PHILOSOPHY OF

LOGICS

SUSAN HAACK
Reader in Phílosophy, University of Warwick

CAMBRIDGE UNIVERSITY PRESS

CAMBRIDGE
LONDON • NEW YORK • MELBOURNE
Published by the Syndics of the Cambridge University Press
The Pitt Building, Trurnpington Street, Cambridge CB2 IRP
Bentley House, 200 Euston Road, London NWI 2DB
32 Ea1t 57th Srreet, New York, NY 10022, USA
:zQ6 Beaconsfield Parade, Middle Park, Melboume 3206, Australia

� Cambridge University Press 1978

First published r 978

Printed in Great Britain at the

Universíry Press, Cambridge

Lilrrary of Congrm Cataloguing 111 Publication Data

Haak, Susan.
Philosophy of Iogics.
Bibliography: p.
Includes index.
J. Logic. I Title.

BCr.H15 16o 77-17071

ISBN o 5:zr :11988 '1- hard c,:rvers
JSBN o 521 :z93:z9 '1- paperback
for RJH
 ACKNOWLEDGMENTS

This book is based, in large part, on lectures in the philosophy of logic

given at the University of Warwick since 1971. Thanks are dueto ali
the colleagues and friends with whom I have discussed the issues
raised here; especially to Nuel Belnap, Robín Haack, Peter Hemsworth,
Paul Gochet, Dorothy Grover, Graham Priest and Timothy Smiley
for detailed comments on draft material. I am indebted, also, to my
students, who have taught me a great deal; and to Jeremy Mynott, for
editorial advice and support.
 CONTENTS

Preface xi n and abbreviations xv

1 'Philosophy of logics' 1

Logic, philosophy of logic, metalogic 1

The soope of logic 3

z Validity u
Assessing arguments 11

Deductive validíty: with some brief comments about

inductíve strength 13
Validity in a system; Extra-systematic validity;
Logica utens ami logica docens ; Inductiue strength
Formal logical systems: the 'L' in 'valid-in-L" 17
Notational oariants; Altematioe primitive constants;
Axiomatic and natural deduction formulations; Altematiue
axioms and/or rules
Validity and logical fonn 22

3 Sentence connectives 28

Formal considerations 28
Adequate sets oj connectiues: Junctional completeness ;
Characteristic matrices: decidability; Many-oalued logic
The meanings of the connectives 30
Formal languages and informal readings; 'tonk' ,· Aims of
Jormalisatüm; '&' and 'and', 'v ' and 'or', etc.
vili Contente
4 Quantifiers 39
The quantifiers and their interpretation 39
Metaphysical interlude: Quine on quantification and
ontology 43
The criterion of ontological commitment; Substitutional
quantification ami ontology
The choice of interpretation 50
Substitutional quantifiers ami truth ; Too few names?;
Tense; Modality; Second-order quantification

5 Singular terms 56
Singular terms and their interpretation 56
Names 57
Names as purely denotative; Names assimilated to
descriptions
Descríptions
Non-denoting names: fiction

6 Sentences, statements, propositions 74

Three approaches 74
Sentence, statement, proposition 75
'Sentence letters', 'propositional variables', or what? 78
Truth-bearers 79
Truth-bearers ami theory of truth
The problem reformulated 83
Validity again

7 Theories of truth 86
A summary sketch 86
Dejinitions oersus criteria of truth
Correspondence theories 91
Coherence theories 94
Pragmatic theories 97
The semantic theory 99
Adequacy conditions on definitions of truth ; Material
adequacy; Formal correctness ; Tarski's dejinition of truth;
Informal account; Formal account
Commentary on the semantic theory IIO
Tarski's own estimate; Popper's claims on behalf of
Tarski' 1 theory; Da'Vidson' s use of Tarski' 1 theory
 Contente ix

The redundancy theory 127

Ramsey ; Truth-bearers ; The object language/metalanguage
distinction; The quantifiers : '(p) (if he asserts that p, p)';
The 'prosentential theory of truth'

8 Paradoxes 135
The Liar and related paradoxes 135
"Set-theoretical' versus 'semantic' paradoxes?
'Solutions' to the paradoxes 138
Requirements on a solution; Russell' s solution : the theory
of types, the vicious circle principie; Tarski's solution:
the hierarchy of languages; Kripke' s solution : groundedness
Paradox without 'false'; sorne remarks about the
redundancy theory of truth; and thc V.C.P. again 149

9 Logic and logics 152

'Classical' and 'non-classical' logics 152
Responses to prcssure to change the standard formalism 153
First case-study: the logic of temporal discourse 156
Sccond case-study: precisification versus 'fuzzy logic ' 162
Postscript: degrees of truth

10 Modal logic 170

Necessary truth 170
Modal systems 175
Extmsions of classical logic; Historical remarks ;
A formal sketch; Relations betuieen the modal systems
Criticisms of modal logic 178
Moda/ logic ' was conceived in sin'; Moda/ logic is not
needed; The interpretation of modal logic is [raught with
difficulties
Semantics for modal logics 187
Formal semantics - a sketch; 'Pure' and "depraoed'
semantics; Approaches to possible worlds; Approaches to
possible individuals: transtoorld identity; Quine's doubts
confirmed?
Prospects 195
Implication again: a postscript on 'relevance logic' 197
The 'paradoxes' of strict implication,· Releoance logic
 Contents

u Many-valued logic 204

Many-valued systems 204

Restrictions of classical logic: deviant logics; Historical
remarks; Formal sketcb
Phílosophical motivations 208
Future contingents; Quanium mechanics ; Semantic
paradoxes; Meaninglessness; Sense toithaut denotation;
Undecidable sentences
Many-valued logics and truth-values 213
Non-truth-functional deviant logics 215
Superoaluations ; lntuitionist logic

12 Sorne metaphysical and epistemological questions

about logic 221

Metaphysical questions 221

Monism, pluralism, instrumentalism; The issues
summarised; Comments
Epistemological questions 232
What is fallibilism?; Does fallibilism extend to logic? ,·
A digression: ' Two dogmas' again; Reoision of logic
Logic and thought

Glossary 243
Advice on reading 253
Bibliography 255
Index 267
 PREFACE

The century since the publication of Frege's Begriffsschrift has seen

a tremendous growth in the development and study of logical systems.
The variety of this growth is as impressive as íts scale. One can dis-
tinguish four major areas of development, two in formal, two in
philosophical studies: (i) the development of the standard logical
apparatus, beginning with Frege's and Russell and Whitehead's pre-
sentation of the syntax of sentence and predicate calculi, subsequently
supplied with a semantics by the work of e.g. Post, Wittgenstein,
Lówenheim and Henkin, and studied metalogically in the work of e.g.
Church and Gódel ; (ii) the development of non-standard calculi, such
as the modal logics initiated by C. I. Lewis, the many-valued logics
initiated by Lukasiewica and Post, the Intuitionist logics initiated by
Brouwer. Alongside these one has (iii) philosophical study of the
application of these systems to informal argument, of the interpreta-
tion of the sentence connectives and quantifiers, of such concepts as
truth and logical truth; and (iv) study of the aims and capacities of
formalisation, by those, such as Carnap and Quin e, who are optimistic
about the philosophical significance of formal languages, by those,
such as F. C. S. Schiller and Strawson, who are sceptical of the pre-
tensions of symbolic logic to philosophical relevance, and by those,
such as Dewey, who urge a more psychological and dynamic con-
ception of logic over the prevailing one.
I see sorne philosophical significance in the fact that these develop-
ments took place in parallel rather than in series; for it is salutary to
remember that 'non-standard' logics have developed alongside the
standard systems, and that there have always been critica, too, not only
of specíñc formal systems, but of the aspirations of fonnalisation itself.
xii Pre:face

Developments in the four areas I have distingu�shed ":ere =. of

course, independent of each other; and I see philosophical signifi-
canee, also, in the interplay between them. For example, �lthough
sorne of the key ideas of both modal and many-valued logics were
anticipated by MacColl as early as 1880, their syst�matic for�al
development carne, respectively, in 1918 after the canonical formalisa-
tion of non-modal calculi in Principia Mathematica, and in 1920 after
the provision of truth-table semantics for 2-valued logic. However,
the motivation for the development of non-standard calculi derived
not only from the mathematical appeal of the prospect of extensions
and modifications of classical logic, but also from philosophical criti-
cism: in the case of modallogics, of the claim of the material conditional
to represent implication, and, in the case of many-valued logics, of the
assumption that every proposition is either true or else false. And one
development in non-standard logic prompted another: doubts about
the success of modal logics in formalising the intuitive idea of entaíl-
ment led to the development of relevance logics, while the mathe-
matical appeal of modal systems encouraged the development, by
analogy, of epistemic, deontic and tense logics; or again, reflection on
the philosophical motivation for many-valued logics led to the idea of
supervaluations. Formal innovations, in turn, have given a new dimen-
sion to philosophical questions originally raised by standard calculi:
as, for instance, issues about the interpretation of quantifiers and their
relation to singular terms arose in a new and acute form whcn the
intelligibility of modal predica te logic was challenged; or, as old
worries about whether logic deals with sentcnces, statcments or pro-
positions turned out to be implicated in the challenge to bivalence
posed by many-valued systems. Sometimes new formal systems have
even challenged, explicitly or implicitly, and more or less radically,
accepted assumptions about the aims and aspirations of formal logics:
relevance logic, for instance, questions not only the adequacy of the
material and strict conditionals but also the classical conception of
validity; the distinctive character of Intuitionist logic derives in part
from a challenge to the 'logicist' presumption of the priority of logic
over mathematics; and fuzzy logic breaks with the traditional principle
that formalisation should correct or avoid, but not compromise with,
vagueness. And, as the last example reminds one, new formal develop-
m��ts have some�es aspired to overcome what both supporters and
entres of formal logic had tak.en to be its inherent limitations - such as
its supposed incapacity, stressed by both Schiller and Strawaon, to
 Preface xiii

deal with the pragmatic features which affect the acceptability of

informal reasoning, perhaps overcome, at least in part, by the 'formal
pragmatics' initiated by Montague.
My concem, in this book, is with the philosophy, rather than the
history, of logic. But my strategy has been devised with an eye to the
history of the interplay of formal and philosophical issues which
I have just sketched. 1 begin with a consideration of sorne problems
raised by the standard logical apparatus - the interpretation of sen-
tence connectives, sentence letters, quantifiers, variables, individual
constants, the concepts of validity, truth, logical truth; 1 turn, from
ch. 9 onwards, to a consideration of the way sorne of these problems
motiva te formal innovations, 'extended' and 'deviant' logics, and to
the ways in which these new formalisms lead, in turn, to a re-
evaluation of the philosophical issues; and I conclude, in the final
chapter, with sorne questions - and rather fewer answers - about the
metaphysical and epistemological status of logic, the relations between
formal and natural languages, and the relevance of logic to reasoning.
And two recurring themes of the book also reflect this historical
perspective. What seem to me to be the vital philosophical issues in
logic are focussed by consideration (i) of the plurality of logical
systems and (ii) of the ways in which formal calculi bear on the
assessment of informal argument. More specifically, 1 shall be urging
that, in view of the existence of alternative logics, prudence demands
a reasonably radical stance on the question of the epistemological
status of logic, and that the interpretation of formal results is a delicate
task in which judicious attention to the purposes of formalisation is
highly desirable,

I have tried to produce a book which will be useful asan introduction

to the philosophical problems which logic raises, which will be intel-
ligible to students with a grasp of elementary formal logic and sorne
acquaintance with philosophical issues, but no previous knowledge of
the philosophy of logic. But I haven't offered simple answers, or even
simple questions; for the interesting issues in philosophy of logic are
complex and difficult. 1 have tried instead to begin at the beginning,
to explain technicalities, and to illustrate highly general problems
with specific case studies. To this end I have supplied, for those new
to the subject, a glossary of possibly unfamiliar tenns used in the text,
and sorne advice on finding one's way about the literature; while, for
those anxious to go further, 1 have included a generous (but I hope
xiv Prefac«

not intimidating) bibliography. The response of my students has

encouraged me to believe that it is unnecessary, as well as undesirable,
to oversimplify. I have aspired - though the result, I fear, inevitably
falls short of the aspiration - to produce a book which may be of sorne
use to the student, and at the same time of sorne interest to the
teacher.
It is, 1 find, irritating to be unsure whether, or how, an author has
modified views he previously put forward; but, on the other hand, it
is tedious to be subjected to frequent discussions of an author's earlier
mistakes. By way of compromise, therefore, 1 indica te her e, briefly,
where, and how, 1 have modified the ideas put forward in Deviant
Logia. First: 1 have, 1 hope, made the distinction between meta-
physical and epistemological questions about the status of logic rather
clearer; and this has led me to distinguishmore carefully between the
question of monism versus pluralism, and the quesrion of revisability,
and to support a qualified pluralism rather than the monism sorne-
what confusedly assumed in Deoiant Logic. Second ; 1 have come to
appreciate that the consequences for ontology of the substitutional
interpretation of the quantifiers are somewhat less straightforward
than I used to suppose; and this has led me to a more subtle, or at any
rate more complex, account of the respective roles of quantifiers and
singular terms. I dare say, though, that I shall have missed sorne old
mistakes, besides making sorne new ones.
 NOTATION ANO ABBREVIATIONS

A,B . metavariables, ranging over sentence letters

p,q . sentence letters
negation (' it is not the case that ')
V disjunction (' or '); sometimes called 'vel'
& conjunction (' and '); 'ampersand'
material implication ('if')
material equivalence (' if and only if')
x,y ... individual variables
(3) existen tia! quantifier (' at least one ')
( ) universal quantifier (' for ali')
(1x) . definite dcscription (' the x such that ... ')
F,G . predicate letters (R, ... for polyadic predicates)
a.b . singular terms
= identity
L necessarily
M possibly
-3 strict implication
relevant implication
:::::, . entailment
1 Intuitionist negation
{ } set
{xi ... x} the set of xs which are ...
( ) sequence (ordered pair, triple ... n-tuple)
E set membership
j ... ¡ the value of ...
< less than
> greater than
xvi Notation and abbreoiations

� less than or equal to

;l!: greater than or equal to
iff if and only if
wff well-formed formula
V.C.P. vicious circle principie
� syntactic consequence
I= semantic consequence
MPP modus ponens (from A and A-+ B to infer B)
RAA reductio ad absurdum
 1

' Philosophy of logics'

There is no mathematical substitute for philosophy.
Kripke, 19']6

Logic, philosophy of logic, metalogic

The business of philosophy of logic, as I understand it, is to
investigate the philosophical problems raised by logic - as the busi-
ness of the philosophy of science is to investigare the philosophical
problems raised by science, and of the philosophy of mathematics to
investigate the philosophical problems raised by mathematics.
A central concern of logic is to discriminate valid from invalid argu-
ments; and formal logical systems, such as the familiar sentence and
predicate calculi, are intended to supply precise canons, purely formal
standards, of validity. So among the characteristically philosophical
questions raised by the enterprise of logic are these: What does it mean
to say that an argument is valid] that one statement follows from
another] that a statement is logically true? Is validity to be explained
as relative to sorne formal system? Or is there an extra-systematic idea
that formal systems aim to represent] What has being val id got to do
with being a good argument? How do formal logical systems help one
to assess informal arguments? How like 'and' is ' & ', for instan ce, and
what should one think of 'p' and 'q' as standing for? Is there one
correct formal logic? and what might 'correct' mean here? How does
one recognise a valid argument or a logical truth? Which formal
systems count as logics, and why? Certain themes recur: concern
with the scope and aims of Iogic, the relations between formal logic
and informal argument, and the relations between dífferent formal
systems.
The sphere of the philosophy of logic is related to, but distinct from,
that of metalogic. Metalogic is the study of formal properties of formal
logical systems; it would include, for instance, proofs (or disproofs) of
:¡ Philosophy of logics

their consistency, completeness or decidability. Philosophy of logic

likewise concerns ítself with questíons about formal logical systems -
but with philosophical rather than purely formal questíons. Take the
relatíons between the standard, z-valued, and many-valued sentence
calculi as an example: the philosopher will want to know in what,. if
any, sense many-valued logics are alternatives to z-valued logic;
whether one is obliged to choose between many-valued and z-valued
calculi, and if so, on what grounds; what would be the consequences
for the concept of truth íf a many-valued system were adopted, and
so forth. Metalogical results may well help one to answer questions of
this kind: for instance, it is presumably a necessary, though not a
sufficient conditíon of a many-valued logic's being a serious alterna-
tive, that it be consistent; and it may be pertinent to questions of their
relative status that (most) many-valued logics are contained in
z-valued logic (i.e. that all their theorems are theorems in z-valued
logic, but not vice-versa). A second difference is that philosophy of
logic is not wholly occupied with questions about formal logics;
informal argument, and the relations between formal system and in-
formal argument, are also within its spherc. The developrnent of
formal systcms, indeed, greatly increases thc depth and rigour of
logical studies; but the study of informal argumcnt is often an indis-
pensable preliminary to such developments, and succcss in systcrnat-
ising informal arguments a test of thcir uscfulness, It is pcrtincnt
that Frege, one of the pioneers of modero formal logic, was prompted
to develop his /Jpgriflssrhrift ( 1879) bccuusc he nccdcd a kss um higuous
and cumbersomc medium thun Gcrman in which to givc properly
rigorous arithmetical proofs.
The locution 'philosophy of logic' is, I think, much to be prcferrcd
to 'philosophical logic ', which is apt to convoy the unfortunate im-
pression that there is a peculiar, philosophical way of doing logic,
rat�er than that there are peculiarly philosophical problems about
lo�c. (I observe that, unlike "philosophical logic ', "philosophical
science ' and 'philosophical mathematics' have never gained
curr�ncy.J My examples have already shown, however, that philo-
sophical mterest attaches to the fact that there is not just one,
but a plurality of formal logics; and so 'philosophy of logics ' is, I
hope, better yet,
 'Philosophy of logics'
3

2 The scope of logic

. Among the problems of the philosophy of science are ques-
tíons about the scope of science: what domains of k.nowledge (or
'k.nowledge ') are to count as sciences? - for example, should alchemy,
or astrology, or sociology, or psychology count as bona fide sciences?
And what grounds could be given for including or excluding a given
domain of inquiry? Similarly, among the problems of the philosophy
of logic are questions about the scope of logic, and hence about the
scope of the philosophy of logic: what is a logic? which formal
systems are systems of logic? and what makes them so?I
Because I have to begin somewhere, I shall take for granted an
intuitive idea of what it is to be a formal system. But I shall indicate
what range of formal systems I have in mind when I speak of formal
logics.
It is relevant to distinguish, at the outset, between interpreted and
uninterpreted formal systems: uninterpreted, a formal system is just
a collection of marks, and cannot, therefore, be identified as a formal
logic rather than, say, a formalisation of a mathematical or physical
theory. The claim of a formal system to be a logic depends, I think,
upon its having an interpretation according to which it can be seen as
aspiring to embody canons of valid argument: I count many-valued
'Iogícs ' as logics, for example, because they have interpretations ac-
cording to which their values are 'truth-values', their variables
sentences, their operators negation, conjunction etc. (They also have
otlatr interpretations - e.g, in terrns of electrical circuits; the iso-
morphism between the logical and the electrical interpretations is
relevant to the way computers work. See Rescher 1969 p. 61 for
references.] So, in speaking of various formalisms as logics, I shall be
making an implicit appeal to their usual interpretations.
In deciding which formalisms to count as logics I have adopted, for
the present, the hospitable policy of giving the benefit of any doubt -
subsequently, though, I shall give sorne attention to arguments why
systems I have included ought to be excluded. One reason for this
policy is that it lessens the danger of dismissing a formal system as
'not really a logic', when one ought to be asking seriously whether it
is a good or useful system. I fear for instance that Quine ( 1970 ch. S),
1 The 11ígnificance of such questiona as these will, I hope, beeome
íncreuin¡ly apparent as the book proceeda. Readers who find thÍII
aection hard ¡oin¡ may prefer to retum to it at the end of the book.
4 Philosophy of logics

who excludes second-order predicate calculus because �f wh�t he

takes to be its commitment to an ontology of abstract, mt�n�1onal
objects - properties - may have succumbed to this da�ger. (Similarly,
I should distrust definitions of what it is for something to be a work
of art which encouraged evasion of questions about bad works of art.)
Anyway, as formal logics I shall include:
'traditional' logic - Aristotelian syllogistic
'classical' logic - z-valued sentence calculus
predicate calculus1
'extended' logics - modal logics
tense logics
deontic logics
epistemic logics
preference logics
imperative logics
erotetic (interrogative) logics
'deviant' logics - many-valued logics
lntuitionist logics
quantum logics
free logics
'inductive' logics
The intention is to distinguish between formal logics and systems
of, say, arithmetic or geometry, or axiomatisations of biology, physics
and so forth. The demarcation is not based on any very profound ideas
about 'the essential nature of logic' - indeed, I doubt that there is
any such 'essential nature '. But it is not wholly arbitrary; it corre-
sponds reasonably well, I hope, to what writers on philosophy of logic
usually have in mind when they speak of 'logics'; and it has, at least,
the following pragmatic rationale.
Those formal systems which are known as the 'standard' or
'classical' logic ( and taught in courses in elementary formal logic)
must surely count as logics if anything does. It then seems appro-
priate to admit also as logics those formal systems which are analogous
to these. Among such 'analogous' systems I include: extensions of
classical logic, systems, that is, that add new logical vocabulary
(' necessarily' and 'possibly' in modal logics, 'it used to be the case
1
In accordance with the 'benefit of the doubt' policy, I take thia to
include identity theory (i.e. axioma or rules for '= ') and second-order
predicare calculus (i.e, quantification binding 'F' ... etc. u well aa
'x' ... ete.) besides first-order predicare calculua.
 'Philosophy of logics' 5
that' and 'it will be the case that' in tense logics, 'ought' and 'may'
� deontic logics, 'knows' and 'believes' in epistemic logics, 'prefers'
m preference logics) along with new axioms or rules for the new
vocabulary, or which apply familiar logical operations to novel items
(imperative or interrogative sentences); deviations of classical logíc,
i.e. systems with the same vocabulary but different (usually more
restricted) axioms or rules; and inductive logics, which aim to
forrnalise a notion of support analogous to, but weaker than, logical
consequence. Their similarity to classical logic - not just formal simi-
larity, but also similarity in purpose and intended interpretation -
makes it natural to regard these systems as logics. (Alternatively,
I could have begun with traditional Aristotelian logic, of which the
modern 'classical' logic is an extension, and proceeded from there by
a similar process of analogy.)
However, the idea of a system's being sufficiently similar to the
classical logic is obviously pretty vague; and one might reasonably
wonder whether the scope of logic could be delimited in sorne less
pragmatic, and more precise, fashion.
The traditional idea that logic is concerned with the validity of
arguments as such, irrespective, that is, of their subject-matter - that
logic is, as Ryle neatly puts it, "topic-neutral ' - could be thought to
offer a principie on which to delimit the scope of logic. On this
account those systems which are applicabk to reasoning irrespectioe of
its suhject-matter would count as logics. This idea is one with which
I sympathise; 1 doubt, though, that it is really appreciably more
precise than the notion of analogy to classical logic with which I began.
What does it mean, fi.rst, to say that a formal system is 'applicable' to
reasoning on such-and-such subject-matter? Presumably, that its
principies are intended to be true of such reasoning. And now what is
one to understand by 'irrespective of its subject-matter '? lt could be
suggested that while sentence and predicate calculi are indifferent to
subject-matter, arithmetic, for example, is not topic-neutral because
it is specifically about numbers; but this raises awkward questions
about 'about' (is fi.rst-order predicate calculus 'about individuals' ?).
lt is suggested, again, that logic applies to reasoning irrcspective of its
subject-matter beca use it is concerned with the form of argumenta
rather than their ctmtent. Again, 1 think, the idea is helpful, though it
is still imprecise. How is one to distinguish between the form of an
argument and its content? Tense logic is applicable to tensed sen-
tcnces, impcrative logic to imperative sentencee, and the tense or
6 Pliiwsophy of logics

mood of a sentence could, not implausibly, be regarded as a matt�r of

its form rather than its content; but other cases are less straight-
forward - the idea of form would need refi.nement before it was clear
that a sentence's being about belief was a matter of form, but its being
about numbers a matter of content, for example.
However, the vagueness of the idea of topic-neutrality and the
related distinction between form and content isn't necessarily objec-
tionable; as I said, I am doubtful that logic has a precisely specifi.able
'essential character'. When I judged that modal logics, for example,
are enough like classical logic to be included within the scope of logic,
I was implicitly relying on the idea that the adverbs 'necessarily' and
'posaibly' are topic-neutral enough to count as 'new logical vocabu-
lary '. So the idea of topic-neutrality can certainly help to fortify one' s
intuitions about what formal systems are relevantly analogous to
classical logic. It is also significant that where to draw the line between
logics and other formal systems is more doubtful and more contro-
versia} in sorne cases than in others. For example: sorne mathematical
theories, notably set theory, are very general in application, and seem
to have strong affinities to logic; while epistemic or preference logics
seem more specifi.c as to subject-matter than the standard logical
formalisms, and not to have quite so strong a claim to inclusion.
Briefly, one gets more doubtful about the exclusion of a 'mathe-
matical' fonnalism, the more general its application, and more doubt-
ful about the inclusion of a 'logical' formalism, the less general ita
application; this suggests that topic-neutrality is vague in tke right way.
These ideas will prove important aubsequently. The dístinction
between form and content will receive sorne closer scrutiny when, in
the next chapter, I discuss the thesis that the validity of an argument
depends upon its form; and the idea that logic is characteristically
topic-neutral will be relevant when, in ch. 12, I tackle the question of
monism versus pluralism in logic, i.e. whether there is, so to speak,
one correct logic, or whether different logics might each be appro-
priate to different areas of discourse.
Sometirnes a purely formal, metalogical criterion is suggested to
demarcate logical from other formal systems. Kneale, for instance,
urges that only complete systems be allowed within the scope of logic.
The upshot of adopting such a criterion would be to restrict my
?ospitable list; sine� second-order predicate calculus is not
complete
m the usual sense, it would, by these standards, be excluded. Thia
proposal ha.a the advantage of preciaion; one ia entitled to ask,
 'Philosophy of logics' 7
though, what rationale it could have - why should completeness be
the criterion of a system's being a logic? Kneale (1956 pp. 258-9)
argues like this: the fact that a theory is in complete shows that its
basic concepta cannot be fully formalised, and this, in view of the
essentially formal character of logic, justifies excluding such theories
from its scope. So, interestingly, Kneale is proposing completeness as
the test of a system's being 'purely fonnal'; he connects the precise
idea of completeness with the vaguer notion of topic-neutrality. How-
ever, I fear that Kneale's argument may depend upon an equivocatíon
over 'formal': the sense in which the incompleteness of set theory
shows its basic concept, membership, not to be 'formal' is, simply,
that that concept cannot be completely characterised by a set of
axioms and rules which yield ali the truths which involve it essen-
tially; it is not obvious why it should be thought to follow that such
a concept is not 'formal' in the sense that it belongs to the content
rather than the form of argumenta.
My feeling is that the prospecta for a well-motivated formal
criterion are not very promising (but cf. p. 19 n. below). Another
example supports this hunch: if one placed particular weight on the
role of logic as a guide to reasoning, as a means of assessment of
informal argumenta, one might see sorne point in requiring that
logical systems be decidable, that there be a mechanical procedure for
settling whether or nota formula is a theorem. But this would restrict
the scope of logic very severely indeed, for though sentence calculus is
decidable, predicate calculus is not.
It is notable that practically every non-standard 'logic' has, at sorne
time, been subject to criticism on the ground that it isn't really a logic
at ali; which raises the suspicion that a restrictive view of the scope of
logic may disguise a conservatism that would be questioned if it were
more openly proclaimed.
Nevertheless, it may prove instructive to look at sorne argumenta
for excluding systems which, in accordance with the "benefit of the
doubt' policy, I have included. Dummett has urged (1973 pp. 285-8;
and cf. Kneale and Kneale 1962 p. 610) that epistemic 'logics' aren't
really logics, because belief and knowledge are ineradicably vague
notions. It is true that an important element in the motivation for the
formularion of logic has been to increase precision, and consequently
vagueness is normally to be avoided in the logician's choice of con-
stants, though it is more debatable whether vagueness absolutely
debars a concept from logical employment. Of course, the logician's
8 Philosophy of logics

treatment of 'not' or 'and' or 'or' or 'if' already in vol ves a not

inconsiderable tidying-up of informal negation, conjunction, etc.
(cf. ch. 3 §2); the issue is not, I think, simply wheth�r :know�' and
'believes' are vague, but whether their vagueness is ineradicable,
whether, that is, they necessarily resist regimentation. And it must be
conceded that the epistemic logics to be found in the literature ( cf.
Hintikka 1962) are somewhat disappointing, and for a reason to which
Dummett draws attention: that one is apt to find an axiom to the
effect that if s believes that p, and q follows from p, then s believes
that q. The ordinary, vague concept of belief, in other words, gets
replaced by a logical understudy, perhaps called 'rational belief',
which allows the construction of a formally interesting system, but
quite severely limits its relevance to informal arguments about belief.
Others, again, Lesniewski for instance, have suggested that many-
valued systems shouldn't really count as logics (see Rescher 1969
p. :215). It is true that sorne many-valued systems were devised and
investigated out of purely formal interest, or for purposes of com-
puter technology; but it is also true, and important, that such pioneers
as Lukasiewicz and Bochvar quite clearly regarded themselves as pre-
senting logical systems as alternatives to the classical apparatus. Still,
the claim of a formal system to be a logic depends, I allowed, on its
having a certain kind of interpretation; and a reason that might be
given for excluding many-valued systems is that they require too
radical a change in the theory of truth, or perhaps of truth-bearers, to
be sufficiently analogous to classical, z-valued logic. How much
weight one gives to this kind of argument depends, obviously, on how
radical one believes the effect of many-valuedness on the concept of
truth to be (cf. Haack 1974 ch. 3 for relevant discussion),
I gave both epistemic and many-valued systems the benefit of the
doubt about their status as logics. In each case, however, the doubts
that are raised are based on considerations the relevance of which
I concede: in the case of epistemic logics, the difficulty of eliminating
the vagueness of the new operators; in the case of many-valued logics,
the difficulty in supplying an appropriate interpretation of the new
values. The relevance of these considerations is that they throw into
question the strength of the analogy of episternic or many-valued
'lo�cs ''. in .respect of purpose and interpretation, to classical logic.
My mc.lmat10n, neverthcless, is to admit these systems as logics, at the
sarne time, of course, submitting their credentials as alternatives to
elassical logíc to stringent scrutiny. This tolerance will help to
 'Philosophy of logics' 9
counteract any conservatism inherent in the procedure of delineating
logic by analogy with the classical systems.
What difference <loes it make, one might reasonably ask, exactly
how one delimits the scope of logic? Sometimes the issue has been
thought crucial to a philosophical thesis; the case of logicism provides
an interesting example.
Logicism is the thesis (suggested by Leibniz, but worked out in
detail by Frege) that arithmetic is reducible to logic: that is, that
arithmetical statements can be expressed in purely logical terms, and
arithmetical theorems can then be derived from purely logical axioms.1
The truth or falsity of this thesis will depend, in part, on what one
allows to coun t as 'purely logical '. Specifically, it will matter whether
one is prepared to allow set theory, or second-order predicate calculus
(which for this purpose is comparable in power) as belonging to
logic; for it is certain that arithmetic cannot be reduced to first-order
logic. On the face of it, the step from first-order quantification
(as in '(x)(Fx-+Fx)') to second-order quantification (as in
'(F) (x) (Fx-+ Fx) ') seems - as Frege took it to be - quite small and
quite natural. But, for various reasons, it has been urged that second-
order predicate calculus ought not to count as belonging to logic, but
is, rather, itself a mathematical rather than a logical theory. As I've
observed, the test of completeness would rule it outside logic; and
Quine argues against its inclusion on the grounds of its excessive onto-
logical commitments. (However, I'vc also already given reasons for
having reservations about completeness as a criterion of a system's
counting as logical; and in ch. 4 §2 I shall urge sorne scepticism about
Quine's critcrion of ontological commitment.)
So here is a case whcre the fate of a philosophical theory seems to
depend on the demarcation of logic; if second-order 'logic' is logic,
arithmetic is reducible to logic, if not, not. But isn't it rather dis-
maying to think that the truth of logicism should depend on so prag-
matic a question as I have taken that of the scope of logic to be? Not,
I think, once one goes a little deeper, and asks why it should be
thought to matter whether arithmetic is really purely logical. The
1
Frege devised the first fully worked-out fonnal logical system as a
preliminary, as he hoped, to establislúng the truth of logicism by
actually deriving the Peano postulates for arithmetic from his logical
axioms. He developed the logical apparatus in 1879, supplied the
appropriate logical definitions of arithmetical terms in 1884, and the
derivations in 1893 and 1903 : see Camap 1931 for a straightforward
introduction to the logicist philosophy of mathematice.
Io Philosophy of logics

really important issue is, or so it seems to me, obscured by

putting the question as if the scope of logic were the key point.
Why did Frege think it was important to show that arithmetic is
reducible to logic? The motivation for logicism was at least in part
epistemological; the principies of logic, Frege thought, are self-
evident, so that if the laws of arithmetic can be shown to be derivable
from them, they are thereby shown to be epistemologically secure -
they acquire innocence by association, so to speak. It turned out,
however, that Frege's logic (or 'logic') was inconsistent- Russell's
paradox ( cf. ch. 8) is derivable in it. Frege's response to the discovery
of the inconsistency was to concede that he'd never really thought
that the relevant axiom was quite as self-evident as the others - a com-
ment which may well induce a healthy scepticism about the concept
of self-evidence. The relevance of this story to present concerns,
though, is this: that since Frege's basis - logic or not - hasn't the
epistemological standing he thought, the epistemological point of his
programme is lost regardless of the decision about the demarcation
of logic.

One thing, at least, should be quite clear by now: that whether or not
a formal system should count as a logic is itself a question which
involves quite deep and difficult philosophical issues. It is ali to the
good that the pervasiveness of philosophical problems in logic be
evident at the outset. For the very rigour that is the chief virtue of
formal logic is apt, also, to give it an air of authority, as if it were above
philosophical scrutiny. And that is a reason, also, why I emphasise the
plurality of logicalsystems; for in deciding between alterna ti ves one is
often obliged to acknowledge metaphysical or epistemological pre-
conceptions that might otherwise have remained implicit.
 2

Validity

Assessing arguments
Argumenta are assessed in a great man y ways; sorne, for
instance, are judged to be more persuasive or convincing than others,
sorne to be more interesting or fruitful thanothers, and so forth. The
kinds of assessment that can be made can be classified, in a rough and
ready way, like this:
(i) logical: is there a connection of the appropriate sort between
the premises and the conclusion?
(ii) material: are the premises and conclusion true?
(iii) rhetorical: is the argument persuasive, appealing, interesting
to the audience?
I have given only the vaguest indication of the kinds of question
characteristic of each dimension of assessment, but a rough indication
should be adequate for present purposes. The separate category given
to rhetorical considerations is not intended to suggest that the validity
of an argument, or the truth of its premises, is quite irrelevant to its
persuasiveness; it is intended, rather, to allow for the fact that, though
if people were completely rational they would be persuaded only by
valid arguments with true premises, in fact often enough they are
persuaded by invalid arguments or arguments with false premises,
and not persuaded by sound (cf. p. 14 below) arguments (see e.g.
Thouless 1930, Stebbing 1939, Flew 1975, Geach 1976 for discussion
of, and advice on how to avoid, such failures of rationality).
In what follows I shall be almost exclusively concerned with the
first, logical, dimension of assessment. Within this dimensión, in turn,
I need to distinguish different standards of assessment which may be
employed: an argument, that is, m.ay be judged to be deducti'Ve/y oalid,
12 Phil.osophy of logics

or deductively invalid but inductively strong, or neither. Deductive

standards, as this indicates, and as we shall see in more detail la ter, _are
more stringent than inductive ones - the connection between premises
and conclusion has to be, as it were, tighter for deductive validity than
for inductive strength.1
Sometimes it is suggested (e.g. Barker 1965, Salmon 1967) that
there are tsoo kinds of arguments, deductive arguments, on the one
hand, and inductive arguments, on the other. This 'distinction ', at
least as it is usually explained, only confuscs matters. One is told that
'deductive arguments' are 'explica tive' or 'non-ampliative ', that is,
they 'contain nothing in the conclusion not already contained in the
premises '. If this is, as it seems to be, in tended as an explanation of
what it is for an argument to be deductively valid, it is apt to tum out
either false, if 'contains nothing in the conclusion not already con-
tained in the premises' is taken literally (for while 'A and B, so A'
meets this condition, 'A, so A V B', which is also deductively valid,
does not) or else trivial, if 'contains nothing in the conclusion not
already contained in the premiscs' is taken metaphorically (for what
is the test for 'A V B' being implicitly 'contained in' 'A', if not that
'A V B' follows deductively from 'A'?). 'Inductive arguments ', by
contrast, one is told, are 'ampliative' or 'non-explicative ', that is to
say, 'their conclusions go beyond what is contained in their premises '.
This makes matters worse, because it cannot be taken, symmetrically
with the explanation of 'deductive argument ', as an explanation of
what it is for an argument to be inductively strong. For ali it says about
inductive arguments is that they are not deductively valid; but not ali
deductively invalid arguments are inductively strong.
So I (with Skyrms, e.g. 1966 ch. 1) prefer to put the matter this
way: it is not that there are two kinds of argument, but that argu-
ments may be logically assessed by different, deductive or inductive,
standards; they may be deductively valid, inductively strong, or
neither. And this makes it clear what the next questions should be:
What is an argument? What conditions must an argument meet if it is
to count as deductively valid or inductively strong?
What is an argument? W ell, one recognises that sorne stretches of
discourse are intended as supporting a conclusion by means of
premises, as arguing to a conclusion from premises; in informal dis-
1
Sorne writers, notably Peirce and, more recently, Hanson, think there
are other logical standards, "abductive ' standards, aa well, Cf. Haack
I977b for sorne relevant discussion.
 Validity 13

course in natural languages this intention may be signalled by marking

the passage from one statement to another by means of such locutions
as 'so', 'hence', 'it follows that', 'because' and so forth; in formal
logic, by the presentation of a string of formulae with an indication on
each line that it is claimed to follow by such-and-such a rule of infer-
ence from such-and-such a previous line or lines. What one judges to
be valid or invalid, though, can be thought of simply as a stretch of
discourse: if one is considering formal argument, a sequence of wffs
of a formal language, or, if one is considering informal argument, a
sequence of sentences (or perhaps statements or propositions; cf.
ch. 6) of natural language. (Similarly, sorne of the things people say
are intended assertively - the speaker means to claim their truth - and
others are not; but it is tohat is said that is true or false.)

2 Deductive validity: with sorne brief comments about

inductive strength
Validity in a system
Within a formal logical system, validity can be defined both
syntactically and semantically, i.e. in terms of the axioms or rules of
the system, and in terms of its interpretation. I will represent a formal
argument as a sequence of well-forrned formulae (i.e. grammatical
sentences of a formal language; hereafter 'wffs ') A1 ... An-i, An,
(n � 1) of which A1 ... An-l are the premises, and An the conclusion.
Syntactic validity can then be explained along the following lines:
A1 ••• A"_1, An is valid-in-L just in case An is derivable from
A1 ... An-l• and the axioms of L, if any, by the rules of
inference of L.
This is usually represented: A1 ... An-l 1-L An.
Semantic oalidity can be explained along the following lines:
A1 ... An-1' An is valid-in-L just in case An is true in all
interpretations in which A1 ••• An-l are true.
This is usually represented: A1 ... An-l• l=L An.
The 'L' in ' 1-L ' and 'l=L' serves to remind one that both these con-
ceptions of validity are system-relatioe.
Corresponding to the syntactic and semantic ideas of validity of
sequences of wffs are the ideas of theoremhood and logical truth,
respectively, of wffs. Yo u may have noticed that I allowed the possi-
bility of argumenta consisting of just one wff (sometimes these are
14 Philosophy of logics

called 'zero-premise conclusions '). If the ideas º.f validity just

sketched are applied to this special case, the upshot is:
A is valid-in-L (is a theorem of L) just in case A follows
from the axioms of L, if any, by the rules of inference
of L (1-LA)
and
A is valid-in-L (is a logical truth of L) just in case A is
true in ali interpretations of L (l=L A).
I have represented theoremhood and logical truth as, as it were,
special cases of, respectively, syntactic and semantic validity. It would
also have been possible to approach the matter the other way round,
and explain validity as theoremhood of the corresponding conditional.
The former approach has the advantage of stressing logic's concern
with the connection between premises and conclusion, which is why
I chose it.
How do the syntactic and semantic ideas fit together? W ell, one
naturally aspires to have a formal system in which just those wffs
which are syntactically valid are semantically valid (soundness1 and
completeness results show that theoremhood and logical truth
coincide).

Extra-systematic validity
The conceptions of validity, syntactic and semantic, con-
sidered thus far, are system-relative and apply only to formal argu-
ments. What is going on, though, when one judges an informal
argument to be valid? One is claiming, I take it, that ita conclusion
follows from its premises, that its premius couldn't be true and its
conclusion false. (If, besides being valid, an argument has true
premises - and so, being valid, true conclusion too - it is said to be
sound.) When, intuitively, we judge sorne ordinary, informal argu-
ments good and others bad, something like this conception of validity
is probably being deployed. Of course, judging an argument 'good' is
apt to involve more than judging it valid; but we recognise that
validity is an important, though not the only, virtue of an argument.
The question arises, whether there is also an informal extra-
systematic conception corresponding to the system-relative 'notions
of theoremhood and logical truth. I think that there is, though I sus-
I A dífferent sense of ' aound', applying not to logical a}'ltema but to
argumenta, will be defined below.
 Validity 15

pect that it is somewhat less developed and central than the extra-
systematic idea of validity (another reason for treating logical truth
as a special case of validity rather than vice-versa). The extra-
systematic idea of a valid argument as one such that its premises
couldn't be true and its conclusion false, adapted to the case of a single
statement (as the formal definitions were adapted to the case of
'zero-premise conclusions ') yields the notion of a statement that
couldn't be false - the notion, in other words, of a necessary truth.
And something like this idea is indeed to be found at the informal
level. For example, one judges that sorne statements are 'tautologous';
this, in the non-technical sense, means that those statements are
trivially true, they just (as the etymology of 'tautologous' suggests)
say the same thing twice, and consequently -they couldn't be false.
The informal notion of a tautology, of course, is broader than the
technical usage, which includes only logical truths of truth-functional
logic. And the informal idea of a necessary truth is also broader than
the formal idea of a logical truth ( cf. ch. 10 § 1 ). It should occasion no
great surprise that these informal conceptions have themselves been
refined with the development and study of formal logical systems.
But what is one to say of the connection between the system-
relative conceptions of validity, applicable to formal arguments, and
the extra-systematic conception, applicable to informal arguments?
Something like this: formal logical systems aim to formalise informal
argumente, to represent them in precise, rigorous and generalisable
terma; andan acceptable formal logical system ought to be such that,
if a given informal argument is represented in it by a certain formal
argument, then that formal argument should be valid in the system
just in case the informal argument is valid in the extra-systematic
sense.

Logica utens and logíca docens

In fact, there is likely to be a quite complex process of adjust-
ment. One may begin to develop a formal system on the basis of
intuitive judgments of the extra-systematic validity of informal argu-
ments, representing those arguments in a symbolic notation, and
devising rules of inference in such a way that the formal representa-
tions of informal arguments judged (in)valid would be (in)valid in the
system. Given these rules, though, other formal arguments will turn
out to be valid in the system, perhaps formal argumente which repre-
sent informal arguments intuitively judged invalid; and then one may
16 Philosophy of logics

revise the rules of the system, or one may, instead, especially i_f the �
- le
ofinfo_rmal mvahdtty
isagreeablysimple and plausible and the in�i�on
not strong, revise one's opinion of the vahdity of the mform�l argu-
ment, or else one's opinion of the appropriateness representmg t_hat
of
lo?tc�l
informal argument in this particular way. And_ on_ce a form�l
like_ly_ that �t will m
system becomes well-established, of co�r�, it t�
tum tutor one's intuitions about the validity or invalidity of informal
arguments. Following Peirce (who in turn borrowed the_ �
ter nology
from the medieval logicians] one may call one's unreflective Judgment
of the validity of informal arguments the logica utens, the more
rigorous and precise judgments developed as, through reflection on
these judgments, formal systems are devised, the logica docens. The
picture is somewhat like fig. 1.

Jogica utens logica docens

informal arguments formal arguments
symbolic
representation
of informal
argument
extra-systematic system-relative
validity validity
Fig. I

Sorne writers have doubts about the adequacy of the extra-

systernatic conception of validity as I explained it above. What, specifi-
cally, they object to in the idea that an argument is valid if it is
impossible for its premises to be true and its conclusion false, is the
'and '. On this account, if the premises of an argumentare impoesible,
or if its conclusion is necessary, then, since a fortiori it is impossible
that its premises should be true and its conclusion false, that argu-
ment is valid; and this is so, of course, even if its premises are quite
irrelevant to its conclusion. Proponents of 'relevan ce logic' therefore
challenge this conception of validity; and because of this challenge
they urge the adoption of a non-classical formal logic which requires
relevance of premises to conclusion (see Anderson and Belnap 1975
§22.2. 1, and cf. ch. 10 §7); so their dissatisfaction with the usual
informal conception of validity is intimately conneeted with their
challenge to classical logic. (Conventionally, consideratione of rele-
 Validity 17
vanee are apt to be relegated to the rhetorical rather than the logical
dimension of assessment of arguments.)

Inductioe strength
Inductive strength could be characterised, syntactically or
semantically, relatively to formal systems of inductive logic. How-
ever, since there is no formal system of inductive logic which has
anything approaching the kind of entrenchment that classical deduc-
tive logic enjoys, the extra-systematic idea has, in the case of inductive
strength, an especially central role. The idea is that an argument is
inductively strong if its premises give a certain degree of support,
even if less than conclusive support, to its conclusion: if, that is, it is
improbable that its premises should be true and its conclusion false.
(Notice that if one puts it this way all deductively valid arguments
would count as inductively strong; deductive validity will be a limit
case of inductive strength, where the probability of the premises'
being true and the conclusion false is zero.)
It is worthy of notice, however, that in his characterisation of the
extra-systernatic idea of inductive strength Skyrms (1966 pp. 9-11)
insista on the formulation: 'it is improbable, given that the premises
are true, that the conclusion is false', beca use he doesn't want to
allow that the high probability of its conclusion or the low probability
of its premises should be sufficient, of themselves, for the inductive
strength of an argument. Significantly, then, his view of inductive
strength is closely analogous to the relevance logicians' conception of
deductive validity.1

3 Formal logical systems: tbe 'L' in "valid-In-L'

I distinguished, above, systern-relative conceptions of validity
applicable to formal arguments and an extra-systernatic conception
applicable to informal arguments. An adequate account of the former
- of validity-in-L - will obviously require sorne explanation of how
one identifies and individuates formal systems. The problem may be
illustrated by considering the sentence logic to be found in, say,
Principia Mathematica (Russell and Whitehead 1910) and Beginning
Logic (Lemmon 1965): if one was concerned with the difference
1 Notoriously, of course, there is a problem about the justification of
induction. Nothing in what I have saíd shows that there are any
argumenta which are (deductively invalid but) inductively strong. In
fact, 1 think that deduction and induction are more symmetrical than is
generally auppoaed; cf, Haack 1976a.
18 Philosophy of logics

between two-valued and many-valued logics, one would naturally

regard them as altemative formulations of the same (two-valued)
system, whereas if one was concemed with the contrast between axio-
matic and natural deduction techniques (see below, p. 19), one
might take them as examples of different systems.
For the sake of having sorne suitably neutral terminology, 1 shall
call a specific presentation of a system a 'formulation' of a logical
system. Now, differences between formulations are of two main kinds:
differences in vocabulary, and differences in axioms and/or rules of
inference. 1 shall first sketch sorne significant differences between
formulations, and then offer two accounts of 'the same system ', one
broader and one narrower.

N'otational variants
Typographically different expressions may be used for the
same operations (e.g. for the same truth-functions), Among the com-
monest current notational variants one finds:
for negation: -p, -p, p, Np
for disjunction: p v q, Apq
for conjunction: p & q, p · q, p II q, Kpq
for material implication: p --+ q, p ::, q, Cpq
for material equivalence: p = q, p ..... q, p - q, Epq
for universal quantification: (x), (Vx), Ax, TI.t"
for existential quantification: (3x), (Ex), Vx, LX
The last notation, in each case, is Polish notation, which has the
advantage of being bracket-free; operators precede the formulae they
govern, and scope is determined without brackets.

Altematioe primitive constants

Different sets of constants are equivalent in expressive power;
e ·g , '&' an d' - , or • V ' an d. - ' to express z-valued .
truth-functtons,
'(3x) and ' - ' or '(x)' and ' - ' for existential and universal quantifi-
cation. Sorne formulations take e.g. • &' and • - ' as primitive and
define ' V ' and '--+'; others take ' V ' and ' - ' as primitiva and define
'&' and '--+', and so forth. Principia Mathematica, for example,
has only negation and disjunction as primitive, whereas Beginning
Logic has negation, disjunction, conjunction and material implication.
 Validity 19

Axiomatic and natural deduction formulations

An axiomatic system of logic (e.g. Principia Mathematica)
includes, besides one or more rules of inference, a privileged set of
wffs, the axioms, which may be used at any point in an argument, and
the truth of which is unquestioned in the system. The axioms are
included among the theorems of the system, since, trivially, they are
derivable from themselves. (An axiomatic system must have at least
one rule of inference, since no derivations or proofs would be possible
without the means to move from one wff to another.)
A natural deduction formulation (e.g, Begi"nning Logic), by contrast,
relies just on rules of inference. (A rule of assumptions will enable one
to get started without the need of axioms from which to begin.) It is
worth observing that natural deduction rules have an indirect, even
quasi-metalogical character; consider the rule of vel elimination: if
you have derived C from assumption A (plus possibly other assump-
tions) and derived C from assumption B (plus possibly other assump-
tions), you may derive C from the assumption that A v B (plus any
other assumptions used in deriving C from A and from B).1
Sometimes axioms not known to be true, or even known to be false,
are adopted simply with the airn of investigating their consequences.
A famous example comes from the history of geoJetry. Saccheri took
the contradictory of Euclid's parallel postulate as an axiom, hoping to
show that the result was an inconsistent system, and hence, that the
parallel postulate was deducible from Euclid's other axioms. Since
this postulate is actually independent of the others, he didn't succeed
in this aim. (Cf. the discussion of Prior's rules of inference for 'tonk ',
ch. 3 §2.)
The very same valid arguments and theorems can be generated
either axiomatically or by means of natural deduction rules: by the
axioms of Principia Mathematica or the rules of Begi'nning Logic, for
instance. But, of course, this is not to say that the difference between
natural deduction and axiomatic techniques is of no significance.
Kneale for example argues (1956 §4) that natural deduction formu-
lations better reflect the central concern of logic with the validity of
1 Cf. Blanché I962 and Prawitz I965 for detailed discussion of, respec-
tívely, axiomatic and natural deduction techniques, The pioneer natural
deduction preaentation, in Gentzen I934, includes one axiom. Gentzen
also devised a metalogical calculus, the sequent calculus; see Hacking
I97 + for an interesting attempt to dernarcate the acope of logic
formally by reference to the sequent calculus.
20 Philosaphy of logics

arguments. An unfortunate effect of the axiomatic formulation of the

Begriffsschrift and Principia, Kneale suggests, was a shift of attention
from validity of arguments to logical truth of wffs, And Blumberg
suggests (1967 p. 24) that natural deduction formulations highlight
the difference between formal logic and other formal theories, such as
geometry or biology, say, which require special axioms relating to
their special subject matter, over and above a common basis of logical
rules of inference. I agree in stressing logic's concern with arguments,
and I agree that with natural deduction presentations of formal logic
this concern is thrown into relief, However, since validity of argu-
ments and logical truth of formulae are intimately connected, an axio-
matic formulation needn't necessarily distort one's perspective.
(Camap 1934 points out that one can think ofaxioms as rather pecu-
liar rules of inference, to the effect that one may infer the given wff
from any premises or none at ali.) And the distinction between logical
and other formal systems needn't be lost in an axiomatic presentation
of logic, either, since there remains room for a distinction between
logical and proper (i.e. geometrical, biological or whatever) axioms.
It is pertinent that sorne instrumentalist philosophers of science
have urged that one regard scientific laws rather as rules than as
axioms.

Alternatioe axioms and/or rules

If two formulations are notational variants, thcir axioms
and/or rules wíll differ at least typographically; again, if they take
different constants as primitíve, each will, usually, employ its primi-
tive constants in its axioms and/or rules. (Sometimes, though, a
system is formulated in such a way that defined constants appear in
the axioms/rules; in Principia only ' - 'and 'V' are primitive, but '--+'
also appears in the axioms.)
Sorne formulations employ axiom schemata rather than axioms and
a �ubstitution rule. The difference is that between having, say, the
axiom:

(P-+ q)-+ ((q-+ r)-+ (p--+ r))

and the rule that any substitution instance of an axiom is a theorem,

and having the schema :

(A--+ B)-+ ((B--+ C)-+ (A--+ C))

where the use ofthe 'metavariables' 'A', 'B', 'C' indicates that no
 Validity :u

matter what wff of the language is put for these letters, the resulting
wff is an axiom.
Quite apart from divergences of notation and presentation already
mentioned, different formulations may simply have different sets of
axioms/rules, even when notational differences are allowed for: their
sets of axioms/rules may overlap or even be entirely distinct. As an
example, compare Mendelson's and Meredith's axiom schemata, both
in'-' and '-+', for z-valued sentence calculus:
Mendelson's set:
1. (A -+ (B-+ A))
2. ((A-+ (B-+ C))-+ ((A-+ B)-+ (A -+ C)))
3. ((-B -+-A) -+ ((-B -+ A) -+ B))
Meredith's:
1. ((((A-+ B)-+ (-C-+-D))-+ C)-+ E)-+
((E-+ A)-+ (D-+ A))
(And see Prior 1955 pp. 301ff., Mendelson 1964 pp. 40-1 for altema-
tive axiom sets.)
The example just given is of altemative axiom sets for the z-valued
sentence calculus; the alternative formulations yield just the same
sets of theorems and valid inferences. Another way in which formula-
tions may differ is that they may result in different theorems or valid
inferences; for example, Intuitionist sentence logic lacks sorne
classical theorems, including double negation and excluded middle.

At this point I have sufficient material to retum to my original problem,

when to treat alternative formulations as formulations o/ the same
system. I shall suggest two accoun ts of ' the same system ', one
broader and one narrower, each suitable for certain purposes.
The narrower sense: L1 and L2 are alterna ti ve formulations of the
same system if they have the same axioms and/or rules of i'nference
once allowance has been made for differences of notation (e.g.
replacing ' & ' by '. ') and of primitive constants ( e.g. replacing
'p & q' by' -(-p V -q)').
The broader sense: L1 and L2 are altemative formulations of the
same system if they have the same theorems and oalid inferences once
allowance has been made for differences of notation and of primitive
constants.
An example: the formulations of Principia Mat�a and of
22 Philosophy of logics

Beginning Logic are formulations of different systems in the narrower

sense (one has axioms plus modus ponens, the other only rules of
inference) but of the same system in the broader sense (they generate
the same theorems and inferences ).
These two senses of 'same system' will help, 1 hope, to reconcile
sorne conflicting intuitions. The narrower of these senses seems suit-
able to use in the definitions of validity-in-L, whereas the broader will
be more useful in contrasting, for example, z-valued and many-
valued logics. An advantage of the narrower sense for the account of
validity is that it avoids a circle which otherwise threatens, in which
'theorem' and 'valid inference' are defined relatively to a system, and
'system' relatively to sets of theorems and valid inferences.
The narrower sense will also be useful for the discussion of incon-
sistent formulations. Since, except in sorne unconventional systems,
anything whatever follows from a contradiction, in virtue of the
theorem 'A -+ ( -A -+ B) ', all inconsistent systems will count as the
same system in the broader sense. The narrower sense enables one to
respect the intuition that sorne but not ali inconsistent formulations
are nevertheless of considerable philosophical interest; an exarnple
would be Frege's, in which Russell's paradox is a theorem.

4 Validity and logical form

One cannot tell whether an informal argument ia (in the
extra-aystematic sense] valid merely by inveatigating the truth-valuea
ofita premises and conclusion. If the argument haa true premiaea and
false concluaion, that shows that it ia invalid ¡ but if it has true
premises and true conclusión, or false prernisea and true conclusion,
or false premises and false conclusion, that doesn't show that it'a valid.
For it is valid only if it couldn't have, not just doem't have, true
premises and false conclusion. A technique that is often useful to
show that an argument is invalid, even though, as it happens, it
doesn't have true premises and false conclusion, is to find another
argument which is of the same form and which does have true premises
and false conclusion. For example, to show that: 'Either Godel's
proof is invalid, or arithmetic is incomplete, so arithmetic is incorn-
plete ', though it has true premise and true conclusion, is nevertheless
invalid, it could be pointed out that the structurally similar argument:
'Either 7+ 5 = 12 or dogs meow, so doga meow' has true premise
�d �al� conclusion. This, of course, is a better method of showing
mvalidity than of showing validity; if one can't find an argument of
 Validity 23

the same form with true premises and false conclusion, that isn't con-
clusive proof that an argument is valid (cf. Massey 1974).
What one seeks, to show an argument is invalid, is a structurally
similar argument with true premises and false conclusion; and this
suggests that there is sorne truth in the dictum that arguments are
valid or invalid 'in virtue of their form '. And formal logical systems
are devised to represent in a schematic, generalised way the structure
which we judge to be shared by a group of informal arguments, and
to be the basis of their validity or invalidity. This is apt to suggest, in
tum, a picture of informal arguments as having a unique, recognisable
structure, as composed, as it were, of a skeleton, the expressions which
constitute its form, clothed in flesh, the expressions which constitute
its content; and of the formal logician as simply devising symbols to
represent the 'logical constants ', the structural components. But this
oversimplifies. A better picture, 1 think, is this. One recognises struc-
tural similarities between informal arguments, similarities charac-
teristically marked by the occurrence of certain expressions, such as
'and' or 'unless' or 'every '. (But one shouldn't expect each informal
argument necessarily to have a unique place in this pattern.) The
formal logician selects, from among the expressions whose occurrence
marks structural similarities, those which (for various reasons, truth-
functionality, for instance; cf. ch. 3 §2) are promising candidates for
formal treatment.
This picture - sketchy as it is - already begins to explain why it
should be that attempts to specify which expressions of English
should be counted as 'logical constants' should have a tendency to
conclude with a somewhat uncomfortable admission that not all
suitably 'topic-neutral' expressions (Ryle 1954), not all the expres-
sions which seem to be essential to the validity of informal arguments
(von Wright 1957) are represented in the symbolism of formal logic;
for instance, 'several' is as topic-neutral as, and may be as essential to
an argument as, 'all'; the formal logicians' equipment includes an
analogue of the latter but not of the former. Compare Ouine's
enumeration of the logical constants: ' ... basic particles suc.h as "is",
"not", "and", "or", "unless", "if", "then", "neither", "nor",
"sorne", "all", etc.' (1940 p. 1); it is notable that the list comprises
just those English expressions that can comfortably be represented
within the classical sentence and predicate calculi, and that it excludes
'necessarily' and 'possibly', for instan ce, no doubt beca use of Quine's
scepticism about the intelligibility of modal logic. The 'etc.', of
24 Philosophy of logics

course, helps not at all, since one is given no indication what would
count as a permíssible addition to the list. .
The relation between informal arguments and their formal repre-
sentations is, as this would lead one to expect, not straightforward!y
one-one. An informal argument may be appropriately represented tn
dífferent ways in different formalisms; for instan ce:

Every natural number is either greater than or equal to

zero, and every natural number is either odd or even, so
every natural number is either greater than or equal to
zero and either odd or even
could be correctly represented in sentence calculus as:
p
q
and in predicate calculus as:

(x) Fx & (x) Gx

(x) (Fx & Gx)
(Notice that the availability of alternative representations needn't
depend upon any ambiguity in the oríginal, though if an informal
argument is ambiguous that will naturally mean that it has more than
one formal representation; cf. Anscombe's splendidly ambiguous 'if
you can eat any fish, you can eat any fish '.)

