structures are now seen to take
on part of the burden of explanation of hitherto problematic facts. For that
reason it is important to have as clear an idea of what primary anaphora
amounts to as can be achieved in the light of what has so far been found.
One way of doing that is the following. In the context of the principle of
optimization of sense, truth and actuality, the OSTA principle discussed in
Section 3.4.2 of Volume I, it was observed that, in cases of possible referential
ambiguity, pronouns and definite descriptions automatically key on to the
reference object that makes for truth. An example, given in (3.31) of Section
3.4.2 of Volume I, is the case of two professors of English in the same
department, one a Dutch and the other a British national. That being so,
the pronoun he in the sequence of sentences There is a professor of English in
this department. He is British automatically selects the British national as
reference object because that is the reference object that makes for truth.
The first sentence is true of two individuals in the situation at hand; the
second sentence provides the information needed for a proper referential
keying of its own subject term he, which is an instance of primary anaphora.

‘Every time Mary goes out with one or more Frenchmen, they pay for her drinks’, respectively. What
the appropriate conditions are for such a reading will then be a matter of further investigation. See also
Neale (1990: 222–63) for an interesting but inconclusive discussion of much the same cluster of
problems within the terms of Russell’s Theory of Descriptions.
 Primary and donkey anaphora 305

Given the situation as described, the second sentence He is British cannot

be false because there is no way the pronoun he can be made to refer to the
professor who is a Dutch national. In general, in cases of referential ambiguity
such as these, reference fixing takes place the moment enough information is
provided to identify the correct reference object. This procedure can be
caught in terms of a simple algorithm, which is presented in Figure 9.1.
Figure 9.1 shows the algorithm for the fixing of definite singular reference
after address closure. The general principle is that reference is fixed for a
definite term as soon as the predications stored for an address in D determine
a unique object Æ in the verification domain OBJR. Once reference has been
fixed for a definite term t—that is, once æ(t) has Æ as a unique value—that
value remains unchanged for the remainder of the discourse, provided no
second object is added to or discovered in OBJR also satisfying the conditions
satisfied by Æ so as to qualify as æ(t).
If the predications stored before address closure fail to determine a unique
object Æ because more objects satisfy the criteria, as when I speak of John’s son
when John has more than one son, then the predications after closure are called
upon to secure reference. If that fails to yield the desired result, no reference
object is fixed so that no proposition comes about and, strictly speaking, no
truth value results, though the future possibility of truth remains open.
But if no reference value can be fixed because no object satisfies the
predications provided, as when I speak of John’s son when John has no son,
then, as is argued in Section 9.3, the normal consequence is that radical falsity
ensues. In such a case truth is possible only when the main predicate is
intensional with respect to the term position occupied by t, as when John
has no son and I say that John’s son is imaginary.
Thus, when OBJR contains exactly one donkey owned by Socrates,
the sentence Socrates owns a donkey is true of one donkey and, therefore,
j [[Own(S,x)]] \ [[Donkey(x)]] j ¼ 1 (the vertical bars around a set denota-
tion stand for the cardinality of the set in question). This condition suffices
for the reference fixing of the primary anaphoric pronoun it in the subsequent
sentence He feeds it. Consequently, if that is the way it is in OBJR, this latter
sentence is true just in case that one donkey owned by Socrates is indeed fed
by him, and false otherwise.
But now suppose Socrates owns two donkeys, one of which he feeds while
the other he does not. Now the sentence Socrates owns a donkey is true of two
donkeys and the subsequent sentence He feeds it is in danger of failing to pick
a referent for it. For this eventuality, the algorithm calls in the after-closure
predication Feed(Socrates,x) to help fix reference. This measure is success-
ful if OBJR contains exactly one donkey owned by Socrates and fed by him,
306 The Logic of Language

as we have stipulated it does. In such a situation, remarkably, the sentence He

feeds it CANNOT BE FALSE, because its very truth is a necessary condition for the
fixing of reference. When no reference can be fixed (for example because
Socrates does not feed either of his two donkeys), there is either no truth value
or the third value of radical falsity—depending on the theory one embraces—
but there is certainty that it cannot be true.
If Socrates owns three donkeys but he feeds only two of them while the
third is left starving, the sentence He feeds it simply fails to pick a referent for
it and the sentence must remain without a truth value (or be radically false)
until reference is fixed in a possible later sentence, which then again cannot be
false because its truth is a necessary condition for the reference fixing of it. If
OBJR contains three donkeys, all owned by Socrates but only two being fed
by him, as before, while of the two donkeys fed by Socrates only one is brown,
the sentence It is brown fixes reference and is automatically true. Clearly, this
same story can, in theory, be repeated indefinitely, though in practice appeals
to after-closure predications for the fixing of reference will be relatively rare
and restricted to one after-closure appeal.
The definite-singular-reference-fixing algorithm is a partial, but only a
partial, formalization of the OSTA Principle. In the cases where a sentence
containing a primary anaphor cannot be false because the very reference
fixing for the anaphor makes the sentence true, the algorithm guarantees
the optimization of truth: it would be absurd to say that sentence (2) in
Figure 9.1 is false because the speaker wants it to refer to the donkey not fed by
Socrates. Yet it must be borne in mind that many other factors play a role in
reference fixing that are not taken into account in the algorithm.

REFERENCE FIXED FOR it : TRUE IFF:

(1) Socrates owns a donkey

(NO REFERENCE FIXING) [[Own(S,x)]] Ç [[Don(x)]] ≠ Ø

(2) He feeds it
[[Own(S,x)]] Ç [[Don(x)]] = 1 → a Î [[Feed(S,x)]]
[[Own(S,x)]] Ç [[Don(x) ]] Ç [[Feed(S,x)]] = 1 → a Î [[Feed(S,x)]] (cannot be false)

(3) It is brown
[[Own(S,x)]] Ç [[Don(x)]] = 1 → a Î [[Brown(x)]]
[[Own(S,x)]] Ç [[Don(x)]] Ç [[Feed(S,x)]] = 1 → a Î [[Brown(x)]]
[[Own(S,x)]] Ç [[Don(x)]] Ç [[Feed(S,x)]] Ç [[Brown(x)]] = 1 → a Î [[Brown(x)]] (cannot be false)
… and so on
NB: a is the one member of set X when X = 1.

FIGURE 9.1 The definite-singular-reference-fixing algorithm and truth conditions

 Primary and donkey anaphora 307

a. b. in search of truth value for:

reference "Black(it)"
d–5 [a Cat(a) ] secondary input

truth value for

(Russellian): d–5 [a Cat(a) // Black(5)]
∃ x(Cat(x))
the a(Cat(a))
antecedent
FIGURE 9.2 Function structure of an open and a closed address
As a matterof interest, Figure 9.2 schematically shows the function structure of (a)
the open address for There is a cat (¼ (8.1) in Section 8.1.1) and (b) the closed address
for There is a cat. It is black (corresponding to (8.4) in Section 8.1.1). In Figure 9.2a,
which it is useful to compare with Figure 7.10 in Chapter 7 of Volume I, the predicate
Cat(a) (where a is a bound variable) is fed into the existential quantifier a yielding
truth just in case [[Cat(a)]] 6¼
. In Figure 9.2b, the predicate Cat(a) is again
fed into a, but a is no longer an existential quantifier but has become a
definite determiner as a result of address closure. The definite determiner a
searches the verification domain OBJR for a proper reference value (æ-value)
for the address d–5, which has now become an object representation in D. At
the same time, d–5 is the antecedent of the pronoun ‘5’ serving as the subject
term for the predicate Black. ‘Black(5)’ is now a proposition with a truth
value, provided the reference search in OBJR has been successful. If the search
results in indeterminacy because there is more than one cat, the proposition
Black(5) is called upon as a secondary input to the reference function a,
which will then no longer search for just a cat but for a black cat.

9.6 The solution

We can now proceed to a discussion of the donkey-anaphora problem proper.
It would seem that the solution to this problem is to be sought in reducing all
forms of donkey anaphora to primary anaphora, so that the mechanism of
address closure can be applied to all donkey anaphors. We deal with the three
kinds of donkey anaphora as exemplified in (9.16a,b,c) in that order.

9.6.1 Donkey anaphora under disjunction

Let us start with donkey anaphora under disjunction, as exemplified in
(9.16a), repeated here as (9.32):
(9.32) Either Socrates does not own a donkey or he feeds it.
Here the solution was already formulated in Section 3.4.2, where it was
argued, on grounds of basic-natural set theory, that a surface-structure
308 The Logic of Language

disjunction of the form ‘P OR Q’ should be tacitly understood as (P AND NOT-

Q) OR (NOT-P AND Q), making the disjuncts mutually exclusive. As explained
in Section 8.2.3, this ‘tacit’ reading is given concrete form in the result of the
incrementation procedure (IP) for OR. This turns (9.32) into (9.33):
(9.33) Either Socrates does not own a donkey and does not feed it, or he
owns a donkey and feeds it.
The second it has now found its antecedent and turns out to be an instance of
primary anaphora. Yet, as was pointed out in Section 3.4.2, the first it, which
does not occur in the surface form (9.32), is problematic since its would-be
antecedent is barred from the discourse by the negation, so that anaphora
should be excluded. The only way to bestow a coherent interpretation on the
first disjunct of (9.33) is to interpret not as the radical, presupposition-
cancelling negation. The first disjunct of (9.33) improves when the word
therefore is inserted after the first and, indicating that what follows is a
metalinguistic comment, and the negation is given heavy accent, indicating
that it is the radical, rather than the minimal, negation that is being used.
Donkey anaphora under disjunction is thus seen to be a case of primary
anaphora requiring address closure.
One notes that intervening operators do not nullify this analysis. Sentence
(9.34a), repeated from (9.23a), survives the expansion of the type ‘(P AND NOT-
Q) OR (NOT-P AND Q)’ without a hitch, as one sees in (9.34b), where both
occurrences of it are instances of primary anaphora (we take it, as is argued in
Section 10.4, that minimal NOT over still gives no longer, and minimal NOT over
no longer gives still):
(9.34) a. Either Socrates no longer owns a donkey or he still feeds it.
b. Either Socrates no longer owns a donkey and he no longer feeds it,
or he still owns a donkey and he still feeds it.
The solution to the donkey-anaphora problem in disjunctions is thus
to be sought not in the logico-semantic analysis of disjunctive sentences—
their semantic analyses or SAs—but in the instruction for their incre-
mentation.

9.6.2 Donkey anaphora in conditionals

Donkey anaphora in conditionals was exemplified in (9.16b) above, repeated
here as (9.35):
(9.35) If Socrates owns a donkey, he feeds it.
 Primary and donkey anaphora 309

Here again, the reduction to primary anaphora in terms of discourse incremen-

tation is unproblematic. As is argued in Section 8.2.4, conditional if activates the
instruction that in case the if-clause is incremented to the current D, the
subsequent incrementation of the main clause is licensed. That is, if Socrates
owns a donkey is incremented, he feeds it may also be incremented, giving rise to
the sequence Socrates owns a donkey (and) he feeds it—again a case of primary
anaphora, this time embedded in the subdomain reserved for conditional if-
clauses. Intervening operators present no problem. As in (9.14) above, the
pronoun it in (9.36) is a case of address closure across (sub)domains:
(9.36) If Socrates claims that he owns a donkey, I want to see it.

9.6.3 Donkey anaphora under universal quantification

Finally, we come to donkey anaphora under universal quantification, as it
occurs in the example sentence (9.16c), repeated here as (9.37):
(9.37) Every farmer who owns a donkey feeds it.
According to the instruction associated with ALL specified in Section 8.1.2.2
(and assuming an isomorphy between ALL and EVERY), EVERY licenses the closure,
if necessary, of any still open address for a representative of the restrictor set—that
is the closure of any address containing the information stored in (9.38a): ‘there is
a farmer such that he owns a donkey’. The address in (9.38a) induces, in virtue of
inferential bridging as shown in (8.10) and (8.11) of Section 8.1.1, the setting up of
an address d–m for the donkey owned by the farmer in question. This second
address d–m reads: ‘there is a donkey such that some farmer owns it’.
(9.38) a. d–n [aj Farmer(a), [b j Donkey(b), Own(a,b)]]
b. d–m [a j Donkey(a), [b j Farmer(b), Own(b,a)]]
The quantifier EVERY in (9.37) now licenses the closure of d–n plus the
addition of the information that the farmer in question feeds the donkey he
owns, and similarly for every address for a donkey-owning farmer to be found
or yet to be set up in D in conformity with the verification domain (situation)
at hand. That is, (9.38a) is expanded to (9.39), which is read as ‘the farmer
feeds the donkey he owns’:
(9.39) d–n [a j Farmer(a), [b j Donkey(b), Own(a,b)] // Feed(n,m)]
And likewise for any other address for a donkey-owning farmer. At the same
time, as we have seen, the instruction for the universal quantifier blocks the
addition of ¬Feed(n,m) to any address d–n, thus blocking any possible coun-
terexamples due to the OSTA principle and the reference-fixing algorithm.
310 The Logic of Language

The SA-structure postulated for sentence (9.37) is something like (9.40),

where the asinus symbol @ represents the problematic donkey pronoun.
(9.40)
So

NP1 NP2
Pred _
EVERY x
_ Det_
S1 S2 S3
ˆx
the x
Pred NP Pred _ Pred
NP NP
_
_ NP :: Farmer x x
:: Feed x @
::S4

Pred NP3 NP4

AN y
^y S4 ^y S5

Pred NP NP Pred NP
_
:: Own x y Donkey y

It is clear that @ lacks a proper variable binding so that it remains marooned

and should lead to uninterpretability. This is so not only in the technical
notation adopted here but also in any other notational or notional variety of
predicate logic. Therefore, the fact that sentence (9.37) is fully and effortlessly
interpretable cannot be explained by an appeal to the logico-semantic analysis
of sentences. It seems, however, that it can be explained in terms of the
incrementation procedures associated with sentences of this type. The prob-
lematic @ is referentially resolved through the operation of address closure
and thus through primary anaphora as occurring when the instruction
associated with the universal quantifier is applied to specific cases.
It is important to realize, in this connection, that we are forging ahead into
largely uncharted territory. That being so, one must be ready to admit that not
all questions can be answered right away and that it is as yet early days for a
complete formalization of the machinery envisaged. In fact, one must reckon
with the possibility that the aim of complete formalization will be unrealistic
for quite some time to come and perhaps forever, given the complexity of the
system and the manifold ways in which it links up with and is integrated into
the whole of the human mind. We are still finding our feet.
Yet there are plenty of signs that we are not totally off course. The machin-
ery that has been set up or envisaged is beginning to yield explanations, or at
least possible explanations, for facts that have been troublesome for a long
time because they would simply not adjust to the theories developed so far.
Thus, we now see that the donkey-anaphora problem seems to find its
solution in terms of the machinery of discourse incrementation and not in
terms of logical analysis.
 10

Presupposition and
presuppositional logic

10.1 Presupposition as an anchoring device

It is now time to pass on to presupposition, the second main device for the
maintenance of coherence in discourse. Though central to the theory of
language, the phenomenon of presupposition did not become the subject of
systematic study till the second half of the twentieth century. When it did, it
quickly became one of the wedges that helped to prise open the hitherto
intractable mysteries of context-sensitivity in language.
Essentially, presuppositions are discourse restrictors for the sentences car-
rying them. A presupposition P carried by a sentence Q restricts the universe
of discourse to what can be true or false on the assumption that P is true.
Normally, in the flow of a natural discourse, subsequent increments narrow
down the universe U of all admissible situations to an ever more restricted UR
to which new increments can again be added. A presupposition P, being part
of the semantics of its carrier sentence Q, automatically shortens that process
in that it makes its carrier sentence fit for use only in the restricted class of
discourses where P has been or can be incremented. When we say that a
sentence must have ‘the right papers’ for a given discourse D, part of what is
meant is that D must (be able to) contain, or satisfy, P.
The situations that belong to the complement of UR in U are, so to speak,
put on hold for the time being. If necessary, they can be fallen back on, as
happens, for example, when the discourse must be repaired owing to the need
to remove an incorrect or undesirable previous increment from the current
discourse. But on the whole, the situations that fall outside the UR at any stage
in a discourse play no role in the communicative process.
All the indications are that presuppositions are generated by the language
system, not by whatever factors happen to play a role in any contingent
speech situation. This appears in particular, but not exclusively, from the
fact that a sentence carrying a presupposition evokes a context where that
312 The Logic of Language

presupposition is fulfilled without that context being actually provided. This

phenomenon goes by the name of accommodation (Lewis 1979) or post hoc
suppletion (Seuren 1988). Thus, when I say John is divorced, it is not necessary
for me to say first that he was married before, because that information
is accommodated, or supplied post hoc, by the semantics of the predicate
be divorced. Since the actual physical realization of speech takes a great deal
of time and energy compared with the mere cognitive processing involved,
presupposition is seen to be a powerful device for the saving of time and
energy.
The latter point is of great functional importance. One grand way in which
the language system enables speakers to economize on the effort of speaking
and hit directly on the hic et nunc target of linguistic interaction is the device
of presupposition. We leave implicit a great deal that could be said explicitly,
not only, as is well known, because of shared knowledge (Clark 1992, 1996),
but also because the lexical meanings of the predicates we use are more
often than not reserved for special classes of situations, so that the mere
use of such a predicate carries the information that what is said is meant to
be restricted to the class of situations associated with the predicate at
hand. This latter device is encoded in the form of preconditions in the
satisfaction conditions of predicates, discussed earlier in Sections 3.1.5 and
9.4 of Volume I. It is these preconditions that generate the presuppositions.
From a strictly logical point of view, presuppositions are entailments.
When I say that a sentence like John is divorced presupposes that he was
married before, I also say that John is divorced entails that he was married
before, because it is implicit in the meaning of the predicate be divorced that
one must have been married before one can get divorced. But the strictly
logical point of view is insufficient to account for what presupposition really
is. The essence and raison d’être of presupposition lies in its function of
context abridger and guardian of contextual coherence.

10.1.1 Some early history

As is our custom, we first take a brief look at the history of the topic at hand.
The first to see that presuppositions pose a threat to Aristotle’s celebrated
Principle of the Excluded Third (PET) was his somewhat older contemporary
Eubulides of Miletus (405–330 BCE; see Kneale and Kneale 1962: 113–17),
who taught philosophy at Megara, not far from Athens (the Megarian school
of philosophy was one of the main sources of the great Stoic tradition in
ancient philosophy). Eubulides is mainly known for his four paradoxes, the
Liar, the Sorites, the Electra, and the Horned Man, and for the bitter personal
animosity between him and Aristotle.
 Presupposition and presuppositional logic 313

Strangely, the significance of his paradoxes was never properly understood

in the history of logic. Probably as a result of the enormous influence exerted
by Aristotle, Eubulides has consistently been depicted as a fool who
made flippant jokes without any real depth. Yet the Liar paradox was a
much discussed ‘insoluble’ during the Middle Ages. It was rediscovered
about a century ago, when Russell and subsequent mathematical logicians
saw how profoundly significant the Liar paradox is for logic and semantics
(the currently standard Russellian answer to it—the prohibition to mix object
language and metalanguage—is still open to serious doubt; see Seuren 1987).
But the historians of logic had meanwhile lost the connection with modern
mathematical logic, so that the sudden revival of the Liar paradox failed
to alert them to the relevance of Eubulides’ work. The Kneales are an
exception. In their standard 1962 work on the history of logic, they begin to
suspect that something is afoot in this respect, even though they fail to see
exactly what. They write (1962: 114–15) that Eubulides ‘must surely have
been trying to illustrate some theses of Megarian philosophy’ and that ‘the
Megarian study of the paradoxes was a serious affair and not mere perversity’.
In our day, most professional logicians do not even realize that their cherished
Liar paradox originated with Eubulides of Miletus.
Nowadays, in the context of modern semantics, we are in a position to see
that the four Eubulidean paradoxes constitute a highly significant head-on
attack on Aristotelian truth theory and logic, in particular the thesis of
truth as correspondence and the Principle of the Excluded Third. They also
summarize the core of well-nigh the entire research programme of twentieth-
century semantics, including the problems of vague predicates (Sorites),
intensional contexts (Electra), and presuppositions (Horned Man). It is a
surprising and highly significant fact that a contemporary of Aristotle
already spotted the main weaknesses of the Aristotelian paradigm (see Seuren
2005 for a detailed discussion).
What interests us here is Eubulides’ paradox of the Horned Man. It runs
as follows (Kneale and Kneale 1962: 114):
What you have not lost you still have. But you have not lost your horns.
So you still have your horns.
This paradox rests on presupposition. Read B for You have lost your horns and
A for You had horns. Now B presupposes A (B >> A), because the predicate
have lost has the precondition that what has been lost was once possessed.
Eubulides implicitly assumed (like, much later, Strawson) that presupposi-
tional entailments are categorically preserved under negation: B ‘ A and NOT
314 The Logic of Language

(B) ‘ A. Under PET, this would make A a logically necessary truth, which is
absurd for a contingent sentence like You had horns.1 To avoid this, PET would
have to be dropped, very much against Aristotle’s wish. Although Aristotle
himself was unable to show Eubulides wrong—his grumpy reaction was to say
that Eubulides’ paradoxes were just silly (átopa)—there is a flaw in the
paradox. It lies in the incorrectly assumed entailment in the first premiss
‘What you have not lost you still have’. For it is possible that a person has not
lost something precisely because he never had it.
To the best of present knowledge, there was no explicit awareness of
presuppositional phenomena until Frege, or perhaps more accurately, until
well into the twentieth century. We now know that the solution to Eubulides’
Paradox of the Horns lies in an adequate analysis of presuppositions,
but there are no signs that Eubulides himself was aware of that fact. All
the same, the proper answer to Eubulides’ Paradox of the Horns is still in
conflict with the classic Aristotelian Principle of the Excluded Third, as
is shown below.
The issue is not raised in any of the Ancient literature on logic or the
philosophy of language. Nor did the medievals, otherwise so resourceful and
so creative, have much to say about presuppositions. Occasionally, however,
they came close. In an anonymous text, Ars Meliduna, probably written
between 1154 and 1180 (Nuchelmans 1973: 165), Aristotle’s celebrated Principle
of the Excluded Third is called into question. One of the grounds for doubt
in this respect consists in the fact that utterances may be neither true nor
false but ‘nugatory’. In De Rijk’s edition of the Ars Meliduna we read (De Rijk
1967: 363):
. . . enuntiables such as that ‘Socrates is white because it is him’ or that ‘he loves his
son’ appear to become nugatory when Socrates is no longer white or no longer has a
son. [ . . . ] We must, therefore, posit that such enuntiables may become nugatory
[ . . . ] even if that goes against Aristotle . . . .

No further attempt, however, is made in this treatise to investigate the

concepts of truth and falsity within given contextual conditions.
Another near miss was the medieval doctrine of exponibles—expressions
that require special explication in terms of separate propositions. A proposi-
tion containing such an expression is an exponible proposition. Examples are
given in an anonymous Tractatus Exponibilium (incorrectly attributed to

1
It is a safe bet that Eubulides meant to tease the prudish Aristotle by confronting him with the
absurd and somewhat disconcerting consequence that his logic made it a necessary truth for every
man to be a cuckold. See Seuren (2005), where it becomes apparent that Aristotle was not amused.
 Presupposition and presuppositional logic 315

Peter of Spain by Mullally 1945):2 ‘exclusives’ (only), ‘exceptives’ (except),

‘reduplicatives’ (insofar as), time-aspectual verbs (begin, stop), comparatives,
superlatives, etc. Though some of these, in particular the exclusives, had been
known at least since Abelard in the early twelfth century (Mullally 1945: lxxv),
exponibles did not become a topic until about 1340.3
Some space is devoted to the exclusives (only-sentences) in the logical
treatise Syncategoreumata (edition De Rijk 1992), probably authored by
Peter of Spain (1212–1277). In the fourteenth century we have, besides
the Tractatus Exponibilium mentioned above, William of Ockham (1285–
1349) writing on exponibles in his Summa Logicæ and Walter Burleigh
(1275–1357) doing the same in his De Consequentiis (Green-Pedersen
1980) and in separate tracts on exclusives and exceptives (De Rijk 1985,
1986). All these authors divide an exponible proposition into two exponent
propositions, the propositio praeiacens (‘proposition that comes first’) and
the propositio superveniens (‘proposition that comes after’). Thus we read
Walter Burleigh on exclusive propositions (De Rijk 1985: 49–50):
. . . the proposition only man walks is exposed in the following way: ‘man walks and
nothing but man walks’. The one exponent, i.e. ‘nothing but man walks’, it has in
virtue of the exclusive expression; but the other, ‘man walks’, it has in virtue of its
prejacent. For this is the prejacent: ‘man walks’.

The two exponents are conjoined by and to give the meaning of the whole
exponible proposition. Therefore, in Peter of Spain’s words: ‘every true

2
According to De Rijk (1972: xcix) it begins to crop up at the end of manuscripts of Peter of Spain’s
Summulae Logicales starting from about 1350.
3
An observation by Horn (1985: 123; 1996: 300, and elsewhere) has caused some confusion in this
respect. Horn notes that the term praesupponere occurs in the Tractatus Exponibilium mentioned
above (Mullally 1945: 112), which he, following Mullally, incorrectly attributes to Peter of Spain (it was
written about a century later by an unknown author; see De Rijk 1972: xcix). Horn takes over Mullally’s
translation ‘presuppose’. But this cannot be correct. The term is used in the context of a sentence-type
called ‘reduplicatives’, such as: insofar as man is rational, he is capable of weeping. My best translation of
the passage in question is:
The first rule is that a reduplicative word [‘insofar as’; PAMS] anticipates (praesupponit) that some predicate inheres
in some entity and says (denotat) that the clause to which it is immediately attached expresses the cause of that
inherence.

That is, the expression insofar as anticipates that some predicate (in this case ‘capable of weeping’)
inheres in some entity (‘man’), and means in addition that ‘insofar as man is rational’ expresses the
cause of man’s being capable of weeping. Since there can be no question of ‘man is capable of
weeping’ being presupposed by the sentence mentioned, one must conclude that praesupponere is
used here in a different sense from what presuppose means today, just as supponere—the medieval
Latin term for ‘referring’—does not mean what suppose means today. As a matter of fact,
praesupponere does not occur anywhere else in the whole of the philosophical literature written in
Latin.
316 The Logic of Language

exclusive proposition leaves its prejacent true’ (De Rijk 1992: 110–11), or:
presuppositions are entailed by their carrier sentences.
Although the exclusives were the most discussed among the exponibles,
nothing suggests an awareness of their specific, discourse-related, presuppo-
sitional character. Burleigh applies, in principle, standard propositional logic
to exposed exclusives. Since only man walks is equivalent with man walks and
nothing but man walks, its negation, not only man walks, is true if at least one
of the conjuncts is false. He writes (Green-Pedersen 1980: 119):
Note that the opposite of an exclusive proposition has two grounds for truth: because
no man walks, or because something other than man walks.

He thus denies the entailment from not only man walks to man walks, despite
the natural intuition that it does hold. Had he followed intuition and thus
done justice to language, he would have found that, in this respect, language is
in conflict with standard logic, and he might have embarked on an analysis of
presuppositions. Unfortunately, however, this did not happen.
Subsequent centuries do not even come close to presuppositions. Till Frege
(1892), there is no development at all on the presuppositional front. Strawson
(1950, 1952, 1954) follows up on Frege, but specifically with regard to existen-
tial presuppositions and only in a strictly logical perspective. Like Eubulides
and Frege, Strawson assumed full entailment of presupposition under nega-
tion for all cases and concluded that PET had to go. In Strawson’s view,
nonfulfilment of a presupposition leads to both the carrier sentence and its
negation lacking a truth value altogether.
Frege (1892) had come to the same conclusion, though from a different
angle. In a sentence like (10.1) the subject term lacks a referent in the actual
world, though the existence of such a referent is presupposed in virtue of the
existential precondition on the subject term of the verb run:
(10.1) The unicorn ran for its life.
This makes it impossible to test the truth of (10.1): given that there is no
actually existing unicorn, there is no way to check whether it (whatever this it
may stand for) actually ran. Therefore, Frege, like Strawson more than half a
century later, concluded that (10.1) lacks a truth value.
This posed a profound problem for standard logic in that the applicability
of standard logic to, say, English would have to be made dependent on
contingent conditions of existence—a restriction no logician will accept. In
the effort to solve this problem two traditions developed, the Russell tradition
and the Frege-Strawson tradition.
 Presupposition and presuppositional logic 317

10.1.2 The Russell tradition

In his famous 1905 article ‘On denoting’, Russell proposed a new analysis for
sentences with definite terms, like (10.2a). Putting the new theory of quantifi-
cation to use, he analysed (10.2a) as (10.2b), or ‘there is an individual x such
that x is now king of France and x is bald, and for all individuals y, if y is now
king of France, y is identical with x’ (for a more philosophical critique, see
Section 2.1.1 in Volume I):
(10.2) a. The present king of France is bald.
b. ∃x[KoF(x) ∧ Bald(x) ∧ 8y[KoF(y) ! x ¼ y]]
In order to save bivalence, Russell thus replaced the time-honoured subject–
predicate analysis with an analysis in which the definite description the present
king of France no longer forms a constituent of the logically analysed sentence,
but is dissolved into quantifiers and propositional functions.
The linguistic negation of (10.2a) is (10.3a). Accordingly, one would expect
its logical analysis to be (10.3b)—that is, (10.2b) preceded by the negation
operator. However, Russell observed, speakers often prefer, for reasons best
known to themselves, to interpret (10.3a) as (10.3c), with internal negation
over Bald(x):
(10.3) a. The present king of France is not bald.
b. ¬∃x[KoF(x) ∧ Bald(x) ∧ 8y[KoF(y) ! x ¼ y]]
c. ∃x[KoF(x) ∧ ¬Bald(x) ∧ 8y[KoF(y) ! x ¼ y]]
This makes sentences like (10.3a) ambiguous.
This analysis, known as Russell’s THEORY OF DESCRIPTIONS, was quickly ac-
cepted by logicians and philosophers of language, as it saved strict bivalence—
a principle most were afraid to tamper with and hence dear to their hearts.
At the same time, however, it drove logicians and linguists apart, first because
it defies any notion of sentence structure as conceived of by the linguists of
the day, but then also on account of the strident unnaturalness of taking
(10.2b) as the logico-semantic analysis of (10.2a).
Russell tried to save natural intuitions of truth and falsity as much as
possible by dispensing with the—otherwise very natural and intuitive—no-
tion of ‘aboutness’. It is with that purpose in mind that he presented his
famous theory of descriptions, which is, in fact, a programme aimed at the
elimination of all definite referring terms, like the present king of France, from
the expressions of logic, including those that purport to render the meaning
of natural language sentences.
318 The Logic of Language

Besides all kinds of linguistic, semantic, and pragmatic objections, his

theory of descriptions fails to account for the natural intuition that a sentence
like (10.4a) is simply true. In Russell’s analysis, (10.4a) should be read as ‘there
is a god Apollo who was worshipped in the island of Delos and who is
identical with nothing else’, rendered formally as (10.4b):
(10.4) a. The god Apollo was worshipped in the island of Delos.
b. ∃x[God Apollo(x) ∧ Be worshipped in Delos(x) ∧ 8y[God
Apollo(y) ! x ¼ y]]
Nobody conversant with ancient mythology will deny that (10.4a) is true, as
opposed to (10.5), which is historically and factually false:
(10.5) The god Poseidon was worshipped in the island of Delos.
Yet, Russell’s analysis makes them both false, since neither Apollo nor Posei-
don ever enjoyed actual existence. For Russell, as for almost the entire body of
standard twentieth-century logic, the existential quantifier ‘there is’ has
existential import: it induces an entailment of actual, tangible, existence for
individual objects. Sentence (10.4a), therefore, should entail the actual, tangi-
ble, existence of the god Apollo. And since Apollo never tangibly existed,
(10.4a) turns out false in Russell’s analysis.
The conclusion must be, therefore, that Russell’s attempt at saving
natural intuitions was a failure. It may perhaps be thought to account for
the natural intuition that (10.6) is false, since no-one will doubt that it is
not true that there is a god Apollo who lives in Kathmandu and is unique
of his kind, but it fails, as a matter of principle, to account for the truth
of (10.4a):4
(10.6) The god Apollo lives in Kathmandu.
Russell was not totally unaware of this difficulty, but one has the impression
that he tended to be dismissive with regard to the facts of natural language.
One reads (Russell 1905: 491):

4
In his (1905: 485), Russell quips about the king of France’s alleged baldness:
By the law of the excluded middle, either ‘A is B’ or ‘A is not B’ must be true. Hence either ‘the present King of France
is bald’ or ‘the present King of France is not bald’ must be true. Yet if we enumerated the things that are bald, and
then the things that are not bald, we should not find the present King of France in either list. Hegelians, who love a
synthesis, will probably conclude that he wears a wig.

We may paraphrase this for the case at hand, saying that if one searches among the inhabitants of
Kathmandu one will not find Apollo there; yet if one looks among those who live elsewhere one will
not find him either. A Hegelian synthesis that makes him be of no fixed abode will not be of much
use either, because Apollo will be equally absent from the vagrants of this world.
 Presupposition and presuppositional logic 319

All propositions in which Apollo occurs are to be interpreted by the above rules. If
‘Apollo’ has a primary occurrence [has large scope; PAMS], the proposition
containing the occurrence is false; if the occurrence is secondary [has small scope;
PAMS], the proposition may be true.

But the theory of descriptions fails to show how Apollo in sentence (10.4a) can
possibly be assigned small scope in such a way that it turns out true. Nor is it
likely that any form of linguistic or semantic analysis will achieve such a feat,
as (10.4a) is in no way ambiguous. I fully agree with Zalta (1988: 11):
[O]ne might offer Russell’s infamous theory of descriptions as the means of analyzing
away the propositions in question. Unfortunately, this theory not only fails to do
justice to the apparent logical form of the propositions in question, but more
importantly, when applied generally, it fails to preserve the intuitive truth value of a
wide range of other propositions. For example, it turns the historical fact that Ponce
de León searched for the fountain of youth into a falsehood. Results such as this
suggest that the theory of descriptions is, at best, not general and, at worst, false.

The solution lies, of course, in the fact that a predicate like be worshipped is
nonextensional (intensional) with regard to its subject term, while a predicate
like search for is nonextensional with regard to its object term. But if one
wants to account for the facts at hand by an appeal to the occasional
nonextensionality of predicates with regard to their terms, the semantic
definition of the existential quantifier must be changed so as no longer to
imply actual existence but rather ‘being’ in a wider sense than mere existence.
This, in turn, requires, besides an extension of the logical machinery, a
thorough revision of the concomitant ontology, not found in the relevant
logical literature.5
Further objections may be raised as regards what is known as the ‘unique-
ness clause’ in (10.2b): 8y[KoF(y) ! x ¼ y], meant to say that only one king
of France exists. Russell added the ‘uniqueness clause’ in order to account for
the uniqueness expressed by the definite determiner the. In fact, however, the

5
Richard Montague’s model-theoretic possible-worlds semantics comes closest, but it fails
irreparably on account of its inability to account for substitutivity salva veritate in intensional
contexts. Dowty et al. write: (1981: 175):
We must acknowledge that the problem of propositional attitude sentences is a fundamental one for possible world
semantics, and for all we know, could eventually turn out to be a reason for rejecting or drastically modifying the
whole possible world framework.