'p, so q' is invalid, but '(x)Fx & (x)Gx, so (x) (Fx & Gx)' ia valid¡
and since the latter reveals more of the structure of the original,
informal argument than the former one might be tempted to think
that the best formal representation will be the one that exhibits the
most structure, But my informal argument can be represented, again
in the symbolism of predicate calculus, with further structure
revealed, as:
(x) (Fx V Gx) & (x) (Hx v Jx)
(x) ((Fx V Gx) & (Hx v lx))

It is clear that there is a sense in which this exhibits more structure

than one needs¡ it is preferable to think of the optimal formal repre-
sentation as the one which reveals the least structure consistently with
supplyíng a formal argument which is valid in the systern if the in-
formal argument is judged extra-systematically valid, This is Quine'a
 Validity 25
maxim of shallota analysis (196oa p. 160): 'where it doesn't itch,
don't scratch '.
In the interplay between logica utens and logica docens, 1 suggested
(p. 16), one may judge it worthwhile to sacrifice pre-formal judg-
ments of validity to smoothness of formal theory, or modify one's
formal theory to accommodate assessments of informal arguments,
or - and it is this point I want to pursue here - to revise one's view of
the appropriate way to represent an informal argument in formal
logic. One criterion by which one judges whether an informal argu-
ment is correctly represented by a given formal argument is that
intuitive judgments of validity be respected. For example, one's confi-
dence that 'Somebody is Prime Minister and somebody is Queen, so
the Prime Minister is Queen' is invalid would lead one to resist
representing it by a formal argument valid in predicate calculus, as:
a=b
a=c
b=c
and to require something like the invalid:
(3x) Fx & (3x) Gx
(1x) Fx=(1x) Gx
lf, on the other hand, one judges an informal argument to be valid,
one will seek a representation by means of a valid formal argument.
For example, within the confines of the standard predicate calculus,
adverbially modified predicates are normally represented by means of
new predicate letters, so an argument like:
The President signed the treaty with a red pen.
So, the President signed the treaty.
would be represented as:
Fa
Ga
where 'a' representa 'the President ', 'F' represents 'signed the treaty
with a red pen' and 'G ', 'signed the treaty '. This, of course, is an
invalid argument in predicate calculus; and so, in view of the pre-
sumed validity of the original, informal argument, it has been urged
that sorne more perspicuous means of representing adverbial modifi-
cation, which does not simply obliterate the logical connection be-
26 Philosophy of logics

tween an adverbially modified predicate and its unmodi.fi.ed fo�,

be devised. Davidson, for example (1968a), proposes a representatlon
along the lines of:
(3x) (x was a signing of the treaty by the President
and x was with a red pen)
(3x) (x was a signing of the treaty by the President)

which, like the original, is valid. Notice that this supplies the original
argument with a representation within the standard predicate calculus
by quantifying over events and treating adverbs as predica tes of events;
another possibility would be to extend the standard formalism, e.g. by
the addition of predicate operators to represent adverbs. In the case of
modal adverbs, 'necessarily' and 'possibly ', this k.ind of extension of
the vocabulary of formal logic has already taken place.

Sorne philosophers of logic have urged the claims of a neater picture,

according to which each informal argument has a unique logical forro -
perhaps not immediately recognisable - which the correct symbolic
representation will exhibit. Such a view was held, for instance, by
Wittgenstein and by Russell during their logical atomist periods (see
e.g. Russell 1918, Wittgenstein 1922; and cf. comments on Russell's
theory of descriptiona in ch. 4 §3); for they aspired to devise a unique,
idcally perspicuous language in which logical form would be perfcctly
exhibited, More recently, Davidson has taken a similar stance: for
Davidaon, the logical form of an argument is ita representation in a
formal language for which truth can be defined in accordance with
the constraints irnposed by Tarski's theory ( ch. 7 § 5). Ruaaell thought
that the grammatical form of a sentence was apt to be misleading as to
its logical forro; sorne recent writers, impressed by Chomsky's postu-
lation of a deep gramrnatical structure underlying, but perhaps quite
different from, surface grammatical structure (see Chomsky 1957),
have suggested that the logical forro of an argument could be identi-
fied with its deep grammatical structure (see e.g. Harman 1970). The
relcvant deep grammatical/logical structure would have, presumably,
to be universal as bctween languages, since otherwise one would be in
danger of allowing that an argument could be, say, valid in Hebrew
but invalid in Hindi; and it is, to my mind, doubtful whether one is
e�titled �o expect that linguists will evcntually discover a suffi.ciently
n�, �versal, grammatical structure. So I cannot be altogether
optim.iatic about the prospecta for thi.a - admittedly gratifyingly tidy -
 Validity 27
picture. But nonetheless I see no reason for dismay at the inter-
dependence between intuitive, informal judgments of validity, hunches
as to the essential structural features of informal arguments, and the
development of formal logical systems. Rather, one might even feel
sorne satisfaction at the way this explains why the central questions in
philosophy of logic should cluster around the issue of the fit between
informal arguments and their formal representations: an issue which,
wíth respect to connectives, quantifiers and singular terms, the next
three chapters will investigate more thoroughly.
 3
Sentence connectives

I Formal considerations
I shall begin by sketching sorne important formal features of
the sentence connectives, and proceed to a consideration of sorne
philosophical questions about the meanings of the connectives.

Adequate sets of connectioes: functional completeness

The connectives - '- ', '&','V','-+-', and '= ' - of classical
sentence calculus are truth-functional: the truth-value of a compound
sentence formed by means of them depends only on the truth-values
of its components. A set of connectives is adequate if it can expresa all
truth-functions. There are 16 (221) z-valued truth-functions of two
n
arguments.1 Each of the sets { - , -+-}, { - , v }, { - , &}, { 1 }, and {
('AIB' is 'not both A and B' and 'A+B' ia 'neither A nor B') ia
adequate to expresa them all. A formal aystem is functiona/ly complttt
if it has an adequate set of connectives. For inatance, Principia, with
' - ' and 'V' as primitive, is functionally complete, whereaa the impli-
cational fragrnent of sentence calculus, with '-+-' alone, is not. Many
formulations - e.g. Lemmon 1965 - have more connectives than are
needed for functional completeness. lt is because there are alterna-
tive adequate sets of connectives that one has formulations of sentence
calculus with different sets of primitives. Given any adequate set, the
other connectives can be defi.ned. For example, with '-' and '-+-' as
1 vis:
A B 1 2 3 4 s 6 8 9 JO JI
7 12 13 14 15 16
t t t t t t t t
t f f f f f f f f
t f t t t t f f f f t t t t f f f f
f t t f f t f f t t f f t t f f
f f f t f f t f t f t f t f t f
 Sentence connectioes

primitive, 'A V B' can be defined as' -A� B ', and then 'A & B' as
' - ( -A V - B)'; with 'I ' or ',¡.' primitive ' - A' can be defined as
'AIA' or 'A,1-A '. Sorne formulations employ a constant, 'F', which is
always to have the value /, and define ' -A ' as 'A � F'. In each case
the correctness of the definitions can be checked by comparing the
truth-tables of definiens and definiendum and observing that they
correspond to the same truth-function.

Characteristic matrices: decidability

A matrix, or set of truth-tables, M, is characteristic for a
system S iff ali and only the theorems of S are designated on M, and
ali and only the valid inferences of S are designation-preserving in M.
Any value may be designated, but usually the point is to designate
the 'truth-like' value, or, perhaps, in the case of many-valued logics,
values; in z-valued logic, of course, 't' is designated. A wff is
designated on M iff it takes a designated value whatever assignment is
made to its atomic parts; a rule, from A ... An to infer B, is designa-
tion-preserving iff B takes a designated value whenever ali of
A ... A,. do. For example, the z-valued truth-tables are characteristic
for classical sentence calculus.
Finite truth-tables provide a decision procedure, that is, a mechanical
method for determining, for any wff of the system, whether it is a
theorem.

Many-valued logic
Of course, it would be possible to devise many-valued
matrices characteristic for z-valued sentence calculus. I call z-valued
sentence logic "z-valued ' rather than 'many-valued' beca use this is
the smallest number of values which can supply a characteristic
matrix. By an 'n-valued logic' I shall understand a system which has
a characteristic matrix with n values and no characteristic matrix wíth
m values for m < n, Sorne of the systems I referred to as 'deviant' have
finite characteristic matrices; not unexpectedly, one motivation for
devising such systems has been the belief that sorne sentences within
the scope of logic are neither true nor false, but either are truth-
valueless, or, perhaps, have an intermediate truth-value: a belief
which will receive closer attention in ch. 11. Other deviant systems,
such as Intuitionist and sorne quantum logics, have no finite but only
infinite characteristic matrices. In what follows, 'many-valued logic'
will mean 'n-valued logic for 2 < n < oo', except when I specifically
speak of infinittly many-valued systems.
30 Philosophy of logics

In an n-valued logic, any given place in a truth-table can be

occupied by any of n values, so, since the truth-tabl� for a k-place
connective has nk entries, the number of truth-functlons of k argu-
ments in an n-valued logic will be nnk - a number which increas.es
enormously with small increases in n. Lukasíewicz's 3-valued log1c'.
= ',
with ' - ', ' & ', 'y ', '-+' and ' is functionally in complete; Slu peck.i
showed that it becomes functionally complete with the addition of
a new r-place connective T (for 'tertium ') such that ' TA' takes the
intennediate value whatever the value of 'A'. &:, is to be expected,
the familiar interdefinability relations are apt to break down in many-
valued logic; for example, in Lukasiewícz'a 3-valued logic 'A Y B'
does not, as in 2-valued logic, have the same truth-table as' -A -+ B';
it can be defi.ned, instead, as ' ( A -+ B) -+ B '.

2 The meanings oí the connectives

Formal languages and informal readings
One can look at the sentence calculus from, as it were, four
levels:
(i) the axioms/rules of inference
(ii) the formal interpretation (matrices)
(iii) the ordinary language readings of (i)
(iv) the informal explanation of (ii)
(i) is the level of syntax; levels (ii) and (iv) are dubbed by Plantinga
1974 pp. 126-8 'pure' and 'depraved' semantics, respectively. Levela
(i) and (ii), being formal, are agreeably rnanageable; but levels (iii)
and (iv), though trickier, are no lesa important. l observed in ch. I
that the identification of a system as a syetem of logic requires appeal
to its (intended?) interpretation. To identify a system as a sentence
calculus one does not only need to know the axioms/rules and their
formal interpretation by means of matrices; one also needs to know
that the values are to represent truth and falsity, the letters 'p', 'q', etc.
to represent sen ten ces, ' - ' negation, ' & ' conjunction, 'v' disjunc-
tion, and M forth. One's understanding of the connectives must
presumably derive somehow from sorne or all of these levels.
One's view about how the connectives get their meaning will affect
one's attitude to a number of issues. For instance, it has been held
that the connectives in deviant Iogics differ in meaning from the
typographically identical connectives of classical logic, so that when
a deviant logician denies 'A v -A', say, what he denies is not,
 Sentence connectioes 31

contrary to appearances, what the classical logician asserts when he

asserts 'A V -A'. One argument for this 'meaning-variance' thesis
( cf. ch. 12 § 1) would be that the meaning of the connectives is given
simply by the axioms/rules and/or matrices of the system (levels (i)
and (ii)), from which it follows that the connectives of a many-valued
logic must differ in meaning from those of z-valued logic, since the
axioms/rules and the matrices differ. Another dispute concems the
appropriateness of the English readings of the connectives, how accu-
rately, for instance, 'and' represents '&', or 'if', '-+'. What seems to
be at issue here is whether the axioms/rules are true/truth-preserving,
and the matrices correct, if thought of as characterising the English
expressions used as readings: whether, since 'A & B' takes t iff 'A'
takes t and 'B' takes t, 'A and B' is true iff 'A' is true and 'B' is true.
This in turn raises another question: does it matter if there is a
discrepancy?

'tonk'
Prior has argued that the meanings of the connectives can-
not derive from the axioms/rules of the system in which they appear,
nor from their truth-tables, but must be given by their English
readings. (If he were right, of course, the 'meaning-variance' view of
many-valued logics, mentioned above, would be refuted.) Prior (1960,
1964) presents a purported reductio ad absurdum of the thesis that there
are 'analytically valid inferences ', inferences, that is, the validity of
which arises solely from the meanings of the logical constants con-
tained in them. According to this thesis, the inference from 'A & B'
to 'A' is analytically valid, for the meaning of '&' is completely given
by the rules of inference of &-introduction and &-elimination. Prior
argues that 'in this sense of "analytically valid" any statement may be
inferred, in an analytically valid way, from any other statement'
(1960 p. 130). Suppose the meaning of 'tonk' to be given by the rules
of inference:
(T1) from 'A' to infer 'A tonk B' ('tonk-introduction')
(T2) from 'A tonk B' to infer 'B' ('tonk-elimination')
Using these rules, A � B, for any A and B:
(1) A assumption
(2) A tonk B (1), (Ti)
(3) B (2), (T2)
32 Philosophy of logics

So, of course, a system with (T1) and (T2) would be inconsistent.

Nothing vital depends on Prior's using rules of inference rath�r th_an
axioms; the axioms 'A-+ (A tonk B)' and '(A tonk B)-+ B , with
the rule to infer B from A -+ B and A (modus ponens, hereafter MPP)
would lead to equally alarming consequences; see Prior 1964 P· 197·
Prior believes himself to have shown that the notion of an analyti-
cally valid inference is a confusion, and that 'an expression must have
sorne independently determined meaning before we can discover
whether inferences involving itare valid or invalid' (1960 pp. 129-30).
Prior argues that since the rules (T1) and (T2) cannot give the
meaning of 'tonk' the meanings of the connectives cannot, in general,
be given by the axioms/rules in which they occur. However, one
might reply that (T1) and (T2) fail to specify the meaning of 'tonk'
for the sufficient reason that they are defective rules. They allow that
A 1- B for any A and B; and no system in which anything is derivable
from anything has any prospect of discriminating acceptable from
unacceptable inferences (cf. Belnap 1961, Stevenson 1961). Prior has
not shown that acceptable rules of inference could not give the
meaning of connectives occurring in them.
I suggested earlier that a major objective of the construction of
formal systems of logic is to give axioms/rules such that the informal
inferences expressible in the language of the formalism which are
intuitively judged valid in the extra-systematic sense are valid in the
system. LT would be so defective that it has no prospect of success in
this enterprise.

AiSI offormalisation
Sorne more needs to be said, however, about the way formal
logieal systems aim to represent intuitively valid inferences. One
could think of a formal logical system as being devised in something
like the following way. Sorne informal arguments are intuitively
judged to be valid, others invalid. One then constructs a formal
language in which the relevant structural features of those arguments
can be schematically represented, and axioms/rules which allow the
intuitively approved, and disallow the intuitively disapproved, argu-
ments. This, of course, is at best a very sketchy 'rational reconstruc-
tion ' and is not intended as detailed, serious history. Still, while
I concede that formal logics have sometimes been devised simply out
of mathematical curiosity, 1 think that something like the procesa
I have described was at work when, for instance, Frege devised hia
 Sentence connectioes 33
Begriffsschrift. Of course, the standard logical languages are now so
familiar that one is no longer very conscious of how and why they
were first constructed. But the same process can be seen in recent
attempts to devise new formalisms for hitherto neglected kinds of
argument; see, for example, the procedure adopted by D. K. Lewis
1973 in devising his analysis of counterfactuals.
Well, supposing that this is roughly right, what is its significance
for questions about the meanings of the connectives? Something,
I think, like this: first, both the syntax and pure semantics (levels (i)
and (ii)) and the informal readings and depraved semantics (levels (iii)
and (iv)) may be expected to contribute to the meanings of the con-
nectives; for part of the object of the enterprise is to have levels (i)
and (ii) adequately represent (iii) and (iv).
However, if formal logic faithfully followed informal arguments in
ali their complexity and vagueness there would be little point in
formalisation; one aims, in formalising, to generalise, to simplify, and
to increase precision and rigour. This means, 1 think, that one should
neither expect nor desire a direct formal representation of all the
informal arguments judged, extra-systematically, to be valid. Rather,
pre-syatematic judgments of validity will supply data for the con-
struction of a formal logic, but considerations of simplicity, precision
and rigour may be expected to Iead to discrepancies between informal
arguments and their formal representations, and even in sorne cases
perhaps to a reassessment of intuitive judgments. One uses intuitive
judgments of sorne arguments to construct a formal theory which
gives verdicts, perhaps quite unexpected verdicts, on other argu-
ments; and one might eventually sacrifice sorne of the original
judgments to considerations of simplicity and generality. These
points relate, of course, to the interdependence of one's judgments of
the correctness of a translation of an informal argument into a formal
Ianguage and one's pre-systematic view of its validity, remarked in
ch. 2. (An example would be the standard rendering of 'Ali Fs are Gs'
as '(x) (Fx-+ Gx)', which is true if its antecedent is false, i.e. if there
are no Fs. It is pretty doubtful whether one would pre-systematically
have agreed that, say, all unicorns are purple, and pretty certain that
one would not have agreed that all unicorns are purple and ali
unicoms are orange.)
One should recognise, then, that a failure on the part of a formal
system to represent ali the knoba and bumps of the informal argu-
menta it aystematises is not necesearily objectionable. On the other
34 Philosophy of logics

hand, one must be wary of assuming that ali adjustments are acce�t-
able · one needs to ask whether the gains in simplicity and generahty
eompensate for the discrepancy. Sorne of the knobs and bumps of
English may be important. These remarks may seem disagreeably
vague; I shall try to make them more specific by considering a couple
of examples.
Why do the usual formal logics have, for example, ' & ', to be read
'and ', but no formal analogue of 'because' or 'but'; and '(3 ... ) ', to
be read 'at least one ', but no formal analogue of 'several' or 'quite a
few'? Two features of the favoured expressions suggest themselves:
truth-functionality and precision.
'&' is truth-functional; and truth-functions are especially readily
amenable to formal treatment - notably, they allow the possibility of
a mechanical decision procedure. This is no doubt why the formal
logician has an analogue of 'and' but none of 'because' or 'but';
'and ', in at least a large class of uses, is truth-functional, whereas the
truth-value of 'A because B' depends not only on the truth-values of
'A' and 'B', but also on whether Bisa reason for A, and the truth-
value of 'A but B' also upon whether the combination of A and Bis
surprising. 'At least one' and 'all' are not truth-functions (though in
the special case of a finite universe they are equivalent to 'Fa V Fb V
... v Fn' and 'Fa & Fb & ... & Fn' respectively). But they are -
unlike 'several' and 'quite a few' - precise. It is notable that one
common reading of '(3 ... )', 'sorne', is vaguer than '(3 ... )' itself;
'at least one' is a more accurate reading. (Other sciences share logic'a
tendency to precisify and idealise ; compare the extensionlesa points
of geometry or the frictionlesa surfaces of mechanics.)
However, while it is clear that truth-functional and precise exprés-
sions are preferable from the point of view of simplicity and rigour to
non-truth-functional or vague expressions, it is not so clear that this
preference is overriding. For non-truth-functional operators - 'L'
and 'M' for 'necessarily' and 'possibly', for example - are used by
formal logicians. Von Wright has suggested (1963) a system with a
sentence connective ' T', to be read 'and then ', which preserves the
temporal sense which 'and' sometimes has in English.! And, while
the standard predicate calculus restricts itself to 'all' and 'at least
I
A related point ís that the standard logical apparatus ís insensitive to
temporal conaiderationa; one is uaually advised to understsnd ita
'p'a and 'q's tenaeleaaly. Sorne proposala for ternporal logica are
discuaaed in ch. 9 § 3.
 Sentence connectioes 35
one', Altham (1971) has devised a logicwith quantifiersfor 'many'
and 'few'.
The desirability of truth-functionality is quite uncontroversial;
but it is equally clear that a logic restricted to truth-functions would
be unacceptably limited. How essential precision is to the formal
logical enterprise is more controversial. Dummett's objection to
allowing epistemic 'logics' as genuinely logics, you will recall, was
that 'k.nows' and 'believes' are inherently vague. Other logicians,
however, have deliberately made use of vague ideas. For example, in
an analysis of counterfactual conditionals (' If it had been the case that
A it would have been the case that B') D. K. Lewis (1973, especi-
ally ch. 4) proposes to employ the admittedly vague idea of similarity
between possible worlds (roughly, 'in all those possible worlds most
similar to the actual world, but in which A, B '); he defends his
compromise with vagueness by observing that the vagueness of the
analysans is unobjectionable since the analysandum is itself vague.
Zadeh, with his 'fuzzy logic' (see e.g. 1975), proposes an even more
radical departure from logic 's traditional concern with precision. I
doubt whether such departures are justified by their results so far, but
much more argument would be needed to show that this doubt is
well-founded; cf. ch. 9 §4.

'&' and 'and', 'v' and 'or', etc.

Of the readings 'not' (of ' - '), 'and' (of '&'), 'or' (of 'v')
and 'if ... then ---' (of '--+'), Strawson has remarked (1952 p. 79) that
'the first two are the least misleading' and the remainder 'definitely
wrong '. Certainly there are discrepancies.
Whereas ' - ' in sentence calculus is a sentence-forming operator
on sentences, 'not' in English may negate either an entire sentence or
else its predica te. This distinction (between 'externa!' and 'interna}'
negation) has been thought important for the understanding of
allegedly meaningless sentences; for instance, it has been suggested
that 'Vírtue is not triangular', like 'Virtue is triangular', is meaning-
less, whereas 'It is not the case that virtue is triangular' is true. It has
also been observed that in colloquial speech double negations do not
always 'cancel out', but may be used as emphatic negatives. 'And ', as
I have already observed, is sometimes used in the sense of 'and then ',
whereas ' & ' is indifferent to temporal order.
Sorne have argued that 'or' has two senses, one inclusive and the
other exclusive; but this would not be too serious a divergence from
36 Philosophy of Iogics

the "v ' of sentence calculus, since an exclusive disjunction could ?e

defined as '(Av B) &-(A & B)'. A second argument for a dis-
crepancy between 'or' and 'V ' appeals to the fact that, in ordinary
speech, it might well be seriously misleading to assert, say, 'John has
the book or Mary has it' if one was in a position to assert 'J ohn has
the book '. However, it could be held that the oddity of the analogue
of the rule ofv-introduction (from 'A' to infer 'A V B') in ordinary
discourse is rather a matter of what Grice has called conoersational
implicature than of validity. According to Grice's account, a speaker
conversationally implies that B if his asserting that A gives his hearer
reason to believe that he believes that B. Sin ce asserting 'A or B'
when one is entitled to assert 'A' (or 'B') contravenes one of Grice's
principies of conversational candour: that one should not make a
weaker when entitled to make a stronger assertion - a speaker who
asserts 'A or B' conversationally implies that he doesn't know
whether it is A or B which is true. This explanation, since it doesn't
concern the truth-values of assertions, allows one to agree that 'A or
B', like 'A V B', is true just in case 'A' is true or 'B' is true, and so
would explain away the apparent discrepancy.
The discrepancies between ·�· and 'if' have generally been
thought the most serious. It seems to be pretty much agreed that if
'If A then B' is true, then 'A� B' is true, but it is highly contro-
versial whether, if 'A 4 B' is true,' If A then B' is true. Faris ( t 962)
argues that 'If A then B' is derivable from 'A -+ B ', so that 'A -+ B'
and 'If A then B' are interderivable, if not synonymous. He assurnes
that a necessary and sufficient condition for the truth of 'If A then R'
is condition E: there is a set S of true propositione such that R is
derivable from A together with S. If 'A --.. B' is true, Faris goes on,
there is a set of true propositions, namely, the set of which 'A 4 B' is
the sole member, from which, with A, B is derivable; so E is satisfied,
and 'If A then B' is true. Faris' argument has been attacked at various
points; understandably, the objectors seem confident that the conclu-
sion is mistaken, but less certain where exactly the flaw is in the
argument (see e.g. Baker 1967, Clark 1971, L. J. Russell 1970). It may
be worth observing that Faris' argument depends heavily on a notion
of 'de:ivability' which spans natural and formal languages in a sorne-
w_hat trre�lar manner. Other writers have argued that the apparent
d1Screpanc1es between ·�· and 'if' are rather a matter of conversa-
tional implicature than of truth-conditions. Their explanation would
go something like this: it's not that 'If A then B' is false if 'A' is false
 Sentence connectioes
37
or 'B' true but rather that when there is no connection between 'A'
and 'B' it would be pointless and misleading to assert 'If A then B'
if one was entitled to assert '-A' or 'B' (see e.g. Johnson 1921,
Moore 1952). Others, again, have suggested that 'if' has severa! uses
in English, one of which may correspond closely to'-+', but the others
of which require a different representation (see e.g. Mack.ie 1973
where nine uses, and six accounts, of English conditionals are
distinguished).
Modero logic offers, in fact, more than one kind of conditional.
The material conditional, which I have been discussing so far, is
truth-functional; and 'A-+B' is true if either 'A' is false or 'B' is
true. So it has the theorems:

A-+(B-+A)
-A-+ (A-+B)
(A-+B) V (B-+A)
These are the 'paradoxes of material implication '. The 'paradoxes'
result if one reads '-+' as 'if' or 'implies'; C. l. Lewis comments that
the third of the above theorems says that if one takes any two sen-
tences at random from a newspaper, either the first will imply the
second, or the second the first. Reflection on these 'paradoxes' led
Lewis to propose a stronger conditional, 'A --3 B' where ' --3' is
strict implication, defined as 'Necessarily (A-+B)'. 'Necessarily
( A --+ B) ', given the standard semantics for modal logic, is supposed
to be true if B is true in all possible worlds in which A is true. Other
implication relations, modelled on strict implication, have been offered
in analysis of counterfactuals (see Stalnaker 1968, D. K. Lewis 1973).
However, strict implication has its own paradoxes; briefly, justas
a false proposition materially implies any proposition, and a true one
is materially implied by any proposition, so an impossible proposition
strictly implies anything, and anything strictly implies a necessary
proposition. ReJevance logicians consequently propose a stricter con-
ditional yet, which requires a relation of relevance between antecedent
and consequent (see Anderson and Belnap 1975 § 1). These logicians
object to calling '-+' 'material implication', as well as to reading it
'If ... then --- '; 'immaterial negation ', they suggest, would be no
more inappropriate. They also extend their critique of truth-
functional logic to disjunction - remember that in the standard system
'A-+B' is equivalent to' -A V B' - arguing that the informal 'or'
is, like 'if ', intensional.
38 Philosophy of logics

One issue here is which conditional best corresponds to 'if'; to

which, of course, the answer may be that different formal conditionals
correspond best to different uses or senses of 'if'. Another is, granted
that material implication, being truth-functional, is the simplest of
the formal conditionals, whether resort to strict or subjunctive or
relevant conditionals brings advantages to compensate for the loss in
simplicity. And here, I think, the purposes for which formalisation is
undertaken may be crucial. If one is concerned only to represent
formally the valid arguments which are used in mathematics, for
example, it might be that a truth-functional implication would be
adequate; though even this is disputable ( cf. Anderson and Belnap
1975 §3). If, on the other hand, one is also concerned to represent
arguments in empirical science, it may be that, since science is ap-
parently deeply committed to dispositions, and so to subjunctive
conditionals (' x is soluble' or 'If x were put in water, it would
dissolve ') one is apt to need something stronger; but this too is dis-
putable (see e.g. Goodman 1955, or Quine 1973 pp. 8-16). So the
significance of the discrepan cíes between 'if' and '--)>' will depend on
the answers to at least two further questions: for what purpose(s) is
the formalisation intended? and, does that purpose require something
stronger than the material conditional? Both - as we shall see in the
course of a closer examination of the strict and relevant conditionals
in ch. 10 - are deep and difficult questions,
 4
Quantifiers

The quantifiers and their interpretation

'(x) Fx' is usually read long the lines of 'For all x, Fx', and
'(3x) Fx' along the lines of 'For sorne x, Fx' or, more accurately,
'For at least one x, Fx'; '( ... )' is generally known as the universal,
'(3 ... )' as the existential, quantifier. A variable inside the scope of
a quantifier, such as 'x' in '(3x) Fx', is said to be bound, a variable
not bound by any quantifier, such as' x' in' Fx', or 'y' in '(3x) Rxy",
to be free. A formula with one or more free variables is called a
(1-, 2- ... n-place) open senience, a formula without free variables a
closed sentence ( or 'o-place open sentence '). So prefi.xing a quantifier,
'(x)' or '(3x)', toan open sentence, such as 'Fx', withjust 'x' free,
yielda a closed sentence, '(x)Fx' or '(3x)Fx'; in general, prefixing
a quantifier binding one of ita free variables to an open sentence with
n free variables yields an open aentence with n - 1 free variables.
Sorne formulations of the predicate calculus have singular temu,
'a', 'b ', 'e' ... etc. as well as variables; these are individual constants,
each denoting sorne specific individual. By dropping a quantifier and
replacing the variable(s) it bound by singular terms, one obtains an
instance of the quantified formula, as e.g. 'Fa � Ga' is an instance
of ' ( x) (Fx � Gx) '. One could think of bound variables as playing a
role analogous to that of the pronouns which, in natural languages,
secure cross-reference, and of singular terms as playing a role
analogous to that of the proper names which, in natural languages,
refer to individuals (but cf. ch. 5).
In modero logic, as I've just indicated, quantifiers and singular
terma belong to quite different syntactical categories, Frege, who
invented quantification theory (Frege 1879; the quantifiers were also
40 Philosophy of logics

devised, independently, by Peirce and Mitchell; see Peirce 1885) laid

great stress on the importance of shifting attention from the subject-
predicate distinction to the function-argument distinction. One con-
sequence of this, essential to the adequacy of the formalism to repre-
sent mathematical argument, is to allow for relations, since one can
have functions of more than one argument. Another, which is most
relevant to our present purposes, is to allow for second-level functions,
the category of the quantifiers. For instance, to say that three-legged
dogs exist, according to Frege, is to say that the concept three-legged
dog is not empty; the quantifier '(3 ... )'isa concept which applies
to concepts, a second-level fu{!ctLoI!_(see Frege 1891, 1892). However,
sorne writers have thought that natural language quantifiers, 'sorne',
'ali', 'every' and so forth, behave very much like names. Russell, for
instance, once tried to treat these 'quantifiers' as 'denoting phrases';
'sorne hoy' was like 'J ohn' except for denoting an 'ambiguous' indi-
vidual; but subsequently he settled for an account in the Fregean
style (Russell 1903; and cf. criticisms in Geach 1962 ). La ter writers,
notably Montague 1973, have pursued the idea of treatipg quantifiers
as name-like (and cf. Hintikka's defence of this approach in 1976, and
the comments by Fogelin and Potts). 1, however, shall have to confine
my discussion to the regular, _' Fregean' quantifiers.
In first-order predicate calculus only 'individual' variables,
'x', 'y' etc., may be bound by quantifiers; in second-order calculi
'F', 'G' etc. may also be bound, as in '(x) (F) Fx'. A sentence
letter, 'p', 'q' ... etc., can be thought of as a limit case of a predicate
letter; 'R' in 'Rxy' is a z-place predicate, 'F' in' Fx' a r-place predi-
cate, and 'p', in 'p', a o-place predicate. So quantífied sentence
calculus, which allows quantifiers binding 'p ', 'q' ... etc., as in
'(p) (p V -p) ', is a kind of second-order calculus. Calculi with dif-
ferentstyles of variable, each varying over different kinds of thing, such
as a formalism with one style of variable for natural, and another for
real, numbers, are known as many-sorted theories.
With the help of the quantifiers numerical statements - "There are
n xs which are F' - can be formulated. "There is at least one x which
is F' is:

(3x) Fx

and 'There is at most one x which is F' is:

(x)(y)(Fy = x =y)
 Quantifiers

(if this isn't obvious, observe that you could read the above formula
as 'If there are two Fs, they're the same '); so 'There is exactly one x
which is F' is:
(3x) (y) (Fy = x = y)
and 'There are exactly two xs which are F' is:
(3x) (3y)(z) (Fz = x = z v y = z)
and so forth.1 Less specific numerical quantífiers, such as 'many' and
'few', have also received formal treatment (Altham 1971), along the
lines of 'at least n' and 'at most n' for variable n.
The distinctions made in the previous chapter between the informal
readings of the symbols of a formal language (level (iii)), their formal
interpretation (level (ii)), and the informal explanation offered of the
formal semantics (level (iv)), applies, of course, to the quantifiers as
well as the sentence connectives. Whereas in the case of the connec-
tives the main controversy centred around the question, how ade-
quately the truth-functional connectives represent their English
analogues, in the case of the quantifiers a key issue concerns their
appropriate formal interpretation, lt is often observed that the
universal quantifier is analogous to conjunction:
(x) Fx = Fa & Fb & Fe & ... etc.
and the exístential quantifier to disjunction:
(3x) Fx = Fa v Fb v Fe V ••• etc.
Indeed, for a theory for which the domain is finite (e.g. where the
variables range over the members of the British government) a uni-
versally quantified formula is equivalent to a finite conjunction and
an existentially quantified formula to a finite disjunction. However,
for a theory with an infinite domain (e.g. where the variables range
over the natural numbers) quantified formulae can be represented
only by infinitely long conjunctions or disjunctions - the ' ... etc.' is
ineliminable. So an acceptable interpretation will have to supply the
1 Part of the logicist programme consisted in the definition of the natural
numbers as certain sets; o as the set of o-membered sets, 1 as the set
of r-rnembered sets, ... nas the set of n-membered sets, for inatance.
Notice how this defines the nominal use of numbers (as in '9 > 7') in
terrns of the adj�ctival use (as in 'There are 9 planeta'), which can,
as I have just explained, be expressed in terma of quantifiers and
identity.
42 Philosophy of logics
interpretati?n
requisite generality. And in fact two distinct styles of
have been offered for the quantifiers. The objectual interpretatton
appeals to the oalues of the variables, the objects over which the
variables range:
'(x) Fx' is interpreted as 'For all objects, x, in the
domain, D, Fx'
'(3x) Fx' interpreted
is as 'For at least one object, x, in the
domain, D, Fx.'
On the 'model-theoretic' approach the domain, D, is a set of objects
assigned as the range of the variables - as it might be, the natural
numbers, persons, fictional characters, or whatever; the 'absolute'
approach, however, requires D to be 'the universe ', i.e. all the o bjects
there are. The restricted domains assigned on the model-theoretic
approach aren't necessarily subsets of 'the universe', though; the
set of fictional characters, for instance, wouldn't be (ch. 5 §.4.). The
substitutiona] interpretation appeals, not to the values, but to the
substituends for the variables, the expressions, that is, that can be
substituted for the variables:
'(x) Fx' is interpreted as' All substitution instances
of 'F ... ' are true'
'(3x) Fx' is interpreted as 'At least one substitution
instance of 'F ... ' ie true.'
The objectual interpretation is championed by - among others -
Quine and Davidaon; the substitutional interpretation by - among
others - Mates and Marcus. Both interpretatione have a pretty long
history; Russell's explanations of the quantifiers, for instance, are
sometimes of the one and sometimes of the other character. However,
it would be fair to say, I think., that the objectual interpretation is
generally thought of as standard, the substitutional as a challenger
whose credentials stand in need of scrutiny. There are, as this suggests,
two possible views about the status of the two styles of interpretation:
that they are rivals, only one of which can be 'right' ; or that they
may both have their uses. I, in the company of e.g. Belnap and Dunn
1968, Linsky 197z, Kripke 1976, shall tak.e the second, more tolerant,
view of the matter.
But this is not to say that nothing hangs on which interpretation is
chosen, On the contrary, the choice may have important philosophical
consequences. I shall not be able to consider all the ramificationa in
detail; I shall, however, sketch an account of the crucial role played
 Quantifiers
43
by the objectual interpretation in Quine's ontological views, This will
be worthwhile both for the sake of illustrating the metaphysical issues
which are apt to be entangled with questions about the interpretation
of formal languages, and also with a view to showing subsequently
(ch. 10) how Quine's ideas about quantification and ontology effect
his attitude to the intelligibility of modal logic.

2 Metaphysical interlude: Quine on quantification and

ontology
Ontology may be characterised as that part of metaphysics
which concerns the question, what kinds of thing there are. Quine's
views on ontology may be seen as the product of two key ideas, the
ideas expressed in two of his best-known slogans: 'to be is to be the
oalue of a variable' and 'No entity toithout identity' (see fig. 2). The
first slogan introduces Quine's criterion of ontological commitment, a
test of what kinds of thing a theory says there are; the second intro-
duces his standards of ontological admissibility - only those entities
should be tolerated for which adequate criteria of identity can be
supplied. I shall concentrate on the first of these ideas, the criterion
of ontological corrunitment, for it is to this that Quine's support for
objectual quantification is primarily relevant.
One or two brief comments on the second idea will, however, be
useful. Criteria of identity give conditions for things of a given kind to
be identical, as: sets are the same if they have the same members, or
aa: two physical objects are the same if they occupy the same spatio-
temporal positíon. Notice that the requirement that only those kinds
of entity be allowed for which criteria of identity can be given is rather
strong (we're pretty confident that there are people, for instance, but
there is a notorious problem about supplying criteria of personal
identity). Quine holds that intensional (meaning) notions are hope-
lessly unclear; consequently, identity conditions stated in intensional
terms are not, by Quine's standards, adequate; and so k.inds of sup-
posed entity which can only be individuated by appeal to meaning -
properties or propositions, for example - are not, by his standards,
admissible.

TM criterion of ontological commitmmt

What, then, is Ouine's criterion of ontological corrunitment,
and how is it connected with the objectual interpretation of the
 "'�
i::
··-�o u.
"'o (1)
o.. �
...
o
.... "'(1)
o..·�
o ....
... &
"'
i:: o
....
o o..
•t
(1)

:E'
o

...e
"'
(1)
ni �
........
e
.2
....
fj
� :o . .o. t.:·=e
(1)
l;¡ §
in
.s
·p
�
E,.;
�
"'
·¡: e �
"' E..
.... Cj
"'
> o
'fi(1) ....C"
s.["' 'o ...
(1)

.s .....:¡ 'o"' ·=
(1)
(1)
"O
.... o ....

·;:::
.§... cil cS (1)
;:l 1

u
"' "'
i:: ...
..e:: ...u
;:l
c3 ]
"O
i::
o
(1)

:"¡¡§ ...E a u
...
<l.)
"'CI (1)

:E' .....:¡ .e <l.)

'o ...... o "'

f_J
...
..e::
i_J � ..o
� o
a
(1)

...
...
<l.)
..e:: .23
� <l.)
·"' :;a"' ..e
1:s e � o
�
...
V
N
� E-<"' e» tlD
¡¡;
 Quantifiers 45
quantifiers? The criterion is put in various, not always equivalent,
forms:
entities of a given sort are assumed by a theory if and only if
some of them must be counted among the oalues of the
variables, in order that the statements affirmed in the theory
be true. ( 1953a p. 103)
to say that a given existential quantification presupposes
objects of a given kind is to say simply that the open
sentence which follows the quantifier is true of sorne
objects of that kind and none not of that kind. (1953a p. 131)

The idea is - roughly- that one tells what a theory says there is by
putting it in predicate calculus notation, and asking what kinds of
thing are required as values of its variables if theorems beginning
'(3x) ... ' are to be true. (So a theory in which '(3x) (x is prime and
x > 1 ,000,000)' is a theorem is committed to the existence of prime
numbers greater than a million, and afortiorito the existence of prime
numbers and to the existence of numbers.) It is obvious that the
criterion applies only to interpreted theories. It is important, also, that
the criterion is to apply only when the theory is expressed in primitive
notation; if quantification over numbers is only an abbreviation for
quantification over classes, for instance, then the theory is committed
to classes but not to numhers. Quine's criterion is a test of what a
theory says there is, not of what there is. What there is is what a true
theory says there is. The refusal to admit intensional entities acts as
a sort of preliminary filter; theories which say there are intensional
entities are not, in Ouine's view, really intelligible, so a fortiori they
are not true.
Quine's explanation of his criterion leaves a good deal to be
desired. AB Cartwright observes in 1954, sorne formulations - the first
one quoted above, for instance - employ such locutions as 'has to',
'must', 'require '; yet these are intensional idioms, and Quine offi-
cially urges that these should be shunned. Quine has explicitly claimed
that his criterion is extensional (1953a pp. 15, 131);1 and sorne formu-
1 He allows that when the criterion is applied to a theory not in
quantificational fonn, an intensional element will be introduced in the
fonn of an appeal to a correct translation of thst theory into predicate
calculus (1953a, p. 131). 'Ibis concession, together with Quine's thesis
of the indetenninacy of translation (1g6oa ch. a) leads to the thesis of
ontological r�/atwity (19Ó8). But my preserrt concern is whether the
criterion itaelf can be put in an a:tensional way.
46 Philosaphy oj logics

lations - the second quoted above, for instance - are given purely in
extensional terms. The question is whether extensional forrnulations
are adequate. Scheffler and Chomsky (1958) argue rather persuas�v.ely
that they are not. The problem is how to understand the condition
'the open sentence which follows the quantifier is true of sorne objects
of a certain kind and none not of that kínd' in the extensional formu-
lation. lf it is read '3 objects of kind k such that the open sentence is
true of them and of non e not of that kind ', it follows that it is impos-
sible to say that a theory says there are objects of kind k, without
oneself saying that there are objects ofkind k, for '3 objects of kind k .. ."
itself involves ontological commitment. But if it is read 'If the open
sentence is true of any objects, it is true of sorne objects of k.ind k and
of none not of that kind ', it follows that any theory which is com-
mitted to anything that doesn't exist is thereby committed to every-
thing that doesn't exist, for if the antecedent is false, the conditíonal
is true. But if the criterion cannot be adequately stated extensionally
it fails by Quine's own standards. Sorneone who didn't share Quine's
scruples, of course, might find the criterion acceptable notwith-
standing its intensional character. But there are further questions to
be asked about the reaaons Quine offers for the criterion.
One important reason why Quine locates ontological commitment
in the variables is that he thinks that the tliminability of singular terms
shows that the ontological cornmitment of a theory cannot reside in
its narnes. This raises two questiona: is Quine right to claim that
singular terma are eliminable] and, is he right to think that if they are,
ontological commitment must be carried by bound variables? 1 shall
take these questions in turn.
Quine's proposal for the elimination of singular terma has two
stages: first, singular terms are replaced by definíte descriptions, and
then the definite descríptions are eliminated in favour of quantifiers
and variables.
(i) In the case of sorne proper names at least, one can supply a
definite description which denotes the same thing: 'the teacher of
Plato' for 'Socrates ', for example. To avoid the difficulties which
rni�ht sometimes be encountered in finding an ordínary predicate
rehably true of just the individual denoted by a name, Quine proposes
the construction of artificial predica tes, and defines 'a' (as,' Socrates ')
as' (1x) Ax' (as, 'the x which socratises'). One could think of the new
predicate, 'A', Quine suggests, as meaning ' = a• ( so ' ... socratises '
rneans ' ... is identical with Socrates '). However this unofficial com-
 Quantifiers 47
ment is not to be thought of as defining the new predicates, for the
whole point of introducing them is to get rid of names altogether; it is
merely an intuitive explanation of predicates which must be taken as
primitive.
(ii) The second stage is to use Russell's theory of descriptions to
eliminate the definite descriptions which now replace singular terms.
This eliminates definite descriptions in favour of quantifiers, variables,
and identity ( details in ch. 5 § 3), thus:
The x which is F is G = df. There is exactly one F and
whatever is F is G
i.e. in symbols:
G((1x)Fx) = df, (3x) ((y) (Fy = x = y) & Gx)
Thus, sentences containing names (as, 'Socrates took poison ') can be
replaced by sentences containing descriptions (' The x which socratises
took poison '), and then by sentences containing only quantifiers and
varia bles (' There is just one x which socratises and whatever socratises
took poison ').
Quine draws the conclusion (1953a p. 13) that, since 'whatever we
say with the help of narnes can be said in a language which shuns
names altogether ', it cannot be names, but must be the quantified
variables, which carry ontological commitment.
The thesis of the eliminability of singular terms has come in for
criticism (see e.g. Strawson 1961). But the real doubt concerns, not so
much the formal feasibility of Quine's proposal, as its philosophical
significance. The fact that Quine can supply an appropriate definite
description to replace a name only by the use of predicates which,
though officially unanalysable, are unofficially explained with the help
of names ('A' means '= a') scarcely reassures one that the
eliminability of singular terms really shows them to be ontologically
irrelevant.
Equally unsettling is the discovery that not only singular terms, but
also quantifiers and variables, are eliminable. In the combinatory logic
due to Schonfinkel and Curry - and, ironically enough, explained by
Quine himself in 196ob - variables are supplanted by predicate
operators called 'combinators '. The predica te operator 'Der ', for
'derelativisation', turns an n-place predicate into an (n- 1)-place
predicate. lf 'F' is a r-place predicate, say ' ... is a dog', 'Der F' is an
o-place predica te - a closed sen ten ce - 'Something is a dog'; if
48 Philosophy of logics

' ... R---' is a z-place predi cate, say ' ... bites --- ', 'Der R' is a I -place
predicate, ' ... bites something', and 'Der Der R' a o-place predicate,
'Something bites something '.
'Jnv ', for 'inversion ', reverses the order of the places of a z-place
predicate; so '((lnvR) ... , ---)' means '---R ... '. 'Ref", for 'reflexive',
tums a z-place predicate into a r-place reflexive predicate; so 'Re/ R'
means ' ... has R to itself'. The procedure is generalised to polyadic
predicates and to compound predications; and the upshot is a transla-
tion without quantifiers of the formulae of quantification theory, in
which inversion provides for permutation of the order of variables,
reflexion for repetition of variables, and derelativisation for
quantification.
Quine concedes that his criterion doesn't apply directly to com-
binatory logic, but observes that it can be applied indirectly, via the
translation of combinatory into quantified formulae. But this is only
to obscure the issue, which is that if the eliminability of singular terms
were a good reason for denying that they carry ontological cornmit-
ment, the eliminability of quantifiers should presumably be as good
a reason for refusing them ontological significance.
This makes clearer, I think, the very considerable importance which
Quine's insistence on the objectual interpretation of the quantifiers has
for his ontological criterion. Though the same theory could be
expressed using singular terms as well as quantifiers, or combinatory
operators instead of quantifiers, its quantificational form reveals its
ontological commitments most transparently, Quine thínks, because
a senience of the form '(3x) ... ' says that there is something whirh ...
To insist on the correctness of the criterion ... is, indeed,
merely to say that no distinction is being drawn between
the 'there are' of 'there are universals ', 'there are
unicorns ', 'there are hippopotami ' and the 'there are' of
'(3x) ', 'there are entities x such that ... '
And deviation from the objectual interpretation would threaten the
criterion:
To contest the criterion ... is simply to say either that the
familiar quantificational notation is being used in sorne
new sense (in which case we need not concern ourselves]
or else that the familiar 'there are' of 'there are universals'
et al. is being used in sorne new sense (in which case again
we need not concern ouraelves). (1953a p. 105)
 Quantifters 49
On the objectual interpretation '(3x) Fx' means that there is an
object x, in the domain, D, which is F. Now, if one takes D to be 'the
universe' - everything there is- which, it seems, is what Quirie
assumes, then, indeed, '(3x) Fx' means that there is an ( existent, real)
object which is F; cf. Quine's use of 'entity' in the passage just
quoted.
If '(3x) Fx' means "There is an (existent) object which is F' then,
if it is a theorem of a theory that (3x) Fx, then that theory says that
there is an object which is F; and if one says that there are Fs, one is
committed to there being Fs. The objectual reading of the quantifier
does indeed locate ontological commitment in the bound variables of
a theory. Perhaps I may rewrite Quine's slogan: to be said to be is to
be the value of a variable bound by an objectual quantifier; it's less
memorable, but truer, that way ! Notice, though, that Quine's criterion
now begins to look oddly obligue: as if one discovered that a theory
which says there are so-and-so's is ontologically committed to so-and-
so's by first translating it into predicate calculus notation, and then
appealing to the objectual interpretation of the quantífiers to show
that its existential theorems say that there are so-and-so's.
The serious work has to be done in deciding which ostensibly
existential assertions of a theory need remain in primitive notation,
and which are eliminable by suitable paraphrase. An example would
be Morton White's proposal (1956) to reduce "There is a possibility
that James will come', which seems to assert the existence of possí-
bilities, to 'That James will come is not certainly false', which does
not. (There are still tricky philosophical questions to be asked about
the significance of paraphrase here, though I shan't pause over them
now; but see Alston 1958 for a critique of the idea that paraphrase
may elimínate ontological commitment, and cf. Lewis 1973 ch. 4,
where it is assumed that paraphrase preserves ontological com-
mitment.)

Substitutional quantification and onto/ogy

The substitutional interpretation does not give a negative
answer to ontological questions; rather, it postpones them. On the
substitutional account, '(3x) Fx' means 'Sorne substitution instance
of 'F ... ' is true'; questions of existen ce now depend u pon the condi-
tions for the truth of the substitution instances. lf, for instance, 'Fa'
is true only if 'a' is a singular term which denotes an ( existent) object,
then there will have to be an object which is F if '(3x) Fx' is to come
50 Philosaphy of logics

out true; but it is not inevitable that the truth-conditions for t�e
appropriate substitution instances will bring an ontological comm1t-
ment. An example: the presence in predicate calculus of theorems
such as:
(3.x)Fxv-Fx
which on the objectual interpretation says that there is at least one
object which is either F or not F, i.e. that there is at least one object,
is embarrassing if one thinks it oughtn't to be a matter of logic that
anythi"ng exists. Would a substitutional interpretation avoid the exis-
tential commitment of the embarrassing theorems? Well, on that
interpretation the theorem means that:
At least one substitution instance of 'F ... V - F ... ' is true.
If only names which denote an object are admitted as substituends,
then on this interpretation too predicate calculus will require at least
one object. But if non-denoting terms, like 'Pegasus ', are allowed as
substituends, then ontological commitment may be avoided. This
illustrates the way that substitutional quantification defers ontological
questions, shifting them from the quanti.fiers to the names. Quine is
apt to suggest that this relocation of existential questions is a deplor-
able evasion of metaphysical responsibility ! But I shall suggest
subsequently that it may have advantages.
Since on the substitutional reading '(3x) Fx' means that at least one
substitution-instance of 'F.. .' is true, if this metalinguistic quantifier
is interpreted objectually, it will be committed to the existence of the
appropriate expressions, the substitution-instances. But this will not
be so if it too is interpreted substitutionally.

3 The choice of interpretation

Is one interpretation of the quantifiers the 'correct' one? Or
may one choose between the two according to one's purposes? And
if so, what are the strengths and weaknesses of each ?

Substitutional quantifiers and truth'

. Which interpretation of the quantifiers is adopted will make a
difference to the definition of truth for quantified sen ten ces. 1'11 he brief
now, since there will he more sustained discussion in ch. 7 §§ 5 and 6.
If the_ quantifiers are interpreted suhstitutionally, then the truth of
quan�1fied forrnulae can he defined directly in terms of the truth of
atomic formulae (as, "(3x) Fx' is true iff sorne suhstitution instance
 Quantifiers

of 'F ... ' is true'); if the quantifiers are interpreted objectually the
defi.nition of truth will be less direct. Now Tarski proposes, as a
'material adequacy condition' on definitions of truth, that any
acceptable defi.nition must have as consequence all instances of the
'(T)-schema': 'S is true iff p ', where 'S' names the sentence 'p '; and
Wallace 1971 fears that if a substitutional interpretation is adopted
the defi.nition of truth will not meet this requirement. But Kripke has
argued ( 1976) that Tarski's condition isn't violated; and anyway there
may be reservations about the requirement itself. So I shall assume
that the substitutional interpretation isn't objectionable on this score.

Too few names?

This leaves the question of whether the substitutional and
objectual interpretations are always equally suitable. The answer is,
pretty clearly, no. lt is of course a requirement on an acceptable
interpretation that the theorems of the theory being interpreted come
out true. The substitutional interpretation will obviously make exis-
tentially quantified wffs true only if suitable substituends are avail-
able. For instance, '(3x) (Fx V -Fx)' is a theorem in first-order predi-
cate calculus; in a formulation with quantifiers but no singular terms,
however, the substitutional interpretation could not, for want of
appropriate substitution-instances, make such a wff come out true
(so that the elimination of singular terms will preclude a sub-
stitutional account). Another situation in which a substitutional
interpretation would be precluded would be a formal system in which
'(3x) Fx' was a theorem but for every substitution instance '-Fa'
was provable; for the substitutional interpretation couldn't make the
quantified theorem true without making at Ieast one of its instances
true (this possibility is discussed by Quine 1968; and cf. Weston
1974).

Tense
Its supporters argue, however, that sometimes the substitu-
tional interpretation offers advantages over the objectual. F or instan ce,
Marcus suggests that a substitutional reading will avoid difficulties
about tense. Strawson had asked (1952 pp. 15�1) how to represent
'There was at least one woman among the survivors': 'There is (was?)
at least one x such that x is (was?) a woman and ... '? 1 think it
unlikely that this problem will be solved by a substitutional reading:
'At least one aubstitution inatance of ' ... ' is (was?) true'. lt is true,
52 Philosophy of logics

and important, that tense matters to the (in)�di� of ��ormal argu-

ments and that the usual logical apparatus is insensitrve to tense
(cf. ch. 9 §3); but the substitutional interpretation doesn't seem to
help.

Modality
Marcus also suggests, with, 1 think, more justice, that
substitutional quantification could resolve sorne problems about the
interpretation of modal predicate logic. From the presumably true
sentence:
Necessarily (the Evening Star = the Evening Star)

by presumably valid predica te calculus reasoning there follows:

(3x) Necessarily (x = the Evening Star)
that is, on the objectual interpretation,
There is at least one object, x, such that necessarily x is
identical with the Evening Star.
But this is hard to understand; indeed, Quine bases u pon it an argu-
ment that the whole enterprise of modal predicate logic is misguided;
for what is this object which is necessarily identical with the Evening
Star? Not the Evening Star; for that is the Morning Star, and the
Morning Star isn't necessarily, but only contingently, identical with
the Evening Star. But Quinc's awkward questions are avoidcd by
reading the bothersome sentence substitutionally:
At least one substitution instance of 'Necessarily
( ... = the Evening Star)' is true
when it seems (since 'Necessarily (the Evening Star = the Evening
Star)' is true) unproblematically true.
The substitutional interpretation also seems to offer certain
advantages when one tums to second-order quantification.

Second-order quantification
If, as on the objectual interpretation, '(3x) ... ' says that there
is an object such that ... , and '(x) .. .' that for all objects, ... , then it
is to be expected that the appropriate substituends for bound vari-
ables should be expressions whose role is to denote objects, that is to
say, singular terms. Quine, indeed, sornetimes defines a singular term
 Quantifiers 53
as an expression which can take the position of a bound variable. On
the substitutional interpretation, however, quantification is related
directly not to objects, but to substituends; and so there is no parti-
cular need to insist that only expressions of the category of singular
terms rnay be bound by quantifiers. The substitution class could be
the class of singular terrns, but could equally be the class of predicates,
or the class of sentences ... etc.
On the objectual interpretation, therefore, just as a first-order
quantification like:

1. (3x) Fx
says that there is an object (individual) which is F, a second-order
quantification, like:

2. (3F)Fx
says that there is an object (property) which x has,
and:

3· (3p)(p--+-p)
says that there is an object (proposition) which materially implies its own
negation. The natural restriction on the substitution class obliges one
to construe bound 'F' and 'p' as syntactically like singular terrns; it
will be observed that this strains the reading, since if 'F' is a singular
terrn, 'Fx' rnust be read as 'x HAS F' to make a grarnrnatical sentence.
And the interpretation in terrns of objects obliges one to regard
second-order quantification as cornrnitting one to (abstract) objects.
Not caring for such supposed objects as properties or propositions,
but obliged by the objectual interpretation to allow that second-order
quantification would commit hirn to their existence, Quine prefers
not to indulge in second-order quantifi.cation at all, but to confine
hirnself to first-order theories.
However, with the substitutional interpretation one is not restricted
to singular terms as substituends; and while in the case of first-order
quantification singular terrns would be suitable, in the case of
second-order quantification as in 2 and 3 predicate letters or sentence
letters, respectively, would be the appropriate substituends. 2 will
say that sorne substitution instance of ' ... x' is true, 3 that sorne
substitution-instance of ' ... �- ... 'is true. Now it is no longer neces-
sary to strain the reacling of the bound variables to make them
54

�
�
E..;
>.
� bO
o
� o...
t t
e::o

-s.
u, -e
� $ tí11) e::<si
� �
i::
:E' t"'
'"4
o �...;
s E-,, o
::,

\,,)
l"l
�
¡:x:¡ bO
a ti:
 Quantifiers 55
name-like, and of course there is no commitrnent to intensional
objects, since there is no commitment to objects (see fig. 3).
These thoughts have sorne relevance, 1 think, to metaphysical
issues. Nominalists admit the exístence only of particulars, whereas
realists, or Platonists, also allow the reality of universals. C. S. Peirce
thought that the influence of nominalism on the history of philosophy
since Ockham had been so pervasive that nominalism and 'nomin-
alistic Platonism ', the view that universals are a sort of abstract
particular, had come to seem the only alternatives; rejecting both, he
urged a realism which, rather than assimilating them, allowed the
difference between particulars and universals. Now if only names are
substitutable for bound variables one is, as it were, obliged to choose
between a kind of nominalism (allow only first-order quantification,
with variables replaceable by names of particulars: Quine's position)
anda kind of nominalistic Platonism (allow second-order quantifica-
tion, with variables replaceable by names of abstract objects, prop-
erties or propositions: Church's position). Allowing substituends
from other syntactical categories, however, gives a third option; which
may well be, from a metaphysical point of view, appealing.
Other important issues also depend upon the provision of an
acceptable interpretation of second-order quantifiers. One of these is
the feasibility of the redundancy theory of truth; and the considera-
tions of the present section will be highly pertinent when I come, in
ch. 7 §7, to discuss that question in detail.
 5
Singular terms

I Singular terms and their interpretation

Sorne formulations of the predicate calculus employ singular
terms ('a', 'b' ... etc.) as well as variables. lf the quantifiers are to be
ínterpreted substitutionally, of course, the presence of singular terms
in the language to supply the appropriate substitution instances is
essential. What, in informal argument, corresponds to singular tenns
in formal logic? Singular terms are usually thought of as the formal
analogues of proper names in natural languages. (Where the variables
range over numbers, the numerals would correspond to the singular
terms.) The formal interpretation of singular terma assigns to each
a speeific individual in the domain over which the variables range ; and,
in natural languagea, proper names are thought to work in a similar
way, each standing for a particular person (or place or whatever).
So while in the case of the quantifiers the main controversy sur-
rounds the question of the most suitable formal interpretation, in the
case of singular terms the problems centre, rather, on the under-
standing of their natural language 'analogues '. The formal interpre-
tation of singular terms in straightforward extensional languages is
uncontroversial; however, rival views about how to understand proper
names in natural languages have been used in support of altemative
proposals about the formal interpretation of singular terms in less
straightforward, e.g. modal, calculi. Among the disputed questions
about how, exactly, proper names work are, for instance: just which
expressions are bona fide proper names? For example, are 'names' for
mythical or fictional entities ('Pegasos', 'Mr Pickwick') to count?
And if so, what is one to say of the truth-value of aentences con-
taining such 'non-denoting' names] Eapecially, how ia one to explain
 Singular terms 57
the in tui ti ve truth of negative existentials like 'Pegasus never
existed '? If the role of names is purely and sim ply to denote an indi-
vidual, how can a true identity statement (like 'Cicero = Tully') be
informative, and how can substitution of one name for another
denoting the same individual sometimes change the truth-value of a
sentence (as, the presumably true 'One needs no knowledge of Roman
history to know that Tully = Tully ', and, the presumably false
'One needs no knowledge of Roman history to know that Tully =
Cicero ')? A central issue, and the one on which I'll concentra te, is
whether proper names have meaning (' sense ', 'connotation ') as well
as denotation, and if so, what meaning they have. Those who think
that proper names do have meaning generally associate their meaning
more or less closely with the meaning of co-designative definite
descriptions. The first view, that names have denotation but no
meaning, makes for a sharp differentiation between proper names
(' Socrates', 'Bismarck' etc.) and descriptions (' The teacher of Plato',
'The Chancellor responsible for the unifi.cation of Germany '),
whereas the second view sees names as rather like descriptions. This
leads to a second key question: how do definite descriptions work?

2 Names
I'll take the first question first. The alternatives are sketched
in table 1.
Sorne preliminary comments about the distinction between proper
names and definite descriptions may be in order. Ordinarily the dis-
tinction isn't hard to draw, but there are sorne expressions which are
tricky to classify; for example, although the Morning Star isn't a star,
but a planet (Venus), it is still called 'the Moming Star', so that 'the
Morning Star' seems to have become more like a name and less like
a description than, perhaps, it was originally; the capital letters may
be significant of this intermediate status. (The University of Warwick
isn't at Warwick, either.) Furthermore, not all names are of the same
kind; logicians are apt to take as their examples names of persons or,
less commonly, places, but there are also titles of books, brand
names, trade names ... ( and notice how trade names have a ha bit of
tuming into common nouns, and even verbs, e.g. 'hoover'). It is also
worthy of attention that logicians are fondest of the names of famous
people ('Aristotle', 'Napoleon' etc.); and salutary to recall that there
are no doubt many Aristotles and Napoleons, and only a background
of shared infonnation to make us all think of the same one. Drawing
58 Phiwsophy of logics

. ?
TABLE 1. Do proper names haoe meaning as toell as denotation ·

Yes No
Frege proper names have Mill proper names
Russell the sense of sorne have denotation
(Quine) co-desígnative definite but no connota-
description known to tion
the speaker Ziff and are not part
of the language
Wittgenstein proper names have
Searle the sense of sorne Kripke proper names
indeterminate sub- are' rigid
set of sorne open- designators';
ended set of causal account
co-designative of the correct
descriptions use of names
Burge proper names as
(Davidson) predicates

attention to the variety of kinds of proper narne may induce sorne

caution about assuming that there is such a thing as th« way in which
proper names work.

Na�, a, puuly dmotativt

One view is that proper names, by contrast with definite
descriptions, are, so to speak, mere labels; they serve simply to stand
for a person, place, or thing. 1 don'tthinkthatpeople who take this view
mean, or need, to deny that in the case of personal names there are
conventions about which names are given to males and which to
females, for example, about a child's taking his father's sumame, and
so forth; nor that narnes have a 'sense' derived from their etymology,
as,' 'Peter' meaos 'a rock". The name a person has may, in virtue of
conventions of the kind mentioned, convey sorne information about
him; what is being denied, rather, is that the name describes him.
According to Mill 1843, proper names have denotation but no
connotation, that is, no meaning. Ziff 1900 subscribes to aomething
like thia view; proper names have no meaning, and, indeed, in sorne
 Singular terms 59
sense are not even part of the language. Another writer who denies
that proper names have meaning is Kripke, who in 1972 sketches an
account both of the semantic and of the pragmatic aspects of proper
names. Proper names are 'rigid designators'; that is to say, they have
the same reference in all possible worlds. For example, the name
'Aristotle' designates the very same individual in all possible worlds,
whereas the definite description 'the greatest man who studied under
Plato', though it designates Aristotle in the actual world, may desig-
na te other individuals in other possible worlds; for it is possible that
Aristotle should not have studied under Plato. The idea is this:
a proper name simply designates a specific individual, and since it
doesn't describe that individual, it designates it, not in virtue of its
being the individual which ... , but simply qua that specific indi-
vidual; and so, no matter how different the individual that the name
designates were to be from the way it actually is, the proper narne
would still designate that individual - and this is what Kripke means
by saying that it designates the same individual in all possible worlds.
(Kripke would apparently identify an individual by virtue of its
origin; in the case of persons, by their date of birth and parentage.)
Kripke doesn't deny that the reference of a name may be fixed by
means of a definite description, that one might introduce a name to
denote the referent, in the actual world, of sorne definite description,
fixing the reference of 'Fido', say, as the fi.rst dog to go to sea. What
he denies is that the definite description gives the sense of the name;
Fido, presumably, might not have been the first dog to go to sea, but
in a possible world in which he isn't, while 'the first dog to go to sea'
designates a different dog, 'Fido' still designa tes Fido, i.e. the dog
that, in the actual world, was the first to go to sea.
The semantic account is supplemented by a causal explanation of
the pragmatics of naming; the object is to explain how a speaker can
use a narne correctly even though he is quite unable to give a descrip-
tion that applies uniquely to the individual named - who knows of
Feynman, for instance, only that he is a physicist. According to
Kripke, a speaker uses a name correctly if there is a suitable chain of
communication linking his use of the name with the individual
designated by the name in an initial 'baptism '. A baby is born, his
parents name him, other people meet him, he becomes a physicist,
writes papers which other people read and write about ... and so
forth; then a speaker uses the name ' F eynman' correctly, to re fer to
Feynman, if bis use of this name is causally linked in an appropriate
6o Philosophy of logics
way to the chain of communication which goes back to Fe��n
himself. Of course, there needn't literally have been an initial
baptism, and the chain of communication may be very long indeed, as
in our use of 'Julius Caesar' for instance. AB Kripke is aware,
'causally linked in an appropriate way' stands much in need of amplifi-
cation; sin ce he does not o:ffer any further explanation, there is not
yet a guarantee that the causal account won't turn out either trivial
or false.
The connection between the pragmatic and semantic strands of
Kripke's account is, presumably, that his criteria for correct use of
a name malee no appeal to the speaker's knowledge of or beliefs about
the individual designated, but require only that his use of the name
be appropriately connected, causally, with that individual; this is con-
sonant with the insistence, in the semantic account, that a name only
designates and does not describe. However, if the reference of a narne
may, as Kripke allows, be fixed by means of a definite description,
a gap could open between the semantic and the pragmatic accounts;
for if I fix the reference of a proper name by means of a definite
description which, in fact, though I don't know it, designates nothing
(e.g. if I decide to call the man who stole my suitcase 'Smith ', when
actually my suitcase hasn't been stolen, but only moved by a porter)
there can't be an appropriate causal chain to the bearer of the name,
since there is no bearer.
lt follows from the theais that proper names are rigid designators
that all true identity staternents of the form 'a • b ', where 'a' and 'b'
are names, are necessary. If 'a' and 'b' are names, and 'a • b' ia
true, so that 'a' and 'b' designate the same individual in the actual
world, then, since both names, being rigid designators, designare the
same individual in ali possible worlds, 'a == b' is true in ali possible
worlds, that is to say, it is necessarily true.

Nemes assimilated to descriptions

Now, it was precisely because of a problern about identity
statements that Frege introduced ( 1892a) his distinction between sense
(Sinn) and reference (Bedeutung), and argued that proper names have
sense as well as reference. How, Frege asks, can:
(i) a = b
differ in 'cognitive value' from, i.e. be more informative than:
(ii) a = a
if a ú b? His answer is that, while, if a is b, the reference of 'a' is the
 Singular terms

same as the reference of 'b' (they stand for the same object), the sense
of 'a' is different from the sense of 'b ', and this difference accounts
for the greater informativeness of (i) over (ii).1
Frege explains the informativeness of true statements of the form
'a = b' as arising from the difference of sense of the names 'a' and 'b '.
How would Kripke, who doesn't admit that names have senses, and
according to whom ali true identities are necessary, explain it? His
explanation is that, though statements of the form 'a = b' are neces-
sary, not ali necessary statements can be known a priori; that is, it
may be a discovery, for ali that it is necessary, that a is b. For instance,
the name 'Hesperus' was given to a certain heavenly body seen in the
evening, and the name 'Phosphorus' to a heavenly body seen in the
morning; both are rigid designators, and designators, as it turns out,
of the very same heavenly body (the planet Venus); but astronomers
had to discooer, and didn't know a priori, that they designate the same
heavenly body. (Kripke comments that there is nothing especially
remarkable about one's knowing of sorne proposition that it is neces-
sary iJ it is true, and yet not knowing whether it is true; Goldbach's
conjecture would be an example.)
Frege, however, regards proper names as having sense as well as
reference. By 'proper name' Frege understands both ordinary names
and definite descriptions (he says that a name is any expression that
1 Although the distinction is originally introduced specifically for names,
it is extended to apply to predicates, and then, on the principie that the
aense (reference) of of a compound expression is to depend on the
aense (reference) of its parts, to sentences. Thus:
expreuion seme referenc«
proper narne meaning of the narne object
preclicate meaning of the predicate expression concept
sentence proposition truth-value
The reference of a sentence must be ita truth-value, Frege argues,
since if sorne component of a sentence is replaced by another with
a clifferent sense but the same reference (as 'The Morning Star is
a planet '/' The Evening Star is a planet ') it is the truth-value that
remains unchanged, Always strongly anti-psychologistic, Frege stresses
that the sense, or meaning, of an expression is to be clistinguished from
the idea that may happen to be associated with that expression; so
when he saya that the sense of a sentence is the 'thought' (Gedanke) it
expresses, he means 'proposition' rather than 'idea'. In 'oblique'
contexts, Frege adds (Le. intensional contexts, e.g, inclirect speech),
sentences have, not their customary, but an 'indirect' reference, the
direct reference being the customary sense, i.e, the proposition
expressed. So in 'Tom said that Mary would come', the reference of
'Mary would come' is not its truth-value, but the proposition that
Mary would come.
62 Philosophy of logics

refers to a definite object, though in fact he also envisages the P.ossi-

bility of names, like 'Odysseus', that don't denote a real object).
And he equates the sense of an ordinary name with the sense of
a definite description which refers to the same object. Which co-
designative definite description? Apparently (1892a p. 58n, and cf.
1918 p. 517) one that the speaker has in mind, or happens to know.
Frege realises that this has the consequence that different people may
attach different meanings to a name, depending on what they know
about the person named; he comments that such variations of sense,
though they should be avoided in a perfect language, are tolerable so
long as the reference remains the same. In view of the fact that one of
the objections he frequently brings against identifying the sense of an
expression with the associated idea is that this would mean that the
sense varied from person to person, this tolerance is surprising.
Russell, like Frege, identifies the meaning of ordinary proper names
with the meaning of sorne relevant definite description (though, as will
appear below, he differs from Frege both in his view of meaning, and
in his view of how definite descriptions, in their turn, should be
explained). And, again like Frege, Russell saw that it followed that
names have a different meaning for different speakers.
Russell also distinguished, however, a special category of logically
proper names: these are expressions whose role is purely to denote
a simple object, and the meaning of which is the object denoted ( so in
the case of logically proper names Russell equates meaning and
reference). In Russell's version of logical atomism, the 'simple objects'
are 'objects of acquaintance', so logically proper names denote
objects of acquaintance; according to Russell we are dircctly
acquainted not with ordinary objects, people, etc., but only with
sense-data. So the only expressions he allows to be logically proper
names are 'this ', 'that' and ( during the period in which he believed
in a directly introspectible ego) '1 '. No ordinary proper names are
logically proper names, for no ordinary proper names denote objects
of acquaintance. Russell sometimes uses 'acquaintance' in a more
commonsensical fashion, distinguishing between persons and places
one has actually met or visited and those one has only heard about,
and treats names of persons or places with which one is acquainted in
this sense as logically proper names. But it is clear that this is a loose
usage, and that the strict theory according to which no ordinary
names are logically proper names is the one to be taken seriously.
Identifying the meaning of a proper name with sorne co-designative
 Singular terms

description known to the speaker has, as Frege and Russell realised,

the uncorrúortable consequence that the meaning of the name is vari-
able between speakers. This difficulty could be avoided by identifying
the meaning of the name, instead, with the set of all descriptions true
of the bearer. But this has the unhappy consequence that every true
statement of the form 'a is (was) the person who ... ', where 'a' is
a proper name, is analytic, and every false statement of that form,
contradictory, for 'a', on this view, just means 'the person who ... '
for all descriptions true of the bearer. This problem, in turn, could be
deflected by loosening the connection between the meaning of the
name and the set of descriptions of its bearer. An idea of this kind is
found in the Philosophical lnvestigations (Wittgenstein 1953), where
it is suggested that a name hasn't a fixed, unequivocal meaning, but is
loosely associated with a set of descriptions; by 'Moses' one may
mean the man who did most, or much, of what the Bible relates of
Moses, but how much, or what part, of the story has to be false before
one says that there was no such person as Moses, is not fi.xed (§79).
Something like this is also proposed by Searle 1969: while no one of
what are thought of as established facts about a need necessarily be
true of him, nevertheless the disjunction of them must be (p. 138). It
isn't analytic that Moses was found in the bullrushes, nor that he led
the Israelites out of Egypt, nor ... etc., but it is analytic, according to
Searle, that either Moses was found in the bullrushes, or ... etc. Like
Wittgenstein, Searle stresses that how many of the disjuncts may be
false before it is true to say that a never existed is indeterminate.
Thus far, then, there are these alternatives:
let d1 ... d,. be all the descriptions (supposedly) true of a
then either:
the meaning of 'a' is sorne member(s) of the set
or:
the meaning of 'a' is the conjunction d1 & d2 & •.. d¿ of
all the members of the set
or:
the meaning of 'a' is sorne subset of the set of descriptions,
there being indeterminacy about which, and how many,
of the d¿ to include.
These proposals identify, or more loosely associate, the sense of
a proper name with that of related definite descriptions. Another
64 Philosophy of logics

proposal, in somewhat similar spirit, is offered by Burge 1973 (�nd

endorsed by Davidson); on this account, instead of a narne's ?emg
regarded as abbreviating a definite description, it is held to be, itself,
a predicate. Burge points out that proper names rarely, in fact, stand
for a unique object, that they take plural endings ('there are three
J acks in the class ') and the definite and indefinite articles (' the J ack
who wrote this ', 'there is a Mary in the class but no Jane'). Burge is
concemed with literal, not with metaphorical, uses of names, with
'Callaghan is a James' rather than 'Callaghan is no Churchill '. 'J ack
is tall', on Burge's account, is best regarded as a sort of open sentence,
with 'Jack' as a predicate governed by a demonstrative, 'that J ack is tall'
(like 'that book is green ')the reference of which is fixed by the context.
Regarded, thus, as a predicate, 'J ack' is, according to Burge, true of
an object just in case that object is a Jack, that is to say, just in case
that object has been given that name in an appropriate way. Burge's
account has sorne affinities with a suggestion to be found in Kneale
1962a, that the meaning of the narne 'a' is 'person called 'a".
Kripke objects to Kneale's proposal that it is viciously circular. Burge,
however, points out that his treatment of proper names as predicates
could be supplemented by a theory of naming, a theory which would
fill in the conditions in which an object is a Jack, the conditions, that
is, in which it is true to say that it 'has been given the narne 'J ack' in
an appropriate way'. There is no reason, of course, why the kind of
causal relation that Kripke stresses shouldn't have a role to play at
this level.
There is a tendency to think of proper names as, 110 to speak, the
means by which language gets its most direct grip on the world ; and
perhaps for this reason there is strong motivation to give a neat and
tidy picture of the way naming works. In the theories I have sketched,
two kinds of picture emerge of the connection between-names and the
individuals named: the purely denotative, or 'harpoon ', picture, and
the descriptiva, or 'net', picture. (1 derive the useful metaphor from
Fitzpatrick; but I've changed his 'arrow' to 'harpoon' to give a place
to the role of the causal chain of naming in Kripke's account.) I've
already suggested that ordinary proper names in natural languages
are very various, and that they work against a background of shared,
or partly shared, information, or misinformation. Sorne confirmation
for my suspícion that there may not be only one way in which ali
narnes work may be found in the way that the two pictures, officially
presented as rivals, seem actually to complement each other: the
 Singular terms 65
harpoon picture explains how we can manage to talk about someone
even though we are ignorant, or misinformed, about him - he' d slip,
as it were, through our net; the net picture how we can talk, without
confusion, about one of several or many people of the same name.
The details of the net picture will depend, obviously, on what
account is given of the descriptions that, on this view, are associated
with a name: It is to this question that I tum next.

3 Descriptions
Though Frege and Russell both equate the meaning of
(ordinary) proper names with that of corresponding definite descrip-
tions, they give quite different explanations of the way these
descriptions work.
According to Russell's theory of descriptions (Russell 1905) defi.nite
descriptions, such as 'the tallest mountain in the world', are 'incom-
plete symbols ', that is to say, they are contextually eliminable. Russell
gives, not an explicit definition enabling one to replace a definite
description by an equivalent wherever it appears, but a contextual
definition, which enables one to replace sentences containing defi.nite
descriptions by equivalent sentences not containing definite descrip-
tions:
El (1x) Fx = df. (3x)(y)(Fy = x = y)
i.e. 'the F exists' means 'rhere is exactly one F', and
G((1x) Fx) = df. (3x) (y) (Fy � x = y) & Gx
i.e, 'the F is G' means 'there is exactly one F and whatever is F is G '.
The latter will consequently have two 'negations':
- (3x) (y) (Fy = x = y) & Gx
i.e. 'It is not the case that (there is exactly one F and whatever is F
is G)', and:
(3x) (y) (Fy = x = y) &-Gx
i.e. 'There is exactly one F and whatever is F is not G'.
Of these, only the first is the contradictory of "The F is G'; the latter
is its contrary. (In general, in fact, it has to be indicated what scope
a definite description has when it is in a compound sentence.)
Russell rernarks that the grammatical form of sentences like "The
tallest mountain in the world is in the Himalayas' is misleading as
to their logical form: what he means is that, whereas the English
66 Philosophy of logia

sentence contains an expression, 'the tallest mountain in the world ',

which looks as if its role is to designate an object, its formal representa-
tive contains no singular terms at all, but only bound variables, predi-
cates and identity. And this enables Russell to deal with the problem
of definite descriptions, such as 'the present King of France', whích
aren't true of anything. The problem, as Russell sees it, is this: if "The
present King of France is bald' is, logically, as it is grammatically,
a subject-predicate sentence, then its subject term, 'the present King
of France ' must be a logically proper name, the meaning of which is
the object it denotes; but since there is no present King of France,
either 'the present King of France' denotes an unreal object, or else
it denotes nothing, and hence it, and consequently the entire sentence,
is meaningless, Reluctant to accept either of these conclusions,
Russell solves the problem by denying that 'The present King of
France is bald' is, logically, of subject-predicate form at ali; logically,
it is an existential sentence. In the end, then, Russell denies that any
ordinary proper names (or definite descriptions) are properly repre-
sented by the singular terms of his formal language; this privilege is
restricted to logically proper names,
Russell regarded his theory of descriptions as ontologically liber-
ating; for it freed him of the necessity to admit a domain of unreal
entities as the denotation of apparently non-denoting names. (See his
criticisms (1905) of Meinong, who did admit non-existent objects,
and cf. §4 below.) After developing the theory, in fact, Russell quite
severely pruned his ontological commitments. Early on, in revolt
against Bradley's monism, he had admitted a luxuriantly pluralist
ontology, belíeving, as he put it, in everything Bradley disbelieved in;
but subsequently, influenced by Whitehead's advocacy of Ockham's
razor, and equipped with the theory of descriptions which freed him
ofthe need to admitan object as denotation to secure the rneaningful-
ness of every apparent name, he dismissed not only Meinongian
objects, but classes, properties, even physical objects, as 'fictions '.
(Cf. Quine 1966b for details of the development of Russell's onto-
logical views.)
. Quine's proposal (discussed in ch. 4 §2) to eliminate singular terms
m favour of co-designative definite descriptions is clearly in the spirit
of Russell's approach to proper names; Quine does not acknowledge
a special category of logically proper names, and neither would he
accept th� epistemological assumptions underlying Ruesell'a doctrine
of acquamtance, but he would, 1 think, sympathise with Russell's
 Singular ierms 67
view of the theory of descriptions as an instrument of ontological
restraint.
For Frege, who has no special category of logically proper names
the meaning of which is identified with their denotation, the problem
of non-denoting names looks somewhat different. Frege can allow
that sentences containing non-denoting names or descriptions never-
theless have a perfectly good sense (express a bona fide proposition).
However, given his principie that the reference of a compound expres-
sion depends on the reference of its components, he is obliged to
admit that a sentence like 'The present King of France is bald ', the
subject of which has no reference, itself lacks reference, that is to say,
it has no truth-value. So, whereas according to Russell's analysis
'The present King of France is bald' entails that there is a present
King of France (for that there is, is part of what it says), according to
Frege's account "The present King of France is bald 'presupposes that
there is a present King of France; that is, it is neither true nor false
unless 'the present King of France exists ' is true. An adequate formal
treatment of presupposition, pretty obviously, would require a non-
bivalent logic, a logic in which truth-value gaps were allowed, i.e. in
which sorne wffs are neither true nor false. However, Frege does not
offer such a logic (but see Smiley 1960 and van Fraassen 1966 for
formal reconstructions of Frege's idea); for he thinks of non-denoting
singular terms as an imperfection of natural languages which ought
not to be permitted in a logically perfect language, and so recorn-
mends that, in formal logic, all singular terms be guaranteed a denota-
tion, if necessary by artificially supplying an object - he suggests the
number o - as their referent. (The choice of the number o may be
a little unfortunate, since it would presumably have the consequence
that 'The greatest prime number is less than 1 ', for instance, was
true.) Anyway, whereas on Russell's theory definite descriptions and
ordinary proper names aren't genuinely singular terms, but are con-
textually eliminated, Frege treats ordinary names and descriptions as
bona fide singular terms, each with a single referent, with 'rogue'
terms like 'the greatest prime number ' referring to o. (A Fregean
formal theory is to be found in Camap 1942.)
In his influential critique of Russell's theory, Strawson 1950 (and
cf. Nelson 1946 which anticipates sorne of Strawson's points) employs
a notion of presupposition which is reminiscent of Frege's analysis.
There are differences, though, to be sorted out first, deriving in large
part from Strawson's stress on the distinction between sentences and
68 Phi'losaphy of logics

staternents. According to Strawson, while it is linguistic expressions

which have rneaning, it is uses of linguistic expressions which refer,
and, in particular, uses of sentences - staternents - which are �ru� or
false. So his diagnosis of the problern of non-denoting descnpttons
goes like this: though the expression 'the present King of France' is
quite rneaningful, a use of that expression fails to refer, and con-
sequently a use of a sentence containing that expression fails to rnake
a true or false staternent. Strawson is ambiguous about whether his
diagnosis is that a use of the sentence 'The present King of France is
bald' fails to make a staternent, or that such a use rnakes a staternent,
but a staternent which is neither true nor false. (The ambiguity is set
out clearly in Nerlich 1965.) There is also an ambiguity in Strawson's
thesis that an utterance of "The present King of France is bald' does
not, as Russell thinks, entail, but presupposes, that there is a present
King of France: sorne passages suggest that it is the speaker who pre-
supposes that there is a present King of France, others, that presup-
position is not this kind of episternological relation, but a logical
relation holding between the staternent that the present King of
France is bald, and the staternent that there is a present King of
France. In later papers (1954, 1964) Strawson settles for the second
thesis: presupposing is a logical relation between staternents, such
that S1 presupposes S2 just in case S1 is neíther true nor false unless
S, is true. Since logical relations only hold, accordíng to Strawson,
between staternents, thís also resolves the first ambiguity noted above -
an utterance of "The present King of France is bald' must be allowcd
to constitute a staternent, but a staternent that is neither true nor false.
Notice, first, that, but for the insistence that it is a relation between
staternents, Strawson's account of presupposition is just like Frege's;
and, second, that íf an utterance of "The present King of France is
bald' does, after all, constitute a staternent, Strawson's criticisrn that
Russell's rnistake was to fail to distinguish between sentences and
staternents cannot be sustained. (On this second point cf. Russell's
(1959) reply to Strawson.)
I arn doubtful whether the question, whether "The present King of
France is bald' should be accounted false or truth-valueless could, or
even should, be settled by appeal to 'what we would ordinarily say'.
The issue tums, rather, on whether one is pre pared to tolerate sorne
artificiality (either, in the case of Russell's theory of descriptions, in
the translation frorn natural languages into the formalism, or, in the
case of Frege's preferred theory, in the choice of refercnt for other-
 Singular terms

wise non-denoting expressions) in order to conserve bivalence, since

the Fregean 'presupposition' theory advocated by Strawson would
require a non-bivalent base logic. And, if, of course, one thought
there were other reasons for doubting bivalence, this would be rele-
vant to one's estímate of the relative costs and benefits. (The com-
ments about rival strategies in formalisation in ch. 9 § 1 are pertinent
to this particular choice.)
Strawson is careful to say that it is uses of expressions which refer.
But there is, again, sorne ambiguity about what he takes to be the
conditions for successful reference. Sorne passages hint at a pragmatic
account, according to which it is a sufficient condition for a use of an
expression to succeed in referring to an object that the speaker have
a certain object in mind, and his use of the expression bring that
object to the hearer's attention - regardless, that is, of whether the
expression used actually denotes that object (cf. Strawson 1959 ch. 1
and 1964). But generally Strawson prefers a semantic account,
according to which it is necessary, for a use of an expression to
succeed in referring to an object, that the expression denote that
object.
Donellan 1966 brings the pragmatic notion of reference to the fore.
Donellan distinguishes between attributioe and referential uses of
definite descriptions. (The same definite descriptíon can be put to
either use.) A definite description is used attributively if the speaker
wants to assert something about whoever or whatever fits the descrip-
tion; referentially if, rather, he wants to draw his audience's attention
to sorne particular person or thing and assert something about that.
Donellan gives as example the use of the sentence 'The man who
murdered Smith is insane', attributively, to convey that anyone who
murdered Smith must be insane, or referentially, to convey that J ones
(whom the speaker and the audience know to have been convicted of
the murder - perhaps wrongly) is insane. And one can use a definite
description referentially, in Donellan's sense, even if it is not true of -
and even if speaker and hearer know it is not true of - the person or
thing referred to; for the criterion of successful referential use ia
simply that the speaker should manage to bring to the audience's
attention the person or thing he has in mind. Strawson's aceount,
Donellan suggests, is applicable only to attributive and not to
referential uses.
It is true - as I had already suggested earlier - that Strawson's
theory is, in the end, more semantic and less pragmatic than bis
70 Philosaphy of logics

official stress on the use of expressions might have led one to expect.
It might be a helpful artífice to distinguish between reference an_d
denotation or designation, and to use the former for the pragmatic
notion (what speakers do) and the latter for the semantic one (what
expressions do); then one can, if one wishes to adopt Donellan's
standards of successful reference, say that a speaker may refer to a
person or thing by the use of an expression which doesn't denote that
person or thing. One advantage of this is to make clear that one needn't
regard Donellan's account of the 'referential use' of definite descrip-
tions as a rival to Frege's or Russell's theory.

4 Non-denoting names: fiction

The issues here are complex and tangled, and I can't hope
to comment on them all. Some - the relations between singular terms
and bound variables, and the possibility of eliminating the former in
favour of the latter - have been touched upon already (ch. 4 §2).
Others will receive more attention in subsequent chapters - the role
of singular terms in modal contexts and the consequences of rival
theories of naming for problems about the identity of individuals
across possible worlds in ch. 10, theories of presupposition in ch. 1 I.
One - the question of non-denoting names - l'U tackle now.
It emerged in the above discussion that there are two kinds of
discrepancy between proper names in natural languages and singular
terma in formal languages: while singular terma are each aasigned
just mu individual in the domain, proper names sometimee have more
than mu bearer, and sometimes ,io,u. It is not unknown for writers
simply to dísmiss these discrepancies, assuming 'for the aake of
argument' that ordinary proper names reliably denote a unique indi-
vidual (e.g. McDowell 1977); but some interesting questions are
evaded if they are so lightly disregarded. I shan't discuss, here, the
fi.rst discrepancy, that proper names (' J ohn Smith ') frequently have
several, or many, bearers, though it is worth observing that, of the
theories I have sketched, Burge's is the one that takes this possibility
most seriously. For the present, I shall confine myself to sorne com-
ments about the other discrepancy, the phenomenon of non-denoting
names, and, relatedly, to some thoughts about fictional discourse.
The problems raised by non-denoting names can be thrown into
relief by considering them from the point of view of Ruasell's theory.
Take the name of a fietional character, 'Bherlock Holmes ', for inatance.
According to Russell, the meaning oí a bonafide ('logically') proper
 Singular terms

name is to be equated with its denotation, so, if 'Sherlock Holmes'

were a bona fide name, it would, sin ce it is non-denotíng, be meaning-
less, and so, too, would all sentences about Sherlock Holmes, includ-
ing sorne, like 'Sherlock Holmes never existed ', which one takes,
surely with sorne justification, to be straightforwardly true (the
'problem of negative existentials', cf. Cartwright 1960 ). Russell
would avoid this difficulty by denying that 'Sherlock Holmes ' is a
genuine name; it is a disguised definite description, and sentences
about Sherlock Holmes are disguisedly existential, perfectly meaning-
ful, and either straightforwardly true or straightforwardly false:
'Sherlock Holmes never existed' is true, while other statements about
Sherlock Holmes, like 'Sherlock Holmes was a detective' or 'Sherlock
Holmes was a policeman ', are false.
Russell's account provides an explanation of how it is possible for
us to speak meaningfully about non-existents, and to say truly that
they are non-existent, and at the same time a simple solution to the
problem of the truth-values of such statements. But sorne have felt
that the assignment of 'false' alike to, say, 'Sherlock Holmes was
a detective' and 'Sherlock Holmes was a policeman', is too crude,
and takes too little account of the intuition that the fonner is 'right'
in sorne sense in which the latter is not.
Sherlock Holmes is a fictional character, and according to the works
of fiction in which he features, he was a detective, and not a police-
man. It has been suggested (cf. Routley 1963) that a formal language
appropriate to represent discourse about Holmes might require a
dornain of fictional entities, so that the name 'Sherlock Holmes' does
denote, only it denotes a fictional, nota real, object. (Such systems are
known as 'free logics ', free, that is, of existential commitment; see
Schock 1968 and cf. remarks about alternative choices of domain in
ch. 4 § r.) This approach, interestingly, is in the spirit of Meinong's
theory of objects, which allows for meaningful discourse about non-
existents by admitting not only real, spatiotemporal objects, such as
physical objects and persons, and subsistent, non-spatiotemporal,
objects such as numbers and properties, but also non-existent, non-
subsistent, and even impossible objects as all, genuínely, objects (see
Meinong 1904 and cf. Parsons 1974). Russell 1905 recognises, as we
saw, that this offered an alternative to his own theory, but thought it
ontologically objectionable, maybe because of its affinities with onto-
logical extravagances in which he had himself once indulged (1903).
The curious way in which free logics represent non-denoting terms as
72 Philosopñy of logics

terms denoting unreal objects (somewhat as the third 'value' of s�me

3-valued logics is intended to represent a lack of truth-value) might
be thought, similarly, to exhibit a certain ontological ambivalence.
Now although the story tells usa good deal about Holmes, there are
also a good many statements about him the truth of which is not fixed
by the story - whether he had an aunt in Leamington Spa, for
instance. So there is sorne motivation not only to adjust the domain
to allow fictional entities, but also to allow that while sorne statements
about Holmes are true and others false, others, again, are neither.
And this means that a suitable formal language might need to aban don
the principie of bivalence, the principie that every statement is either
true or false. In such a formal language there would be scope for the
representation of the Fregean relation of presupposition, which, as
I pointed out above, calls for a non-bivalent logic.
There is a question, of course, to what extent all talk of non-
existents should be seen on the model of talk about fictional entities;
though Sherlock Holmes and the greatest prime number are alike in
not existing, it is debatable whether they are alike in ali logically
relevant respects. But for the present I shall confine myself to the con-
sideration of fictional entities. It is clear, at any rate, that there is an
important distinction between, on the one hand, discourse about
fiction, and, on the other, discourse in fiction. (1 don't mean to sug-
gest, of course, that ali discourse in or about fiction is discourse about
fictional entities.] What I do when I talk about Sherlock Holmes is
presumably not altogether on ali fours with what Conan Doyle did in
writing the Holmes stories. In particular, whereas in the former case
one can see sorne grounds for the intuition that there is a sense in
which what I say may be right or wrong, in the latter case it seerns
more appropriate to say that the question of Doyle's getting it right or
wrong simply doesn't arise. The kind of response just considered,
I think, looks more promising with respect to discourse about fiction
than with respect to discourse in fiction.
What's unusual about discourse in fiction, 1 suspect, isn't semantic
at ali, but pragmatic. Uttering ( or writing) sentences in the course of
telling a story differs from uttering sentences in the course of reportíng
an actual event; one is not, in the former case as one is in the latter
assertíng, that is claiming the truth of, the' sentences one utters'
(cf. Plantinga 1974 ch. 8 §4, Woods 1974, Searle 1975, Haack 1976b).
Whereas one might feel the need for a free logic for discourse about
fiction, one míght reasonably hope to handle the distinctive features
 Singular temu 73
of discourse in fiction by means of a theory of pragmatics, For, if my
hunch is right, the most significant difference between telling a story
and making a report, so to speak, is not in the difference between the
story and the report, but in the difference between the telling and
the making.
Sometimes it is taken for granted that if the distinctive features of
sorne kind of discourse are pragmatic, that necessarily puts it beyond
the scope of formal logical methods. The ubiquitous importance of
pragmatic aspects of all discourse in natural languages has been a
recurring theme with critics, such as Schiller and Strawson, who
regard formal methods as seriously inadequate to the subtleties of
natural language. So perhaps I should stress that, in urging that the
distinctive characteristics of discourse in fiction may be rather prag-
matic than semantic, I do not take for granted that this necessarily
excludes the possibility of formal treatment.
 6
Sentences, statements,
propositions

I Three approaches
A recurrent issue in the philosophy of logic concems the
question, with what kind of item logic deals, or perhaps primarily
deals, The altematives offered are usually sentences, statements and
propositions, or, more rarely these days, judgments or beliefs. 1 have
put the question in a deliberately vague way, since more than one
issue seems to be involved. Once again, as with the issue about the
meanings of connectives, quantifiers, etc., the problem concems the
relation between formal and informal arguments: what in informal argu-
menta corresponds to the well-formed formulae of formal languages?
It may be useful to distinguish three approaches to the question:
(i) syntactic: what, in natural languages, is the analogue of
the 'p', 'q' of formal logic?
In speaking thus far of 'sentence calculus ', 1 did not mean to beg this
question. Sorne prefer to speak of 'propositional calculus', "proposi-
tional variables', "propositíonal connectives'; and so far I have said
nothing to justify my preference for the former usage.
(ii) semantic: what kind of ítem is capable of truth and falsity?
Since formal languages aim to represent those informal arguments
which are valid extra-systematically, that is, which are truth-
preserving, this will relate closely to the first issue.
{iii) pragmatic:1 what kinds of item should one suppose
1
1 ca11 this the pragmatic approach because pragmatics is concerned
with relations between expresaions and the usen of thoee expresaiom
(' ayntax' and 'semantics' were explained in ch. s), I derive thi• way
of aeparating the iasuea from Gochet 197:a.
 Sentences, statements, propositions
75
to be the 'objects' of belief, knowledge, supposition,
etc.?
(' Know ', 'believe ', 'suppose ', etc. are sometimes called the verbs of
'propositional attitude'.) Since one can know, believe or suppose
either something true or something false, the third will relate quite
closely to the second issue.
For the present, however, 1 shall not discuss (iii) (but cf. pp. 124-'7
and ch. 12 §3); 1 shall comment first, very briefly, on (i), and then, at
greater length, on (ii).

2 Sentence, statement, proposition

A necessary preliminary, however, is to specify what I shall
mean by 'sentence ', 'statement' and 'proposition'; for one reason
why discussion of these issues is often confusing is that there is scant
uniformity of usage.
By a sentence I shall mean any grammatically correct and complete
string of expressions of a natural language. For example, 'Snow is
white', 'Shut the door', 'Is the door shut?' are sentences; 'Sat by in'
and 'Is pink' are not. I hope this rough and ready account is sufficient
to convey the idea I have in mind; it is, of course, imprecise to the
extent that there is uncertainty about which strings of expressions are
to count as grammatical. 1 shall need to distinguish between sentence
types and sentence tokens. A sentence token is a physical object, a series
of marks on paper or of sound waves, constituting an inscribed or
spoken sentence. Sometimes, however, one thinks of two or more
tokens as inscriptions or utterances of, in sorne sense, the same
sen ten ce; 'same sentence ', here, means 'same sentence type '. For
example, the two inscriptions:
Ali philosophers are slightly insane
Ali philosophers are slightly insane
are tokens of the same type. One could think of a sentence type either
as a pattern which similar tokens exemplify, or as a class of similar
tokens. The question of what to take as criteria of identity for sentence
types is disputed; sorne would require typographical or auditory simi-
larity (presumably one would also need to specify the conditions in
which an utterance was of the same sentence type asan inscription),
others would requíre sarneness of meaning. 1 shall stick to the former
criterion, and allow the possibility of ambiguous sentence types.
76 Philosophy of logics

I need, again, to distinguish, among sentences, those which are inter-

rogative or imperative, for example, frorn those which are 'declara ti ve'.
Sentences with their main verb in the indicative mood are declarative,
but 'declarative' is meant to be rather broader than 'indicative ', to
include, for example, conditionals whose main verb is in the sub-
junctive. Intuitively, one might say that declarative sentences are
those eligible for truth and falsity, whereas non-declarative sentences
are not; but to define "declarative ' thus, in the present context, would
obviously be question-begging.
By a statement I shall mean what is said when a declarative sentence
is uttered or inscribed. In its non-technical employment, 'statement'
is ambiguous between the event of the utterance or inscription of a
sentence, and the content of what is inscribed or uttered. Only the
second sense is relevant to present concerns. The question now arises,
whether every utterance or inscription of a declarative sentence will
make a statement. Strawson seems to think that sorne uses of declara-
tive sentences - his examples include utterances or inscriptions used
in the course of acting a play or writing a novel - do not make state-
ments. He also, as we saw in the previous chapter, seems to hint that
utterances of sentences whose subject terms do not denote anything
fail to make statements, though at other times he suggests, rather,
that such utterances are statements, but statements which are neither
true nor false. These questions will obviously be important to the
issue about truth-bearers. When, now, do two utterances or inscrip-
tions make the same statement? It is usually said that they do so just
in case they 'say the same thing about the same thing '. This account
works well enough in simple cases. For instance, the utterances:
You are hot (said by x to y)
I am hot (said by y)
J'ai chaud (said by y)
would, by these standards, make the same statement. Making the
criterion precise, however, threatens to be difficult, for it may not
always be easy to specify when two utterances are about the same
thing, and it could be harder still to specify when they say the same
thing about their subject, since this will require an appeal to the
notoriously tricky notion of synonymy.
By a proposítion I shall understand what is common to a set of
synonymous declaratíve sentences. In this sense of "proposition ' two
sentences will expresa the same proposition if they have the same
 Sentences, statements, propositions 77
meaning; so here again, as with statements, the problem of synonymy
will have to be faced, Another account, popular since the advent of
possible-worlds semantics for modal logics, identifies a proposition
with the set of possible worlds in which it is true, or with a function
from possible worlds into truth-values. It isn't clear, however, that
this comes to anything very different from the account I gave before,
sin ce one distinguishes the possible world in which p from the possible
world in which q by means of distinguishing p from q. (If 'J ack and
Jill have one parent in common' expresses the same proposition as
'J ack and Jill are step-siblings ', then ali possible worlds in which the
first holds are possible worlds in which the second holds, and if not,
not.) Another account, which delimits a different idea, identifies the
proposition with the common content of sentences in different
moods. So:
Tom shut the door.
Tom, shut the door I
Has Tom shut the door?
have as common content the proposition Tom's shutting the door,
Propositions in this sense are unlikely candidates as truth-bearers,
and for that reason I shall pay them rather little attention here. They
do, however, have sorne relevance to the interpretation of e.g, impera-
tive logic, on which I shall offer sorne brief comments below.
It is quite easy to check that sentences, statements and propositions,
as characterised here, are distinct, that is, that one could have same
sentence/different statement/different proposition, same statement/
different sentence/different proposition, same proposition/different
sentence/different statement (cf. Cartwright 1962).
One's attitude to statements or propositions may well be coloured
by one's metaphysical views. Nominalists, who dislike abstractobjects,
or extensionalists, who suspect that meaning notions suffer from a
crippling unclarity, are likely to be ill-disposed towards statements
and propositions and better disposed towards sentences, while
Platonists, admitting abstract objects, and intensionalists, comfort-
able with the theory of meaning, could admit statements or proposi-
tions with equanimity. (Compare Quine 1970 ch. 1 with Putnam
1971 chs. 2, 3, 5, for contrasting attitudes.) It is necessary to observe,
however, that though sentence tokens are physical objects, sentence
types are abstract; and that, while the identity criteria for both state-
ments and propositions require appeal to synonymy, the identity
78 Philosophy of logics

criteria for sentence types require appeal to the not altogether

unproblematic notion of similarity. (See Goodman I?7? �or
sorne of the problems which beset atternpts to define similarity
precisely.)

3 'Sentence letters', 'propositional variables', or what?

How one understands the 'p', 'q' ... etc. of sentence logic
will obviously depend upon whether one allows sentence letters to be
treated as genuine variables to be bound by quantifiers, and, if one
does, how one interprets those quantifiers.
The usual presentations of sentence logic do not use quantifiers.
On the face of it, however, it seerns reasonable to suppose that
unquantified sentence calculus has an implicit generality which quanti-
fied sentence calculus rnerely rnakes expli'cit. A theorem lik.e
'P-+ (p V q)' is usually understood to hold for all instances of 'p' and
'q', just as, in the usual, unquantified presentations of algebra,
'a+ b = b +a' is understood to hold whatever a and b rnay be. So the
alternatives are either to regard the usual, unquantified forrnulation
as sirnply an abbreviated version of quantifi.ed sentence logic, or else
to find sorne other way of accounting for the irnplicit generality of
the unextended calculus.
Quine, for reasons already touched upon in ch. 4 §3, prefers the
second alternative. He proposes that 'p ', 'q' etc. not be treated as
genuine, bindable variables, but be construed, instead, as "schernatic
letters '. A wff of sentence calculus, such as 'p V - p ', is to be thought
of 'not as a sentence but as a scherna or diagram such that all actual
statements of the depicted forrn are true' (1953a p. 109).
If, however, one does treat sentence letters as genuine variables, one
is then faced with the question of the interpretation of the quantifiers.
If one adopts an objectual interpretation, one is next faced with the
question of what kind of object the quantifiers are to range over:
propositions are the most usual candidates, though Quine 1934
argued for a dornain of sentences. (If one is concemed only with the
usual, truth-functional sentence calculus, one might even construe such
quantifiers as ranging over truth-oalues, that is to say, over the two
values t and /; for in truth-functional sentence logic only the
truth-values of the cornponents are relevant to the truth-value of the
cornpound. The addition of non-truth-functional sentence operators,
' ne cessan·1y ' , ' perha ps ' , or • B b e 1·ieves t hat ', however, would rule out
this altemative.) An adjustment to the usual reading is then needed:
 Sentences, statements, propositions 79
in '(p) (p V - p) ', if the quantifier is read 'for all propositions p ', then
'p v-p' must be construed as a singular term denoting a compound
proposition (' the disjunction of a proposition with its own negation '),
and an implicit predica te (' is true') has to be supplied to make the
reading grammatical. If, on the other hand, one adopts a substitu-
tional interpretation, '(p) (p v-p)' will be read 'AH substitution
instances of ' ... V- ... ' are true', where appropriate substitution
instances result from putting the same sentence in each of the blanks.
(It will not have escaped notice, I'm sure, that Quine's 'schematic
letters' look much like bound variables of substitutional quantification
with sentences as the substitution class.)
At this level, then, there seem to be several options. But what about
the question of truth-bearers?

4 Truth-bearers
If an argument is valid, then if its premises are true, its con-
clusion must be true too; so presumably premises and conclusions
need to be the kind of item capable of being true or false. So, many
writers have considered it very important to decide whether it is
sentences, statements or propositions which are properly called 'true'
or 'false'. The issue has many ramifications; it has been suggested,
for example, that confusion about truth-bearers underlies the
semantic paradoxes (Bar-Hillel 1957, Kneale 1971), that it motivates
the proposals for many-valued logics (Lewy 1946, Kneale and Kneale
1962, Kripke 1975 p. 7000), that it vitiates Russell's theory of
descriptions (Strawson 1950). I have already commented (ch. 5 §3) on
the last of these; I'll have something to say about the first in ch. 8,
and about the second in ch. 11.
The dispute about truth-bearers usually proceeds along sorne such
lines as the following: since truth is presumably a property, one
should be able to identify the type of item which possesses it; it is
usually taken for granted either that only one of the candidates
can be the truth-bearer, or that one is primary and the others
somehow derivative. The ensuing debate about which items are the,
or the primary, truth-bearers, however, has been, to my mind,
neither very conclusive nor very fruitful. You will shortly see what
I mean.
Several writers (Strawson 1950, Pitcher introduction to 1964,
Putnam 1971, for example) have claimed that it is irnproper, or even
meaningless, to speak of sentences as being true or false. The argu-
80 Philosaphy of logics

ments offered for this claim, however, seem pretty inconclusive. One
is that if sentences were true or false sorne sentences would be some-
times true and sometimes false; another is that sorne sentences, non-
declarative sentences, for instance, just aren't capable of truth or
falsíty, so not ali sentences could be true or false. But a gate, after ali,
can quite properly be called red or green, though it may have one
colour one year and another the next; and sorne glass, stained glass for
instance, can properly have colour predicates ascribed to it, despite
the fact that sorne glass is colourless (cf. Lemmon 1966, R. J. and S.
Haack 1970).
Though these arguments certainly don't show that sentences can-
not properly be called true or false, they may suggest an apparently
more promising line of thought: that whatever items be ch osen as
truth-bearers, they should be such that (i) they can be relied upon not
to change their truth-value, and (ii) all items of the relevant kind are
either true or false. The acceptability of these desiderata will need
investigation, of course. But even leaving that question aside for the
present, it tums out that statements and propositions are scarcely
more successful than sentences in these respects.
(i) Whether a statement can change its truth-value depends,
obviously, on how exactly one understands 'saying the same thing
about the same thing'. But on an intuitive understanding, at least,
two utterances conceming the same Jones, half a minute apart, of
'Jones is wearing an overcoat ' would presumably say the same thing
about the sarne thing. Yet one utterance could be true and thc other
false, if J ones put on or took off his overcoat in the interval. Of course,
one could guarantee against statements changing their truth-velues
by tightening up the criteria for statement identity so far that no
non-simultaneous utterances count as making the same statement.
But this would, in effect, correlate statements one-one with sentence
tokens, and then one might justifiably wonder what the point was of
introducing statements as distinct from sentences.
Since the sense of a sentence may remain stable over a considerable
period, the proposition expressed by a sentence could also, pre-
sumably, change its truth-value; for instance, the proposition
expressed by the sen ten ce 'Louis XIV is dead' was once false and is
now true. Sorne writers (Frege 1918, Moore 1953, Kneale 1971, for
instance) have responded to this difficulty by tightening up the
criteria of propositional identity so as to disallow change of truth-
value; this seems to be vulnerable to a similar objection to that made,
 Sentences, statements, propositions 81

above, to a comparable manoeuvre to prevent statements changing

their truth-values.
(ii) Since it is uncertain whether every utterance of a declarative
sentence is supposed to make a statement, it is not clear, either,
whether every statement must be either true or false. Strawson con-
cedes, however, that it is not part of the definition of 'statement' that
every statement is either true or false (1952 p. 69); and, as we saw,
there are traces in 1950, andan explicit claim in 1964, that utterances
of 'reference failure' sentences make statements which are neither
true nor false. So sorne statements will lack truth-value.
In sorne cases where a sentence is neither true nor false, one could
plausibly argue that there is no corresponding proposition, and to
that extent propositions fare better than sentences at satisfying (ii).
Among the sentences which are neither true nor false, there are, so it
is often said, sorne which though grammatically correct are meaning-
less (' Virtue is triangular' for instan ce); being meaningless, such sen-
tences express no proposition. Imperative and interrogative sentences
also presumably fail to be true or false, and it could, again, be claimed
that such sentences do not expresa propositions. However, it is
doubtful whether one could specify which kinds of sentence do
express propositions except by restricting oneself to declarative
sentences (as in §2 above); so this argument does not show proposi-
tions to be better off with respect to (ii) than sentences. And sorne
declarative sentences (vague sentences and future contingent sen-
tences for instan ce) are thought by sorne writers to be neither true
nor false, and yet, being meaningful, expresa propositions which are
therefore themselves neither true nor false.
I am not suggesting, of course, that sentences fare better than
statements or propositions with respect to (i) and (ii). I have already
mentioned, above, severa! kinds of sentence which may fail to have
any truth-value, so sentences don't meet (ii). As for (i): many sen-
tence types obviously change their truth-value (' I am hungry ', for
instance, would be true in sorne mouths at sorne times, false in others
and at others); even sorne sentence tokens can be shown capable of
changing their truth-value. (A token of "There is a person in this
room', written on the blackboard in my office, would usually be true
at 12 noon and false at 12 midnight.) Quine has pointed out that we
can specify a class of sentence types which do not change their truth-
value; it would include both sentences stating physical and mathe-
matical laws, to which temporal considerations are, he argues,
82 Philosophy of logics

irrelevant, and sentences completely specified as to time and place,

with tensed verbs and indexicals like 'now' replaced by tenseless
verbs, dates and times. Quine calls these stable types 'eternal
sentences' (cf. ch. 9 §3).

Truth-bearers and theory of trutb

One argument which might be offered in favour of allowing
sentences as truth-bearers is this: sorne theories of truth, certain
versions (Wittgenstein's hut not Austin's, for example) of the corre-
spondence theory, and, most notably, Tarski's semantic theory,
exploit grammatical structure in the definition of truth ( details in
ch. 7). Sentences, of course, have grammatical structure; statements
and propositions, however, being extra-Iinguistic, do not. And since
the same statement may be made by uttering, and the sarne proposi-
tion expressed by, sentences in different languages with different
grammatical structures, it will be hard for statements or propositions
to 'borrow' a structure from the sentences which make or express
them. However, while sorne would regard the plausibility of Tarski's
theory as a reason for regarding sentences as truth-bearers, others, on
the basis of their conviction that sentences cannot be truth-bearers,
are disposed to reject Tarski's theory (see e.g. White 1970 pp. 94-9).
Sorne, again, would argue that the fact that Tarski's definition of
truth has to be language-relative, that he defines 'true-in-L' rather
than 'true', is a point against it. And others propose modifying
Tarski's theory in ways which would make it applicable to proposi-
tions (Popper 1972) or atatements (Davidaon 1967).
One can see, on reflection, that auch requirementa as (i) and (ii),
which those who object to sentencea implicitly impose on truth-
bearers, themselves relate to assumptions - which turn out to be
questionable - about the theory of truth : that a correct theory will be
bivalent and will make truth timeless. I shan't discuss the question
of truth's supposedly timeless character here, but simply refer the
reader to Putnam 1957 and Haack 1974 pp. 69-70. One or two brief
comments about bivalence may be in order. Undue weight placed on
the idea that every ítem of a kind to have a truth-value should be
either true or false often líes behind an undesirable kind of con-
servatism about deviant logics. For sorne writers react to the sugges-
tion that certain sentences, being neither true nor false, perhaps call
for a non-bivalent logic, by retorting that such sentences cannot make
statements or cannot expresa propositions, and so, aince it is with
 Seniences, statements, propositions

statements or propositions that logic is concerned, are outside its

scope (see e.g. Lewy 1946, and cf. Kripke 1975 p. 7000; the 'no ítem'
thesis is discussed in Haack 1974 pp. 47-53). This reaction is apt to
trivialise serious issues.
Sorne theories of truth - the descendants of Ramsey's 'redundancy'
theory - suggest a radical solution to the problem about truth-
bearers. The question what truth is a property of arises on the -
natural enough - assumption that truth is a property. But these
theories ( ch. 7 §7) deny that truth is a property, and hence bypass the
question, what it's a property of. One could be excused for thinking
that, in view of the unsatisfactory state of that question, it is a
virtue of such theories to avoid it.

5 The problem reformulated

Arguments against sentences seem to impose requirements
on truth-bearers which statements and propositions also fail, and
which are themselves questionable anyhow; sorne argue in favour of
sentences as truth-bearers that Tarski's truth theory requires them,
others reject Tarski's theory because it requires sentences as truth-
bearers ... One begins to suspect that the formulation of the problem
may be in need of improvement. I think the problem which underlies
the debate can be reformulated in a way which makes it rather more
manageable. I began, you may remember, by observing that the issues
about sentences, statements, propositions and so forth arose, as philo-
sophical problems in logic are apt to, from questions about the
relations between formal and informal arguments. Now suppose one
has a formal argument in sentence calculus such as:
pv-q
:::P__
-q
and wants to know what informal arguments can properly be regarded
as instances of it. Obviously this is something one needs to know if
formal logic is to help one assess informal arguments. One question
which needs to be answered now is what can stand, in an informal
argument, for the 'p' and 'q'. Well, one wants to say, any sentence
you like can correspond to 'p' and 'q ', provided the same sentence
corresponds to each occurrence. That's a start, but more constraints
are needed: declarative sentences can correspond to 'p' and 'q ', but
not imperative or interrogative sentences; if the sentences corre-
84 Philosophy o/ logics

sponding to 'p' and 'q' are tensed, then the time reference should
stay constant throughout the argument; if they contain indexicals
like 'I', 'he', 'now', their reference should stay constant through the
argument; and if they're ambiguous, they should be used in the same
sense throughout the argument. Otherwise, though the formal argu-
ment is valid, its supposed analogue in informal argument is apt to
be invalid; if, for instance, the last condition is not observed, one has
a 'fallacy of equivocation '.
This way of putting the problem has the advantage of metaphysical
neutrality, so that it raises neither nominalist nor Platonist hackles;
and yet it seems to ask the right questions about how to apply formal
logic to informal argument. And the fact that the constraints on what,
in informal argument, can stand where 'p' and 'q' do in formal logic,
reflect the identity conditions proposed for the various candidate
truth-bearers is sorne confirmation of my claim to have reformulated
the original problem rather than replace it by a different one. (But it
turns out, interestingly, that in its reformulated version the problem
arises even on the theories which do not regard truth as a property.)
The reformulated problem does not refer to 'truth-bearers'
directly, but, rather, asks what vagaries of the sentences put for 'p'
and 'q' may interfere with validity. In the case of classical sentence
logic this amounts to asking what may prevent the same sentence's
having the same truth-value at different occurrences within an argu-
ment; it is because an ambiguous sentence may be true in the
premises and false in the conclusion that equivocation interferes with
validity. But the greater generality of the reformulated problem
should prompt another look at the relations between validity and
truth.

Validity again
I observed earlier that to insist that logic deal only with items
which are either true or false is to give uncharitably short shrift to
non-bivalent logics. Furthermore, ifsuch enterprises as imperative or
erotetic logic (logic of questions) are to be feasible, it must be con-
ceded that logic may deal with sentences incapable of truth and
falsity. Now the extra-systematic account of validity that I gave in
ch. 2 was in terms of truth-preservingness, lf, however, one takes
serious account of the possibility of logics which deal with non-truth-
bearing iterns, one is likely to need an extended conceprion of validity.
If, for example, one wants to handle imperative sentences, it may
 Sentences, statements, propositions 85
prove appropriate to define an analogue of truth ( a 'designated value '
if you like) which is applicable to them. Ross 1968 suggests: 'pi' is
satisfied1 iff 'p' is true. (E.g. 'Close the door ! 'is satisfied iff' The door
is closed' is true.) Validity, for imperative logic, would then be
preservation of satisfaction rather than preservation of truth.
It should not be dismaying - nor even very surprising - that
developments such as non-bivalent or imperative logic may call for
changes or extensions of the intuitive conception of validity to which
the standard logical apparatus gives formal expression. It is often
enough by modification or extension of its key ideas that a science
grows.
1 This use of 'satisfaction' should be kept distinct from that to be
introduced in the next chapter, in the discussion of Tarski's theory
of truth.
 7
Theories of truth

1 A summary sketch1
The object of this section is to sketch the main kinds of
theories of truth which have been proposed, and to indicate how they
relate to each other. (Subsequent sections will discuss sorne theories
in detail.)
Coherence theories take truth to consist in relations of coherence
among a set of beliefs. Coherence theories were proposed e.g. by
Bradley 1914, and also by sorne positivist opponents of idealism, such
as Neurath 1932; more recently, Rescher 1973 and Dauer 1974 have
defended this kind of approach. Correspondence theories take the
truth of a proposition to consist, not in its relations to other propoai-
tions, but in its relation to the world, ita correspondence to the facts.
Theories of this kind were held by both Ruesell 1918 and Wittgenstein
1922, during the period of their adherence to logical atomism ; Auatin
defended a version of the correspondence theory in 1950. The prag-
matist theory, developed in the works of Peirce (see e.g. 1877),
Dewey (see e.g. 1901) and James (see e.g. 1909) has affinities with
both coherence and correspondence theories, allowing that the truth
of a belief derives from its correspondence with reality, but stressing
a1so that it is manifested by the beliefs' survival of test by experience,
its coherence with other beliefs; the account of truth proposed in
�ummett 1959 has, in tum, quite strong affinities with the pragmatist
view.
1
Proponents of the theories I shall discuss take clifferent views about
what kinds of items are truth-bearers, In what follows I ahall apeak
variously - depending upon which theory I am diacuaing - of
' belíefs ', "aenteneee ', 'propoaitiona' etc. u true or falte; only when
the difference maket a dífference ahall I draw attmtion to it.
 87

·::!

1
1
1
1
1
1
1
1
1
\
1
\
1
1
1
1
1
' -� ·- e::
\ -;::�
=------------ �o
�;� -Q�
!;:! ...

\
1

I
I

t'�
,'
I
'1
\
, 'C1)""'
'-'

- ·5
...
------------ p..�
l a ·a�
Cl'J
;:l e:
C1)
�
l:E�
tr;i::t::----·-
> -in
c::,
� ¡;, -e
;:l 1
1

-i1
1
1

-i���
1 1
1
1

-�
1
.s -5
� g
,l,j

s
ss
C1)

� �� � of ------------- �
(.)
e:C1) .....o
o
;:l
� .E
...........
,:¡;:
.s ·e:fs
i..