Meanwhile, a quarter century has passed, but no solution has appeared at the horizon. I take it,
therefore, that Montague’s programme of ‘extensionalisation of intensions’ has foundered on the
cliffs of the human mind.
320 The Logic of Language

definite article implies no claim to uniqueness of existence, only to discourse-

bound uniqueness of identifying reference. Saying John and Harry met in
the pub after work in no way implies that the world contains just one pub.
Since the world is full of pubs, its Russellian translation would make it false.
Yet it may well be true, provided John and Harry met in a particular pub
whose identity has been fixed by shared speaker–hearer’s knowledge and is
thus taken for granted. Russell is his usual off-handed self when he writes
(1905: 481):
Now the, when it is strictly used, involves uniqueness; we do, it is true, speak of ‘the
son of So-and-so’ even when So-and-so has several sons, but it would be more correct
to say ‘a son of So-and-so’.

But this superior, if not arrogant, attitude with regard to the facts of natural
language blinded him to the basic truth that definite reference in language is
unrelated to uniqueness of existence and fully related to uniqueness of
identification.
Then, this analysis is limited to definite descriptions and is unable to
account for other kinds of presupposition. Factive and categorial presupposi-
tions, and those derived from words like all, still, or only, fall outside its
coverage.
To account for other than existential presuppositions some have proposed
to change Russell’s analysis into (10.7) or ‘there is a king of France, and he is
bald’.
(10.7) ∃x[KoF(x)] ∧ Bald(he)
He is now no longer a bound variable but an instance of primary anaphora
outside the scope of the existential quantifier. With a logical mechanism for
such anaphora (as in Kamp 1981 or Groenendijk and Stokhof 1991), this
analysis can be generalized to all categories of presupposition. A sentence
BA (that is, B presupposing A) is now analysed as A AND BA, and NOT(BA),
though normally analysed as A AND NOT(BA) with the negation restricted to
the second conjunct, can also, forced by discourse conditions and marked by
special accent (see below), be analysed as NOT(A AND BA), with the negation
over the whole conjunction. This analysis, which saves PET, is known as the
CONJUNCTION ANALYSIS for presupposition.
Anaphora is needed anyway, since Russell’s analysis fails for cases like
(10.8), where quantifier binding is impossible for it, which is in the scope of
I hope, while I hope is outside the scope of I know:
(10.8) I know that there is a dog and I hope that it is white.
 Presupposition and presuppositional logic 321

The conjunction analysis, however, still cannot account for the fact that
(10.9a) is coherent (though perhaps a little ponderous) but (10.9b) is not:
(10.9) a. There is a dog and it is white, and there is a dog and it is not white.
b. !There is a dog and it is white and it is not white.
(10.9a) speaks of two dogs, due to the repetition of there is a dog, but (10.9b)
speaks of only one. Yet the conjunction analysis cannot make that difference,
since the repetition of there is a dog makes no logical or semantic difference
for it.
Attempts have been made to incorporate this difference into the logic (e.g.
Kamp 1981; Heim 1982; Groenendijk and Stokhof 1991) by attaching a memo-
ry store to the model theory which keeps track of the elements that have
so far been introduced existentially. Though this is no doubt a move
in the right direction, it still falls short of what is needed, logically, philosoph-
ically, and linguistically. And even when these needs are satisfied, the con-
junction analysis still postulates existence for term referents whose existence
is denied:
(10.10) Santa Claus does not exist.
(One notes that the negation not in (10.10) in no way needs to be marked by
special accent nor be forced to take large scope by discourse factors.)

10.1.3 The Frege-Strawson tradition

Some time before the advent of Russellian logic, the German mathematician
Gottlob Frege (1848–1925) was already deeply engrossed in questions of
meaning and reference. One may rightly say that the modern history of
presupposition theory started with a footnote in Frege (1884: 87–8), which
runs as follows:
The expression ‘the largest real fraction’, for example, has no content because the
definite article has a claim to the possibility of pointing at a unique object. [ . . . ] If
one were to determine, by means of this concept, an object that falls under it, two
things would no doubt have to be shown first:

1. that there is an object falling under this concept;

2. that there is no more than one object falling under it.

Since the first of these assertions is already false, the expression ‘the largest real
fraction’ makes no sense.
322 The Logic of Language

This is clearly reminiscent of the Ars Meliduna mentioned earlier, but the
difference is that this time the observation was followed up. Even so, however,
it still took some time for presupposition theory to flourish.
The follow-up started in Frege’s famous 1892 article ‘Ueber Sinn und
Bedeutung’ (On sense and reference). There he discusses, among other things,
what truth values are with regard to sentences and how truth values are
assigned to them. For Frege, the use of a definite term normally presupposes
(setzt voraus) the actual existence of its reference object. When we say The
moon is smaller than the earth we presuppose that there is an actual moon and
an actual earth, and we say of the former that it is smaller than the latter
(Frege 1892: 31). Only if this presupposition is fulfilled can the sentence have a
truth value. If not, the sentence may still have a sense or meaning, as in
fictional contexts, but it lacks a truth value.
Frege takes the medieval distinction between the extension and the inten-
sion of predicates as his point of departure and extends this distinction
to cover argument terms and sentences as well (see Section 6.1 in
Volume I). He considers a sentence to be composed of a predicate and its
argument terms, and he follows Aristotle in saying that when the referents of
the argument terms possess the property expressed by the predicate, the
sentence is true; otherwise it is false.
For Frege, the extension of a (definite) argument term is an individual
reference object, while its intension (or sense) is the way by which a speaker-
hearer cognitively arrives at the reference object—the search or reference
procedure. The extension of a predicate is a set of individual objects, while
its intension is the corresponding concept. And, surprisingly, the extension of
a sentence (Satz) is its truth value, while its intension is defined as the
underlying thought. We read (Frege 1892: 32–3; translation mine):
Let us assume, for the time being, that the sentence has a reference! If we replace one
of its words with another word that has the same extension (¼reference object;
PAMS) but a different sense, such replacement will have no bearing on the
extension of the sentence. But we see that the thought does change; for the thought
underlying a sentence like The morning star receives its light from the sun is different
from the thought underlying The evening star receives its light from the sun. Someone
who does not know that the morning star is identical with the evening star, might take
the one thought to be true and the other to be false. Therefore, the thought cannot be
the extension of the sentence. Rather, we take the thought to be its sense. But then,
how about its extension? Is it anyway appropriate to ask that question? Maybe the
sentence as a whole only has a sense and no extension? One may anyhow expect such
sentences to occur, just as there are sentence parts with a sense but without an
extension. Sentences with nominal expressions that lack a reference will be of that
 Presupposition and presuppositional logic 323

nature. The sentence Odysseus was put ashore at Ithaca while sound asleep obviously
has a sense. But since it is doubtful that the name Odysseus occurring in this sentence
has an extension, it is equally doubtful that the whole sentence has one. Yet one thing
is certain: if one seriously takes this sentence to be true or false, one also assigns an
extension to the name Odysseus, and not just a sense. For it is to the extension of this
name that the predicate is assigned or denied. [ . . . ]
Why do we want every name to have not only a sense but also a reference? Why is
the thought alone insufficient? Because, and in so far as, we care about its truth value.
This is not always the case. For example, when we listen to an epic poem, it is, besides
the euphony of the language, only the sense of the sentences and the images and
feelings aroused by them that will captivate us. But as soon as we ask about the truth
of the story, we leave the precinct of aesthetic pleasure and enter upon the territory of
scientific investigation. As long as we take the poem as no more than a work of art, the
question of whether a name like Odysseus has an extension may remain a matter of
total indifference to us. It is, therefore, the effort to achieve truth that invariably drives
us forward from the sense to the reference.

Obviously, this is a suggestive, not a conclusive, argument. Yet the overall

coherence and force of this suggestive picture, together with its Aristotelian
basis and the lack of an equally powerful alternative, gained it the hearts and
minds of logicians and philosophers of language during the century to come.
The Fregean point of view that a predicate is a function from entities to truth
values (type e,t) became the basis of twentieth-century Categorial Grammar.
It follows, in this perspective, that when a nominal expression fails to refer
to an actually existing object, the predicate function lacks an input and
therefore cannot produce an output: the sentence in question will thus lack
a truth value. For the sake of clarity, the Fregean system of extensions and
intensions for terms, predicates and sentences is schematically rendered in
Figure 10.1.
Frege never directly opposed Russell’s theory of descriptions, which dates
from 1905, though he must have had grave reservations about it. The first to

type of categorial extension intension

expression type (Bedeutung) (Sinn)
term e entity search procedure

predicate e→t set of entities concept

sentence t truth value thought

FIGURE 10.1 Frege’s system of extensions and intensions for terms, predicates, and
sentences
324 The Logic of Language

oppose Russell directly was Peter Geach in a curious little article of 1950, in
which Geach deals with presuppositions (the term is used only in passing),
referring to the footnote in Frege (1884) quoted above. The article is a critique
of Russell’s theory of descriptions, first on account of its failure to recognize
presuppositions in ordinary language, then on account of Russell and White-
head’s defective definition of the iota operator in Principia Mathematica.
About the former, Geach writes (Geach 1950: 84–5):
On Russell’s view ‘the King of France is bald’ is a false assertion. This view seems to me to
commit the fallacy of ‘many questions’. To see how this is so, let us take a typical example of
the fallacy: the demand for ‘a plain answer – yes or no!’ to the question ‘have you been
happier since your wife died?’ Three questions are here involved:

1. Have you ever had a wife?

2. Is she dead?
3. Have you been happier since then?

The act of asking question 2 presupposes an affirmative answer to question 1; if the true
answer to 1 is negative, question 2 does not arise. The act of asking question 3 presupposes
an affirmative answer to question 2; if question 2 does not arise, or if the answer to it is
negative, question 3 does not arise. When a question does not arise, the only proper way of
answering it is to say so and explain the reason; the ‘plain’ affirmative or negative answer,
though grammatically possible, is out of place. (I do not call it ‘meaningless’ because the
word is a mere catchword nowadays.) This does not go against the laws of contradiction
and excluded middle; what these laws tell us is that if the question arose ‘yes’ and ‘no’
would be exclusive alternatives.
Similarly, the question ‘Is the present King of France bald?’ involves two other
questions:

4. Is anybody at the moment a King of France?

5. Are there at the moment different people each of whom is a King of France?

And it does not arise unless the answer to 4 is affirmative and the answer to 5 is negative.

What Geach describes here is, of course, a gapped bivalent logic, which
does violate the Principle of the Excluded Third (not the Law of the Excluded
Middle, since what is at issue is not a truth value between ‘true’ and ‘false’, but,
rather, the absence of a truth value). But the interesting thing about this article
is that the existential presupposition is brought in line with other kinds of
presupposition, a step that was repeated soon afterwards in the linguistic
literature.
 Presupposition and presuppositional logic 325

Geach’s article appeared at roughly the same time as Strawson (1950),

which was further elaborated in Strawson (1952, 1954, 1964). Strawson rein-
stated the traditional subject–predicate analysis, which had been destroyed by
Russell’s theory of descriptions. He discussed only existential presuppositions,
probably because he considered the other presupposition categories less
relevant to philosophy, or perhaps because he just was not aware of them.
His main thesis revolves around negation. Strawson considered negation to
be presupposition-preserving, which means that both BA and NOT(BA)
entail A. In order to escape from the unwanted consequence that this would
make A a necessary truth, Strawson gave up PET and introduced, besides
‘true’ and ‘false’, the possibility of lacking a truth value, realized in cases of
presupposition failure.
He remained unclear, however, as to what ‘truthvalueless’ should be taken
to mean. If taken literally, it is the absence of a truth value for sentences
with presupposition failure. This, however, is problematic not only because it
is possible to make such sentences true by the use of emphatic not, as
in (10.12a–c) below, but also because the whole machinery of logic is based
on the notion that its L-propositions express token propositions that are
bearers of truth values.
It seems sensible, therefore, to treat Strawson’s ‘truthvalueless’ as a truth
value after all, but one that is ‘infectious’ in that it produces a lack of
truth value whenever it figures in the input to any of the truth-functional
propositional functions. The resulting propositional calculus is known as
GAPPED BIVALENT PROPOSITIONAL CALCULUS, shown in Figure 10.2, where the
negation operator is symbolized as ‘~’ rather than ‘¬’, since the negation is
no longer the standard bivalent operator of standard logic. The alleged
absence of a truth value is symbolized as ‘*’.
One sees that, in so far as the values T and F are assigned, Gapped Bivalent
Propositional Calculus preserves standard propositional logic, while * is
‘infectious’ and makes any truth function inert (cf. Dummett 1973: 432).

B B
∧ ∨
~A A A T F * A T F *

F T T T F * T T T *

T F F F F * F T F *

* * * * * * * * * *

FIGURE 10.2 Strawson’s Gapped Bivalent Propositional Calculus

326 The Logic of Language

Strawson’s definition of presupposition is strictly logical. For him, ‘B

presupposes A’ (B >> A) simply means that both B and ~B entail A. When
A is false, both B and ~B ‘lack a truth value.’ This, purely logical, definition of
presupposition is given in (10.11):
(10.11) B >> A ¼Def B ‘ A and ~B ‘ A
This Frege-Strawson paradigm was attacked by Wilson (1975) and Boër and
Lycan (1976), who side with Russell and his theory of descriptions. Their
argument is that natural language negation is not invariably presupposition-
preserving, as Frege and Strawson have it, but can very well be used to cancel
presuppositions. This is correct, at least for many cases but not across
the board (see Section 10.2). Wilson (1975) provides an effusion of examples
illustrating this point, but fails to look for counterexamples. Thus, (10.12a–c)
are well-formed and coherent, though they require some sort of emphatic,
discourse-correcting accent, normally both on NOT and on the finite verb of
the correcting sentence, as in (10.12a–c), or only on the latter when the
correcting sentence takes the form of a because-clause, as in (10.13a–c):
(10.12) a. The present king of France is NOT bald. There IS no king of France!
b. Jill has NOT forgotten that Jack is her student. Jack ISN’T her
student!
c. Jack is NOT divorced. He never MARRIED!
(10.13) a. The present king of France isn’t bald because there IS no king of
France!
b. Jill hasn’t forgotten that Jack is her student because Jack ISN’T her
student!
c. Jack isn’t divorced because he never MARRIED!
Wilson, as well as Boër and Lycan, consider classical bivalent logic adequate
for natural language: presuppositional entailments differ from classical en-
tailments only ‘pragmatically’. Their argument, however, is specious, no
matter how forceful it may appear. It fails on several counts. First, these
authors appear a little too gullible when accepting Russell’s theory of descrip-
tions without further ado and despite the weighty arguments against it. Then,
their appeal to a ‘pragmatic’ explanation of presupposition-cancelling (radi-
cal) negation would be more convincing if such an explanation were available.
Sadly, however, all one is presented with in Wilson (1975), as in whatever else
has been written on a possible pragmatic account of radical negation, is a
collection of suggestive, rhetorical, and sometimes just mistaken disquisi-
tions, but no systematic observation of data and no modus explanandi. Had
 Presupposition and presuppositional logic 327

Wilson (and Boër and Lycan) looked more carefully, they would have found,
for example, that the presupposition-cancelling ‘echo’ negation NOT is not
always possible, or that, by contrast, it is sometimes the only negation
possible. Since that is clearly so, the conclusion must be that their ‘pragmatic’
way out is basically flawed. The matter is important enough to deserve a
closer look. This is done in Section 10.4, but first we must provide some clarity
on the nature and structural basis of presuppositions and on the operational
criteria for recognizing them.

10.2 The origin and classification of presuppositions

The question of the origin of presuppositions has never been answered
satisfactorily. It has been customary to speak of ‘presupposition inducers’,
which, supposedly, could be listed but not, or hardly, explained or at least
reduced to some stable parameters in linguistic descriptions. These presup-
position inducers could be lexical predicates, as in the case of the predicate be
divorced, or adverbials, as with only, or still, which induces the presupposition
that the propositional content was true before the reference time and entails
in the standard way that it is true at the time of reference. The quantifier all
and the definite determiner the were likewise—mistakenly—taken to induce
an existential presupposition with regard to the class quantified over or the
intended reference object, respectively. Ever since this question arose, I have
found this answer unsatisfactory, no matter how widely accepted it may have
been, mainly because it is merely taxonomic and ad hoc, but also because it
fails to integrate presuppositions into the machinery of semantics—quite
apart from the factual errors it is based on.
Instead, I have always attempted to reduce presuppositions to the lexical
meanings of the predicates occurring in a sentence, whereby the principle
holds that every lexical item is a predicate at the level of semantic analysis
(SA), including not only adjectives and verbs but also those elements that
make their appearance in surface structure as prepositions, conjunctions,
adverbs, quantifiers, negation, and other logical connectives. The main or
matrix predicate is, on the whole, the predicate figuring as the main finite
verb in surface structure, but it may also be an infinitival or participial, when,
for example, an auxiliary predicate is realized in surface structure as the finite
verb attracting finite-verb morphology (as with the perfect-tense auxiliary
have in English). It is a matter of considerable delicacy to decide what
presuppositions should be assigned to what predicates.
Presupposition inducers such as the adverbs only or still and many other
items that do not have verbal status in surface structure are thus considered
328 The Logic of Language

to be underlying predicates at the SA-level of representation. This, of

course, requires a machinery to transform SAs into well-formed surface
structures. The viability of this account of the origin of presuppositions
thus depends in part on whether or not there is a chance that a syntactic
theory reducing all presupposition inducers to ‘deep’ or underlying predicates
in SAs will carry the day. The programme of developing such a syntactic
theory is known as Generative Semantics, or, more properly, as Semantic
Syntax (Seuren 1996). The requirement of a sound theory of syntax is
thus seen to be far from trivial.
This approach has the double advantage of unifying presupposition
theory and placing it in a functional context. It also provides for a uniform
format for the specification of presuppositions, which are taken to be induced
by certain given preconditions of lexical predicates. This can be shown
as follows. A normal way of specifying lexical meanings is by means of
a specification of the extension of any given lexical predicate F, defined by
the conditions to be fulfilled by any object or n-tuple of objects to qualify
for F. These conditions are called the satisfaction conditions of F. For example,
one may think of specifying the satisfaction conditions of the predicate Feed
as follows:
[[Feed]] ¼ {<x,y> j x causes y to eat by bringing food to y’s mouth or body}
to be read as ‘the extension of Feed equals the set of all pairs of objects x, y,
such that x causes y to eat by bringing food to y’s mouth or body’. This may
be satisfactory up to a point, but it fails to express the fact that only
existing objects can feed and that only living objects can be fed. These
conditions can be incorporated into the meaning description of Feed
by splitting up the satisfaction conditions into, on the one hand, PRECONDI-
TIONS and, on the other, UPDATE CONDITIONS. For the predicate Feed we do this
in the following manner:
[[Feed]] ¼ {<x,y> : x and y actually exist; y is an animate being j x causes
y to eat by bringing food to y’s mouth or body}
to be read as ‘the extension of Feed equals the set of all pairs of objects x, y,
such that (a) x and y actually exist and y is an animate being, and (b) x causes
y to eat by bringing food to y’s mouth’. Conditions listed under (a) are
preconditions; conditions listed under (b) are update conditions.
Similarly for, for example, Be divorced:
 Presupposition and presuppositional logic 329

[[Be divorced]] ¼ {x : x actually exists; x has been married till time t |x’s
marriage has been dissolved by legal procedure since time t}
or: ‘the extension of Be divorced equals the set of all objects x such that (a) x
actually exists and x has been married till the time t of the process of getting
divorced, and (b) x’s marriage has been dissolved by legal procedure
since time t’. (Clearly, the predicate Be married again has its preconditions,
and so on.)6
Predicates that carry a precondition of actual existence with respect to a
term a are called extensional with respect to a. Since most predicates are
extensional with respect to their terms, we consider that to be the default case,
which can be left without special notation. Some predicates, however, are not
extensional with respect to some term. For example, the predicate think about
is extensional with respect to its subject term, but not with respect to its object
term, because one can think about anything at all, including nonexisting
entities, such as mermaids, unicorns, or dodoes. When a predicate F is
nonextensional (intensional) with regard to a term a, we asterisk that term
position in the semantic specification. Thus, in the specification of the
predicate Think about we asterisk the object-term position:
[[Think about]] ¼ {<x,y*> : x is endowed with cognitive powers j . . . }
This notation makes it clear that Think about is extensional with respect to its
subject term but not with respect to its object term.
When predicates are used in sentences, their preconditions become pre-
suppositions of the sentences in which they are used. This is how a sentence
like Jack is feeding the dog acquires its presuppositions that both Jack and
the dog actually exist and that at least the dog is an animate being. This is
also how a sentence like Jack is divorced evokes a context in which it has
been established that he was married until the moment the divorce became
effective.
Often predicates are ‘misused’ in the sense that they do not fit into the
current discourse. For example, when I say The trees whispered in the wind, in
a context about a picnic in the woods, then, clearly, the verb whisper is out of

6
Isidora Stojanovic correctly pointed out that the precondition of existence and that of having
been married before, though presented as being on a par, intuitively seem to have a different status.
I do not expand on this difference in the present text, but it seems to be reducible to the fact that
the contextual role of existential presuppositions differs considerably from that of categorial
presuppositions induced by specific lexical items such as be divorced. Denials of existence tend to
have a much more profound effect on discourse construction than denials of specific lexically bound
properties. Moreover, existential preconditions do not lend themselves to metaphoric use, as
categorial preconditions do.
330 The Logic of Language

place, because it carries the precondition that the one who does the whisper-
ing must be a living being capable of speech. Since trees do not satisfy that
condition, the wrong context is evoked and (radical) falsity should ensue. The
current term for such a ‘misuse’ is CATEGORY MISTAKE. Some category mistakes,
however, are evocative in a way that turns out to be inspiring or amusing or
perhaps even moving, in that the object failing the precondition in question is
regarded, for the purpose of the current discourse, as satisfying that precon-
dition. The fact that the wrong context is evoked is exploited precisely for the
purpose of evocation and association. When this is the case, one speaks of
METAPHOR. In the sentence at hand, for example, the trees in question are
regarded, for the purpose of the current discourse, as living beings capable of
speech, which evokes a magical world of comparisons and associations.
A presupposition is thus a semantically defined property of a sentence
making that sentence fit for use in certain contexts and unfit for use in others.
This property is partly based on the fact that if a sentence Q presupposes a
sentence P (Q >> P), then Q entails P (Q ‘ P): whenever Q is true, P is
necessarily also true—given the same situational reference points—in virtue
of the meanings of Q and P. Presuppositions are thus a subclass of entail-
ments, which, for the purpose of the present discussion, we call P-ENTAILMENTS.
Entailments that are not presuppositional are called CLASSICAL or C-ENTAIL-
MENTS. (10.14) illustrates a C-entailment (‘c); (10.15a–d) illustrate P-entail-
ments or presuppositions (>>):
(10.14) Jack has been murdered. ‘c Jack is dead.
(10.15) a. Jack lives in Manchester. >> Jack exists.
b. Jill has forgotten that Jack is her student. >> Jack is Jill’s student.
c. Jack is divorced. >> Jack was married before.
d. Only Jack left. >> Jack left.
(10.15a) is an instance of existential presupposition: to be murdered one must
be an actually existing entity. (10.15b) exemplifies factive presuppositions
(Kiparsky and Kiparsky 1971): the factive predicate have forgotten requires
the truth of the that-clause.7 (10.15c) is a case of categorial presupposition,
derived from the lexical meaning of the main predicate be divorced. (10.15d)
belongs to a remainder category, the presupposition in question being due to
the particle only. For such cases, the generalization that presuppositions are

7
Some predicates are WEAK FACTIVES, in that the truth requirement of the factive complement clause
is not absolute but can be overruled, albeit with some difficulty. Examples of weak factive predicates
are regret, surprise, anger, as in:
 Presupposition and presuppositional logic 331

induced by lexical preconditions can be upheld only in terms of a well-

founded theory that reduces surface particles like only to a predicate in the
semantic analysis of any sentence in which it occurs.

10.3 Operational criteria for the detection of presuppositions

There are various differences between P-entailments and C-entailments. The
first intuitively striking difference is that when Q >> P, P is somehow ‘prior’
to Q. On further analysis it appears that this intuition arises from the fact
that P restricts the domain within which Q is interpretable. Then, presuppo-
sitions present themselves specifically, dependent as they are on lexical mean-
ing descriptions, whereas C-entailments are ‘unguided’ in that they are
infinite in number and need not be semantically connected, as in the case of
inconsistent sentences (like All living animals are dead), which, technically
speaking, entail any arbitrary sentence. C-entailments thus lack the function
of restricting the interpretation domain. This makes presupposition relevant
for the cognitive aspects of linguistic information transfer.
But let us look now more specifically at the question of how one recognizes
a presupposition. Since presuppositions are part of the language system, and
not of the pragmatics of use, they are detectable (‘observable’) irrespective of
actual token use. Like C-entailments, P-entailments can be read off isolated
sentences, regardless of special context. This makes it possible, as has been
observed, for P-entailments to evoke a context all by themselves—something
C-entailments cannot do. (10.15a), for example, requires it to be contextually
given that there actually exists someone called ‘Jack’ and thus evokes such a
context, while it asserts that he lives in Manchester. (10.15b) evokes a context
where Jack is Jill’s student, while asserting that Jill has forgotten that. (10.15c)
requires a context where Jack was married, and asserts that the marriage has
been dissolved. And (10.15d) requires a context where Jack left, while asserting
that no one else did. This, together with the criteria for detecting entailments
to be discussed presently, provides a set of operational criteria to recognize
presuppositions.
First, as has been said, if Q >> P then Q ‘ P. But on what grounds does one
conclude that one sentence entails another (apart from the fact that in
standard logic every sentence entails itself)? The usual heuristic criterion for

(i) Harold was under the illusion that his son had failed and this (namely that his son had failed)
angered him.
Weak factive predicates invariably involve an emotive factor. See Gazdar (1979: 119–22) for some
discussion.
332 The Logic of Language

an entailment Q ‘ P is the incoherence of the juxtaposition of NOT(P) with Q.

On the whole, this works well as a criterion for entailment in the general
sense. For example, (10.16a) does not entail, and therefore does not presup-
pose, (10.16b), since (10.16c) is still coherent.
(10.16) a. Lady Fortune neighs.
b. Lady Fortune is a horse.
c. Lady Fortune is not a horse, yet she neighs.
But this criterion overkills when the entailing sentence is qualified by an
epistemic possibility operator, like English may, as in (10.17a), which does
not entail (10.17b) even though (10.17c) is incoherent. Epistemic possibility
requires compatibility of what is said to be possible with what is given in
discourse or knowledge (see Section 7.2.1.3). Therefore, if Q ‘ P, then with
NOT[P] in the knowledge base, Possibly[Q] results in inconsistency, even
though Possibly[Q] does not entail P. This means that the criterion for
entailment must be refined. A viable refinement, without loss of generality,
consists in testing the (in)coherence of the juxtaposition of Possibly [NOT[P]]
with Q, as in (10.17d). Since (10.17d) is coherent, (10.17a) does not entail, and
therefore does not presuppose, (10.17b) (‘!’ marks incoherence):
(10.17) a. Jack may have been murdered.
b. Jack is dead.
c. !Jack is not dead, yet he may have been murdered.
d. Jack may not be dead, yet he may have been murdered (and thus
be dead).
By contrast, (10.18a) entails (10.18b), because one cannot coherently say
(10.18c):
(10.18) a. Jack has been murdered.
b. Jack is dead.
c. !Jack may not be dead, yet he has been murdered.
Further criteria are needed, however, to distinguish P-entailments from
C-entailments. First there is the PROJECTION CRITERION: if Q >> P and Q stands
under an entailment-cancelling operator like Possibly or NOT or Believe,
P survives not as a P-entailment but as a more or less strongly INVITED
INFERENCE (>) which can be overridden. Generally, O[QP] > P, where ‘QP’
stands for ‘Q presupposing P’ and ‘O’ for an entailment-cancelling operator.
In standard terminology: the presupposition P of Q is PROJECTED through the
operator O as an invited inference. To work out the conditions under which
 Presupposition and presuppositional logic 333

and the form in which presuppositions of embedded clauses are projected

through higher operators constitutes the PROJECTION PROBLEM OF PRESUPPOSI-
TION, discussed in Section 10.5.
Projection typically comes with P-entailments, as in (10.19a), not with
C-entailments, as in (10.19b):
(10.19) a. Jill believes that Jack is divorced > Jack was married before
b. Jill believes that Jack has been murdered > / Jack is dead
The projection criterion is mostly used with NEGATION as the entailment-
cancelling operator. Strawson (1950, 1952) held, incorrectly, that presupposi-
tion is always preserved as entailment under negation. In his view, a sentence
like (10.20a) still presupposes, and thus entails, that there exists a king of
France, who therefore, if (10.20a) is true, must lack wisdom. However,
according to the entailment criterion specified above, Strawson was wrong,
because one can coherently say (10.20b):
(10.20) a. The present king of France is not wise.
b. The present king of France may not exist, but he is, in any case,
not wise.
Although presuppositions are, in fact, normally weakened to invited infer-
ences under negation, there are cases where they are not but remain presup-
positions in the full sense of the term. More is said on this issue in Section 10.4
below. For the moment it suffices to say that, apart from such special cases,
presuppositions are normally weakened to invited inferences under negation,
as they are under other entailment-cancelling operators.
In Strawson’s day, it was still customary to take what is in fact the default
function of negation in presupposition-carrying sentences for a categorical
function. As a result, Strawson’s ‘negation test’ became the standard test for
presupposition. We now know that this test is unreliable, for reasons set out in
Section 10.4. If we still want to use it, it has to be heavily qualified.
We fare better with the DISCOURSE CRITERION: a discourse bit P and/but QP
(with allowance for anaphoric processes) is felt to be consistent with the
context given, orderly, and informative—that is, SEQUENTIAL. The condition of
sequentiality is used to characterize stretches of acceptable text that have their
presuppositions spelled out (‘S’ signals sequentiality):
S
(10.21) a. There exists someone called ‘Jack’, and he lives in Manchester.
S
b. Jack is Jill’s student, but she has forgotten that he is.
334 The Logic of Language
S
c. Jack was married, but he is divorced.
S
d. Jack left, and he is the only one who did.
C-entailments and inductive inferences lack the property of sequentiality.
When they precede their carrier sentence the result may still be acceptable, yet
there is a qualitative difference, as shown in (10.22a,b), where a colon after
the first conjunct is more natural (‘C’ marks nonsequential but coherent
discourse):
(10.22) a. CJack is dead: he has been murdered.
b. CJack earns money: he has a job now.
The discourse criterion still applies through projection: P AND/BUT O[QP]
is again sequential (the entailment-cancelling operators are printed in
bold face):
S
(10.23) a. Jack really exists, and Jill believes that he lives in Manchester.
S
b. Jack is Jill’s student, but she has probably forgotten that he is.
S
c. Jack was married in the past and he has not got divorced.
S
d. Jack left, and he is not the only one who did.
These tests reliably set off P-entailments from C-entailments. They thus
form an operational test for the detection of presuppositions.

10.4 Some data that were overlooked

It appears that neither the Russell tradition (including the analysis proposed
in Wilson 1975 and Boër and Lycan 1976) nor the Frege-Strawson tradition can
carry the day. The former holds that natural-language negation always cancels
all entailments (except those that are necessarily true in virtue of the meaning
of the entailing sentence or sentences); the latter holds that negation always
leaves presuppositional entailments intact and it seeks a way out of the
predicament of having to call all presuppositions necessary truths by allowing
an infringement of PET. The problem with both traditions is that they have
both overlooked the fact that natural language negation does not categorically
either cancel or preserve presuppositional entailments. Close inspection of
the facts shows that, although most of the time natural language negation
allows for the cancelling of presuppositions (provided it is invested with
emphatic accent), it sometimes precludes and sometimes requires presupposi-
tion cancelling. Moreover, the criteria for optional or obligatory preservation
or cancelling of presuppositions are of a systematic linguistic nature, which
excludes, as a matter of principle, any kind of pragmatic account.
 Presupposition and presuppositional logic 335

Let us have a closer look. Although no claim of completeness can be made,

and further cases will probably come to light as research proceeds, we can
say that in the following classes of cases negation is per se presupposition-
preserving:
A. MORPHOLOGICALLY INCORPORATED NEGATIONS
Negative prefixes like un-, in-, dis-, a-, cannot fulfil the cancelling role of NOT.
Thus, (10.24a) is felt to be inconsistent, whereas (10.24b) easily allows for the
cancelling interpretation (initial ‘!’ indicates inconsistency):
(10.24) a. !Tim is UNrealistic about the risk. He doesn’t know there to be
one!
b. Tim is NOT realistic about the risk. He doesn’t know there to be
one!
B. NEGATIONS IN NONCANONICAL POSITIONS
By ‘canonical position’ is meant the position of ordinary, unmarked sentence
negation. For English, this is the position in construction with the finite verb
form, with or without do-support. Remarkably, negations in any other posi-
tion than the canonical one are necessarily presupposition-preserving, even
when they are logically speaking the highest operator and thus function as
sentence negation:8
(10.25) a. !NOT only Harry laughed. He didn’t laugh at all!
b. It is not true that only Harry laughed. He didn’t laugh at all!
(10.26) a. !NOT all doors were locked. There wére no doors!
b. All doors were NOT locked. There wére no doors!
Native intuition clearly says that the first sentence of (10.25a), with or without
the emphatic accent on not, entails that Harry laughed, just as the same
sentence without the initial not. As a whole, therefore, (10.25a) is incoherent,
with or without the emphatic accent on not. This, in fact, is one of the clearest
and most convincing cases showing that sentential negation in natural
language is sometimes obligatorily presupposition-preserving. All one can
do to make the negation cover the presupposition as well is to say something
like (10.25b).

8
The only exception, as was pointed out to me by Larry Horn (p.c.), is the English construction
with not even, as in Not even John was disappointed. This semantically and grammatically complicated
construction is discussed in Section 11.4.
336 The Logic of Language

As regards (10.26), both (10.26a) and (10.26b) are to be understood with the
negation as the highest operator, followed by the universal quantifier. (10.26a)
poses no problem for SMPC: since the first sentence of (10.26a) entails the
existence of doors, owing to the equivalence in SMPC of ¬A and I*, it is
incompatible with the second sentence. But (10.26b), which should have the
same analysis, does pose a problem for SMPC, precisely because its two
sentences are compatible to the native speaker.
(10.26b) is of particular relevance because it conflicts with standard
modern predicate logic and, therefore, with the conjunction analysis
discussed above. For SMPC, NOT[ALL DOORS were LOCKED] is equivalent
with SOME DOORS were NOT LOCKED and should, therefore, entail the
existence of doors. This would make the conjunction There were no doors
and not[all doors were locked] inconsistent. In fact, however, it is not, provided
the negation is placed in the canonical position and is provided with heavy
emphatic accent, as in (10.26b).