/ � 8
'-' tlC1) ..e:
1
... E--
i =g>- ---------------- � :.a
1

.....
� �� J �
!JO
88 Philosophy of logics
Aristotle had observed that 'to say of what is that it is not, or of
what is not that it is, is false, while to say of what is that it is, or of
what is not that it is not, is true'. In proposing his semantic theory of
truth, Tarski 1931, 1944 aims to explícate the sense of 'true' which
this dictum captures. Truth, in Tarski's account, is defi.ned in terms
of the semantic relation of satisfaction, a relation between open
sentences (like 'x > y') and non-Iinguistic objects (such as the num-
bers 6 and 5). The truth theory recently proposed by Kripke 1975 is
a variant of Tarski's, modified essentially to cope in a more sophisti-
cated way with the semantic paradoxes. Popper's account of truth and
his theory of verísímílitude or nearness to the truth is based upon
Tarski's theory, which Popper regards as supplying a more precise
version of traditional correspondence theories.
The redundancy theory of truth, offered by Ramsey 1927, claims
that 'true' is redundant, for to say that it is true that p is equivalent
to saying that p. It is evident that this account has sorne affinities with
Aristotle's dictum, and consequently with sorne aspects of Tarski's
theory. There have been several recent variants of Ramsey's theory:
Strawson's 'performative' account (1949); the 'simple' theory of
truth suggested by Prior 1971 and amplified by Mackie 1973 and
Wílliams 1976; and the 'prosentential' theory presented by Gro ver,
Camp and Belnap 1975.

Definitions versus criterio of truth

A distinction is commonly made (by e.g. Russell 1908b,
Rescher 1973 ch. :2, Mackie 1973 ch. 1) between definitions of truth
and criteria of truth; the idea is, roughly, that whereas a definition
gives the meaning of the word 'true', a criterion gives a test by means
ofwhich to tell whether a sentence (or whatever) is true or false - as,
for example, one might distinguish, on the one hand, fixing the
meaning of 'feverish' as having a temperature higher than sorne given
point and, on the other, specifying procedures for deciding whether
someone is feverish.
This distinction needs careful handling. One's suspicions may be
aroused by the existence of disagreement about which theories of
truth count as definitional and which as criterial: for instan ce, while
Tarski himself disclaims any interest in supplying a criterion of truth,
and Popper regards it asan advantage of the semantic theory that it is
definitional rather than criterial, Mackie counts Tarski's theory as -
and criticises it for - aspiring to supply a criterion. And one's suspi-
 Theories of trutlt

cions should be confirmed by some clearly inappropriate uses of the

distinction. For exarnple, Russell accused the pragmatists of having
confused the definition and the criterion oftruth, when the pragmat-
tists held that the meaning of a terrn is correctly given precisely by
supplying criteria for its application. (It is not at all unusual, I'm
afraid, for a philosopher who deliberately identifies As and Bs to find
himself facing the criticism that he has 'confused' As and Bs.)
However, one cannot simply decide to refrain from using the dis-
tinction, problematic as it is, because of its importance to such
questions as whether the coherencc and correspondence theories need
be regarded as rivals, between which one is obliged to choose, or as
supplementing each other, correspondence supplying the definition
and coherence the criterion. This question is at issue even between
proponents of the coherencc theory. Thus Bradley, conceding that
"Truth to be truth must be true of something, and this something
itself is not truth' ( 1914 p. JZS), seems to allow that an account of the
meaning of truth may requirc appeal to something like correspon-
dence, while coherence is rather a mark, a test, of truth. Blanshard,
by contrast, insists that truth consists in coherence, which is a defini-
tion as well as a criterion. This insistence seems to be based on the
conviction that there must be sorne intimate connection between
a dependable criterion and what it is a criterion of. Coherence could
not be the test, but correspondence the meaning, of truth, he argues,
for then there is no explanation why coherent beliefs should be the
ones that correspond to the facts; if coherence is to be a reliable test of
truth, it must be because it is constitutive of the meaning of truth
(see Blanshard 1939 p. 268).
Rescher proposes (1973 chs. 1 and 2) to deflect this argument by
distinguishing between guaranteeing (infallible) and authorising
(fallible) criteria, and arguing that only in the case of guaranteeing
criteria need there be the connection with the definition that
Blanshard thinks inevitable. This distinction illuminates sorne issues
touched upon earlier. Rescher counts C as a guaranteeing criterion
of X if:
necessarily (C íff x obtains)
But - as Rescher observes - in this sense, any definition of truth
would also supply an infallible criterion of truth. For instance, if
truth consists in correspondence to the facts, then, necessarily, if 'p'
corresponds to the facts, 'p' is true, so correspondence is an infallible
90 Philosophy o/ logics

criterion.! (The idea that Tarski gives a criterion of truth may derive
from this conception of criteria.)
So: if one has a definition, one thereby has a 'guaranteeing'
criterion. The converse, however, is a bit less straightforward. It is
a guaranteeing criterion of a number's being divisible by three, for
example, that the sum of its digits be divisible by three, but this,
I take it, is not what it means for a number to be divisible by three.
Rather: if one has a guaranteeing criterion, then either it is, or else it
is a logical consequence of, a definition.
An authorising criterion, however, is fallible: it is not necessarily
the case that (C iff x obtains); so, either it is true, though not neces-
sarily, that C iff x obtains, or perhaps it is not invariably true that
C iff x obtains. (Rescher considers the second but not the first kind
of case.) So an authorising criterion of x is distinct from a definition
of x - it need not be logically related to the meaning of 'x '. 2
But now why, if any definition supplies a guaranteeing criterion,
should one ever want an authorising críterion] The answer is rather
clear, I think, but hard to put precisely: if one wants to fi.nd out
whether x obtains, one would like, ideally, a reliable indicator of the
presence of x which is easier to discooer to obtain than x itself. A defini-
tion gives an indicator which is perfectly reliable, but exactly as diffi-
cult to discover to obtain as x itself; an authorising criterion gives an
indicator which may be less than comp1etely re1iable, but which, by
way of cornpensation, is easier to discover to obtain. For example,
one might think of the characteristic spots as an authorising criterion
of measles; not a foolproof test, since it is not logically necessary
that one has the spots iff one has the meaales, but much more easily
discovered than, say, the presence of a given bacterium which ia (or

1 If one identifies meaning and criterion - as the pragmatista do - then

one is obliged to hold the criterion to be guaranteeing. This will be
pertinent to the discussion in §6 below of Popper's argument that the
pragmatiet theory of truth threatens fallibilism.
I
Rescher does nor explicitly arnplify the ' necesearily' in his account of
a guaranteeing criterion, but contextual clues indicate that he has in
mind logícal necessity, which is the interpretation I have used. If
physically necessary tests were included, the previoua and sorne
subsequent paragraphs would have to be rewritten to allow criteria
which are related to that of which they are II test by physical neceseity,
aa well as critería related by logical neceasity, to count aa guaranteeing.
Of course, the diatinction between lo&ical and phyaical necesaity - and
índeed, the dietinction between the neceseary and the contin¡ent - ia
not unproblematic.
 Theories o/ truth 91
so I shall suppose for the sake of argument) the guaranteeing
criterion.
So far, then, Rescher's vindication of Bradley's view of coherence
as a criterion (an authorising criterion, that is) but nota definition, of
truth, against Blanshard's argument for an inevitable connection
between definition and criterion, is successful. However, it is perti-
nent that a weaker version of Blanshard's idea seems to work even for
authorising criteria. It seems plausible to argue that, if C is an
authorising criterion (even in the least favourable case when its
presence isn't invariably correlated with that of x), then there ought
to be some kind of connection - not, indeed, a logical connection, but
perhaps a causal connection, for example - between x and C. Con-
sider, again, spots as an authorising criterion of measles; there is a
causal connection between the spots and the disease of which they
are the symptom. And, indeed, this is relevant to a feature of
Bradley's account which Rescher neglects. lt is plausible to think that
Bradley believed there to be a connection between one's beliefs' being
coherent, and their corresponding to reality (i.e. between the author-
ising criterion and the definition), for he holds that reality is coherent.
The concept of truth is as important to epistemology as to philo-
sophy of logic. Sorne theories of truth have an important epistemo-
logical component, are con cerned with the accessibility of truth; and
the search for a criterion of truth is often a manifestation of this
concem. It is noticeable that on the whole the theories on the left side
of the sketch of truth-theories (fig. 4) take the epistemological dimen-
sion more seriously than those on the right, with the coherence and
pragmatist theories epistemologically rich, but redundancy theories,
at the other extreme, with virtually no epistemological 'meat' ( as
Mackie puts it) on them.

2 Correspondence theories
Both Russell and Wittgenstein, during their 'logical atomist'
periods, 1 offered definitions of truth as the corresponden ce of a
proposition to a fact.
Propositions, according to Wittgenstein, are verbal complexes;
molecular propositions (such as 'Fa V Gb ') are composed truth-
functionally out of atomic propositions (as,' Fa'), The world consists
1 Wittgenstein was the originator oí logical atomiam, but Ruuell'a
veraion appeared first, in hia 1918 lectures, while Wittgenstein'• was
presented in 19:u in the Traetatus,
92 Philosophy of logics

of simples, or logical atoms, in various complexes, or arrangements,

which are facts. And in a perfectly perspicuous language the arrange-
ment of words in a. true, atomic proposition would mirror the
arrangement of simples in the world; 'correspondence' consists in
this structural isomorphism. The truth-conditions of molecular pro-
positions can then be given; '-p' will be true just in case 'p' is not
true, 'p v q' will be true just in case either 'p' is true or 'q' is true,
and so forth.
Wittgenstein's version of logical atomism is austere; Russell aug-
mented it with an epistemological theory according to which the
logical simples about whose character Wittgenstein is agnostic are
sense-data, which Russell took to be the objects of direct acquaint-
anee, and the meaningfulness of a proposition is supposed to derive
from its being composed of names of objects of acquaintance. These
epistemological additions do not vitally affect the core of the account
of truth; but sorne other differences between Russell's and Wittgen-
stein's versions are more relevant. Russell's account has the virtue of
recognising the difficulties in regarding ali molecular propositions,
notably belief propositions and universal quantifications, as truth-
functions of atomic propositions. Other features of Russell's version,
however, seem to create unnecessary difficulties; for instan ce, he
allows (though with less than complete confidence, because of the
adverse reaction this thesis received at Harvard !) negative as well as
positive facts, so that the truth of the negation of p can consist in its
correspondence to the fact that not p, rather than p's failure to corre-
spond to the facts; and the suggeerion that there are two correspon-
dence relations, one of which relates true propoaitions and the other
false propositions to the facts, seems gratuitous, indeed, in view of the
admission of negative facts, doubly so.
Numerous critics have observed that the trouble with the corre-
spondence theory is that its key idea, correspondence, is just not made
adequately clear. Even in the most favourable cases the required iso-
morphism between the structure of a proposition and that of thc fact
in vol ves difficulties; considcr:
The cat is to the left of the man (the proposition)

7l (the corresponding fact)

even here (as Russell concedes, pp. 315-16) it looks as if the fact has
 Theories of truth 93
two components, the proposition at least three; and of course the
difficulties would be much severer in other cases ( consider 'a is red',
'a is married to b ', or for that matter 'the cat is to the right of the
man'). The interpretation of correspondence as a structural iso-
morphism is intimately connected with both the theory about the
ultimate structure of the world and the ideal of a perfectly perspicuous
language, characteristic theses of logical atomism. The question arises,
therefore, whether the correspondence theory can be divorced from
logical atomism, and, if it can, what account could then be given of
the correspondence relation.
Austin 1950 supplies a new version of the correspondence theory,
a study of which offers sorne answers. Austin's version does not rely
either on an atomist metaphysics or on an ideal language; the corre-
spondence relation is explicated, not in terms of a structural iso-
morphism between proposition and fact, but in terms of purely
conventional relations between the words and the world. Correspon-
dence is explained via two kinds of 'correlation':
(i) 'descriptive conventions' correlating words with types of
situation
and
(ii) 'demonstrative conventions' correlating words with
specific situations
The idea is that in the case of a statement such as '1 am hurrying',
uttered by s at t, the descriptive conventions correlate the words with
situations in which someone is hurrying, and the demonstrative con-
ventions correlate the words with the state of s at t, and that the state-
ment is true if the specific situation correlated with the words by (ii) is
of the type correlated with the words by (i). Austin stresses the con-
ventional character of the correlations; any words could be correlated
with any situation; the correlation in no way depends on isomorphism
between words and world.
A difficulty with this account of correspondence, which essentially
appeals to both kinds of correlation, is that it applies directly only to
statements made by sentences which are indexical, since the demon-
strative conventions would have no role to play in the case of sentences
like 'J ulius Caesar was bald' or 'Ali mu les are sterile ', which can't be
used in statements referring to different situations. (Austin's corn-
ments on these cases, p. 23n, are none too reassuring.)
On the other hand, Austin's version, I think, makes an improve-
94 Philosophy of logics

rnent on Russell's account of 'the facts '. The point is hard to put
clearly, but it is significant enough to be worth putting even sorne-
what vaguely. Russell is apt to speak as if the truth of p consists in its
correspondence to the fact that p; but the trouble with this is that the
relation between 'p' and the fact that p is just too close, that 'p'
couldn't fail to correspond to that fact. His evasiveness about the
criteria of individuation of facts rnay indicate that he felt this dis-
comfort. Austin's version, however, locates the truth of the staternent
thatp notin its correspondence to the fact thatp, but rather in thefacts'
being as 'p' says, or, as Austin puts it, in the dernonstrative conven-
tions' correlating 'p' with a situation which is of the kind with which
the descriptive conventions correlate it. (Austin is aware of this
difference; see 1950 p. 23; and cf. Davidson 1973 and O'Connor
1975.)

3 Coherence theories
A coherence theory of truth was held by the idealists (1 shall
discuss Bradley's account, but related views were held by his Gerrnan
philosophical ancestors Hegel and Lotze) and also by sorne of their
logical positivist opponents. So the relation between coherence theories
and idealisrn is rather like that between correspondence theories and
logical atornisrn - in that in each case the theory of truth became
divorced frorn the rnetaphysical outlook with which it was originally
characteristically associated.
It will be useful - because this way sorne significant relations
between coherence and correspondence theories can be highlighted -
to begin in the rniddle, with Neurath's defence of a coherence view.
A little history will not go amiss: the logical positivista, under the
influence of Wittgenstein's Tractatus, originally subscribed to a cor-
respondence view of the character of truth. They were, however,
strongly rnotivated by episternological concems, and consequently
desired a test (authorising criterion) of truth - a way to tell whether
or not a sentence indeed corresponds to the facts. Carnap and Schlick
tackled the problern in two parts; staternents reporting immediate
perceptual experience, they argued, are incorrigible, that is to say,
we can directly verify that they correspond to the facts, and the truth
of other statements can then be tested by rneans of their logical
relations to these. Already a characteristic feature of the correspon-
dence theory - that truth líes in a relation between beliefs and the
world - is modified: the test of the truth of all but perceptual state-
 Theories of truth 95
ments derives from their relations with other statements, the percep-
tual ones, which are supposed to be verified by direct confrontation
with the facts. Neurath, however, raised doubts about the supposed
incorrigibility of 'protocols ', and having thus denied the possibility
of a direct check of even perceptual beliefs' correspondence to the
facts, held the only test of truth to consist of relations among beliefs
themselves. Our search for knowledge requires a constant readjust-
ment of beliefs, the aim of which is as comprehensive a belief set as
consistency allows. (This is strongly reminiscent of the 'method of
maxima and minima ' in James' epistemology (James 1907); Quine's
position in 'Two dogmas of empiricism' (1951), where he endorses
Neurath's metaphor of the process of acquisition of knowledge as
repairing a raft while afloat on it, is similar. Cf. Hempel 1935 for an
excellent account of the development of the positivists' view of truth,
and Scheffler 1967 ch. 5 for a lively 'blow by blow' report of the
controversy between Schlick and Neurath.)
Neurath's final position has much in common with Bradley's
account of the test of truth as 'system ', which he explains as requiring
both consistency and comprehensioeness of the belief set. And in
Bradley as in Neurath the appeal to coherence is connected with the
denial that our knowledge has any incorrigible basis in the judgments
of perception. However, Bradley's theory has intimate connections
with his absolute idealism. Briefly and roughly, reality, according to
Bradley, is itself essentially a unified, coherent whole. (Russell's
pluralistic logical atomist metaphysic was motivated by reaction
against the idealists' monism.) And while Bradley conceded something
to the idea of truth as correspondence to reality, he held that, strictly
speaking, nothing short of the fully comprehensive, consistent belief
set at which we aim is really true; at best, we achieve partial truth -
part of the truth is not fully true. The point of these remarks is to
bring out a point anticipated earlier (§ 1) - that the connections
between Bradley's view of truth and his view of reality are close
enough for it to be somewhat mísleading simply to regard him as
offering coherence as the test, while leaving correspondence as the
definition, of truth; rather, the explanation of the success of co-
herence as the test derives from an account of reality as itself
essentially coherent.
A persistent difficulty with the correspondence theory, as I observed
above (§:i) has been the diffi.culty of supplying a precise account of
'corresponds '. A similar problem dogs the coherence theory; it needs
96 Philosophy of logics

to be specified exactly what the appropriate relations between beliefs

must be for them to be 'coherent' in the required sense. U nsympa-
thetic critics of coherence theories - Russell, for example - have been
apt to assume that simple consistency is sufficient; Bradley, however,
was already insisting (as early as 1909, against criticism from Stout;
see Bradley 1914) that comprehensiveness as well as consistency is
required.
Rescher, who defends a coherentist epistemology ( coherence as the
test of truth) offers a detailed explication of the twin requirements of
'system'; consistency and comprehensiveness. The problem facing
the coherentist, as Rescher sees it, is to supply a procedure for
selecting, from incoherent and possibly inconsistent data ('truth-
candidates', not necessarily truths) a privileged set, the warranted
beliefs, those one is warranted in holding true. A 'maximal consistent
subset' (M.C.S.) of a set of beliefs is defined thus: S' is an M.C.S. of
S if it is a non-empty subset of S which is consistent, and to which no
member of S not already a member of S' can be added without
generating an inconsistency. But the data-set is lik.ely to have more
than one M.C.S.; this is the basis of Russell's criticism that coherence
cannot distinguish the truth from a consistent fairy tale. To avoid this
difficulty Rescher proposes that the M.C.S.s of the data-set be
'filtered' by means of a plausibility index, dividing data into those
which are, and those which are not, initially plausible, and thus
reducing the number of eligible M.C.S.s. However, this may be
insufficient to single out a unique M.C.S.; so Rescher recornmends
the adoption of the disjunction of those M.C.S.s perrnitted by the
plausibility filter.
Though Rescher's work has contributed significantly to the
detailed working-out of a coherentist epistemology, difficulties
remain. An obvious problem is the specification and justification of
the standards of plausibility (Schlick's appeal to the alleged íncor-
rigibility of protocols could be seen as an alternative response to a
related difficulty). A less obvious, but also important, difficulty is that
the recommended procedure is, so to speak, static in character: it tells
one how to selecta privileged, 'warranted', subset from an initial set
of data, but correspondingly underestimates the importance of seeking
new data. (Bradley's insistence that only the most fully comprehensive
belief set - the whole truth - is strictly speaking true could be seen as
a response to this difficulty.) Coherence will surely form part, but not
the whole, of a satisfactory epistemology.
 Theories o/ truth 97
Thus far, I have followed Rescher (with sorne qualifications in
Bradley's case) in taking coherence to be intended as a test of truth,
as playing an epistemological role, while allowing correspondence the
metaphysical part. (Cf. the large role played by coherence in Quine's
epistemology, from 1951 to 1970, with his adoption of the semantic
definition of truth, 1970 ch. 3). The pragmatists, however, challenge
this distinction with their characteristic criterial theory of meaning.

4 Pragmatic theories1
Peirce, James and Dewey offer characteristically 'pragmatic'
accounts of truth, which combine coherence and correspondence
elements.
According to the 'pragmatic maxim' the meaning of a concept is
to be given by reference to the 'practica!' or 'experimental' con-
sequences of its application2-'there can be no difference ' as James
put it (1907 p. 45) 'that makes no difference'. So the pragmatists'
approach to truth was to ask what difference it makes whether a belief
is true.
According to Peirce, truth is the end of inquiry, that opinion on
which those who use the scienrific method will, or perhaps would if
they persisted long enough, agree. The significance of this thesis
derives from Peirce's theory of inquiry. Very briefly: Peirce takes
belief to be a disposition to action, and doubt to be the interruption
of such a disposition by recalcitrance on the part of experience ;
nquiry is prornpted by doubt, which is an unpleasant state which one
tries to replace by a fixed belief. Peirce argues that sorne methods of
acquiring beliefs - the method of tenacity, the method of authority,
the a priori method- are inherently unstable, but the scientific method
enables one to acquire (eventually) stable beliefs, beliefs which will
not be thrown into doubt. For the scientific method, Peirce argues,
alone among methods of inquiry, is constrained by a reality which is
independent of what anyone believes, and this is why it can lead to
consensus. So, since truth is the opinion on which the scientific
method will eventually settle, and since the scientific method is con-
strained by reality, truth is correspondence with reality. It also
follows that the truth is satisfactory to believe, in the sense that it is
stable, safe from the disturbance of doubt.
1
This section draws upon Haack 1976c.
I
Peirce stressed the connection of 'pragmatic' with Knnt's use of
'pragmatische' for the empiricslly conditioned, James the connection
with the Greek 'praxiJ ', action.
98 Philosophy of logics

James' major contribution was an elaboration on this idea. The

advantage of holding true beliefs, he argued, was that one was thereby
guaranteed against recalcitrant experience, whereas false beliefs would
eventually be caught out(' Experience ... has ways of boiling ooer ... ',
1907 p. 145). James' account of the way one adjusts one's beliefs as
new experience comes in, maximising the conservation of the old
belief set while restoring consistency - strikingly like Quine's 1951
view of epistemology - introduces a coherence element. True beliefs,
James comments, are those which are verifiable, i.e. those which are,
in the long run, confirmed by experience.
Thus far, I have stressed the continuities between Peirce's and
James' views, but there are sorne differences which should be men-
tioned. First, while Peirce was a realist, James was inclíned towards
nominalism (cf. Haack 197 + ), and therefore embarrassed by the
possible-but-not-yet-realised verifications to which the view of truth
as verifiability committed him; consequently, although in principle he
allows that beliefs are true (false) though no one has yet verified
(falsified) them, in practice he is sufficiently persuaded of the point-
lessness of dwelling on this that he slips into speaking, inconsistently,
as if new truths come into existence when beliefs get verified. (The
idea that truth is made, that it grows, was taken up by the English
pragmatist, F. C. S. Schiller.) Second, James often speaks ofthe true
as being the 'good ', or the 'expedient' or the 'useful' belief ( e.g. 1907
pp. 59, 145). Unsympathetic critics (e.g. Russell 1908b, Moore 1908)
have taken James to be making a crass, not to say morally objection-
able, identification of the true with the congenial belief. The com-
ments which provoked this critica! fury, when taken in context, can
often be read, much more acceptably, as pointing to the superiority of
true beliefs as saje from falsification (cf. James' own defence, 1909
p. 192 - 'Above all we find consistency satisfactory'). But James is also
making another claim: that since at any given time the evidence
available to us may be insufficient to decide between competing
beliefs, our choice may depend upon such grounds as simplicity or
elegance (1907 p. 142); a claim which does have connections with
his 'will to believe ' doctrine.
Dewey adopts Peirce's definition as 'the best definition of truth'
( 1938 p. 3450). He prefers the expression 'warranted assertibility' to
'truth', and adds the thesis that it is precisely warranted assertibility
that characterises those beliefs to which we give the honorific title,
knowledge (cf. Ayer 1958). Dummett's view of truth, the direct
 Theori'es of truth 99
inspiration for which derives from the work of the later Wittgenstein
and from Intuitionism in the philosophy of mathematics, resembles
Dewey's in its stress on assertibility; see Dummett 1959.
The main theses of the pragmatic account can be summarised as
follows:
truth is:
the end of inquiry ) Peirce)
correspondence with reality
satisfactory ( stable) belief James
coherence with experience - Dewey
verifiability
what entitles belief to be called
'knowledge'

5 The semantic theory

Tarski's has been, of late, probably the most influential and
most widely accepted theory of truth. His theory falls into two parts:
he provides, first, adequacy conditions, i.e. conditions which any
acceptable definition of truth ought to fulfil; and then he provides a
definition of truth (for a specified formal language) which he shows
to be, by his own standards, adequate. Both parts of this programme
will be examined. The detailed statement of the theory is to be found
in Tarski 1931; 1944- is a good introduction.
It isn't hard to see why Tarski's theory should have been so influ-
ential. For one thing, his adequacy conditions on definitions of truth
promise a kind of filter to discriminate, among the embarrassingly
numerous theories of truth, those which meet minimal conditions of
acceptabílity, which therefore have sorne prospect of success. Further-
more, the method employed in Tarski's definition of truth can be
applied to a large class of formal languages. But the very features of
Tarski's theory which contribute most to its appeal also, as we shall
see, create problems for it: can Tarski's adequacy conditions be given
independent motivation? and: have his methods any interesting appli-
cation to the problem of truth for natural Jangua ges?

Adequacy conditions on definitions of trutk

The problem which Tarski sets himself is to give a definition
of truth which is both material/y adequate and formally correct; the
first of these conditions sets limits on the possible content, the second
on the possible form, of any acceptable definition.
100 Philosophy of logics

Material adequacy ·
Tarski hopes that his definition will 'catch hold of the actual
meaning of an old notion' (1944 p. 53). However, the 'old' notion of
truth is, Tarski thinks, ambiguous, and even doubtfully coherent. So
he restricts his concern to what he calls the 'classical Aristotelian con-
ception of truth ', as expressed in Aristotle's dictum:
To say of what is that it is not, or of what is not that it is,
is false, while to say of what is that it is, or of what is not
that it is not, is true.
And he proposes, as material adequacy condition, that any acceptable
definítion of truth should haoe as consequence all instances of the (T)
schema:
(T) S is true iff p
where 'p' can be replaced by any sentence of the language for which
truth is being defined and 'S' is to be replaced by a name of the
sentence which replaces 'p'. An instance of (T) would be, e.g.:
'Snow is white' is true iff snow is white
where the sentence on the right-hand side is referred to by its
'quotation-mark name' on the left-hand side.
Tarski emphasises that the (T) schema is not a definition of truth -
though in spite of his insistence he has been misunderstood on this
point. It is a material adequacy condition: all instances of it must be
entailed by any definition of truth which is to count as 'materially
adequate'. The point of the (T) acherna is that, if it is accepted, it
fixes not the intension or meaning but the extension of the term 'true'.
For suppose one had two definitions of truth, D1 and D2, each of
which was materially adequate. Then D1 would entail all instances of:
S is true¡ iff p
and D2 all instan ces of:
S is true, iff p
so that D1 and D2 are co-extensive, Or, to put essentially the same
point in another way, the material adequacy condition would rule out
certain definitions of truth, those, that is, which did not entail
instances of the (T) schema.
But exactly what kinds of definition will the material adequacy
 Theories of truth IOI

condition rule out? In answering this question I shall use a weakened

version of the criterion: not that all instances of the (T) scherna
be deducible from any acceptable truth definition (Tarski's version),
but that the truth of all instances of the (T) schema be consisteni with
any acceptable truth-definition, The reason for this modification is
simply that the weakened adequacy condition is much more readily
applicable to non-formal definitions of truth. Now it is to be hoped -
and perhaps even expected - that it will allow the sorts of definition
which have been seriously proposed, and disallow what one might
call 'bizarre' theories. But matters turn out rather oddly. Consider
the following definition of truth, which seems to me definitely bizarre:
a sentence is true iff it is asserted in the Bible. N ow it might be sup-
posed that this definition (I shall call it 'DB' for short) does not entail
all instan ces of the (T) schema, not, for instance:

'Warsaw was bombed in World War II' is truej, iff

Warsaw was bombed in World War II.

Now it is indeed the case that someone who did not accept DB might
deny:
'Warsaw was bombed in World War II' is asserted in
the Bible iff Warsaw was bombed in World War II.

But further reflection makes it clear that a proponent of DB could

perfectly well maintain that his definition does entail all instances of
(T); he may allow that 'Warsaw was bombed in World War II' is
true, but insist that it is asserted in the Bible (in an obscure passage in
Revelation, perhaps], or if he agrees that 'Warsaw was bombed in
World War II' is not asserted in the Bible, he will also, if he is wise,
maintain the falsity of the right-hand side of the above instance of
the schema. So, rather surprisingly, Tarski's material adequacy con-
dition cannot be relied upon to be especially effective in ruling out
bizarre truth-definitions.
The material adequacy condition does, however, apparently rule
out a certain important class of truth theories, those, that is, according
to which sorne sentences (statements, propositions, wffs or whatever)
are neither true nor false. For suppose 'p' to be neither true nor false;
then the left-hand side of:
'p' is true iff p
102 Philosophy of logics
will be, presumably, false, while the right-hand side will be neither
true nor false. So the whole biconditional will be false, or at any rate
untrue. (This argument could, however, be avoided if one were pre-
pared to allow that metalinguistic assertions such as ''p' is true'
might themselves be neither true nor false.) It is arguable that Tarski's
material adequacy condition would rule out at least sorne versions of
the coherence theory; arguably it would not rule out a pragmatist
theory, since the pragmatist view of meaning would rule meaningless
any sentences which are neither verifiable nor falsifiable, so that there
could be no meaningful but truth-valueless sentences. It certainly
seems rather extraordinary to rule non-bivalent theories of truth out
of court.
Toe idea behind Tarski's material adequacy condition is, pre-
sumably, that the truth of the (T) schema is so certain and obvious
that it is proper that one should feel confident in rejecting any theory
of truth which is inconsistent with it. For myself, I fi.nd the initial
certainty and obviousness of the (T) schema somewhat modified when
it turns out that not only sorne of the seriously propounded theories
of truth, but also sorne very bizarre theories, are consistent with it,
while sorne other serious theories are inconsistent with it (but see
Davidson 1973 for a defence of 'convention T ').

Formal correctness
The formal requirement which Tarski lays down concerns
the structure of the language in which the definition of truth should
be given, the concepts which may be employed in the definition, and
the formal rules to which the definition must conform.
It is notorious that semantic concepta, incautiously handled, are apt
to give rise to paradoxes ( e.g. the Liar - 'This sen ten ce is false';
Grelling's paradox - '' not true of itself' is true of itself iff it is
not true of itself', and so forth). Tarski investigates the Liar paradox
in sorne detail, and argues that the antinomy arises frorn the
assumptions:
(i) That the language used contains, in addition to its
expressions, (a) the means of referring to those expressions
and (b) such semantic predicates as 'true' and 'false'.
Such a language Tarski calls 'semantically closed '.
(ü) That the usual logical laws hold.
Being unwilling to reject assumption (ii), Tarski concludes that
 Theories of truth 103

a formally correct definition of truth should be expressed in a language

which is not semantically closed.
Specifically, this means that the definition of truth-in-O, where
O is the object language (the language/or which truth is being defined),
will have to be given in a metalanguage, M (the language in which
truth-in-O is defined). The definition of truth will have to be, Tarski
argues, relative to a language, for one and the same sentence may be
true in one language, and false, or meaningless, in another. The
danger of the semantic paradoxes can be avoided by resort to a meta-
language; the Liar sentence, for instan ce, will then become the harrn-
less "Thís sentence is false-in-O', which is, of course, a sentence of M,
and consequently not paradoxical. The object/metalanguage distinc-
tion is, of course, a relative one, and a whole hierarchy of languages
would be required to define truth at every level. Since all equivalences
of the form (T) must, by the material adequacy condition, be implied
by the definition of truth, M must contain O or translations of all
sentences of O as part, plus the means to refer to expressions of O;
for instances of (T) have, on the left-hand side, an expression denoting
a sentence of O, and, on the right-hand side, a sentence of O or a
translation of a sentence of O. Notice that, in specifying, in the meta-
metalanguage, that the metalanguage, M, should contain either the
object language, O, itself, or a translation of each sentence of O,
semantic notions are employed (explicitly in the latter case, and
implicitly in the former, since M must contain the same expressions
of O with the same interpretations as they have in O).
It is also required that the structure of O and M should be 'formally
specifiable '. For in order to define 'true-in-O' it will be essential to
pick out the wffs of O, since these are the items to which 'true-in-O'
applies. (This is one of the reasons which Tarski gives for feeling
sceptical about the possibility of defining "true-in-English ' - or 'true'
for any natural language; the sentences of natural languages are not,
he thinks, formally specifiable. Later followers of Tarski, notably
Davidson, feel more optimistic on this point. It is one I shall need to
investigate more closely.)
Tarski also requires that 'the usual formal rules of definition are
observed in the metalanguage' ( 1944 p. 6i ). These rules include:
(i) no free variable may occur in the definiens which does not
also occur in the definiendum
ruling out e.g. "F» =df(x+y = o)', and
104- Philosophy of logics

(ii) that no two occurrences of the same variable may occur

in the definiendum
ruling out e.g. 'Fxx = df Gx'. Condition (i) prevents definitions
which could lead to contradiction; condition (ii) prevents definitions
in which the definiendum is ineliminable ( cf. Suppes 1957 ch. 8).
Any acceptable definition of truth must, then, according to Tarski,
satisfy both the material adequacy and the formal correctness condi-
tions. He gives a definition, and shows that it is, by these standards,
acceptable.

Tarski.'s dejinition of truth

It might be thought that the (T) schema, though not itself a
definition of truth, pro vides an obvious way of giving such a definition.
Tarski himself points out that one could think of each instance of (T)
as a partial definition of truth, in that each instance specifies the
truth-conditions of sorne one specific sentence; so that a conjunction
of all instances of the (T) schema, one for each sentence of O, would
constitute a complete definition. Tarski, however, argues that it is not
possible to give such a conjunctive definitíon, for the number of
sentences of a language may be infinite, and in this case it is impos-
sible actually to give all the required instances of the (T) schema.
Neither, Tarski argues, can the (T) schema be turned into a definition
of truth by universal quantification. It might be supposed that, using
on the left-hand side a quotation mark name of the sentence used on
the right-hand side, one could straightforwardly generalise to obtain:
(O) (p) ('p' is true., iff p)
which would apparently constitute a complete definition, and one,
furthermore, guaranteed to be materially adequate, since all instances
of (T) are instances of it. But Tarski rejects this suggestion because
he believes that the result of quantifying into quotation marks is
meaningless. For, according to Tarski (and also according to Quine),
the expression obtained by writing quotation marks around an expres-
sion is an indivisible unit, analogous to a proper name, so that:
Snow is white
is no more part of:
'Snow is white'
than (to use Quine's example) 'rat' is of 'Socrates'. Tarski concedes
 Theories of truth 105

that if it were feasible to regard quotation as a function, then (D)

would be no less well-formed than e.g.:
(x) (x2 = x.x)
He thinks, however, that there are overwhelming objections to
treating quotation as a function, and, in consequence, that (D) is no
more well-formed than e.g.:

(x) (Texas is large)

So Tarski thinks that the (T) schema not only is not, but also
cannot be turned into a definition of truth. So he constructs his own
definition by a more roundabout route. He takes it as a desideratum
that no semantic terms should be taken as primitive, so that any
semantic notion in terms of which 'true' is defined should itself
previously be defined. Since he is to define 'true' using the concept
of satisfaction, which is a semantic one, this means that he must first
define 'satisfies '.

Informal account
The procedure is as follows:
(a) specify the syntactic structure of the language, O, for
which truth is to be defined

(b) specify the syntactic structure of the language, M, in which

truth-in-O is to be defined; M is to contaín
(í) either the expressions of O, or translatíons of the
expressions of O
(ii) syntactical vocabulary, including the names of the
primitive symbols of O, a concatenation sign (for
forming 'structural descriptions' of compound expres-
sions of O), and variables ranging over the expressions
of O
(iii) the usual logical apparatus
(e) define 'satisfies-in-0', and

( d) define 'true-in-O' in terms of 'satisfies-in-0'

Why does Tarski first define 'satísfies '? Well, first, because he con-
siders it desirable to employ, in hís definition of truth, no semantic
primitives; for he considers that semantic notions are non e of them,
J 06 Philosophy of logics

pre-theoretically, sufficiently clear to be safely employed. But why

'satisfies'? This is a suitable notion in terms of which to define 'true'
beca use closed, compound sentences are formed out of open sentences,
rather than closed, atomic sentences. For exarnple, '(3x) Fx V Gx' is
formed out of 'Fx' and 'Gx' by the opera ti o ns of disjunction and
existential quantification; and the open sen ten ces 'Fx' and 'Gx' are
not true or false, but satisfied, or not, by objects. The definition of
satisfaction is recursioe - that is, definitions are given first for the
simplest open sentences, and then the conditions are stated in which
compound open sentences are satisfied. (The definition could, how-
ever, be turned into an explicit one.) This procedure will provide
a definition of truth applicable to ali sentences of O.
'Satisfies': open sentences are not true or false, they are satisfied,
or not, by certain things, pairs of things, triples of things, etc. For
instan ce: 'x is a city' is satisfied by London; 'x is north of y' is satis-
fied by (London, Exeter); 'x is between y and z' is satisfied by
(London, Exeter, Edinburgh) ... etc.('( ... , ... )' indicates the ordered
n-tuple of the n items which appear between the pointed brackets.)
The order of the items is obviously important, since {London,
Exeter) satisfies 'x is north of y' but (Exeter, London) does not.
Satisfaction is a relation between open sentences and ordered n-tuples
of objects. To avoid the difficulties arising from the fact that open
sentences may have 1, 2, ... or any number of free variables, Tarski
defines satisfaction as a relation between open sentences and infinitt
sequences, under the convention that 'F(x1 ... x,,)' is to be satisfied
by the sequence {01 ... On, 0,.+1 ... ) just in case it is satisfied by the
first n members of the sequence; subsequent members are ignored.
The negation of an open sentence S1 will be satisfied by just those
sequences which do not satisfy S1; and the conjunction of S1 and S2 by
just those sequences which satisfy S1 and satisfy S2• The existential
quantification of an open sentence will be satisfied by a sequence of
objects just in case there is sorne other sequence of objects, differing
from thefirst in at mosttheith place (where the ith is thevariable bound
by the quantifier) which satisfies the open sentence resulting from drop-
ping the quantifier, For instance, the sequence (England, London,
Edinburgh ... ) satisfies '(3x) (x is a city between y and z)' because
e.g. the sequence {York, London, Edinburgh) satisfies 'x is a city
between y and s',
'T�': Closed sen ten ces are special cases of open sen ten ces, those,
namely, with no free variables. The first member of a sequence, and
 Theories of truth 107

ali subsequent rnernbers, are irrelevant to whether or not the sequence

satisfies a o-place open sentence, i.e. a closed sentence. So Tarski
defines a sentence as true just in case it is satisfied by ali sequences, and
as false just in case it is satisfied by none. This procedure rnay be rnade
less rnysterious by considering an exarnple. The z-place open sen-
tence 'x is north of y' is satisfied by e.g. ali sequences (Edinburgh
London, ... ), whatever their third and subsequent rnernbers. The
r-place open sentence 'x is a city' is satisfied e.g. by all sequen ces,
(Edinburgh, ... ) whatever their second and subsequent rnernbers.
And the (true) o-place open sentence '(3x) (x is a city)' is satisfied by
ali seq uences ( ... , ... , ... ), whatever their first and subsequent mem-
bers; for there is a sequen ce, (Edinburgh, ... ) for instance, which
differs from any arbitrary sequence in at rnost the first place, and
which satisfies 'x is a city'. Any closed sentence will be satisfied by
ali sequences or by none, and can't be satisfied by sorne and not
others. Consider a rather austere language: first-order predicate
calculus without singular terms. In the sirnplest case, a closed sentence
is forrned by existential quantification of a r-place open sentence.
Such an existentially quantified sentence is satisfied by an arbitrary
sequence only if there is another sequence, differing frorn it in the
first place at rnost, which satisfies the r-place open sentence which
results frorn dropping the initial existential quantifier; and so, if the
existential sentence is satisfied by any sequence, it will be satisfied by
e,;ery sequence. So a closed existential sentence will be satisfied either
by ali sequences or by none. The negation of a closed exísrential
sentence, by the negation clause of the satisfaction definition, will be
satisfied by a sequence iff the negated sentence is not satisfied by that
sequence and so, once again, will be satisfied either by ali sequences
or by non e; and sirnilarly for the conjunction of two closed existential
sentences, which will be satísfied by a sequence iff both conjuncts are
satisfied by that sequence, and so, also satisfied by ali sequences or
by none. But why is 'true' defined as 'satisfied by ali sequences', and
'false' as 'satisfied by none'? Well, consider again the closed sentence
'(3x) (x is a city)': let X be an arbitrary sequence of objects. By the
clause of the definitíon of satisfaction which covers existentially
quantified sentences, X satisfies this sentence iff there is sorne
sequence Y differing frorn X in at most the first place which satisfies
'x is a city'; now an object o satisfies 'x is a city' just in case o is a
city, so there is such a sequence just in case there is sorne object
which is a city. Thus '(3x) (x is a city)' is satísfied by ali sequences
108 Philosophy o/ logics

just in case sorne object is a city. (Consult Rogers 1963 for further
informal discussion of Tarski's definition.)
Two features of Tarski's definition deserve explicit mention at
this point. First, it imposes an objectual interpretation of the quanti-
fiers; as the previous example indicates, '(3x) Fx' is true iff sorne
object is F. A substitutional interpretation would avoid the need for
the detour via satisfaction, for it would permit truth of quantified
sentences to be defined directly in terms of the truth of their substitu-
tion instances (cf. ch. 4 §1). Second, in his original paper, Tarski
gives an absolute rather than a model-theoretic definition; 'satisfies ',
and hence 'true', is defined with respect to sequences of objects in
the actual world, not with respect to sequences of objects in a model
or 'possible world' (e.g. 'there is a city north of Birmingham' is true,
absolutely, but false in a model in which the domain is, say, {London,
Exeter, Birmingham, Southampton}; cf. pp. 1 15, 122 below ).1
Formal account
Tarski gives his definition of truth for a class calculus (the
object language), and uses a formalised metalanguage. 1 shall give,
instead, a definition of truth for a more familiar object language, the
first-order predicate calculus, and I shall use English plus the object
language (cf. (b)(i), p. 105) as metalanguage. This truth-definition
will, however, follow Tarski's in all essentials. (lt follows Quine's
account in 1970 ch. 3 rather closely.)
Syntax o/ O
The expressions of O are:
variables: x1, x2, x8 •.. etc.
predicate letters: F, G ... etc. (each taking a given number
of arguments)
sentence connectives: - , &
quantifier: (3 ... )
brackets : ( , )
1
In 1957 Tarski and Vaught give a model-theoretic definition. The
significance attached to the difference between absolute and model-
theoretic definítions will depend, in part, on one's attitude to possible
worlds (see pp. 190 ff, below). Those who think of the actual world
as just one possible world among others will think of the absolute
definition as simply a special case of a model-theoretic definitícn,
However, since a model-theoretic definition uses semantic prímitives
�the notíon of the interpretation of expressions) in the metalanguage,
rt does. not satisfy ali the constraints Tarsk.i used in his 193 1 paper ;
�nd this seems to sorne (Davidson for example; see below) to be an
Impottant reason to prefer an absolute definition.
 Theories of truth

In terms of this austere primitive vocabulary, of course, the other

truth-functions and the universal quantífier can be defined. I am also
assuming that singular terms have been eliminated. The advantage of
choosing such a minimal vocabulary is, as will become apparent, that
it much reduces the work which has to go into the truth definition.
The atomic sentences of O are those strings of expressions which
consist of an n-place predicate followed by n variables.
(i) Ali atomic sentences are well-formed formulae (wffs)
(ii) If A is a wff, -A is a wff
(iii) If A, B are wffs, (A & B) is a wff
(iv) If A is a wff, (3x) A is a wff
(v) nothing else is a wff

Definition of 'satisfies'
Let X, Y range over sequences of objects, A, B over sentences
of O, and let Xi denote the ith thing in any sequence X.
Then satisfaction can be defined for atomic sentences, by giving
a clause for each predicate of the language.
I. for r-place predicates:
for ali i, X: X satisfies 'Fxi' iff Xi is F
For z-place predicates:
for ali i, X: X satisfies 'Gxix/ iff Xi and
X1 stand in the relation G
and so on for each predicate.
2. for ali X, A: X satisfies ' -A' iff X does not satisfy 'A'
3. for ali X, A, B: X satisfies 'A & B' iff X satisfies A and
X satísfies B
4. for ali X, A, i: X satisfies '(3xi) A' iff there is a sequence
Y such that X1 = Yj for ali j =I i and Y satisfies 'A'
(Notice how each clause of the definition of satisfaction corresponds
to a clause in the definition of a wff. This is why it is so convenient to
work with minimal vocabulary.) A closed sentence is a wff with no
free variables; closed sentences will be satisfied either by ali sequences
or by none.

Definition of 'true': a closed sen ten ce of O is true iff it is satisfied by

ali sequences.
Tarski shows that his definition is both materially adequate and
formally correct. He also shows that it follows from his definition of
11o Philosophy of logics

truth that of each pair consisting of a closed sentence and its negation
one, and only one, is true. This was to be expected in view of the
fact, already observed, that the material adequacy condition rules out
non-bivalent theoríes of truth.

6 Commentary on the semantic theory

Tarski's theory has the distinction of having been criticised
both for saying too little:
the neutrality of Tarski's definítion1 with respect to the
competing philosophical theories of truth is sufficient to
demonstrate íts lack of philosophical relevance. (Black 1948
p. 260)
and for saying too much:
The Tarskian theory belongs to factual rather than
conceptual analysis Tarski's theory has plenty of meat
to it, whereas a correct conceptual analysis of truth has
very little. (Mackie 1973 p. 40)
The question of the philosophical significance of Tarski's theory is
evidently a hard one; 1 shall tackle it in three stages: first by discuss-
ing Tarski's own estimate of his theory's significance, and then by
discussing the use made of the theory by two writers - Popper and
Davidson - who have more ambitious hopes of it than Tarski himself.

(a) Tarski's own estimate

Tarski expresses the hope (194+ pp. 53-4) that his dcfinition
will do justice to the Aristotelian conception of truth, but sees little
point in the question whether that is the 'correct' concept, offering,
indeed, to use the word 'frue' rather than 'true' should the decision
go against him on that issue (p. 68).
Tarski is aleo modest about the epistemological pretensions of his
theory; he doesn't really understand, he says, what the 'philosophical
problem of truth' might be (p. 70), but anyway:
we may accept the semantic conceptions of truth without
giving up any epistemological attitude we may have had,
1
Here Black apparently confusea the material adequacy condition with
the defínition, though elsewhere in the same paper he maltea the
distinction clearly enough,
I
The context suggeata that Tankí ia here concemed primarily with hís
material adequacy condition.
 Theories of truth III

we may remain naive realista or idealista, empiricists or

metaphysicians ... The semantic conception is corn-
pletely neutral toward ali these issues, (p. 71)

Field suggests (1972) that Tarski may have attached metaphysical

importance to the constraint on which he insisted (but cf. p. 108n),
that truth be defined without the use of semantic primitives: a con-
straint he justified (1931 pp. 152-3) by urging the superior clarity of
syntactic notions, A comment in another paper, "The establishment
of scientific semantics ', suggests that he may also have had a deeper
significance in mind; after repeating that the use of semantic
primitives would threaten clarity, he goes on:

this method would arouse certain doubts from a general

philosophical point of view. It seems to me that it would
then be difficult to bring this method into hannony with
the postulates of the unity of science and of physicalism
(since the concepts of semantics would be neither logical
nor physical concepts). (1936 p. 406)

Field's conjecture is that Tarski's intention was to bring semantics

into line with the demands of physicalism, the thesis that there is
nothing but physical bodies and their properties and relations; and
that this is to be achieved by defining such non-physical concepta as
truth and satisfaction. It is confirmed by a passage, 1944 pp. 72-4,
where Tarski defends the semantic conception of truth against the
criticism that semantics involves metaphysical dementa, by stressing
that his definition uses as primitives only logical terms, expressions of
the object language, and names of those expressions. The further
question, whether Tarski's theory indeed has this significance, is
tricky. Field believes that Tarski does not really succeed in reducing
semantics to physicalistically acceptable primitives. Tarski defines
satisfaction for complex open sentences recursively, in terma of satis-
faction for atomic open sentences, but he defines satisfaction for
atomic open sentences enumeratioely, a clause for each primitive predi-
cate of the object language (as it were, 'X satisfies 'x, is a city' iff X,
is a city, X aatisfies 'x, is north of xJ' iff X, is north of XJ ... ' and
so forth). Since Field holds that a succeseful reduction requires more
than extensional equivalence of definiens and definiendum, which is all
Tarski's definition guarantees, he finds that Tarski does not, as he
hoped, vindicate physicalism. It seems worth observing that there is
112 Philosophy o/ logics

a strong tendency for physicalists to be extensionalists, and sorne

reason, therefore, to suppose that Tarski would have thought exten-
sional equivalence a sufficient constraint. The question rernains, of
course, whether extensional equivalence really is a sufficient con-
straint upon reductions, or whether, as Field suggests, sorne stronger
requirernent is proper.
(b) Popper's claims on behalf o/ Tarski's theory
Popper welcornes Tarski's theory as having:
rehabilitated the correspondence theory of absolute or
objective truth. . . He vindicated the free use of the
intuitive idea of truth as corresponden ce to the facts ...
(1960 p. 224-)
and he uses Tarski's ideas in developing his own account of the role
of truth as a regulative ideal of scientific inquiry.1

Is Tarski's a correspondence theory?

According to Popper, Tarski has supplied just what was lacking with
the traditional corresponden ce theories - a precise sen se for 'corre-
sponds' (196o p. 223, 1972 p. 320). Initially, at least, this is puzzlíng,
for Tarski explicitly cornrnents that the correspondence theory is un-
satisfactory (194-4- p. 54-), and observes that he was 'by no rneans
surprised' to learn that, in a survey carried out by Ness, only 15 %
agreed that truth is correspondence with reality, while 90% agreed
that 'It is snowing' is true if and only if it is snowing (194-4- p. 70; and
see Ness 1938).
So what is it that leads Popper to think of Tarski as having vindi-
cated the correspondence theory? Sorne comments (e.g. 1960 p. 224)
suggest that what he specifically has in rnind is Tarski's insistence on
the need for a rnetalanguage in which one can both refer to expressions
of the object language and say what the object language says. It is as
if he thinks of the left-hand side of each ínstance of the (T) scherna,
such as:
'Snow is white' is true iff snow is white
as referring to the language, and the right-hand síde to the facts. But
this seerns a pretty inadequate reason for tak.ing Tarski's to be a
correspondence theory, for the material adequacy condition, though
1
Thia section amplifiea and modifiea sorne pointa from Haack 1976d;
and cf. Sellara r967 ch. 6 for sorne pertinent díscuAion.
 Theories of truth 113

its role is to rule out sorne definitions, certainly does not single out
the correspondence theory as uniquely correct; presumably it permits,
for instance, a redundancy definition such as Mackie's:
(p) (the statement that p is true iff p)
lt is just for this reason that Tarski himself stresses the epistemo-
logical neutrality of the (T) schema.
However, though Popper does not explicitly refer to thern, there
are features of Tarski's definition of truth which are rerniniscent of
correspondence theories, A difficulty here is that it isn't very clear
what is required for Tarski's to count as really a version of the corre-
spondence theory; and it is aggravated by Popper's insistence that
until Tarski there had been no genuine, no satisfactory, correspon-
dence theory. Still, one can rnake sorne progress by cornparing Tarski's
definition first with the logical atornist version given by Russell and
Wittgenstein, and then with Austin's version.
Tarski defines truth in tenns of satisfaction, and satisfaction is a
relation between open sentences and sequences of objects; the
account of satisfaction bears sorne analogy to Wittgenstein's view of
truth as consisting in the correspondence between the arrangernent of
narnes in a proposition and the arrangernent of objects in the world.
On the other hand, Tarski 's definition of truth rnakes no appeal to speci-
fic sequences of objects, for truesentences aresatisfied by ali sequences,
and false sentences by none. It is syrnptornatic that logical as well as
factual truth is embraced by Tarski's definition; it is surely less plausi-
ble to suppose that logical truth consists in correspondence to the facts
than that 'factual' truth does. Two historical observations seern called
for here: first, that Wittgenstein thought that quantified wffs could be
understood as conjunctions/disjunctions of atornic propositions, and
that if this were indeed so, Tarski's detour through satisfaction would
be unnecessary; and secondly, that Russell allowed 'logical facts '.
Tarski's exploitation of the structure of sentences in the recursive
definition of satisfaction is, then, an analogy wíth Russell's and
Wittgenstein 's gloss on 'corresponds '. It is equally a disanalogy with
Austin's account. Austin insists that staternents, not sentences, are
the prirnary truth-bearers. This has at least two relevant consequences:
Tarski ignores problerns raised by sentences containing indexicals
such as '1' and 'now', upon which Austin concentrates; and while
Tarski's definition of satisfaction relies on the syntactic structure of
open sentences, Austin's account of correspondence stresses its
114 Philosophy o/ logics

purely conventional, arbitrary character - in another language, the

statement that nuts could, he says, be true in just the circumstance
that the statement in English that the National Liberals are the
people's choice is true.1 There is, however, one point of analogy
which deserves mention. Austin's account, 1 suggested earlier, avoids
locating correspondence in the too intimate connection between the
statement that p and the fact that p, explaining it rather as consisting
in the situation to which the statement th.at p refers being of the kind
which the statement says it is. Here one can see, without too severe
a strain, a resemblance to Tarski's enumerative account of satisfac-
tion for atomic open sentences: for example, X satisfies "x¡ is white'
iff the ith thing in the sequence X is white.
So: Tarski does not regard himself as giving a version ofthe corre-
spondence theory, and his material adequacy condition is neutral
between correspondence and other definitions. However, Tarski's
definition of satisfaction, if not of truth, bears sorne analogy to
correspondence theories: the clauses for atomic open sentences to
Austin's, the clauses for molecular open sentences to Russell's and
Wittgenstein's version.

1s Tarski' s theory "absolute' and "objectioe'Y

Whether or not one considers the affinities strong enough for Tarski's
to count as a version of the correspondence theory, it is worth asking
whether the semantic definition of truth has, anyway, what Popper
considers to be the major virtues of the correspondence theory, ita
'absolute' and 'objective' character.
Tarski stresses that truth can be defined only relatioely to a
/anguage - what he defines is not 'true' (period), but 'true-in-O'.
This is for two reasons; that the definition must apply to sentences
(which, unlike such extralinguistic items as propositions, have the
syntactic structure which it exploits) and one and the same sentence
can be true in one language and false, or meaningless, in another; and
that only a hierarchy of object language, metalanguage, and meta-
metalanguage can avoid the semantic paradoxes. In this sense, there-
fore, Tarski's is not an absolute, but a relative, truth definition.
1
Here, then, is a case where the issue about truth-bearers acquires a real
significance. (1 shall not resiat the temptation to draw attention to
Auatin'a complaint (1950 p. 30), that hia fellow-symposiast, Strawaon,
had failed to make the crucial distinction between sentence and state-
ment.) 1 •hall touch on the question, how Tanki's theory miaht be
adapted to deal with indexical aentencn, in the aection on Davidaon.
 Theories o/ truth II5
Popper, however, who is apt to take a somewhat cavalier attitude to
the question oftruth-bearers (1972 pp. 11, 45, 319n) is unconcerned
with this sense of 'absolute '. He takes no interest, either, in the fact
that Tarski's original definition is absolute rather than model-
theoretic.
Popper seems, rather, to equate 'absolute' and 'objective', con-
trasting both with 'subjective ', that is, 'relative to our knowledge or
belief'. In this respect Popper believes the correspondence theory to
be superior to
the coherence theory ... [which] mistakes consistency for
truth, the evidence theory ... [which] mistakes 'known to
be true' for 'true', and the pragmatist or instrumentalist
theory [which] mistakes usefulness for truth. (1960 p. 225)
I needn't comment, 1 think, on the accuracy of Popper's characterisa-
tion of the rival theories; anyhow, the core of his argument, fortu-
nately, does not depend on these details. The rival theories, Popper
argues, are founded on the 'widespread but mistaken dogma that a
satisfactory theory should yield a criterion of true belief' ( 1960 p. 225).
And a criterial theory of truth is subjective because it cannot allow the
possibility of a proposition's being true even though no one believes
it, or false even though everyone believes it,
What exactly does Popper find objectionable about criterial truth-
theories? Popper doesn't make this very clear; but I thin.k the
problem can be focussed. The crucial difficulty lies, not in the
attempt to supply a criterion of truth, in itself, but in the adoption of
a criterial theory of the meaning of 'true'. (His attitude is perhaps
clearest in the appendix to the 1961 edition of vol. 2 of The Open
Society and its Enemies.) lf one gives the meaning of 'true' in terms of
our criteria of truth, one cannot leave room for the possibility that
a proposition be false though it passes our tests of truth, or true
though it fails them. This is a particular problem for the pragmatista,
since it poses a threat to their official fallibilism; though there is still
room for mistakes in the application even of infallible tests of truth.
lnfallibilism in itseif is not subjectivist; but the further claim that to
say that a proposition is true (false)just means that it passes (fails) our
tests poses a threat to objectivism.
Tarski expressly disclaims the aspiration to supply a criterion of
truth ( 1944 pp. 71-:2); and his definition certainly makes no reference
to our tests of truth. (lronically, the passage in which Tarski draws
116 Philosophy of logics

attention to these features is intended as a rebuttal of the 'objection'

that his is a kind of correspondence theory which involves logic in
'a most uncritical realism' !)
So Tarski's is an objective theory in Popper's sense. But why does
Popper attach so much importance to this point? The explanation líes
in the epistemological use to which he proposes to put the concept
of truth.

Truth as a regulatioe ideal: verisimilitude

Popper describes himself as a 'fallibilistic absolutist': fallibilist
because he denies that we have any guaranteed method of acquiring
knowledge; absolutist beca use he insists that there is such a thing as an
objective truth to which scientific inquiry aspires. Tarski's theory is
to supply a suitably objective account of this 'regulative ideal' of
science.
This requires, of course, that Tarski's theory be applicable to the
languages - presumably, fragments, more or less completely regi-
mented, of natural and mathematical languages - in which scientific
theories are expressed. I shall not here discuss the questions raised by
this requirement; partly because Tarski himself expresses (1944- p. 74)
a cautious optimísm about the applicability of his work to the
empírica! sciences, and partly because in the next section, when I dis-
cuss Davidson's use of Tarski's work, 1 shall have to consider the
reasons Tarski gives for doubting whether his methods apply to
'colloquial' language.
According to Popper, the business of science is to devise and test
conjectures; scientists can't be confident that their current con-
jectures are true, nor even that they will ever reach the truth or would
know, if they did reach the truth, that they had. But if truth is to be
not just an ideal, but a guiding or 'regulative' ideal, it should be pos-
sible to tell whether, as one theory replaces another, science is getting
closer to the truth. So Popper's problem is to explain in what sense,
of two theories both of which may be false, one can be closer to the
truth than another. His solution is his extension of Tarski's ideas in
the theory of 'verisimilitude ', or truth-likeness.
Popper's account of verisimilitude goes:

Assuming that the truth-content and the f'alsity-content of

tt.oo theorie« ti and t1 are comparable, we can say that t1 is
more closely similar to the truth ... than ti if and onJy if nther:
 Theories of truth

(a) the truth-content but not the falsity-content of t2 exceeds

that of t1
[or]
(b) the falsity-content of t1 but not its truth-content
exceeds that of t2• (1963 p. 233)

The truth- (falsity-) content of a theory is the class of all and only its
true (false) consequences. The truth- or falsity-content of one theory
can exceed the truth- or falsity-content of another only if its truth-
or falsity-content set-theoretically includes the other's, so this
account applies only to theories which overlap in this way. Popper
also suggests (1963 pp. 393-6, 1972 pp. 51, 334) measures of truth-
and falsity-contents in terms of logical probability, so that any two
contents can be compared. But I shall concentrate on the former,
'qualitative ', rather than the latter, 'quantitative' version.
The definition of verisimilitude cannot show that science does
progress towards the truth: but Popper hopes (1972 p. 53) that it
supports his falsificationist methodology, which recommends that one
choose the more falsifiable conjecture, the one with more content, for
a theory with more content will have greater verisimilitude, unless,
Popper adds, it has more falsity-content as well as more truth-
content.
However, it has been shown 1 that a theory t2 has greater veri-
similitude than another, t1, in accordance with Popper's (a) and (b)
only if t2 is a true theory from which the truth-content of t1 follows.
This means that Popper's definition of verisimilitude does not apply to
comparisons betueen theories both of which are fa/se; but that, of course,
was the principal objective of the theory, which therefore fails of its
epistemological purpose. This failure is, 1 think, important to the
question of the feasibility of fallibilistic absolutism ( see Haack
1977b); and it should also, to my mind, support Tarski's rather
modest, as against Popper's rather more ambitious, assessment of the
epistemological significan ce of the semantic theory of truth.
1 Miller 1974; and cf. Tichy 1974 and Harris 1974. Very brieñy, Miller's
strategy is first to show, if t1 and t1 are comparable by truth-content,
how they are also comparable by falsity-content; and then to show
that for ta to be nearer the truth than t., ta must be a true theory from
which the truth-content of t1 follows, since otherwise ta will exceed 11
in f1.lsity- 1.1 well as truth-content, so that their verisimilitudes will not
be comparable.
118 Philosophy o/ logics

(e) Davidson's use o/ Tarski.'s theory

Truth and meaning. Any adequate theory of meaning,
Davidson thinks, must explain how the meanings of sentences depend
u pon the meanings of words ( otherwise, he argues, the language
would be unlearnable). A theory of meaning must be consistent
with - or, he sometimes says, explain - 'semantic productivity':
speakers' ability to produce, and understand, sentences they have
never heard before. What this amounts to, he claims, is that the
theory should yield all sentences of the form:
S means m
where 'S' is a structure-revealing description of a sentence of the
language for which the theory is being given, and 'm' a term denoting
the meaning of that sentence. But the appeal to meanings implicit
here, he suggests, con tributes nothing useful; and reformulating the
requirement thus:
S means that p
where 'p' is a sentence that has the meaning that the sentence
described by 'S' has, lea ves a problem with the 'means that ', which,
therefore, Davidson reformulates as 'is T iff' where ' T' is any
arbitrary predicate which, given the above conditions on 'S' and 'p',
satisfies:
Sis T iff p
But, of course, any predicate satisfying this condition will be, by
Tarski's standards, a materially adequate truth-predicate. Davidson
concludes that what is required by a theory of meaning is, precisely,
a definition of such a truth-predicate (Davidson 1967).

Meaning as truth-conditions
Though the route by which Davidson reaches this conclusion is
somewhat indirect, the terminus - that the meaning of a sentence
can be given by specifying its truth-conditions - is not unfamiliar.
What is novel in Davidson's version is the imposition of 'Tarskian'
constraints upon the account of truth-conditions. •
1
Dummett urges 1. theory of meaning in terma rather of aaaertibility-
conditions than truth-conditions (again 1. comparison with the
pragnuÍtiata, now with their criterial theory of meaning, su¡¡est8 itaelf).
For critic1.l discu1&iona lff Huck 1974 pp. 103fl'. and cf. Bnndom 1976.
 Theories of truth 119

The appeal of a truth-condition theory of meaning may perhaps be

appreciated by recalling Quine's classification of semantic notions
into two groups, the extensional, which he takes to constitute the
business of the 'theory of reference' and the intensional, which he
takes to constitute the business of the 'theory of meaning', thus:
semantic ideas
theory of reference ------- .....__ theory of meaning
( extensional) (intensional)
e.g. 'designa tes' e.g. 'is significant'
'satisfies' 'is synonymous with'
' is true' 'analytic'
'means'
Quine argued in 1953a that the theory of reference was in consider-
ably better shape than the theory of meaning. An appealing feature
of the truth-condition theory is that it promises an explanation of
meaning (from the more problematic right-hand side) in terms of
truth (from the less problematic left-hand side).

Theory of interpretation
La ter ( I 97 4) Davidson appends sorne further theory, of interpretation
of another's discourse, in another or even in the same language as
one's own; essentially, this consists in an account of how to tell
when 'p' is a sentence that has the meaning that 'S' describes.
Briefly, the idea is that to test, empirically, whether a sentence of the
form
'Es regnet' is true iff it is raining
is a T-sentence, that is, meets Tarski's specification that the sentence
on the right translates the one named on the left, one tests whether
speakers of the language concemed (here, German) hold true 'Es
regnet' iff it is raining. The point of the appeal to what native speakers
hold true is to get at the meaning of their utterances by, so to speak,
holding their beliefs constant. In consequence an assumption, the
principie of charity, to the effect that speakers of other languages
generally agree with us about what is the case, is required. The holist
character of Davidson's account, the insistence that the 'unit of
interpretation' is the entire language, may derive at least in part from
epístemic holism, the Duhemian idea, also stressed by Quine, that
beliefs are verified/falsified not alone but in a corporate body.
120 Philosaphy of logics

Though there are many important questions to be asked about this

theory of interpretation, I shall concentrate, in what follows, on
Davidson's account of meaning, since it is there that Tarski's theory
of truth plays the crucial role.
If the task of a theory of meaning is indeed, as Davidson thinks, to
define a Tarskian truth-predicate, what work over and above that
already accomplished by Tarski would be needed? Davidson is seeking
a theory of meaning/or natural languages, such as English; Tarski, of
course, is thoroughly sceptical about the applicability of his theory to
natural languages. So a first task, if Davidson's programme is to be
feasible, is to show that Tarski's methods can be extended. This is an
important question even independently of Davidson's special ambi-
tions for Tarski's methods, for the concept of truth is of philosophical
significan ce in man y contexts where 'true' must be allowed to apply
to sentences of natural languages - in epistemology, for instance.
Despite Tarski's official modesty on this score, it seems to me that the
usefulness of his work would be sadly restricted if the concept he
defines turns out to be quite different from the concept of truth in
natural languages.

Is Tarshi's theory applicable to natural languages?

According to Tarski:
The oery possibility o/ a consistmt use o/ the expression
'true sentence' which is in harmony with the laws o/ logic and
the spirit o/ eoeryday language seems to be very (Jtllstionable,
and consequently the same doubt attaches to the possibility
o/ c<mstructing a correet lkfinition o/ this expression.
(1931 p. 165)
Tarski's pessimism has two main sources: his formal correctness con-
dition rules out the possibility of an adequate definition of truth for
languages which are neither (i) semantically open nor (ii) formally
specifiable. Natural languages, Tarski argues, fail on both seores, so
there is no prospect of an adequate definition of truth for them.
(i) Tarski suggests that natural languages contain their own meta-
languages, so that truth cannot be defined without running into
paradox; though sometimes he hints, rather, that beca use natural
languages are not formally specifiable, the question of their semantical
closure cannot be answered. Davidson has no very satisfactory answer
to this problern, but urges that 'we are justified in carrying on without
 Theories of truth 121

having disinfected this source of conceptual anxiety' (1967 p. IO). He

seems to propose that work proceed on those semantically open frag-
ments of natural languages where the danger of paradox does not
arise. There is sorne difficulty in squaring Davidson's attitude to the
paradoxes (don't worry too much about them, concentrate on the rest
of the job) with his holism, the insisten ce that an adequate theory of
meaning must be a theory for a whole language; though he also hints
that he doubts whether natural languages really are universal.
(ii) There seems to be a whole family of difficulties here; the
problem of giving a precise account of just what strings count as
sentences of a natural language, aggravated by the fact that natural
languages are not static, but growing; and the prevalence in natural
languages of such phenomena as vagueness, ambiguity, indexicality.
Tarski is gloomy:
\Vhoever wishes, in spite of all difficulties, to pursue the
semantics of colloquial language with the help of exact
methods will be driven first to undertake the thankless task
of a reform of this language ... It may however be
doubted whether the language of everyday life, after being
'rationalised' in this way, would still preserve its natural-
ness and whether it would not rather take on the
characteristic features of the formalised languages.
(1931 p. 267)
The core of Davidson's reply to this is that, though sorne 'tidying
up' will be needed before Tarski's methods can be applied to a natural
language, this need not be such as to transform it out of all
recognition. He would hold, I think, that work in transfonnational
grammar (see e.g. Chomsky 1957) promises to overcome the first
problem; and he is optimistic that more fragments of natural
languages can be brought within the scope of Tarskian methods,
rather as Frege's work on '(x)' and '(3x)' has already suitably regí-
mented 'all', 'none' and 'some'.1
What Tarski regards as a 'thankless task' Davidson undertakes
gladly, observing that 'It's good to know we shan't run out of work'.
His main task, in fact, is to supply a suitable analysis of those locutions
1 It is not beyond dispute that Frege's account does properly regiment
natural language quantifiers; recall (ch. 4 § 1) that Montague and
Hintikka, like the early Russell, stress their affinities with singular
terrns, whereas according to Frege they belong to an entirely difl'erent
syntactic category.
122 Philosophy of logics

of natural languages which are initially recalcitrant to Tarskian treat-

ment. And it is on his success or failure at this task that one's assess-
ment of Davidson's response to Tarski's scepticism must be based.
It is worth observing that Davidson insists on using the 'absolute'
rather than a model-theoretic concept of truth; and that sorne of these
problems ( e.g. problems created by the introduction of new predicates
as a natural language grows) are harder on an absolute than they
would have been on a model-theoretic approach; cf. Field 1972.

Logical form
Davidson describes himself as seeking 'the Jogical form ' of natural
language locutions. For example, recall ( ch. 2 §4) that, according to
Davidson, adverbial constructions in natural language are best repre-
sented as involving quantification over events, with adverbs construed
as adjectives of event terrns. The logical form of 'J ohn buttered the
toast with a knife ', Davidson claims, is something like "There is an
event which is a buttering of the toast by J ohn and which is performed
with a knife '. Davidson's confidence that each natural language con-
struction has a unique logical form springs from the belief that a
formal representation to which Tarski's method of defining truth
applies represente essential structure in an ideally perspicuous fashion.
(The analogy with Russell's and Wittgenstein's project, in their
logical atomist periods, of devising an ideal language which would
represent the real form of natural languages, is striking.) Interestingly,
Cargile has asked (1970; and cf, Davidson's reply in the aame volume)
why the connection between a predicate and its adverbially modified
form need necessarily be assumed to be a matter of form rather
than content. It isn't, he suggests, as obvious as Davidson seems to
suppose what one should count as skeleton, and what as flesh; he
urges, in fact, a more flexible conception of logical forrn, closer to the
one presented in ch. 2.

Damdson's programme
So Davidson regards it as the task of a theory of meaning to analyse
the structure of sentences, not to supply an account of the meaning
of individual words. (This isn't quite right, because sorne particles -
'un' for instan ce - have a structural character.) For example,
Davidson does not require a theory of meaning to give the rneaning of
'good ', but he does require it to analyse the structure of e.g, 'Bardot
is a good actress' in such a way as to explain why it is not equivalent
 Theories of truth 123

to 'Bardot is good and Bardot is an actress' as 'Bardot is a French

actress' is to ' Bardot is French and Bardot is an actress' ( cf. 'small
elephant ', and the ambiguous 'poor violinist '). The appeal of Tarski's
method, which is to define satisfaction for complex open sentences in
terms of satisfaction of simple open sentences, is its promise of an
explanation of how the meanings of compound sentences depend on
the meaning of their parts; the challenge is to analyse sentences líke
'Bardot is a good actress' so that Tarski's method applies to them as
well as to the less recalcitrant 'Bardot is a French actress '. Davidson
admits that the task is considerable, that:
a staggering list of difficulties and conundrums remains.
(1967 p. 321)
He includes ('to name a few') counterfactuals, subjunctives, prob-
ability statements, causal statements, adverbs, attributive adjectives,
mass terms, verbs of belief, perception, intention, action. Obviously
my consideration of details of the programme will have to be selective.

Indesicals
Tarski's theory needs to be relativised to speakers and times,
Davidson suggests, because natural languages contain indexicals.
The revised (T) schema will call for the theory to entail sen ten ces like:
'I am tired' (s, t) is true iff sis tired at t
Truth, Davidson says, is a predicate rather of utterances than
sentences. (This suggestion is relevant to the claim, canvassed by
Strawson and, before him, by Schiller, that formal methods are
inherently inadequate to deal with the context-dependence of state-
ments in natural languages.)
But Davidson's concem with indexicals is also directed towards the
problema raised by the analysis of quotation and verbs of 'proposi-
tíonal attítude' ('says that', 'knows that' etc.); for he thinks that
these constructions all involve concealed demonstratives. An analysis
of these indexicals ('this', 'that') given by Weinstein 1974 has been
endorsed by Davidson. In this account, "That is a cat', say, is true just
in case the object indicated by the speaker at the time of utterance
satisfies ' .. .is a cat '.
124 Philosophy o/ logics

Oratio obliqua
While truth-functional compounds raise no problems, there will
obviously be a difficulty about applying Tarski's methods to com-
pound English sentences the truth-values of which do not depend in
any obvious way upon the truth-values of their parts. Oratio obliqua
sentences are of this problematic, intensional kind; for the truth-value
of 'Galileo said that the earth moves ', for instan ce, does not depend
in any direct way on the truth-value of 'the earth moves'; and there is
failure of substitutivity, for from "Tom said that the moon is round'
and 'The moon = the sole planet of the earth' one cannot safely infer
"Tom said that the sole planet of the earth is round '.
The first step in the right direction, Davidson urges, is to parse:
Galileo said that the earth moves.
along the lines of:
Galileo said that.
The earth moves.
The 'that' is to be construed not as a relative pronoun, but as a
demonstrative pronoun referring to an utterance - rather as I might
say 'I wrote that', pointing to a message on the notice-board.1
Of course, Galileo didn't utter the very utterance which the speaker
produces; indeed, Galileo didn't speak English; so sorne more
explanation is needed. Davidson amplifies his analysis thus:
The earth moves.
(3x) (Galileo's utterance x and my last utterance make
us samesayers)
Galileo and I are samesayers, we are told, just in case he uttered a
1
Two points of comparison are worth making. I have already mentioned
the affinities between Tarsk.i's definition of satisfaction and
Wittgenstein's Tractatus account of truth; the verbs of proposicional
attítude, which presenta problem to Wittgenstein's as to Davidson's
approach, are discussed at 5.542. Wittgenstein's analysis is, however,
notoriously obscure, Alun Jones has pointed out to me that Davidson's
list of 'difficulties and conundrums' for his enterprise, and Anscombe's
list (1959) of the problems for Wittgenstein's, are very similar. An
analysis of indirect discourse strikingly like Davidson's was suggested
by Kotarbiáski (¡955). Kotarbióski's airn was to support the thesis
that only material bodies exist (' pansomatism ') by analysing away
apparent references to such abstract objects as propositions; this, in
view of the conjecture that Tarski was motivated by sympathy with
materialism, may be si¡nificant.
 Theories of truth 125

sentence which meant in his mouth what sorne utterance of mine

meant in my mouth.
The application of Tarski's methods, as extended by Weinstein to
cope with indexicals, gives a result along the lines of:
'Galileo said that the earth moves' (s, t) means that}
is true iff
Galileo uttered at t' (t' earlier than t) a sentence which
meant in his mouth what the utterance demonstrated by
s at t" (t" just after t) meant in s's mouth, where the
demonstrated utterance is of 'The earth moves '.
It may be useful to pause briefly to contrast Davidson's with sorne
alternative accounts of the (so-called) 'propositional' attitudes. Frege,
for instance, would regard 'that p' in 's said (believes) that p' as
referring to a proposition (see ch. 5 §2). Carnap would analyse 's said
(believes) that p' as 's uttered (is disposed to assent to) sorne sentence
intensionally isomorphic to 'p' in English' (see 1947). Scheffier treats
'that p' rather as an adjective than a noun: 's said (believes) that p'
amounts to 's uttered (believes) a that-p utterance' where there is
a separate predicate corresponding to each sentence 'p' (see 1954);
Quine goes yet further in the same direction, treating the whole of
'said (believes)-that-p' as a predicate of s (see 196oa §44).
Davidson believes his account to have the advantages that: unlike
analyses which treat 'that r as referring to a proposition, it doesn't
require appeal to intensional entities; unlike Carnap's analysis, it does
not require explicit reference to a language; and unlike analyses which
treat 'says (believes)-that-p' as a single predicate, it allows that what
follows the 'that' is a sen ten ce with 'significant structure ', structure
a theory of meaning can exploit.
His account, Davidson argues, allows, as seems proper, that 's said
that p' entails 's said sornething ', for the analysis goes 's uttered
a sentence which ... '. At the same time, it explains, as is also
required, why 's said that p' does not entail 'p ', for what seemed to
be a sen ten ce ('p ') within a sentential operator (' s said that ') becomes
a single sentence (' s said that ') containing a demonstrative (' that ')
which refers to an utterance of another sentence ('p '). And just
as, although cats scratch, the sentence, 'that is a cat', which
refers to a cat, doesn't scratch, so, although 'p' entails 'p', the sen-
tence 's said that p ', which refers to an utterance of 'p ', doesn 't
entail 'p'.
126 Philosophy of logics

Of course, as the last example brings out, in the regular cases

considered by Weinstein, what 'that' refers to is a non-linguistic
ítem, a cat, for instan ce. When the account is extended to 'that 's in
indirect speech, the referents will be utterances of sentences. And
these sentences have significant structure (among the instances of
's said that p' would be e.g. 's said that q and r' and 's said that s' said
that q') in virtue of which their meaning would be given.
Sorne comments are in order, here, about how Davidson's differs
from Carnap's analysis. On Davidson's account 's said that p' in vol ves
reference to an utterance of the speaker's related to sorne utterance
of s's by samesaying; in Carnap's, reference to a sentence related to
a sentence of English by intensional isomorphism. An utterance (here,
Davidson makes it clear that he means a speech act, the event of
uttering a sentence) is an utterance of sorne sentence in sorne specific
language with sorne specific context; and so the need to specify the
relevant language is avoided.1
This gives Davidson's account an unexpected character - for the
concept of utterance (speech act) belongs rather to pragmatics than
to semantics, Equally surprising, and methodologically also dis-
quieting, is that Davidson's account, like Carnap's, requires a semantic
primitive (respectively, samesaying and intensional isomorphism) in
the metalanguage. s and s' are samesayers, Davidson explains, just in
case sorne utterance of s's means the same as sorne utterance of s"s.
Now, Davidson insists that the truth-conditions be given in terms of
an absolute definition of truth, a definition, that is, which uses no
semantic primitives. And he avoids 'S means m' and the formula
'S means that p' because of their intensional character. Davidson
apparently regards the appeal to samesaying as admissible because
local; the general account of meaning appeals only to Tarskian truth-
conditions, though the specific account of 'says that' requires same-
saying as semantic primitive. It is questionable, though, whether the
appeal is local in the relevant sense; for surely 'says that' counts as
structure rather than vocabulary in the sense in which the dependen ce
of the meaning of 'good' on the meaning of 'actress' in 'Bardot is a
1
Davidson sometimes speaks as if it is the reference to an utterance
(rather than a sentence) that prevents '1 said that p' entailing 'p'.
But this is surely sufficiently explained by appeal to the fact that
(an utrerance of) 'p' is refnrtd lo by, and not comained in • , said that p '.
The sense of' utterance' in which, according to Davidson, truth is
a property of utterances, hu, presumably, to be the 'content', and not,
as in thia context, the ' act' aense,
 Theories of truth 127

good actress' is structural {Davidson objects to the Fregean account

of indirect discourse because it requires intensional objects). The
problem is what exactly the constraints should be on Davidson's
enterprise: what apparatus should he be perrnitted to use, and where?
lt is pertinent that the appeal of his enterprise derives in large part
frorn the austerity of rnethod it appears, at the outset, to prornise.
Since the enterprise was launched, Davidson and his followers
have tackled, with various degrees of success, rnany of the 'difficulties
and conundrurns' pointed out in 1967. By 1973 Davidson speaks of
'fairly irnpressive progress ', pointing to work on propositional atti-
tudes, adverbs, quotation (Davidson 1967, 1968a, b), proper narnes
(Burge 1973), 'ought' (Harrnan 1975), rnass terms and cornparatives
(Wallace 1970, 1972).
The success of Davidson's prograrnrne would vindicate, in large
rneasure, the applicability ofTarski's theory to natural languages; but
the assessrnent of his programme obviously depends on the detailed
study of the specific analyses offered. And as I have suggested with
reference to the analysis of oratio obliqua, this study in its turn raises
sorne rnethodological questions which are at any rate tricky enough
that one cannot say with any confidence that Davidson has shown that
Tarski's theory applies to English.

7 The redundancy theory

Ramsey
The redundancy theory (though suggested earlier by sorne
remarks of Frege in 1918) derives prirnarily from the work of F. P.
Ramsey in 1927. Ramsey offers his sketch of a theory in a very brief
passage (pp. 142-3) in the course of a discussion of the proper analysis
of belief and judgment; the context is significant of Ramsey's estima te
of the importance of the issue: 'there is' he thinks, 'really no separate
problern of truth, but merely a linguistic muddle '.
Briefly, his idea is that the predicates 'true' and 'false' are
redundant in the sense that they can be eliminated from ali contexts
without semantic loss;1 he allows that they have a pragmatic role, for
'emphasis or stylistic reasons'. Ramsey considers two kinds of case
where 'true' and 'false' typically occur. The cases he uses to intro-
duce the theory are of the more straightforward kind, where the
1 There ia an allusion here to Russell's doctrine of 'incomplete symbols',
aymbola, that ia, which are contextually eliminable. cr. ch. s § 3 for • diacus-
aion of this doctrine with refereeee to Russell'• theory of deecriptiona.
128 Philosophy of logics

proposition to which truth or falsity is ascribed is explicitly given:

'it is true that p ', Ramsey argues, means the same as 'p ', and ' It is false
thatp' means the same as 'notp'. Cases where the relevant proposition
is not actually supplied but only described present rather more initial
difficulty, for, as Ramsey realises, one cannot simply eliminate 'is
true' from, for instance, 'what he says is always true'; this difficulty
he proposes to overcome by using the apparatus of propositional
quantification, to give, in the case mentioned, something along the
lines of 'For allp, if he assertsp, thenp'.1
Whether the second-order quantifi.ers which Ramsey needs can be
suitably explicated is a key question, as it turns out, for the feasibility
of the redundancy theory; but I shall begin by pointing out sorne of
the advantages of the theory before turning to its problems.

Truth-bearers
In view of the embarrassments caused by the trappings -
facts and propositions - of the correspondence theory the austerity of
the redundancy theory is appealing. Ramsey understandably regards
it as a virtue of his theory that it avoids the questions raised by a
correspondence account about the nature and individuation of facts.
"It is a fact that ... ', he urges, has the same semantic redundancy, and
the same emphatic use, as 'It is true that ... '.
Again, since the effect of Ramsey-style theories is to deny that in
'It is true that p', ' ... is true ... ' is to be thought of as a predicate
ascribíng a bona fide property to whatever 'p' stands for, the question
of the truth-bearers is similarly bypassed; if truth isn't a property,
one needn't ask what it's a property of. 1 observe, however, that what
I argued (ch. 6 § 5) to be the real issue lying behind disputes about
truth-bearers - the question of the appropriate constraints on
instances of sentence letters, i.e. what one can put for 'p' - does still
arise. (Ramsey's preference for the locution 'It is true thatp', rather
than ''p' is true' is of sorne significan ce in this regard.) I should
count it an advantage of my diagnosis of the issue about truth-bearers
that it is applicable even to redundancy theories, and an advantage

1
Ta111ki writes (1944 pp, 68-<}) as if Ramsey's theory sirnply has no way
to handle this kind of case; Ramsey would presumably analyse the two
problematic cases Tarskí gives - 'The first sentence written by Plato
is true' and 'All consequences of true sentences are true' - u '(p) (if
the first thing Plato wrote was that p, then p)' and '(p) (q) (if p, and
if p then q, then q) '.
 Theories of truth 129

of the redundancy theory that there the issue arises in its funda-
mental form.
Of course, this will be a genuine economy only if it is certain that
one doesn't need propositions (or whatever) for other purposes
besides truth-bearing. Those who believe that we need propositions
as objects of belief, for instance, are liable to be less impressed by the
redundancy theory's ability to do without them as bearers of truth.
It is significant, therefore, that Prior, who accepts Ramsey's theory,
urges (1971 ch. 9) an account of belief according to which 's believes
that ... ' in 's believes that p' is a sentence-forming operator on
sentences like 'It is not the case that ... ', rather than 'believes' being
a relation symbol with arguments 's' and 'that p ', the latter denoting
a proposition. Again, one might suppose that propositions ( or what-
ever) may be required as bearers of other properties, and that the
redundancy theory is therefore in danger of sacrificing the analogy
between ' ... is true', and, say, ' ... is surprising' or ' ... is exag-
gerated' without, in the end, any compensation by way of genuine
ontological economy. And it is significant, in this regard, that Grover
et al., in a paper (1975) urging the claims of a redundancy-style
theory, argue that it is only a misleading appearance that ' ... is true'
and ' ... is surprising' are ascriptions to the very same thing.

The object language /metalanguage distinction

The redundancy theorist denies that' 1 t is true that p' is about
the sentence 'p': 'It is true that lions are timid ', like 'It is not the case
that lions are timid ', is in his view about lions, not about the sentence
' Lions are timid '. This means that he sees no need for insisten ce on
the distinction between object language and metalanguage which is so
vital to Tarskian semantics (Prior shows most awareness of this point;
e.g. 1971 ch. 7). This raises sorne questions about the redundancy
theory's capacity to handle problems where the object language/meta-
language distinction apparently plays an important role.
The idea that truth is a metalinguistic predicate seems, for example,
to contribute to the usual explanations of the semantics of the sen-
tence connectives, as: " - p' is true iff 'p' is false', 'p v q' is true íff
'p' is true or 'q' is true'. How adequate an alternative theory can the
redundancy theory offer? Since that theory equates both 'It is false
that p • and 'It is true that -p' with '-p •, ali that remains of the
'explanation' of negation seems to be '-p iff -p '. The redundancy
theorist might urge that there is indeed less than meets the eye to the
130 Philosophy of logics

usual explanations of negation, for there is, according to him, less

than meets the eye to the usual explanations of truth. (Cf. Dumrnett
1958, and Grover et al.'s acknowledgment that 'It is not the case
that ... ' may not be eliminable.)
Another, related, difficulty is that the redundancy theorist seems
to be unable to allow an apparently genuine distinction between the
law of excluded middle ('p v -p ') and the metalinguistic principie of
bivalence (' for allp, 'p' is either true or else false'). For if "p' is true'
means the same as 'p', and "p' is false' means the same as '-p',
then "p' is either true or else false' means 'p v - p '. Once again, the
redundancy theorist might accept the consequence, that this is a
'distinction' without a difference ; but since it is a distinction with,
apparently, sorne explanatory power, this leaves him with sorne
explaining to do. (For instance, would he insist that van Fraassen's
'supervaluational' languages, where 'p v -p' is a theorem but the
semantics allow truth-value gaps, must be confused? Cf. Haack 1974
pp. 66 ff. and ch. 11 §4 below.)
I pointed out above (pp. 101-2) that the (T) schema seems to
require bivalence, and this raises the question whether a redundancy
theory isn't also committed to the thesis that 'p' must be either true
or else false. But this consequence is avoidable, for the redundancy
theorist may deny that, if it is neither true nor false that p, it is false
that it is true that p; after all, since his theory is that 'it is true that
p' means the same as 'p ', he could reasonably insist that, if it is neither
true nor false that p, it is also neither true nor false that it is true that p.
So he isn't obliged to deny the posaibility of truth-value gaps and
hence, the prevíous argument doesn't entail that he is obliged to
insist on the law of excluded middle.
In Tarski's work, of course, the most important role of the object
language/metalanguage distinction was to secure formal adequacy,
specifically, to avoid the semantic paradoxes. So its capacity to deal
with the paradoxes will be a pretty crucial question for one's assess-
ment of the feasibility of the redundancy theory. This question must
wait till ch. 8; but sorne of the considerations about propositional
quantifiers, to which I now turn, will be relevant to it.

The quantifers: '(p) (if he asserts that p, p)'

Ramsey proposes to elimina te "true.', where what is said to be
true is not explicitly supplied but only obliquely referred to, by means
of second-order quantification: 'What he saya is always true', for
 Theories o/ truth 131

instance, is to be explained as meaning 'For ali p, if he asserts p,

then p '. He admits that there is sorne awkwardness in this analysis,
for, he thinks, English idiom seems to call for a final 'is true' (as:
'(p) (if he asserts p, then p is true)' to make the final 'p' into a
bona fide sentence; but this apparent obstacle to elimination is over-
come, he argues, if one remembers that 'p' is itself a sentence, and
airead y contains a verb. Supposing that all propositions had the logical
form 'a R b ', he suggests, one could observe the grammatical pro-
prieties by writing 'For ali a, R, b, if he asserts a R b, then a R b '. But
of course, as Ramsey is well aware, ali propositions are not of the
form 'a R b ', and neither is there much prospect of giving a finite
disjunction of ali possible forms of proposition, so this scarcely solves
the problem.
Ramsey's discomfort is understandable, for the problem is real. If,
in his formula:
(p) (if he asserts p, then P)
the quantifier is interpreted in the standard, objectual style, one has:
For ali objects (propositions?) p, if he asserts p, then p
Here the bound 'p 's are syntactically like singular terms, and the
final 'p' has, therefore, to be understood ellíptically, as implicitly
containing a predicate, to turn it into something of the category of
a sentence, capa ble of standing to the right of 'then ', along the lines of:
For ali propositions p, if he asserts p, then p is true
But if the analysis turns out to contain the predicate 'is true', truth
hasn't, after ali, been eliminated, and it isn't, after ali, redundant.
(This is the difficulty Ramsey sees; it is stated rather clearly, with
reference to Camap's version of the redundancy theory, in Heidel-
berger 1968.) If, on the other hand, the quantifier is interpreted
substitutionally, one has:
Ali substitution instan ces of ' If he asserts ... then ... '
are true
and once again 'true' appears in the analysis, and so, hasn't really
been eliminated.
So this much is clear: if Ramsey's theory is to work, sorne other
explication of the second-order quantifiers will be needed, since on
either of the usual interpretations, 'true' seems not to be eliminated.
132 Philosophy of logias

Prior sees the difficulty as the result of a defi.ciency in English, which

lacks suitable colloquial locutions for reading second-order quanti-
fiers, and obliges one to resort to such misleadingly nominal-sounding
locutions as 'Everything he says ... '. He therefore suggests ( 197 I
p. 37) 'anywhether' and 'somewhether' as readings of '(p)' and
'(3p) ', and reads '(p) (p-+ p )', for instance, as 'lf anywhether, then
thether'.
Grover also thinks that the quantifiers can be supplied with suit-
able readings, and offers sorne further grammatical apparatus to this
purpose. The difficulty of giving an appropriate reading arises, as
Prior suggests, from the lack of words and phrases to stand in for
sen ten ces in the way that pronouns stand in for names and descriptions;
what is needed, as Grover puts it, is prosentences.
Pronouns and prosentences are two kinds of proform; cf. pro-
verbs like 'do', and proadjectives like 'such '. A proform must be
capable of being used anaphorically, for cross-reference, either like
pronouns of laziness (Geach 1967) as in 'Mary meant to come to the
party, but she was ill', or like 'quantificational' pronouns, as in 'lf
any car overheats, don't huy it '. Prosentences are like pronouns in
occupying positions that sentences could occupy, as pronouns occupy
positions that nouns could occupy, and fulfil a similar anaphoric role.
Grover's proposal is that one read '(P) (if he asserts that p, then p)' as:
For all propositions, if he asserts that thatt, then thatt
where 'thatt' is a prosentence. Notice that what is proposed is a novel
reading; it is, Grover argues, compatible with either an objectual or
a substitutional account at the level of formal interpretation.
This ingenious proposal raises a number of questions, to which
I can offer only tentative answers. First, remember that the problem
with which I began was whether it is possible to give a reading of
Ramsey's propositional quantifiers which is grammatical, and which
doesn't re-introduce the predicate 'true'. Does Grover's reading
meet these requirements? W ell, it would be somewhat odd to ask
whether her reading is grammatical, since it isn't, of course, English;
it expressly calls for an addition to English. It would be more appro-
priate to ask whether there are sufficiently strong grammatical
analogies to justify her innovation ; but this, in view of the 'suffi-
ciently strong', is none too precise a question. English, as Grover
allows, doesn't have any atomic prosentences - though it does, I think,
have compound expressions that play such a role: 'I t is', for instance,
 Theories o/ truth 133
which one might describe as a prosentence composed of a pronoun
anda proverb. And the second part of the question, whether Grover's
reading genuinely eliminates 'true', is equally tricky. In fact, there
are two points to be raised here. The first is that even if a suitable
reading is supplied, this leaves open a question about whether there
isn't still an implicit appeal to truth at the level of formal interpreta-
tion. (And what exactly must one eliminate 'true' from to show that
it is redundant?) The second question is whether one's understanding
of 'thatt' implicitly requires the notion of truth.

The 'prosententia/ theory o/ truth'

Sorne light may be shed on this problem by Grover's own
application of her account of propositional quantification to the
theory of truth. Grover et al. 1975 propose a modifi.ed version of the
redundancy theory according to which 'that is true' is explained as
being itself a prosentence. Truth-ascriptions, on their account, are
eliminable in favour of 'It is true' asan atomic prosentence, i.e. one in
which 'true' is not a separable predicate.1
What does this show about whether the 'prosentential' theory
really eliminates 't�e'? 'True', one is told, is eliminable; not from
English, to be sure, but from English + 'thatt '. But how are we to
understand 'thatt'? Well, there's nothing exactly like it in English,
but it works like 'That's true', except for being atomic rather than
compound ....
lt is open to doubt, 1 think, whether Ramsey's hope of eliminating
talk of truth altogether has been vindicated. Nevertheless, there is
something important to be learned from the discussion of the pro-
sentential theory: that the truth-predicate plays a crucial role in
enabling us to talkgenerally, to talk, that is, about propositions which

1
Ramsey thought that ali truth talk is eliminable; Grover et al. aclmit
that there is a resídue, In sorne cases the elimination of 'true' calls for
modíñcation of the contaíned sentence, as' It used to be true that
Rorne was the centre of the k.nown world '/' Rorne used to be the centre
of the k.nown world' or, 'It rníght be true that there is life on
Mars 'f' There rnight be life on Mars '. And where this phenornenon is
cornbined with quantification, as in 'Sorne sentences used to be true
but are true no Ionger ', they are obliged to introduce new connectives,
as • (3p) (Ir-used-to-be-the-case-thet p but ir-is-no-longer-the-case-
that p) •, which they aclmit to be, in effect, truth-locutions. Their
commenta about • It might be true that', on the other hand, •uaest an
interesting altemative to the idea that necessary truth, like truth, is
a property of sentences or propositions.
I 34 Philosophy of logics

we don't actually exhibit, but only refer to indirectly, a role it shares

with the apparatus of second-order (' propositional' or 'sentential ')
quantifiers. This similarity of function will turn out to be relevant to
the diagnosis of the semantic paradoxes.
 8
Paradoxes

The Liar and related paradoxes

The importance of the Liar paradox to the theory of truth
has already become apparent; for Tarski's formal adequacy condi-
tions on definitions of truth are motivated, in large part, by the need
to avoid it. It is time, now, to give the Liar and related paradoxes
sorne direct attention on their own account.
Why the 'Liar paradox'? Well, the Liar sentence, together with
apparently obvious principies about truth, leads, by apparently valid
reasoning, to contradiction; that is why it is called a paradox (from
the Greek, 'para' and "doxa ', 'beyond belief').1
The Liar comes in severa] variants; the classic version concerns
the sentence:
(S) This sentence is false
Suppose Sis true; then what it says is the case; so it is false. Suppose,
on the other hand, that S is false; then what it says is not the case,
so it is true. So S is true iff S is false. Variants include indirectly
self-referential sentences, such as:
The next sentence is false. The previous sentence is true.
and the 'postcard paradox ', when one supposes that on one side of a
postcard is written:
The sentence on the other side of this postcard is false
and on the other:
The sentence on the other side of this postcard is true.
1 The • plll'lldoxee' of material and stric:t implication - diacuued at len¡th
in ch. 11 - are, at wont, counter-intuitive, and not, like the Liar,
contradictory; hence the acare quotes.
136 Phi/osophy o/ logics

Another variant, the 'Epimenides' paradox, concerns a Cretan called

Epimenides, who is supposed to have said that all Cretans are always
liars. If a liar is someone who always says what is false, then if what
Epimenides said is true, it is false. The Epimenides is, however,
somewhat /ess paradoxical than the Liar, since it can be consistently
supposed to be false, though not to be true ( cf. Anderson 1970).
There are also 'truth-teller' (' This sentence is true') and imperative
(' Disobey this order') variants.
Other paradoxes involve 'true (false) of ... ' rather than 'true
(false)'. 'Heterological' means 'not true of itself'; so e.g. 'German ',
'long', 'italicised' are heterological, while 'English ', 'short ',
'printed' are autological, true of themselves. Now, is 'heterological'
heterological? Well, if heterological is heterological, it is not true of
itself; so, it is not heterological. If, however, it is not heterological, it
is true of itself; so, it is heterological. So 'heterological' is heterologi-
cal iff 'heterological' is not heterological ( Grelling' s paradox ).
Others again involve 'definable' or 'specifiable'. The number ten
is specifiable by a name of one syllable, the number seven by a name
of two syllables, the number seventeen by a name of three syllables.
Consider, then, the least number not specifiable in fewer than twenty
syllables. That number is specifiable in nineteen syllables, by 'the
· least number not specifiable in fewer than twenty syllables' (Berry's
paradox). Let E be the class of decimals definable in a finite number of
words, and let its members be ordered as the first, second, third ...
etc. Now let N be the number such that if the nth figure in the nth
decimal in E is m, then the nth figure in N is m + 11 or o if m = 9.
Then N differs from every member of E, and yet has been defined in
a finite number of words (Richard's paradox).
Other paradoxes involve the concept of set. Sorne sets are members
of themselves, while others are not (e.g. the set of abstract objects,
being itself an abstract object, is a member of itself; the set of cows,
not being itself a cow, is not). Now consider the set of sets which are
not members of themselves. Is ita member of itself or not? If it is a
member of itself, then it has the property which ali its members have,
that is, it is not a member of itself; if, on the other hand, it is not a
member of itself, then it has the property which qualifies a set for
membership in itself, so it is a member of itself. So the set of all sets
which are not member of themselves is a member of itself iff it is not
a member of itself (Russell's paradox). Other set-theoretical para-
doxes include Cantor's paradox: no set can be Jarger than the set of
 Paradoxes 137
ali sets, but, for any set, there is another, the set of ali its subsets,
which is larger than it is; and Burali-Forti's: the series of all ordinal
numbers has an ordinal number, O, say, but the series of all ordinals
up to and including any given ordinal exceeds that ordinal by one, so
the series of all ordinals up to and including Q has the ordinal
nurnber n+ l.
This by no means exhausts the ·range of paradoxes to be found in
the literature (cf. Russell 1908a, Mackie 1973 appendix, for more
examples). I hope, however, that my list is sufficiently representative
to illustrate the kind of problems with which a solution to the para-
doxes must deal; the point of considering a number of variants is to
enable one to check whether proposed solutions are sufficiently
broad in scope.

"Set-theoretical' versus 'semantic' paradoxes?

Though sorne of these paradoxes had been known long
before, they began to be of serious philosophical concern after
Russell's discovery of his paradox. Frege had reduced arithmetic to
sentence calculus, predicate calculus, and set theory. Russell, how-
ever, showed that his paradox was actually a theorem of Frege's
system, which was, therefore, inconsistent. (Since Frege had hoped to
supply foundations for arithmetic by reducing it to self-evident
principles, the fact that his 'self-evident' logic axioms turned out to
be contradictory was, naturally, a pretty severe epistemological shock;
cf. ch. I §2.) The paradoxes cannot be dismissed as mere tricks or
puzzles, for they follow from intuitively obvious set-theoretical
principies and thus threaten the very foundations of set theory. In
view of the fact that anything whatever is derivable from a contra-
diction, the consequences of paradoxes for a theory in which they are
derivable are quite intolerable (but cf. ch. I I §6 for further thoughts
about 'p & -p 1- q'). Russell's paradox operates as a key constraint
on attempts to devise consistent set theories; the Liar paradox,
similarly, operates as a key constraint on attempts to devise consistent
semantic theories.
But this raises an important, though difficult, question. As the
comment about the anaJogy between the role of Russell's paradox
in set theory and the role of the Liar paradox in semantic theory
suggests, it is possible to classify the paradoxes in two distinct groups,
those which essentially involve set-theoretical concepts, such as 'e'
and 'ordinal number', and those which essentially involve semantic
138 Phi/osophy of /ogics

concepts, such as 'false', 'false of ... ', and 'definable '. In fact, it is
commonplace to distinguish the set-theoretical and the semantic
paradoxes (the distinction goes back to Peano; its currency derives
from Ramsey's charnpionship in 1925):

set-theoretical semantic
paradoxes paradoxes
(Rarnsey: 'logical') (Ramsey: 'epistemo/ogical')
Russell's paradox Liar paradox and variants
Cantor's paradox Grelling's paradox
Burali-Forti's paradox Berry's, Richard's paradoxes
(Essentially involve (Essentially involve
'set', 'E', 'ordinal nurnber') 'false', 'false of', 'defina ble')

The second group is the one which is of irnmediate concern for

semantic theory.
Russell hirnself, however, didn't think of the paradoxes as falling
into two distinct groups, because he thought that they ali arose as the
resulto/ onefal/acy, from violations of the 'vicious circle principie'. If
one supposes that sorne paradoxes arise because of sorne peculiarity
of set-theoretical concepts, and others because of sorne peculiarity of
semantic concepts, the classification into two groups will be accept-
able; but if, like Russell, one thinks that the trouble lies in something
deeper, cornmon to ali the paradoxes, one will find it misleading. It is
hard to den y, I think, that ali the paradoxes sketched do have a prima
[acie affinity with each other and that a solution to them ali would
surely be more satisfying than a solution to only sorne; and in view of
this, the safest course seems to be not to beg, by concentrating exclu-
sively on the 'semantic' paradoxes, questions which could be left
open.

2 'Solutions' to the parados:es

Requirements on a solution
Before attempting to assess the solutions which have been
offered, it is wise, I think, to try to get a bit clearer just what would
constitute a 'solution'. Well, what exactly is the problem?- that
contradictory conclusions follow by apparently unexceptionable
reasoning from apparently unexceptionable premises. This suggests
two requirernents on a solution; that it should give a eonsistent
formal theory ( of semanties or set theory as the case may be) - in
 Paradoxes 139

other words, indicate which apparently unexceptionable premises or

principie of inference must be disallowed (the formal solution); and
that it should, in addition, supply sorne explanation of why that
premise or principie is, despite appearances, exceptionable (the
philosophical solution). It is hard to malee precise just what is re-
quired of such an explanation, but roughly what is intended is that it
should be shown that the rejected premise or principie is of a kind
to which there are independent objections - objections independent
of its leading to paradox, that is. It is important, though difficult, to
avoid supposed 'solutions' which simply label the offending sentences
in a way that seems, but isn't really, explanatory. Further require-
ments concern the scope of a solution; it should not be so broad as to
cripple reasoning we want to keep (the 'don't cut off your nose to
spite your face' principie); but it should be broad enough to block all
relevant paradoxical arguments ( the 'don 't jump out of the frying pan
into the fire' principie); the 'relevant ', of course, glosses over sorne
problems. At the formal Jevel, the latter principie urges simply that
the solution be such as to restore consistency. Frege's response to the
inconsistency found by Russell in his set-theory was a formal restric-
tion which avoids Russell's paradox but still allows closely related
paradoxes, and thus breaches this requirement (see Frege 1903,
Quine 1955, Geach 1956). At the philosophical level, the 'frying pan
and fire' principie urges that the explanation offered go as deep as
possible; this, of course, is what underlies my hunch that a solution
to both 'semantic' and 'set-theoretic' paradoxes, if it were possible,
would be preferable to a solution local to one group.
The force of these requirements may perhaps be appreciated by
looking briefly at sorne proposed solutions which fail to meet them.
It is sometimes suggested that the paradoxes be resolved by
banning self-reference; but this suggestion is at once too broad and
too narrow. It faJls foul of the 'don't cut off your nose to spite your
face' principie: for not only are many perfectly harmless sentences
('This sentence is in English', "This sentence is in red ink') self-
referential (cf. Popper 1954, Smullyan 1957), but also sorne mathe-
matical argument, including Godel's proof of the incompleteness of
arithmetic, malees essentiaJ use of self-referential sentences (cf. Nagel
and Newman 1959 and Anderson 1970); so that the consequences
of a han on self-reference would be very serious. And yet, since not
ali the variants of the Liar are straightforwardly self-referential
(neitber sentence in 'The next sentence is false. The prevíous
140 Philosophy of logics

sentence is true' refers to itself) this proposal is, at the same time, too
narrow.
The argument to a contradiction from the Liar sentence uses the
assumption that "This sen ten ce is false' is either true or false; and so,
unsurprisingly, it has often been suggested that the way to block the
argument is to deny this assumption. Bochvar proposed (1939) to
deal with the Liar by adopting a 3-valued logic in which the third
value, 'paradoxical ', is to be taken by the recalcitrant sentences.
(See also Skyrms 197oa, 197ob, and ch. 11 §3.) This proposal, too, is
in danger of being both too broad and too narrow: too broad, because
it requires a change in elementary (sentence calculus] logical prin-
cipies; and yet still too narrow, for it leaves problems with the 'Streng-
thened Liar' paradox - the sentence:
This sentence is either false or paradoxical
which is false or paradoxical if true, true if false, and true if paradoxical.
Another approach also denies that the Liar sentence is true or
false, without, however, suggesting that it has a third truth-value, by
arguing that it is not an item of the appropriate kind to have a truth-
value. Only statements, it is argued, are true or false, and an utterance
of the Liar sentence wouldn't constitute a statement. (See Bar-Hillel
1957, Prior 1958, Garver 1970; and cf. - mutatis mutandis with
'proposition' for 'statement' - Kneale 1971.) This kind of approach
suffers, I think, from inadequate explanatoriness - it doesn't supply a
suitable rationale for denying the offending sentences a truth-value.
Even granted for the sake of argument that only statements or
propositions can be either true or false (but granted only for the sake
of argument - cf. ch. 6) one would need an argument why in the case
of the Liar one does not have an ítem of the appropriate kind. After
all, the Liar sentence suffers from no obvious defi.ciency of grammar
or vocabulary. The mínimum requirements would be, first, a clear
account of the conditions under which an utterance of a sentence
constitutes a statement; second, an argument why no utterance of the
Liar could fulfil these conditions; third, an argument why only
statements can be true or false. Otherwise, one is entitled to complain
that the solution is insufficiently explanatory.
 Paradoxes 141

Russell's solution: the theory o/ types, the vicious circle

principie
Russell o:ffers (1908a) both a formal solution, the theory of
types, and a philosophical solution, the vicious circle principie.
Nowadays, it is customary to distinguish in Russell's formal
solution the simple and the ramified theory of types. The simple
theory of types divides the universe of discourse into a hierarchy:
individuals (type o), sets of individuals (type 1), sets of sets of indi-
viduals (type 2), ... etc., and correspondingly subscripts variables
with a type index, so that x0 ranges over type o, x1 over type 1 ... etc.
Then the formation rules are restricted in such a way that a formula
of the form 'x E y' is well-formed only if the type index of y is one
higher than that of x. So, in particular, 'xn E xn' is ill-formed, and
the property of not being a member of itself, essential to Russell's
paradox, cannot be expressed. The ramified theory o/ types imposes a
hierarchy of orders of 'propositions' (closed sentences) and 'propo-
sitional functions' ( open sentences), and the restriction that no
proposition (propositional function) can be 'about ', i.e. contain a
quantifier ranging over, propositions (propositional functions) of the
same or higher order as itself. 'True' and 'false' are also to be sub-
scripted, depending on the order of the proposition to which they are
applied; a proposition of order n will be true (false) n+ 1. The Liar
sentence, which says of itself that it is true, thus becomes inex-
pressible, justas the property of not being a member of itself did in the
simple theory. (I have simplified considerably; see Copi 1971 for a
more detailed account.)
Russell himself, however, did not see the paradoxes as falling into
two distinct groups; he believed that ali the paradoxes arose from one
and the same fallacy, from violations of what he, following Poincaré,
called 'the vicious circle principie' (V.C.P.):
'Whatever involves ali of a collection must not be one of
the collection'; or, conversely, 'If, provided a certain
collection had a total, it would have members only
definable in terms of that total, then the said collection
has no total'. [Footnote: I mean that statements about all
its members are nonsense.] (1908a p. 63)
He states the V.C.P. in several, not obviously equivalent, ways: e.g. a
collectíon mustn't 'involve', or, 'be definable only in terms of'
itself. The V.C.P. motivates the type/order restrictions imposed upon
14-2 Philosophy of logics

the formal theory, by showing that what the formulae ruled ill-formed
say is demonstrably meaningless. It is important that the very same
philosophical rationale is given for both the simple and the ramified
theories. Indeed, since Russell held that sets are really logical con-
structions out of propositional functions, he saw the restrictions of the
simple theory as a special case of those of the ramified theory ( cf.
Chihara 1972, 1973).
At both the formal and the philosophical levels, Russell's account
runs into difficulty. Formally, there is sorne danger that Russell has
cut off his nose to spite his face; the restrictions avoid the paradoxes,
but also block certain desired inferences. Remember that Russell was
trying to complete the programme, begun by Frege, of reducing
arithmetic to 'logic ', i.e. to sen ten ce calculus, first-order predicate
calculus, and set theory. However, the type restrictions block the
proof of the infiníty of the natural numbers, and the order restric-
tions block the proof of certain bound theorems. In Principia Mathe-
matica these proofs are saved by the introduction of new axioms,
respectively, the axiom of infinity and the axiom of reducibility; this
ensures the derivability of the Peano postulates for arithmetic; but
the ad hoc character of these axioms lessens the plausibility of the
claim that arithmetic has been reduced to a purely logical basis. Still,
it could be thought that these difficulties, though they cast doubt on
the feasibility of Russell's logicism, don't necessarily show his solu-
tion to the paradoxes to be misguided.
But one's suspicions are confirmed by difficulties at the philo-
sophical level. In the first place, the V.C.P. certainly isn't stated with
all the precision that might be desired; and it is correspondingly
difficult to see what, exactly, is wrong with violations of it, Ramsey
com.mented that he could see nothing objectionable about specifying
a man as the one with, say, the highest batting average of his team - a
specifi.cation apparently in violation of the V.C.P. Not ali the circles
ruled out by the V.C.P., he urged, are truly vicious (notice the analogy
to the difficulties in the proposal to han all self-referential sentences).
However, despite these diffi.culties, Russell's diagnosis and solution
have continued to be influential; later, in §3, I shall argue that
Russell's approach is, indeed, in certain respects, on the right lines.
But my immediate concern is with other solutions which resemble
Russell's in interesting ways. His diagnosis is echoed in Ryle's
approach. Ryle 1952 argues that 'The current statement is false'
must be unpacked as 'The current statement (namely, that the
 Paradoxes

current statement. .. [namely, that the current statement. .. {namely

, .. etc.}]) is false', and no completely specified statement is ever
reached. Like Russell, Ryle thinks that the "self-dependence ' of the
Liar sentence somehow robs it of sense. Mackie 1973 agrees with
Russell and Ryle that the problem lies in the Liar's 'vicious self-
dependence ', but prefers to say, for the good reason that the Liar
sentence is apparently correctly constructed from bona fide compo-
nents, that the upshot is not meaninglessness but 'lack. of content'.
However, since he is careful to distinguish 'lack. of content' from lack
of meaning and from lack of truth-value, one is left somewhat at a
Ioss to understand just what lack of content is lack of. And Tarski's
approach to the semantic paradoxes, to which I turn next, has sorne
significant similarities (observed by Russell 1956; and cf. Church
1976) to the Russellian hierarchy of orders of propositions.

Tarski's solution: the hierarchy of /anguages

Tarski diagnoses the semantic paradoxes (to which his
attention is restricted) as resulting from the two assumptions:

(i) that the language is semantically closed, i.e. contains (a)

the means to refer to its own expression, and (b) the
predicates 'true' and 'false'
(ii) that the usual logical laws hold

and, being reluctant to deny (ii) (but cf. the comments, above, on
Bochvar's proposal) denies (i), proposing as a formal adequacy
condition that truth be defined for semantically open languages. So
Tarski proposes a hierarchy o/ languages:
the object language, O,
the metalanguage, M,
which contains (a) means of referring to expressions of O
and (b) the predicates 'true-in-O' and 'false-in-O',
the meta-metalanguage, M',
which contains (a) means of referring to expressions of M
and (b) the predicates "true-in-M", and 'false-in-M.',
etc.

Since, in this hierarchy of languages, truth for a given level is

always expressed by a predicate of the next level, the Liar sentence
can appear only in the harmless form 'This sen ten ce is false-in-O',
144 Philosophy of logics

which must itself be a sentence of M, and hence cannot be true-in-O,

and is simply false instead of paradoxical.
Though the appeal of Tarskí's theory of truth has won this
proposal a good deal of support, there have also been criticisms of its
'artificiality '. The language hierarchy and the relativisation of 'true'
and 'false' avoid the semantic paradoxes, but they seem to lack
intuitive justification independent of their usefulness in this regard.
In other words, Tarski's approach seems to give a formal, but not a
philosophical, solution. The reason Tarski gives for requiring semantic
openness, is, simply, that semantic closure leads to paradox. There is
an independent rationale for the relativisation of 'true' and 'false' to
a language - that Tarski is defining 'truth' for sentences (wffs), and
one and the same sentence (wff) can have a different meaning, and
hence a different truth-value, in different languages; but this rationale
does not supply any independent justification for insisting that
'true-in-L' always be a predicate, not of L, but of the metalanguage
ofL.
Intuitively, one does not think of 'true' as systematically ambiguous
in the way Tarski suggests it must be. Perhaps this counter-intuitive-
ness would not, by itself, be an overwhelming consideration. But
Kripke (1975) points out that ordinary ascriptions of truth and falsity
cannot even be assigned implicit levels. Suppose, for instance, that
Jones says:
Ali of Nixon's utterances about Watergate are false.
This would have to be assigned to the next level above the highest
level of any of Nixon's utterances about Watergate; but not only will
we ordinarily have no way of determining the levels of Nixon's
utterances about Watergate, but also in unfavourable circumstances
it may actually be impossible to assign levels consistently - suppose
that among Nixon's utterances about Watergate is:
Ali of Jones' utterances about Watergate are false
then Jones' utterance has to be at a level one higher than ali of
Nixon's, and Nixon's at a level one higher than all of Jones'.
Tarski's approach, Kripke argues, fails to take adequate account
of the 'risky' character of truth-ascriptions. Quite ordinary assertions
about truth and falsity, he points out, are apt to tum out paradoxical
if the empírica} facts are unfavourable. Suppose e.g. that Nixon had
said that all of Jones' utterances about Watergate are true; then
 Paradoxes

Jones' assertion that all of Nixon's assertions about Watergate are

false would be false if true and true if false ( cf. the 'postcard paradox'
in § 1 ). The moral, he suggests, is that one can scarcely expect the
recalcitrant sentences to be distinguished by any syntactic or semantic
feature, but must seek a rationale which allows that paradox may
arise with respect to any truth-ascription if the facts turn out badly.1

Kripke' s solution: groundedness

Kripke seeks to supply an explanation of the source of
paradox which is more satisfactory in this respect, and then to build a
formal theory on this basis. (My hunch is that this is the right way
round to go about it.) His proposal depends upon the rejection of the
idea - taken for granted by Tarski - that the truth-predicate must be
totally defined, that is to say, that every suitably well-formed sentence
must be either true or false. It thus has affinities both with Bochvar's
proposal of a j-valued logic, and with the no-ítem proposals discussed
above. But Kripke stresses that his idea is not that paradoxical
sentences have sorne non-classical truth-value, but that they have
no truth-value.
The key idea in the explanation of how ordinary sentences are
assigned truth-values - and how extraordinary sentences fail to get a
value - is the concept of groundedness, first introduced by Herzberger
1970. Kripke explains the idea as follows:
Suppose one is trying to explain the word 'true' to someone who
doesn't understand it. It could be introduced by means of the prin-
cipie that one may assert that a sentence is true just when one is
entitled to assert that sentence, and one may assert that a sentence
is not true just when one is entitled to deny it. Now given that the
learner is entitled to assert that:

Snow is white

this explanation tells him that he is entitled to assert that:

' Snow is white ' is true.