C. NEGATIONS IN NONASSERTIVE CLAUSES OR IN THE SCOPE OF HIGHER OPERATORS

Negations in nonassertive main sentences and in many subordinate clauses or
infinitivals cannot cancel presuppositions:
(10.27) a. p ffi seems NOT to be back. He hasn’t been away at all.
!Tim
b. Tim does NOT seem to be back: he hasn’t been away at all.
(10.28) pffi may NOT be divorced. He never got married in the first place.
a. !Tim
b. Tim can’t be divorced: he never got married in the first place.
(10.29) !Do NOT go back to your wife. You haven’t even left her.
D. NEGATIONS WITH CERTAIN QUANTIFIERS
As was demonstrated in (10.26b), the negation with the quantifier ALL
can be used to cancel presuppositions, provided it is placed in the canonical
position. Not so, however, it seems, with, for example, each (of the), or both
(of the):
(10.30) !Each of the children was NOT given a sweet. There wére no children!
(10.31) !Both of his children are NOT spoiled. He hás no children!
E. NONEXTRAPOSED FACTIVE SUBJECT CLAUSES
When a factive that-clause in semantic subject position is preposed (or, if one
prefers, nonextraposed) in surface structure, the negation over the factive
 Presupposition and presuppositional logic 337

main verb is, though in the canonical position, unable to cancel the factive
presupposition, as shown in (10.32a), though it may affect other presupposi-
tions, as is made clear by (10.32c). Only if the factive subject clause is
extraposed, as in (10.32b), can the (canonically placed) negation cancel the
factive presupposition:
(10.32) a. !That Tom is clever does NOT irritate Joanna. He ISN’T clever!
b. It does NOT irritate Joanna that Tom is clever. He ISN’T clever!
c. That Tom is clever does NOT irritate the king of France. There IS no
king of France.
One notes, moreover, that when the factive clause is pronominalized by
means of that, the factive presupposition still remains intact under negation,
as is shown in (10.33a). But when the negation is reinforced with epistemic
possibility and comes out as cannot, the factive presupposition can be can-
celled, as in (10.33b):
(10.33) a. !That does NOT irritate Joanna. He ISN’T clever! (cf. (10.32a))
b. That CANNOT (possibly) irritate Joanna. He ISN’T clever!
These observations were not made in either Wilson (1975) or Boër and
Lycan (1976). Had they been made, they would have undermined their
analysis.
F. CLEFT AND PSEUDOCLEFT CONSTRUCTIONS
As is well known, cleft and pseudocleft constructions have a specific existen-
tial presupposition associated with the clefted WH-constituent: if in the non-
cleft version of the sentence this constituent requires a really existing
object for the sentence to be true, so does the clefted constituent, whether
in cleft or in pseudocleft constructions. This presupposition is uncancellable
by negation:
(10.34) !What he said was NOT ‘Damn!’. He said nothing at all!
Here the existential presupposition applies to the WH-constituent what he
said, since for something to be said it must, albeit for a brief moment, actually
exist. But other presuppositions not directly associated with the clefted
constituent are fully cancellable:
(10.35) Who wrote the letter was NOT Mr. Davis. Mr. Davis doesn’t exist!
Here it is presupposed that someone wrote the letter, since in order to write a
letter one must actually exist. But Mr. Davis’s existence is not presupposed,
338 The Logic of Language

because in the semantic analysis of the cleft sentence be Mr. Davis is the value-
assigning predicate bev.
G. CONTRASTIVE ACCENTS
Contrastive accents form an exact parallel to the (pseudo)cleft constructions.
In sentences with contrastive accent the accented constituent serves as a
predicate establishing the identity of the entity mentioned in the nonaccented
part. This latter entity is presupposed to exist in all cases where it is in
the corresponding sentence without contrastive accent. This presupposition
cannot be cancelled under negation:
(10.36) !The WAITER did NOT start the argument. Nobody did!
Again, however, other presuppositions, such as those associated with
the accented part functioning as an underlying predicate, are freely can-
cellable:
(10.37) The WAITER did NOT start the argument. There wás no waiter!
H. NEGATIONS WITH NEGATIVE POLARITY ITEMS
As is well known, every language has a, usually large, number of so-called
‘negative polarity items’ (NPIs). These are words, constructions, or expres-
sions which, mostly for unknown reasons, require a negation or, for some
NPIs at least, a negative word, when used in simple declarative sentences.
(Their behaviour in other clause-types differs in ways that have as yet never
been exhaustively studied.) Some, but not all, NPIs allow for emphatic
auxiliaries (do-support when there is no auxiliary) as a form of negativity.
In the examples below the NPIs are italicized. (10.38a) is a standard case.
In (10.38b,c) one has NPIs with negative words (hardly, difficult). (10.38d) is
a case of emphatic do-support:
(10.38) a. She couldn’t possibly have known that.
b. She could hardly breathe any more.
c. It was difficult (*easy) for him to go on any longer.
d. It DOES matter that Jones is an alcoholic.
The negation required in simple assertive clauses with NPIs (if there is no
other negative word and no auxiliary emphasis) is per se presupposition-
preserving, for all presuppositions in the sentence. Thus, the examples of
(10.39) are all felt to be inconsistent, if not outright ungrammatical:
 Presupposition and presuppositional logic 339

(10.39) a. !It does NOT matter that Jones is an alcoholic. He ISN’T! (factive)
b. !Jones does NOT live in Paris any more. He doesn’t exist!
(existential)
c. !He did NOT at all acknowledge my presence. I wasn’t there!
(factive)
NPIs have a counterpart in so-called ‘positive polarity items’ (PPI). When a
PPI stands directly under negation, the sentence loses its default property of
inviting presuppositional inferences and acquires what is known as an ‘echo-
effect’ to an even stronger degree than in presupposition-cancelling sentences
without a PPI: it sounds as if the same sentence but without the negation has
been uttered (or strongly suggested) in immediately preceding discourse,
preferably by a different speaker. Take, for example, the PPI still, which
induces the presupposition that what is said in the rest of the sentence, if in
the present tense, was true at least till the moment of utterance, and the
sentence as a whole asserts that that situation continues to obtain. Contrast
this with the NPI any more, which induces the same presupposition but
lets the sentence, with the obligatory negation, assert that that situation has
ceased to obtain. Thus, given a sentence with the PPI still, its natural negation
will not be that sentence with the default-cancelling and ‘echoing’ NOT
but rather that sentence with still replaced by not . . . any more, as in the
following pair:9
(10.40) a. Harold still lives in Paris.
b. Harold doesn’t live in Paris any more.
The test is now that the presuppositions of (10.40a) have not become invited
inferences but are cancelled altogether when simple not is inserted, whereas
those of (10.40b) are not cancellable, just as in (10.39b):
(10.41) a. Harold does NOT still live in Paris: he has never set foot in France.
b. !Harold doesn’t live in Paris any more: he has never set foot in
France.
Examples of English PPIs are (see also Seuren 1985: 233): rather, far from,
hardly, terrific, daunting, ravenous, staunch, as fit as a fiddle, at most, at least,
perhaps, already, certainly, surely, awful, even, each, both, most, some, several,

9
For the reader’s reassurance, I have often tested this out with my students. The reader, if in a
teaching position, might do the same. He or she will then find out that, when the students are asked to
give the negation of a sentence like (10.40a), their answer will be (10.40b).
340 The Logic of Language

few, not. Note that the negation word not is itself a PPI: a succession of two or
more occurrences of not has the effect of cancelling all presuppositions and
creating an echo. But if there is no stark succession of two nots, as in (10.42)
below, they can both be presupposition-preserving.
Thus, generally, when a PPI stands in the immediate scope of NOT it cancels
the presuppositions of the sentence, not leaving even an invited inference. It
then also produces an echo-effect. However, Baker (1970) observed that,
interestingly, this is not so when there is double negation (other than stark
succession of nots), as in (10.42), with the PPI rather:
(10.42) There is nobody here who wouldn’t rather be in Montpelier.
This sentence carries no echo-effect and contains no radical negation, despite
the occurrence of rather. Baker’s observations are tantalizing, but still unex-
plained.
A further unexplained complication is that some, but not all, PPIs can
stand under an unaccented not when an explicit or implicit comparison is
made:
(10.43) a. You are not still building (as we are).
b. She hadn’t already finished (as you had).
Such sentences have a (slight) echo-effect, but preserve presuppositions in so
far as these are not induced by still or already.
PPIs are generally excluded in the scope of implicitly negative operators (or,
if one prefers, operators with underlying negation), such as the comparative
than, as shown in (10.44a). In (10.44b), the PPI some is outside the scope of
than (as opposed to any in the same position); this sentence is interpreted as
‘there are some of her colleagues who she is richer than’ (‘*’: ungrammatical):
(10.44) a. *She is richer than you already/still are.
b. She is richer than some of her colleagues.
If the comparative particle than is analysed as containing an underlying
negation (see Seuren 1973), this agrees with the observation made in A
above that morphologically incorporated negations are necessarily presuppo-
sition-preserving and cannot take PPIs in their immediate scope.
Just as it is, for the most part, unknown what causes the phenomena
mentioned under A–H, it is not known what system or mechanism is
responsible for the emergence of polarity items, whether positive or negative,
and their behaviour. Nor is much known about the question of what factors
 Presupposition and presuppositional logic 341

lie behind the fact that often the negation word cancels presuppositions as
entailments but leaves them as invited inferences, while in certain classes of
cases it preserves some or all of the presuppositions in the sentence at hand,
and in other classes of cases it eliminates even the invited inference of the
presuppositions—the projection problem. It would seem that a theory of
topic–comment modulation might lay bare the grounds of the necessary
preservation of presuppositions in the categories E (non-extraposed factive
clauses), F ((pseudo)clefts) and G (contrastive accents). Sentences that fall
under these categories have a grammatically fixed topic–comment structure
built into them in such a way that the presupposition adheres to the topic,
and presupposition-cancelling can probably be shown to be incompatible
with topic-hood. Yet on the whole, our theoretical insights still fall short of
an explanation of the facts concerned.
Even so, however, the answer cannot be merely that the negation operator
in language is just the simple bivalent truth-functional operator known from
standard logic, somehow modified by pragmatic factors. Pragmatic types of
analysis are in principle unable to cope with the clear-cut difference between
the cases where presuppositions are necessarily preserved and those where
they are necessarily cancelled. The minimal conclusion to be drawn is that
there are at least three systematically differing ways of using the negation:
(i) with the presuppositions necessarily preserved, (ii) with the presupposi-
tions reduced to invited inferences, and (iii) with even the invited inferences
removed. The question is now: what theory has the best chance of coming
to grips with the facts observed above? A Gricean pragmatic theory may
be considered for certain peripheral parts of the question, but it does not
seem the first choice for the central problems, given the known failure, so
far, of such theories in those areas. The observed facts are anyway too
linguistically structural to be a natural object for pragmatics, whose typical
hunting ground is the nonlinguistic interactional aspects of communication
by means of utterance tokens.
If presupposition is not a pragmatic phenomenon, is it a logical phenome-
non? Again, the answer appears to have to be negative. If, as we have posited,
presuppositions originate in the preconditions of predicates, and if the
raison d’être of these preconditions is to restrict the use of predicates to
certain classes of situations, then presupposition is primarily a semantic
property of sentences, whose function it is to restrict the use of sentences
to certain classes of contexts (discourses). Presupposition is thus a discourse-
semantic phenomenon with, as one might expect, consequences for the logic
of language. These consequences, however, are epiphenomenal on the true
nature of presuppositions; they do not define presuppositions. This enables us
342 The Logic of Language

to modify the Strawsonian definition of presupposition given in (10.11) above

in the following way (again, one notes that the symbol used for negation is ‘~’,
not the standard bivalent ‘¬’; the symbol ‘~
_’ stands for the radical, presuppo-
sition-cancelling negation):
(10.45) If B >> A then B ‘ A and ~B ‘ A and (~A or ~A) ‘ ~B.10
What this means for propositional logic is explained in Sections 10.6
and 10.7.

10.5 Presupposition projection

10.5.1 What is presupposition projection?
As was said in Section 7.2.2.2, presupposition projection is the phenomenon
that presuppositions of embedded clauses (E-PRESUPPOSITIONS) tend to perco-
late upward into higher domains, either as fully entailed presuppositions, or
as default inferences. In the former case they require, in the latter case they
prefer, a higher D admitting them. If the requirement is not met, the sentence
in question is unfit for the D at hand and is rejected. If the preference is not
met, the sentence in question is still acceptable in D but the E-presupposition
stays within its own subdomain (or, as is explained below, is relegated to a
more general recipient subdomain).
The driving force behind this phenomenon is the Principle of Maximal
Unity of discourse domains, discussed in Section 7.2.2. This principle makes
listeners keep their Ds as compact as possible. Since maximization of com-
pactness is a general feature of human interpretative processes (Ockham’s
razor is one manifestation of it, and so is the principle of induction),
presupposition projection is a direct consequence of the way the mind deals
with its environment in a general sense. For that reason, presupposition
projection is maximal: it will make concessions only when forced to.
A powerful means of upward percolation of E-presuppositions is provided
by the process of accommodation or post hoc suppletion discussed in Section
10.1. It takes little effort to see that in cases where presuppositions have not

10
The inverse does not hold. It is possible for both B and ~B to entail A, without A being a
presupposition of B (or ~B). For example, both B and ~B entail any necessary truth T, but T is most
probably not a presupposition of B (or ~B). Or consider the fact that if B and/or ~B entail A, they also
entail A ∨ C (the entailment schema of addition; see Section 3.3.2), where C is any arbitrary sentence,
but A ∨ C is highly unlikely to be a presupposition of A. Moreover, as is shown in Section 10.6,
conjunctions of the form B ∧ BA cannot be taken to presuppose A, even though (minimal or radical)
falsity of A leads to the radical falsity of B ∧ BA.
 Presupposition and presuppositional logic 343

been spelled out explicitly, their post hoc suppletion will take them as far as
possible into the higher domains. That process is stopped only when the
introduction into a higher domain leads to inconsistency either with the
domain itself or with independently available situational or world knowledge.
In other words, E-presuppositions will percolate upward as long as they
are compatible with any higher domain they are about to invade. It is for
that reason that presupposition projection is, normally speaking, a default
process, which can be overruled by contrary information. It is stopped by
inconsistency and enforced when the presupposition in question is entailed in
the domain in question. In between there is a gliding scale of possibilities,
which is analysed in some detail in the present section.
For sentences that are looked at in isolation, regardless of any specific
context they may occur in, the question of whether an E-presupposition
makes it upward or not, and if so in what form, depends on the predicate
or the instruction that has created the subdomain, given the principles of
D-construction. Following Karttunen (1973: 178), the literature has made a
distinction between (a) HOLES: predicates which cannot stop presuppositions
and let them through as full entailing presuppositions because they entail and
often presuppose their argument clauses, (b) FILTERS: predicates which let
presuppositions through but in the weakened form of default inferences,
and (c) PLUGS: predicates which categorically stop presuppositions from
creeping upward.
During the 1970s and 1980s, presupposition theory was entirely dominated
by the question of what formalism would account for presupposition projec-
tion, whereby no account was taken of the fact that the entire projection
mechanism is driven by, and becomes transparent in the light of, the Principle
of Maximal Unity that governs discourse incrementation processes. Instead,
one looked for a strictly formal calculus, very much in the tradition of formal
semantics that was en vogue during that period. The total disregard for the
ecology of the phenomena at issue led to a situation where the debate became
sterile and fruitless, as a result of which it inevitably petered out. Now that a
more realist course is taken in discourse semantics and some greater clarity
exists regarding D-structures and incrementation processes, the question can
be looked at in a new and more explanatory light.

10.5.2 Projection from lexical subdomains

Let us look at sentence (10.46), containing the main verb Hope, which creates
(or continues) an intensional subdomain.
344 The Logic of Language

(10.46) Joan hoped that her son’s new girlfriend would have better manners
than the previous one.
Taken in isolation (10.46) carries the following default inferences: (a) Joan
had a son, (b) Joan’s son had a girlfriend, (c) Joan’s son had a previous
girlfriend, and (d) her son’s previous girlfriend’s manners could be improved.
In a context containing the information (a), (b), (c), and (d), (10.46) is
perfectly sequential—that is, consistent with the context given, orderly, and
informative. Yet none of these inferences are entailed, as they can all be
cancelled. Let us start with (d). (10.46) is still fully sequential if preceding
context says that Joan’s son’s previous girlfriend had perfectly good manners
but Joan herself thought less well of the manners of this girl. The default
inference (c) is equally vulnerable, because the preceding context may contain
the information that Joan’s son’s actual girlfriend was his first one, in which
case the default inference (c) is cancelled and prevented from projecting, but
the text remains fully sequential. Likewise for the default inference (b),
because preceding text may have told the listener/reader that Joan had been
misinformed and that, in fact, her son never had a girlfriend at all, which still
leaves (10.46) fully sequential. Similarly again for the default inference (a),
which can likewise be eliminated by preceding context. D may contain the
information that Joan lives in a world of her own making. Poor Joan never
had any children but in her own fantasy she had a son, who once had an ill-
mannered girlfriend. In such a context, (10.46) may well describe Joan’s latest
delusion. In sum, the more default inferences are scrapped, the more Joan
seems to be out of touch with reality. The conclusion is that the verb Hope is
a so-called ‘filter’, letting through its E-presuppositions as default inferences
as long as they are not stopped by contrary information stored in higher
domains or in available knowledge.
One would expect presuppositions that do not make it into a higher
domain to stay put as presuppositions of the domain in which they have
been generated. While this is true for many subdomain-creating predicates, it
is not true for all. As was pointed out in Section 7.3.3, there appears to be a
class of predicates that do not let their subdomains take in nonprojected
presuppositions but, instead, send them to a subdomain created by a predi-
cate that is higher in the subdomain hierarchy. Hope is one such predicate. In
a context where the default inference (d) of (10.46) is blocked, (10.46) is not
interpreted as saying something like (10.47a), but rather as saying something
like (10.47b). The blocking of the default inference (c) requires a specification
of a subdomain for Joan’s belief about her son’s previous girlfriend or
girlfriends:
 Presupposition and presuppositional logic 345

(10.47) a. Joan hoped that her son’s previous girlfriend’s manners could be
improved and that her son’s new girlfriend would have better
manners.
b. Joan believed that her son’s previous girlfriend’s manners could
be improved and she hoped that her son’s new girlfriend would
have better manners.
Those subdomains that need an appeal to a different subdomain for the
storage of nonprojected presuppositions are called subsidiary subdomains,
while the receiving subdomain is named recipient subdomain (Section 7.3.3).
As far as can be seen, given the present state of the enquiry, it looks as if
subsidiary subdomains are created or continued either exclusively or typically
by emotive complement-taking predicates.
These include predicates like hope or fear, but also the weak factive pre-
dicates like regret, or surprise. In the case of strong or weak factives, projection
does not primarily affect the E-presuppositions of the weak factive comple-
ment clause but the clause itself in its entirety or any of its semantic, including
presuppositional, entailments. Weak factives differ from ordinary strong
factives in that a negation of their presupposed that-clauses in D or in
available knowledge strongly resists, but in the end does yield to, the incre-
mentation of the sentences in the superordinate D (see note 7). In Sections 3.2
and 6.2.3.2 of Volume I it was observed that weak factives do not allow for
substitition salva veritate of topic–comment modulation and the following
examples were given, repeated here as (10.48a,b):
(10.48) a. It surprised/angered Ann that JOHN (and not Kevin) had sold the car.
b. It surprised/angered Ann that John had sold THE CAR (and not the
speedboat).
It requires little effort to see that (10.48a) differs truth-conditionally from
(10.48b)—a fact that has so far remained undiscussed in the semantics or
pragmatics literature.
Weak factive predicates are thus almost ‘holes’, but not quite. Contextual
overruling seems possible, since sentences like (10.49a,b) are still sequential—
that is, consistent and informative—albeit perhaps with some difficulty:
(10.49) a. Kevin had been falsely told that John had left the country and it
surprised/angered him that John had done that.
b. Someone had whispered into Ann’s ear that John had sold the car,
although John had done nothing of the sort, and it surprised/
angered her that JOHN had done that.
346 The Logic of Language

What we now see is that nonprojection of the presupposed complement

clause or any of its presuppositional or other semantic entailments on account
of contextual blocking requires a subdomain for Ann’s belief, not for her
surprise or anger, to store the nonprojected subordinate clause or its entail-
ments. Suppose someone had whispered into Ann’s ear that John had sold the
car, whereas John had done nothing of the sort, the proper interpretation of
(10.48a) is not (10.50a) but (10.50b):
(10.50) a. It surprised/angered Ann that John had sold the car and that it
had been JOHN to do so.
b. Ann had been made to believe that John had sold the car and it
surprised/angered her that it had been JOHN to do so.
We thus have subsidiary subdomains, created by complement-taking emo-
tive predicates and needing a recipient subdomain for the storage of non-
projected presuppositions. A subclass of the complement-taking emotive
predicates creating subsidiary subdomains is the class of weak factives,
which are still admissible when the current D contains the information that
the factive that-clause is false, in which case they relegate the presupposed
factive that-clause to the recipient subdomain of what the repository of the
emotion in question believes to be true. It also seems that, in general, the
emotive predicates creating subsidiary subdomains disallow the substitution
salva veritate of topic–comment modulation in their complement clauses.
Now consider the causative predicates. In sentence (10.51), the embedded
infinitival her seek shelter in the basement of her house carries the E-presuppo-
sitions that (a) she had a house and (b) the house had a basement. These E-
presuppositions must be projected upward on pain of the discourse becoming
inconsistent and thus incoherent. This is so because, given the metaphysical
fact that what has been physically caused has actual being, predicates like
make or cause require full incrementation of the entire content of the object
clause in the superordinate D. As a result, they are not intensional with regard
to their sentential object term:
(10.51) The bombing made her seek shelter in the basement of her house.
Sentence (10.52) is different, in that here the main predicate realize is a
strong factive—that is, it presupposes the truth of its object clause with all its
E-presuppositions and other entailments, even though it is intensional with
regard to its object clause and thus does not allow for substitution salva
veritate in it. Therefore, when used in the commitment domain, (10.52)
 Presupposition and presuppositional logic 347

requires the incrementation of its object clause before it can itself be incre-
mented, and, with its object clause, also the presuppositions thereof.
(10.52) She realized that she would never get her husband to give up
smoking.
In Karttunen’s terms, the predicate realize is, therefore, also a ‘hole’, but not
for the same reason that makes causative predicates ‘holes’.
Now to the epistemic modal predicates may and must. Epistemic may,
discussed earlier in Section 7.2.1.3, strongly projects its E-presuppositions into
any D that is compatible with them. It has the peculiar property that, on the
one hand, it does not entail its argument clause or that clause’s E-presupposi-
tions, while, on the other hand, the whole modal sentence is refused in a D
(with the concomitant knowledge base) where that argument clause or any of
its E-presuppositions have been declared false, as is shown by (10.53b) below.
Incrementation in an incompatible D of a sentence with a main predicate of
epistemic possibility results in a strong intuition of inconsistency. Yet such a
sentence does not entail its argument clause. When the current D is not
explicit on the truth or falsity of the complement clause and/or its presuppo-
sitions and thus leaves open the possibility that they are true, the E-presup-
positions are projected into a subdomain of epistemic possibility. We shall see
in a moment that the truth-functional disjunctive operator OR shares with
MAY the property of being blocked by negative information on either of its
disjuncts in D while not entailing them, merely requiring compatibility. Just
as MAY, OR strongly projects its E-presuppositions in any compatible D, but
does not entail them. What is entailed is that the disjuncts, together with their
projected E-presuppositions, are possible.
The following sentences illustrate the projection properties of epistemic
MAY:

(10.53) a. Andy may have left his car in the garage.

b. !!Andy doesn’t have a car but he may have left it in the garage.
c. Andy may have a car and he may have left it in the garage.
d. Andy may not have a car but he may (also) have left it in the
garage.
In (10.53a), the embedded infinitival ‘Andy have left his car in the garage’
presupposes that Andy has a car. This E-presupposition, though not entailed
under MAY, is not only blocked from projecting if it is negated in the higher D
but such blocking makes the whole modal sentence unacceptable, as is shown
348 The Logic of Language

by (10.53b), which is strongly felt to be inconsistent. The reason for this is that,
as argued in Section 7.2.1.3, epistemic MAY requires truth of the relevant
present knowledge state K, besides the requirement that the embedded
infinitival clause is consistent with K. This makes it inconsistent to say first
that Andy has no car and then that he left his car in the garage. But it does
allow for a knowledge state that is not explicit on whether Andy has a car,
because Andy having left his car in the garage is consistent with such a
knowledge state.
Sentence (10.53b) is thus felt to be inconsistent. (10.53c) shows that the E-
presuppositions of MAY project into a subdomain of possibility unless blocked
by negative information in D. And (10.53d) shows that ‘Andy has a car’ is not
entailed by (10.53a).
Like MAY, epistemic MUST requires a knowledge state that contains or still
admits the incrementation of ‘Andy has a car’. But unlike MAY, epistemic MUST
entails its embedded clause, because it requires for truth that the relevant
knowledge state K of the speaker is correct and that the deduction schema
followed for the MUST-statement is valid. These two conditions suffice for the
entailment of the embedded clause complete with its presuppositions. The
difference between epistemic MAY and MUST shows up in (10.54a–d). (10.54a)
and (10.54b) run parallel with (10.53a) and (10.53b), respectively. (10.54c) is
incoherent on account of the fact that it violates the domain hierarchy for
epistemic inference shown in Figure 7.3 in Section 7.3.3. The incoherence of
(10.54d) shows that (10.54a) entails ‘Andy has a car’ (according to the opera-
tional criterion developed in Section 10.3):
(10.54) a. Andy must have left his car in the garage.
b. !!Andy doesn’t have a car but he must have left it in the garage.
c. !Andy may have a car and he must have left it in the garage.
d. !Andy may not have a car but he must have left it in the garage.

10.5.3 Projection from instructional subdomains

· Conjunction 11

The conjunction predicate AND (sometimes realized as BUT) immediately and

obligatorily secures the incrementation of both conjuncts to the current D,
in the order in which they are presented. That is, the first conjunct is

11
A large part of the recent literature on the projection properties of the propositional connectives
is, though formally very elaborate, in fact irrelevant, as it fails to take into account the ecological
embedding of language in cognition and society. To take just one example, Schlenker (2007) criticizes
the dynamic approach proposed in Heim (1990) and in principle endorsed in the present book on the
grounds that (Schlenker 2007: 327–8):
 Presupposition and presuppositional logic 349

incremented first while all its non-incremented presuppositions are supplied

post hoc (accommodated). Then the second conjunct is incremented, while
all its non-incremented presuppositions are supplied post hoc in the D as
it is after the full processing of the previous conjunct. And so on for any
following conjuncts.
This means that conjunctions project their E-presuppositions conjunct by
conjunct, each conjunct being a separate incrementation unit. This is impor-
tant in the light of the often heard objection that if AND is taken to project
its E-presuppositions the way other ‘holes’ do, primary-anaphora cases
like (10.55) lead to the paradox that conjunction sentences both classically
and presuppositionally entail what is asserted in the first and presupposed in
the second conjunct:
(10.55) John has a car and [his car/it] is in the garage.
This objection loses its force when it is stipulated that AND does not inherit
any presuppositions of its argument conjuncts but treats them as independent
but successive incrementation units. Such a stipulation is, of course, not
arbitrary but follows from the fact that AND is the primary conjoiner of
successive increments.
In many cases, as the discussion in Section 8.2.1 makes clear, the order of
incrementation has an extra iconic force in that it reflects the temporal, causal,
or motivational order of the events described. In such cases we speak of
ordered interpretation. The precise conditions of ordered interpretation of
AND-conjunctions are not clear, though it appears that, at least for temporal
ordering, the conjuncts must have an event, and not a situational, interpreta-
tion. It, moreover, looks as if nonidentity of the main predicates in the
conjuncts involved plays a major role, in that it appears to be a necessary
but not sufficient condition for an ordered interpretation that the two main
predicates involved must not be identical (see examples (8.41) to (8.46) in
Section 8.2.1). Be that as it may, as far as the ‘dynamic’ projection is concerned

Heim’s dynamic semantics is just too powerful: it can provide a semantics for a variety of operators and connectives
which are never found in natural language. To make the point concrete, it suffices to observe that in Heim’s
framework one could easily define a deviant conjunction and* with the same classical content as and but a different
projection behavior […] with the order of the conjuncts reversed.

It should be obvious that Schlenker’s and* is unnatural for reasons that have nothing to do with the
formalities of presupposition projection and everything with the fact that the order of
incrementation corresponds with the order of presentation of successive utterances (with post hoc
suppletion as a functional artifice making it possible to economize on the effort of speaking). One
thus sees how research can get out of touch with reality as a result of an all too narrow formalist
approach.
350 The Logic of Language

of the presuppositions of the successive conjuncts of AND, it is safe to assume

that across the board the treatment is as sketched above, no matter whether or
not the conjunction is open to an ordered interpretation: the conjuncts are
anyway incremented in the order of occurrence and their presuppositions are,
if necessary, obligatorily projected along with them, but always after the
incrementation of the preceding conjunct.

· Negation
In the light of what has been said in Section 10.4, we can be brief about the
projection properties of the minimal and radical negation operators. The
minimal negation entails and thus obligatorily projects the E-presuppositions
of its argument L-proposition. When any such E-presupposition is incom-
patible with D, the negative sentence is inadmissible in D. In such cases, the
speaker/listener must fall back on the radical negation, which takes its argu-
ment L-proposition as a linguistic object (hence the echo-effect induced by
radical NOT) and declares it inadmissible in D on account of presupposition
failure.

· Disjunction
The disjunctive operator OR, as has been said, is in many ways like epistemic
MAY. It does not entail its disjuncts and projects their E-presuppositions as
default inferences so as to ensure maximal unity in the overall discourse
domain. But it requires a superordinate D that is compatible with the
disjuncts as well as with their negations, since there is no point in presenting
the disjuncts as possible increments if D has already banned them. When the
superordinate D is compatible with but silent about the E-presuppositions of
a disjunct, the E-presuppositions in question are projected into D as relatively
strong default inferences and they are entailed in a subdomain of epistemic
possibility: P (AND NOT-Q) OR (NOT-P AND Q) entails that both P (AND NOT-Q)
and (NOT-P AND Q) are epistemically possible.
The OR-expansion defined in (8.67) of Section 8.2.3 incorporates AND but
this makes no difference for the projection mechanism. All one has is two or
more alternative incrementation packages.

· Implication
In Section 8.2.4 we surmised that the conditional structure IF P then Q is, at a
basic-natural level, expanded to the biconditional IF P then Q AND IF Q then
P and at a strict-natural level to IF P then P AND Q, which is logi-
cally equivalent with the standard material implication. Two alternative
 Presupposition and presuppositional logic 351

subdomains are set up, one for P AND Q and one for NOT-P AND NOT-Q at a
basic-natural, and for just NOT-P at a strict-natural level. The difference with
OR is that, for implication, the relation between the overt antecedent and the
overt consequent clause is the dynamic relation that exists between successive
conjuncts. The disjunctive operator OR has no such relation between the overt
disjuncts.
The projection properties of natural language implication follow from this
description. The antecedent clause requires a D that is compatible with it and
hence with its E-presuppositions, so that the whole conditional sentence is
inadmissible in a D that contains information contrary to what is said or
presupposed in the antecedent clause. The consequent clause requires com-
patibility with D as it is after the incrementation of the antecedent clause.
Primary anaphora is thus allowed in conditionals, as shown in (10.56a). In
(10.56b), the antecedent clause presupposes that Nancy has a husband. If D
contains information that is incompatible with the news that Nancy has a
husband, the whole of (10.56b) is inadmissible in D. (10.56b) thus entails that
it is possible that Nancy has a husband and it projects the invited inference
that she has one, in accordance with the Principle of Maximal Unity proposed
in Section 7.2.2. The consequent clause of (10.56b) has the factive, and thus
entailed, E-presupposition that Nancy’s Norwegian husband is faithful. This
E-presupposition is obligatorily projected not in the D as it was before
IP started work on (10.56b) but in the D as it will be after the antecedent is
incremented, ensuring that a text like ‘Nancy has a husband. He is Norwegian.
He is faithful. She knows that he is.’ is fully sequential—that is, consistent with
the context given, orderly, and informative.
(10.56) a. If Nancy has a husband, he is Norwegian.
b. If Nancy’s husband is Norwegian, she knows that he is faithful.
All the, partly far-fetched, examples adduced in Gazdar (1979: 83–7)
and later literature regarding the projection properties of conditionals are
accounted for by the incrementation mechanism proposed here and the
ecological context in which it is placed.