N ow he can extend his use of 'true' to other sentences, e.g. as 'Snow

• Kripke also maltea the technical objection that Tarski's hierarchy has
not been extended to transfinite levels, and that, furthermore, there
are difficultiea about so extending it.
146 Philosophy of logics

is white' occurs in Tarski 1944-, the explanation allows hirn to assert

that:
Sorne sentence in 'The semantic conception of truth' is
true.
And he can also extend his use of 'true' to sentences which already
contain 'true', e.g. to assert that:
' ' Snow is white' is true' is true
or:
'Sorne sentence in 'The semantic conception of truth' is
true' is true.
The intuitive idea of groundedness is that a sentence is grounded just
in case it will eventually get a truth-value in this process. Not all
sentences will get a truth-value in this way; arnong the 'ungrounded'
sentences that won't are:
This sentence is true
and:
This sentence is false.
This idea has affinities with the notion - expressed in Russell's
V.C.P. and by Ryle and Mackie - that what's wrong with paradoxical
sentences is a kind of vicious self-dependence. However, ungrounded
sentences are allowed to be rneaningful, whereas Russell's idea is that
violation of the V.C.P. results in rneaninglessness.
Formally, this idea is represented (1 sirnplify considerably) in a
hierarchy of interpreted languages where at any level the truth-
predicate is the truth-predicate for the next lowest level, At the
lowest level, the predicate 'T' is cornpletely undefined. (This corre-
sponds to the initial stage in the intuitive account.] At the next
level, the predicate 'T' is assigned to wffs which don't thernselves
contain 'T'. lt is assurned that this assignrnent will be in accordance
with Kleene's rules giving the assignrnent of values to cornpound
wffs given the assignrnent - or lack of assignrnent - to their com-
ponents: '-p' is true (false) if 'p' is false (true), undefi.ned if 'p' is
undefined; 'p Y q' is true if at least one disjunct is true (whether the
other is true, false, or undefined), false if both disjuncts are false,
otherwise undefined; '(3x) Fx' is true (false) if 'Fx' is true for sorne
(false for every) assignrnent to x, otherwise undefined. (This corres-
ponds to the first stage, in which the leamer assigns 'true' to a
 Paradoxes 147

sentence if he is entitled to assert the sentence.) At each level the

wffs assigned 'T' and 'F' ata previous level retain those values, but
new wffs, for which ' T' was previously undefined, are assigned val-
ues - ' T' gets more defined as the process goes on. But the process
doesn't go on indefinitely with new sentences getting values at each
level; eventually- at a 'fixed point' - the process stops. Now the
intuitive idea of groundedness can be formally defi.ned: a wff is
grounded if it has a truth-value at the smallest fixed point, otherwise
ungrounded. The smallest or 'minimal' fixed point is the fi.rst point at
which the set of true (false) sentences is the same as the set of true
(false) sentences at the previous level. Ali paradoxical sentences are
ungrounded, but not ali ungrounded sentences are paradoxical;
a paradoxical sentence is one that cannot consistently be assigned a
truth-value at any fixed point. This supplies sorne explanation of
why 'This sentence is true' seems to share sorne of the oddity of
"This sentence is false', and yet, unlike the Liar sentence, is con-
sistent. A truth-value can be given to 'this sentence is true', but
only arbitrarily; a truth-value cannot consistently be given to 'This
sentence is false'. The picture also allows for the 'riskiness' of
truth-ascriptions: for the paradoxical character of a sentence may be
either intrinsic (as it would be with "This sentence is false') or
empirical (as it would be with 'The sentencc quoted on p. 147
ll. 22-3 is false').
I observed above that the relaxation of the requirement that 'true'
be fully defined, the admission of truth-value gaps, gave Kripke's
idea sorne analogy, also, to proposals, like Bochvar's, that the seman-
tic paradoxes be avoided by resort to a 3-valued logic. This raises the
question, how Kripke avoids the criticisms made earlier of Bochvar's
solution. Kripke himself stresses that he does not regard his use of
Kleene's '3-valued' valuation rules as a challenge to classical logic.
Whether the use of 3-valued matrices necessarily carries such a
challenge, is a difficult question, on which I shall have more to say in
ch. 11 § 3; for the present I shall allow Kripke's claim that his pro-
posals are compatible with logical conservatism. What, though, of the
Strengthened Liar?
Kripke doesn't tackle this issue directly, but it is possible to work
out what he would say about it. The notions of 'groundedness' and
'paradoxicality ', he says, unlike the concept of truth, don't be long in
his hierarchy of language levels. (Consider again the intuitive picture
of a leamer having the concept of truth explained to him. His
148 Pbilosophy o/ logics

instructions give hím no way to assígn a truth-value toan ungrounded

sentence líke 'This sentence is true'; but he cannot conclude that
'This sentence is true' is not true, for his instructions do tell him
that he may deny that a sentence is true only if he is entitled to deny
that sentence.) Now if 'paradoxical' belongs, not in the hierarchy of
language levels, but in the metalanguage of that hierarchy, then
Kripke can draw the teeth of the Strengthened Liar, 'Thís sentence
is eíther false or paradoxical' in much the way that Tarski draws the
teeth of the Liar. But this may occasion sorne dissatisfaction; for it is
a little disappointing to find that the novelty of Kripke's approach to
the Liar must be compromised by a neo- Tarskian dismissal of the
Strengthened Liar. (Is it indifferent whether one is hung for a sheep
ora lamb?)
It will be worthwhile to summarise the main points of comparison
and contrast between Kripke's approach, Russell's theory of types,
and Tarski's language hierarchy:

RUSSELL TARSKI KRIPKE

formal solution
híerarchy of hierarchy of hierarchy of
orders of languages (problems language levels (with
propositions with transfinite levels) limit leve Is)
systematíc dístínct truth and single, unívoca! truth-
ambíguity of falsity predícates at predícate, wíth
'true' and each leve! applícation extended
'false' up to mínima! fixed
point
'true' and 'true' and 'false' 'true' and 'false' only
'false' completely completely defined partially defined
defined
'Thís sentence is 'Thís sentence is 'This sen ten ce is
false' meaníng- false-in-O' false- false' neither true nor
less in-M false.
rationale
V.C.P. (language- groundedness
relatívísatíon of
'true')
 Paradoses

3 Paradox without 'false,; some remarks about the

redundancy theory of truth; and the V.C.P. again
I shan't, I'm afraid, be able to offer, in conclusion, a novel
solution to the paradoxes. The purpose of the present section is rather
more modest: to redeem the promise (pp. 130, 134) to comment
upon the consequences for the paradoxes of the redundancy theory of
truth, with its resistance to the idea of truth as a metalinguistic
predicate; a consequence of considerations which this investigation
brings to light, however, will be sorne support for a proposal which, as
I shall argue, has affinities with the V.C.P.
One of Tarski's reasons for refusing to countenance the treatment
of quotation as a function, and hence, for denying that truth could be
defined by generalising the (T) schema, to obtain '(p) ('p' is true iff
p)', was, if you recall (p. 104 above) that with quotation functions
paradox would ensue even without the use of the predicates 'true'
and 'false'. (And Tarski's semantic openness requirement, of course,
would be powerless to cope with paradox generated without semantic
predicates.) Tarski's argument goes as follows:
Let 'e' ah breviate 'the sentence numbered 1 '.
Now, consider the sentence:
I. (p) (e = 'p' � -p)
It can be established empirically that:
2. e = '(p) (e = 'p' � -p)'
and so, assuming that:
3· (p) (�) ('p' = 'q' � p = q)
'by means of elementary logical laws we easily derive a contradic-
tion' (1931 p. 162).1 Notice that here one has a paradox that arises, not
intrinsically in the nature of a single statement, but extrinsically,
due, as Kripke would put it, to the facts' tuming out badly. Tarski's
diagnosis is that quotation functions are the root of the trouble, and
must not be allowed. Sorne writers have, in response, suggested that,
rather than quotation functions being disallowed altogether, certain
restrictions should be imposed upon them; Binkley, for instance,
1 Tarski does not give the derivation, but it would preaumably go along
the following linea. From 1, if e = '(p) (e = 'p'-+- -p) ', then
-(p) (e= 'p' -+--p), so, given a, -(p) (e= 'p' -+--p); hence by
RAA, -1. If -1, then (:lp) (e =- 'p' & n: Suppoae for ínstance
that e= 'q' & q; then 'q' "''(p) (e= 'p' -+--P)' since both = e,
hence, by 3, q= (p) (e"' 'P'-+--p). But q; so (p) (e= 'P'-+--p),
i.e. 1. Hence, 1 & -1.
150 Philosophy o/ logics

suggests (1970) a 'no-mixing' rule, which prevents one and the same
quantifier from binding both variables inside, and variables outside,
quotation marks, and hence disallows I above. But neither Tarski's
diagnosis, nor this kind of response, can be quite right; for an ana-
logous paradox can be derived without the use of quotation marks:
Let '§' be an operator forming a term from a sentence;
it could be read e.g. 'the statement that ... '
Let 'e' abbreviate 'the statement made by sentence
numbered 1 '.
Now, consider the sentence:
I. (p) (e = § P-+ -p)
I t can be established empirically that:
2. e= §(p)(c = §p-+ -p)
anda contradiction follows as before.1 Now one might try, again, to
impose restrictions on terrn-forming operators like '§'; for instan ce,
following the example of Harman 1971, one might rule that if 'p'
belongs to L, '§p' must belong, not to L, but to the metalanguage of
L. But this kind of manoeuvre - quite apart from its disagreeably
ad hoc character - again seems not to go to the heart of the problem;
for an analogous paradox can be derived without the use of '§'; one
has only to let 'e' abbreviate the sentence numbered 1 (instead of
'the sentence numbered 1 '; 'e' now abbreviates a sentence, not a
term):
r, (p) ((e = P)-+ -p)
so that, in virtue of the abbreviation,
2. (e = (p) ((e = p)-+ -p))
and once again a contradiction is derivable.
1 Sorne comments are called for about the moral to be drawn about
quotation marks. Tanki holds (and Quine agrees) that the result of
enclosing an expression in quotation marks is an expression denoting
the enclosed expression, but o/ fllhu:h the endosed expressio« is not
genuinely a part. The idea that quotation fonns a aort of 'logical block',
that 'dog' isn't part of "dog", Ieads to very curious consequences, and
is quite counter-intuitive (cf. Amcombe 1957). So it is a relief to find
that the failure of Tanki's diagnosi, of the parados leaves one free to
treat quotation as a function; cf. Belnap and Grover 1973, Haack 1975
for more detailed discussion.
 Paradoxes

This shouldn't be too surprising. For the effect of a truth-

predicare can, as investigation of the redundancy theory (pp. 13er-4
above) showed, be achieved by using second-order (propositional)
quantifiers; and adding negation gives the effect of 'false.'. So the
fact that a Liar-type paradox is derivable without explicit use of
semantic predicates, provided propositional quantifiers and negation
are available, was to have been expected.
But how are paradoxes of this kind to be avoided? Suppose the
propositional quantifiers are interpreted - as I recommended in
ch. 4 §3 -substitutionally. On a substitutional interpretarion, a
quantified formula, A, of the form (v) <I> (v), is true just in case all its
substitution instances, <I>(s), are true. Since in the case under con-
sideration the quantifier binds sentence letters, the substituends for v
will be wffs, and may, therefore, themselves contain quantifiers. Now,
the usual conditions of definitional adequacy require that only substi-
tuends which contain no more quantifiers than A itself be allowed;
otherwise inelirninability would result (see Marcus 1972, and cf.
Grover 1973). This restriction is in no way ad hoc, since it is only a
special case of quite ordinary conditions on definitions; but at the
same time it is sufficient to block the paradoxical argument where the
= =
wff substituted for 'p' in '(p) ((e P)-+ -p)' is '(p) ((e P) -+-p)'.
It wouldn't be altogether fanciful, I think, to see affinities between
this idea and the theory of types, with its hierarchy of propositions
ordered according to what propositional quantifiers occur in them;
nor to see affinities between the motivation for the restriction on
substituends for sentential variables, and the V.C.P. Russell's argu-
ment why a proposition 'about all propositions' can't itself be a
member of that totality is that it 'creates' a new proposition not
previously belonging to that totality, which is unconvincing since it
assumes, what it is intended to prove, that the proposition about all
propositions isn't already a member of that totality; Ryle and Mack.ie,
however, urge, in favour of the V.C.P., that violations of it lead
to a 'vicious self-dependence ' which results in iMliminability. And,
finally, the fact that paradoxes can be generated without semantic
predicates rnight be thought to suggest that there might, after all, be
something in Russell's hunch that the paradoxes weren't to be handled
in distinct groups according as semantic or set-theoretic predicates
occurred essentially in them, but were to be handled together, as all
the result of one fallacy.
 9
Logic and logics
... since one never knows what will be the line of advance,
it is always most rash to condemn what is not quite in
the fashion of the moment.
Russell 1906, cited in Rescher 1974

'Classical' and 'non-classical' logics

There are a great many formal logical systems. In fact, ever
since the 'classical' logical apparatus was formulated, there have
been those who urged that it should be improved, modified or re-
placed. An instructive example can be taken from the history of the
material conditional; anticipated by the Stoics, 'material implication'
was formalised by Frege 1879 and Russell and Whitehead 1910 and
supplied with a suitable semantics by Post 1921 and Wittgenstein
1922. As early as 1880, however, MacColl had urged the claims of a
stricter conditional; 'strict implication' was formalised by Lewis
1918; and after that dissatisfaction with its claims to represent entail-
ment led to the introduction of 'relevant implication' (see ch. 10 §7).
My present object is to get sorne perspective on the great variety of
logical systems, to approach such questions as how they relate to each
other, whether one must choose between them, and, if so, how. My
strategy will be to consider the various ways in which the standard
logical apparatus has been modified, and the various pressures in
response to which such modifications have been made. An initial
note of caution should, however, be sounded: this strategy, of looking
at 'non-standard' by contrast with the 'standard' logical apparatus,
carries the danger of inducing an overly conservative attitude to
logical innovations. (Wolf 1977 puts the point well by reminding one
that 'possession is nine-tenths of the law '.) Awareness of this danger
may, I hope, of itself help to sorne degree in avoiding it. And it will
also be salutary to bear in mind that today's 'classical logic' was once
itself a 'logical innovation '. Kant, after ali, urged (1800) that logic
was a completed science, finished, in ali essenrials, in the work of
 Logic and logics 153
Aristotle; the next century, however, saw the development, with the
work of Boole, Peirce, Frege and Russell, of new, stronger and more
rigorous logical techniques. Recall, too, that Frege confidently
supposed that the principles of his logical system were self-evident,
until Russell showed that they were inconsistent!

2 Responses to pressure to change the standard formalism

The pressures to change the standard z-valued sentence
calculus and first-order predicate calculus have come from worries
about the apparent inadequacy of the standard apparatus to represent
various kinds of informal argument, and about the interpretation and
application of that apparatus. The reactions to such pressures have
been very varied; 1 shall first sketch, and then illustrate, sorne of the
commonest responses:
r. Informal arguments to which the standard apparatus does
not comfortably apply may be ruled outside the scope of
logic. For instance, pressure for a 'logic of
meaninglessness' may be resisted on the grounds that
meaningless sentences are simply outside the proper sphere
of logical formalisation. 1'11 call this the delimitation of the
scope of logic response.
2. Problematic informal arguments rnay be admitted as within
the scope of logic, and the standard apparatus retained;
but adjustments made in the way that the awkward
informal arguments are represented in the formalism. For
instance, Russell's theory of descriptions proposes that
sentences containing definite descriptions be represented
not in the obvious way, as 'Fa', but as existentially
quantified formulae. I'll call this the novel paraphrase
strategy. (Since Russell comments that the grammatical
form of such sentences conceals their logical form, in
1974 1 called it the misleading form strategy. But l'd
prefer not to seem to subscribe to his view that each
argument has a unique 'logical form. ')
3. A third response, like the second, admits the problematic
arguments as within the scope of logic, and retains the
standard apparatus without any change at the level of
syntax; however, the interpretation of that apparatus is
modified, in such a way that initially recalcitrant informal
154 Philosophy of logics

locutions are after ali adequately represented. For example,

concern about the apparent ontological commitments of the
predicate calculus may be met by the proposal that the
quantifiers be interpreted substitutionally, and empty
terms allowed as bona fide substituends, so that ontological
neutrality is secured. I'll call this the semantic innovation
response.
4. The standard apparatus may be extended to obtain a
formalism applicable to informal arguments which were
previously inaccessible to formal treatment. For example,
new operators may be added - tense operators or modal
operators for instance - and axioms/rules governing them;
or the standard operations may be extended to cover
novel items - imperative or interrogative sentences, for
instance. I'll call this the extended logic response.
5. Alternatively, the standard apparatus may be restricted,
in that, while its vocabulary remains the same, its axioms/rules
of inference are restricted in such a way that classical
theorems/inferences cease to be valid. For example,
concern to avoid anomalies in quantum mechanics has led
to proposals in which certain 'classical' principies, the
dístrihutive law for instance, no longer obtain. I'll call
this the restricted logic response; its upshot is a 'deviant
logic' (Haack 1974).

Sometimes novel formulations are proposed which at once extend

and restrict classical logic - they add new operators and new
principies governing them, but at the same time restrict the principies
governing old operators. 'Relevance logics' which introduce a new
conditional, while at the same time rejecting sorne classical laws,
such as modus ponens for the material conditional, would be an
example.
I distinguished 4 and 5 from 2 and 3 because the former involve
modifications at the level of syntax, while the latter leave the standard
syntax untouched. But, of course, an extension or a restriction of the
standard syntax would, in turn, require semantic modification, so that
an interpretation is supplied which verifies the extended or restricted
set of theorems/inferences. Indeed, restrictions of logic have quite
often been motivated by semantic considerations - as, for exarnple,
challenges to the assumption that every sentence within the scope of
 Logic and logics 155

logic must be either true or false led to the development of many-

valued logics, which characteristically lack classical theorems such as
'pV -p'.
As one would expect, extensions are most usually proposed in
response to a supposed inadequacy, restrictions in response to a
supposed incorrectness, in the standard formalism.

6. Innovations in the logical formalism are sometimes

accompanied by - and sometimes motivated by -
innovations at the level of metalogical concepts. For
example, the Intuitionists (who propose a restriction of the
standard apparatus) do so in part because they challenge
the concept of truth presupposed in classical logic; the
relevance logicians challenge the classical conception of
validity. I'll call this the challenge to classical metaconcepts.
7. Finally - andas, so to speak, the converse of the first
response - there are challenges to the standard conception
of the scope and aspirations of logic. These are often
enough associated with challenges to classical metaconcepts,
as in 6. For example, the Intuitionists not only restrict
classical sentence calculus so that 'p V -p', for instance,
is no longer a theorem, and not only offer an alternative
to the classical conception of truth; they also take a
radically different view from most classical logicians of the
role of logic, which they regard as secondary to
mathematics, rather than as underlying reasoning about
all subject-matters whatever. I'll call this the reoision of
the scope of logic response. (An Intuitionist, however,
would see the classical logician as revising the scope of
logic.)

Roughly speaking, I suppose, it would be right to think of these

responses as increasingly radical. But only roughly speaking. For
instance, although it is usually thought that a reinterpretation of the
standard apparatus is more conservative than an extension of it, there
is surely a sense in which the conservatism of 3 is nominal- l mean,
that the system only looks the same, but, since it's been reinterpreted,
the upshot is little different from the introduction of new symbolism.
lt is worth observing, for instance, that sorne urge, in the interests of
clarity, that we use a different notation for substitutional than for
156 Phi/osophy o/ /ogics

objectual quantifiers. And I have drawn attention to the way that

quite deep challenges to classical metaconcepts or to classical con-
ceptions oí the aim of formalisation may often underlie proposa1s
for extending or restricting the classical fonnalism; in view of this, it
is not altogether surprising that, as we shall see, such systems have
sometimes been held, by conservatives, not to be really logias at
ali.
Later (ch. 12) I shall be using the distinctions drawn here to try to
understand the epistemological issues raised by the existence of a
plurality of logics. For the present, though, my concem is mainly
to supply sorne kínd of framework for lookíng at that plurality.
Strategies 1-7 are not exclusive (nor, probably, exhaustive); it is of
sorne interest to notice that sorne problems, the problems raised by the
possibility of non-denoting singular terms, for example, have pro-
voked several of them: Strawson proposes to rule sentences con-
taining such terrns outside the scope of logic, Russell to supply a
novel translation that reveals their real logical form, Hintikka to
devise a restricted logic.
Since I can't possibly consider all the issues raised by the choice
between these strategies, I shall, instead, t�e a closer look at two
examples which illustrate sorne of the issues very well. I begin with
the problem of how to handle tense formally.

3 Fint cue-atudy: the logic of temporal diacoune

The pioneers of modern formal logic were primarily moti-
vated by the desire to represent mathematical arguments in a rigorous
way. Consequently, because of the irrelevance of considerations of
tense to the (in)validity of mathematical arguments, they were able
largely to ignore the fact that, in informal argument on non-mathe-
matical subjects, tense is sometimes crucial.
While this problem is often enough dismissed - along with related
problems about indexical expressions - with the comment that one
must take care, in representing informal arguments in symbolic
form, that tense remains constant throughout the argument (a sort of
casual version of the no-item response), sorne writers have made more
serious attempts to allow for tense. And two quite distinct strategies
have been proposed: Quine urges that temporal discourse be repre-
sented within the standard apparatus by interpreting the variables of
the predicate calculus as ranging, not over enduring spatiotemporal
individuals, but over 'epochs'; Prior urges that temporal discourse
 Logic and logics 157
be accommodated by an extension of the standard apparatus by the
addition of tense operators.
So, on the one hand, Quine proposes to deal with the problem by
means of a semantic innovation, while, on the other, Prior proposes
an extended logic. Another difference between the two strategies is
important: though both are attempts to accommodate considerations
of time, Prior's approach does this by taking tense seriously, while
Quine's approach tries to achieve the same end in an untensed
formalism. And a consequence of this is that Quine needs to make
adjustments in the way informal tensed discourse is represented
formally, as well as in the way the formalism is adjusted. That is to
say, his approach combines semantic innovation with the novel
paraphrase strategy.
Quine's approach (196oa §36; his ideas have been developed in
more detail in Lacey 1971, on which I'll also rely) is to represent
what is logically relevant in the tensed discourse of informal argu-
ments within the standard logical formalism. Although Quine admits
the relevance of tense to the validity of informal arguments, he regards
it as really inessential, a reflection of the bias of ordinary language
towards the temporal perspective of the speaker. So he proposes to
replace tensed by tenseless verbs with 'temporal qualifiers' such as
'now', 'then ', 'before t', 'at t' and 'after t'. The variables' t', 'u' ...
etc. are construed as ranging over what Quine calls 'epochs ', which
are stretches of space-tirne of any chosen duration, an hour, say, ora
day. An epoch, Quine explains, is a 'slice of the four-dimensional
material world, exhaustive spatially and perpendicular to the time
axis' (196oa p. 172). Reference to ordinary spatiotemporal individ-
uals, such as persons, is to be replaced by reference to 'time-slices'
of individuals, such as a person through a given time. So ordinary
tensed sentences get rewritten along these lines:
Mary is a widow Mary at now is a widow
George married Mary (3t) (t is before now and George
at t mames Mary at t)
George will marry Mary (3t) (t is after now and George
at t marries Mary at t)
The notational conventions are that tenseless verbs are to be written
in the present tense form, but italicised; the variable 't' is to range
o ver epochs; ' George at t' and ' Mary at t' refer to time-slices of the
spatiotemporal individuals George and Mary.
158 Philosophy of logics

'Now', of course, retains the indexical character of ordinary

language; but eventually Quine will elimina te this too, by means of
singular terms denoting epochs. So 'now' will be replaced by the
appropriate date, and the last trace of tensed discourse eliminated,
as:
Mary is a widow Mary at 12 March 1977 is a
widow
The result is that tensed sentences the truth-value of which varies
over time are superseded by what Ouine calls eternal sentences, the
truth-value of which remains constant. (Eternal sentences are, of
course, Quine's answer to the supposed need for propositions, which,
in view of their intensional character, he will not admit.)
It should be clear by now that Quine's proposal calls for consider-
able departures from ordinary language locutions as well as for con-
siderable innovations in the interpretation of the variables, singular
terms, and predicates of the predicate calculus. Nevertheless, Quine
would regard his proposal as, in an important sense, a conservative
one; because its point is to allow the representation of tensed dis-
course within an extensional formalism. This is why Quine - who
regards extensionality as a touchstone of intelligibility - attaches so
much importance to retaining the standard syntax.
However, Quine sees his proposal as having, besides, another
virtue: its consonance with modern physics. For while ordinary
tensed discourse singles out time, Quine's representations treat the
temporal dimension quite on a par with the three spatial dimensions.
The temporal parts of a thing are treated in the same way as its
spatial parts (a point which Quine exploits (p. 171) in arguing how
his approach illuminates the problern of personal identity: why should
one expect a person's temporal parts to be alike, since his spatial
parts, e.g. his head and his feet, are not?). Einstein's discoveries,
Qui ne comments, leave 'no reasonable altemative to treating time as
spacelike' (p. 172).
Príor's approach (see 1957, 1967, 1968) is interestingly different.
He accommodates temporal considerations not by adjusting the
tensed locutions of ordinary language to fit within a tenseless,
extensional symbolism, but by extending the standard symbolism to
accommodate tensed locutions. Prior begins from a regular sentence
calculus, in which, however, the sentence letters are to be understood
as standing for sentences uniformly in the present tense. (And hence
 Logic and logics 159

for items vulnerable to change of truth-value, by contrast with Quine's

tenseless eternal sentences.) He then enriches the symbolism with
tense operators 'F' and 'P', which are sentence-forming operators on
sentences, the former turning a present tense sentence into a future
tense sentence, the latter tuming a present tense sentence into a past
tense sentence. Prior reads 'F' as 'lt will be the case that ... ' and
'P' as '1 t used to be the case that ... ' Compound tenses are built up
by iterating these operators. For instance, if 'p' is 'George is marry-
ing Mary ', one has:
George married Marry Pp
George will marry Mary Fp
George will have married Mary FPp
The tense operators are not extensional; the truth-value of 'Fp' or
'Pp' does not depend solely u pon the truth-value of 'p '.
Axioms are supplied governing the new operators. In fact, Prior
offers alternative axiom sets, each suitable, he suggests, to rival
metaphysical views about time, such as whether time has a beginning
and/or end, whether it is linear or circular, whether determinism is
true, and so forth (see Prior 1968).
Prior observes that tense operators, instead of being taken as
primitive, might be defined in terms of quantification over instants of
time; ' It will be the case that p ', for instan ce, would be 'For sorne
time t later than now, p at t'.1 This would perhaps lessen the con-
trast with Quine's approach to sorne degree. But 'instants' are tem-
poral, not, like Quine's 'epochs ', spatiotemporal. And Prior says,
anyway (1968 p. 118) that he prefers to think of tense operators as
primitive, and of instants of time as 'mere logical constructions out of
tensed facts. '
Thus, Prior's approach achieves simplicity of paraphrase of infor-
mal arguments into formal symbolism, but at the same time increases
the complexity of the formalism, requiring, in particular, the loss of
extensionality. And it also contraste with Quine's approach at the
metaphysical level; for, though one is offered altemative axiom sets
between which to choose on the basis of one's view of time, the very

1 Prior's tense logica are closely modelled on C. l. Lewis' modal

systema (cf, ch. 10); and the definability of tense operaton via
quantifien over instanta corresponda to the account, in the U1Ual
aemantica for those modal logica, of necessíty (poaibility) n truth in
all (sorne) poasible worlds.
160 Philosophy of /ogics

TABLE 2

Quine's approach Prior's approach

semantic innovation,
novel paraphrase extended logic
strategies
eliminates tense introduces tense operators
eternal sentences, tensed sentences, change
no change of truth-value of truth-value
extensional formalism intensional formalism
substantial 'regimentation' conformity to ordinary
of informal arguments language
consonant with Newtonian in spirit
relativity theory
ontology of 4-dimensional ontology of objects
space-time world occupying space and
enduring through time

syntax of the system conforms to a 'Newtonian' conception of time

as quite unlike space, The major points of contrast between the two
approaches are summarised in table 2.

Prior's treatment is most consonant, I have suggested, with a view

of time as categorically different from space, Quine's with a view of
time as space-like. Not surprisingly, then, it has sometimes been
suggested that there are metaphysical reasons for preferring one or
other approach.! Quine, as I have already reported, thinks that
modern science 'leaves no reasonable alternative' to his approach.
Geach, on the other hand, argues (1965) that Quine's ontology of
epochs and four-dimensional spatiotemporal objects is defective
because it entails that there is no such thing as change. But this is
1
Cf. MacTaggart 1908 where a distinction is drawn between the
'A-series', in which events are ordered according as they are past,
present, or future and the 'B-series ', in which they are ordered as
earlíer than, simultaneous with, later than each other. Prior'a approach
emphasiaes the former, Quine's the latter. See alao Strawaon 1959 for
a defence of the former, and Whitehead 1919 for a defence of the
latter metaphyaical stance.
 Logic and logics 161

false; Quine's approach allows change, ali right, it is just that it

represents what one would ordinarily call change in an enduring
object over time as a difference between the earlier and later time-
slices of that object - as, for instance, my hair's turning grey would be
represented by a difference in the hair colour of my earlier and later
time-slices.
My present concern, though, is not with these metaphysical
questions, but with sorne methodological issues raised by the choice
of strategies.
In general, as in the present case, it seems reasonable to expect that
the price of sticking (like Quine) to an austere symbolism will be a
loss of naturalness of paraphrase of informal arguments. (To put it in
Russellian terms: the fewer logical forms available, the more grarn-
matical forms will have to be diagnosed as ' misleading '.) If one
attaches great significance to sorne degree of austerity - in Quine's
case, to extensionality - in one's formalism, one will have to accept a
divergence from natural language. If one attaches great significance
to conformity with natural language forms - as Geach does - one
will need a richer formalism. F or myself, I concede the desirability
both of austerity of symbolism (after ali, part of the point of formalis-
ing at ali is to systematise, to have relatively few rules to cover rela-
tively many cases) and of simplicity of paraphrase (for another part of
the point of formalisation is to supply a technique for evaluating
informal argumenta}; 1 fear that it is just a fact of logical life that
these are competing desiderata.
A factor that may eometimes help one to decide such a competition
is that a sacrifice either of austerity of formalism or of simplicity of
paraphrase will be better justified the broader the scope of the advan-
tage to be gained by it. For example, one would hope that a fonnalism
equipped to cope with temporal discourse rnight also be able to repre-
sent discourse about action and discourse about causation - and
clearly, if only one approach succeeded here, that would be reason to
prefer it. (Bee Lacey 1971 for relevant discussion: and cf. Davidson
1968a where it is argued that to represent action and causal state-
ments one needs to quantify over events, Recall (p. 124) that David-
son, like Quine, is committed to restricting himself to an extensional
formalism.)
Quine appeals to the character of current physical theories to
support his approach; Geach, on Prior's behalf, urges that it Í8 quite
improper to adjust logic to suit science. The issues here are tangled.
162 Philosophy of logics

Geach's attitude derives, in part, from the fact that he apparently

regards relativity theory as incoherent, since it involves denying what
he takes to be a 'difference of category' between space and time. And
his conviction that there is such a difference of category derives in
turn from our ordinary concepts of space and time, as embodied in
our ordinary, tensed discourse. Those who, like myself, allow that
developments in physics may well lead to conceptual revision, will
resist this facile diagnosis of relativity theory as 'conceptually
confused.'
But, quite apart from the question of the coherence or otherwise of
relativistic physics, there is a deeper point at issue. Quine expressly
aims, in his choice of logical formalism, for a 'language adequate for
science', and regards logic as, so to speak, continuous with science;
Geach regards logic as autonomous of, indeed prior to, science. The
history of logic offers sorné- support for the former view; for example,
the logic devised by Frege and Russell, unlike Aristotle's syllogistic,
can express relations as well as properties; and it is just because of this
superiority of expressive power that modern logic is capable - as
Aristotelian logic was not - of representing the sorts of argument
essential to modem mathematics. One needs, though, to distinguish
the question of the expressive power of logic from the question of its
doctrinal content ; I mean, that while it seems unexceptionable to
modify the expressive power of one's formalism to enable it to
express styles of argument characteristic of science, it is a more
serious matter to give up a supposed law of logic (as the quantum
logicians, for instance, urge that one give up the distributive law)
because of developments in science. This suggests that extensions of
logic are less radical, epistemologically speaking, than restrictions of
it; a point to which I shall return in ch. 1:2.

4 Second case-study: precisification versus 'fuzzy logic'

A good deal of informal discourse is, in sorne degree, vague.
And so the question arises whether, and if so, how, logicians should
take account of this fact.
A first point to make is that an important reason for constructing
formal systems of logic is to supply precise canons of validity - a
major advantage of formal logic over unregimented informal argu-
ment is its much greater rigour and exactness. In view of this it is not
surprising that Frege and Russell should have regarded vagueness as
a defect of natural languages, to be banished from an acceptable
 Logic and logics

formal language. (And no doubt it is also relevant here, as to their

neglect of tense considerations, that they were primarily concemed
with the formalisation of mathematical argument.)
This perhaps suggests that it would be appropriate simply to
exclude vague sentences as ineligible for logical treatment. But this
strategy is too crude, I think, because it is clear that vague sentences
can occur in informal arguments without threatening their validity.
There is a significant contrast, here, with the case of meaningless
sentences. An argument must be composed of meaningful sentences:
a meaningless string of symbols wouldn't be an argument, and a
sequence of meaningful sentences with a meaningless string inter-
posed would, if regarded as an argument at all, be valid or invalid
independently of the meaningless string. So it's entirely reasonable
to exclude meaningless sen ten ces from the scope of logic; 'logics of
meaninglessness' ( e .g. Halldén 1949, Routley I 966, 1969) are, to my
mind, neither necessary nor desirable.1 But a vague sentence can
play a genuine role in an argument (' J ohn likes capable girls; Mary
is capa ble and intelligent; so J ohn will like Mary '); and so logicians
must take vagueness more seriously.
However, vague sentences do seem to present certain difficulties
for the application of the standard logical apparatus. Formal logical
systems are supposed to be relevant to the assessment of informal
arguments; but the classical logical systems, in which every wff is
eithcr true or else false, seem inappropriate for the assessment of
informal arguments with premises and/or conclusions which, because
of their vagueness, one hesitates to call either definitely true or
definitely false. Once the problem has been put in this way, there
seem to be two natural approaches to its solution: to tidy up the
vague informal arguments before submitting them to assessment by
the standards of classical, z-valued logic, or, to devise sorne altema-
tive formal logical system which will apply to them more directly.
The first approach calls for informal arguments to be regimented
so that the standard logical apparatus can be used. (The procedure
would be quite analogous to the accommodations regularly made to
take account of the discrepancies between the truth-functional
connectives and their English readings.) Camap proposes (1950
1 I don't mean to deny that there may be sorne ínteresting philosophical
issues about the character and sources of meaninglessneaa (consider the
role played by the alleged meaninglessness generated by 'category
mistakes' in Ryle 1949, for example).
164 Philosophy of Iogics

ch. 1) what he calls a programme of precisification: before formalisa-

tion vague should be replaced by precise, e.g. qualitative by compara-
tive or quantitative, predicates, in such a way that (usually but not
invariably) the precise terms correspond in extension, in all clear,
central cases, with the vague ones they replace, but also have well-
defined application in cases which were borderline for the vague
terms. This proposal involves elements of both the first and the third
strategies distinguished in §2: informal arguments are tidied up
before receiving formal representation (strategy 2), but in such a
way that the regimented arguments always avoid the vagueness of the
originals (hints of strategy l ).
Sorne writers (e.g. Russell 1923, Black 1937) have urged that
natural languages are wholly vague; and if this were so, of course,
Carnap's programme couldn't be carried out. However, no very con-
vincing arguments have been offered why precision is impossible
in prínciple (cf. Haack 1974 ch. 6), and I shall proceed on the assump-
tion that precisification is feasible.
But granted that it is possible, it is desirable? Sorne support for a
different strategy - of altering classical logic to fit informal argument,
rather than informal argument to fit classical logic - has derived from
the belief that the sort of successive refinement of scientific concepts
urged by Camap may result in restricted applicability and unmanage-
able complexity. Indeed, it is significant that the writer responsible
for the most influential proposals for a revised logic of vagueness is an
electrical engineer whose earlier work (Zadeh 1963, 1964) had been
devoted to refining such concepts as 'static' and 'adaptive', but who
eventually concluded (Zadeh 197:2) that "fuzzy thinking' may not
be deplorable, after all, if it makes possible the solution of problems
which are much too complex for precise analysis '. The idea that
increase of precision may not be an unmixed blessing is not new;
Duhem pointed out (1904 pp. 178-9) that the statements of theoreti-
cal physics, just because they are more precise, are less certain, harder
to confirm, than the vaguer statements of common sense. Popper
(1961, 1976) has also suggested that precision may be a 'false
ideal',
What is the altemative to precisification? Well, if informal argu-
ments are not to be regimented so that classical logical apparatus can
be applied, perhaps the logical apparatus can be modified in sueh a
way that it can be applied to unregimented informal arguments. It
has been suggested, for example, that a 3-valued logic would be more
 Logic and logics

suitable than classical z-valued logic (Korner 1966); the idea is that
the trouble with vague predicates like 'tall' is that there are border-
line cases, i.e. cases of which the predicate is neither straightforwardly
true nor straightforwardly false, and that this problem may be solved
by allowing a third category, distinct from 'true' and 'false', to
accommodate borderline cases. But this doesn't solve the problem at
all satisfactorily; for it requires a sharp line to be drawn between
border-line cases and central, true or false, cases. Yet surely to insist
that at a given height a man ceases to be a borderline case aad be-
comes definitely tall, no less than to insist that at a given height a
man ceases to be not tall and becomes definitely tall, imposes an
artificial precision.
Zadeh also recommends that a non-standard logic be adopted, but
his 'fuzzy logic' represents a much more radical departure from the
classical. I'll first briefly sketch the salient formal features of fuzzy
logic. (For fuller details the reader is referred to Zadeh 1975, and
Gaines' survey, 1976.) Zadeh's non-standard logic is devised on the
basis of a non-standard set theory, 'fuzzy' set theory. Whereas in
classical set theory an object either is or is not a member of a given
set, in fuzzy set theory membership is a matter of degree; the degree
of membership of an object in a fuzzy set is represented by sorne real
number between o and 1, with o denoting no membership and I ful/
membership. (A fuzzy set will thus consist of all those objects which
belong to it in any degree, and two fuzzy sets will be identical íf the
same objects beJong to them in the same degree.) Now fuzzy set
theory can be used to characterise, semantically, a non-standard
logic; as values of sentence letters, instead of the two classical values,
one has the indenumerably many values of the interval [o, 1J, and
sentence connectives can be associated with set-theoretic operations
in the usual way ( e.g. negation with set complementation, implication
with set inclusion, etc.). The upshot is an indenumerably many-
valued Iogic. The exact character of this logic will depend upon the
characterisation of the fuzzy set-theoretical operations; one quite
natural set of assumptions yields the indenumerably many-valued
extension of t.ukasiewicz's 3-valued logíc (p. 2o6). Fuzzy logic is
constructed on the basis of one or another indenumerably many-
valued logic. There is thus a family of fuzzy logics, each with ita own
base logic. The indenumerably many truth-values of the base logic
are superseded by denumerably many fuzzy truth-values, which are
fuzzy subsets of the set of values of the base logic, characteriaed as:
166 Philosophy of logics

true, false, not true, very true, not very true, more or less
true, rather true, not very true and not very false ...
(Zadeh 1975 p. 410)

True is defined as a specified fuzzy subset of the set of values of the

base logic, and the other linguistic truth-values are then defined;
very true, for instance, is true"; if degree of truth o·8 belongs to true to
degree 0·7, it belongs to oery true to degree 0·49.
What this amounts to, at an intuitive level, is something of the
following kind. A vague predicate is taken to determine, nota classical
set, but a fuzzy set; for example, a person a may be tall to sorne
degree. lf, say, a belongs to degree 0·3 to the set of tall people, then
the sentence 'a is tall' would receive, in the base logic, the value 0·3
(' x is tall' is true to degree n iff x E tall to degree n). But according to
Zadeh 'true' itself is vague, and so it receives analogous treatment;
the degree of truth which 'p' has may be quite low, rather high, not
very high ... etc. The linguistic truth-values of fuzzy logic can be
thought of as corresponding to rather low ('not very true'), rather
high ('very true'), not very high ('more or less true') degrees of truth
in the base logic. So, to return to the example, if a E tall to degree
0·3, so that 'a is tall' has value 0·3 in the base logic, it will have, say,
the value not very true in fuzzy logic, since its degree of truth is
rather low.
In brief, one could think of fuzzy logic as the result of two stages of
'fuzzification': the move from z-valucd to indenumerably many-
valued logic as a result of allowing degrees of membership to sets
denoted by object language predicates, and the move to countably
many fuzzy truth-values as a result of treating the metalinguistic
predicate ·' true.' as itself vague. The term 'fuzzy logic' is sometimes
also used of the non-standard base logics; but I havefollowed Zadeh's
own, more restricted, usage, in which 'fuzzy logic' denotes a family
of systems with fuzzy truth-values. And, according to Zadeh, the
second stage of 'fuzzification' has radical consequences -. Among the
most notable - not to say alarming - are these. It tums out that the
set of truth-values of fuzzy logic is not closed under the operations of
negation, conjunction, disjunction and implication: for example, the
conjunction of two sentences each of which has a linguistic truth-
value in that set may not itself have such a value. So fuzzy logic has
'fuzzy truth-values ... imprecise truth-tables ... and ... rules of infer-
ence whose validity is approximate rather than exact' (1975 p. ,4,07).
 Logic and logics

Consequently, Zadeh claims, in fuzzy logic such traditional concerns

as axiomatisation, proof procedures, consistency, and completeness
are only 'peripheral' (Zadeh and Bellman 1976 p. 64). Fuzzy logic,
in brief, is not just a logic for handling argumenta in which vague
terms occur essentially; it is itself imprecise. I t is for this reason that
I said that Zadeh's proposal is much more radical than anything
previously discussed; for it challenges deeply entrenched ideas about
the characteristic objectives and methods of logic, For the pioneers of
formal logic a large part of the point of formalisation was that only
thus could one hope to have precise canons of valid reasoning. Zadeh
proposes that logic compromise with vagueness.
One is faced, here, with a striking example of strategy 7, a radical
challenge to the traditional conception of the scope and aims of
formal logic. In fact, we have seen that responses to vagueness have
ranged ali the way from the most conservative (attempts to exclude
vague sentences altogether from the scope of logic) through the
moderately innovative (proposals for a 3-valued logic of vagueness)
to the most radical (the proposal that logic abandon its aspirations to
precision).
Precision is certainly too central and important a desideratum of
formalisation to be lightly surrendered. And in the present instance
I think one is justified in asking whether the benefits can be expected
to outweigh the costs. Obviously, the adoption of a fuzzy logic would
result in a pretty serious loss in terms of simplicity (Zadeh himself
admits that fuzzy logic is in sorne ways much less simple even than its
non-standard base logic); and if one recalls that the reason Zadeh
gives for preferring to make logic imprecise rather than to make
informal arguments precise is that the latter is apt to introduce un-
manageable complexity one is likely to feel even more doubtful
whether the game is worth the candle. For another thing, it isn't even
clear that fuzzy logic avoids the artificial imposition of precísíon. In
the base logic, though one isn't obliged to insist that (say) Jack must
be either defi.nitely tall or definitely not tall, nor that he must be
either definitely tall or definitely not tall or definitely borderline, one
will be oblíged to insist that he be tall to degree 07 or tall to degree
o· 8, or ... etc.; and in the resulting fuzzy logic one will be obliged to
insist that, if 'J ack is tall' is true to degree o·8, it should count as
very true, or only as true but not very true, or ... etc. Zadeh propases
to define true as:
true = 0·3/0·6+0·5/0·7+0·7/0·8+0·9/0·9+ 1/1
168 Philosophy of logics

i.e., as the fuzzy set to which degree of truth o·6 belongs to degree
0·31 0·7 to degree 0·5, o·8 to degree 0·7, 0·9 to degree 0·9, and I to
degree 1 ( 1975 p. 411); is not this an artificial imposition of precision?
lt is hard to avoid the suspicion that Zadeh's programme brings only
doubtful benefits, and at excessive cost.

Postscript: degrees o/ truth

Zadeh's second stage of fuzzification - the extension of
fuzzy set theory to 'true' and 'false' -is based on the idea that truth
is a matter of degree, and is reflected in his list of linguistic truth-
values, where adverbial modifiers such as 'not very' and 'more or
less' (which he calls 'hedges') are attached to 'true' and 'false'. But
Zadeh's list of linguistic truth-values is extremely odd: for example,
though 'very true' and 'more or less true' sound acceptable, 'rather
true', 'slightly true', and, for that matter, 'not very true' seem to me
quite bizarre. This prompts me to look a little more closely at the
linguistic evidence.
Among the adverbial modifiers which do apply to 'true' one has
'quite' and 'very'. Now 'quite' and 'very' apply to predicates of
degree, i.e. predicates which denote properties which come in degrees
(quite tall, heavy, intelligent ... , very tall, heavy, intelligent ... )
where they indicate possession of the property in, respectively,
modest or considerable degree. And Zadeh apparently thinks that,
analogously, 'quite true' indica tes the possession of a modest degree of
truth, and 'very true' the possession of a high degree of truth. But
whereas 'quite tall (heavy, intelligent)' can be roughly equated with
'rather (fairly) tall (heavy, intelligent) ', 'quite true' certainly doesn't
mean anything like 'rather true' or • fairly true'. For 'rather' and
'fairly ', like other adverbs which typically modify adjectives of
degree, just don't apply to 'true' (I follow linguists' practice of
starring unacceptable locutions):
• rather true
• fairly true
• somewhat true
• slightly true
• extremely true
In fact, 'quite true' can be roughly equated with 'perfectly true' or
'absolutely true', and (so far from contrasting with it) 'very true'.
Again, when "quite.' ( or 'rather' or 'fairly ') is attached to a predicate
 Logic and logics 169
of degree, as in 'quite (rather, fairly) tall (heavy, intelligent)' it cannot
be preceded by 'not' (' not quite tall' is unacceptable); whereas
when it is attached to an absolute predi cate, as in 'quite ready ', it can
('not quite ready'). The behaviour of 'quite' and 'very' with 'true',
so far from supporting the hypothesis that 'true' is a predicate of
degree, indicates that it is an absolute predicate.
But what about other adverbial modifiers which apply to 'true',
such as 'wholly ', 'completely ', 'substantially ', 'largely ', 'partially ',
'more or less ', 'approximately ', 'essentially ', 'not strictly ', 'not
exactly' ... and so forth? I conjecture that it may be possible to
explain such locutions without treating truth as a matter of degree ¡
roughly, one might expect something along the lines of' 'p' is wholly
true iff the whole of 'p' is true', "p' is partially true iff part of 'p' is
true',' 'p' is approximately true iff 'approximately p' is true' ... etc.
These issues will receive further attention in ch. 11 §3.
 10

Modal logic

Necessary truth
Modal logic is intended to represent arguments involving
essentially the concepts of necessity and possibility. Sorne preliminary
comments about the idea of necessity, therefore, won't go amiss.
There is a long philosophical tradition of distinguishing between
necessary and contingent truths. The distinction is often explained
along the following lines: a necessary truth is one which could not
be otherwise, a contingent truth one which could; or, the negation of
a necessary truth is impossible or contradictory, the negation of a
contingent truth possible or consistent; or, anecessary truth is true in
all possible worlds (pp. 188:ff. below), a contingent truth is true in the
actual but not in all possible worlds. Evidently, such accounts aren't
fully explanatory, in view of their 'could (not) be otherwise', '(im)-
possible', 'possible world '. So the distinction is sometimes intro-
duced, rather, by means of examples: in a recent book (Plantinga
1974 p. 1) '7 + 5 = 1:2 ', 'If all menare mortal and Socrates is a man,
then Sócrates is mortal' and 'If a thing is red, it is coloured' are
offered as examples of necessary truths, and 'The average rainfall in
Los Angeles is about 12 inches' as an example of a contingent truth.
The distinction between necessary and contingent truths is a
metaphysical one; it should be distinguished from the epistemological
distinction between a priori and a posteriori truths. An a priori truth
is one which can be known independently of experience, an a
posteriori truth one which cannot, These - the metaphysical and the
epistemological - are certainly di:fferent distinctions. But it is con-
troversial whether they coincide in extension, whether, that is, all and
only necessary truths are a priori and all and only contingent truths
 Modal logic

a posteriori. Opinion on this question has fluctuated: Kant thought

there were contingenta priori truths; the logical positivists insisted
on the coextensiveness of the necessary and the a priori, and the
contingent and the a posteriori; Kripke has recently (1972) urged that
there are, after ali, contingent a priori (and necessary a posterior,")
truths. I shan't enter into this question here, where necessary truth
is the main preoccupation; it will have sorne relevance when I come,
in ch. 12 §3, to the question of the epistemological status of logic.
Among necessary truths, it is also traditional to distinguish
physically necessary truths (truths which physically could not be
otherwise, the negations of which are physically impossible, true in
all physically possible worlds) and logically necessary truths (truths
which logically could not be otherwise, the negations of which are
logically impossible, true in all logically possible worlds). Sometimes
physical necessity is explained by means of logical necessity, as
logical compatibility with the laws of nature. Or, again, one may
resort to examples: 'Any two material bodies attract each other with
a force proportional to their mass' can serve as an example of a
physically necessary truth, 'If any two material bodies attract each
other with a force proportional to their mass, then any two material
bodies attract each other with a force proportional to their mass' as an
example of a logically necessary truth. Sorne philosophers are scepti-
cal about this distinction; see e.g. Kneale 1962a, Molnar 1969, and
cf. Quine, 'Necessity', in 1966a. And of course the question whether
there are any physically necessary truths raises important issues in
the philosophy of science. But modal logics were designed, primarily,
with the object of representing logical rather than physical necessity
and possibility, which is why I only mention, and do not answer, the
intriguing questions raised by the idea of physical necessity.
The distinction between logically necessary and logically contin-
gent truths has sometimes been thought to rest, in tum, on that
between analytic and synthetic truths. 'Analytic' and its opposite
'synthetic' have been variously defined: Kant defined an analytic
truth as one the concept of whose predicate is included in the concept
of its subject or - arguably not equivalently - as one the negation of
which is contradictory; Frege defined an analytic truth as either a
truth of logic, or a truth reducible to a logical truth by means of
definitions in purely logical terms (thus, logicism is the thesis that the
truths of arithmetic are, in this sense, analytic). More recently,
analytic truths have been characterised as 'true solely in virtue of
172 Philosophy o/ logics

their meaning ', synthetic truths as 'true in virtue of facts'; with

truths of logic being thought of as a subclass, true in virtue of the
meaning of the logical constants, of the larger class of truths in
virtue of meaning. (Hintikk.a 1973 is informative on the history of
'analytic'. Notice the characteristic shift from Kant's quasi-psycho-
logical account, in terms of the concepts involved injudgments to more
recent linguistic characterisations in terms of the meanings of the
component words of sentences.)
Analyticity is thought to explain the grounds of necessary truth,
what makes a necessary truth necessarily true. So the necessary/
contingent and analytic/synthetic distinctions are supposed to
coincide, the idea being that an analytic truth, being true solely in
virtue of its meaning, couldn't be otherwise than true, and so, is
necessary.1
Now Quine is sceptical of the analytic/synthetic distinction ;2 and
his scepticism is, as we shall see, one of the reasons for his distaste for
modal logic.
The critique of analyticity in 'Two dogmas' is directed, primarily,
at the second disjunct of a roughly 'Fregean' account of analyticity,
as:
1 But words can change their meaning; and if they do, may not
previously analytic sentences becorne synthetic or false? The supporters
of analyticity might reply that though one and the same sentence may
at one time expresa an analytic truth, and at another time a synthetic
truth, or perhaps a falsehood, the proposiüo« originally expressed by
the sentence remains analytic though the sentence ceaaes to express it,
I
Historical note: Quine didn't always reject the analytic/synthetic
dístinctíon; in 1947 he used the concept of analyticity, though he
comments in a footnote that Goodman had been urging scepticism
about it. And before 'Two dogmas' appeared Morton White (1950)
had already attacked the analytic/synthetic distinction as an 'untenable
dualism'. lt was Quine's attack, however, that proved the most
influential. By 196oa Quine's scepticism about synonymy was buttressed
by hís thesis of the indetenninacy aj translation: the thesis that meaning
notions are not just, as he urged earlier, obscure and of doubtful
empírica) content, but demonstrably indetenninate. In 1973, however,
though remaining officially sceptical about meaning, Quine offered
his own ersatz conceptíon of analyticity: a statement is analytíc, in thís
sense, if everyone in the Iinguistic community learns that it is true by
learning to understand it; notice that this conception is, charac-
teristically, both genetic and social. For the present I shall be concerned
only with Quine's straightforwardly sceptícal viewa, with the period
between 1951 and 1973; the reader is referred to Haack 1977c, where
I have argued that his new conception of analyticity is apt to collapse
into the traditional conception of truth-in-virtue-of-meaning.
 Modal logic
173

A is analytic iff either:

(i) A is a logical truth
or
(ii) A is reducible to a logical truth by substitution of
synonyms for synonyms

lt �'.ll be convenient. to call the class of statements falling under (i)

or (u) _broadly analytic, and those which qualify under (ii) narrotoly
analytzc; broa? _analyticity, in this terminology, is Iogical truth plus
n�rro�. anal�1c�ty. The picture Quine rejects is portrayed in fig. 5;
his critique rs aimed at the concept of narrow analyticity.

(logically) (logically)
versus
necessary contingent

1
i.e.
broadly i.e.
anal y tic synthetic

»<.
logically narrowly
--E
Quine's
true analytic attack
Fig. 5

Specifically, Quine's strategy is to argue that no satisfactory expla-

nation can be given of the second clause, or of the conception of
synonymy on which it relies. The explanations which have been
offered, he claims, either fail correctly to characterise ali the supposed
analytic truths (for instance, Camap's account of analyticity as truth
in ali state-descriptions, he argues, applies only to logical truths and
not to narrowly analytic truths which should qualify under clause
(ii)) or else tum out to depend, overtly or covertly, upon an under-
standing of sorne other intensional notion no clearer than analyticity
itself. If, for example, clause (ii) is explained in terms of substitution
on the basis of definitions, this involves an indirect appeal to the
synonymies upon which the definitions are based; nor, again, can
synonymy be explained as substitutability in ali contexts salea
TJeritate (i.e, without change of truth-value), unless contexts such as
'Necesaarily .. , ' are taken into account. In brief, explanations of
174 Philosophy of logics

analyticity can never break out of an 'intensional circle' of concepts

no clearer than what is being explained (see fig. 6 ).
analyticity
�
/ reducibility to logical
(
truths plus semantic rules
)
reducibility to logical
truth plus synonymy

reducibility to logical
truth plus definitions
\ /
synonymy as substitivity
Fig. 6 in ali (including modal) contexts salva veritate

This is not the place for a full-scale discussion of the argument of

"Two dogmas' (I shall have sorne more to say about it in ch. 12 § 3);
my present object is, rather, to bring out sorne points which are
especially relevant to Quine's attitude to modal logic.
First: Quine's attack, however successful, threatens only narrow
analyticity: logical truths, which qualify as analytic under clause (i),
are unaffected. The distaste Quine feels for the concept of narrow
analyticity does not extend to the concept of logical truth. This will
be relevant to discussion (pp. 182, 193 below) of whether the necessity
operator in the usual modal logics can be understood as representing
logical truth, or whether it must correspond to a broader conception
of necessity.
Quine characterises a logical truth as 'a statement which is true and
remains true under all reinterpretations of its components other than
the logical particles' (1951 pp. 22-3; and cf. 1970 ch. 4). Here, as
elsewhere, Quine is careless of the distinction between the system-
relative idea of the logical truth of a wff of a formal language, and the
extra-systematic idea of the logical truth of a statement of a natural
language. 1 suggested (pp. 14-15) that the extra-systernatic idea of
logical truth amounts, initially, only to a none-too-precise idea of a
statement which is trivially true. However, if this idea is refined in
the same way that the idea of a valid argument as one whose premises
couldn't be true and its conclusion false is refined by the insight
that an argument is valid if there is no argument of the same form
 Modal logic 175

with true premises and false conclusion, the result is the idea of a
statement such that no other statement of the same form is false:
which is very close to Quine's characterisation.1
Second: Quine's objection to narrow analyticity rests, at bottom,
on the idea that no explanation can be given of it except by means of
other terms from the 'intensional circle ', and that ali such tenns are
unclear. This will be relevant to discussion (pp. 189-90, 193-4 below)
of whether, in the interpretation of modal systems, it is realistic to
hope for a non-modal explanation of modal terms.
First, though, I shall be concerned with a syntactical characterisa-
tion of modal logics.

2 Modal systems
Extensions o/ classical logic
One system is an extension of another if it shares the vocabu-
lary of the first, and has the same theorems and valid inferences
involving only the shared vocabulary, but also has additional vocab-
ulary, and additional theorems and/or valid inferences involving that
vocabulary essentially; an 'extended logic' is a system which is an
extension of classical logic ( ch. 1 §2, ch. 9 § 1 ). Extensions of classical
logic are often motivated by the belief that the standard sentence and
predicate calculi are, though unobjectionable, less than fully ade-
quate: their theorems are logically true, and their valid sequents
truth-preserving, but there are other logical truths and/or valid
arguments which involve operations for which they Iack the vocabu-
lary, which they cannot even express.
Modal logic adds to the classical vocabulary the r-place operators
'L ', to be read 'necessarily ', and 'M', to be read 'possibly ', and the
z-place operator' -3', to be read 'strictly implies'. (Other extended
logics, such as epistemic logic, which adds the operators 'K', to be
read '» knows that ', and 'B ', to be read 'x belíeves that'; deontic
logic, which adds the operators '0', to be read 'It ought to be the

1 Straweon has urged (1957) agaín1t Quine, that the explanation of

logical truth, líke the explanation of narrow analyticity, requires appeal
to synonymy. How is one to lcnow that ' lf he is sick then he ia sick'
i1 logically true, he 88ks, unless one has ueurance that ' he is sick'
mean, the aame at each occurrence? The reply should be, 1 think, that
where appeal to meaning may be neceeeary ia, rather, with respect to
the queetion, whether ' lf he ia 1ick then he is sick' is appropriately to
be repraented, formally, u 'p-+ p'.
176 Philosophy o/ logics

case that', and 'P', to be read 'It is permitted that'; and tense logic
(ch. 9 §3) are modelled rather closely on modal logic.)

Historical remarks
The logic of modal sentences was discussed by Aristotle and
by medieval logicians; in the present century Hugh MacColl (1880,
1906) contributed formal as well as philosophical proposals. But
sustained formal development carne in the present century, in thewake
of Frege's and Russell's development of non-modal sentence calculi.
The first axiomatisations of modal sentence logic were given by
Lewis 1918. The extension to modal predicate logic carne with
Marcus 1946.
The primary motivation for Lewis's development of modal
logics was dissatisfaction with the notion - central to the logic of the
Begriffsschrift and Principia Mathematica - of material implication.
Since 'p' materially implies 'q' if either 'p' is false or 'q' is
true, one has the theorems, the so-called 'paradoxes' of material
implication:

P-+(q-+p)
-p-+(P-+q)
(p -+ q) V (q -+ P)

The material implication of classical logic, Lewis held, is quite

inadequate to the intuitive notion of implication, which requires.
not just that 'p' is not true and • q • false, but that 'p • could not be
true and 'q' false. So he proposed that the logic of Principia should
be enriched by a new operator, for strict irnplication, which could be
defined as necessity of material implication.

A formal sketch
Only one modal operator need be added, as a primitive, to the
vocabulary of classical logic; with • L • (' necessarily ') as primitive,
• M' ('possibly') is standardly defined as:

MA =df -L-A
and' -3' as:
A -3 B = df L(A -+ B)

Or 'M' may be taken as primitive and 'LA' defined as' -M-A'.

 Modal logic

In the usual modal logics the formation rules allow 'LA' as a wff
whenever 'A' is a wff; this, of course, allows for iterated modalities,
such as' LMp' or 'LLp'.
Now there is not one, but a whole range, of modal logics, di:ffering
from each other with respect to the strength of the axiorns they admit
to govern the modal operators. I shall sketch sorne of the better-
known systems, in order of increasing strength.
So· 5, one of the weakest modal systems, results from the addition
of the axioms:
r, Lp--> p

and:
2. L (p --> q) --> (Lp --> Lq)
together with the rule:
(R) If A is a theorem of sentence calculus, then 1-80.5 LA
The system T results from the strengthening of (R) to:
(RN) If 1-T A, then 1-T LA
(so that 1-LA not only when A is a theorem of sentence calculus (as
with (R)) but also when it is one of the added axioms I and 2, etc.;
another contrast with (R) is that (RN) is iterable, so that one gets
1-LLA, 1-LLLA, etc.).
The systern 84 results from T by the addition of the axiom:
3. Lp--+ LLp
and the system S 5 from 84 by the addition of:
4. Mp->LMp
There are other modal systems as well, weaker, stronger and in
between these. The exact character of the quantified modal logic, also,
may differ according to sorne variations in the presentation of the
underlying predicate calculus. (For fuller details, consult Hughes
and Cresswell 1968.)

Relations bettoeen the modal systems

The proliferation of modal systems immediately raises the
question whether one is obliged to choose between them, whether,
that is, each aims at - and so, at most one succeeds in - capturing
178 Philosophy of logics

just the logical truths/valid inferences involving the notion of neces-

sity, or whether, perhaps, each aims - and so, all may succeed - at
capturing a sense of 'necessary '. This, to anticipa te an idea I shall
discuss in more detail subsequently (ch. 12 §1), is the question,
whether the various modal systems should be regarded as rivals of
each other. Lemmon has argued (1959) for a tolerant, pluralistic
approach; each of the modal systems, he thinks, can be seen as
formalising a different idea of necessity: for example, So· 5 the idea
of tautologousness, 85 the idea of analyticity. Others, however,
(e.g. Cargile 1972) doubt the feasibility of Lemmon's interpretations.
Among those who believe that there is a correct modal logic, the
stronger systems, 84 and 85, seem to be most often favoured.