10.5.4 Summary of the projection mechanism

All the properties illustrated in Sections 10.5.1 to 10.5.3 are epiphenomenal
on (a) the structure of Ds and the processes involved in incrementation, and
(b) the meanings of, or the instructions associated with, the predicates
in question. An overall survey of the entailment and projection properties
of S-embedding predicates is given in Figure 10.3, where the boxes contain
352 The Logic of Language

Entail-
ment Entailed Not entailed
scale
Projec- Requires compatible D Not refused Indifferent Requires
tion scale Refused in contrary D in contrary D to contrary D contrary D

strong 1 2 Weak 3 4 5
(emotive)
Causatives factives
Epistemic
Factives MAY Belief verbs

Projected Epistemic Emotives

OR (2x)
MUST (hope,
Antecedent complain)
Minimal
IF of antifactives
not
conditionals
counter-
weak factuals

Not pro- Verbs of Radical

jected saying NOT

FIGURE 10.3 Projection chart for E-presuppositions

predicates creating or continuing recipient subdomains in the sense defined

above.
Figure 10.3 shows the relation between entailment and projection. All
entailing predicates project obligatorily, in a descending order of strength:
causatives more strongly than factives, factives more strongly than epistemic
MUST, and epistemic MUST, it seems, more strongly than minimal NOT. Non-
entailing predicates also project, except verbs of saying and radical NOT. Verbs
of saying do not project because of the very fact that they report on what
someone has said or will say, which may be in a D that has no relation to the
current D. As a result, they are indifferent to any possible contrary D. Radical
NOT goes further in this respect. For a sentence under radical NOT to be
properly anchored, a D is required that is incompatible with one or more
E-presuppositions. For example, for (10.57) to be properly anchored, it
is necessary that D contain, or be made to contain, the information that the
E-presupposition of the L-proposition embedded under NOT saying that John
has a dog is false:
(10.57) John’s dog has NOT been put in quarantine. He hás no dog!
 Presupposition and presuppositional logic 353

The predicates that do not induce the entailment of their E-presupposi-

tions yet may project their E-presuppositions under certain conditions are
more interesting. Of these, some require a D that is compatible with their
E-presuppositions, while others allow for a D that is not. The former class
comprises epistemic MAY, OR and the conditional IF in antecedent clauses.
These pose the heaviest restrictions on D in that they cannot be anchored, let
alone project their E-presuppositions, in a D that contains information that
leads one to conclude that one or more of their E-presuppositions are false.
When D does not contain the positive information that an E-presupposition
P of any of these predicates has been stored in it (though it may be), these
predicates induce the entailment that P is possible in D, with the result that P
ends up in the recipient subdomain of epistemic possibility. Thus, sentence
(10.58a) is incoherent because, owing to the satisfaction conditions of MAY (see
Section 7.2.1.3), D must be at least compatible with the information that John
has a dog for John’s dog may be in quarantine to be incrementable. Likewise for
OR and IF of the antecedent of conditionals, as is demonstrated in (10.58b) and
(10.58c), respectively.
(10.58) a. !!John has no dog but his dog may be in quarantine.
b. !!John has no dog but his dog is either in quarantine or in the
park.
c. !!John has no dog but if his dog is in quarantine it will be well fed.
That the second sentences in (10.58a–c) do not entail that John has a dog
appears from the coherence of (10.59a–c) (see Section 10.3):
(10.59) a. John may have no dog but he may have one and then it may be in
quarantine.
b. John may have no dog but he may have one and then it is either
in quarantine or in the park.
c. John may have no dog but he may have one and then if it is in
quarantine it will be well fed.
By contrast, the predicates listed in column 3 of Figure 10.3 do allow for
anchoring in a D that is inconsistent with their E-presuppositions. Consider
the following sentences:
(10.60) a. John has no dog but he thinks he has one and that it is in
quarantine.
b. John has no dog but he thinks he has one and he regrets that it is
in quarantine.
354 The Logic of Language

c. John has no dog but he thinks he has one and he hopes that it is in
quarantine.
d. John has no dog but he thinks he has one and he is under the
illusion that it is in quarantine.
e. John has no dog but, if he had one and it was in quarantine, he
would miss it.
Sentence (10.60a) is a classical example of projection blocking: since D is incon-
sistent with the E-presupposition that John has a dog, this E-presupposition is
not projected but stays with John’s belief-domain. (10.60b,c,d) are examples
of projection blocking whereby the blocked E-presupposition is confined
to the recipient subdomain of John’s beliefs. (10.60d) instantiates the ANTIFACTIVE
main verb be under the illusion. Antifactives are predicates inducing a presuppo-
sition not of the truth but of the falsity of their embedded clauses. Examples
are be under the illusion that, lie that, falsely suggest that, or German wähnen
(used by Frege in his 1892 article). They form a neglected category, yet it is a
real one. In so far as antifactives involve a sincere but false belief, as with be
under the illusion that, their E-presuppositions are relegated to a recipient sub-
domain of belief.
(10.60e) is a counterfactual construction. Counterfactual IF carries the
precondition that the clause C in its scope has been coded in D as being
false. Thus, the sentence If John’s dog were in quarantine, John would miss
it presupposes that John’s dog is not in quarantine, while the counterfactual
IF-clause has the E-presupposition that John has a dog. Now in cases where
D says or entails that John has no dog, it also entails that John’s dog cannot
be in quarantine, thus providing a proper anchoring base for (10.60e). The
E-presupposition that John has a dog is now prevented from projecting and
must stay within the subdomain created by counterfactual IF.

10.6 The presuppositional logic of the propositional operators

Although, as has been stressed repeatedly, presupposition is not a logical but a
discourse-semantic phenomenon, it does allow for a logic to be distilled from
it. This logic is trivalent in that it has, besides truth, two kinds of falsity and
two negations. MINIMAL FALSITY (F1) ensues when an update condition of
the main predicate (which generates a classical and not a presuppositional
entailment) is not satisfied; it is repaired to truth by the default MINIMAL
NEGATION (~). RADICAL FALSITY (F2) ensues when a precondition of the main
predicate (generating a presuppositional entailment) is not satisfied; this
kind of falsity is repaired to truth by the marked RADICAL NEGATION (’). This
 Presupposition and presuppositional logic 355

presupposition-related distinction between two kinds of falsity was made, but

not further elaborated, in Dummett (1973: 425–6) and further elaborated in
Seuren (1980, 1985, 1988, 2000).
We repeat the discourse-semantic definition of presupposition in terms of a
logical condition given in (10.45) above:
(10.45) If B >> A then B ‘ A and ~B ‘ A and (~A or ~A) ‘ ~B.
Together with the semantics of radical NOT for presupposition failure, the
principle that F2 has priority over F1 and F1 over T for ∧, and that T has
priority over F1 and F1 over F2 for ∨, makes for the following truth tables for
the propositional operators concerned in strict-natural or standard proposi-
tional logic:

B B
A ~A ~A
_ ∧ T F1 F2 ∨ T F1 F2
A 1 2 4 A 1 2 4
T F1 F1 T T F1 F2 T T T T
2 3 5 2 3 5
F1 T F1 F1 F1 F1 F2 F1 T F1 F1
4 5 6 4 5 6
F2 F2 T F2 F2 F2 F2 F2 T F1 F2

FIGURE 10.4 Three-valued Presuppositional Propositional Calculus (PPropC3)

The minimal negation operator (~) makes minimal falsity true and leaves
radical falsity unaffected, whereas the radical negation operator (’) makes
radical falsity true and leaves minimal falsity unaffected, both making truth
minimally false. The ~ operator yields truth when all preconditions of the
main predicate are satisfied but at least one update condition is not. The ’
operator says that at least one precondition of the main predicate is not
satisfied. The standard bivalent negation ¬, though probably not occurring
in natural language, remains operative in that it yields truth when either a
precondition or an update condition is not satisfied. Thus, for any sentence
A, ¬A
~A ∨~A.
Conjunction (∧) is defined by the condition that radical falsity (F2) is
infectious in the sense that when at least one conjunct is radically false, so is
the conjunction. Then when neither conjunct is radically false, minimal falsity
(F1) is infectious in that, when at least one conjunct is minimally false, so is
the conjunction. A conjunction is true only when both disjuncts are true.
Conversely, disjunction (∨) is defined by the condition that truth (T) is
infectious in the sense that when at least one disjunct is true, so is the
disjunction. Then when neither disjunct is true, minimal falsity (F1) is
356 The Logic of Language

infectious in that, when at least one disjunct is minimally false, so is the

disjunction. A disjunction is radically false only when both disjuncts are
radically false. This yields the truth tables of Figure 10.4.
It is important to note that, if the condition that falsity of A leads to the
radical falsity of B were a defining condition of B >> A, then a conjunction of
the form A ∧ BA would presuppose A, because the falsity of A makes the
whole conjunction radically false. Yet we do not want to say that A ∧ BA >>
A, because then A ∧ BA would both assert and presuppose A. Thus, even
though it follows from the logic of conjunction that conjunctions of the form
A ∧ BA are radically false when A is (minimally or radically) false, this does
not mean that B presupposes A. The reason is that presupposition is not a
logical but a discourse-semantic notion, and the latter takes precedence over
the former. The conjunctive sentence operator AND may have a logic as an
emerging property, but it is defined as a discourse-incrementer: ‘A and B’
carries the instruction, and thus means, ‘increment A first and then B’, as
specified in Section 8.2.1. And since presupposition is defined as a condition
on preceding discourse, A ∧ BA can only have a presupposition C if C is
required in the discourse that precedes A ∧ BA as a whole.
In Figure 10.4, the slots for conjunction and disjunction have been given
numbers 1–6. This has been done to render it possible to state the various
valuation spaces and hence the entailment relations holding in the system.
There are nine valuations (32 for two L-propositions and three truth values),
but only six VSs are needed, since the operators ∧ and ∨ are symmetrical. The
numbers in the slots of Figure 10.4 correspond to the following combinations of
T, F1, and F2:

VS: 1 2 3 4 5 6
A T F1 T F1 F2 T F2 F1 F2
B T T F1 F1 T F2 F1 F2 F2

We use the following notation:

AND: A∧B ~AND: ~(A ∧ B)
OR: A∨B ~OR: ~(A ∨ B)
AND*: ~A ∧ ~B ~AND*: ~(~A ∧ ~B)
OR*: ~A ∨ ~B ~OR*: ~(~A ∨ ~B)
’AND: ’(A ∧ B) AND*: ’ A ∧ ’B
’OR: ’(A ∨ B) OR*: ’ A ∨ ’B
’AND*: ’(~A ∧ ~B) ………
’OR*: ’(~A ∨ ~B) ………
This gives the VS-model of Figure 10.5. The circle around space 3 in Figure
10.5 has been printed bold because the valuation spaces 1, 2, and 3 form
 Presupposition and presuppositional logic 357

AND*_

_~AND _
~AND*
_~AND*
_~AND

_~AND _
~AND*

~AND AND*

~AND ~AND*

AND ~AND*
6 5 4 3 2 1 2 3 4 5 6
OR ~OR*
OR OR*

~OR OR*

OR ~OR*

~OR UR OR*
_
~OR OR*_ _
~OR*

OR*_
U
OR*_

FIGURE 10.5 Valuation-space representation of strict natural PPropC3

the presuppositionally restricted universe UR of situations causing no pre-

supposition failure and thus no radical falsity. Figure 10.6 gives the
polygonal representation of PPropC3, but without the radical negation (the
superscript ‘R’ over the metalogical relations restricts the validity of the
relation to the spaces 1, 2, and 3). Within UR standard propositional logic is
valid (with standard ¬ for the presuppositional minimal negation ~). When
all spaces are taken into account, one sees that ~(A ∧ B) ‘ ~A ∨ ~B but not
vice versa, while ~A ∧ ~B ‘ ~(A ∨ B) but not vice versa. Also, ~A ∨ ~B
entails none of the other admissible conjunctions or disjunctions or their
minimal or radical negations. With minimal negation, De Morgan’s laws thus
do not hold but have been reduced to one-way entailments.
A polygonal representation of the entire system may result in an aestheti-
cally pleasing picture of a polygon with 18 vertices, each standing for a basic
expression with and without minimal or radical external and/or internal
negation, but it would seem that such a labour of love should rather be left
to the true devotees who are interested in the logical system as such, regardless
of what is required for the use it is put to in natural linguistic interaction.
358 The Logic of Language

~AND {2,3}

{1} CDR SCR {1,4}

AND ~OR*
C

C C CDR

SCR C
{1,2,4} {2,3,5}
OR SCR OR*
C
C

CDR

SCR

~OR AND*
SCR {3}
{3,5} CDR
CDR
~AND*
{1,2}

FIGURE 10.6 Octagonal representation of strict-natural (= standard) PPropC3 without

radical negation

Meanwhile, the reader will quickly ascertain that the following expression
types of PPropC3 correspond with the valuation spaces specified:
/AND/ = {1} /~AND/ = {2,3} /~AND/ = {4,5,6} /AND*/ = {6}
/OR/ = {1,2,4} /~OR} = {3,5} /~OR} ¼ {6} /OR*/ ¼ {4,5,6}
/AND*/ ¼ {3} /~AND*/ ¼ {1,2} /~AND*/ ¼ {4,5,6} ... ... ...
/OR*/ ¼ {2,3,5} /~OR*} ¼ {1,4} /~OR*} ¼ {6} ... ... ...
One also finds that the following duality relations hold between AND and OR:
~OR
AND*
~OR* ~AND
OR*
~AND*
That is, AND and OR are duals only under the radical negation. Moreover,
De Morgan’s laws do hold for the radical negation ~ and also for the standard
negation ¬ but not for the minimal negation ~, as is easily shown. This means
that PPropC3 is isomorphic with standard bivalent propositional calculus for
the negation operators ~ and ¬, but not for the minimal negation ~. Figure
10.6 shows immediately that, for the minimal negation, AND ‘ ~OR* but not
 Presupposition and presuppositional logic 359

B B
∨ T F1 F2 • T F1 F2
A 1 2 4 A 1 2 4
T F1 T T T F1 F1 F2
2 3 5 2 3 5
F1 T F1 F1 F1 F1 T F2
4 5 6 4 5 6
F2 T F1 F2 F2 F2 F2 F2

FIGURE 10.7 Truth tables for OR (∨) and NOR (•) in basic-natural PPropC3

vice versa, and ~AND* ‘ OR but not vice versa. Moreover, ~AND ‘ OR* and
AND* ‘ ~OR, but not vice versa.
As is shown in Weijters (1985), PPropC3 can be expanded to PPropCn with
n–1 negations and n truth values, for n > 1. Standard propositional calculus is
thus seen to be just the extreme minimal instance of PPropCn, while PPropC3
represents the three-valued variant. In Seuren et al. (2001) it is shown that the
Kleene calculus (Kleene 1938, 1952) is likewise the three-valued variant of a
system with, in principle, an unlimited number of truth values between, not
beyond, any given pair of successive truth values. The Kleene set of many-
valued propositional logics can be combined with the presuppositional set
of many-valued logics, resulting in a system of propositional logic with an
unlimited number of definite truth values and an equally unlimited number
of values that are intermediate between any pair of successive definite truth
values. This rather extends the number of possible propositional logics
natural language could choose from when it started on its evolutionary path.
Basic-natural PPropC3 differs in certain respects from its strict-natural or
standard counterpart. The operators ~, ~ and ∧ are not affected, but the
operator ∨ is. Moreover, as one recalls from Section 3.4.2, there is the further
·
operator NEITHER NOR, symbolized here as ‘ ’, and occurring in the sentence
type NOR, which is defined as ~A ∧ ~B. The three-valued truth tables for
basic-natural PPropC3 are as given in Figure 10.7.
Basic-natural (exclusive) OR differs from strict-natural or standard OR in that it
produces truth only when one of the constituent L-propositions is true. Given
this, it is still so that T has priority over F1 and F1 over F2. Basic-natural NOR is
defined on the basis of the assumption that the negations involved are minimal
negations. On this assumption, NOR produces truth only when both (all) constit-
uent L-propositions have the value F1—that is, in space 3. Given this, the principle
holds that F2 has priority over F1 and F1 over T (as for the conjunctive operator
∧). The corresponding VS-model is shown in Figure 10.8.
Here we see that NOR is equivalent with AND*, both sharing the VS {3}.
Moreover, OR* entails ~AND but not vice versa, since {2}  {2,3}. This is of
360 The Logic of Language

AND*_

_~AND _~AND*

_
~AND*
_~AND

_~AND _~AND*

~AND AND*

~AND ~AND*

AND ~AND*
6 5 4 3 2 1 2 3 4 5 6
~OR ~OR*
~NOR
OR OR*

~OR ~NOR
~OR*

OR NOR ~OR*
R
~OR U OR*_
OR*
_
~OR _
~NOR _
~OR*
OR*_
_
U ~NOR
OR*_
_
~NOR

FIGURE 10.8 Valuation-space representation of basic-natural PPropC3

special interest in the light of examples (3.24) and (3.25) of Section 3.4.2,
repeated here as (10.61) and (10.62):
(10.61) a. He doesn’t like planes or trains.
b. He doesn’t like planes and he doesn’t like trains.

(10.62) a. He doesn’t like planes and trains.

b. He doesn’t like planes or he doesn’t like trains.
The point is that (10.61a) is naturally reduced to (10.61b), while a reduction of
(10.62a) to (10.62b) is highly unnatural and requires a good deal of reflection.
In Section 3.4.2 it was shown that in bivalent basic-natural propositional logic
(10.61a) is equivalent with (10.61b), whereas (10.62a) is not equivalent with
(10.62b). Here we see that the same holds in trivalent basic-natural proposi-
tional logic, so that the explanation given in Section 3.4.2 for this difference
can be maintained.
But is natural language negation really ambiguous between a minimal and
a radical negation as defined above? Horn (1985) does not think it is. Horn
 Presupposition and presuppositional logic 361

first posits that a distinction must be made, in natural language, between a

‘descriptive’ and a ‘metalinguistic’ negation, the former being negation as
used in the normal flow of speech and meant to deny the truth of the
proposition in its scope, while the latter is used to deny the appropriateness
of specific words or expressions, the phonology or the pronunciation, or
indeed of the whole sentence. Examples of metalinguistic negation, in
Horn’s sense, are:
(10.63) a. No Johnny, aunt Bessie isn’t SPLITTING tomorrow, she is LEAVING.
b. He doesn’t hate SOME of his pupils. He hates them ALL.
c. He is not in his 0læb´r´ 0tori but in his l´ 0bor´t´ri.
d. It is NOT sad that she died so young. She is still very much alive.
While it is certainly correct and very necessary to distinguish between a
descriptive and a metalinguistic use of negation and to classify the radical
negation as being metalinguistic on account of the echo-effect it provokes, it
is highly doubtful whether the radical negation, as exemplified in (10.63d) (as
well as in (10.12), (10.13) and many other examples above), is sufficiently
accounted for by placing it in one ‘natural class’ along with cases of lexical
selection or phonological realization such as (10.63a–c). The reason for this
doubt lies in the fact that, as has been pointed out, the negation that cancels
presuppositions cannot occur in any other position in the sentence than what
has been called the canonical position, which, for English, is the position in
construction with the finite verb form (see category B in Section 10.4:
negation in noncanonical positions necessarily preserves presuppositions).
The point is that the metalinguistic negation as exemplified in (10.63a–c) does
not have to be in the canonical position, witness, for example, sentences like
the following:
(10.64) a. Not several but all guests left after the row.
b. Not Lizzy, please, but Her Majesty the Queen is wearing a
funny hat. (cf. Horn 1985: 133)
In (10.64a,b) the negation precedes the surface subject, and is, therefore,
not in the canonical position. One notes, moreover, that the quantifier several,
which is, as we have seen, a PPI, does not function as a PPI here, apparently
because the word used is not several but its quoted counterpart ‘several’. If it
had been a PPI here, the negation would have had to occupy the canonical
position, and would have been presupposition-cancelling. This difference is
real, and considerably weakens Horn’s thesis that all cases of utterance
correction form one natural class, which must, therefore, be accounted for
uniformly. Note, for example, the difference between (10.65a), which is an
362 The Logic of Language

acceptable case of presupposition-cancelling, and (10.65b,c), which are not,

because of their being inconsistent:
(10.65) a. He did NOT only lose a thousand pounds. He lost nothing at all!
b. !Not/NOT only did he lose a thousand pounds. He lost nothing at
all!
c. !He not/NOT only lost a thousand pounds. He lost nothing at all!
A theory like Horn’s will have to explain why (10.65b,c) do not work, while
(10.64a,b) do.
This is not just a grammatical problem (though, if it were, it would be
serious enough), it is also a semantic problem. For, contrary to what this
theory predicts, the negation over ‘ordinary’ cases of metalinguistic negation,
correcting lexical choice or phonological realization, does not cancel presup-
positions. Take, for example, (10.63a,b) above, and try replacing the second
sentence—that is, the correction—by a presupposition denial. The result is
unacceptable:
(10.66) a. !No Johnny, aunt Bessie isn’t SPLITTING tomorrow. There ı́s no
aunt Bessie!
b. !He doesn’t hate SOME of his pupils. He doesn’t exist!
It seems unlikely, therefore, that presupposition denials belong in one
single class with the other cases of metalinguistic negation. A further distinc-
tion appears to be called for, setting presupposition denials apart from other
cases of metalinguistic negation.
Then, there is the problem why English, and with it all known languages,
does not systematically distinguish between the two functions reserved for
negation—a problem raised by Gazdar (1979: 65–6):
But no language, to the best of my knowledge, has two or more different types of
negation such that the appropriate translation of John doesn’t regret having failed
could be automatically ‘disambiguated’ by the choice of one rather than the other.

This is a problem that plagues all theories of ambiguous not. One might
counter this argument by saying that such ‘universal’ ambiguities do occur.
No language is known, for example, to distinguish formally between the two
senses of:
(10.67) There’s a fly in the middle of the picture.
Such ambiguities are likely to come with the language machine the human
race is natively endowed with. Even so, however, it must be admitted that a
theory that manages to subsume all different varieties of negation under one,
 Presupposition and presuppositional logic 363

formally precise umbrella definition, covering the logical differences as well as

the differences of a more pragmatic nature, is preferable to a theory that
makes negation (multiply) ambiguous without any unifying formula. No
such umbrella definition, however, has been found to date.

10.7 The presuppositional logic of quantification

10.7.1 The presuppositional version of the Square and of SMPC
We define the universal quantifier in trivalent presuppositional predicate
calculus (PPredC3) as in (10.68) (one recalls that the colon is followed by
the preconditions and the upright stroke by the update conditions of the
predicate defined):
(10.68) [[8]] ¼ {< [[G]],[[F]] > : [[F]] satisfies the preconditions of G j [[F]] 
[[G]]}
(The universal quantifier 8 denotes the set of all pairs of predicate
extensions [[G]] and [[F]] such that (precondition) all elements of
[[F]] satisfy the preconditions of the predicate G, and (update
condition) [[F]] is included in [[G]].)
The quantifier 8 thus inherits the preconditions of the matrix predicate G,
which are to be satisfied by the members of the set denoted by the restrictor
predicate F. As a result of the presuppositionalization of the universal quantifier
the distinction we were forced to maintain between strict-natural ABPC and
constructed SMPC has now vanished. When 8 in SMPC is presuppositionalized
as in (10.68), it turns out to be identical with ABPC, whose undue existential
import has now been subsumed under, and solved by, the general precondition
that the elements in [[F]] should satisfy the preconditions of the matrix predicate
G. For, if [[F]] ¼
, then the truth value of type-A sentences will depend on
whether the G-predicate is or is not extensional with respect to the restrictor
term. If G is extensional with respect to the restrictor term and thus requires that
[[F]] ¼
6
(in a purely extensional version of the logic concerned), then ALL F is G
will be radically false when [[F]] ¼
. And likewise for all other possible
preconditions of G. This is the force of the Cartesian argument Cogito ergo
sum, because cogitare (think) is extensional with regard to its subject term.
In fact, when the logic is properly intensionalized, as it should be, the
F-class will never be null, because the thought underlying any properly
contextually and situationally integrated utterance by definition creates in-
tensional objects, which allow for reference and quantification as naturally as
actually existing extensional objects do. There will, therefore, always be a
364 The Logic of Language

nonnull intensional F-class to fall back on for reference or quantification. All

that remains in a properly intensionalized predicate logic is a precondition of
actual existence for extensional argument positions. The problem of UEI has
thus been generalized to preconditions of predicates.
One notes, in particular, that we do not follow the advice, usually given by
well-meaning standard logicians (see Section 4.1) to presupposition theorists,
to define 8 as in (10.69), as if the universal quantifier induced an entailment of
actual existence, thereby avoiding truth for cases where [[F]] ¼
:
(10.69) [[8]] ¼ {< [[G]],[[F]] > j [[F]]  [[G]] and [[F]] 6¼
}
One reason for not following this advice is simple: it is not the universal
quantifier but the matrix predicate G that is held responsible for the problems
arising in connection with the situations where [[F]] ¼
. In fact, the
standard-logicians’ advice is a little careless, because following it up would
mean that NOT ALL F is G would come to mean ‘either some F is not-G or
there are no Fs’. And if one then wants to keep the Conversions, SOME F is
NOT-G would mean the same, which is absurd (see Section 5.2.4 for a more
detailed discussion).
Our solution to the problem of UEI is different. In accordance with the
hypothesis that all presuppositions, including those of actual existence, are
derived from the preconditions of matrix predicates, we say that the problem
of UEI is solved by an appeal to the semantic description of the matrix or
G-predicate. The source of existential import is thus located in the existential
precondition of the matrix predicate. A sentence like (10.70a) has existential
import with respect to the set of all houses because it presupposes the
existence of at least one house (or, rather, at least two houses, since ALL is a
plural determiner), not because of the semantics of the universal quantifier
but because the matrix predicate be damaged is semantically (lexically) de-
fined as being extensional with regard to its subject term.
(10.70) a. All houses were damaged.
b. All mermaids are a product of some writer’s imagination.
By contrast, (10.70b) does not have existential import and does not presup-
pose the actual existence of mermaids, because the matrix predicate be a
product of some writer’s imagination is nonextensional (intensional) with
regard to its subject term. Thus, sentence (10.70a) is radically false when
there are no houses, but (10.70b) is not radically false just because there are
no mermaids. In the world as it is, (10.70b) is either true or minimally false,
depending on whether it is indeed so that all mermaids are a product of some
 Presupposition and presuppositional logic 365

writer’s imagination. If some particular mermaid was not thought up by a

writer, but, say, by a sculptor, (10.70b) is minimally false.
This system of truth-value assignments is thus not restricted to cases of
existential preconditions. It covers all preconditions of any lexical predicate in
matrix position. Thus, a sentence like (10.71) is likewise radically false, not
because of a failure to satisfy the existential precondition of the predicate sing,
which requires the actual existence of at least some singer(s), but because of
the failure to satisfy some other precondition of the same predicate, in this
case the precondition that those who do the singing should be entities capable
of making vocal noises—a property one can hardly assign to apples other than
in a metaphorical sense.
(10.71) All apples were happily singing cantatas.
Failure to satisfy the preconditions of the matrix predicate is considered to
be the only source of radical falsity. The lexical matrix predicate imposes its
preconditions on its terms regardless of whether these terms are universally
quantified or definite. If this is a valid principle, as I propose it is, the
quantifiers have no bearing on the presuppositional properties of sentences;
only the lexical matrix predicate has. The presuppositional variety of predi-
cate logic is thus not only relevant for just the logical constants but has a
direct and all-pervasive bearing on the logical analysis of language as a whole,
since presuppositional preconditions adhere to almost every predicate in the
lexicon of every language. Logic has thus become an integral part of the
semantic description of languages, not as a modelling tool, as in model-
theoretic formal semantics, but as part of the language machinery itself.
So far, we have only looked at the universal quantifier, which was sufficient
to eliminate the problem of UEI. But since we must define the whole of
PPredC3, the question now presents itself whether the existential quantifier
should likewise be taken to ‘inherit’ the preconditions of the matrix predicate
or whether it should stay as in nonpresuppositional ABPC or SMPC. This is a
choice we are facing: either we don’t do anything and define ∃ as in standard
predicate logic, as in (10.72a), or we add the same precondition as has been
added to the universal quantifier, as in (10.72b):
(10.72) a. [[∃]] ¼ {<[[G]],[[F]]> j [[G]] \ [[F]] 6¼
}
(The existential quantifier ∃ denotes the set of all pairs of
predicate extensions [[G]] and [[F]] such that the intersection of
[[G]] and [[F]] is nonnull.)
b. [[∃]] ¼ {<[[G]],[[F]]> : [[F]] satisfies the preconditions of G j [[G]]
\ [[F]] 6¼
}
366 The Logic of Language

(The existential quantifier ∃ denotes the set of all pairs of

predicate extensions [[G]] and [[F]] such that (precondition)
all elements of [[F]] satisfy the preconditions of the predicate G,
and (update condition) the intersection of [[G]] and [[F]] is
nonnull.)
It is hard to decide between the two. The assignment of radical falsity to a
sentence like No mermaid has a bank account or its paraphrase There is no
mermaid that has a bank account appears counterintuitive. But if these
sentences are considered true and some is considered to be preconditionless,
which seems to agree with natural intuitions, then No mermaid does not have
a bank account or its paraphrase There isn’t a mermaid that hasn’t a bank
account should also be true, according to Figure 10.9a. But this runs counter to
natural intuition, since, for natural intuition, the latter two sentences are
equivalent with All mermaids have a bank account. According to Figure 10.9b,
this equivalence does indeed hold, since, in Figure 10.9b, /A/ ¼ /~I*/. But then
Some mermaid has a bank account and No mermaid has a bank account should
both be reckoned to be radically false.
Perhaps we should opt for two varieties of the existential quantifier,
one represented by bare some and one by some of the, whose negation is
none of the. Bare some would then be preconditionless and be defined as in
(10.72a), corresponding to the VS-model in Figure 10.9a, while some of
the would be defined as in (10.72b), corresponding to the VS-model in
Figure 10.9b. Minimal falsity would then be assigned to Some mermaids
have a bank account and truth to No mermaid has a bank account. But
both Some of the mermaids have a bank account and None of the
mermaids has a bank account would be radically false. If this solution is
adopted, the radical falsity of Some of the mermaids have a bank account
parallels that of All mermaids have a bank account and The mermaids
have a bank account, which has a definite, nonquantified subject term. The
Conversion law A
~I* would then be restricted to some of the and not
extend to bare some.
It seems sensible, given the still tender state of the theory and in particular
our still deficient knowledge of the difference between some and some of the
and between all and all of the, to leave the question undecided for the time
being. Either way, the general principle regarding the source of existential
import can be maintained:
Existential import as regards the extension of an argument term t of a
predicate G derives solely from the fact that G is extensional with
 Presupposition and presuppositional logic 367

a. ~ ~A b. ~
– ~A
–
~A ~I* ~A ~I*
–
– –
~A I* ~A I*
~A I* ~A I*
A ~I* A ~I* ~– ~I*
4 3 2 1 2 3 4 4 3 2 1 2 3 4
I ~A* ~
– ~I I ~A*
I ~A* I ~A*
~I A* ~I A*
~I U' ~A*
– ~I
– U' ~A*
–
~ ~A* ~
– ~A*
U – U

FIGURE 10.9 Two candidates for the VS-modelling of PPredC3

regard to t. When G is nonextensional (intensional) with regard to t,

there is no existential import with regard to t.
The difference between bare some and some of the would then be that for
bare some existential import is a C-entailment whereas for some of the it is a
P-entailment.
The spaces 1, 2, and 3 form the restricted universe UR, the class of situations
where all presuppositions are fulfilled: the presuppositional subuniverse of UR.
Space 4 in Figure 10.9b contains only radically negated positive or negative
sentences. In Figure 10.9a, by contrast, the radical negations in space 4 are
restricted to A and A* (and their minimal negations) whereas I and I* are
minimally negated, as a result of the fact that (10.72a) specifies no precondi-
tions for ∃ so that I and I* cannot be radically false. In both cases, UR
comprises precisely those situations where all presuppositions are satisfied.
ABPC (the Square) holds as long as the logic is restricted to UR. If this analysis
is correct, we have saved the Square from logical disaster: UEI no longer
spoils the system. And it does not correspond too badly with natural intui-
tions. We can thus eat our pudding and have it.
One notes that both the internal and the minimal external negations are
otiose when preceded by the external radical negation: ’A
’A*
’~A

’~A*. This follows from the fact that the radical negation ’ merely says that
its argument proposition contains a presuppositional error, so that the
discourse as built up through prior utterances must be repaired. This analysis
is consistent with the observation made above (category C in Section 10.4)
that negations in the scope of a higher operator are necessarily presupposi-
tion-preserving. That being so, the internal negation must be minimal.
368 The Logic of Language

Consider the A-form ALL F is G and the corresponding A*-form ALL F is NOT-
G. Since NOT in NOT-G is presupposition-preserving, the complement of [[G]]
is restricted to UR, within which all preconditions of the matrix predicate G
are satisfied. Given that under radical negation truth results only when one or
more of the preconditions of G are not satisfied, and since the preconditions
of G and NOT-G are identical, it follows that the radical negation makes all
eight basic sentences types of the calculus not containing the radical negation
equivalent.

10.7.2 The presuppositional version of BNPC

Let us now see what happens when a presuppositional component is added to
the basic-natural predecessor of ABPC, BNPC, resulting in basic-natural
PPredC3 or BNPPredC3. As before, when the elements in [[F]] fail to satisfy
the preconditions of the G-predicate, radical falsity is assigned to A and ~A
but truth to ~I and ~I*, and also to N and N*. The satisfaction conditions for
the quantifiers ALL, SOME, and NO in BNPPredC3> are thus as follows ([[F]] and
[[G]] are distinct natural sets):
(10.73) a. [[8]] ¼ {< [[G]],[[F]]> : [[F]] satisfies the preconditions of G j [[F]]
 [[G]]}
(The universal quantifier 8 denotes the set of all pairs of
predicate extensions [[G]] and [[F]] such that (precondition) all
elements of [[F]] satisfy the preconditions of the predicate G, and
(update condition) [[F]] is properly included in [[G]].)
b. [[∃]] ¼ {< [[G]],[[F]]> j [[G]] O
O [[F]] or [[G]]  [[F]]}
(The existential quantifier ∃ denotes the set of all pairs of
predicate extensions [[G]] and [[F]] such that [[G]] and [[F]] are in
mutual partial intersection or [[G]] is properly included in [[F]].)
c. [[N]] ¼ {< [[G]],[[F]]> j [[F]] OO [[G]]}
(The negative existential quantifier N denotes the set of all pairs
of predicate extensions [[G]] and [[F]] such that [[G]] and [[F]] are
in total mutual exclusion.)
Again, the variables [[F]] and [[G]] do not have to be restricted to natural
sets: the logic will simply assign a truth value to any given admissible
expression keyed to any given situation. One notes that 8 is defined as
requiring proper inclusion of [[F]] in [[G]], while ∃ requires that [[F]] and
[[G]] M-partially intersect. This makes /A/ and /I/ mutually exclusive, so that
A and I are contraries.
 Presupposition and presuppositional logic 369

a. b.

[[F]] = Ø
~A ¬A

~A ¬A
~I N ¬I N
~I ~A N ¬I ¬A N
I ~N I ¬N
A A
~I ~N ¬I ¬N
4 3 2 1 2 3 4 4 3 2 1 2 3 4
N* ~I* N* ¬I*
~N* ~A* I* ¬N* ¬A* I*
~N* ~I* ¬N* ¬I*
~A* ¬A*
N* ~I* N* ¬I*
A* A*

UR ~A* ¬A*
U U

Figure 10.10 VS-model for BNPPredC3, compared with bivalent BNPC

Figure 10.10a shows the VS-model for BNPPredC3. Figure 10.10b, which
represents bivalent basic natural predicate calculus (BNPC) extended to a full
logic, is repeated from Figure 3.8b to enable the reader to compare the two.
The spaces 1 to 3 in Figure 10.10a,b are reserved for situations where the
preconditions of the G-predicate are satisfied, while space 4 covers those
situations in which they are not (in Figure 10.10b only existential presupposi-
tions are taken into account). As with bivalent BNPC, the spaces 1 to 3
are defined for the following conditions: space 1 for [[F]]  [[G]], space 2 for
O [[G]] or [[G]]  [[F]], space 3 for [[F]] OO [[G]].
[[F]] O
Now we can revert to the problem discussed in Section 4.4 of the intuitive
equivalence of ~A with both I and I* (intuition 7). If it is assumed that, at least
prototypically, NOT ALL F is G is interpreted as the topic–comment (cleft)
structure ‘the F that is G is not all F’, then NOT ALL F is G presupposes and
thus entails SOME F is G and also SOME F is NOT-G, as the two are equivalent.
This establishes the entailment from ~A to both I and I*. Conversely, both I
and I* entail ~A because that is how they are defined in the system of Figure
10.10a: /I/ ¼ /I*/ ¼ {2,4} while /~A/ ¼ {2,3,4}, which means that I/ I* ‘ ~A. We
thus have entailment both ways between ~A and I or I*, and hence equiva-
lence.
For PpredC3 this result is not attainable. As one can read from Figure 10.9a,
the conjunction of ~A with its presupposition I gives the VS {2}, whereas
/~A/ ¼ {2,3}, which makes for a one-way entailment from I to ~A, and no
equivalence. Of course, ~A and I* are equivalent because the Conversions
370 The Logic of Language

hold in PpredC3 (with minimal negation). Nor is this result attainable for any
presuppositional form of AAPC, as is easily checked.
The conclusion is that, provided one lets ~A presuppose I on account of
topic–comment structure, we may add intuition 7 (NOT ALL F is G
SOME F
is G
SOME F is NOT-G) to the score of BNPC, now transformed into
BNPPredC3, as specified in Section 4.4, which makes BNPPredC3 the clear
overall winner of the systems considered. The only intuition still missing is
intuition 3 (SOME F is G
SOME G IS F or I
I!), but, as has been said, here
we have no remedy. All we can say is that this particular intuition is weaker
than the others and may be due to the fact that cases where SOME F IS G is true
while [[G]]  [[F]] are relatively rare and untypical, which may lead intro-
specting subjects to overlook them. But even if that is so, we have no
explanation for the fact that PNST–5 apparently does not translate directly
into BNPC or its presuppositional counterpart BNPPredC3.

10.7.3 The victorious Square

The upshot of our trek through the jungle of predicate logic is more or less as
follows. The presuppositional variant of BNPC, BNPPredC3, is the richest
logic discovered so far, logically sound and empirically the most successful.
But it suffers from the fact that it requires full knowledge of the verification
domain before existential statements can be valued ‘true’ or ‘false’ (since
A must have been valued before I can be). This is not a logical defect but
it hampers inductive cognitive processes, and thus the advancement of
knowledge.
The presuppositional version PpredC3 of its more advanced sister, ABPC
(the Square) is now also logically sound, owing to the added presuppositional
component. And it is still next in line as regards empirical success (Section
4.4). This has made it worthwhile to see if the Square cannot be, or should not
be, salvaged from the logical scrapheap to which it has been relegated by
modern logicians on account of its suffering from UEI. This rescue operation
has proved remarkably rewarding, in that presupposition theory appears to be
able not only to restore the Square to logical soundness by the addition of a
presuppositional component but also to provide an explanatory basis for the
fact that utterances fit naturally into specific given contexts and/or situations.
In fact, language proves to be tailored to the needs of contingency in that its
predicates are, for the most part, purpose-made for specific classes of refer-
ence objects and hence of situations. The mere use of a specific predicate may
suffice to limit the set of objects one wishes to talk about and the set of
situations the utterance in question fits into (as one composes music but writes
 Presupposition and presuppositional logic 371

a song). The most powerful predicate logic allowing for existential statements
without complete situational knowledge, the Square of Opposition, is thus
seen to be operative precisely where it needs to be, in the actual, restricted
contexts and situations that utterances are used in. Psychologists rightly speak
about shared speaker–hearer’s knowledge to explain the fact that so much can
be left unsaid. Here we have another device that makes language as efficient as
it is.
But there is a price to be paid for the logical soundness of the Square. First,
it appears necessary to distinguish three truth values for natural language,
true (T), minimally false (F1) and radically false (F2), the two varieties of
falsity being contingent upon whether or not it is a lexical update condition or
a lexical precondition that has remained unsatisfied. Correspondingly, two
varieties of natural language negation are distinguished, the minimal negation
to repair minimal falsity, and the radical negation to repair radical falsity.
Radical falsity is now interpreted as a cry for discourse correction, which
explains the metalinguistic ‘echo’ the radical negation provokes in actual use.
The resulting trivalent propositional calculus (which differs from that pre-
sented in Kleene 1938, 1952) incorporates standard propositional logic as long
as one stays within the boundaries set by presuppositional conditions. Anal-
ogously, the resulting new trivalent predicate calculus incorporates the Square
as long as there is no presupposition failure. The Square has now been cured
of its logical disease while, at the same time, it has been made to fit into
a flexible, dynamic system of discourse-restricted linguistic interaction—
something no human or superhuman engineer would have thought of,
presumably. In this perspective, the trivalence of natural logic is not really a
concession to the facts of language and cognition. On the contrary, it is a
small investment yielding a huge profit.
A second price to be paid arises from the fact that the logico-semantic
system envisaged and partly developed here requires a nonextensional ontol-
ogy containing ‘Meinongian’ virtual or intensional objects which derive their
being from the fact that they are the product of the cognitive powers of
imagination and representation. They have been thought up one way or
another and thus combine specific identity with a specification that is not
complete on all relevant parameters, as opposed to the complete specification
that is a necessary characteristic of actually existing objects. This will perhaps
cause some consternation among Anglo-Saxon philosophers trained in the
tradition of Russell and Quine. But then, there are plenty of reasons why the
purely extensional ontology that is standard in the Western world is anyway
hopelessly inadequate if one wants to achieve a proper semantic theory of
natural language, regardless of any requirement issuing from a trivalent
372 The Logic of Language

logical system. Here again, therefore, if there is a price to pay, that price is not
a concession to Ockham’s principle of minimal assumptions or Quine’s
(anyway unlovable) ‘desert landscape’, but much more an investment that is
necessary for an adequate insight into the nature of human language and
cognition.

10.8 The attempt at equating anaphora with presupposition

In circles associated with Kamp’s Discourse Representation Theory (DRT;
Kamp 1981, Kamp and Reyle 1994), the notion has spread that presupposition
is a form of anaphora. This notion is due in the first instance to Van der Sandt
(1992), where it is proposed that there is one single mechanism underlying
both anaphora resolution and presupposition projection. Those readers who
have seriously worked through Chapters 9 and 10 will be startled at this
proposal, as the two do not look as if they had anything in common. And
indeed, it is argued here that this identification is just an instance of inexpert,
unprofessional botching-up of otherwise good notions.
Rolling anaphora and presupposition into one is a priori an extremely
unlikely and hazardous enterprise which will require a very solid underpin-
ning indeed for it to be successful. However, Van der Sandt (1992) provides no
underpinning at all. On the contrary, he equates anaphora with external, or
perhaps with primary, or perhaps even with donkey anaphora, and he equates
presupposition with presupposition projection, thus confusing the phenom-
enon itself with a mechanism associated with it. Having thus blurred the
notions concerned, he then claims that the two are identical. This alleged
‘identity’ can, therefore, only be blurred itself.
DRT-practitioners accepted this all too readily, as they have always felt
diffident with regard to presuppositions, either because they consider pre-
suppositions to be of a nonformal, pragmatic nature and thus to be someone
else’s business or, if presuppositions are formally tractable, because they
threaten to drag one down the slope of nonstandard logics and perhaps
even of nonstandard mathematics. Van der Sandt’s ‘solution’ thus couldn’t
have come at a better moment.
Van der Sandt is curiously vague about the nature of presupposition. His
paper opens as follows (Van der Sandt 1992: 333): ‘The traditional view on
presupposition has it that presuppositions are referring expressions.’ This is
odd, to say the least. Since Strawson (1950)—the starting point of presuppo-
sitional studies (see Section 10.1.1)—there have been various approaches or, if
one likes, traditions. One can distinguish a tradition that continues the lines
drawn by Strawson and is Fregean in nature (Section 10.1.3), a tradition
 Presupposition and presuppositional logic 373

continuing the lines set out in Russell (1905) for the analysis of definite
descriptions (Section 10.1.2), a tradition attempting to reduce presuppositions
to pragmatic phenomena (Wilson 1975) and a discourse-semantic tradition,
defended in the present study, reducing presuppositions to lexical precondi-
tions, together with a trivalent logic and two kinds of falsity (Dummett 1973:
425–6; Seuren 1985, 1988, 2000, and many other publications). Given this
diversity of opinions on presuppositional phenomena, one is mystified by
Van der Sandt’s expression ‘the traditional view’.
Moreover, calling presuppositions ‘expressions’, and ‘referring expressions’
at that, is something that no current of thought on presuppositions, tradi-
tional or not, has, to my knowledge (based on forty years of research in
presupposition theory), ever been daring enough to do. It would have helped
if Van der Sandt had mentioned some references, to jog his readers’ memory
or let them find out what could be meant, but, of course, he does not, because
there aren’t any. Nor does he give any further explanation of this highly
original thought. In other words, the first sentence of Van der Sandt’s paper
is simply bizarre.
But perhaps we shouldn’t worry too much about the supposedly ‘tradi-
tional’ but in fact unidentifiable, view, because a few lines down on the same
page we read: ‘For Frege it is referring expressions that give rise to presuppo-
sitions.’ In the light of the previous quote, this should be read (in the
presumed but unidentifiable ‘traditional view’) as: ‘For Frege it is referring
expressions that give rise to referring expressions.’ This is an inauspicious start
of what is presumably meant to be a serious paper. Since the remainder of the
paper hardly adds anything, we can safely let it rest.
Geurts and Beaver (2007), an encyclopedia article on DRT, is equally
evasive as regards the notion of presupposition. These authors define, or
describe, presupposition as follows (Section 5.2): ‘Presuppositions are chunks
of information associated with particular lexical items or syntactic construc-
tions.’ This is a very different notion from the one put forward by Van der
Sandt, but equally unhelpful. An English word like pal and an English
construction like ain’t are ‘associated with’ the ‘chunk of information’ that a
colloquial register is being used. Does that make this information presuppo-
sitional? Presumably not. But then, what distinguishes this information from
the information conveyed by presuppositions? The answer is not to be found
either in this encyclopedia article or anywhere else in the DRT literature.
Geurts and Beaver (2007) thus shares the fuzziness with Van der Sandt (1992)
but it directs its diffusing bundle of darkness in a different direction.
Yet, despite these notional unclarities, the authors in question maintain
that there is one unified mechanism underlying both external-anaphora
374 The Logic of Language

resolution and presupposition projection. This should entail that where an

external anaphor is unresolvable, presupposition projection is blocked and
vice versa. A priori one is inclined to think that such a view is implausible,
because anaphora resolution is a matter of referential identification, whereas
presupposition projection is a matter of semantic consistency within a dis-
course domain and its subdomains. It would be most remarkable if these two
processes were to be seen to be subject to the same mechanism. And, of
course, they are not.
Geurts and Beaver (2007, Section 2) argue for the same-mechanism posi-
tion on the strength of examples like (10.74) (the exclamation mark indicates
incoherence):
(10.74) Pedro doesn’t own a donkey. !It is grey.
In their perspective, a parallel can be drawn with (10.75):
(10.75) Pedro doesn’t own a donkey. !Juanita regrets he does.
Just as (10.75) is incoherent because the (factive) presupposition that Pedro
owns a donkey, carried by the second sentence in (10.75), is denied in the first
sentence, in the same manner (10.74) is thought to be incoherent because the
it in the second sentence in (10.74) requires an accessible address to latch on
to, which is absent owing to the negation in the first sentence.
In fact, however, the incoherence of (10.74) has nothing to do with anaph-
ora resolution: the anaphoric relation between it and a donkey is clear enough,
as one can see immediately when one replaces grey with, for example, a fiction
of his mind. On the contrary, it has everything to do with the fact that the
predicate grey induces a presupposition of actual existence. Grey is extensional
with regard to its subject term, which means that, for truth to arise, this term
must refer to an actually existing entity: for something to be grey it must first
exist (see Section 3.5 in Volume I).
Since extensionality or intensionality of a given predicate with respect to
some term position is a matter of lexical preconditions and hence of pre-
suppositions, the incoherence of (10.74) is not a matter of anaphora but
of presupposition. The anaphora aspect of (10.74) is unproblematic. The
Principle of Maximal Unity for discourse domains (Section 7.2.2) says that
entity representations (addresses) in any (sub)domain are retrievable from
any (sub)domain. Therefore, the anaphor it has free access to any other
domain or subdomain in D, including the subdomain under negation created
by the first sentence in (10.74).
 Presupposition and presuppositional logic 375

The often heard claim that definite determiners, including those thought to
be present in definite pronouns, induce existential presuppositions is false.
As has been amply shown in Seuren (1985, 1988, 1994, 2000) and again in
Section 3.5.1 of Volume I and Section 10.7 of the present volume, existential
presuppositions originate from the extensional character of argument posi-
tions and not from definite determiners. Yet Geurts and Beaver (2007, Section
5.2) still claim that existential presuppositions derive from definite determi-
ners, even though they are fully aware of the arguments put forward in many
publications by me over the past thirty years showing that this position is
untenable. But much as one may try to ignore arguments or theories out of
existence, it doesn’t necessarily make them go away.
Since the first sentence in (10.74) denies the actual existence of a donkey
owned by Pedro, the sequence as a whole is semantically incoherent, though
with a perfectly straightforward anaphora resolution. In (10.76a–d), there is
again no problem as regards anaphora resolution (although, according to
Geurts and Beaver, there should be one in (10.76a)), but they differ from
(10.74) in that they are all fully consistent. Their consistency derives from the
fact that not all predicates involved in the second clauses of (10.76a–d) are
fully extensional with regard to their object-term positions or with regard to
the subdomains they introduce:
pffi
(10.76) a. John doesn’t own a Ferrarii. He has simply invented iti.
b. John may own a Ferrarii, but I have never seen iti.
c. John may own a Ferrarii but he may also have simply invented iti.
d. Geert has announced that he is going to make an anti-Islam filmi
but I doubt that iti will ever materialize.
Example (10.76a) illustrates the fact that anaphora resolution does not
require an antecedent address in the main or truth domain of the discourse
at hand. All it requires is the presence of an appropriate antecedent address in
any domain or subdomain (in this case, the extensional subdomain under
negation). This is quite different from presupposition projection, which is
restricted by conditions of overall semantic consistency. That no clash arises
in (10.76a) is simply due to the fact that the predicate invent (in the sense of
‘pretend there to be’) is intensional with respect to its object term, so that it
does not need an antecedent for iti in the extensional truth domain. Analo-
gous analyses apply to (10.76b,c,d).
Of course, this analysis requires an ontology that allows for nonexisting,
virtual entities, which is something most Western philosophers, burdened as
they are with the legacy left by Russell and Quine, have great difficulty with.
They ought, however, to consider that this is apparently the way humans
376 The Logic of Language

naturally construct their ontologies, as appears from the fact that humans
refer to and quantify over virtual entities with the same ease and naturalness
as they do with regard to actually existing ones (see Chapter 2 of Volume I for
extensive discussion). Refusing to admit an intensional ontology means inter
alia the inability to explain the coherence of sentences like (10.76a–d).
Closer inspection quickly reveals that there are many cases where anaphora
resolution proceeds without a hitch but where incoherence arises owing to the
blocking of presupposition projection. Consider the following examples:
(10.77) a. ! Johni pretends that he has left but hei has come back.
b. ! I know Johni hasn’t left, so hei must have come back.
c. ! John never had a wifei. He simply divorced heri.
d. ! If Johni pretends to have childrenj, theyj must be staying with
hisi sister.
The first clauses in (10.77a,b) entail that John hasn’t left while the second
clauses presuppose that he has. This clash of entailments blocks the obligatory
upward projection of the embedded presupposition of the second clause and
thus causes the incoherence of (10.77a,b). Yet there is no problem with the
resolution of the anaphor hei in the second clauses of (10.77a,b), which is an
external anaphor with John as its antecedent.
Similarly in (10.77c), where the first clause denies a presupposition of the
second. Yet there is no problem with the anaphoric pronoun heri, which
resolves (by primary or donkey anaphora) into a wife in the first clause. In
(10.77d), the first clause entails that John has no children, whereas the second
presupposes that he has, owing to the fact that the predicate stay with one’s
sister is extensional with regard to its subject term and must is a ‘hole’, giving
rise to existential import. But the resolution of the (donkey) anaphor theyj is
entirely unproblematic.
Further examples illustrating free anaphoric access across subdomains are:
(10.78) a. John has no children. So they (the poor creatures) can’t be on
vacation.
b. If Juan had any children, they (the poor creatures) would speak
Spanish.
c. A farmer who has no donkey can’t feed it (the animal).
d. A farmer who has no donkey can still dream about it (the animal).
(The possibility of epithet anaphora (Section 9.2) shows that these are anyway
cases of external and not of bound-variable anaphora, which precludes an
analysis in terms of Geach (1969, 1972); see Section 9.4.1.)
 Presupposition and presuppositional logic 377

This suffices to show the untenability of the position defended by Van der
Sandt, Geurts and Beaver, and all too easily adopted by other adherents of
DRT. But there is more. Presuppositions are recoverable from their carrier
sentences on account of the lexical meanings of the predicates inducing them,
while no such help is available for anaphors. Thus, while (10.79a) allows for
the conclusion So he was married, no corresponding counterpart is available
for (10.79b):
(10.79) a. Harold wanted to get divorced. (So he was married!)
b. Harold wanted to see her. (Who?)
Van der Sandt’s reply to this argument is that (Van der Sandt 1992: 341)
. . . unlike pronouns, <presuppositions> contain descriptive content which enables
them to accommodate an antecedent in case discourse does not provide one.

But this won’t do. The difference is that presuppositions consist of proposi-
tional content while definite descriptions have descriptive content. Epithet
pronouns, as has been shown, also have descriptive content, yet such descrip-
tive content does not suffice to provide an antecedent:
(10.80) Harold wanted to see the old girl. (Who?)
We must conclude that Van der Sandt’s claim that (Van der Sandt
1992: 341):
. . . presuppositions are just anaphors. They can be treated by basically the same
mechanism that handles the resolution of pronominal and anaphoric expressions.

is confused and bizarre. Not only does it require notional obfuscation for it to
be palatable to outsiders such as the school of DRT-practitioners, it also
confuses the phenomenon itself with a mechanism associated with it. In the
end, DRT still lacks an account of presupposition. Anaphora is anaphora and
presupposition is presupposition. The twain meet only in so far as they link
arms in context-driven utterance interpretation.
 11

Topic–comment modulation

11.1 What is topic–comment modulation?

11.1.1 The Aristotelian origin of topic–comment modulation
To understand what is meant by topic–comment modulation, also known as
topic–comment structure or information structure, one can hardly do better
than go back to Aristotle’s definition of the notion of proposition as explained
in Section 3.1.2 of Volume I, because the Aristotelian notion of proposition is,
in fact, the notion of topic–comment structure, even though this was not
discovered until the nineteenth century. One recalls that Aristotle defines a
proposition as (PrAn 24a16):
A proposition (prótasis) is an affirmative or negative expression that says
something of something.

What matters here is that what Aristotle had in mind was the attribution of
a specific property to some specific entity, as in for example, ‘Mr. G. has the
property of being obnoxious’. This notion of proposition is schematically
rendered in Figure 11.1 (where one notices that Aristotle had no term for what
we now take to be the subject term in a propositional structure).
This analysis implies that the something of which something is said—the
reference object of what was later to be called the subject term—must be given
for a proposition to be conceived, uttered, and interpreted. And there are
precious few ways in which an object can be given—that is, intentionally
focused on by the speaker and identifiable by the listener.
A traditional view holds that reference objects are identified in linguistic
interaction through the meaning of the predicate embedded in the referring
expression (mostly a noun phrase). Husserl writes (1900: 49):
. . . daß es also mit Recht heißt, der Ausdruck bezeichne (nenne) den Gegenstand
mittels seiner Bedeutung.
( . . . that it is, therefore, correct to say that the expression denotes (names) the object
by means of its meaning).
 Topic–comment modulation 379

Proposition
(prótasis)
Thought/
Language
Predicate
? (Subject term) –
(kategoróumenon)

true or false

fact or fiction

Situation/ Entity Property

(hypokéimenon, –
(symbebekós,
World
suppositum) accidens)

FIGURE 11.1 The Aristotelian notion of proposition

But if this is intended to imply that the meaning of the predicate embedded in
a definite NP suffices to identify the reference object (æ-value), it is wrong and,
in fact, rather naı̈ve. The meaning of the embedded predicate in a definite NP
suffices to identify the reference object only if the object in question is unique
in the speaker–hearer world of experience, such as the sun or the moon. But
such cases are rare compared with those where the reference object needs ad
hoc situational and/or contextual information to be identified. Pointing is, of
course, one device, possibly combined with the use of a pronoun, but in the
vast majority of cases æ-values are established in that the cognitive-linguistic
context of utterance enables speaker and listener to home in on the intended
reference object. (One recalls from Section 9.5 that in some cases of primary
anaphora the truth of the utterance in question makes for the fixing of
reference.)
An Aristotelian proposition is thus the mental assignment of a specific
property to one or more objects that have been given enough salience to act as
reference object. When such a proposition is expressed linguistically, the
property is expressed by means of a predicate, while that to which the property
is assigned is expressed by what is now known as the subject term. This analysis
was adopted, uncritically one has to admit, by the earliest grammarians in the
Greek world and mistaken for what we now know as the grammatical or
syntactic analysis of sentences. Thus, the first syntactic analysis of sentential
structure consisted of a distinction between a subject term and a predicate,
which were seen as the linguistic counterparts of the Aristotelian something of
380 The Logic of Language

which something was said, respectively. Luckily for the earliest grammarians,
what they discovered was real, since there can be no doubt that syntactic
structure, with a subject term (along with a main verb and other argument
terms plus any number of adverbial modifiers), is real. But they were less
lucky in that they remained fixated on subject-predicate structure, failing
to see the roles of further subdivisions within what they considered the
‘predicate’.

11.1.2 The discovery of the problem in the nineteenth century

As amply demonstrated in Section 3.2 of Volume I, it was not until well into
the nineteenth century, when the difference between syntactic structure and
topic–comment modulation was discovered, that syntactic analysis was un-
shackled and became free to explore those aspects of syntactic structure that
had remained hidden for twenty or so centuries. Critical and inquisitive
investigators began to see the discrepancy between the original Aristotelian
notion of proposition and what had been made of it by the grammarians. And
this discrepancy or, as some authors called it, incongruity gave rise to a
terrible theoretical predicament that has not found a final solution till the
present day. Syntactic structure was the object of fruitful investigation and
many positive results have been chalked up (despite the many changes of
direction that have taken place over the past fifty years), but the topic–
comment counterpart was entirely forgotten by the mainstream theorists
and stayed in the remit only of pragmatically oriented investigators, who,
on the whole, have little sympathy for the more formally oriented work of
syntacticians, semanticists, or linguistic theoreticians.
As a result, there is a lively research area with many practitioners occupying
themselves with questions of what is called information structure. Yet, apart
from notable and important exceptions, such as the work done by the Prague
school of linguistics ever since World War II (see below) and the work by
Carlos Gussenhoven of Nijmegen University (see, for example, Gussenhoven
1984, 2005), questions regarding the relation between sentence structure and
topic–comment modulation are, on the whole, dealt with in more pragmati-
cally oriented circles, which, unfortunately, means that foundational ques-
tions regarding the semantic status of topic–comment modulation are not or
hardly envisaged. It is, therefore, well worth our time to delve a little into the
history and the essence of the problem.
The first observations around the incongruity between syntactic structure
and the providing of new information regarding given concrete or abstract
 Topic–comment modulation 381

entities were made around 1850, mostly by German philosophers of language

and philologists. Heymann Steinthal observes (1855: 199) that in a sentence
like The patient slept well the grammatical subject is the patient and the
grammatical predicate is slept well. But if the sentence is interpreted
simply as the attribution of the property of sleeping well to the individual
described as the patient, an important fact is overlooked, namely that often,
‘what one wants to say is that the patient’s sleep was good’. Therefore, an
analysis is wanted, different from the syntactic surface structure, in which
the patient’s sleep is the reference object of the subject term (Aristotle’s
hypokeı́menon) and what has come out in the syntactic structure as the adverb
well is the underlying predicate.
Somewhat later Georg von der Gabelentz observed (1869: 378):1
What does one wish to achieve when one speaks to another person? The answer is that
one wants to arouse a thought in him. In my view this implies two aspects: first, one
has to direct the interlocutor’s attention (his thinking) to something, and secondly,
one makes him think this or that about it. I call that of or about which I want my
addressee to think the psychological subject, and that which he should think about it
the psychological predicate. In the sequel it will become clear how much these cate-
gories often deviate from their grammatical counterparts.

In his (18911/19012), he defended the view that the grammatical subject

expresses what should be considered the hypokeı́menon in a logical analysis
(though he does not indicate what logic he has in mind), while the psycholog-
ical subject consists in a mental representation of that which comes to mind
first when one is in the middle of the speech process. This psychological
subject will normally also come first in the spoken or written utterance. Thus,
in the sentence (19012: 370) Mit Speck fängt man Mäuse (with bacon one
catches mice) the psychological subject is mit Speck (with bacon), and the
psychological predicate is what one does with bacon, namely catch mice. He
does, however, point at the difficulty of providing observational support for
his thesis (Von der Gabelentz 19012: 370):
But if one wants to give the inductive proof for all this, one has to be careful with
examples. For the phenomena to do with positions in the sentences of different
languages are not unambiguous or equivalent.

Around the same time we find Wilhelm Meyer-Lübke, a prominent

Romance philologist, reacting to Von der Gabelentz, confusingly using the
term logical for what Von der Gabelentz had called psychological and not for

1
All translations are my own.
382 The Logic of Language

what Von der Gabelentz had called logical, which for him was the grammatical
structure of the sentence (Meyer-Lübke 1899: 352):
I want to stress that ‘subject’ is used here in a purely grammatical sense, and
designates, therefore, the agent of the action. Admittedly, this goes against the original
meaning of this term, which, as one knows, originated in logic. From the point of view
of logic there can be no doubt that in the sentence il arrive deux étrangers [two
foreigners arrive] the subject is il arrive while deux étrangers is the predicate [. . .]. But
from the point of view of grammar the relation between Noun and Verb remains
unchanged, no matter which comes first in the sentence.

Interestingly, Meyer-Lübke specifies the grammatical subject of a sentence as

designating ‘the agent of the action’, not unlike many modern attempts at
providing a semantics for lexical argument functions in terms of thematic
roles.
Von der Gabelentz’s notion of psychological subject was taken up and
further developed by a number of scholars, in particular Philipp Wegener
(1885: 21):
It is the function of the subject [die Exposition] to state the position [die Situation
klar zu stellen], so that the logical predicate becomes intelligible.

He was followed by Theodor Lipps, who introduced the notion that the
‘psychological’ predicate is in fact the answer to a question about the hypo-
keı́menon that has arisen in the current context (Lipps 1893: 40):
The grammatical subject and predicate of a sentence now agree now do not agree with
those of the judgement. When they do not, the German language has intonation as a
means of marking the predicate of the judgement. The subject and predicate of the
associated judgement are best recognised when we bring to mind the question to
which the sentence is an answer. That which the full and unambiguous question is
about is the subject, while the information required is the predicate. The same
sentence can, accordingly, serve to express different judgements, and hence different
subjects and predicates.

Meanwhile, in Britain, some philosophers were working along the same

lines. The Cambridge philosopher George Stout compares the progress of
thought as expressed in language with the steps one takes while walking
(Stout 19093, Vol. 2: 213–4):
Predication, from this point of view, just consists in the definition and specification of
what is, at the outset, indefinite and indeterminate. It is because this process takes
place gradually by a successive concentration of attention, that language is divided
into sentences. The predicate of a sentence is the determination of what was
 Topic–comment modulation 383

previously indeterminate. The subject is the previous qualification of the general topic
or universe of discourse to which the new qualification is attached. The subject is that
product of previous thinking which forms the immediate basis and starting-point of
further development. The further development is the predicate. Sentences are in the
process of thinking what steps are in the process of walking. The foot on which the
weight of the body rests corresponds to the subject. The foot which is moved forward
in order to occupy new ground corresponds to the predicate. [. . .] All answers to
questions are, as such, predicates, and all predicates may be regarded as answers to
possible questions. If the statement, ‘I am hungry’ be a reply to the question, ‘Who is
hungry?’ then ‘I’ is the predicate. If it be the answer to the question, ‘Is there anything
amiss with you?’ then ‘hungry’ is the predicate. If the question is, ‘Are you really
hungry?’ then ‘am’ is the predicate. Every fresh step in a train of thought may be
regarded as an answer to a question. The subject is, so to speak, the formulation of the
question; the predicate is the answer.

Wilhelm Wundt, one of the founding fathers of modern psychology, also

took part in this debate. He denies that there is a disparity between ‘psycho-
logical’ and grammatical structure, in whatever terminology. For him, the
grammatical subject always expresses the Aristotelian hypokeı́menon. Gram-
matical structure thus faithfully reflects the logical analysis imposed by
Aristotle on sentences. This is, for him, the only legitimate use of the terms
subject and predicate (Wundt 19224: 266):
The fact that the judgement consists of subject and predicate results from an analysis
of judgements, and this is an insight that has rightly passed untrammelled from
Aristotelian logic [. . .] into the more modern forms of logic. The subject is the
thing the proposition is about, that which forms the basis, hypokeı́menon; the predi-
cate is the content of the proposition, the katēgórēma, as Aristotle called it.

Although it is probably correct, he says, to distinguish, as Von der Gabe-

lentz did, between that which comes to mind first and that which is added as
new information, that should not be labelled as the subject-predicate distinc-
tion. Instead, Wundt proposed the term dominant representation (dominier-
ende Vorstellung) for what Von der Gablelentz, Lipps, Stout, and others had
called the psychological subject, but Wundt’s term never gained acceptance
(Wundt 19224: 269–70):
Suppose I transform the sentence Caesar crossed the Rubicon into The Rubicon was
crossed by Caesar, does that mean that the subject Caesar has become a remote object,
and has, conversely, the original object the Rubicon now become the subject? And
when I say The crossing of the Rubicon was achieved by Caesar, has now the original
predicate become the subject?
384 The Logic of Language

These are the questions that have led, in our new linguistics, to a kind of distinction
that has found a rather widespread acceptance, but which, in my eyes, has increased
rather than solved the confusion resulting from the mixing of logic, grammar, and
psychology. If we are to believe G. von der Gabelentz we should distinguish between a
logical, a grammatical and a psychological subject and predicate. The logical subject
and predicate keep the function they have in logic. The psychological subject is seen as
‘the representational complex that occurs first in the consciousness of speaker and
hearer’, while ‘the content that is added to this prior representation’ should be the
predicate. Or, as v.d. Gabelentz formulates it from the teleological point of view, the
psychological subject is ‘that about which the speaker wants the hearer to think, to
which he wants to direct his attention, while the psychological predicate consists of
that which the hearer should think about the subject’. [. . .]
When one says that the two sentences Caesar crossed the Rubicon and The Rubicon
was crossed by Caesar have the same logical subject but different grammatical subjects,
one has already lost sight of the notion of subject in the Aristotelian sense, namely as
that on which the assertion is based, and surreptitiously introduced a psychological
consideration, namely that the subject must be an agent. Obviously, the agent in both
sentences is Caesar. But only in the first sentence, and not in the second, is he the basis
on which the proposition is grounded. The former is an assertion about Caesar, the
latter about the Rubicon.

Although one may disagree with Wundt on several counts, he makes some
important points, such as the difference between the genesis and the sub-
stance of a propositional thought, and the necessity to create a separate
terminology for the grammatical distinction of subject and predicate on the
one hand and the ‘psychological’ distinction of ‘what comes to mind first’ on
the other.
As one sees from the quotations given, there was a great deal of confusion
about this issue around the turn of the century, and the parties involved were
unable to settle on an agreed solution. In fact, the confusion was such that
Theodor Kalepky exclaimed (1928: 20): ‘Such a confusion simply cries out for
relief ’ (Eine derartige Wirrnis schreit förmlich nach Abhilfe).2 After 1930 the
subject-predicate debate, which had dominated linguistic theorizing for al-
most a century, disappeared from the limelight, mainly due to the lack of
empirical support and the general unclarity of the issues concerned, but also
because the new structuralism in linguistics had different interests.

2
Kalepky belonged to a group of linguists who felt that a theory of grammar should be set up
without any notion of subject and predicate at all. Others belonging to this movement were Svedelius
(1897) and Sandmann (1954). This movement, however, petered out without leaving as much as a
trace.
 Topic–comment modulation 385

The only place where the debate was continued was Prague, largely owing
to a tradition of loyalty to good work done by local scholars. Anton Marty, a
disciple of the German phenomenologist Franz Brentano and professor of
philosophy at Prague by the end of the nineteenth century, made important
contributions to the subject-predicate debate. According to him, logic de-
serves no place in semantics, all semantics being psychological. Besides an
abstract propositional meaning, every sentence has an ‘inner form’ which
expresses the way the propositional meaning is to be integrated into running
discourse. He follows Lipps, Stout, and others in saying that this ‘inner form’
is determined in principle by question–answer structure. Unlike Wundt, he
maintains that the terms subject and predicate are most appropriately used at
this ‘inner form’ level, since it is here that the Aristotelian meaning of these
terms is immediately applicable. Despite some unclarities, this makes a great
deal of sense, as will be clear in a moment.
Marty’s work was continued by the Czech scholar Vilém Mathesius, pro-
fessor of English at Prague University and founder, in 1926, of the Prague
Linguistic Circle. Mathesius followed Wundt in wishing to see a separate
terminology for subject and predicate on the level of grammatical analysis
on the one hand, and the ‘known-new’ distinction found to exist at a more
psychological level by Lipps, Stout and company on the other. Not wishing to
upset existing terminology, he felt that the terms subject and predicate should
go on being used in grammar, no matter what confusions had occurred in
recent literature, and proposed a new term pair for the Aristotelian distinc-
tion, which is realized at the ‘psychological’ level. For the latter he proposed a
Czech term pair that has been rendered variously as theme versus rheme, topic
(or focus) versus comment, the former pair member indicating the Aristotelian
hypokeı́menon, the latter the Aristotelian predicate. The structure into which
both are combined is not called ‘proposition’ but the functional sentence
perspective (Mathesius 1939).
Although the question of the disparity between syntactic structure and
topic–comment modulation dominated all discussions about the nature of
language for well over half a century, it disappeared from the theoretical
agenda when the new structuralism made its appearance around 1930. This
was no doubt due to the fact that structuralism in linguistics, in particular the
American variety, was strongly focused on grammatical form and tried to
dispense with meaning as an object of ‘scientific’ enquiry altogether. In this
perspective, introspection-based talk about ‘what comes to the mind first’ and
things like that was considered unscientific and an improper intrusion of
phenomenological psychology into the much more ‘scientific’ arena of lin-
guistics and behaviourist psychology.
386 The Logic of Language

Needless to say, after 1960 the psychologists struck back and joined forces
with pragmaticists to develop a pragmatically oriented discipline of discourse
and text analysis, sometimes called ‘conversation analysis’, studying questions
of ‘information packaging’ and ‘information structure’ in the intuitive terms
of personal, introspective experiences. It is as if the practitioners of conversa-
tion analysis have turned their backs on formal grammar and even more on
everything to do with logic and the more formal aspects of meaning. They
represent extreme ecologism as described in Section 1.3.3 of Volume I. It is our
purpose here, in the context of discourse semantics, to redress the balance
somewhat and shed some light on the more formal and theoretical aspects of
topic–comment modulation.