3 Criticisms of modal logic

Doubts go deeper, however, than the dispute about whether
there is a correct modal logic, or which modal logic is the correct one.
For the feasibility, and even the intelligibility, of the whole enterprise
of modal logic has been questioned; Quine, most notably, has long
challenged it (but cf. also e.g. Bergmann's criticisms in 1960).
Quine's objections are threefold: that the motivation for the
development of modal systems rested u pon a confusion; that modal
logics are anyway not needed for any of the legitimate purposes of
formalisation; and that the interpretation of modal logics presents
insuperable difficulties. Underlying these objections, of course, is
Quine's deep-seated scepticism about the concept of analyticity. It is
against this background of scepticism about the status of modal
notions that one should see Quine's objections to modal logics.

Modal logic 'soas conceived in sin'

' � ', or 'strictly implies ', Lewis argued, was needed because
of the excessive weakness of '-+' or 'materially implies '. Now Ouine
points out that 'materially implies' is anyway a grammatically
improper reading of '-+'; for '-+' is a sentence-forrning operator
upon sentences, while 'materially implies' is a z-place predicate.
So modal logic was 'conceived in sin', the sin of confusing use (as in
'P-+ q') and mention (as in "p' materially implies 'q"). It seems
that Lewis did, indeed, succumb to this confusion, helped, no doubt,
by Russell's bad example; it is also pretty clear, however, that this
grammatical misdemeanour need not vitiate the enterprise of modal
logic (and cf. Belnap 1974 for arguments why grammatical deviance
 Modal logic 179
may be positively desirable in logical innovations). Grammar, as
Quine insists, deplores the reading of '�' as 'materially implies';
nevertheless, there is a relation that holds between two sentences
'p' and 'q' just when p � q, a weak relation which may, in ali
grammatical propriety, be referred to as 'material implication'; and
modal logic formalises another relation that holds between two sen-
tences 'p' and 'q ', a stronger relation which may be referred to as
'strict implication '.

Moda/ logic is not needed

Modal logics are, as I explained, extensions of classical logic;
Quine suggests (e.g. in 196oa §,p) that such extensions are not
needed. The question arises, of course, 'needed for what?'. Quine
holds that the aim of formalisation, or, as he puts it, 'regimentation'
of informal argument is to achieve a precise language 'adequate for
science'; and for the purposes of science, he believes, modal notions
are not required.
The assumption that the aim of formalisation is a language 'ade-
quate for science' may be challenged, even though Quine, I think,
understands 'science' pretty broadly, including mathematics as well
as physics, chemistry, biology, psychology and the social sciences,
and commonsense cognitive discourse as well as the official talk of
professional scientists. Certainly sorne logicians regard it as part of
their job to devise a language adequate, also, to represent argument
in, for example, moral discourse (cf. Smiley 1963) or discourse in
fiction (cf. ch. 5 §4). The claim that modal notions are inessential to
scientific discourse is, again, controversial. lt is particularly difficult
to get an undistorted perspective on this issue, because Quine him-
self is - naturally enough - inclined to apply specially stringent
standards when considering claims that scientific discourse requires
the broad conception of necessity which he anyway rejects. In other
words, Quine's claim that these concepts are not needed, and his
claim that they are empty, inevitably no doubt, interact.

An example: dispositions and the subjunctive conditional

The best way to understand what is at issue here may be, therefore,
to consider in sorne detail a case where it has been held, but where
Quine denies, that certain locutions are (i) essential to scientific dis-
course and (ii) inexplicable except in modal tenns. A family of
locutions which is apparently very deeply entrenched in the language
180 Philosophy of logics

of science is the idiom of dispositions and its close relation, the sub-
junctive conditional. To say, for instance, that x is water-soluble, is to
say that if x were placed in water, x would dissolve. The material
conditional of truth-functional sentence calculus is inadequate to
represent the subjunctive conditional, for 'A --+ B' is true if 'A' is
false, while one does not suppose that 'x is soluble' or 'if x were
placed in water, x would dissolve' need be true just because x has
never been placed in water. Some writers believe that an adequate
formal representation of the subjunctive conditional requires modal
apparatus, specifically, appeal to possibilities. Modal analyses of
subjunctive conditionals have been offered by Stalnaker 1968 and
by D. K. Lewis 1973. Quine, of course, takes a somewhat dim view
of such proposals, just beca use they use this modal apparatus; but
he has also, more relevantly to the present issue, argued that sub-
junctive conditionals can be accommodated without it. At one time,
it seems, Quine admitted that dispositional terms must be part of a
language of science, and offered an extensional analysis of them:
'x is soluble', for instance, was explained along the lines of '(3y) (x
has an interna! structure like y, which has been placed in water and
which dissolved)'. Sometimes, Quine observed, as in the case of
solubility, the relevant structure is known; sometimes, as in the case
of irritability, the reference to an interna! structure is no more than a
'promissory note' (see 196oa §46). It has been argued, however, that
this account does not allow the surely genuine possibility that all
things of a certain kind should have a given disposition, and yet none
of them should ever have rnanifested it, as, perhaps, all nuclear power
stations have a disposition to explode in certain circumstances, even
though, thus far, safety precautions have ensured that those circum-
stances have not arisen, so that none have ever exploded (Mellor
1974). Subsequently, anyway, Quine seems to suggest that disposi-
tional terms do not really belong, after all, to the language of science;
they are essential only while the enterprise of science is uncompleted,
but can be dropped once the relevant structures are known. One
might feel that Quine's attempt to exclude dispositional idioms from
a 'regimented theoretical language' would be more convincing had
it not been preceded by an abaridoned attempt to include them in an
extensional guise; and one may well feel discomfort with the appeal
to a finished, by contrast with an ongoing, science, for such a dis-
tinction fits especially ill with Quine's usually pragmatic approach to
the philosophy of science.
 Modal logic

As I have observed, Quine's conviction that we can do without

modal notions also rests upon his belief that the interpretation of
modal logic is so beset with difficulties that the use of such apparatus
is not really helpful anyway. So it is time to look at these difficulties.

The interpretation of modal logic is fraught with difficulties

These criticisms fall into two groups: the difficulties which
Quine finds in the interpretation of modal sentence logic, and the
additional difficulties which he finds in the interpretation of modal
predicate logic.
In 1953b Quine distinguishes what he calls the "three grades of
modal involvement', viz:
(i) the use of 'necessary' as a predicate of sentences. Here
'L' would apply to names of, or quoted, sentences, as in
'L '2 + 2 = 4"; it could be read ' ... is necessarily
true' and would have a strong analogy to ' ... is true' in
Tarski's theory, where is it treated as a predicate of
sentences;
(ii) the use of 'necessarily' or 'it is necessary that' as a
sentence-forming operator on sentences, as in
'L(2 + 2 = 4) ', where 'L' is treated as syntactically
analogous to 'it is true that ... ';
(iii) the use of 'necessarily' as an operator both on closed
sentences, as in (ii), and on open sentences, as in
'L(2 + 2 = x)' and its existential generalisation,
'(3.x) L(2+2 = x)'.
Modal sentence Iogic will require at most the second grade of
modal involvement, whereas modal predicate logic requires the
third. In "Three grades' it is clear that Quine regards (i) and (ii) as,
though by no means unproblematic, at least preferable to (iii); and
this corresponds to his view of modal sentence logic as, though by no
means unproblematic, at least preferable to modal predicate logic.

Difficulties in the interpretation of modal senience logic

In modal sentence logics of the conventional kind, 'L' and 'M' are
sentence operators, as in grade (ii). However, at least so long as one
sticks with single modal operators, one can always regard, say,
'L(2+2 = 4) ', as a syntactic variant on 'L '2+ 2 = 4", as in grade
(i).
182 Philosophy of logics

Since he dislikes the notion of analyticity, Quíne is less than happy

even about the use of' necessary' as a sentence predicate. However, he
admits the concepts of theoremhood and its semantic counterpart,
logical truth, so that the interpretation of 'LA' as ' 'A' is logically
true (a theorem)' is available to him. But this kind of interpretation
allows for only a fragment of the usual modal sentence logics, since it
leaves the status of iterated modalities in question. It does, however,
suggest an interesting line of thought: that if 'LA' were interpreted
as "A' is a theorem ( valid formula) of L ', where L is sorne formal
theory, then 'LLA' míght be interpreted as ''LA' is a theorem of M ',
where M is the metalanguage of L. In other words, íterated modal
operators would not be univoca}, but each would refer to theorem-
hood or logical truth in one of a hierarchy of theories, Modal logics
along these lines - motivated by the sorts of consideration just
discussed - have been devised (Príest 1976). The usual modal logics,
however, with their univocally interpreted íterated modal operators,
are not amenable to thís kind of approach.

Difficulties in the interpretation of modal predicate logic1

If the adjunction of modal operators to sentence logic is dubious, the
mixture of modal operators and quantifiers is, Quine argues, dis-
astrous.
Quine's difficulties with quantified modal logic derive, funda-
mentally, from the intersection of his views about quantifiers and his
views about modality. According to Quine (ch. 4 §2), since singular
terms are eliminable, it is the quantifiers which carry ontological
commitment; the quantifiers are the basic device by means of
which we talk about things. On the other hand, he takes modal
locutions to be talk, not directly about things, but about our ways of
talking about thi'ngs: 'necessity ', he comments, 'resides in the way we
say things, and not in the things we talk about' (1953b p. 174-). To
put it another way, Quine thinks that modality, insofar as it is
intelligible at all, is de dicto and not de re; 'necessary' and 'possible'
are predicates of sentences, not of extra-linguistic things: "2 + 2 = 4'
is necessary' (grade (i)) is comprehensible, but '2 + 2 necessarily = 4'
(grade (iii)) is not (cf. Plantinga 1974 ch. 1 §2 and ch. 2). Given his
view of the contrasting roles of quantifiers and modal operators, the
main theme of Quine's criticisms of quantified modal logic should
1
A useful preaentation and discuasion of these criticisms w to be found
in Fellesda) 1969; there is a reply by Quine in the aame volume,
 Modal logic

come as no surprise: when quantifiers and modal operators are

combined, it is hopelessly unclear what we are talking about.
Sorne of the difficulties show up in the anomalous behaviour of
singular terms in the scope of modal operators, Modal operators, as
Quine puts it, are ref'erentially opaque ( or intensional); substitutivity
(Leibniz's law) fails in modal contexts, that is to say: within the scope
of a modal operator, substituting one singular term for another which
denotes the very same object can change the truth-value of the
resulting sentence; so singular terms within the scope of the modal
operator are not purely ref'erential, do not serve, that is, solely to pick
out their referents. (In point of referential opacity, Quine argues,
modal operators are like quotation marks or epistemic operators.)
For example (Quine 1943, 1947, 1953b):
9 = the number of the planets

is true, yet substitution on the basis of this identity into the true
sentence:
L (9 > 7)
yields the presumably false sen ten ce:
L (the number of the planets > 7)

Since, however, Quine allows no very fundamental significance to

singular terms, which, after ali, he thinks can and should be elimi-
nated, the core of his objection lies in the anomalous behaviour of
quantifiers and bound variables within the scope of modal operators.
In non-modal predicate calculus:
(3x) (x > 7)
follows by existencial generalisation from:

9 > 7
and in modal predicate calculus, analogously,
(3x) L (x > 7)
follows from :
L (x > 7)
But Quine cannot accept that tker« is something rDlaich is Mcmarily
greater than 7 ('(3x) L (x > 7)'); the 'something', he argues, cannot
184 Philosophy of logics

be the number 9, for that is the number of the planets, and the num-
ber of the planets isn't necessarily, but only contingently, greater
than 7. Being necessarily greater than 7, Quine urges, can't be a
property of a number; it is only that that a number is greater than 7
necessarily follows if it is specified in certain ways (e.g. as the
number 9, oras the sum of 5 and 4) but not if it is specified in certain
other ways (e.g. as the number of the planets], If modal predicate
calculus requires one to accept that the number 9 has the property of
being necessarily greater than 7, it is committed to essentialism, the
thesis that things have sorne of their properties necessarily, or
essentially. But essentialism, according to Quine, is a 'metaphysical
jungle' (1953b p. 174), to which the only appropriate response is
'puzzlement' (196oa p. 199).
The difficulties thus far discussed, Quine allows, could be avoided
if one was prepared to place sufficiently stringent restrictions on the
universe of discourse, specifically, to admit only objects such that
any two conditions specifying them are necessarily equivalent, i.e.:
C: ((y) (Fy =
y = x) & (y) (Gy = y = x))-+
L (y) (Fy = Gy)1
Condition C restores substitutivity; i.e., given C:
(x) (y) ((x = y & Fx) -+ Fy)
However Quine points out (1953b pp. 155-6) that substitutivity,
together with the presumably true:
L (x = x)
yields the conseq uence that:
(x) (y) (x = y-+ L (x = y))
that is, that all identities are necessary, And this, Quine thinks, is
doubtful. (It has, for example, been held by sorne proponents of the
physicalist theory that the identity they claim between the mind and
1
Quine points out (1953a pp. 152-3, contra Church 1943) that restricting
the universe of discourse to intensional objects, e.g, numerical concepta
rather than numbers, wouldn't be sufficient to restare substitutivity,
For if a is such an intensional object, and p a sentence which is true
but not necessarily true, then:
a = (,x) (p & x = a)
But this identity i, not analytic, and ita two sídes are not ínter-
changeable in modal contexts ,alva VtTitate.
 Moda//ogic

the brain is contingent, not necessary; it is like, for instan ce, the
identity between lightning and electrical discharges in the atmosphere.
Such contingent identities, they assume, are commonplace in science.)
This is rather closely related to the first problem discussed, with
'L ( ... = x)' replacing 'L ( ... > 7) '. Indeed, the 'Moming Star
paradox' is another well-known version of the original problem:
necessarily (presumably) the Moming Star = the Morning Star; but,
though the Moming Star = the Evening Star, it is not necessary,
but contingent, that the Moming Star = the Evening Star.
The further consequences of imposing condition C, anyway, are,
Quine argues, even worse: modal distinctions collapse, for with
condition C one can prove that:
p�Lp
which, in view of the axiom 'Lp � p' means that Lp = p, so that
'L' is redundant.!
So the strategy of Quine's critique is this: adjunction of modal
operators leads to anomalous behaviour on the part of singular terms
and bound variables; these difficulties may be avoided by a restriction
on the universe of discourse, but at the cost of the collapse of modal
distinctions. The collapse of modal distinctions could not, of course,
be tolerated by the supporters of modal logic; the question is, there-
fore, whether they can avoid or explain what Quine sees as 'mis-
behaviour' on the part of singular terms and bound variables in
modal contexts. In one way or another, their responses consist, as
one would expect, in claiming that what Quine takes to be false (or
perhaps doubtfully intelligible) consequences of quantified modal
logic are, in fact, when properly understood, true. For instance, de re
modalities and essentialism are defended (e.g. Plantinga 1974) and it
is argued that all identities are, indeed, necessary [Marcus 1962,
Kripke 1972).
I can't look at all the replies which have been made to Quine's
criticisms, but will confine myself to a couple which serve quite well to
illustrate what is at issue. Several writers (e.g. Smullyan 1948,
Fitch 1949) argue that the apparent failure of substitutivity in modal
contexts can be shown, once one takes adequate care about the
I The argument goes: let p be any true sentence, and let F be
'p & y = x' and G be' y = x'; then from C it follows that
L(y) (p & y = x = y = x), whence, in particular,
L(p & x = x ;;; x = x), and so Lp.
1 86 Philosophy of logics

distinction between names and descriptions, to be mere/y apparent.

Smullyan argues thus:
9 = the number of the planets
is nota straightforward identity statement of which the two terms are
bona fide names, but, rather, has the form:

9 = (1.x) Fx
And the sentence
L (the number of the planets > 7)
which Quine takes to be straightforwardly false, is ambiguous;
depending on the scope (p. 65) given to the definite description it
can be understood eithcr as:
The number of the planets is necessarily > 7
oras
It is necessary that the number of the planets is > 7.
Of these, Smullyan argues, the first follows from 'L (9 > 7)' and
'9 = the number of the planets', but that is ali right, because it's
true; while the second is false, but that's also all right, because it
doesn't follow.
Smullyan's distinction blocks Quine's original argument in quite a
neat way. However, his solution requires one to accept the truth of
"The number of the planets is necessarily greater than 7' which has,
when the definite description is eliminated, the form:
(3x) ((y) (y numbers the planets = x ""' y) & L (x > 7))
But Quine, no doubt, would object to this, where a quantifier (thc
initial '(3x) ') binds a variable (the 'x' in "x > 7 ') inside a modal
context; this after ali is just his example of the misbehaviour of
bound variables in modal contexts. Smullyan's solution would be, in
Quine's eyes, unacceptably essentialist.
Marcus, however, denies (1962) that there is really anomalous
behaviour on the part of the quantifiers in modal contexts. Quine's
difficulties spring from his reading the quantifier objectually, as
'There is at least one object, x, such that x is necessarily greater than
7', and then asking what this object could be. Marcus proposes,
instead, that the quantifier be read substitutionally, as 'Sorne sub-
stitution instance of 'L ( x > 7)' is true' ; and this, she argues, is
 Modal logic

straightforwardly true, since 'L(9 > 7)', for example, is a true

substitution instance.
But Quine, of course, rejects the substitutional interpretation of the
quantifiers. Furthermore, he assimilates proper names to contex-
tually eliminable definite descriptions. So Quine's attitudes to quanti-
fiers and singular terms are such that (i) they blur the distinction on
which Smullyan's response depends and (ii) they assume a priority of
quantifiers over singular terms which is directly opposed to Marcus'
substitutional interpretation of quantification. The debate runs true
to form: Quine's criticisms are answered by rejection of the premises
on which they rest. Quine thinks quantifiers talk about things;
according to the substitutional interpretation, quantifiers talk about
talk about things. Quine thinks that modality is talk about talk about
things; according to essentialism, modality is talk about things.
Quine's views about quantification and necessity aren't sacrosanct,
of course - indeed, I have already expressed sorne reservations about
them. But this doesn't make the tendency for the debate between
Quine and the defenders of modal logic to degenerate into assertion
and counter-assertion less disagreeable, especially in view of the fact
that rival views about names, for instance, are apt to be defended by
appeal to 'essentialist' intuitions (e.g. Kripke 1972, Plantinga 1974).
What prospects are there for a more independent resolution?

4 Semantics for modal logics

Quine's criticisms of modal logic are to the effect, not that it is
notf'ormally feasible, but that its interp,etation in vol ves serious philoso-
phical difficulties. These criticisms should be seen in the light of the
fact that modal logic was initially developed syntactically, by the
introduction of new modal vocabulary, formation rules and axioms;
and that for a long time after its syntactic development no semantics
was available. However, after the publication, in the 194-os and 195os,
of Quine's critique, a formal semantics was developed for modal logic
(Kanger 1957a,b; Kripke 1963; Hintikka 1969); that is to say, a
formal model was devised - comparable to the trurh-table semantics
for non-modal sentence logic, for instance. And it has, understand-
ably enough, been thought by sorne that this settles the question of
the interpretability of modal logic, and shows Quine's fears to have
been unnecessary. It tums out, as we shall see, that this is very far
from obvious.
188 Philosophy o/ logics

Formal semantics - a sketch

A model structure is an ordered triple (G, K, R), where K
is a set of which G is a member and on which R is a relation; for T, R
is to be a reflexive relation, for 84, reflexive and transitive, for 85,
reflexive, transitive and symmetrical. A quantified model structure is
an ordered pair of which the fi.rst member is a model structure as
already described, and the second a function 'Y(w), assigning to each
w in Ka set of individuals. Conditions are specifi.ed for the valuation
of formulae in each member w of K; and then this set-theoretical
construction provides a definition of 'valid formula' for each of the
systems treated: a formula A is valid in the system 8 iff the valuation
of A is true for all w in K in the quantified model structure.
Thus far, a set-theoretical construction has been supplied in terms
of which validity can be defi.ned and the consistency of the modal
systems established. More is needed, however, to establish that these
systems, besides being formally feasible, have a plausible claim to
represent modal reasoning, reasoning in which the notions of necessity
and possibility play an essential role. Intuitively, Kripke suggests,
one could think of Kas a set of possible worlds w1 ••. wn, of G as the
actual world, of R as the relation of accessibility, which holds between
w1 and w2 when w1 is possible relative to w2, and of 'Y( wi) as the set o/
individuals existing in the possible world w1• On this understanding,
the formal semantics tell one that, for instance, 'LA' is true just in
case 'A' is true in all possible worlds, and 'MA' just in case 'A' is
true in sorne possible world, so that one may with sorne plausibility
takc 'L' as corresponding to 'necessarily' and 'M' to 'possibly '.

'Pure' and 'depraved' semantics

I distinguished (ch. 3 §2) four aspects relevant to one's
understanding of ordinary, non-modal sentence logic; the distinction
applies, equally, to modal logic. One has:
(i) (ii) (iii) (iv)
syntax of informal formal informal
the formal readings semantics account of
language of (i) for (i) (iii)
('pure ('depraved
semantics ') se man tics ')
In the case of the sentence calculus, the formal semantics (iii) supplies
a mathematical construction in which one of t, f is assigned to wffs of
 Modal /ogic

the calculus, and in terms of which (semantic) validity is defined and

consistency and completeness results proved. For ali the formal
semantics tells one, however, the calculus could be a notation repre-
senting electrical circuits, with 't' standing for 'on' and '/' for 'off'.
(Indeed (ch. 1 §2) interpretations of this kind, of 2- and of many-
valued calculi, are both feasible and useful.) But the claim of the
calculus to be a sentence logic, to represent arguments the validity of
which depends upon their molecular sentential structure, depends
upon one's understanding the formal semantics in such a way that
't' represents truth and '/' falsehood; it depends, in other words, on
the informal account of the formal semantics - level (iv),
The questions I want to raise now concern the status of depraved
semantics. First, do we need it? Well, I have already urged that the
pure semantics, by itself, is not sufficient; to justify the claim of a
formal system to be a modal logic (sentence logic) sorne intuitive
account of the formal semantics, connecting that set-theoretical
construction with the ideas of necessity and possibility (truth and
falsity) seems essential. In urging this view, I am, of course, opposing
a purely formalist conception, according to which logic consists of
sheer uninterpreted formalism (but contrast Curry 1951; there is also
sorne pertinent discussion, couched in less familiar terms, in Derrida
1973). Second, how seriously must we take depraved semantics? It has
been suggested that it is appropriate to regard the intuitive explanation
given of the formal semantics as a picture or metaphor, a heuristic
device to make the pure semantics a little more palatable. But I think
we need to take the intuitive account somewhat more seriously than
this, to regard the explanation of possible worlds and their possible
inhabitants as aspiring to literal truth (to be the 'sober metaphysical
truth ' in Plantinga's memorable phrase). This is clear from the
non-modal case; the explanation of 't' and '/' as 'true' and 'false',
after ali, is scarcely to be dismissed as merely metaphorical.
A third methodological question which arises at this point is, I
fear, as difficult as it is important. Perhaps the best way to introduce
it is by means of D. K. Lewis' criticisms of explanations of possible
worlds in terms of consistency. Such accounts, Lewis argues ( 1973
ch. 4) are apt to be objectionably circular. Suppose it is said that to is a
possible world only if there is a consistent description of w; if this
means, only if there is a description of w which is possibly true, it
fails to explain 'possible' in a suitably independent fashion. Lewis
claims, furthermore, that his own realist account of possible worlds is
190 Philosophy of logics

explanatory, non-circular; indeed, he proposes to use it as a test of

disputed modal principies, such as the S4 principie. Critics, however,
have urged that Lewis' account is no more successful, on this score,
than the altemative he criticises (Richards 1975, Haack 1977a).
But, as I now see it, there is a deeper question to be asked: is one
entitled to require, as Lewis does, that the intuitive account with
which one is supplied at the level of depraved semantics give a non-
circular, explanatory informal account of the formal semantics?
What is being required, I think, is that the depraved semantics be
given in terms which are, so to speak, epistemologically independent
of the readings of the modal operators, that one should be able to
tell whether there is a possible world in tohich A independently of one's
beliefs about tohether possibly A. But is this feasible? One's suspicions
may be aroused, initially, by the fact that Lewis' account, like its
rivals, seems to fail the epistemological requirement. They may be to
sorne degree confirmed by the following considerations. The syntac-
tical operators of a formal logical system are given both natural
language readings and a formal semantics, which then has, in turn,
to be 'interpreted'. At this stage, I take it, further formal interpre-
tation would only postpone the issue; one needs, as I argued above,
an informal account. But now either the informal account (I'll call it
the 'patter') will be closely related to the natural language readings
of the system's operators, or not. If not, one is likely to regard the
patter as somewhat inappropriate ( consider the suggestion that w is a
possible world just in case it is a country in the southem hemisphere,
for instance; then why should 'L ', i.e. 'true in all possible worlds ',
be read 'necessarily '?). But if the patter is close to the readings, it is
apt to violate the requirement of epistemological independence. It is
too much to ask that neither 'necessary' nor 'possible' nor any
equivalents thereof appear in the patter; explanations of meaning
must end somewhere. This isn't to say, of course, that there is no
point in giving patter which elabora tes on the original readings: one
can after all be helped to understand something by being told the
same thing another way. ·

Approaches to possible worlds

It is notable that even among those who take possible worlds
seriously there is disagreement about what kinds of thing possible
worlds are. At least three approaches can be distinguished:
 Modal logic

(i) the linguistic approach, which construes talk about

possible worlds as talk about rnaxirnally consistent sets of
se?tences (e.g. Hintikka 1969), and in which consistency
. . might be understood either syntactically or sernantically
(u) the conceptualist approach, which construes talk about
possible worlds as talk about ways in which we could
... conceiv � the world to be different (see Kripke 1972)
(111) the realist approach, which takes talle about possible
worlds at its face value, as talk of real, abstract entities
wholly independent of our language or thought (see
D. K. Lewis 1973 ch. 4).1

Appror!ches to possible indioiduals: transuiorld identity

However possible worlds are construed, sorne account needs
to be given of when individuals in different possible worlds are to
count as the sarne ; for the truth-conditions of such sentences as:
(3.x) lyl (Fx) ('there is an x which is possibly F')
or :
M(Fa) (' a is possibly F')

will go 'in the actual world there is an individual which in sorne

possible world is F' and 'in sorne possible world a is F', respectively,
and thus seem to require that one be able to identify an individual
as the sarne in different possible worlds. Consider, for example, a
sentence like 'Socrates might have been a carpenter'; its truth-
conditions would be given as 'There is a possible world in which
Socrates is a carpenter'. But what determines sohat individual in
another possible world is Socrates? Suppose, for e:xample, that in
w,. there are two possible individuals, one just like Socrates but for
being a cobbler instead of a philosopher, the other just like Socrates
but for being a carpenter instead of a philosopher; which is to be
identified with the actual Socrates? (see Chisholm 196¡).
Now the problern of transworld identity has proved notably
intractable, and there remains considerable disagreernent about how
it is best tackled. The alternatives seern to be:
( 1) Certain of the properties of an individual are regarded as
essential to its being that individual, and the criterion for
1 The different approaches are quite strongly analogoua to formaliat,
Intuitionist and logicist views, in the philoeophy oí mathematica, about
the status oí numben.
192 Philosophy o/ logics

an individual in another possible world being the same

individual is that it possess those properties. (This is the
'net' model of ch. 5 §2.)
(2) The burden of the problem is shifted off predicates and
on to names. Thus, Kripke denies that the proper names of
individuals are equivalent in sense to any set of descriptions
of their denotata, and bypasses the question, how much of
such a set of descriptions an individual in another possible
world would have to satisfy to be identical with, say,
Socrates, in this world. Proper names are rigid designators,
denoting the same individual in all possible worlds; the
correct answer to the question which individual in
another possible world is Socrates, is, simply, 'Socrates ',
that individual. (This is the 'harpoon' model.)
(3) The terms of the original difficulty are rejected. It is
denied that it is necessary, in order for it to make sense to
say that individuals in different worlds are one and the
same, that criteria be supplied by means of which one
could pick out which individual in another world is the
same as a given individual in this world. The requirement
that 'criteria of identity' be given is, according to
proponents of such an approach (e.g. Plantinga 1974
ch. 6), both impossibly and undesirably demanding
(cf. ch. 4 §2). After all, Plantinga observes, it makes
sense to say that Georg Cantor was once a precocious
baby, even though we may be entirely unable to 'Iocate '
or 'pick out' that baby, or specify what properties an
individual must have to be the infant Cantor.
(4) Others reject the terms of the original problem, not because
they consider the requirement that criteria of identity be
given if it is to be meaningful to identify individuals
across possible worlds to be too stringent, but because they
deny that the same individual can exist in different possible
worlds, so that the problem doesn't arise. Leibniz, the
originator of the possible-worlds metaphysic, thought that
each individual exists in only one possible world.
D. K. Lewis 1968 adopts this line, but elaborates it with
what he calls 'counterpart theory '. Each individual,
according to this theory, exista in just one possible world,
but has counterparts in other possible worlds (not
 Modal logic 193
necessarily in all other possible worlds, and perhaps more
than one in sorne possible worlds); and the truth of
assertions such as 'Socrates might have been a carpenter'
now depends, not upon whether there is a possible world
in which Socrates is a carpenter, but upon whether there
is a possible world in which a counterpart of Socrates is a
carpen ter.

Quine' s doubts confirmed?

Quine's doubts about modal logic antedate the development
of possible-worlds semantics; Quine, however, clearly doesn't think
that that development justífies confidence about the philosophical, as
opposed to the purely formal, feasibility of modal logic (see e.g.
Quine 1976). I shall suggest, in what follows, that the philosophical
problems raised by the metaphysics of possible worlds and their
possible inhabitants turn out to illuminate, and to sorne extent to
confirm, Quine's earlier doubts (and that reservations about the
views of modality and quantification on which Quine's original
criticisms rested can be to sorne extent bypassed).
(i) First, Quine had suggested that if modality were - as he
urges - understood as essentially a metalinguistic concept, the
standard modal systems would not be appropriate. Montague 1963
investigates in detail the restrictions which would be imposed by a
syntactical treatment of modality, concluding that 'virtually all of
modal logic, even the weak system Si, must be sacrificed ' (p. 294).
This means, furthermore, that the prospects for interpreting the
conventional modal systems by understanding possible worlds in
syntactic style, as suggested in the first approach to possible worlds,
are poor.
(ii) Second, Quine doubted whether an account could be given of
modal locutions which did not, eventually, tum out to require an
understanding of just the ideas it purported to explain. Quine's
refusal to be content with an explanation in terms of the 'intensional
circle' ( analyticity - synonymy - definition - semantic rule) of • Two
dogmas' can, I think, be seen as rather strongly analogous to an
insistence on the requirement of epistemological independence,
discussed above. The only account of possible worlds which shows
much promise of meeting this requirement is a purely syntactic
linguistic account, such that w is a possible world only if there is a
consistent descriptíon of it, where 'consistent' is understood purely in
194 Plzilosophy of logics

syntactic terms, as 'no formula of the form 'A & - A' is derivable'.
But such an account - as I observed under (i) above - leads to a
conception of necessity weaker than that formalised by the usual
modal systems. (The fact that Quine's scepticism about analyticity
doesn't extent to logical truth is pertinent here.) The rival accounts
of possible worlds seem all to be apt to violate the independence
requirement: the semantic linguistic approach because 'consistent' is
explained as 'possibly true'; the realist because (since, despite Lewis'
habit of speaking of other possible worlds as if they were like distant
places, Australia or Mars, we can't visit other possible worlds, nor
have we, to borrow a phrase of Kaplan's, a J ules-Verne-o-scope
through which to inspect them) it gives no test of which worlds are
possible; the conceptualist beca use someone who claims to be a ble to
imagine a world in which A is apt to be told he has wrongly des-
cribed what he imagines if A is inconsistent. However, 1 suspect that
the requirement of epistemological independence may be unaccept-
ably stringent; and if so, those critics of Quine (e.g. Grice and Straw-
son 1956) who conunented that, in "Two dogmas', he has asked for
the impossible and complained when it wasn't forthcoming, are, in
sorne measure, vindicated.
(iii) Third, Quine found it hopelessly obscure what it was that
quantified modal logics quantify over. Now the problems about the
transworld identity of possible individuals can reasonably be seen, 1
think, as confirming sorne of Quine's suspicions on this score. For
of the 'solutions' sketched above, (4) amounts rather to giving up the
problem than to solving it, (3) depends upon the rejection of the
requirement that one quantify only over items for which one can give
adequate identity conditions, (2) depends on a distinction between
names and descriptions which Quine rejects, and ( 1) seerns to
require a form of essentialism - essentialism not about kinds of
thing, but about indioiduals (cf. Parsons 1969). lf, that is, one
accepts Quine's views about quantification then the problem of trans-
world identity of individuals is soluble only at the cost of (individual)
essentialism. Of course, this leaves one with the options of rejecting
Quine's views about quantification or accepting essentialism, besides
the option Quine recommends, of abandoning the enterprise of
modal logic.
 Modal logic
195
5 Prospects
I should say, to make it quite clear that the following remarks
should not be taken in the spirit of a Strawsonian critique of the
inadequacies of formal languages to the subtleties of English, that
formalisation, inevitably, involves a certain amount of sirnplificatíon ;
it is a legitimate aim of modal logics to aspire to represent what is
vital to reasoning about possibility and necessity while ignoring
inessential features of modal discourse in ordinary language. How-
ever, in view of the metaphysical burden that conventional modal
logics carry, there may be profit, 1 think, in a fresh look at the infor-
mal argument that they are intended to formalise.
There are many features of modal discourse in English to which
these modal logics are quite insensitive. For example, 'possible'
takes modifiers, as: It is perfectly (quite, entirely, distinctly, remotely,
just, barely, ... ) possible that. .. ; it is a distinct (remote, real, ... )
possibility that. . . Sorne of these locutions suggest a link with
'probably' or 'likely' (e.g. 'It is just possible that I shall be late'
seems close in meaning to 'It is possible, but highly unlikely, that I
shall be late'). 'Necessary' doesn't take the same modifiers -which
may of itself occasion sorne doubts about 'Mp = - L-p' - but it
can be qualifi.ed in other ways, as: It is absolutely necessary (quite
essential ... ) to ...
These features may turn out to be logically significant. But I feel
more discomfort about logicians' disregard for sorne other features of
English1 modal discourse, which seem to have quite a strong prima
facie claim to logical relevance. In English one needs to pay attention
both (i) to the tense of a modal operator and (ii) to the tense and the
mood of the verb in the contained sentence. The conventional modal
logics are wholly insensitive to both tense and mood; yet it seems to
make a difference whether one reads:
M (3x) (Fx)
for example, as:
It is possible that there is an F (There may be an F)
oras:
It is possible that there should be an F (There might be
(have been) an F)
I It would be a pertinent question whether they are shared by other
languaaes.
196 Philosophy of logics

or whcther one reads:

M(Fa)
as:
I t is possible that a is F ( a may be F)

oras:
It is possible that a should be F (a might be (have been)
F).
Or, again, consider the difference between:
It is possible that he has hadan accident
said when the visitor is late and has not yet arrived, and:
It was possible that he should have hadan accident
when the visitor has arrived late, and the delay is known to have been
due to a traffic jam ; or, the significance of the tense of the modal
operator in:

I "bl {that the government would save the pound

t was possr e for the government to save the pound
but it failed to act in time.
I t is notorious that philosophers find themselves unable to agree
on the truth-values of formulae of modal logic, especially on those
involving íterated modal operators. This is scarcely surprising in view
of the fact that, without paying attention to considerations of tense
and mood, one has difficulty in understanding even modal staternents
with single modal operators. So I conjecture that it might prove
profitable for logicians to try to devise modal systems which build
upon an underlying apparatus in which tense and mood can be repre-
sented. Obviously, however, there would be dangers in attempting
to build on the tense Iogics presently available - for those systems
are themselves constructed by analogy with the conventional modal
logics, the inadequacies of which prompted this suggestion in the
first place. And more will be needed than merely formal ingenuity,
even in combination with sensibility to those complexities of un-
formalised modal discourse which seem to be inferentially relevant;
for instance, the interplay between modality and tense may well
raise metaphysical questions about determinism. But then, it is just
this kind of interdependence between formalism, informal argument,
 Modal logic

and philosophical argument, that makes the philosophy of logic

interesting.

6 Implication again: a postscript on 'relevance logic'

The 'paradoxes' o/ strict implication
An important motivation for the development of modal logic
was, as we saw, to introduce a stronger implication relation, and so
deflect the impact of the 'paradoxes' of material implication, And, of
course, strict implication is, indeed, stronger than material implica-
tion (since A -3 B =1 L (A-+ B)); whether it altogether meets the
need for which it was introduced, however, remains controversia]. For
strict implication has 'paradoxes' of its own, since in the regular
modal systems one has the theorems:

Lp-+ (q -3 p)
L-p-+ (P -3 q)

i.e, a necessary proposition is strictly implied by any proposition

whatever, andan impossible proposition strictly implies any proposí-
tion whatever. It isn't hard to see how this happens: for one propo-
sition strictly implies another just in case it is impossible that the
first be true and the second false; and so, in particular, if it is im-
possible that the first be true, or if it is impossible that the second be
false.
Lewis himself held that these consequences, however surprising,
must be accepted as true. (In this, he is followed by e.g. Kneale
1945-6, Popper 1947, Bennett 1954.) He thought it still appropriate
to identify strict implication as the formal counterpart of the intuitive
idea of 'implication' or 'entailment'. For entailment, he proposed, is
the converse of deducibility (A entails B iff there is a valid deduction
o{ B from A); and the 'paradoxes' are truths about entailment since
there is, he argued, a valid deduction of any necessary conclusión
from an arbitrary premise, and of an arbitrary conclusion from an
impossible prernise ; for example, in the latter case, as follows:

(1) p & -p [impossible premise]

(2) p from (1)
(3) p V q from (2)
(4) -p from (1)
(5) q from (3) and (4) (arbitrary conclusion]
198 Philosophy of logics

(This argument is, of course, valid in the standard sentence logic. You
will recall that it is just because, in standard logical systems, anything
follows from a contradiction, that consistency is of such overriding
importance.) Lewis' challenge to critics of strict implication is to say
what step of this argument, and its twin for the other 'paradox ',
could possibly be rejected.
Other writers, however, find the 'paradoxes' of strict implication
quite as shocking as Lewis thought the 'paradoxes' of material
implication. lt has been said that they are 'so utterly devoid of
rationality [as to be] a reductio ad absurdum of any view which involves
them' (Nelson 1933 p. 271), that they are 'outrageous' (Duncan-
Jones 1935 p. 78). These writers won't allow that strict implication
adequately represents the intuitive idea of entailment. Nelson, for
instance, argues that what is required for A to entail B is not just that
it be impossible for A to be true and B false, but also that there be
sorne 'connection of meanings' between A and B.1 However, the
difficulty is to specify just when there i's a 'connection of meanings'
between propositions, and to justify the rejection of whatever step(s)
of Lewis' 'proofs' of the paradoxes of strict implication are held to
violate this requirement. A further problem is that manoeuvres
adopted to block Lewis' proofs may ramify in perhaps unanticipated
and unappealing ways; sorne critics, for example, found themselves
obliged to deny the transitivity of entailment. Not surprisingly,
perhaps, it has been doubted (e.g. Suppes 1957) whether the idea of
connection of meaning, or, more generally, the idea of the releoance
of one proposition to another, is amena ble to formal treatment.
Relevance logicians, however, think otherwise.

Releoasc« log,"c
As with modal logic, there is not just one, but a whole range
of 'relevance logics '. 1 shall concentra te on R, the system of 'relevant
implication' proposed by Anderson and Belnap ( 1962a, b, 197 5) and
E, the combination of R with the modal system 84 to give a system of
'entailment' (Anderson and Belnap 1975); cf. Smiley 1959 for an
especially clear and helpful account of the alternatives.
Anderson and Belnap agree that entailment is, as Lewis held, the
converse of deducibility; they urge, however, that the standard
1
It may be of more than historical aignificance that Lewis hi.maelf made
sorne proposala along lines similar to Nelson'• in an early paper (1912)
attacking Ruaaell'a notion of implication.
 Modal logic
199
conception of deducibility, because it ignores considerations of
relevance, is defective. Relevance logicians urge that their conception
of deducibility, and not the 'official' notion of classical logicians, is
the one that one's intuitive, uncorrupted sense of what it is for an
argument to be valid requires:

A mathematician writes a paper on Banach spaces, and ...

concludes with a conjecture. As a footnote to the
conjecture, he writes: In addition to its intrinsic interest,
this conjecture has connections with other parts of
mathematícs which míght not immediately occur to the
reader. For example, if the conjecture is true, then the
first order functional calculus is complete; whereas if it is
false, then it implies that Fermat's last conjecture is
correct. . .. the editor counters ... ' ... in spite of what
most logicians say about us, the standards maintained by
this journal require that the antecedent of an "if ... then - "
statement must be releoant to the conclusion drawn.'
..• the fancy that releoance is irreleoant to validity strikes
usas ludicrous, and we therefore make an attempt to
explicate the notion o/ releoance o/ A to B. (Anderson and
Belnap 1975 pp. 17-18, final italics mine)

B is deducible from A, by their standards, only if the derivation of B

genuinely uses, and does not simply take a detour oia A. The idea of a
premise really being used, of course, stands in need of explanation.
But it's easy enough to give examples of the kind of argument that
Anderson and Belnap would describe as 'proving B under the
assumption A' but not 'proving B from the assumption A'; for
instance, in a system with 'p � p' asan axiom:

(1) q assumption
(2) p � p axiom
(3) q � (p � p) (from (1) and (2), by the deduction
theorem: if A 1- B then 1- A� B)

So what Anderson and Belnap propose is, first, to put appropriate

restrictions on deducibility, such that B is deducible from A only if
A is used in the derivation of B; these restrictions are economicaJly
summarised by Fogelin as 'the Rule of No Funny Business'. Then
they construct a system of • relevant implication' such that A rele-
200 Philosophy of logic

vantly implies B just in case B is deducible, in their sense of 'dedu-

cible', from A. The axioms for relevant implication are (I'll write
'� ', to keep a clear distinction from '--+' and ' -3 '):
1.A�A
2. (A � B) � ((C � A)� (C � B))
3. (A � (B � C)) � (B � (A � C))
4. (A � (A � B)) � (A � B)

(This is the 'implicational fragment ', i.e. the axioms involving only
implication, of R.) Entailment, however, they think, requires
necessity as well as relevance; so the connective representing entail-
ment should be restricted by imposing, besides the restrictions on
deducibility which ensure relevance, other restrictions characteristic
of strict implication as specified by 84. The upshot is these axioms,
the implicational fragment of E, for entailment (which I'll write by
an obvious analogy, '=d):
1.A�A
2.A ::j B :::j ((B � C) =! (A � C))
3. A =3 B ::::;j (((A � B) =i C) :::¡ C)
4. (A� (B � C)) � ((A :::j B) =:éj (A :::j C))
The full system E is obtained, finally, by adjoining axioms for the
other sentential connectives.
It remains to show how Anderson and Belnap meet Lewis'
challenge: where, according to them, does Lewis' 'proof' of 'q' frorn
'P & -p' go wrong? They don't, of coursc, deny Lewis' claim that
'q' is deducible from 'p & -p' by 'sorne valid form of infercnce',
in the 'official' sense of "oalid"; what they deny is that 'q' is deducible
from 'p & - p' by a valid form of inference in their sense, i.e., they
take it, the real sense of 'va/id'. Anderson and Belnap centre their
criticism on the step from 'p v q' and ' - p' to 'q '. (In classical logic,
of course, this step is justified, being an instance of what is sometimes
known as the 'disjunctive syllogism '.) Their more detailed diagnosis
of what is wrong with Lewis' argument goes as follows (Anderson and
Belnap 1975 pp. 165-6). 'Or' has two senses, the truth-functional
and the intensional; in the latter but not the former sense, the truth
of 'p V q' requires that the disjuncts be relevant to each other. Now,
they argue, the step from 'p' to 'p V q' is valid only if 'v ' is understood
truth-functionally, while the step from 'p v q' and '-p' to 'q' is
valid only if 'V' is understood intensionally. Once again, I should
 Modal logic 201

point out that by 'valid' here they mean, naturally, valid in their
�ense; they don't deny that if 'p v q' (where 'v' is truth-functional)
is true, and '-p' is true, then, necessarily, 'q' is true, but they do
deny that this is sufficient to show that the argument is valid.
Now, of course, in classical logic, since 'A� B' is equivalent to
'-A V B' the disjunctive syllogism (from '-A' and 'Av B' to
infer 'B') is equivalent to modus ponens (from 'A' and 'A� B' to
infer 'B'). And, indeed, as one might, in view of this equivalence,
expect, modus ponens for material implication fails in E.
The relevance logicians, as should, by now, be apparent, challenge
classical logic in more than one way.
(i) Most fundamentally, their challenge is to the classical concep-
tion of validity. Classical logicians have conceived of relevance as
irrelevant to the validity of an argument; irrelevance, if itis considered
at all, is apt to be relegated to the category of rhetorical defects.
Consequently, relevance logicians give a stricter sense to the notion
of one proposition's being deducible from another, and, hence, to its
converse, the notion of one proposition's entailing another.
(ii) So, relevance logicians introduce a new entailment connective,
' :=;;¡ ', to extend the classical logical apparatus.
(iii) And finally, their diagnosis of a 'fallacy of relevance' in dís-
junctive syllogism, and hence, in modus ponens for material implica-
tion, leads them not only to add a new connective to the classical
logical apparatus, but also to reject certain principies of inference
for classical connectioes.
In the case of relevance logic one has, that is, a challenge to classical
metaconcepts (strategy 6 of ch. 9 §2), an extensum of tke classical
apparatus (strategy 4) and, at the same time, a restriction of it
(strategy 5). Of these, the challenge to the classical concept of
validity is the most basic. How is one to assess this challenge? Well, it
is hard to deny that, at an informal level, irrelevance is seen as a
defect in an argument. The question is, rather, whether it is more
properly regarded as a logical defect or as a rhetorical defect. The
difference between logical and rhetorical concems could perhaps be
indicated, in a rough and ready way, by stressing rhetoric's interest in
the audience for whom an argument is intended; and on this count,
relevance - conceived as a relation between propositions - apparently
has a claim to belong to logic. An important reason why considera-
tions of relevance have tended to be disregarded by logicians, I think,
is that they don't seem, on the face of it, very readily amenable to
:202 Philosophy of logics

formal treatment. Interestingly, Schiller's comment that 'the central

doctrine of the most prevalent logic still consists in a flat denial of
Relevance' (1930 p. 75), quoted with approval by Anderson and
Belnap, is intended as an argument against the pretensions of formal
logic, notas a plea for the formalisation of relevance. If this is so, the
enterprise of relevance logic would be, so to speak, vindicated by
success (rather like Davidson's claíms for the applicability of Tarski's
methods to natural languages) - the formalisabilty of relevance
would be a reason for regarding itas a logical matter. The efforts of
the relevance logicíans have surely gone a long way towards rebutting
the suspicion that relevance is hopelessly recalcitrant to formal
treatment. There may, of course, be grounds for reservations about
the relevance logics presently available (and there is rivalry between
them, too) - sorne find Anderson and Belnap's construction of E
disagreeably ad hoc, and those, like myself, with doubts about
conventional modal logics may not be altogether happy with E's
close alliance with S-4. (But Anderson and Belnap's suggestion, p. 28,
that necessity might be understood in terms of entailment, rather
than conversely, is appealing if one thinks of logic as concerned
primarily with validity, secondarily with logical truth.) And rele-
vanee logic will inevitably be more complex than classical truth-
functional logic; so that one is entitled to ask what advantages one
might expect it to bring.
One reason why the semantic and set-theoretical paradoxes are
seen as cata.strophic is that, since, classically, anything and everything
follows from a contradiction a formal system in which a paradox is
derivable is worthless. Sorne writers have observed, though, that in
informal argument the effects of contradiction are not taken to be so
disastrously global, but are thought of as localised; and sorne of thern,
understandably enough, have hoped that a formalism in which a
contradiction doesn't entail any arbitrary w:ff might have advantages
as a ' logic of paradox '.
But the interest of relevance logic isn't to be restricted to issues in
philosophy of logic (this needn't occasion surprise; remember, after
all, the metaphysical issues to which tense logic is pertinent). I see
sorne prospects for a concept of relevance in sorne interesting issues
in epistemology, for example. Consider Quine's idea, expressed in
"Two dogmas of empiricism ', that the unit of verifi.cation/falsification
must be the whole of science; Quine argues, persuasively enough,
that a single sentence can't be subjected to empirical test in isolation,
 Modal logic :203

and concludes, in what might reasonably seem rather short order,

that it is the whole of science that faces 'the tribunal of experience '.
It wouldn't be too fanciful to suspect that the fact that, of any two
sentences, either the first materially implies the second, or else the
second materially implies the first, may give an air of inevitability to
the dilemma that either an isolated sentence, or else the whole of
science, is the unit of empirical test; and it is interesting to speculate
that a stronger notion of implication might make room for a third
possibility, that a sentence is tested along with those other sentences
that are relevant to it.
I observed, above, that E involves a restriction of classical logic:
modus ponens, for material implication, fails. Many-valued logics, to
which I now tum, also involve a restriction of the classical logical
apparatus.
 II

Many-valued logic

I Many-valued systems
Restrictions of classical logic: deviant logics
One system is a deviation of another if it shares the vocabu-
lary of the first, but has a different set of theorems/valid inferences;
a 'deviant logic' is a system which is a deviation of classical logic.
(A system may involve both an extension and a deviation of classical
logic, if it adds new vocabulary and hence new theorems/valid in-
ferences, but at the same time differs from classical logic with respect
to theorems/valid inferences involving only shared vocabulary
essentially. The system E, considered in ch. 10 §6, would be an
example.) Many-valued logics are deviant; sharing its vocabulary,
they characteristically lack certain theorems of classical logic, such as
the 'law of excluded middle', 'p V -p'. (Sorne also add new vocab-
ulary and so come in the category of extensions as well.)
The many-valued logics I shall consider in this chapter were
devised from two main kinds of motivation: the purely mathematical
interest of alternatives to the z-valued semantics of classical sentence
logic; and - of more philosophical interest - dissatisfaction with the
classical imposition of an exhaustive dichotomy into the true and the
false, and, relatedly, dissatisfaction with certain classical theorems
or inferences. The second kind of motivation is characteristic - as I
observed in ch. 9 §2 - of proposals for restrictions of classical logic.

Historical remarks
Many-valued logics have as long a history as modal logics:
Aristotle already expresses reservations about bivalence (De inter-
pretatione ix); early in the present century Hugh MacColl made both
 Many-oalued logic 205

formal and philosophical proposals. But as in the case of modal

logics, the impetus to detailed formal development carne in the wake
of the formal development of z-valued logic, specifically, of the
truth-table semantics for the logic of the Begriffsschrift and Principia
Mathematica, initiated by Post and Wittgenstein. The earliest many-
valued systems were devised by Lukasiewicz 1920 and Post 1921 (see
Rescher 1969 ch. 1 for detailed historical discussion],
However, in one respect there is an important difference between
the development of modal and the development of many-valued
logics: whereas in the former case the syntax (vocabulary, axioms,
rules) was developed first, and semantics supplied only considerably
later, in the latter case the initial development was semantic and
syntax was supplied only subsequently, axiomatisations of Luka-
siewicz's many-valued logics being devised by jaskowski 1934. Many-
valued logics began, that is, with the development of many-valued
truth-tables; it would be fair to say, however, that the question of the
interpretation of the values of these matrices is still at best only
partially answered. The problem of depraved semantics will be a
major preoccupation, here as in the previous chapter.

Formal sketch
Recall ( ch. 3 § 1) that a system is n-valued if n is the smallest
number of values which any characteristic matrix for that system has.
When speaking of 'many-valued' logics I shall restrict myself to
n-valued logics where 2 < n (so z-valued logic isn't 'many-valued';
this follows standard usage).
Though there is just one system of z-valued logic (in the broad
sense of 'system' explained in ch. 2, in which two systems are the
same if, differences of notation, primitives, and axioms/rules taken
into account, they have just the same theorems/valid inferences)
there are alternati ve systems of j-valued (etc.) logic. This is scarcely
surprising; for, once one has 3 or more values, obviously, alternative
decisions are possible about the value to be assigned to compound
formulae.
I shall offer only a sketch of sorne of the best-known many-valued
logics, and shall concentrate on formal points which bear on the
philosophical issues raised subsequently. (More detailed formal
treatment is to be found in Rosser and Turquette 1952, Ackennan
1()67, or Rescher 1()6g.) My presentation will be semantic rather than
syntactic; this is not only consonant with the history of many-valued
206 Philosophy of logics

logics, but also, I think, brings out the differences between them in
a more perspicuous way.
Lukasiewicz'a 3-valued logic (Lukasiewicz 1920, 1930) is charac-
terised by the following matrices:
B A &B
A -A A t i f
•t f t t i f
i i i i i f
f t f f f f

B A-+ B B AvB
A t i f A t i f
t t i f t t t t
i t t i i t i i
f t t t f t i f

(* indicares the designated value, i.e. the value such that wffs which
uniformly take it count as tautologies.)

Initially, Lukasiewicz had in mind that the third value, which he

read 'in determina te' or 'possible ', was to be taken by future contin-
gent statements, which he, following Aristotle, thought couldn't be
either true or false. Neither the law of excluded middle nor the law of
non-contradiction is uniformly designated on these matrices, so
neither is a theorem in 1:,3; 'p V - p' and ' - (p & - p )' take the val u e
iwhen 'p' does. Since, however, the truth-table for implication gives
'A-+ B' the value t even when antecedent and consequent take i,
the law of identity, 'p-+ p ',isa theorem.1
Kleene's 3-valued logic (Kleene 1952) differs from Lukasiewicz's
with respect to implication. Whereas l:.ukasiewicz, anxious to· save
the law of identity, sets IA-+ BI at t for IAI = IBI = i, Kleene has:
1
This 3-valued logic can be generalised. If one representa the three
valúes by the numbers (1, l, o), then Lukasiewicz's matrices fall under
the rules:
1-AI = 1-IAI
IA V BI = max {IAI, IBI}
IA & BI = min {IAI, IBI}
1 if IAI � IBI
IA ... BI = { 1-IAI +IBI if IAI > IBI
('IAI' meana 'the value of A') and these rules yield matrices for
4, S ... 11, and infinitely many, \'alues. Cf. p. 165.
 Many-oalued logic 207

X A-+B
t i f
i f
i t i i
f t t t
Unlike Lukasiewicz, Kleene didn't think of i as an intermediate
truth-value; rather, it was to represent 'undecidable', and to be
taken by mathematical sentences which, though either true or false,
are neither provable nor disprovable. Kleene's matrices are thus
constructed on the principie that where the truth or the falsity of one
component is sufficient to decide the truth orfalsity of a compound, the
compound should take that value despite having (an)other, undecid-
able component(s); otherwise, the compound is itself undecidable.
Bochvar's j-valued system (Bochvar 1939) was originally intended
as a solution to the semantic paradoxes (ch. 8 §2), and the interpreta-
tion he had in mind for the third value was 'paradoxical' or 'meaning-
less ', On the principie that a compound sentence of which a com-
ponent is paradoxical is itself paradoxical, he offered matrices in
which the third value is, so to speak, 'infectious':
B A&B
A
i f
i i i i
f f i l
B AvB B A-+B
A t i f A t i f
t t i t t t i f
i i i i i i i
f t i f f t i t
(On these matrices, of course, there will be no wffs which take t for ali
assignments to their atomic components, since i input always results
in i output. For in each table the central entries read 'i' throughout.)
So Bochvar adds an 'assertion operator', which I'll write 'T' since
it seems to mean something like 'lt is true that':
A TA
t t
i f
f f
208 Philosophy of logics

This enables him to define 'externa!' connectives thus:

-A =-TA
A &B TA&TB
Av B TA v TB
A-+B TA�TB
The matrices for the externa! connectives, consequently, always
have t or f output; and in fact just the classical, z-valued tautologies
uniforrnly take t for ali assignments to their components. (The
matrices for the externa! connectives are, so to speak, 3-valued
tables for z-valued logic, with both 'i' and '/' as kinds of falsehood.)
Ali the matrices considered so far ssenormal ( Rescher's terminology):
they resemble the familiar z-valued matrices where only classical
input is concerned - where a compound wff has only true or false com-
ponents, the 3-valued matrices give the same value that the classical
table would give. (That is, the 3-valued matrices are like the
classical ones with respect to the corner entries.) Post's many-valued

m
logics are an exception because of his 'cyclical' matrix for negation:

2 Philo1ophical motivatiom
I shan't be able to consider all the arguments which their pro-
ponents have offered in favour of many-valued logics, but will have to
restrict myself to what, 1 hope, is a reasonably representative sample.
Future contingents
Lukasiewicz introduces his j-valued logic by means of an
argument derived from Aristotle, to the effect that unless one allows
that statements about the future are not yet true or false, one will be
committed to fatalism. (Lukasiewicz's interpretation of Aristotle is
disputed, but that dispute needn't concern me here; cf. Haack 1974
ch. 4 for relevant discussion.) Lukasiewicz's argument runs along the
following lines. Suppose it is true now that I shall be in Warsaw at
noon on 21 December next year; then I can't not be in Warsaw at
noon on 21 December next year, that is to say, it is necessary that I be
in Warsaw at noon on sr December next year. Suppose on the other
hand that it is false now that I shall be in Warsaw at noon on·
 Many-oalued logic :209

21 December next year; then I can't be in Warsaw at noon on :21

December next year, that is to say, it is impossible that I be in Warsaw
at noon on 21 December next year. So, if it is either true or false, now,
that I shall be in Warsaw then, it is either necessary or impossible
that I be in Warsaw then. The only way to avoid this fatalist con-
clusion, Lukasiewicz urges, is to deny that such future tense, con-
tingent statements are either true or false in advance of the event.
Bivalence, he concludes, must be rejected.
If this argument were valid, of course, there would remain room
for disagreement about whether to take it as a proof of fatalism or a
disproof of bivalence. (Ali arguments go, in a sense, both ways; I
mean, given an argument to the effect that B follows from A, one
might either accept premise and hence conclusion, or, rejecting the
conclusion, reject the premise too.) However, since I believe that the
argument is invalid, I needn't pause over the question of whether
fatalism is or is nota tolerable conclusion. The argument is invalid, I
think, because it depends on a modal fallacy, the fallacy of arguing from:

Necessarily (if it is now true [false] that I shall be in

Warsaw at noon on 21 December next year, then I shall
[not] be in Warsaw at noon on 21 December next year)

which is, of course, true, to:

If it is now true [false] that I shall be in Warsaw at noon

on :21 December next year, then necessarily I shall [not]
be in Warsaw at noon on :21 December next year

i.e. of arguing from:

L (A--+ B)
to:
A�LB

(If it isn't obvious that this is a fallacy, consider this clearly non-
truth-preserving instance: L ((p & q) � p), so (p & q) � Lp.)
If I am right about this, fatalism doem': follow from bivalence, so,
even if fatalism is an unacceptable thesis, there is no need to reject
bivalence on that account, and Lukasiewicz has not provided a good
reason for adopting his 3-valued logic.
However, other writers have given quite di:fferent arguments in
favour of t.ukaaiewicz's logic.
210 Philosophy o/ logics

Quantum mechanics
Reichenbach argues (1944; Putnam 1957 supports his
proposal) that adoption of 3-valued logic (the one he proposes is just
like Lukasiewicz's but for the addition of further negation and
implication operators) would provide a solution to sorne problems
raised by quantum mechanics. His argument has the following
structure : if classical logic is used, quantum mechanics yields sorne
unacceptable consequences, which he calls 'causal anomalies'
(roughly, statements about quantum mechanical phenomena which
contradict classical physical laws for observable objects); these
causal anomalies can, however, be avoided, without interfering with
quantum mechanics or classical physics, by using a 3- instead of the
z-valued logic. Briefly:
classical physics & quantum mechanics & classical [ogic -+
causal anomalies
classical physics & quantum mechanics & 3-valued logic -+
no causal anomalies
Reichenbach, like Lukasiewicz, reads the third value as 'indeter-
minate"; but the kind of statement which he intends to take this
value is quite different from what Lukasiewicz had in mind. Briefly,
one of the peculiarities of quantum mechanics is this: although it is
possible to measure the position of a particle, and possible to measure
its momentum, it is impossible - it follows from the theory that it is
impossible - to measure both position and momentum at once. Bohr
and Heisenberg had suggested that statements indicating both the
position and the momentumof a particle ata given time be regarded as
meaningless or ill-forrned ; Reichenbach prefers to allow that they are
meaningful(after all, each component is, separately, quite unproblem-
atic) but neither true nor false, but indeterminate. Reichenbach's
argument raises too many questions for me to discuss here; for
example: are the diffi.culties on account of which Reichenbach wants to
modify logic genuine? and is it methodologically proper, anyway, to
modify logic to avoid diffi.culties in physics? (but cf. Haack 1974
ch. 9). However, there seems to be little doubt that, even if Reichen-
bach is right that there is a need for a non-classical logic, the particu-
lar non-classical logic he proposes doesn't meet what he takes to be
the need. The motivation for the adoption of a non-classical logic
was to avoid the causal anomalies without tampering with physics
(see Reichenbach 1944 pp. 159-60, 166); however, since Reichen-
 Many-oalued logic :211

bach intends that ali statements simultaneously indicating position

and momentum be indeterminate, he assigns the value 'indeter-
minate' not only to anomalous statements, but also to certain laws
(e.g. the principie of conservation of energy; 1944 p. 166).
It is doubtful, then, that Reichenbach has given good reason for
the adoption of Lukasiewicz's logic. (Of course, it remains possible
that developments in quantum mechanics do indeed call for the
adoption of a non-classical logic, perhaps the non-truth-functional
system developed by Birkhoff and von Neumann in 1936; cf.
Putnam 1969.)