11.1.3 The dynamics of discourse: the question–answer game

Let us take up the issue where Lipps and Stout left off, accepting that the
normal progress of discourse is driven by a ‘game’ of questions and answers
(see also Seuren 1985: 297–304). Questioning is probably a basic and general
cognitive activity that far exceeds, but underlies, the question–answer game in
topic–comment modulation that is meant to secure textual coherence. Ques-
tioning, whether of the specific (WH-) or the polar (yes/no) variety, pervades
the whole of cognition and is probably the main guiding principle in acquir-
ing new knowledge and insight. What we see of it in discourse incrementation
procedures is but one of the many uses to which the human propensity to be
inquisitive and ask questions is put.
Ideally speaking, the discourse D is unrestricted at the initial point. The
question-answer game may be opened by what in the theatre is called a feeder :
an utterance meant to set a discourse in motion without there being any
anticipation of a question arising in the audience’s mind (other than the
general question ‘What is this person going to tell us?’). Normally, implicit
questions are left out from the actually spoken text, as the speaker anticipates
the listener’s queries. We then speak of an implicit question, to which the
following assertion (which may be a new sentence or part of an ongoing
sentence) is an answer. It also often happens that speakers themselves pro-
nounce the question they think the listener will come up with. This is a typical
feature of what is often felt as ‘patronizing’ speech, as in teaching situations or
in official announcements made by authorities of whatever kind.
Normally, sentences or clauses beginning with and are not answers to
implicit or explicit questions but continuations of answers given (but see
Blakemore and Carston 2005 and note 12 in Chapter 8). It seems that sentences
starting with but (where but is not a comment correction, such as German
 Topic–comment modulation 387

sondern) can be reduced to the speaker’s assumption of the implicit question

‘Is there nothing that diminishes the effect of the previous utterance?’, where
but signals the answer ‘Yes, namely . . . .’.3
Opening sentences, or feeders, typically occur in the daily news (Van
Kuppevelt 1991):
(11.1) The workers of the Slumbo mattress factory in Bubble-on-Rhine have
announced that they will go on strike tomorrow.
This will naturally give rise to a few questions, such as ‘Why?’, ‘Have there
been previous negotiations?’, ‘How does the management react?’, and so on.
And indeed, one typically finds that subsequent sentences after such a feeding
line answer such questions, under the appropriate intonation.4
Existential quantification and OR-disjunctions systematically give rise to
implicit questions, as in (11.2a), where the question is actually anticipated by
the (‘patronizing’) speaker:5
(11.2) a. Some time before World War I the Titanic sank. When did the
Titanic sink? It sank in 1912.
b. Some time before World War I the Titanic sank. It sank in 1912.
c. Some time before World War I, in 1912, the Titanic sank.
A less solicitous or ‘patronizing’ speaker might have said (11.2b) or even
(11.2c). One notes that the underlying ‘entity’ to which the property ‘in 1912’
is assigned is ‘the time of sinking of the Titanic’. We shall not go into the

3
For the contrast between English and and but, see, for example, Lakoff (1971), Bellert (1972),
Blakemore and Carston (2005). For the corresponding German contrast, one may consult Lang (1977),
Abraham (1991), Diewald and Fischer (1998), Fischer (2000). For a comparison between English and
German, see, for example, Asbach-Schnitker (1979).
4
An experiment (Van Kuppevelt 1991) showed that when test subjects read aloud the text of a real-
life news bulletin, which did not contain the anticipated or implicit questions, the intonation patterns
did not differ from readings by the same subjects of the same text but with the implicit questions filled
in as anticipated questions.
5
The existential quantifier FEW appears to do the same and even more (see note 5 of Chapter 9). As
reported in Moxey and Sanford (1986/7), the introduction of a plural discourse address under FEW
systematically induces the setting-up not only of a plural address for the set delimited by FEW but also
for the complement of that set. Moxey and Sanford found that subjects presented with a sentence like
(i) would continue a subsequent sentence starting with They . . . in ways that made it clear that they
were referring to those students that were not at the meeting. A typical continuation would be (ii):
(i) Few students were at the meeting.
(ii) They had (all) gone out with their girl friends.
This strongly suggests that FEW systematically gives rise to an implicit question of the form ‘How
about the others?’, which would also naturally be part of the speaker’s text (in the ‘patronizing’ style
of speech mentioned above). Moxey and Sanford use the term complement anaphora for primary
anaphora to the complement address set up in virtue of that question arising.
388 The Logic of Language

philosophical question of what kind of ‘entity’ this is, but simply accept the
reification procedure underlying expressions such as the time of sinking. The
property assigned to this abstract, reified ‘entity’ is that it was in 1912.6
An important corollary of the assumption that TCM-structure mirrors an
ongoing question-answering game in discourse is that TCM-structure is a
powerful factor in reducing the effort involved in the interpretation of
utterances. Since the topic has already been through the interpretive grinder,
all that remains to be processed in the interpretation is the comment, which is
often just one constituent and sometimes even no more than one single word.
It is surprising that this obvious fact is taken into account so little in the
literature on parsing and the experiments related to it.
Now consider again the example of the sentence John sold the car, discussed
in Section 3.2 of Volume I. Suppose this sentence is uttered in a context where
the (implicit or explicit) question is Who sold the car? Then the answer
requires emphatic accent on John, as in (11.3c), or the corresponding cleft or
pseudocleft, as in (11.3a) and (11.3b), respectively:
(11.3) a. It is JOHN who sold the car.
b. The one who sold the car is JOHN.
c. JOHN sold the car.
We take it (Seuren 1998b) that (11.3a,b,c) have a common underlying SEMANTIC
ANALYSIS (SA) corresponding to the topic–comment modulation or TCM
structure (11.4), which is input to both the grammar of the language in
which the sentence is to be expressed and to the discourse-incrementation
procedure (more is said about the predicate Bev-ind below):
(11.4) Bev-ind JOHN (the x[x sold the car])
In (11.4), the grammatical subject term the x[x sold the car] is the topic and
the grammatical predicate bev-ind JOHN is the comment, as shown in Figure
11.2 (repeated from Figure 7.8 in Chapter 7 of Volume I).
Assume for the moment that, in general, the surface subject term attracts a
grade-1 accent and that what is the predicate of a sentence at SA-level attracts

6
It would appear that quantification in the comment, as in It is not everyone that is granted the gift
of tongues, cannot be handled by the mechanism of underlying cleft structures other than in ways that
are formally so complex as to make their occurrence in language unlikely. Other mechanisms seem to
be at work here, for example, the mechanism of quotation. This would impose a metalinguistic
interpretation on the sentence just quoted, in the sense that everyone should be taken as the quoted
form ‘everyone’: ‘the use of the word “everyone” is inappropriate in the given context’. Note also that
someone or anyone are hardly usable in the sentence quoted: *It is not someone/anyone that is granted
the gift of tongues.
 Topic–comment modulation 389

S1

SPEECH-ACT S2
OPERATOR
TOPIC
<ASSERT> NP Auxiliary
PRED System
be v-ind JOHN Det S3
the x
PRED S4
Matrix-S
PAST
PRED S5
COMMENT ∆

PRED NP NP
sell x the car

FIGURE 11.2 Underlying SA of JOHN sold the car

a coding for sentence-nuclear accent, which is realized as an accenting on the

VP-constituent in surface structure, and that the surface-sentential VP-accent
settles on the last nominal VP-constituent (Gussenhoven 1984). This makes
for an unmarked grade-1 rising intonation on John and the car in (11.5a) and
on John and Mary in (11.5b) (apart from unmarked grade-1 rising tones on the
higher SA-operator Mónday mórning):
(11.5) a. Jóhn sold the cár on Mónday mórning.
b. Jóhn gave the book to Máry on Mónday mórning.
According to this principle, the SA-predicate Bev-ind JOHN in a TCM
structure will receive a coding for grade-1 sentence-nuclear accent. Now
assume that sentence (11.3c) results from the LOWERING of the SA-predicate,
shortened to just JOHN, to the position of the variable x in the Matrix-S.7 On
that assumption, it makes sense to surmise that the (shortened) lowered

7
Such lowering is, of course, subject to the well-known, possibly universal, ISLAND CONSTRAINTS. One
such island constraint forbids the lowering of the comment predicate into a relative clause, as in
(i). No constraint prevents such lowering in sentence (ii):
(i) *Not for JOHN but for BILL are those who work on strike.
p
(ii) Not JOHN’s but BILL’s workers are on strike.
Since parallel observations can be made for the cleft structures (iii) and (iv), it seems that the island
constraints in question should be taken to apply to the relation between lambda-abstracted and non-
lambda-extracted pairs at SA-level:
(iii) *It is not for JOHN but for BILL that those who work are on strike.
p
(iv) It is not JOHN’s but BILL’s workers that are on strike.
To what extent this might be taken to restrict the cognitive process of questioning in discourse is a
matter for further research.
390 The Logic of Language

predicate JOHN will carry its coding for grade-1 sentence-nuclear accent along
to the new Matrix position of surface subject term and that the grade-1 accent
for surface subjecthood is reinforced to a grade-2 accent. This grade-2 accen-
tual peak thus signals the fact that the constituent thus marked is the
comment SA-predicate of the corresponding TCM-structure (11.4)—that is,
the structure shown in Figure 11.2. The accentual peak is reinforced even
further when the comment SA-predicate is contrasted with another such
comment predicate.
One consequence of this analysis is that an elegant parallel can be drawn
between structures underlying WH-questions such as (11.6) and the
corresponding reply (11.4):
(11.6) Bev-ind WHO? (the x[x sold the car])
The open-place question predicate ‘WHO?’ indicates that a value is required in
this position. The answer (11.4) provides the value JOHN.
We now define the topic of a sentence as that element in a situation whose
specific identity is open to question, or as that parameter in a situation whose
value is being requested. The comment then provides the answer by specifying
the element (value) in question. Otto Jespersen already saw this parallel, as
appears from the following quote (Jespersen 1924: 145):
The subject is sometimes said to be the relatively familiar element, to which the
predicate is added as something new. ‘The utterer throws into his subject all that he
knows the receiver is already willing to grant him, and to this he adds in the predicate
what constitutes the new information to be conveyed by the sentence. [. . .] In “A is B”
we say, “I know that you know who A is, perhaps you don’t know also that he is the
same person as B” ’ (Baldwin’s Dict. of Philosophy and Psychol. 1902, Vol. 2.364). This
may be true of most sentences, but not of all, for if in answer to the question ‘Who
said that?’ we say ‘Peter said it’, Peter is the new element, and yet it is undoubtedly the
subject.

It thus makes sense to propose a separate structural analysis for the genesis
of a proposition, as opposed to its actual truth-conditional substance. The
process of genesis—that is, the progress from what has been established in the
discourse to what is added as new information—produces the TCM-struc-
ture, which contains, besides the truth-conditional substance, also informa-
tion about the process that gave rise to the proposition in question.
At this point we need some terminology. Let us call a merely truth-
conditionally presented proposition a flat proposition (fprop), while its
topic–comment modulated variants, each representing a particular history
of its genesis, will be called modulated propositions (modprop). A flat
 Topic–comment modulation 391

Speaker Hearer
Question-Answer
Structure

Expression Interpretation Discourse

modprop Incrementation

modprop Truth
fprop
Calculus
MIND
_______________________________________________
WORLD
Uttered sentence Situation in World

FIGURE 11.3 Discourse-semantic layout of speech process

proposition is like a picture: it represents a mere state of affairs, and cannot

express anything like the topicalization found in TCM-structures. A modprop
thus consists of an fprop plus a topicalization. Its expression at SA-level is
taken to be a TCM-structure as exemplified in (11.4) and in Figure 11.2.
At some stage in the comprehension process, probably immediately upon
the incrementation into the current D, the hearer must distil the strictly truth-
conditional content of an utterance received, the corresponding fprop, from
the information incremented to the current D, as represented in Figure 11.3.
A TCM-structure like (11.4) conveys all the information conveyed by the
corresponding flat proposition. In addition, however, it conveys the informa-
tion requested by the predicate WHO?, as in (11.6), in the flow of discourse. It is
this latter fact that guides the incrementation process. It does so by specifying
the identity of the element that was the object of querying, in this case the
person called John. Figure 11.3 gives a schematic survey of the relation between,
on the one hand, a modprop of the kind that normally occurs in the flow of
speech and the corresponding fprop, whose role has been reduced to the truth
calculus with regard to the situation the utterance is keyed to in the world.

11.2 Phonological, grammatical, and semantic evidence

for TCM
The empirical evidence for topic–comment modulation is a combination of
phonological, grammatical, and semantic observations. More strictly phono-
logical evidence is adduced in, for example, Bolinger (1972) and Gussenhoven
(1984, 2005). Here we look at certain intonational features in combination
with judgements of grammaticality and possible meanings.
392 The Logic of Language

Consider first the fully grammatical sentence (11.7a) (also discussed in

Section 7.2.2 in Volume I) expressing a flat proposition (fprop) with only
grade-1 accentual peaks. This sentence contains the negative polarity item
(NPI) in the least, licensed, in this case, by the negation n’t. An answer is now
needed to the question why (11.7b) is ungrammatical but (11.7c) is well-
formed and meaningful, with grade-2 intonational peaks on both JOHN and
PETER and a grade-1 rise on not:
(11.7) a. pJohn isn’t in the least interested.
b. *JOHN isn’t in the least interested, PETER is.

p
c. JOHN isn’t in the least interested, nót PETER.
(11.8) a. *NOT [Bev-ind JOHN (the x [Be in the least interested(x)])]
p
b. Bev-ind JOHN (the x [NOT [Be in the least interested(x)]])
The answer lies in the fact that, in the corresponding SA, the NPI in the least
requires a negation immediately over the S-structure in which it occurs, as in
(11.8b), which corresponds to The one who isn’t in the least interested is John,
expressing (11.7c).8 (11.8a), however, is unwellformed already at SA-level,
because there is too much intermediate structure between the negation and
in the least, causing the ungrammaticality of (11.7b). These facts are thus
explained by the two assumptions (a) that the grammatical analysis of TCM
is provided by an underlying (SA-level) cleft structure of the type exemplified
in (11.4) and (11.8), and (b) that the NPI in the least requires a negation
immediately over its own clause in SA-structure. (The fact that both assump-
tions happen to be unpopular in certain schools of linguistics does not
diminish their explanatory power.)
Now consider the German examples (11.9) and (11.10) (repeated from note
6 in Section 7.2.1 of Volume I). In German, the use of the conjunction sondern
(meaning ‘but’) is strictly limited to comment-correction. It has a metalin-
guistic flavour in that it says that a comment that was given earlier is to
be replaced by a new comment, as in (11.9). It thus requires a comment
negation, with scope over the whole sentence, and not a negation that belongs

8
It is marginally possible for in the least to be separated from the negation by an intervening factive
verb taking the clause in which in the least occurs as an object complement, as in:
(i) ?She didn’t realize that I was in the least interested.
Cases like (ii) are explained by the rule of NEGATIVE RAISING, which takes the negation out of the
embedded clause in SA-structure and places it in construction with the commanding verb believe :
p
(ii) She didn’t believe that I was in the least interested.
 Topic–comment modulation 393

semantically in the topic-clause, as in (11.7c) and the corresponding (11.8b).

Moreover, German, like Dutch, allows for the comment negation to precede
the comment, as in (11.9). (In English, this is possible only if the correction
follows immediately, as in (11.11a), as against the less grammatical (11.11b).)
But German (like Dutch) also allows the comment negation to be
incorporated into the Matrix-S, as in (11.10), which then has to rely on the
proper intonation contour for the TCM to be expressed—that is, with a
grade-2 peak on HERBERT and on SOHN and a grade-1 final rise on gelacht.
The point is now that with the proper intonation contour for comment
negation, as in (11.10a), the use of sondern is legitimate, but with the intona-
tion pattern appropriate for negation embedded in the topic, as in (11.10b),
which has a grade-1 final fall on gelacht, sondern is not allowed:

(11.9)
Nicht HERBERT hat gelacht (sondern sein SOHN).
not Herbert has laughed (but his son)
It wasn’t HERBERT who laughed (but his son).

(11.10) a. HERBERT hat nicht gelacht (sondern sein SOHN).

Herbert has not laughed (but his son)
p
It wasn’t HERBERT who laughed (but his son).

b. HERBERT hat nicht gelacht (*sondern sein SOHN).

Herbert has not laughed (but his son)
!!It was HERBERT who didn’t laugh (*but his son).
(11.11) a. Not Herb but his son laughed.
b. ?*Not Herb laughed but his son.
The argument is, therefore, that the use of sondern is sensitive to the scope
of the negation in TCM structures of the kind shown in (11.4) and (11.8) and
expressed in different intonational contours. Without a TCM analysis, which
expresses the scope difference explicitly, the facts shown in (11.9) and (11.10)
remain unexplained.
Finally, let us look again at an example quoted in Section 3.2 in Volume I, to
do with reflexivization. Consider the little dialogue between a father and his
young son, who is crying because he has just hurt his knee:
394 The Logic of Language

(11.12) Father: Well-educated boys don’t cry.

Son: I didn’t educate me, YOU did!
It is now widely accepted, and rightly so, that reflexivization is, in principle,
an internal lexical process whereby a transitive predicate is applied to the
subject: self-wash, self-admire, and so on. In some languages reflexivization is
not expressed at all, but often it is expressed by means of verb morphology or
reflexive oblique pronouns, or, iconically, through so-called body-reflexives, as
in the French-based Creole language of the island Mauritius, where one finds
sentences like (Seuren 1995):9
(11.13) Li pe lav so lecôr.
he CONT wash his body
He is washing himself.
In English, reflexivity is expressed mainly through the suffix -self attached
to certain pronouns, but is not expressed in, for example, the possessive
pronouns his, her, their, unlike Latin or Swedish, which have reflexive
forms for reflexive third-person possessive pronouns (suus in Latin; sin
in Swedish). English also allows for a blocking of reflexivization in first-
person uses, as in:
(11.14) a. I hate me.10
b. I don’t know me any more.
In such sentences the non-reflexive nature of the predicate is emphasized, so
that the ego is presented as a third person, which suggests self-alienation. But
there is no truth-conditional difference with the corresponding fully reflex-
ivized sentences:
(11.15) a. I hate myself.
b. I don’t know myself any more.
The point of example (11.12) is that the son’s snappy reply does not express
any form of self-alienation and that it is truth-conditionally different from:

9
Classical Attic Greek had a special set of (pristine Indo-European) third-person reflexive
pronouns (hou for the genitive, hoi for the dative, and he for the accusative) reserved for use in
embedded object clauses where an oblique term refers back to the subject of the main clause, as in
Platoi said that Crito had not listened to himi.
10
This form of non-reflexivity is not possible in, for example, Dutch, where *Ik haat mij is
ungrammatical and must be Ik haat mezelf, also when one wants to express the kind of self-
alienation expressed by (11.14a,b).
 Topic–comment modulation 395

(11.16) I didn’t educate myself, YOU did!

I see no other way of providing a principled answer to this fact than to
assume, for the son’s reply in (11.12), an underlying SA-structure of the
form shown in (11.17a), where the verb educate is not used reflexively, as
opposed to (11.17b), where it is:
(11.17) a. NOT[Bev-ind I (the x [x educated me])]
b. NOT[Bev-ind I (the x [x self-educated])]
The lack of reflexivization observed in (11.12) is thus explained by the
assumption of an SA-level TCM cleft construction with the predicate Bev-ind.
Further evidence can no doubt be adduced, but we shall leave it at this.

11.3 The comment-predicate Bev

We will now say a few words about the specific nature of TCM, in particular
about the predicate Bev (discussed earlier in Section 3.2 of Volume I), even if
it must be admitted that this whole area is fraught with problems, most of
which still elude us. With this proviso, we posit that all cleft and pseudocleft
sentences (but not all sentences with an emphatic-contrastive accent on one
constituent; see Section 11.4) are characterized by the main predicate Bev,
normally realized in English as the verb be, as in example (11.4) cited above.
The specific use of this predicate is to specify values for function arguments.
That Bev is not at all identical with the identity predicate Be expressing
identity (¼), is not generally recognized. In fact (see Section 5.3.2 in Volume
I), Quine (1953: 143–4) mistook the value-assigning predicate Bev for the
identity predicate Be in his famous but fallacious argument that (11.18c) is a
non-sequitur because it is a case of illicit substitution of co-referential terms
salva veritate (SSV), as a parallel to Frege’s even more famous and valid
argument that (11.19c) is a non-sequitur because it is a case of illicit SSV:
(11.18) a. The number of planets is nine.
b. Nine is necessarily greater than seven.
c. Ergo: !The number of planets is necessarily greater than seven.
(11.19) a. The Morning Star is the Evening Star.
b. John believes that the Morning Star is inhabited.
c. Ergo: !John believes that the Evening Star is inhabited.
Frege was right (a) because the predicate Believe introduces an intensional
context, where SSV is blocked and (b) because (11.19a) is an identity state-
ment. Quine was wrong (a) because it is not clear that the predicate
396 The Logic of Language

Necessarily introduces an intensional context (see Section 7.2.1.3) and (b)

above all because (11.18a) is not an identity statement.
Montague (1973: 239), referring to ‘a puzzle due to Barbara Hall Partee’,
discusses the problem of sentences like ‘The temperature is ninety and rising’,
which should carry the absurd entailment that ninety is rising. He seeks a
solution in treating nouns like temperature or price (representing a class of
nouns for which he makes an explicit exception) as denoting ‘individual
concepts’, not individuals, but the solution is ad hoc and Montague still treats
Bev as the identity predicate ‘¼’, failing to see that Bev is a value-assigning
predicate in its own right. Since Montague (1973), no improvement has been
proposed on that faulty analysis.
Meanwhile we distinguish between the two. And we make a further dis-
tinction between a value-specifying predicate for individuals, written Bev-ind,
for categories, written Bev-cat, and for values, written Bev-val. Thus, for exam-
ple, the TCM-structure (11.20a), which may be regarded as an answer to the
question (11.20b), specifies not an individual but the category to which the
thing that Bert sold belongs. The subscript ‘v-cat’ makes it clear that what is
asked for is not the identification of an individual but of a category of
individuals (or of a substance, as when Bert sold milk):
(11.20) a. Bev-cat a car the x [Bert sold x ]
[¼Bert sold A CAR]
b. Bev-cat WHAT? the x [Bert sold x ]
[¼What did Bert sell?]
It is important to note that the constituent a car in (11.20a) does not represent
an underlying existential quantifier but denotes a category. We may regard
(11.20a) as providing the value for the discourse-determined situation at hand
in the function ‘what Bert sold’ (cf. Scharten 1997: 63).11

11
Significantly, Montague wrote (1970: 217):
We have taken the indefinite article ‘a’ always indicating existential quantification, but in some situations it may also
be used universally, and indeed, in precisely the same way as ‘any’; such is the case with one reading of the ambiguous
sentence ‘a woman loves every man such that that man loves that woman’.
That is, Montague held that, apart from cases where it is used generically, the indefinite article always
expresses existential quantification. This implies, as he admits, that John is an American should be
analysed as ‘There is an x such that x is an American and such that John is identical with x’. Needless
to say, this analysis stands in no relation to linguistic reality. It just illustrates again the narrow
fixation on quantification at the expense of other, less studied, semantic categories.
 Topic–comment modulation 397

Besides Bev-ind and Bev-cat, we also have Bev-val, which specifies the value of a
parameter for other than individuals or categories. For example, a sentence like
(11.21a), with the SA (11.21b), specifies the temperature of the room in question,
and (11.22a,b) specifies the cardinality of the set of John’s children:
(11.21) a. The temperature of the room is twelve degrees.
b. Bev-val twelve degrees the x[the temperature of the room is x]
(11.22) a. John has four children. / The number of John’s children is four.
b. Bev-val four the x[the cardinality of the set of John’s children is x]
The temperature parameter in (11.21) involves a function from objects
(possibly places) and times to temperature values. The number parameter
in (11.22) involves a function from sets to cardinality values. In similar fashion
cases can be analysed involving parameters for names, telephone numbers,
dates, etc.
As argued in Seuren (1993) and Scharten (1997), this shows the semantic,
nonpragmatic, nature of number specifications, generally mistaken for cases
of existential quantification. In both the semantic and the pragmatic litera-
ture, sentences like (11.22a) are analysed as existentially quantified, and are
accordingly taken to have the literal meaning ‘John has at least four children’,
which must then be changed into ‘John has precisely four children’ by means
of pragmatic principles that are generally too soft and not always justifiable.
The much more obvious reading in which, literally, a precise value is assigned
to the cardinality function for the set of John’s children is entirely neglected,
owing no doubt to the general neglect of parameters and value assignments in
standard formal semantics.12
The same goes for a question like Which month in the year has 28 days? The
most obvious answer is February, in the reading ‘the month of the year such
that the number of days it has is 28 is February’. But the answer All months
have 28 days is also justifiable, though clearly less probable. For some strange
reason, the latter answer is generally thought to be more ‘logical’ than the
former.
One notes that (11.21a) and (11.22a) are both analysed as TCM-structures.
Yet they have no corresponding cleft or pseudocleft surface structures. Sen-
tences like (11.23a,b) are highly artificial, if they are grammatical at all:

12
A Bev-analysis will also help out on the problem, raised in, for example, Donnellan (1966),
Kripke (1972, 1980), and Neale (1990), of the reference function for the Pope in sentences like The Pope
was Polish in 2000 but is German now, which does not imply that the Pope has changed nationality. The
sentence is now read as ‘the x such that x was the value of the parameter [Bev-cat the Pope] in 2000
was Polish but the x such that x is the value of the parameter [Bev-cat the Pope] now is German’.
398 The Logic of Language

(11.23) a. ?*It is twelve degrees that the temperature of the room is.
b. ?*It is four children that John has./?*It is four that the number of
John’s children is.
This is no doubt because sentences like (11.21a) and (11.22a) already possess a
value-assigning topic–comment structure, providing a value on a lexicalized
parameter such as temperature, price, or number under their SA-predicate
Bev-val. Although it is technically possible to topicalize again an element out
of such structures, leading to ‘double topicalization’, the result will be felt as
unnatural—though languages may well differ in this respect.
In general, it must be observed that, but for a few notable exceptions,
existing grammatical as well as formal semantic theories almost totally neglect
constructions involving the assignment of values to given parameters. This
means that the whole area of measurable gradable adjectives like broad, deep,
high, heavy, far, hot, old, etc., along with measure predicates like weigh, cost,
span, contain, and so on, has been left virtually untouched, which is a serious
handicap for the integration of topic–comment structure into an overall
theory of language.

11.4 Only, even, and Neg-Raising13

Let us now have a cursory look at the words only and even (and their
equivalents in other languages). They are usually called ‘focus particles’,
reflecting the fact that they are intimately related with ‘focusing’—that is,
with TCM. We will now try to penetrate a little more deeply into the
semantics and the grammar of these two particles, in the light of the presup-
position theory developed in the previous chapter and of what we have
glimpsed so far about TCM.
Both only and even require focusing, though in different ways. Both are
treated semantically as sentential operators. They seem to differ in that only
requires the ordinary TCM predicate [Bev NP] (where ‘NP’ stands for a
referring expression and takes a marked accent), whereas even requires an
underlying predicate ‘contains NP as a member’, symbolized as ‘[3 NP]’, the
inverse of the well-known ‘2 C’ meaning ‘is a member of class C’. More
comment follows in a moment.

13
I am indebted to Larry Horn, THE expert on only, even, and Neg-Raising, for useful critiques and
comments.
 Topic–comment modulation 399

I follow the standard semantic analysis of the operator only, which says that
only presupposes the truth of the argument-S and asserts that no other entity
satisfies the main predicate of the argument-S. This analysis appears to
provide the best fit to the available facts (pace those analyses that propose a
laxer, purely existential presupposition ‘someone laughed’ for (11.24a)). Only
JÓHN laughed, as in (11.24a) is thus taken to presuppose that John laughed and
to assert that nobody else did. Sentence (11.24a) is analysed at SA-level as
(11.24b), intuitively paraphrased as (11.24c):14
(11.24) a. Only JÓHN laughed.
b.
S

Pred S
only
Pred NP
Bev-ind John

Det S
the x

Pred S
Past

Pred NP
laugh x

c. The one who laughed was only JÓHN.

[Bev-ind NP] is described semantically as nonextensional (or intensional)
with regard to its subject term, so that, according to the reference hierarchy
discussed in Section 3.5.2 of Volume I, the reference object of the subject term
must have actual existence when the predicate used (laugh) is extensional with
regard to the argument term in question, whereas it may have virtual being
when the predicate in question is intensional with regard to the term at issue.
This makes it possible to use an intensional predicate like imaginary instead of
the extensional predicate laugh: Only JÓHN is imaginary still presupposes that
John is imaginary and asserts that nobody else is. The sentences (11.25a-c)
below thus presuppose that someone actually laughed and assert that this
person was John, whereas (11.24a) and (11.26a,b) presuppose that John
laughed and assert that nobody else did.

14
For details regarding the grammatical theory employed, see Seuren (1996).
400 The Logic of Language

(11.25) a. Who laughed was JÓHN.

b. It was JÓHN who laughed.
c. JÓHN laughed.
How does (11.24b) end up as the surface structure (11.24a) or, for that
matter, as (11.26a) or (11.26b) below? When we demonstrate this process in
terms of the theory of Semantic Syntax (Seuren 1996; see also Section 7.2.1 in
Volume I), we see that, leaving only out of account for the time being, without
Cleft Deletion, the grammar yields either the pseudocleft (11.25a) or the cleft
(11.25b). With Cleft Deletion, which is licensed by the predicate Bev and
lowers the predicate [Bev-ind John] to the position of the variable x (the
subject term of laugh), deleting Bev-ind along with Det[the x] and their
immediately dominating nodes S and NP, the result is (11.25c), where JÓHN
has carried with it the sentence-nuclear accent it naturally has in virtue of
being the main SA-predicate.
Further lowering of only makes only land onto John, the remainder of the
SA matrix predicate, placing it to the left of John, according to the general
right-branching setting for the syntax of English. This gives either (11.24a) or
(11.26a) or (11.26b):
(11.26) a. Who laughed was only JÓHN.
b. It was only JÓHN who laughed.
The description is somewhat oversimplified, but further details will be pro-
vided in a moment, when the particle even is discussed.
First we discuss the ordinary negation of (11.24a):
(11.27) Not only JÓHN laughed.
As was pointed out in Section 10.4, (11.27) carries the same presupposition
as (11.24a): the negation, being in noncanonical position, necessarily pre-
serves the presuppositions of the L-proposition that forms its argument.
No particular problem seems to arise here. The negation word not in (11.27)
clearly represents the sentential (widest possible scope) negation of (11.24a)
above. Accordingly, the SA-tree of (11.27) equals that of (11.24a), except that
S[Pred[not]] is added at the top.
So far so good, or so it seems, which does not mean that we have exhausted
the treatment of only. Thus, we have given no account of the construction only
if, which we know to be far from trivial. Nor have we explained the connec-
tion between the only discussed here and the only found in sentences like Only,
there is still the problem of what we will do next, where only is roughly
equivalent with but (the same phenomenon is observed in German (allein),
 Topic–comment modulation 401

Dutch (alleen), French (seulement), Italian (solo, soltanto), Modern Greek

(móno), and lots of other languages). And many other difficulties remain.15
But for the purpose of the present work, we must leave it at this.
Matters stand rather differently with regard to even. Semantically, a sen-
tence like (11.28b) is not the negation of (11.28a):16
(11.28) a. Even JÓHN laughed.
b. Not even JÓHN laughed.
In fact, as was observed by Horn (1989: 151), there is no way of negating a
sentence like (11.28a) in such a way that (11.28a) and its negation are contra-
dictories. Sentence (11.28b) is anything but the negation of (11.28a). What it is,
we shall see in a moment.
Moreover, whereas English has (11.28a,b), these sentences do not have
cleft or pseudocleft alternatives, since both (11.29a) and (11.29b) are ungram-
matical:
(11.29) a. *It was (not) even JÓHN who laughed.
b. *Who laughed was (not) even JÓHN.
This we account for by assuming that (11.30a) does not contain, at SA-level,
the predicate [Bev-ind NP], but rather the predicate [3 NP] (‘contains NP as a
member’ or ‘includes NP’), which does not allow for clefting and requires the
incorporation of [3 NP] into its subject-S. (It is assumed that the same SA-
predicate [3 NP] also underlies the particles too and also: John too laughed is
analysed as ‘Apart from others, the class of laughers included John’.) Assum-
ing that [Bev-ind NP] may occur independently of any higher-focusing parti-
cle, we can, in principle, explain the ambiguity of JÓHN laughed, which is
either ‘the one who laughed was John’ or ‘John was among the laughers’,
perhaps with a slight intonational difference.
The process of incorporation of [3 NP] into the S-structure of the subject-
NP is closely akin to Cleft Deletion and consists in placing [3 NP] in (11.30b)
(where, as before, the variable x ranges over plural objects or sets) in the
position of the subject term of laugh, thereby deleting 3, leaving only JÓHN
(with marked accent inherited from its being the main SA-predicate), and
deleting the S-node over Pred[3 John] as well as Det[the x ] and its dominating

15
We have, it seems, solved the problem of only some, which clearly excludes all. This is solved by
the assumption, made and developed in Chapter 3, that the meaning of some is rooted in basic-natural
predicate logic, where some excludes all.
16
Larry Horn pointed out to me that sentence-initial not even constitutes a counterexample (the
only one that has come to light so far) to my generalization (Section 10.4) that the negation in
noncanonical position is per se presupposition-preserving (see note 8 in Chapter 10).
402 The Logic of Language

NP-node. This leaves the SA-tree (11.31a), which then undergoes the standard
cyclic treatment, whereby the marked accent on JÓHN is preserved throughout.
(11.30) a. Even JÓHN laughed.
b.
S

Pred S
even
Pred NP
⊃ John
—

Det S
the –
x
Pred S
Past

Pred NP
–
x
laugh

c. The class of those who laughed contained even JÓHN.

(11.31) a.
a. S b. S

Pred S Pred S
even even
Pred S Pred S
Past not

Pred NP Pred S
laugh JOHN Past

Pred NP
laugh JOHN

(11.31b) is taken to underlie (11.28b) after the incorporation of [3 NP] into

the subject-S: not negates laugh, not the whole sentence. (11.28b) is to be
paraphrased as ‘the class of those who did not laugh contained even JÓHN’.
Like [Bev-ind NP], [3 NP] is described semantically as not requiring actual
existence of the reference value of its subject term (see (11.30b)), making it
possible to say (Not) even JÓHN is imaginary in complete parallelism to (Not)
even JÓHN laughed. We say that a sentence like Even JÓHN laughed presupposes
that there is a class of laughers and asserts that John is an unexpected or
unlikely member of that class. Analogously, we say that Even JÓHN did not
 Topic–comment modulation 403

laugh, or, perhaps more idiomatically, Not even JÓHN laughed, presupposes a
class of non-laughers and asserts that John is an unexpected or unlikely
member of thát class (cf. Horn 1989: 151). Semantically, therefore, the negation
in Not even JÓHN laughed belongs in the S-structure of the subject-NP of the
predicate [3 John] and is not a negation creating the contradiction of its
nonnegated counterpart.
But how on earth does the negation in Not even JÓHN laughed land in that
position, while it originates as a negation internal to the subject term of the
SA-predicate [3 John]? What grammatical sorcery has been going on here?
Let us note first that English is well-nigh unique in having the negation in
sentence-initial position in sentences of this nature. Italian comes close in that
it has anche or anzi for ‘even’ and neanche (also (nem)manco or nemmeno) for
‘not even’. But French says Même JEAN n’a pas ri, while German says Sogar
JOHANN hat nicht gelacht and Dutch has Zelfs JAN heeft niet gelachen—all with
the negation in construction with the finite verb.
Given a structure like (11.31b), one is tempted to think of what is known as
the process of NEG-RAISING (NR), much maligned in the (pragmatically orient-
ed) literature yet, in my view, hard to deny as a real grammatical process
found in the grammars of many languages.17 In simple terms, NEG-RAISING is
the phenomenon that a negation which semantically belongs in the scope of a
predicate ends up in surface structure to the left of that predicate and thus
taking scope over it. Well-known examples are sentences like I don’t think he’ll
make it, which does not mean, literally, ‘it is not the case that I think that he
will make it’ but rather ‘I think that he will not make it’. Likewise for a
sentence like I don’t want to die, which does not mean what it says but rather
‘I want not to die’. Nineteenth-century normative grammars told one to avoid
such constructions because they ran counter to logic!
It must be observed that NR is less likely to occur in cases where the that-
clause is the comment in a TCM-structure. For example, when the discourse
has given rise to the (implicit or explicit) question ‘What does Harry believe?’,
then the appropriate answer form is rather one without NR, such as Harry
believes that the crisis will not be over next month. An idiomatic instance, as
was once pointed out to me by my Amsterdam colleague Wim Klooster, is
I thought you’d never come, rather than I never thought you’d come.
Horn argues (1971; 1978; 1989: 330–61) extensively that NR-phenomena are
the result of pragmatic factors such as politeness strategies, understatement,
hedging, or irony. In polite society, one prefers to avoid direct clashes of

17
For an admirable and well-nigh complete survey of the entire complex question of NR and the
literature pertaining to it, see Horn (1978; 1989: 308–30).
404 The Logic of Language

opinion and thus expresses oneself rather in the weaker terms of ‘I don’t
believe that . . .’ than in the crass terms of ‘I believe that not . . . ’. One prefers
to say She is unlikely to come, rather than She is likely not to come, or This is not
good rather than This is bad, etc. etc. (read Horn 1989: it’s very well written).
I fully agree with Horn to the extent that he seeks a pragmatic origin for NR: it
seems to me that his arguments are strong and convincing. I also agree with
his conclusion (Horn 1978: 215–16):
. . . that NR originates as a functional device for signalling negative force as early in a
negative sentence as possible. [. . .] NR must be regarded as a rule in the synchronic
grammar of English and other languages. NR would thus constitute an example of a
pragmatic process which has become grammaticized or syntacticized.18

With Horn, I argue that NR has indeed become part of the grammar and
lexicon of English and many, perhaps all, natural languages. This appears
from a number of facts. First, one sees that NR is, at the level of linguistic
description and not at the noncommittal pragmatic level, associated with
different verbs in different languages. Thus, English hope and its Italian
counterpart sperare do not induce NR but their French, Dutch, and German
cousins do.
Then, as shown in Seuren (2004: 178–81; see also the discussion around
examples (7.19–7.22) in Chapter 7 of Volume I), certain Negative Polarity
Items (NPIs), such as in the least or yet or the slightest, are licensed only when
occurring in the immediate scope of negation:
p
(11.32) a. She knows that John hasn’t arrived yet.
b. *She doesn’t know that John has arrived yet.
p
(11.33) a. Many people are not in the least interested.
b. *Not many people are in the least interested.

18
The same is often found for grammatical categories that find their origin in natural phenomena
but have been extended to becoming autonomous grammatical categories. Grammar often gets
‘started up’ by what takes place in the world or in communicative situations and then ‘takes over’ in
its own right. Consider, for example, grammatical gender distinctions. To the extent that they
correspond with natural gender, the grammatical gender of nouns is almost entirely predictable: la
femme, die Frau (both ‘the woman’) are predictably feminine (though German also has the neuter
nouns das Weib (‘the woman’, used derogatorily) and das Mädchen (‘the girl’), which is a diminutive).
But this natural motivation is no longer valid when these gender distinctions are applied to words
denoting objects that simply have no natural gender. Thus French has the masculine le soleil for ‘the
sun’, but German has the feminine die Sonne, and, ironically, the French and German words for ‘moon’
are the feminine la lune and the masculine der Mond, respectively. Speculations about totally different
French and German ‘popular spirits’ were rife in the nineteenth century but have been abandoned
now.
 Topic–comment modulation 405
p
(11.34) a. Many people don’t show the slightest interest.
b. *Not many people show the slightest interest.
Yet we see that these three NPIs occur naturally and without a hint of
ungrammaticality in sentences like:
p
(11.35) a. I don’t think John has arrived yet.
p
b. I don’t expect John to be in the least interested.
p
c. Don’t expect John to take the slightest interest in your work.
Such facts are readily explained when one assumes SA-forms where the
negation stands over the embedded clauses and not over the main sentence
and one accepts NR as an automatic grammatical process not affecting the
semantics of the sentences involved. But they are hard to explain in purely
pragmatic terms. One would have to present a purely pragmatic explanation
of the occurrence restrictions of NPIs—a feat that has not so far been
achieved. As in the case of NR, the very phenomenon of NPIs may have
had a pragmatic origin, but their subsequent grammaticalization is hard to
deny.
The advantage of the grammaticalization view is that the application range
of NR can now be extended to cover less obvious cases. One may think, for
example, of the propositional operator AND as a Neg-Raiser, whereby and is
converted into its counterpart or. AND (NOT-P, NOT-Q) thus becomes NOT (OR
(P,Q)), as in (11.36a), which becomes (11.36b):
(11.36) a. He doesn’t like planes and he doesn’t like trains.
b. He doesn’t like planes or trains.
It has been noted repeatedly in previous chapters that the converse does not
work:
(11.37) a. He doesn’t like planes or he doesn’t like trains.
b. !!He doesn’t like planes and trains.
This also explains, in principle, the plural were in a sentence like
(11.38a), derived from an underlying (11.38b). One notes that (11.38c) is
ungrammatical:
(11.38) a. I don’t think John or Harry were late.
b. I don’t think John was late and I don’t think Harry was late.
c. *John or Harry were late.
In fact, English so far appears to be a Neg-Raiser, as we say not so far
achieved, and not so far not achieved, which is what it means. In like manner,
406 The Logic of Language

we now posit that English even is a Neg-Raiser (in accordance with Horn 1971:
132). This allows us to posit that an underlying (11.31b), repeated here
as (11.39a), is transformed into (11.39b), after which the syntactic Cycle
is free to operate.
(11.39)
a. S NR b. S
⇒
Pred S Pred S
even not
Pred S Pred S
not even
Pred S Pred S
Past Past

Pred NP Pred NP
laugh JOHN laugh JOHN

There is, however, an important corollary to this solution: it must be assumed

that the syntax of a language does not start with the classic Cycle (which
processes each S-cycle successively, starting from the most deeply embedded
or lowest S and working its way up till the highest S), but that there is a
PRECYCLE that reshuffles SA-trees before the Cycle can start to operate. This is
an almost totally unexplored area of research in the theory of grammar, but
there are plenty of indications that something like a Precycle must be admit-
ted.19 Neg-Raising and Cleft Deletion (along with the deletion of its [3 NP]
counterpart) would then have to be taken to be part of the Precycle. On this
assumption, the problem of the grammar and the semantics of (11.28b) would
be solved.

11.5 Why TCM is a semantic phenomenon: the SSV test

There is one further respect in which TCM-structure has been the victim of
myopia in semantic theory. Whereas Frege’s discovery of the nonsubstitutivity
of co-referring terms in intensional contexts triggered a vast research
programme that dominated twentieth-century semantics, no-one seems to
have noticed the nonsubstitutivity salva veritate of truth-conditionally
19
Larry Horn pointed out that the it in I don’t believe that John paid his taxes and Mary is quite sure
of it pronominalizes ‘that John didn’t pay his taxes’ and not ‘that John paid his taxes’, which would
imply that pronominalization must have occurred before NR. Although it is no longer assumed that
pronominalization is a syntactic process but is more likely to occur before the syntax starts to operate,
in the no-man’s land between propositional thought and language, the observation remains relevant
and interesting.
 Topic–comment modulation 407

equivalent TCM-structures in emotive or evaluative intensional contexts. It

was pointed out in Sections 3.2, 6.2.3.2, and 7.2.2 of Volume I and in Section
10.5.2 of the present volume that emotive factive and nonfactive predicates
create or continue subdomains where the substitution of otherwise truth-
conditionally equivalent TCM-structures makes a truth-conditional differ-
ence. Examples are:
(11.40) a. It surprised Ann that JOHN (and not Kevin) had sold the car.
b. It surprised Ann that John had sold THE CAR (and not the
speedboat).
(11.41) a. It angered Ann that JOHN (and not Kevin) had sold the car.
b. It angered Ann that John had sold THE CAR (and not the
speedboat).
(11.42) a. Ann hopes that JOHN (and not Kevin) has sold the car.
b. Ann hopes that John has sold THE CAR (and not the speedboat).
(11.43) a. It’s a good thing that JOHN (and not Kevin) has sold the car.
b. It’s a good thing that John has sold THE CAR (and not the
speedboat).
Clearly, it is possible for the (a)-sentences to be true while the (b)-sentences
are false, and vice versa. For example, Ann may hope that John and not Kevin
has sold the car because John is her husband and they need the sales
commission, while at the same time she regrets that it is the car that has
been put up for sale and not, say, the speedboat, as the latter would bring in
much more money. Or she may hope that John has sold the car and not the
speedboat because the car takes up too much space in the garage, while the
speedboat is well away from their home. At the same time, she may regret that
John had to do the selling because she knows how bad he is at getting the best
price. In fact, a sentence like (11.44) is entirely consistent:
(11.44) It is fortunate that it is JOHN who has sold the car, but unfortunate
that it is the CAR that has been sold.
Substitution salva veritate of truth-conditionally equivalent TCM-struc-
tures is a fully valid operation in non-emotive intensional contexts, as appears
from cases like (11.45a,b), with the non-emotive epistemic intensional verb
believe:
(11.45) a. Ann believes that JOHN has sold the car.
b. Ann believes that John has sold the CAR.
408 The Logic of Language

Here it is not possible for the one to be true while the other is false, even
though, of course, the two sentences differ as regards their anchoring condi-
tions for given discourse domains.
Yet, as was pointed out in Section 10.4, substitution salva veritate of truth-
conditionally equivalent TCM-structurings is again blocked under sentential
negation. Here, the truth-conditional difference resides in the presuppositional
truth conditions, as appears from the following examples:
(11.46) a. JÓHN hasn’t sold the car: he is a figment of Ann’s mind. PETER did.
b. !! JÓHN hasn’t sold the car: there never wás a car.
(11.47) a. John hasn’t sold the CÁR: there never wás a car. He sold the
SPEEDBOAT.
b. !!John hasn’t sold the CÁR: he is a figment of Ann’s mind.
Clearly, these observations are in full agreement with our analysis, according
to which the constituent under a grade-2 accentual peak is the underlying SA-
predicate Bev-ind, which induces no existential presupposition with regard to
the value specified. But non-accented constituents in the Matrix-S are subject
to the normal preconditions holding for the Matrix-predicate.
The conclusion is, therefore, that TCM is not merely a pragmatic
phenomenon but does contribute to sentence meaning. This again means
that type-level discourse-incremental properties of sentences cannot simply
be relegated to pragmatics but must be considered to be part of a semantic
theory of natural language.
 Bibliography
ABRAHAM, W. (1991) ‘Discourse particles in German: How does their illocutionary force
come about?’, in W. Abraham (ed.), Discourse Particles: Descriptive and Theoretical
Investigations on the Logical, Syntactic, and Pragmatic Properties of Discourse Particles
in German. Benjamins, Amsterdam/Philadelphia: 203–52.
ALFORD, M. (1995) ‘Isaac Newton: the first physicist’. http://www.physics.wustl.edu/~alford/
newton.html
ANDRADE MARTINS, S. (2004) Fonologia e Gramática Dâw. 2 vols. LOT, Utrecht.
ASBACH-SCHNITKER, B. (1979) ‘Die adversativen Konnektoren aber, sondern und but nach
negierten Sätzen’, in H. Weydt (ed.), Die Partikeln der deutschen Sprache. De Gruyter,
Berlin-New York: 457–68.
BARWISE, J. and R. COOPER (1981) ‘Generalized quantifiers and natural language’, Linguistics
and Philosophy 4/2: 159–219.
—— and J. PERRY (1983) Situations and Attitudes. MIT Press, Cambridge, MA.
BÄUERLE, R., U. EGLI, and A. VON STECHOW (eds), Semantics from Different Points of View.
Springer, Berlin-Heidelberg-New York.
BELLERT, I. (1972) ‘On certain syntactical properties of the English connectives and and but ’,
in S. Ploetz (ed.), Transformationelle Analyse. Die Transformationstheorie von Zellig
Harris und ihre Entwicklung. Athenaeum, Frankfurt: 327–56.
BLAKEMORE, D. and R. CARSTON (2005) ‘The pragmatics of sentential coordination with and ’,
Lingua 115: 569–89.
BLANCHÉ, R. (1966) Structures intellectuelles. J. Vrin, Paris.
BOCHEŃSKI, I. M. (1956) Formale Logik. Orbis Academicus, Freiburg/Munich.
BOËR, S. E. and W. G. LYCAN (1976) ‘The myth of semantic presupposition’, Indiana
University Linguistics Club.
BOLINGER, D. (1972) ‘Accent is predictable (if you’re a mind-reader)’, Language 48/3: 633–44.
BORKOWSKI, L. (1970) Jan Łukasiewicz: Selected Works. North-Holland, Amsterdam.
BROUWER, L. E. J. (1966) ‘Life, art, and mysticism’, Notre Dame Journal of Formal Logic 37:
389–429. (Transl. by W. P. van Stigt of L. E. J. Brouwer, Leven, Kunst en Mystiek, 1905.)
BROWN, L. and M. S. DRYER (2008) ‘The verbs for “and” in Walman, a Torricelli language of
Papua New Guinea’, Language 84/3: 528–65.
BUCKNER, E. (2007) ‘The fourth corner’, paper read at the international congress
‘The Square of Opposition’, Montreux, Switzerland, June 1–3, 2007.
http://maverickphilosopher.powerblogs.com/posts/1177106402.shtml.
BURLEIGH, W. (1988) Von der Reinheit der Kunst der Logik. Erster Traktat. Von den
Eigenschaften der Termini. (De puritate artis logicae. De proprietatibus terminorum).
Translated and edited by Peter Kunze, with introduction and commentary. Felix Meiner,
Hamburg.
410 Bibliography

BUSSE, A. (ed.) (1897) Ammonius in Aristotelis De Interpretatione Commentarius. Royal

Prussian Academy of Sciences. Georg Reimer, Berlin.
BUTTERWORTH, B. (1999) The Mathematical Brain. Macmillan, London.
CARNAP, R. (1956) Meaning and Necessity. A Study in Semantics and Modal Logic. Enlarged
edition. The University of Chicago Press, Chicago.
CARROLL, L. (¼ C. L. Dodgson) (1896) Symbolic Logic. Part I. Elementary. Fifth Edition.
Edited, in 1977, with annotations and an introduction by William Warren Bartley, III.
Harvester Press, Hassocks, Sussex, England.
CAVALIERE, F. (2007) ‘Sillogismi sfumati, triangolo oppositivo, quadrilatero numerico, inter-
bivalenza, con un’ appendice storica sulla quantificazione del predicato’. http://www.
arrigoamadori.com/lezioni/AngoloDelFilosofo/AngoloDelFilosofo.htm
CLANCHY, M. T. (1997) Abelard. A Medieval Life. Blackwell, Oxford.
CLARK, H. H. and D. WILKES-GIBBS (1990) ‘Referring as a collaborative process’, in
P. R. Cohen, J. Morgan, and M. E. Pollack (eds), Intentions in Communication. MIT
Press, Cambridge, MA: 463–93.
—— (1992) Arenas of Language Use. The University of Chicago Press, Chicago.
—— (1996) Using Language. Cambridge University Press, Cambridge.
COHEN, L. J. (1971) ‘The logical particles of natural language’, in Y. Bar-Hillel (ed.),
Pragmatics of Natural Language. Reidel, Dordrecht: 50–68.
COMRIE, B. (1986) ‘Conditionals: a typology’, in Traugott et al. (eds): 77–99.
COPI, I. M. (1961) Introduction to Logic. Macmillan, New York.
DE MORGAN, A. (1847) Formal Logic: or, The Calculus of Inference, Necessary and Probable.
Taylor and Walton, London.
—— (1858) ‘On the syllogism III, and on logic in general’, reprinted in P. Heath (ed.), On the
Syllogism and Other Logical Writings. Yale University Press, New Haven (1966): 74–146.
DE RIJK, L. M. (1956) Petrus Abaelardus, Dialectica. First Complete Edition of the Parisian
Manuscript. Van Gorcum/Hak and Prakke, Assen.
—— (1967) Logica Modernorum. A Contribution to the History of Early Terminist Logic.
Vol. II.1: The Origin and Early Development of the Theory of Supposition. Van Gorcum, Assen.
—— (1972) Peter of Spain (Petrus Hispanus Portugalensis) Tractatus, called afterwards
Summule Logicales. Van Gorcum, Assen.
—— (1985) ‘Walter Burley’s tract “De Exclusivis”. An edition’, Vivarium 23: 23–54.
—— (1986) ‘Walter Burley’s “De Exceptivis”. An edition’, Vivarium 24: 22–49.
—— (1992) Peter of Spain (Petrus Hispanus Portugalensis) Syncategoreumata. Brill, Leiden.
—— (2002) Aristotle: Semantics and Ontology. Vol. I: General Introduction. The Works on
Logic. Brill, Leiden.
DECLERCK, R. and S. REED (2001) Conditionals. A Comprehensive Empirical Analysis. Mouton
de Gruyter, Berlin.
DEHAENE, S. (1997) The Number Sense. Oxford University Press, New York.
—— (2005) ‘Evolution of human cortical circuits for reading and arithmetic: the
“neuronal recycling” hypothesis’, in S. Dehaene, J.-R. Duhamel, M. D. Hauser, and
G. Rizzolatti (eds), From Monkey Brain to Human Brain. MIT Press, Cambridge, MA:
133–57.
 Bibliography 411

—— V. IZARD, P. PICA, and E. SPELKE (2006) ‘Core knowledge of geometry in an Amazonian

indigene group’, Science 11 (January 2006): 381–4.
DIEWALD, G. and K. FISCHER (1998) ‘Zur diskursiven und modalen Funktion der Partikeln
aber, auch, doch und ja in Instruktionsdialogen’, Linguistica 38/1: 75–99.
DONNELLAN, K. (1966) ‘Reference and definite descriptions’, Philosophical Review 75:
281–304.
DOWTY, D. (1979) Word Meaning and Montague Grammar. Reidel, Dordrecht.
—— R. E. WALL, and S. R. PETERS (1981) Introduction to Montague Semantics. Reidel,
Dordrecht.
DUMMETT, M. E. (1973) Frege. Philosophy of Language. Duckworth, London.
EBBESEN, S. (1990a) ‘Porphyry’s legacy to logic: a reconstruction’, in Sorabji (ed.): 141–71.
—— (1990b) ‘Philoponus, “Alexander” and the origins of medieval logic’, in Sorabji (ed.):
445–61.
ECO, U. (1995) The Search for the Perfect Language. Blackwell, Oxford.
EVANS, J. St. B. T. and D. OVER (2005) If. Oxford University Press, New York.
FAUCONNIER, G. (1985) Mental Spaces: Aspects of Meaning Construction in Natural Language.
MIT Press, Cambridge, MA.
FILLMORE, Ch. J. and D. T. LANGENDOEN (eds) (1971) Studies in Linguistic Semantics. Holt,
New York.
FISCHER, K. (2000) From Cognitive Semantics to Lexical Pragmatics: The Functional Polysemy
of Discourse Particles. Mouton de Gruyter, Berlin.
FODOR, J. A. (1975) The Language of Thought. Harvester Press, Hassocks, Sussex.
—— (1983) The Modularity of Mind. An Essay on Faculty Psychology. A Bradford Book.
MIT Press, Cambridge, MA.
—— (1998) Concepts: Where Cognitive Science Went Wrong. Oxford University Press,
New York.
FOWLER, TH. (1892) The Elements of Deductive Logic. Clarendon Press, Oxford.
FREGE, G. (1884) Die Grundlagen der Arithmetik: eine logisch-mathematische Untersuchung
über den Begriff der Zahl. W. Koebner, Breslau.
—— (1892) ‘Ueber Sinn und Bedeutung’, Zeitschrift für Philosophie und philosophische
Kritik 100: 25–50. (Also in Patzig 1969: 40–65.)
—— (1969) Nachgelassene Schriften. Edited by H. Hermes, F. Kambartel, and F. Kaulbach.
Felix Meiner, Hamburg.
GAZDAR, G. (1979) Pragmatics. Implicature, Presupposition, and Logical Form. Academic
Press, New York-San Francisco-London.
GEACH, P. T. (1950) ‘Russell’s Theory of Descriptions’, Analysis 10: 84–8.
—— (1962) Reference and Generality. An Examination of Some Medieval and Modern
Theories. Cornell University Press, Ithaca, New York.
—— (1969) ‘Quine’s syntactical insights’, in D. Davidson and J. Hintikka (eds), Words and
Objections. Essays on the Work of W. V. Quine. Reidel, Dordrecht: 146–57. (Also in Geach
1972: 115–27.)
—— (1972) Logic Matters. Oxford, Blackwell.
GEIS, M. L. and A. M. ZWICKY (1971) ‘On invited inferences’, Linguistic Inquiry 2/4: 561–6.
412 Bibliography

GEURTS, B. and D. I. BEAVER (2007) ‘Discourse Representation Theory’, in E. N. Zalta (ed.),

The Stanford Encyclopedia of Philosophy. http://plato.stanford.edu/entries/discourse-
representation-theory/.
GILSON, É. (1978) Héloı̈se et Abélard. J. Vrin, Paris.
GINSBURG, H. P. and B. S. ALLARDICE (1984) ‘Children’s difficulties with school mathematics’,
in B. Rogoff and J. Lave (eds), Everyday Cognition: Its Development in Social Context.
Harvard University Press, Cambridge, MA: 194–219.
GREEN-PEDERSEN, N. J. (1980) ‘Walter Burley’s “De Consequentiis”. An edition’, Franciscan
Studies 40 (annual xviii): 102–66.
GREENBERG, J. H. (1963) ‘Some universals of grammar with particular reference to the order
of meaningful elements’, in J. H. Greenberg (ed.), Universals of Language. MIT Press,
Cambridge, MA: 73–113.
GROENENDIJK, J. A. G. and M. B. J. STOKHOF (1991) ‘Dynamic predicate logic’, Linguistics and
Philosophy 14: 39–100.
GUSSENHOVEN, C. (1984) On the Grammar and Semantics of Sentence Accents. Foris,
Dordrecht.
—— (2005) ‘Semantics of prosody’, in K. Brown (ed.), The Encyclopedia of Language and
Linguistics, second edition. Elsevier, Oxford: 170–3.
HAMBLIN, CH. L. (1970) Fallacies. Methuen, London.
HAMILTON, W. (1866) Lectures on Metaphysics and Logic. Vol. IV (Lectures on Logic. Vol. II,
2nd edition, revised). Edited by H. L. Mansel and J. Veitch. Blackwood and Sons,
Edinburgh-London.
HARRIS, M. B. (1986) ‘The historical development of SI-clauses in Romance’, in Traugott et
al. (eds): 265–84.
HEIM, I. R. (1990) ‘On the projection problem for presuppositions’, in S. Davis (ed.),
Pragmatics. A Reader. Oxford University Press, Oxford: 397–405.
HOEKSEMA, J. (1999) ‘Blocking effects and polarity sensitivity’, in J. Gerbrandy, M. Marx,
M. de Rijke, and Y. Venema (eds), JFAK. Essays Dedicated to Johan van Benthem on the
Occasion of his 50th Birthday. Amsterdam University Press, Amsterdam (not available in
hard copy). http://www.illc.uva.nl/j50/contribs/hoeksema/hoeksema.pdf
HORN, L. R. (1971) ‘Negative transportation: unsafe at any speed?’, in Papers from the
Seventh Regional Meeting of the Chicago Linguistic Society 1971. Chicago Linguistic
Society, Chicago Illinois: 120–33.
—— (1972) ‘On the Semantic Properties of Logical Operators in English’, distributed by
Indiana University Linguistics Club 1976.
—— (1978) ‘Remarks on Neg-Raising’, in P. Cole (ed.), Syntax and Semantics 9: Pragmatics.
Academic Press, New York-London: 129–220.
—— (1985) ‘Metalinguistic negation and pragmatic ambiguity’, Language 61/1: 121–74.
—— (1989) A Natural History of Negation. The University of Chicago Press, Chicago.
—— (1991) ‘Duplex Negatio Affirmat . . . : The economy of double negation’, in Papers
from the 27th Regional Meeting of the Chicago Linguistic Society 1991. Part Two: The
Parasession on Negation. Chicago Linguistic Society, Chicago, Illinois: 80–106.
 Bibliography 413

—— (1997) ‘All John’s children are as bald as the king of France: existential import and the
geometry of opposition’, in Papers from the 33rd Regional Meeting of the Chicago
Linguistic Society 1997. Papers from the Main Session. Chicago Linguistic Society,
Chicago, Illinois: 155–79.
—— (2000) ‘From if to iff: conditional perfection or pragmatic strengthening’, Journal of
Pragmatics 32/2: 289–326.
HUSSERL, E. (1900) Logische Untersuchungen. Vol 2: Untersuchungen zur Phänomenologie
und Theorie der Erkenntnis. Teil 1. Niemeyer, Halle.
ISARD, S. (1975) ‘Changing the context’, in E. L. Keenan (ed.), Formal Semantics of Natural
Language. Cambridge University Press, Cambridge: 287–96.
JANSSEN, TH. A. J. M. (1976) Hebben-konstrukties en indirekt-objektskonstrukties. HES
Publishers, Utrecht.
JASPERS, D. (2005) Operators in the Lexicon. On the Negative Logic of Natural Language. LOT,
Utrecht.
JESPERSEN, O. (1917) Negation in English and Other Languages. Det Kgl. Danske
Videnskabernes Selskab, Historisk-filologiske Meddelelser I,5. Andr. Fred. Hst and
sn, Copenhagen.
—— (1924) The Philosophy of Grammar. Allen and Unwin, London.
JOHNSON-LAIRD, P. N. (1986) ‘Conditionals and mental models’, in Traugott et al. (eds):
55–75.
KALEPKY, TH. (1928) Neuaufbau der Grammatik als Grundlegung zu einem wissenschaftlichen
System der Sprachbeschreibung. Teubner, Leipzig.
KAMP, H. (1981) ‘A theory of truth and semantic interpretation’, in J. A. G. Groenendijk, Th.
M. V. Janssen, and M. B. J. Stokhof (eds), Formal Methods in the Study of Language. Vol.
I. Mathematisch Centrum, Amsterdam: 277–322.
—— and U. REYLE (1993) From Discourse to Logic. Introduction to Model-Theoretic
Semantics of Natural Language, Formal Logic and Discourse Representation Theory.
Kluwer, Dordrecht.
KARTTUNEN, L. (1973) ‘Presuppositions of compound sentences’, Linguistic Inquiry 4/2:
169–93.
KEENAN, E. L. (2003) ‘The definiteness effect: semantics or pragmatics?’, Natural Language
Semantics 11: 187–216.
KING, P. (2005) ‘Ockham’s Summa Logicae’, in J. Shand (ed.), Central Works of Philosophy.
Volume I. Acumen Publishing, Chesham: 242–69.
KIPARSKY, P. and C. KIPARSKY (1971) ‘Fact’, in D. D. Steinberg and L. A. Jakobovits (eds),
Semantics. An Interdisciplinary Reader in Philosophy, Linguistics and Psychology.
Cambridge University Press, Cambridge: 345–69.
KLEENE, S. C. (1938) ‘On notation for ordinal numbers’, Journal of Symbolic Logic 3: 150–5.
—— (1952) Introduction to Metamathematics. North-Holland, Amsterdam.
KLIMA, G. (1988) Ars Artium. Essays in Philosophical Semantics, Mediaeval and Modern.
Doxa Library. Institute of Philosophy, Hungarian Academy of Sciences, Budapest.
KNEALE, W. and M. KNEALE (1962) The Development of Logic. Clarendon Press, Oxford.
KRATZER, A. (1979) ‘Conditional necessity and possibility’, in Bäuerle et al. (eds): 117–47.
414 Bibliography

KRIPKE, S. (1972) ‘Naming and necessity’, in D. Davidson and G. Harman (eds), Semantics of
Natural Language. Reidel, Dordrecht: 253–355.
—— (1980) Naming and Necessity. Blackwell, Oxford (¼ Kripke 1972).
LABOV, W. (1972) Language in the Inner City. Studies in Black English Vernacular. University
of Pennsylvania Press, Philadelphia.
LAKOFF, R. (1971) ‘If ’s, and’s, and but’s about conjunctions’, in Fillmore and Langendoen
(eds): 114–49.
LANG, E. (1977) Semantik der koordinativen Verknüpfung. Akademie-Verlag, Berlin.
LANGENDOEN, D. T. and H. B. SAVIN (1971) ‘The projection problem for presuppositions’, in
Fillmore and Langendoen (eds): 55–60.
LENZEN, W. (2008) ‘Ploucquet’s “refutation” of the traditional Square of Opposition’, Logica
Universalis 2: 43–58.
LEVINSON, S. C. (2000) Presumptive Meanings. The Theory of Generalized Conversational
Implicature. MIT Press, Cambridge MA.
LEWIS, D. (1979) ‘Scorekeeping in a language game’, in Bäuerle et al. (eds): 172–87.
LEWIS, G. L. (1984) Turkish Grammar. Oxford University Press, Oxford.
LIPPS, TH. (1893) Grundzüge der Logik. Dürr, Leipzig.
LÖBNER, S. (1990) Wahr neben Falsch. Duale Operatoren als die Quantoren natürlicher
Sprache. Niemeyer, Tübingen.
LONDEY, D. and C. JOHANSON (1987) The Logic of Apuleius: Including a Complete Latin Text
and English Translation of the Peri Hermeneias of Apuleius of Madaura. Brill, Leiden.
LUKASIEWICZ, J. J. I. (1934) ‘Z historii logiki zdán’, Przegla̧d Filozoficny 37: 417–37. [German
transl.: ‘Zur Geschichte der Aussagenlogik’, Erkenntnis 5 (1935): 111–31. English transl.:
‘On the history of the logic of propositions’, in McCall (1967: 66–87); reprinted in:
Borkowski (1970: 197–217).]
MCCALL, S. (1967) Polish Logic 1920–1939. Oxford University Press, Oxford.
MCCAWLEY, J. D. (1967) ‘Meaning and the description of languages’, Kotoba no Uchu 2/9:
10–18; 2/10: 38–48; 2/11: 51–7. (Also in McCawley 1973: 99–120.)
—— (1972) ‘A program for logic’, in D. Davidson and G. Harman (eds), Semantics of
Natural Language. Reidel, Dordrecht: 498–544.
—— (1973) Grammar and Meaning. Papers on Syntactic and Semantic Topics. Taishukan,
Tokyo.
—— (1981) Everything that Linguists have Always Wanted to Know about Logic* *but were
ashamed to ask. Blackwell, Oxford.
MCLEOD, E. (1971) Héloı̈se. A Biography. Chatto and Windus, London.
MATHESIUS, V. (1928) ‘On linguistic characterology with illustrations from modern English’,
in Actes du Premier Congrès International de Linguistes à La Haye. Reprinted in: J. Vachek
(ed.), A Prague School Reader in Linguistics. Indiana University Press, Bloomington,
Indiana (1964): 59–67.
—— (1939) ‘O tak zvaném aktuálnim cleneni vetném’ (On the so-called functional
sentence perspective), Slovo a Slovesnost 5: 171–4.
MEISER, C. (1880) Anicii Manlii Severini Boetii commentarii in librum Aristotelis Perı̀
Hermēneı́as. Pars posterior. Editio secunda. Teubner, Leipzig.
 Bibliography 415

MEYER-LÜ BKE, W. (1899) Romanische Syntax (Grammatik der romanischen Sprachen III).
Reisland, Leipzig.
MIGNUCCI, M. (1983) ‘La teoria della quantificazione del predicato nell’antichità classica’,
Anuario Filosófico de la Universidad de Navarra 16: 11–42.
MONTAGUE, R. (1970) ‘English as a formal language’, in B. Visentini (ed.), Linguaggi nella
società e nella tecnica. Edizioni di Comunità, Milan: 189–223.
—— (1973) ‘The proper treatment of quantification in ordinary English’, in K. J. J.
Hintikka, J. M. E. Moravcsik, and P. Suppes (eds) Approaches to Natural Language.
Proceedings of the 1970 Stanford Workshop on Grammar and Semantics. Reidel,
Dordrecht: 221–42.
MOODY, A. E. (1953) Truth and Consequence in Mediæval Logic. North-Holland,
Amsterdam.
MOSTOVSKI, A. (1957) ‘On a generalization of quantifiers’, Fundamenta Mathematica 44:
12–36.
MOXEY, L. M. and A. J. SANFORD (1986/7) ‘Quantifiers and focus’, Journal of Semantics 5:
189–206.
MULLALLY, J. P. (1945) The Summulæ Logicales of Peter of Spain. (Publications in Medieval
Studies VIII) The University of Notre Dame Press, Indiana.
NEALE, S. (1990) Descriptions. MIT Press, Cambridge, MA.
NUCHELMANS, G. (1973) Theories of the Proposition. Ancient and Medieval Conceptions of the
Bearers of Truth and Falsity. North-Holland, Amsterdam.
PARSONS, T. (1997) ‘The traditional Square of Opposition. A biography’, Acta Analytica 18:
23–49.
—— (2006) ‘The traditional Square of Opposition’, in E. N. Zalta (ed.), The Stanford
Encyclopedia of Philosophy (October 1, 2006 revision) http://plato.stanford.edu/entries/
square/.
—— (2008) ‘Things that are right with the traditional Square of Opposition’, Logica
Universalis 2: 3–11.
PEIRCE, CH. S. (1974) Collected Works. Vol. II: Elements of Logic. Edited by Charles
Hartshorne and Paul Weiss. Harvard University Press, Cambridge, MA.
PETERS, S. and D. WESTERSTÅHL (2007) Quantifiers in Language and Logic. Oxford University
Press, Oxford.
PICA, P., C. LEMER, V. IZARD, and S. DEHAENE (2004) ‘Exact and approximate arithmetic in an
Amazonian indigene group’, Science 306 (October 2004): 499–503.
PICKERING, M. J. and S. C. GARROD (2004) ‘Toward a mechanistic psychology of dialogue’.
Behavioral and Brain Sciences, 27/2: 169–225.
QUINE, W. V. O. (1952) Methods of Logic. Routledge and Kegan Paul, London.
—— (1953) From a Logical Point of View. Harvard University Press, Cambridge, MA.
—— (1960) Word and Object. MIT Press, Cambridge, MA.
REINHART, T. (1983) Anaphora and Semantic Interpretation. Croom Helm, London.
RESCHER, N. (1969) Many-valued Logic. McGraw-Hill, New York.
ROSCH, E. (1975) ‘Cognitive representations of semantic categories’, Journal of Experimental
Psychology: General 104: 192–233.
416 Bibliography

ROSIER-CATACH, I. (2003a) ‘Abélard et les grammairiens : sur le verbe substantif et la

prédication’, Vivarium 41/2: 175–248.
—— (2003b) ‘Abélard et les grammairiens: sur la définition du verbe et la notion
d’inhérence’, in P. Lardet (ed.), La tradition vive. Mélanges d’histoire des textes en
l’honneur de Louis Holtz. Brepols, Paris-Turnhout: 143–59.
—— (2003c) ‘Priscien, Boèce, les Glosulae in Priscianum, Abélard: les enjeux des
discussions autour de la notion de consignification’, Histoire Épistémologie Langage
25/11: 55–84.
RUSSELL, B. (1905) ‘On denoting’, Mind 14: 479–93.
SANDMANN, M. (1954) Subject and Predicate. A Contribution to the Theory of Syntax.
University Press, Edinburgh.
SANFORD, A. J., E. J. DAWYDIAK, and L. M. MOXEY (2007) ‘A unified account of quantifier
perspective effects in discourse’, Discourse Processes 44/1: 1–32.
SCHILLER, F. C. S. (1912) Formal Logic. A Scientific and Social Problem. Macmillan, London.
SCHLENKER, PH. (2007) ‘Anti-dynamics: presupposition projection without dynamic
semantics’, Journal of Logic, Language and Information 16: 325–56.
SEUREN, P. A. M. (1972) ‘Taaluniversalia in de transformationele grammatika’, Leuvense
Bijdragen 61/4: 311–70. (Partially republished in English as Chapter 12 in Seuren 2001.)
—— (1973) ‘The comparative’, in F. Kiefer and N. Ruwet (eds), Generative Grammar in
Europe. Reidel, Dordrecht: 528–64.
—— (1974) ‘Negative’s travels’, in P. A. M. Seuren (ed.) Semantic Syntax, Oxford Readings
in Philosophy. Oxford University Press, Oxford: 183–208. (Reprinted as Chapter 8 in
Seuren 2001.)
—— (1975) Tussen Taal en Denken. Een bijdrage tot de empirische funderingen van de
semantiek. Oosthoek, Scheltema en Holkema, Utrecht.
—— (1977) ‘Forme logique et forme sémantique: un argument contre M. Geach’, Logique
et Analyse 79/20: 338–47. (Reprinted in English as Chapter 13 in Seuren 2001.)
—— (1980) ‘Dreiwertige Logik und die Semantik natürlicher Sprache’, in Sprache der
Gegenwart. Schriften des Instituts für deutsche Sprache, Band L. Grammatik und Logik.
Jahrbuch 1979 des Instituts für deutsche Sprache. Schwann, Düsseldorf: 72–103.
—— (1985) Discourse Semantics. Blackwell, Oxford.
—— (1986a) ‘Formal theory and the ecology of language’, Theoretical Linguistics 13/1: 1–18.
—— (1986b) ‘Anaphora resolution’, in T. Myers, K. Brown, B. McGonigle (eds), Reasoning
and Discourse Processes. Academic Press, London: 187–207.
—— (1987) ‘Les paradoxes et le langage’, Logique et Analyse 30 (120): 365–83. (Republished
in English as Chapter 6 in Seuren 2001.)
—— (1988) ‘Presupposition and negation’, Journal of Semantics 6/3–4: 175–226. (Reprinted
as Chapter 15 in Seuren 2001.)
—— (1993) ‘Why does 2 mean “2”? Grist to the anti-Grice mill’, in E. Hajičová (ed.),
Proceedings of the Conference on Functional Description of Language, Prague Nov. 24–27,
1992. Prague: Faculty of Mathematics and Physics, Charles University: 225–35.
(Reprinted as Chapter 16 in Seuren 2001.)
 Bibliography 417

—— (1994) ‘Presupposition’, in R. E. Asher and J. M. Y. Simpson (eds), The Encyclopedia of

Language and Linguistics. Oxford: Pergamon Press: 3311–20.
—— (1995) ‘Notes on the history and the syntax of Mauritian Creole’, Linguistics 33: 531–77.
—— (1996) Semantic Syntax. Blackwell, Oxford.
—— (1998a) Western Linguistics. An Historical Introduction. Blackwell, Oxford.
—— (1998b) ‘Towards a discourse-semantic account of donkey anaphora’, in S. Botley and
T. McEnery (eds), New Approaches to Discourse Anaphora: Proceedings of the Second
Colloquium on Discourse Anaphora and Anaphor Resolution (DAARC2). University
Centre for Computer Corpus Research on Language, Lancaster University. Technical
Papers, Vol. 11, Special Issue: 212–20. (Reprinted as Chapter 17 in Seuren 2001.)
—— (2000) ‘Presupposition, negation and trivalence’, Journal of Linguistics 36/2: 261–97.
—— (2001) A View of Language. Oxford University Press, Oxford.
—— (2002) ‘The Logic of Thinking’, Koninklijke Nederlandse Akademie van
Wetenschappen, Mededelingen van de Afdeling Letterkunde, Nieuwe Reeks 65/9.
—— (2004) Chomsky’s Minimalism. Oxford University Press, New York.
—— (2005) ‘Eubulides as a 20th-century semanticist’, Language Sciences 27/1: 75–95.
—— V. CAPRETTA, and H. GEUVERS (2001) ‘The logic and mathematics of occasion
sentences’, Linguistics and Philosophy 24/2: 531–95.
SIGWART, CHR. (1895) Logic. Vol. I. The Judgment, Concept, and Inference (authorized
translation from the German original by Helen Dendy). Swan Sonnenschein and Co.,
London/Macmillan and Co., New York.
SORABJI, R. (2004) The Philosophy of the Commentators, 200–600 AD. A Sourcebook. Vol. 3:
Logic and Metaphysics. Duckworth, London.
—— (ed.) (1990) Aristotle Transformed. The Ancient Commentators and their Influence.
Duckworth, London.
STALNAKER, R. C. (1978) ‘Assertion’, in P. Cole (ed.), Pragmatics. (Syntax and Semantics 9.)
Academic Press, New York-San Francisco-London: 315–32.
STANLEY, J. and Z. SZABÓ (2000) ‘On quantifier domain restriction’, Mind and Language 15/2:
219–61.
STAROBINSKI, J. (1979) Words upon Words. The Anagrams of Ferdinand de Saussure. Transl. by
Olivia Emmet. Yale University Press, New Haven-London.
STEINTHAL, H. (1855) Grammatik, Logik und Psychologie, ihre Prinzipien und ihre Verhältnisse
zueinander. Dümmler, Berlin.
—— (1860) Charakteristik der hauptsächlichsten Typen des Sprachbaues. (Neubearbeitung
von Dr. Franz Misteli.) Dümmler, Berlin.
STOUT, G. F. (18961; 19093) Analytical Psychology. 2 vols. Swan Sonnenschein and Co.,
London/Macmillan and Co., New York.
STRAWSON, P. F. (1950) ‘On referring’, Mind 59: 320–44.
—— (1952) Introduction to Logical Theory. Methuen, London.
—— (1954) ‘A reply to Mr Sellars’, Philosophical Review 63/2: 216–31.
—— (1964) ‘Identifying reference and truth-values’, Theoria 30/2: 96–118.
STRAWSON, P. F. (1974) Subject and Predicate in Logic and Grammar. Methuen, London.
418 Bibliography

SULLIVAN, M. W. (1967) Apuleian Logic. The Nature, Sources, and Influence of Apuleius’s Peri
Hermeneias. North-Holland, Amsterdam.
SVEDELIUS, C. (1897) L’analyse du langage appliquée à la langue française. Almqvist and
Wiksell, Uppsala.
TASMOWSKI-DE RYCK, L. and S. P. VERLUYTEN (1982) ‘Linguistic control of pronouns’, Journal
of Semantics 1/4: 323–46.
THOMPSON, M. (1953) ‘On Aristotle’s Square of Opposition’, The Philosophical Review 62/2:
251–65.
TRAUGOTT, E. C., A. TER MEULEN, J. SNITZER REILLY, and CH. A. FERGUSON (eds) (1986) On
Conditionals. Cambridge University Press, Cambridge.
VAN DALEN, D. (1999) Mystic, Geometer, and Intuitionist: The Life of L. E. J. Brouwer. Vol. 1,
The Dawning Revolution. Oxford University Press, Oxford.
VAN DER AUWERA, J. (1998) ‘Pragmatics in the last quarter-century: the case of conditional
perfection’, Journal of Pragmatics 27/3: 261–74.
VAN DER SANDT, R. A. (1992) ‘Presupposition projection as anaphora resolution’, Journal of
Semantics 9/4: 333–77.
VAN FRAASSEN, B. (1971) Formal Semantics and Logic. Macmillan, New York-London.
VAN KUPPEVELT, J. C. J. (1991) ‘Topic en Comment. Expliciete en Impliciete Vraagstelling in
Discourse’, Ph.D. thesis, Radboud University, Nijmegen.
VAN OIRSOUW, R. R. (1987) The Syntax of Coordination. Croom Helm, London.
VELTMAN, F. (1985) ‘Logics for Conditionals’, Ph.D. thesis, University of Amsterdam.
VON DER GABELENTZ, H. G. C. (1869) ‘Ideen zu einer vergleichenden Syntax. Wort- und
Satzstellung’, Zeitschrift für Völkerpsychologie und Sprachwissenschaft 6: 376–84.
—— (18911; 19012) Die Sprachwissenschaft. Ihre Aufgaben, Methoden und bisherigen
Ergebnisse. Leipzig: Tauchnitz.
WEGENER, C. (2008) A Grammar of Savosavo, a Papuan Language of the Solomon Islands.
Ph.D. thesis, Radboud University, Nijmegen. (¼ MPI Series in Psycholinguistics, 51).
Max Planck Institute for Psycholinguistics, P.O. Box 310, 6500 AH Nijmegen, The
Netherlands.
WEGENER, PH. (1885) Untersuchungen über die Grundfragen des Sprachlebens. Niemeyer,
Halle. [Reprinted 1991, with an introduction by Clemens Knobloch. Benjamins,
Amsterdam-Philadelphia.]
WEIDEMANN, H. (1994) Aristoteles’ Peri Hermeneias. Uebersetzt und erläutert von Herman
Weidemann. Akademie-Verlag, Berlin.
WEIJTERS, A. (1985) ‘Presuppositional propositional calculi’, Appendix to Seuren (1985:
483–525).
—— (1989) ‘Denotation in Discourse: Analysis and Algorithm’, Ph.D. thesis, Radboud
University, Nijmegen.
WHITEHEAD, A. N. and B. RUSSELL (1910–1913) Principia Mathematica. 3 vols. Cambridge
University Press, Cambridge.
WIERZBICKA, A. (1996) Semantics. Primes and Universals. Oxford University Press, Oxford.
WILSON, D. (1975) Presuppositions and Non-Truth-Conditional Semantics. Academic Press,
London-New York-San Francisco.
 Bibliography 419

WUNDT, W. (19001; 19224) Völkerpsychologie. Eine Untersuchung der Entwicklungsgesetze von

Sprache, Mythus und Sitte. Volume 2, Die Sprache, Part 2. Kröner, Leipzig.
ZALTA, E. N. (1988) Intensional Logic and the Metaphysics of Intentionality. A Bradford Book.
MIT Press, Cambridge, MA.
ZWICKY, A. (1973) ‘Linguistics as chemistry: the substance theory of semantic primes’, in
S. Anderson and P. Kiparsky (eds), A Festschrift for Morris Halle. Holt, New York: 467–85.
This page intentionally left blank
 Index
AAPC see predicate calculus, Aristotelian- appellation 198(n)
Abelardian Apuleius, Lucius 152, 155–7, 160
Abelard, P. 15, 88, 122, 136, 141, 150, 160, 162, Aquinas, Thomas 114, 173
172–9, 315 Arabic 171, 241(n)
aboutness 317 Aristotle 6, 10, 15, 17–21, 59–61, 68, 87(n),
ABPC see Square of Opposition 88, 91(n), 93–4, 102–4, 136–7, 141,
Abraham, W. 387(n) 149–60, 165, 167, 171, 173, 175, 194,
accent 259, 312–14, 322, 378–81,
contrastive 338 383, 385
sentence-nuclear 389–90 arithmetic 68–70, 78
accommodation 214, 230, 250, 312, 342 (see Ars Meliduna 314, 322
post hoc suppletion) Asbach-Schnitker, B. 387(n)
actual being 234–6, 240, 285, 399 asinus symbol 297, 310
addition 12, 86 assertion 197–8
address 202, 229–53, 285, 293 Astralabius 172
accessibility of s 207–8, 220 Austin, J. 217
closure 230–3, 236, 243, 248, 288, 306, 310 auxiliary system 206(n)
adióristos 149, 164–6
adjectives, gradable 223–4 Baker, C. 340
Ajdukiewicz, K. 235(n) Bantu 114
Al Huárizmi 68 Barwise, J. 39, 54–5, 287
alchemy 172(n) Beaver, D. 286, 373–5, 377
Alford, M. 171(n) Bellert, I. 387(n)
algebra, Boolean 5, 43–6, 73–4, 79, 128, 183–4 Bernard, St. of Clairvaux 172–3
algorithm 68 Blakemore, D. 255(n), 386, 387(n)
Ammonius 152, 155–7, 170 Blanché, R. 90(n)
anaphora 284, 286, 288–94, 320, 372–7 BNPC see predicate calculus, basic-natural
bound 293 functionality of 190–2
complement 293(n), 387(n) Bochenski, I. 104(n)
donkey 267, 273, 284, 286–8, 291, 293–304, Boër, S. 326–7, 334, 337
307–10 Boethius, Anicius Manlius Severinus 30, 36,
epithet see epithet pronoun 88, 155–8, 160, 170, 174
external 196, 254, 262, 284, 289–93, Bolinger, D. 391
301, 376 Boole, G. 44
internal 289–93 Borkowski, L. 155(n)
primary 232–3, 288, 291–4, 301, 304, Brentano, F. 385
307–10, 379 Brouwer, L. 172(n)
unbound 293 Brown, L 39(n)
resolution 289–93, 296, 372–7, 375–6 Buckner, E. 159, 161(n)
anchor hooks 196 Buridan, J. 161–2
anchoring 8–9, 19–21, 23, 195 et passim Burleigh, W. 162, 300–2, 315
Andrade Martins, S. 39(n) Busse, A. 153
antecedent strenghtening 280 Butterworth, B. 68–9
422 Index

C-command 292 copula verb 175(n)

Callimachus 270 counterfactual 274–5, 354
Carnap, R. 40 Creole languages 288, 394
Carroll, L. 160–1, 167
Carston, R. 255(n), 386, 387(n) Dâw 39(n)
category mistake 330 De Morgan, A. 65, 75, 102(n), 104, 167
Cavaliere, F. xiii, 104 De Morgan’s Laws 33, 59–66, 112, 122, 125,
ceteris-paribus condition 273, 280–2 129–30, 358
Clanchy, M. 173(n) De Rijk, L. 61, 152, 155, 174–5, 314–5
Clark, H. 23, 195, 312 Declerck, R. 277(n)
cleft-structure 100, 337–8, 395, 397–8, 400, definite terms 23
406 Dehaene, S. 68(n), 69
cogito-argument 363 denotation 207 et passim
Cohen, L. 255(n) procedure 237
comment 390 Diewald, G. 387(n)
commitment domain 198–200 discourse 1, 58–9
complement, restricted 81 et passim domain 196–215
compositionality 223–6, 233, 236 Representation theory (DRT) 285–6, 372
comprehension vs interpretation 194–5, 225–6 disjunction 50–2, 108–14, 199, 264–71, 279,
Comrie, B. 274(n), 282(n) 307–8, 347, 387
Conan Doyle, A. 229 basic-natural 108–14, 267
conditional perfection 277–8 exclusive 108–14, 264–70, 279
conditionals 270–82, 308–9 incrementation of 201, 212, 267, 350
incrementation of 201, 272–3, 309, 353 logic of 355–60
phrasal 271 metalinguistic 265, 268
speech-act 271 phrasal 265
conjunction 50–2, 108–14, 199, 216, 254–9 speech-act 264–5, 268
basic-natural 108–9 standard 43–5, 48–9
incrementation of 201, 216, 218, 254–9, reduction 266
348–50 Dodgson, C. 160 (see also Carroll, L.)
logic of 355–60 Donnellan, K. 397(n)
phrasal 257 Dowty, D. 226, 319(n)
speech-act 256 Dryer, M. 39(n)
standard 43–5, 48–9 dual 241(n)
analysis 320 duality 31, 33–7, 124, 132
reduction 213, 257 Dufatanye, A.-A. 114
consistency, transdominial 215–7, 219 Dummett, M. 197, 326, 355, 373
context-sensitivity 194–227 Dutch 119, 206(n), 274(n), 393, 400, 403–4
contradictoriness 27–30, 49, 86, 98, 129, 151, dynamic filtering 23
175–6 et passim
contraposition 29–30, 282 Ebbesen, S. 156
contrariety 27–30, 49, 86, 129, 151, 177 Eco, U. 171
et passim ecologism 3, 199
conversation analysis 386 Elextra paradox 312–13
Conversions 33, 37, 59–66, 94, 98, 124, 132, English 59, 206(n), 228, 274(n), 328, 335,
135–8, 143, 145, 150, 153, 170, 173–4, 176, 387(n), 393–4, 401, 403–4
179, 188–9 BBC radio 257(n)
Cooper, R. 39, 54–5 Cockney 260–1
Copi, I. 5 New York Black 260–1
 Index 423

entailment 7–13, 27, 86, 331–2 et passim Gazdar, G. 255(n), 331(n), 351, 362
C-entailment 330–4 Geach, P. 294–6, 300–4, 324–5, 376
logical 10–13 Geis, M. 277
natural 10–12 gender, grammatical 404(n)
P- entailment 330–4 Genzen, G. 53(n)
epithet German 206(n), 228, 274(n), 386, 387(n),
pronoun 290–1, 376–7 392–3, 400, 403–4
substitution test 290–2, 296–300 Geurts, B. 286, 373–5, 377
Eubulides of Miletus 312–14, 316 Gilso, É, 173(n)
Euclid 68 Ginsburg, H. 68(n)
Euripides 217 golden ratio 171
Evans, J. 43 grammar 16–17, 19
existence see actual being categorial 323
existential import 14–15, 124–7, 150, 158–62, Greek 25, 71(n), 118, 170–1, 172(n), 241(n),
165–70, 173, 235–6 394(n), 401
origin of 175(n), 366–7 Green Pedersen, N. 315–16
undue 14, 93, 122–5, 132–3, 136, 145, Greenberg, J. 282(n)
149–50, 154, 158–9, 162, 173, 190, Grice, H. 71(n), 99–100, 341
363–4, 367, 370 Groenendijk, J. 255, 286, 320–1
exjunction 84, 112 Gussenhoven, C. 380, 389, 391
exponibles 314–16
extension of predicates 322, 329 Hall Partee, B. 396
extreme values 70, 75, 183–6, 193 Hamblin, C. 198(n), 217
Hamilton, W. 102(n), 103–7, 118, 166, 169, 192
falsification 102 Harris, M. 274(n)
falsity Heim, I. 321, 348–9(n)
minimal 15, 22, 32, 97(n), 125, 354 et passim Héloı̈se 172
radical 15, 22, 97(n), 125, 305–6, 326, 354 Hoeksema, J. 116
et passim hole 210–11, 343, 345, 347, 376
Fauconnier, G. 287 homunculus 24
feeder 386–7 Horn, L. 85(n), 99, 102(n), 114–5, 117–9,
Fibonacci numbers 171 158–61, 168, 170, 173, 211, 260, 277,
filter 210–11, 343 315(n), 335(n), 360–2, 398(n), 401,
Finno-Ugric languages 39(n) 403–4, 406
Fisher, K. 387(n) Horned Man, paradox of the 312–13
Fodor, J. 198, 226 Husserl, E. 378
formalism 3, 199 hypokeı́menon 379, 382–3, 385
formalization 3
Fowler, T. 5, 166 identification 76, 85
free variables see open parameters identity 49, 73, 76–7, 82
Frege, G. 15, 41, 44–5, 123, 202, 236, 269, implication 199, 270–282, 350–1
314, 316, 321–6, 334, 354, 372–3, material 27, 270, 275, 279
395, 406 paradoxes of 275, 277–82
French 59, 394, 400, 403–4 implicature 99–100, 255(n)
Fulbert, canon of Notre Dame 172 incrementation procedure 58, 109 et passim
functional sentence perspective 385 independence
logical 50, 72–3, 86, 88, 127, 132
Gapping 257–9 set-theoretic 50, 78, 83–4
Garrod, S. 214 Indo-European 394(n)
424 Index

induction 102, 342 Lenzen, W. 104(n)

inferential bridging 239, 246 Levelt, P. xiii
informativity 184–5, 217–8 Levinson, S. 71(n), 114–5, 117
information structure 380, 386 Lewis, D. 312
information value 116 Lewis, G. 274
intension of predicates 322 lexicalization 117–20, 142(n)
intent 195 liability condition 216
intersection 72–3 Liar paradox 312–13
M-partial 72, 78, 81–3, 86, 88, 90, 109, 231 Lipps, T. 382–3, 385–6
intonation 265 litotes 119–120
intuitionism 172(n) Llull, R. 171
iota operator 324 logic passim
Isard, S. 285, 287 basic-natural propositional 108–14
Islam, 171, 173 Dynamic Predicate 286
island constraint 389(n) fuzzy 22
Italian 401, 403–4 gapped bivalent 324–5
intervalent (¼fuzzy) 92
Janssen, T. 224 mathematical 3
Jasmin, K. xiii modal 10–11, 18
Jaspers, D. xiii, 37(n), 114, 118(n) natural 2, 3, 11
Jespersen, O. 99, 104, 118–9, 192, 390 nonmonotonic 13, 282
Johanson, C. 152, 156 psychological 6
Johnson-Laird, P. 270–1(n), 277(n), 281(n) presuppositional 1, 354–72
referential independence of 13–16
Kalepky, T. 384 specific 13–14
Kamp, H. 285–6, 288, 320–1, 372 trivalent 18
Karttunen, L. 210, 343, 347 of causality 121
Kenya-Rwanda 114 logical constant see operator
keying 7–8, 15, 19–21, 23, 46 et passim logical form 16–17
King, P. 163, 166(n) logical intuitions 3, 67, 70, 101, 144–6, 158
Kiparsky, C. 330 logical power 127–38, 193
Kiparsky, P. 330 Londey, D 152, 156
Kirundi 114 Lull, R. 171
Kleene, S. 359, 371 ukasiewicz, J. 10(n), 18, 155(n)
Klima, G. 159, 170 Lycan, W. 326–7, 334, 337
Klooster, W. 403
Kneale, W. & M. 104(n), 150, 152–5, 160–1, Malay 288
167, 173–5, 270, 312–13 Marty, A. 385
Kratzer, A. 43 material implication 10
Kripke, S. 285, 397(n) Mathesius, V. 385
matrix
L-proposition 4, 7–13, 16, 27, 175(n), 183–4, lexical 40
199–200 et passim greed 40
Labov, W. 260 term 55–8, 63–4
Lakoff, R. 387(n) Mauritian Creole 394
Lang, E. 387(n) Mauritius 182
Langendoen, D. 209 maxim, Gricean 99, 112, 135–6
Latin 118–9, 170, 172(n), 176(n), 290–1, 394 McCawley, J. 1, 39, 52, 282
Leibniz, G. 214 McLeod, E. 173(n)
 Index 425

meaning postulate 40–3 open parameters in lexical meaning 196,

Megarian school of philosophy 312–13 222–8, 284
Meinong, A. 371 operator 10–13, 27, 37–9, 42 et passim
Meiser, C. 158 metalinguistic 91–3
metaphor 329–30 Over, D. 43
Meyer-Lübke, W. 381–2 Oxford English Dictionary 120
modality
agentive 203, 206 Papuan languages 274
epistemic 203–5, 221, 352–3 paradoxes, Eubulidean 312–14
metaphysical 203 Parsons, T. 125, 154, 158, 160–1, 163, 165–6,
semantics of 203–5 167–70, 173
model theory 22 Partee, B. 202(n)
module 43 Peano, G. 53(n), 105
number 68 Pederson, E. 114
Modulo-Key Condition 8, 13, 43 Peirce, Ch. 167
Montague, R. 235(n), 319(n), 396 Perry, J. 287
Moody, A. 10, 158–60, 162, 166, 169 Peter of Spain 315
Moore paradox 217 Peters. S. 250
Mostovski, A. 39, 54 Pica, P. 68
Moxey, L. 292(n), 387(n) Pickering, M. 214
Mullally, J. 315 Plato 10
Mundurukú 68 Platonism 46
mysticism 170–2, 179(n) Plotinus
plug 210–11, 343
Neale, S. 304(n), 397(n) plural power set 240–1, 250–1
negation 10, 27–8, 30–7, 43–4, 48, 50–2, plurality 240–52
84–5, 108, 199, 207, 211. 259–64, 317, polarity item
325, 333–4, 341 et passim negative 338–40, 392, 404–5
canonical position of 335, 361 poasitive 339–40
external 30, 93, 175–6, 260–1, 367–8 polysemy 226
incrementation of 201, 260, 350 Ponce de Léon 319
internal 30–7, 59–66, 93, 129, 165, 175, post hoc suppletion 214–5, 230, 312, 342–3,
367–8 349(n)
lexically incorporated 259 pragmatics 3, 99, 112, 116, 135–6, 158, 170, 226,
metalinguistic 211–2, 259–60, 263, 361–2 254, 263, 270, 277, 283, 289, 291, 327,
minimal 211, 273(n), 308, 352, 355–6, 367 331, 335, 341, 380, 386, 403–5, 408
et passim Prague School 380, 385
morphologically incorporated 335 precondition 23, 312, 328–30, 341
radical 15, 91(n), 211–2, 308, 352, 355–6, 367 predicate 379
et passim antifactive 354
Copying 261 causative 346
test 333 factive 201, 210, 220, 346–7, 352
NEG-Raising 403–6 modal 202–7, 332, 347–8, 352–3
Neo-Platonism 155 lambda 245–6
Newton, I. 171(n) quantification of the 104, 169–70
Nuchelmans, G. 314 value-assigning be 77, 205(n)
weak factive 330–1(n), 345
Object Incorporation 55(n), 142(n) of possession 24, 224–5
Ockham, W. of 161–3, 165–6, 169, 315, 342, 372 of propositional attitude 202–3
426 Index

predicate calculus 10–11 et passim quantifiers 12, 16, 72 et passim

Aristotelian 122, 153–4 distributive 58, 138–43
Aristotelian-Abelardian 89, 94, 97, 128, floating 251(n)
136–8, 145, 149–55, 162 generalized 39, 54–8, 89, 139
Aristotelian-Boethian 99, 103, 108, 122, 145, nonstandard 57, 143
147–9 (see also Square of Opposition) Russellian 52, 54, 58, 139
basic-natural 88–103, 128, 145–6, 368–70 Quine, W. 20–21, 53(n), 60–1, 159, 205(n),
Hamiltonian 103–8, 144–5 285, 294–5, 298–9, 300. 371–2,
prejacent 314–15 375, 395
presupposition 1, 18, 23, 97(n), 181, 196, 284,
286, 288, 311–77, et passim recursivity 78–9
categorial 320, 330 reflexivity 290–2, 394–5
E(mbedded)- 342–54 Reed, S. 277(n)
existential 320, 325, 330 reference 207, 230, 232–3, 250, 288 et passim
factive 201–2, 320, 330, 336–7, 346 opaque 253
operational criteria of 331–4 transparent 253
origin of 327–31 fixing algorithm, 304–7, 309
cancelling 334–42, 362 Reinhart, T. 289(n), 292(n)
denial 260 Rescher, N. 18
failure 325 restrictor term 55–8
inducer 327–8 Reyle, U. 286, 372
projection 209–12, 221, 253, 273, 332–4, rheme 385
341–54 Right-Node Raising 257–9
Principle Romance languages 274(n)
Bivalence 17–19, 22, 126, 317 Rosch, E. 227
Isomorphy 36 Rosier-Catach, I. 175(n)
Modulo*- 33–6, 64 rotation of variables 53
OSTA- 207, 303–4, 306, 309 Russell, B. 6, 38, 53(n), 60, 88–9, 105, 123,
of cognitive independence 22–5 265(n), 285, 294, 297, 300, 304(n), 313,
of maximal unity (PMU) 207–15, 253, 316–20, 323–6, 334, 371, 373, 375
296, 342, 374
of minimal change 214–5 Sandmann, M. 384(n)
of natural set theory 75–9 Sanford, A. 292(n), 387(n)
of the Excluded Middle 18 satisfaction conditions 7, 23, 41, 50–9, 222,
of the Excluded Third 18, 312–14, 320, 224–6, 328
324 Saussure, F. de 172(n)
projection problem 209 (see presupposition Savin, H. 209
projection) scalarity 99–100
proposition 378–80, 384–5 Scharten, R. 396–7
flat 390–2 Schiller, F. 166
modulated 390–2 Schlenker, P. 348–9(n)
propositional attitude 284–5, 287–8 (see also semantic analysis 16
predicate of) Semantic Syntax 328, 400
propositional calculus 10–11, 48–52, 108–14 semantics 16–17, 22 et passim
et passim generative 328
prototypicality 25, 70–1, 227–8 model-theoretic 3, 199, 203, 222, 226, 240,
250, 286, 297, 365
quantification passim lexical 40–3, 222–6
monadic vs multiple 88–9, 106 situational 287
 Index 427

sentence subcontrariety 27, 29–30, 49, 86–7, 129, 152–3,

eternal 20–1, 61, 194 et passim 177 et passim
occasion 20–1, 61–2, 194, 294 et passim subdomain 199–215
sequentiality 333–4, 345 alternative s 252
set, natural 75–6, 184–6, 188, 190, 192 epistemic 203–6
et passim extensional 199, 201–3, 268–9, 274
set theory 2, 45, 47 instructional 252, 348–51
constructed 67, 69 intensional 199, 201–6, 343–4
natural 47, 67–121 lexical 343–8
Seuren, P. 1, 3, 17, 24, 40, 52, 55(n), 111(n), recipient 221, 342, 345–6, 353–4
118(n), 136(n), 172(n), 173(n), 199(n), subordinate 252–4
206(n), 211, 219, 257(n), 260, 269(n), subsidiary 221, 344–6
285, 288, 289(n), 295, 312–14, 328, hierarchies of s 219–22
339–40, 355, 359, 373, 375, 386, 388, unification 212–3
397, 399–400, 404 Subject Incorporation 55(n)
shared knowledge thesis 195–7, 284, Subprinciple
312, 371 of Complete Valuation 18–19, 22
Sigwart, C. 37(n), 166–7 of Binarity 18–19, 22
simplification 86 substitution salva veritate (SSV) 199, 254,
social reality 229 268–9, 406–8
Sophists 227 subtraction 72–3, 77, 85
Sorabji, R. 155 Sullivan, M. 152, 157
Sorites paradox 312–13 surface form 16
speech act 198–9, 216, 256, 264–5, 268, 271 Svedelius, C. 384(n)
square Swedish 290–1, 394
Boethian see Square of Opposition syllogism 104(n)
logical 30 syllogistic 171
natural logical 34, 148 syncategoremata 10
notation 129, 156–8 syntax 328, 379–81, 385, 388–90, 400–3, 406
Square of Opposition 12, 30, 34–7, 87, Szabó, Z. 250
89, 94–5, 99, 103, 108, 124–7, 132–5,
147–8, 156–60, 165–6, 169, 176, 302, Tamil 114
363–8, 370–2 (see also predicate Tasmowski-de Rijck, L. 292(n)
calculus, Aristotelian-Boethian) theme 385
functionality of 181–90 Theodoric the Great 156
Stalnaker, R. 285 theology 172–3
Stanley, J. 250 Theory of Descriptions 317, 323–4, 326
Starobinski, J. 172(n) Thompson. M. 154, 159–60, 166, 169
Steinthal, H. 381 topic 390
Stoa 10(n), 113, 147, 155(n), 175, 176(n), 270. topic-comment modulation 100, 196,
312 267, 273, 284, 288, 341, 378–408
Stojanovic, I. 120(n), 303(n), 329(n) topicalization 63
Stokhof, M. 255, 286, 320–1 transparency, transdominial 207–9, 220
Stout, G. 287, 382–3, 385–6 triangle, natural logical 28–30, 35–6
Strawson, P. 14–15, 21, 160–1, 313, 316, 325–6, True Alternatives Condition 266
333–4, 342, 372 truth table 32, 43
structuralism 385 Turkish 206(n), 274, 282(n)
subaltern 94, 100, 103, 123, 125–6, 130, 136, 138, type–token distinction 4, 8, 10, 17–20, 23,
147, 152, 1889 194–7
428 Index

union viewpoint 226–7

basic-natural 77, 86, 272 virtual being 234, 236, 240, 262, 285–6, 375, 399
full 49–50, 73, 82–4 virtual object attraction (VOA) 236
uniqueness clause 319–20 Von der Gabelenz, G. 381–4
update condition 227, 328–30
Walman 39(n)
vagueness 18, 23, 227 Wegener, C. 274
validity 27 Wegener, P. 382
valuation 46 Weidemann, H. 149(n), 151(n)
valuation space 31, 39, 46, 58, 84, 177, 187–92, Weijters, A. 289(n), 359
275–6, 280–1 et passim Westerståhl, D. 250
modelling 46, 133, 140, 255 et passim Whitehead, A. 53(n), 324
Van Dalen, D. 172(n) Wierzbicka, A. 71(n)
Van der Auwera, J. 277(n) Wilkes-Gibbs, D. 23
Van der Sandt, R. 286, 372–7 Wilson, D. 326–7, 334, 337, 373
Van Fraassen, B. 31 Wittgenstein, L. 172(n)
Van Kuppevelt, J. 387 Wundt, W. 383–5
Van Oirsouw 257(n)
Veltman, F. 275(n), 279, 281–2 Zalta, E. 319
Verluyten, S. 292(n) Zwicky, A. 114, 277