Semantic paradoxes
Bochvar's j-valued logic was intended to supply a solution
to the semantic paradoxes: 'this sen ten ce is false' is true if false and
false if true; Bochvar's proposal is that it be assigned neither 'true'
nor 'false', but a third value, 'paradoxical' or 'meaningless '. I have
already argued (ch. 8 §2) that this kind of approach to the paradoxes
is apt to go from the frying pan - the Liar paradox - to the fire - the
Strengthened Liar (' this senten ce is either false or paradoxical ',
true if false or paradoxical, false or paradoxical if true). As with Luka-
siewicz's logic, however, other reasons than those given by its origi-
nator have also been suggested in favour of a 3-valued logic like
Bochvar's.

Meaninglessness
Halldén's (1949) 'nonsense logic', for instance, has matrices
like those for Bochvar's intemal connectives, in which the third value
(' meaningless ') infects any compound to any component of which it
is assigned. But this, again, doesn't supply any very impressive
reason in favour of Bochvar's logic. For, as I argued in ch. 9 §4, the
whole enterprise of a 'logic of meaninglessness' seems to me funda-
mentally mísconceíved.
I have already commented on the curious, 'infectious' character of
Bochvar's third value, pointing out that it has the perhaps rather
dismaying consequence that there are no wffs using only the intemal
connectives which take the val u e 'true' for ali assignments to their
components, One proposal, however, gives this an interesting
rationale.
ZIZ Philosophy of logics

Sense toithout denotation

Recall (ch. 5 §2) that Frege held that the denotation/sense
of a compound expression depended upon the denotation/sense of
its components; and, consequently, that a sentence containing a
singular term which has no denotation itself lacks truth-value, and a
cornpound sentence of which a component is truth-valueless is itself
truth-valueless. Frege himself preferred, as we saw, to ensure that his
formal language permit no denotationless terms; if such terms were
allowed, however, a non-classical logic would be needed to handle
them in the way Frege's theory requires. Smiley suggests (1960)
that a 3-valued logic like Bochvar's would be the appropriate non-
classical system. The assignment of the third value to a wff indicates,
here, not that it has an intermediate truth-value, but that it has no
truth-value at ali. Now the fact that the matrices for the interna!
connectives assign no truth-value to a compound wff if any component
lacks truth-value corresponds to the Fregean principie that a com-
pound expression lacks denotation if any component lacks denota-
tion. And with the help of the assertion operator the Fregean con-
ception of presupposition ('A' presupposes 'B' if 'A' is neither true
nor false unless 'B' is true) can be defi.ned. Thus, I think, this
proposal succeeds admirably in representing the formal system that
would result from the adoption of Frege's theory of sense and denota-
tion ( contrast W oodruff' s formalisation ( 1 970 ), which doesn 't satisfy
Frege's truth-valueless input/truth-valueless output principie). Of
course, whether one takes it to supply, at the same time, an argument
for the adoption of Bochvar's logic, depends on whether one acccpts
Frege's account of denotationless expressions.

Undecidable sentences
Kleene's j-valued logic is intended, as we saw, to accommo-
date undecidable mathematical statements; the third value repre-
sents 'undecidable ', and the assignment of that val u e to a wff is not
intended to indicate that it is neither true nor false, only that one
can't tell which. Indeed, it is just because Kleene takes it that
undecidable wffs are true or false that he adopts the principie that a
compound wff with an undecidable component should be decidable
if the values of the other components suffice to ensure that the whole
is either true or false (e.g. if IPI = i and lql = t, IP V q¡ = t). So
while the philosophical motivation for Kleene's 3-valued logic seems
unexceptionable enough, what he proposes seems to be less radical,
 Many-oalued logic :213
less of a challenge to classical a-valued logic, than initially appears
(cf. Kripke's insistence (1975) that his use of Kleene's valuation
rules poses no threat to classical logic; see ch. 8 §:2). These reflections
raise sorne interesting questions about how the adoption of a many-
valued logic may be expected to affect the theory of truth.

3 Many-valued logics and truth-values

Not surprisingly, it has sometimes been supposed that the
use of a many-valued logic would inevitably involve a claim to the
effect that there are more than two truth-values: a claim which has -
again, perhaps not surprisingly - sometimes been a major source of
resistance to many-valued logics. But in fact, I think, it is clear that
a many-valued logic needn't require the admission of one or more
extra truth-values over and above 'true' and 'false', and, indeed,
that it needn't even require the rejection of bivalence.
The use made by Smiley of Bochvar's 3-valued logic illustrates
the first point. Assignment of the third value to a wff indicates that it
has no truth-value, not that it has a non-standard, third truth-value,
(If you are tempted to think of 'neither true nor false' as a third
truth-value, on a par with 'true' and 'false', McCall's observation
( 1970) that no-one supposes that 'either true or false' is a third
truth-value, may help to stiffen your resistance.)
Sometimes, again, intermediate values are understood, not as new
truth-values, but as, so to speak, epistemological variants on 'true'
and 'false'. Prior suggests interpretations of the values of a 4-valued
logic as:
1 = true and purely mathematical (or, true and known to
be true)
2 = true but not purely mathematical (or, true but not
known to be true)
3 = false but not purely mathematical (or, false but not
known to be false)
4 = false and purely mathematical (or, false and known
to be false)
And these examples serve to verify, also, my stronger claim, that the
use of a many-valued system needn't even require denial of bivalence.
For this interpretation entails that every wff is either true or else
false.
Another example where the threat to bivalence turns out to be
214 Philosophy of /ogics

only apparent is this: Michalski et al. 1976 propose a 12-valued logic

which is said to be useful for devising computer programmes to
handle material about plant diseases. One might be excused for
feeling a certain bewilderment at this point: however is one to make
sense of the ten extra truth-values? On closer examination, however,
it turns out that what is going on is a good deal less radical, and a
good deal less puzzling, than at first appears. The idea is (I simplify,
but I hope not misleadingly) that instead of classifying information
about the appearance of symptoms in the obvious way, as, say:
Red spots first appear in January - false
Red spots first appear in February - true
Red spots first appear in March - false
etc.
one can classify it, much more economically, thus:
Red spots appear - value 2
The 12 values amount, in effect, to 'true-in-January', 'true-in-
February' ... etc. Notice, here, that what the two classical truth-
values were assigned to ('Red spots first appear in January', etc.)
and what the 12 non-classical values are assigned to (' Red spots
appear') are different.
This leads to an important, more general, point: that what looks
on the face of it like the assignment of a non-standard value to a
standard item may turn out to be best explicable as the assignmcnt
of a standard value to a non-standard item. This may suggest what is
correct in the recurrent criticism (e.g. Lewy 1946, Kneale and
Kneale 1962 pp. 51ff.) that the proponents of many-valued logics are
simply confused about truth-bearers.
The interpretation suggested for Post's many-valued systems
supplies an interesting illustration of this point. The idea is, briefly,
to take the 'sen ten ce letters' to stand for sequences of sentences, and
to take assignments of values to these sequences to depend upon the
proportion of their true to their false members (more accurately: in
n-valued logic, Pis to stand for an (n-1)-tuple, (pi,p2 ••• p,._1), of
regular, z-valued sentences, and P is to take the value i when just
i- 1 of its elements are false). This suggests that Post's logics might
be regarded as giving a formal analogue of the intuitive idea of partial
truth; as, a sentence is partly true if it is complex, and part of it is
true ( cf. p. 169 above; and Haack 1974 pp, 62-4 for more discussion).
 Many-valued logic :ns
What I have been arguing, thus far, is that many-valued logics
needn't call for the admission of intermediate truth-values, nor even
for the rejection of bivalence. This isn't to say, of course, that they
neoer pose this kind of challenge to the classical view of truth. For
example, the use of Bochvar's matrices to represent a Fregean
account of denotationless expressions certainly calls for a denial
of bivalence; assignment of the third value represents precisely the
idea that the formula is neither true nor false. (Recall ( ch. 7 § 5) that
the classical, Tarskian account of truth is bivalent, and indeed that
the (T) schema threatens to rule out non-bivalent truth-theories.]
It will not, perhaps, have escaped notice that, of the philosophical
arguments for the adoption of many-valued logics discussed above,
the most persuasive were those which supported an understanding
of the intermediate values as epistemic variants of classical truth-
values (Kleene), as assignment of classical truth-values to non-
classical items (Post), or as lack of classical truth-value (Smiley).
This may be coincidence; but those who are suspicious of the intel-
ligibility of the idea of intermediate truth-values might find in it sorne
confirmation of their suspicions.

4 Non-trutb-functional deviant logics

Many-valued logics, of course, like classical logic, are truth-
functional; the value assigned to a compound wff depends solely
upon the values assigned to its components. (Modal Iogics, by con-
trast, are not truth-functíonal; the truth-value of a modal formula
does not depend solely upon the truth-value of its components, and
the standard modal logics have no finite characteristic matrices.)
Logicians' preference for truth-functional connectives ( cf. ch. 3 § 2)
is understandable, since the truth-tables supply a simple decision-
procedure for 2- and many-valued logics.
When, however, one reflects upon the motivation for Kleene's
matrices, the assumption of truth-functionality may, I think, be
called into question. Remember that Kleene's argument why IP V qj
should be t if IPI = i and (q( = t is that the truth of one disjunct is
sufficient to determine the truth of the whole disjunction, regardless
of the value of the other disjunct; i.e., that 'p V q' will be true if 'q'
is, whether 'p' is true or false. However, Kleene's matrices assign i to
'p V q' when IPI = (q( = i; and so, in particular, they assign i to
'p V -p• when IPI = 1-PI = i. But, one may observe, whereas
'p V q' can't be relied on to be true regardless of whether 'p' and 's'
216 Philosophy of logics

are true or false, 'p v -p' will be true sohether 'p' is true ar false. And
this suggests that Kleene's principles mightjustify a different assign-
ment to 'p v -p' than to 'p v q', when both disjuncts have i. But
this, of course, would require a non-truth-functional logic.

Supervaluations
Van Fraassen's non-truth-functional 'supervaluational Iang-
uages' (see 1966, 1968, 19l>9) seem to be more consonant than
Kleene's own j-valued matrices with the principles on which
Kleene argues for his assignments. The idea, briefly, is this: a super-
valuation assigns to a compound wff sorne component(s) of which
lack(s) truth-value that value which all classical valuations would
assign, if there is a unique such value, and otherwise no value. Since
all classical valuations - i.e. both those that assign 'true' and those
which assign 'false' to 'p' -would assign 'true' to 'p V -p', the
supervaluation also gives 'true' to 'p v -p '. However, sin ce the
classical valuation which assigns 'false' to 'p' and to 'q' assigns
'false' to 'p v q', while all other classical valuations assign 'true' to
'p v q', there is no unique value assigned by all classical valuations to
'p V q', and the supervaluation gives it no value. It isn't hard to see
that supervaluations will give 'true' to all classical tautologies and
'false' to all classical contradictions, but no val u e to contingent
formulae; however, though van Fraassen's systems will thus have
j ust the same tautologies as classical logic, they di:ffer from classical
logic with respect to the inferences which are allowed as valid - for
instance, 'disjunctive dilemma' (if A � C and B � C, then A v B � C)
fails - which is why they count as deviant.

Intuitionist logic
Another non-truth-functional deviant logic which is of
substantial philosophical and formal interest, is Heyting's Intuitionist
logic.
The Intuitionists claim [see e.g. Brouwer 1952, Heyting 1966) that
classical logic is, in certain respects, incorrect. It is important, how-
ever, that their disagreement goes deeper than their rejection of
certain classical laws. F or, first, the In tuitionist view of the scope and
character of logic is quite distinctive; the Intuitionists think oflogic as
secondary to mathematics, as a collection of principies which are
discovered, a posteriori, to govem mathematical reasoning. This
obviously challenges the 'classical' conception of logic as the study of
 Many-oalued logic 217

principies applicable to all reasoning regardless of subject-matter,

as the most fundamental and general of theories, to which even
mathematics is secondary. However, this di:fferent conception of
logic wouldn't, of itself, explain the Intuitionists' challenge to certain
Iaws of classical logic, if it weren't for the fact that the Intuitionists
also have a distinctive view of the nature of mathematics; for classical
logical laws are supposed, of course, to govern all reasoning, including
classical mathematical reasoning.
According to the Intuitionists, mathematics is essentially a mental
activity, and numbers are mental entities (cf. what I called in ch. 11
§4 the 'conceptualist' view of the character of possible worlds);
relatedly, what it means to say that there is a number with such-
and-such a property is that such a number is constructible. The
Intuitionists' distinctively psychologistic, constructivist view of
mathematics leads them to the conclusion that sorne parts of classical
mathematics - those which deal with completed, infinite totalities,
for example - are unacceptable. And from this restriction of rnathe-
matics there follows a restriction of logic; sorne principies of classical
logic are not, the Intuitionists urge, universally valid. For example,
Brouwer argues, there are counter-examples to the law of excluded
middle. Suppose that it is possible neither to construct a number with
property F, nor to prove that there can be no such number. Then, by
Intuitionist standards, it isn't true that either there is a number which
is F, or there isn't.
Notice, here, an interesting contrast with Kleene's attitude.
Kleene does not take the fact that sorne mathematical statement is
undecidable in principie as any reason to deny that it is, neverthe-
less, either true or false. The Intuitionists, by contrast, regard the
idea that there might be a number which couldn't be constructed as
a piece of hopelessly confused metaphysics (see Heyting 1966 p. 4).
The contrast may usefully serve to draw attention to the fact that the
distinction between what, in the previous section, I called epistnno-
logical values versus genuine truth-values may not be altogether
neutral but may presuppose sorne disputable assumptions about the
relations between metaphysics and epistemology.
Because he regarded mathematics as essentially mental, and hence
thought of mathematical and, a fortiori, logical formalism as rela-
tively unimportant, Brouwer didn't give a formal system of the
logical principies which are Intuitionistically valid. However,
Intuitionist logic was formalised by Heyting, who gives these axioma.
218 Philosophy of logics

r. p--+ (P & p)
2. (P & q)--+ (q & P)
3· (P --+ q)--+ ((p & r)--+ (q & r))
4· ({p--+ q) & (q--+ r))--+ ( p --+ r)
5· q--+ (P--+ q)
6. (p & (P --+ q)) --+ q
7· P --+ (P v q)
8. (P v q)--+ (q v P)
9. ((p--+ r) & (q--+ r))--+ ((p V q) --+ r)
ro. l P --+ (p --+ q)
rr. ((p--+ q) & (p--+ l q))--+ lP

(' l' is the usual symbol for Intuitionist negation.) Notice that this
list contains axioms governing each connective ('&', 'v', '--+', '1');
in Intuitionist logic the connectives are not interdefinable, so all
must be taken as primitive. This is related, of course, to the fact that
Intuitionist logic has no finite characteristic matrix ( cf. cornrnents
on interdefinability of connectives in ch. 3 § r). Heyting's logic lacks
sorne classical theorems; notably, neither 'p V lP', nor '11P--+ p', are
theorems. However, the double negations of ali classical theorems
are valid in lntuitionist logic.
Heyting's is not the only, although it is the best entrenched,
system of Intuitionist logic: in fact, Johansson's logic (1936), which
lacks the tenth axiom, has, arguably, a better claim properly to
represent the logical principies which are acceptable by Intuitionist
standards. But Heyting's logic has sorne unexpected affinities,
affinities which raise questions about the distinction between deviant
and extended logic, with modal logic, which is why it will occupy my
attention for the rest of this section.
There is little doubt that the lntuitionists see themselves as
challenging the correctness of certain theorems of classical logic. This
makes it appropriate that they should propose a restriction of classical
logic, in which the disputed theorems fail. However, Heyting's
calculus, though on the face of it a deviation of classical logic, may
also be interpreted as an extension of classical logic. lf one takes
Intuitionist negation and conjunction as primitive, and defines dis-
junction (P v q = df l ( l p & l q)) implication and equivalen ce in the
usual, classical way, in terms of them, then all classical theorems can
be derived in the Heyting logic; in addition, of course, ali the theo-
rems in Intuitionist disjunction, implication, and equivalence -
 Many-oalued logic

which are not definable in terms of Intuitionist negation and

conjunction - are also derivable. And this makes Intuitionist
logic look less like a restriction, and more like an extension, of
classical logic. (But not ali classical inferences are preserved by the
proposed translation, not e.g. MPP, for since llP � p under the
translation, the validity of MPP would imply llP 1- p.) It is also
possible to interpret the Heyting calculus as a modal logic; if:
m (A) = LA (for atomic sentences)
m(lA) = L-m(A)
m (A v B) = m (A) v m (B)
m (A & B) = m (A) & m (B)
m (A � B) = L (m (A) � m (B))
('m (A)' means 'the translation of A'; the connectives on the left-
hand side are the Intuitionist, those on the right the classical, con-
nectives] it is provable that a wff is valid in the Heyting calculus iff its
translation is valid in 84 (McKinsey and Tarski 1948; cf. Fitting
1969). Of these two 'translations' of Heyting's logic, the latter seems
somewhat more natural than the former; for Brouwer and Heyting
do sometimes read 'l ' as 'it is impossible that ... ', as, for example,
when they read '(3.x) Fx V l (3.x) Fx' as 'Either an F exists, or a
contradiction is derivable from the assumption that an F exists '.
But what, exactly, is the significance of the availability of these
translations ?
It is natural to expect a correlation between, on the one hand,
proposals to restrict classical logic, and the idea that it is in sorne way
mistaken, and, on the other, proposals to extend classical logic, and
the idea that it is in sorne way inadequate. The thought is that a
restricted ( deviant) logic excludes sorne theorems/inferences expres-
sible wholly in classical vocabulary, and thus involves the denial that
sorne classical theorems/inferences are really valid. But now one can
see that the question, whether a non-classical logic really has 'the
same vocabulary' as classical logic, is not so simple as it (perhaps)
seemed. The Intuitionists regard themselves as critics, and the Heyt-
ing calculus as a restriction, of classical logic; the possibility of
representing Heyting's calculus as an extension of classical logic
raises the question whether the Intuitionist connectives differ in
meaning from their classical 'analogues '. For myself, I am inclined
to think that the fact that there is more than one way to represent
Heyting's as an extended logic would justify caution about the idea
220 Philosophy of logics

that the Intuitionists' critique of classical logic can be wholly ex-

plained away as the result of meaning-variance. But the general
issue, about the relevance of considerations of meaning to the dis-
tinction between deviant and extended logic, will prove important
to the argument of the next chapter.
 12

Sorne metaphysical
and epistemological questions
about logic

Metaphysical questions
The purpose of this chapter is to tackle sorne of the questions
about the status of logic which are raised by the existence of a
plurality of logical systems - a plurality I have been exploring in
previous chapters. Sorne of these questions are metaphysical: e.g. is
there just one correct logical system, or could there be several which
are equally correct? and what could 'correct' mean in this context?
Others are epistemological: e.g. how does one recognise a truth of
logic? could one be mistaken in what one takes to be such truths?
1'11 start with the metaphysical questions, since the answers to the
epistemological questions are apt to depend, to sorne extent, upon the
answers to them.

Monism, pluralism, instrumentalism

lt will be useful to start by distinguishing, in a rough and
ready way, three broad kinds of response to the question whether
there is a uniquely correct logical system:
monism: there is just one correct system of logic
pluralism: there is more than one correct system of
logic
mstrumentalism: there is no 'correct' logic; the notion of
correctness is inappropriate

Obviously this needs elaboration and refinement. First, sorne

comments about the conception of correctness which both monism
and pluralism require: this conception depends upon a distinctioo
between system-relative and extra-systematic validity/logical truth;
222 Philosophy of logics

roughly, a logical system is correct if the formal arguments which are

valid in that system correspond to informal arguments which are
valid in the extra-systernatic sense, and the wffs which are logically
true in the system correspond to statements which are logically true
in the extra-systematic sense. The monist holds that there is a
unique logical system which is correct in this sense, the pluralist
that there are several.
Now the significance of the distinction between extensions of
classical logic and deviations from it can be fully appreciated. Prima
facie, at least, the modal logician, for instance, seems to be
claiming that there are valid arguments/logical truths which cannot be
represented within the vocabulary of, and so are not valid arguments/
logical truths of, classical logic; so that, although classical logic is
correct as far as it goes, it doesn't go far enough. The proponent of a
3-valued logic, by contrast, seems to claim that there are valid argu-
ments/logical truths of classical logic the informal analogues of which
aren't valid/logically true, so that classical logic is actual/y incorrect.
(This explains in a more precise way the idea, first adumbrated in
ch. 9 §3, that deviant logics pose a more serious challenge than
extended logics to classical logic.)
lf deviant logics rival classical logic, whereas extended logics
supplement it, this would indicate that a monist attitude would be
suitable to the former (one is obliged to choose between the classical
and deviant systems) and a pluralist attitude to thc latter (one could
accept the classical and an extended logic as both correct). Alterna-
tively, one might regard classical logic and extensions of it (or,
again, of course, sorne deviant logic and extensions of that)
as together constituting • the correct logic '. The point is that the
difference between a pluralism which admits classical logic and its
extensions ( or a deviant logic and its extensions) as both correct
systems of logic, and a monism which admits both classical logic and
its extensions ( or a deviant logic and its extensions) as both fragments
of the correct system of logic, is only verbal.
So I shall concentrate henceforth upon the choice between classical
and deviant logics (similar questions arise about the choice between
one deviant logic and another, and perhaps between one modal logic
and another - cf. p. 178; but I shan't discuss them here), where the
issue between monism and pluralism is substantial. The monist
sees classical and deviant logics as making rival claims about what
formalism correctly represents extra-systernatically valid arguments/
 Metaphysical and eputemological questums 223
logical truths; the pluralist, in brief, clairns that the apparent rivalry
is, in one way or another, mere/y apparent. In fact, there are several
versions of pluralisrn, different ways of disrnissing the apparent
rivalry. Sorne pluralists share with the rnonist the assurnption that
logic should be applicable to reasoning on any subject-rnatter: others,
however, urge that different logics rnay apply to reasoning on differ-
ent subject-matters. So one rnay distinguish between global and
local versions of pluralisrn1; I shall consider the local version first.
According to local pluralism, different logical systems are applic-
able to (i.e. correct with respect to) different areas of discourse;
perhaps classical logic to rnacroscopic phenornena, and 'quanturn
logic' (p. 210) to rnicroscopic phenornen.a, for example, rather as
different physical theories rnay hold for macroscopic than for micro-
scopic phenornena. The local pluralist relativises the extra-systematic
ideas of validity and logical truth, and hence the idea of the correct-
ness of a logical systern, to a specific area of discourse; an argurnent
isn't valid, period, but valid-in-d.
The global pluralist, by contrast, shares the rnonist's assurnption
that logical principies should apply irrespective of subject-matter.
However, while the rnonist takes it that the classical and the deviant
logician disagree about the validity/logical truth in the same sense,
of one and the same argurnent/staternent, the global pluralist denies
either that the classical and deviant logician are really using 'valid '/
'logically true' in the sarne sense, or else that they are really dis-
agreeing about one and the sarne argurnent/staternent. The forrner
idea obviously relates to what 1 called, in ch. 9 § 2, the 'challenge to
classical rnetaconcepts'; the latter, to sorne ideas discussed in
ch. 3 §2 about the rneanings of the connectives.
Roughly, the thought in the second version of global pluralisrn
is this: typographically identical wffs/argurnents in classical and
deviant logics don't have the same rneaning, and hence can't both
represent the very same informal staternents/argurnents. One
argurnent for this view is that the rneaning of the logical constants
depends wholly upon the axiorns/rules of the systern in which they
occur; consequently, when a certain wff, 'p V -p', say, is logically
true in one systern and not in another, then those wffs, though typo-

I The contraat between Boole's idea of logic as a calculus, and Leibnis's

of logic as a universal language, discusaed in van Heijenhoort 1g67b,
may have affinities with the distinction upon which I am preeently
relying, berween local and global approaches to logic.
224 Philosophy o/ logias

graphically the same, have different meanings in the different systems,

the meaning-oariance thesis.1 So what the wfI 'p V - p' says, in
classical Iogic, is Iogically true, but what the very same wfI says, in
3-valued logic (where 'v' and' - ', or perhaps 'p ', have non-classical
meanings) is not Iogically true; so both the classical and the 3-valued
logic are correct. From this point of view the deviant logician seems,
very much Iike, e.g., the modal Iogician, not to be challenging the old,
but as offering new valid arguments/logical truths - he differs from
the modal Iogician only in his disagreeably confusing habit of using
old symbols for his new conception ( cf. the discussion of the trans-
Iatability of Intuitionist into modal Iogic in ch. 11 §4).
The instrumentalist position results from a rejection of the idea of
the 'correctness' of a logical system, an idea accepted by both
monists and pluralists. On the ínstrumentalist view, there is no sense
in speaking of a logical system' s being 'correct' or "incorrect ', though
it might be conceded that it is appropriate to speak of one system's
being more fruitful, useful, convenient ... etc. than another (perhaps:
for certain purposes). The rejection of the concept of correctness is
apt to be based on a rejection of the extra-systematic ideas of logical
truth and validity which that conception requires; if only the con-
cepts of logical truth-in-I: and validity-in-L are intelligible, the
question, whether the wffs/arguments which are logically true/valid-
in-L correspond to statements/arguments which are extra-syste-
matically logically true/valid, simply cannot arise. An instrumentalist
will only allow the 'interna}' question, whether a Jogical system
is sound, whether, that is, ali and only the theorems/syntacticaUy
valid arguments o/ the system are logically true/valid in the
system.
Another version of instrumentalism seems to derive from a refusal
to apply any idea of truth, even a system-relative idea, to logic. Logic,
it is argued, is not to be thought of as a set of statements, as a theory
to be assessed as true or false; rather, it is to be thought of as a set of
rules or procedures, to which the concepts of truth and falsity simply
1
1 deliberately choose this expression to recall Feyerabend's thesis that
the meanings of theoretical terms in science depend upon the theories
in which they occur, so that there is failure of rivalry between
alternative, apparently competing, scientific theories (cf. Feyerabend
1963; and see Haack 1974 pp. 11-14 for further exploration of the
analogy).
 Metaphysical and epistemological questions

Can a logical system be correct or incorrect ?

�
No Yes
/
lNSTRUMiNTALISM
Is there one correct logic?

/�
Yes No
1 1
MONISM PLURALISM
1
global or local?

global pluralism
/�
local pluralism
Fig. 7

don't apply.1 However, the question of correctness would still arise,

on this rule-oriented view, with respect to validity (do the arguments
valid-ín-L correspond to informal arguments which are extra-
systematically valid?) unless the extra-systematic conception of
validity is also rejected. So the initial version of the instrumentalist
position, resting upon a rejection of the extra-systematic ideas in
correspondence to which monists and pluralists suppose correctness
to consist, is the more fundamental. These altematives can be con-
veniently summarised as in fig. 7.
I have aimed rather at mapping out the altematives in as syste-
matic a fashion as I can, than at listing positions held by specific
writers. But in fact it is possible to find examples of writers holding
each of the positions I have identified. Quine seems to take for
granted something like what l've called the monist position when, in
the second half of 'Two dogmas' (1951), he considera the (episte-
mological) question of the revisability of logic; in ch. 6 of Philosophy
o/ Logic ( 1970 ), however, he seems to opt for something like meaning-
1 Analogoualy, the idea that the 'laws' of physica aren't to be thou¡ht of
u true-ar-false statements, but as principies of ínferenee, ia often
taken to be characteristic of an 'i.natrumentalist' phílOIOphy of acience;
see e.g, Toulmin 1953.
226 Philosophy of logics

variance pluralism, using rather complex arguments from his theory

of translation to support the claim that there is change of meaning
sufficient to preclude rivalry; sorne quantum logicians, most clearly
Destouches-Février 1951, but also, probably, Putnam 1969, support
a local pluralism; Rescher's 'relativism' (1969 ch. 3) seems to be
quite close to what I've called instrumentalism, but in 1977 he seems
to try to combine a rule-oriented instrumentalism with the admission
of an extra-systematic notion of validity.

The issues summarised

At any rate, it is clearer, now, what the major issues are:
Does it make sense to speak of a logical system as correct or
incorrect? Are there extra-systematic conceptions of
validity/logi'cal truth by means of whi'ch to characterise what
it is for a logic to be correct?
The instrumentalist position is characterised by a negative answer
to these questions; monists and pluralists answer them positively. (It
should also be clear now why I observed that sorne epistemological
questions depend on the answers to metaphysical questions; unless
there can be a correct logic, the question, how we tell whether a logic
is correct, doesn't arise.)
Must a logical system aspire to global application, i.e. to
represent reasoning irrespectioe of subject-matter, or may a
logic be local/y correct, i.e. correct within a limited area of
discourse?
The local pluralist position is distinguished by the choice of the second
of these options.
Do deoiant logics rival classical logi"c?
The monist answers this question affirmatively, the global
pluralist negatively. Toe issues ali concem the relation between formal
and informal argument, system-relative and extra-systematic validity.
Thus the monist picture may be represented as in fig. 8 (p. 227).
(i) aspires to represent (iii) in such a way that (ii) and (iv) correspond
in 'the correct logic'. The instrumentalist rejects (iv) altogether; the
local pluralist relativises (iv) to specific arcas of discourse; the global
pluralist either denies that the formal arguments of a deviant logic
represent the same informal arguments as those of classical logic,
 Metaphysical and epistemological questions 2a7
(i) Formal argurnent (wff) which is (ii) valid-(logically true-] ín-L
1 1
� represents '+' corresponding to (iii)', being
(iii) informal argument (statement) which is (iv) extra-systematically valid
(logically true)

Fig. 8

i.e, he fragments the relation between (i) and (iii), or else he denies
that validity in the deviant logic is intended to correspond to extra-
systematic validity in the same sense as that to which validity in
classical logic is intended to correspond, i.e. he fragments the relation
between (ii) and (iv).

Comments
It is often enough the case in philosophy that asking the right
questions is half the battle. However, the other half is not to be
shirked; and I shall now offer sorne comments on what I take to be
the major issues. But the questions that have been raised here are, I
think, enormously difficult, and there is a serious problem about
finding a starting point for the argument that begs no pertinent
questions; so I should stress that the next paragraphs are tentative
as well as, no doubt, inconclusive.
I indicated early on ( ch. 2 § 2) that I do take there to be an extra-
systematic idea of validity to which formal logical systems aim to give
precise expression. It is clear enough from the history of formal logic
( consider Aristotle, for instance, or Frege) that the motivation for the
construction of formal systems has been, on the basis of an initial
conception of sorne arguments as good and others as bad, to sort out
logical from other, e.g. rhetorical, features of good arguments, and to
give rules which would permit only the logically good arguments and
exclude the bad. This, therefore, inclines me to answer the first
questions affirmatively, and so to reject the instrumentalist position.
This inclination is strengthened, furthermore, by sorne persistent
doubts about whether an instrumentalist can have anything sensible
to say about how one is to choose between logical systems. The
instrumentalist norrnally concedes that, at least for certain purposes,
one logical system may be judged better than another, perhaps as more
convenient, more fruitful, more appropriate, yielding the desired
inferences ... But no matter how convenient or fruitful it might be if
one could in fer 'A and B' from 'A ', this would, or so it seems to me,
228 Philosophy of logics

be no reason to prefer a system which represented that inference as

valid. I'rn aware, of course, that in making such comments as this I'm
in sorne danger of assuming an extra-systematic conception of
validity, and criticising the instrumentalist for failing to take account
of it, when, of course, he claims that there is no such conception
(rather as Russell and Moore assumed the correctness of a corres-
pondence theory of truth, and criticised the pragmatist theory on
that basis). Nevertheless, I think the fact that Rescher, in presenting
an instrumentalist position, in the end allows that the requirement
that arguments be truth-preserving is overriding may justifiably
confirm my suspicions.
I have also indicated (ch. 1 §2) that I take it to be characteristic of
logic to aspire to present principies which apply to reasoning on no
matter what subject; to be global in scope. I allowed that the notion of
a principle's applying to reasoning irrespective of subject-matter
wasn't perfectly clear or precise - it shares the vagueness of the
extra-systematic conception of validity as holding in virtue of form
rather than content. Still, though I think there is room for doubt
about whether 'believes ', say, or 'prefers ', may legitimately be
counted as form rather than content, I nevertheless feel pretty confi-
dent that principies that held for reasoning about biological subject-
matter, but not for reasoning about physics, for example, wouldn't
be logical (but, I suppose, biological) principies. Consequently, I
should answer the second question affirmatively, and am disinclined
to accept a local pluralism. If, for example, it turns out, as Birkhotf
and von Neumann claimed (1936)1 that where quantum phenomena
are concerned, 'A and (B or C) itf (A and B) or (A and C)' isn't
invariably true, then classical logic, in which the distributive laws are
theorems, isn't correct. (I'm entirely willing to concede that it might
be that while classical principies are, strictly, incorrect, they hold for
all ordinary reasoning about macroscopic phenornena, so that it would
be as reasonable to use classical logic for purposes of ordinary reason-
ing as to use Euclidean geometry for surveying purposes, despite the
fact that Euclidean geometry isn't, strictly, true of our space. However,
I now doubt whether this concession will appease the local pluralist.)
This leaves the options of monism, on the one hand, and sorne
form of global pluralism, on the other. But at this stage, 1 think, the
character of the argument changes; I mean, that while the first two
questions concern the nature and aspirations of Iogic, and can be
answered generally, the last concerns the relations between classical
 Metaphysical and epistemological questions :2:29

and deviant logics, and so it may have no general answer, but perhaps
different answers for different deviant logics, It may be, that is, that
some deviant logicians are using different metaconcepts from the
classical logician, and others the same; or that the meaning-variance
thesis is true of some deviant logics but not of others; or, indeed, both,
A more piecemeal approach is appropriate from here on. Of course,
though, monism and pluralism are asymmetric in a way which is
relevant; even one instance of a deviant logic which could be correctas
well as classical logic would tip the balance to pluralism.
Now, although I urged that there is an extra-systematic idea of
validity which formal systems of logic aspire to represent, I also ob-
served (pp. 14-15) that that idea is by no means fully precise, and that
it may be refined and perhaps modífied as logic develops. The rele-
vanee logician (ch. 10 §6) rejects the principie, from 'A' and 'A� B'
to infer 'B'; modus ponens, he urges, is invalid. He makes it plain,
furthermore, that he is speaking of modus ponens for ordinary,
classical, material, implication. However, he doesn't deny that if 'A'
and 'A� B' are true, then, necessarily, 'B' is true; what he means,
when he says that MPP isn't valid, isn't what the classical logician
means, when he says that MPP is valid, since the relevance logician
would agree that MPP is valid in the classical sense of 'va/id'. This
case, 1 think, gives sorne grounds for a global pluralism ( and it may
be that there is also something to be said for the idea that, in lntui-
tionist logic, a non-classical conception of logical truth is being
employed).
However, to opt unqualifiedly for global pluralism at this point
would be, 1 think, to take much too lightly the relevance logicians'
insisten ce that the classical logicians' conception of validity is not just
different from theirs, but also inadequate. There is real competition,
genuine rivalry, here, not over which arguments are valid in an
agreed sense of 'valid ', but over what conception of validity is most
proper and adequate. (Recall the suggestion made earlier, p. :201, that
relevance logicians could be seen as urging that relevance of prernises
to conclusions, which classical logicians regard as a rhetorical feature
of good by contrast with bad arguments, is really a logical feature of
good arguments, a matter of validity.) Due regard for the significancc
of this disagreement seems to require that one combine a kind of
global pluralism about logical systems with a recognition that there
may be real competition at the level of metaconcepts,
What, now, of the meaning-variance argument for pluralism? It is
230 Philosophy of logics

not plausible, to my mind, to say that when Lukasiewicz, for instance,

denies that 'p v -p' represents a logical truth, his apparent disagree-
ment with the classical logician can be wholly accounted for as the
result simply of his giving a novel meaning to 'V ' or ' - ' or both. I
deliberately put the point in this guarded way; what I am denying is
not that any deviant logic ever involves any change of meaning of the
logical constants - it is reasonable to suspect sorne idiosyncrasy in the
meaning of Intuitionist negation and quantification, for example -
but that any deviance from classical logic inevitably involves such
wholescale meaning-variance as necessarily to preclude real rivalry.
(I have argued this in detail, with specific reference to Quine's
arguments from translation, in 1974 pp. 14-21 and 1977c.)
The question is tricky because there are reasons both for and
against meaning-variance. I argued in ch. 3 § 2 that the meaning of
the connectives can be thought of as deriving in part from the axioms/
rules of the system in which they occur and from its formal semantics,
and in part, also, from the informal readings given to the connectives
and the informal explanation of the formal semantics. The axioms/
rules and the formal semantics of deviant systems are, of course,
different from the classical, and the informal semantics may differ
too ( cf. the discussion of whether intermedia te values in many-valued
systems must be regarded as truth-values, ch. 1 I § 3); this argues for
sorne meaning-variance. On the other hand, deviant logicians usually
employ the same informal readings of their connectives (' not ',
'and ', 'or ', 'if') as the classical logician, which, on the othcr hand,
seems to be a prima facie indication that they intend to offcr rival
representations of the same informal arguments.
But this suggests a thought which has tended to be overlooked in
the debate about meaning-variance (but cf. Quine 1973 pp. 77ff.).
Formalisation involves a certain abstraction from what are taken to be
irrelevant or unimportant features of informal discourse; the logician
feels free to ignore the temporal connotations of 'and ', say, or the
plurality implied by 'sorne'. And this lea ves scope for, so to speak,
alternative formal projections of the same informal discourse; i.e.
scope for the idea that, for instance, material implication, strict impli-
cation, relevant implication, and other formal conditionals might ali
have sorne claim to represent sorne aspect of 'if', or that 2-valued and
3-valued and non-extensional disjunctions might ali be possible pro-
jections of (sorne) uses of 'or'. And this supplies more support for a
pluralist approach, according to which, however, rather than dífferent
 Metaphysical and epistemolopical questions 231

formalisms aspiring to represent different informal arguments, they

may be giving different representations of the same arguments.
Once again, it is likely that there will be disagreement between
deviant and classical logicians - even if their rivalry at the level of the
logical systems can be mitigated as I have suggested - about what is
the best, or perhaps, the proper, way to represent informal arguments.
However, I am sceptical of the idea that one can expect there to be a
unique, ideally perspicuous formal notation in which the unique
logical form of every informal argument is correctly represented
(hence my preference for 'a logical form' over 'the logical form' of an
argument, ch 2 §4). Sorne formal representations may be better than
others, either absolutely, or, for certain purposes, but I'm not confi-
dent that there is a unique best. (It is possible, also, that one formal
representation should be preferable in one area of discourse and
another in another; and if so perhaps something of local pluralism
could be salvaged.)
So, I am inclined to favour a global pluralist position: there can be
severa} logical systems which are correct in the sense I have explained.
The monist picture (fig. 8 above) should be replaced by something
more along the lines of fig. 9:
(i) formal arguments (wffs)• which are (ii) valid-(logically true-] ín-L
\ 1 I
\ 1 / \ 1 I
\ 1 I
represent corresponding to (iii)'s being
\ 1
'\ y y
I \ y y
(iii) informal argument (statement) which is (iv) extra-systematically valid
(logically true)•
Fig. 9

where informal arguments may be represented formally in more than

one way, and when validity-/logical truth-in-L may correspond to
different extra-systematic conceptions of validity/logical truth. How-
ever, I stress fi.rst that this does not mean that one neuer has to
choose between a deviant and the classical logic, only that one son,e-
times need not (so my pluralism is, so to speak, piecemeal although
global); and second that, even in those cases where a deviant and the
classical logic may be both correct, there may nevertheless be compe-
tition between them at the metalogical level, e.g. about how the idea
of validity may properly be understood, or how certain informal
arguments may best be formally represented. (The stars in fig. 9
indicate when such metalogical rivalry is to be located.)
232 Philosophy o/ logics

It may be worthwhile to point out that this position is able to

accommodate at least sorne of the considerations whích have been
taken to indicate monism, local pluralism or instrumentalism. I have
allowed - what the monist primarily stresses - that sorne logícal
systems may really compete with each other, in the strong sense that
they cannot both be correct; I have denied only that logical systems
must always compete in this way. I have also urged recognition of
metalogical competition where I think that logical rivalry can be
defused. And the suggestion that different formal representations
may be best for different purposes perhaps offers sorne comfort to the
local pluralist. I've less, at this stage, to offer by way of concessions to
the instrumentalist. In the next section, however, I shall be arguing
for a quite radical approach to the epistemology of logic, an approach
which would be quite congenial to an instrumentalist. ·

2 Epistemological questions
... no statement is immune from revision. Revision even
of the law of the excluded middle has been proposed as a
means of simplifying quantum mechanics; and what
difference is there in principie between such a shift and
the shift whereby Kepler superseded Ptolemy, or Einstein
Newton, or Darwin Aristotle? (Quine 1951 p. 43)
Quine is claiming that logic is revisable. I think he is right; but the
epistemological issues that this claim raises are far more cornplex than
one might suspect from the elegant but rather perfunctory treatment
they receive in 'Two dogmas'.
One needs, first, to get clear just what is meant by the claim that
logic is revisable - and, equally importantly, what is ,wt meant by it.
What I mean, at any rate, is not that the truths of logic might have
been otherwise than they are, but that the truths of logic might be
other than we take them to be, i.e. toe could be mistaken about sohat the
truths o/ logic are, e.g. in supposing that the law of excluded middle is
one such.
So a better way to put the question, because it makes its epistemo-
logical character clearer, is this: does fallibilism extend to logic?
Even this formulation, however, needs further refinement, for the
nature of fallibilism is often misundersrood.
 Metaphysical and epistemological questions 233

What is fallibilism?
To say that a person (or group of people, 'the scientific
community ', for instan ce) is fallible - I'll use 'fallible' in the sense
of 'cognitively fallible ', that is to say, fallible with respect to beliefs,
and not, for instance, with respect to promises, resolutions, etc. - is to
say that he is liable to hold false beliefs; to say that a method is fallible
is to say that it is liable to produce false results; of course, a person
may be fallible because he uses fallible methods - consultation of
entrails or horoscopes, perhaps - of acquiring beliefs. It seems to me
to be undeniable that people are fallible - we are all liable to hold at
least sorne false beliefs; we know of beliefs that people used to hold
that they are false - once, for instance, people believed that the sun
moves round the earth, that the earth is flat ... etc. - and it is reason-
able, as well as modest, to suppose that we, too, believethings thatare
false, though, of course, we don't know which of the things we believe
are the false ones, and we should naturally stop believing them if we
did.
However, epistemologists have often thought that, with respect to
certain kinds of belief - beliefs about one's own immediate sense-
experience are a favoured example - people may be infallible: they
are liable to have false beliefs about astronomy, geography ... etc., but
they are not liable to be mistaken about whether they are in pain,
seeing a red patch ... etc. And neither, sorne writers have argued,
are we liable to be mistaken abo u t the truths of logic; logic, they think,
has a special epistemological security. Popper, for example, though
he stresses our fallibility where scientific conjectures are concemed,
nevertheless seems confident that logíc is safe; see 196o for his
fallibilism, and cf. 1970 for his refusal to extend fallibilism to logic.

Does f'allibilism extend to logic?

(i) necessity, Now why should one find willingness to admit
that we might be mistaken in what we take to be the laws of physics,
but unwillingness to admit that we might be mistaken in what we
take to be the laws of logic? One important reason - important at
least because it's based on a signíficant confusion - derives from the
presumed necessity of logical laws. The argument would go some-
what like this: the laws of logíc are necessary, that is to say, they
couldn't be otherwise than true; so, sínce a logical law can't be false,
one's belief in a logical law can't be mistaken, and so, is infallible.
I have little doubt that this argument is unsound. (The truths of
�34 Philosophy of logics

mathematics are, also, supposed to be necessary. But nevertheless

we are apt to hold false rnathernatical beliefs, the result of rnistakes in
calculation for example. And if the laws of physics are, as sorne sup-
pose, physically necessary, this is not usually thought to entail that
we are able infallibly to tell what the laws of physics are.) But what
is wtong with the argurnent that, since the laws oflogic are necessary,
fallibilisrn does not extend to logic?
The argument goes wrong in two ways. First, it depends on using
'fallible' as a predicate, not of persons, but of propositions: a predi-
cate tneaning, presurnably, 'possibly false'. Now it is quite true that,
if the laws of logic are necessary, they are not possibly false, and
hence, in this sense, they are 'infallible '. But the thesis that sorne
propositíons are possibly false [which I'll call 'proposition falli-
bilism ') is an uninteresting logical thesis, which should not be con-
fused with the interesting episternological thesis that we are liable to
hold false beliefs (which I'll call 'agent fallibilisrn '). And proposítíon
infallibilism doesn't entail agent infallibilisrn; even if the laws of
logic are not possibly false, this by no rneans guarantees that we are
not liable to hold false logical belíefs. In claiming that we are fallible
in our logical beliefs (that agent fallibilisrn does extend to logic) I arn
not, of course, asserting the contradictory thesis that though, say,
'p v -p' is necessary, we rnight falsely believe thatp v -p; rather,
f arn claiming that, though 'p V -p' is necessary, we rnight falsely
believe that -(p V -p), or else, perhaps, though 'p v -p' is not
necessary, we falsely believe that it is. (I deliberately choose excluded
middle as an example of a purported logical law, since there is, of
course, dispute about its status.) Secondly, the argument is given a
deceptive plausibility by the ease with which the thesis that sorne
propositions are possibly false is confused with the thesis that sorne
propositions are contingent. If the laws of logic are necessary, our
logical beliefs will, indeed, not be contingent, but either necessarily
true or necessarily false. But 'possibly false' should not be equated
with 'contingent', for necessarily false beliefs are possibly false.1
I
If I am right that 11n interesting, genuinely epístemologícal fallibilism
will make 'fallible' 11 predicate of persona rather than propositions, this
has the consequence that Popper's attempt to accommodate fallibilism
within an 'epistemology without a knowing subject' (see his paper of
that title in 1972) is misguided. And if I am right that agent fallibilism
can quite consistently extend to aubject-matten the trutha of which are
necessarily true there is no need for embarraasment (such as even a
'contrite fallibiliat' like C. S. Peirce manifesta) about extending
fallibilism to mathematics.
 Metaphysical and epistemological questions 235
Faith that logic is unalterable has often enough been the basis for
denying that Iogic is revisable. Once it is clear - as I hope it is by
now - that the necessity of logical principies does not show us to be
Iogically infallible, it will be clear, also, that iflogic is unrevisable, it is
not because it is unalterable.
Now one reason for believing that we are fallible where our beliefs
about the world are concerned is that we know that people once
confidently believed what we now (we think) know to be false; and,
though we're sure that they were wrong to think, for instance, that the
earth is flat, the fact that their beliefs turned out to be false is a reason
for us to admit that sorne of our beliefs may, also, turn out to be
mistaken. And similar reasons operate, I should have thought, for a
comparable modesty about our logical beliefs. For example: Kant
wrote that 'In our own times there has been no famous Iogician, and
indeed we do not require any new discoveries in Logic ... ' (1800
p. 11); his confidence that logic was a completed science seems to us -
with the benefit of hindsight, after the enormous advances made in
logic since the Iast quarter of the nineteenth century - to exhibit a
curious and remarkable over-confidence. (Kant's confidence in
Aristotelian logic was based on the belief that Iogic embodies the
'forms of thought', that we can't but think in accordance with these
principies. Discussion of these ideas will be taken up below.) Or,
again: Frege thought that the reduction of arithmetic to Iogic would
guarantee arithmetic epistemologically, because he took the truths of
logic to be self-evident ; we, however, knowing that Frege's 'self-
evident' axiorns were inconsistent, are apt to find his confidence mis-
placed. (Lakatos 1963-4, in a splendid philosophical essay on the
history of mathematics, similarly subverts the tendency to place
mathematics on an epistemological pedestal.) Another reason against
epistemological over-confidence is the knowledge that other people
hold, with as much confidence, beliefs incompatible with one's own.
And this motive opera tes in the sphere of logic, too; the very plurality
of logical systems speaks against our possession of any infallible
capacity to ascertain the truths of logic.

(ii) self-eoidence. StiU, the idea that the truths of logic are self-
evident needs closer examination. What does it mean to claim that
sorne proposition is self-evident? Presurnably, something to the effect
that it is obviously true. But once it has been put Iike this, the diffi-
culty with the concept of self-evidence cannot be disguised. The fact
236 Philosophy of logics

that a proposition is obvious is, sadly, no guarantee that it's true. (It is
pertinent that different people, and different ages, find different
and even incompatible propositions - that sorne men are naturally
slaves, that ali men are equal. .. - 'obvious '.) Whether one says that
Frege's inconsistent axioms only seemed self-evident, but couldn't
really have been, or that they toere self-evident but unfortunately
weren't true, self-evidence must fail to supply an epistemological
guarantee ; beca use either ( on the latter assumption) a proposition
may be self-evident but false, or else (on the former assumption)
though if a proposition is self-evident then it is, indeed, true, one has
no certain way to tell when a proposition is really self-evident.1

(iii) analyticity. Another reason for doubting the revisability of

logic seems to derive from the idea, first, that logical truths are
analytic and then, that analytic truths are, so to speak, manifest. If A
is true in virtue of its meaning, the thought is, then no-one who
understands it can fail to see that it is true. There is room, I think, for
doubt about whether a really convincing argument could be de-
veloped along these lines; for the idea of "truth in virtue of meaning'
is far from transparent, not only because (as Quine has long urged) of
the 'meaning ', but also because of the 'in virtue of'. And even
supposing that it could, there is room for further doubt about whether
its conclusion would seriously damage fallibilism, for even if, if one
understood a logical truth correctly, one could not but recognise its
truth, this would guarantee the correctness of one's logical beliefs
only if one a/so had sorne foolproof way to be sure that one had
correctly understood a candidate logical truth. (The structural
similarity between this comment and the previous criticism of the
"self-evidence ' argument is worth noticing.)

A digression: ' Tsoo dogmas' agmn

This seems the appropriate place for sorne observations about
the structure of Quine's argument in 'Two dogmas'. The paper
opens (I simplify, but not, I hope, misleadingly) with an attack on the
analytic/synthetic distinction, and doses with a plea for the revisa-
bility of logic. What is the connection between the two?
One may interpret Quine as urging the revisability of logic as an
argument against the logical positivists' conception of analyticity.
1
My commenta have much in cornmon with Peirce's very shrewd
critique (1868) of the infallible faculty of 'intuition' which Descartes
IUppoted UI to pos1e11.
 Metaphysical and epistemological questions 237
The positivists take the meaning of a sentence to be given by its
verification conditions; and hence, take a statement to be analytic,
or true in virtue of its meaning, just in case it is verified come what
may. They run together the metaphysical idea of analyticity with the
epistemological idea of a priority; which is why it would be appro-
priate for Quine to attack the claim that logic is analytic, in this
sense, by arguing that logic is revisable. On this interpretation, the
revisability of logic is nota conclusion, but a premise, of the argument
of 'Two dogmas'.
Another possibility is to see the attack on analyticity as premise
and the plea for the revisability of logic as conclusion. However, if the
argument is: if the laws of logic were analytic they would be unre-
visable, but since there are no analytic truths, the laws of logic aren't
analytic, and so, they are revisable - it would be a bad one. lt is
invalid, having the form 'A-+ -B, -A, so B'. One premise is
false, since, as I've just argued, A's being analytic doesn't preclude
our being mistaken about it. And the other premise hasn't been
established; Quine attacks the second disjunct of the 'Fregean'
definition of an analytic truth as either a logical truth, or reducible to
a logical truth by substitution of synonyms, but this can scarcely show
that logical truths aren't analytic, for they qualify under the first
disjunct.
Nevertheless, this interpretation is worthy of sorne attention;
because it enables one to understand Quine's more recent increasing
conservatism about logic. The attack in "Two dogmas' on synonymy
etc. uould threaten an account of logical truths as analytic because
true in oirtu« o/ the meaning o/ the logical constants. Now in Word and
Object Quine renews his sceptical attack on meaning notíons, but
makes an exception in the case of the logical connectives, which, he
claims, do have determínate meaning ( 196oa ch. 2); and this paves the
way for his acceptance (1970 ch. 6) of a meaning-variance argument
to the effect that the theorems of deviant and classical logics are, alike,
true in virtue of the meaning of the (deviant or classical) connectives;
which, in turn, seems to lead him to compromise his earlier insistence
that fallibilism extends even to logic.

Revision o/ logi'c
If fallibilism does extend to logic, if, as I have claimed, we
are apt to be mistaken in our beliefs about logic, then it would be
prudent to be prepared, if need be, to revise our logical opinions,
238 Philosophy o/ logics

But this isn't to say that revisions of logic should be lightly under-
taken, for the extreme generality of Iogical principles means that such
revisions will have the most far-reaching consequences; logic is
revisable, but the reasons for revision had better be good. As I argued
in ch. 11 § 3, the arguments offered in favour of deviant logics have,
too often, been rather weak.

3 Logic and thought

Kant's confidence in the unrevisability of Aristotelian logic
rested on the idea that logical principies represent 'the forros of
thought', that we can't but think in accordance with them: an idea
that raises a host of intriguing questions about just what logic has to
do with 'the way we think'.
Although at one time it was quite usual to suppose that the prin-
ciples of logic are 'the laws of thought' (see Boole 1854), Frege's
vigorous critique was so influential that there has been rather little
support, of late, for 'psychologism' in any shape or form. However,
Frege's arguments against psychologism are, 1 suspect, less con-
clusive, and at least sorne form of psychologism more plausible, than
it is nowadays fashionable to suppose. A full-scale re-assessment of
psychologism would require, however, a fuller and more sophisti-
cated account of the nature of thought than I am able to offer; so
what follows can be sketchy at best.
One can begin by distinguishing - the distinction is pretty crude,
but nevertheless may be serviceable as a starting-point - three kinds
of position:

(i) logic is descriptive of mental processes (it describes how

we do, oc perhaps how we must, think)
(ii) logic is prescriptive of mental processes (it prescribes how
we should think)
(iii) logic has nothing to do with mental processes

One might call these strong psychologism, weak psychologism and

anti-psychologism, respectively. Examples: Kant held something
like (i); Peirce a version of (ii); Frege, (iii).
In what follows I shall present sorne arguments for a form of weak
psychologism rather close to that adopted by Peirce (1930-58, 3,
161ff.): that logic is normative with respect to reasoning. 1 shall then
 Metaphysical and epistemological questions 239

go on to point out sorne advantages of weak psychologism as against

anti-psychologism, on the one hand, and strong psychologism, on the
other.
Logic is primarily concemed with arguments: how, then, can it
relate to the mental processes which constitute reasoning] l'U tackle
this question in two stages, o:ffering, first, a Platonist answer, and then
a nominalist version of that answer; the reason for this strategy is
that the connection between logic and thought is thrown into sharper
relief by the Platonist account, but I think that it is better, though
less simply, explained in the nominalist version.
The Platoníst answer: Logic is concerned with the (in)validity of
argumenta, with the connection between premises and conclusíon;
logical relations are relations, such as entailment or incompatibility,
between propositions. Reasoning is a ( certain sort of) mental process,
such as, coming to believe that q on the strength of one's belief that
P (inferring q from p ), or, coming to recognise that if p were the case,
then q would be the case; and to believe that p, orto wonder whether,
or what if, p, is to stand in a certain relation to a proposition. Hence,
logic is normative with respect to reasoning in this sense: that if, e.g.,
one infers q from p, then, if the argument from p to q is valid, the
inference is saje, in that it is guaranteed not to result in one's holding
a false belief on the basis of a true one.
The nominalist version: that s believes that p, or wonders whether,
or what if, p, can be analysed, ultimately, in terms of a complicated
relation between s and the sentence 'p '; and Platonist talk of belief in
or entertainment of a proposition is to be regarded as a convenient
shorthand for this complicated analysis. Logic is concerned with the
validity of arguments, which, however, are to be conceived (ch 2 § 1)
as stretches of discourse/strings of sentences; and Platonist talk of
logical relations between propositions is, again, to be regarded as a
convenient shorthand (specifically, for quite complicated qualifiea-
tions about what sentences are to be regarded as inter-substitutable,
ch. 6 §4-). Once again, it follows that logic is normative, in the sense
explained above, with respect to reasoning.
The nominalist version of weak psychologism is, I think,
preferable to the Platonist, for reasons which will emerge from
a consideration of an argument of Frege's against psycholog-
ism.
Frege's objections to psychologism are quite complex, and I shall
only consider the argument which is most relevant to the position I
240 Philosophy o/ logics

have defended.' This argument runs as follows. Logic has nothing to

do with mental processes; for logic is objective and public, whereas
the mental, according to Frege, is subjective and private. This is why
Frege is so concerned to stress (see especially Frege 1918; and cf.
p. 6m above) that the sense of a sentence is notan idea (a mental
entity), but a thought (Gedanke: an abstract object, a proposition).
Since ideas are mental, they are, Frege argues, essentially private;
you can no more have my idea than you can have my headache. If the
sense of a sentence were a private, mental entity, an idea in Frege's
sense, there would be a mystery about the relation between one
person's idea and another's:

a's idea ? b's idea

0 a
0 b
Propositions, however, are public; you and I can both 'grasp' the
same proposition, and this is what makes it possible for there to be
objective, public knowledge.8
This argument could be questioned on more than one score: e.g.
why does Frege assume that everything mental is subjective and
private? Is it relevant that the psychology with which he was familiar
was introspectionist? But it is, anyway, pretty clear that the argument
does not oblige one to divorce logic from mental processes in the way
Frege supposes. For the postulation of propositions will only
guarantee the publicity of knowledge if propositions are not only
objectioe, but also accessible, if we can "grasp ' them; and this is just
what the Platonist version of the argument for weak psychologism
requires.
In fact, however, Frege has nothing very substantial to say to
mitigate the mysteriousness of our supposed ability to 'grasp' his
Gedanken:
1
1 shall ignore altogether Frege's arguments against psychologistic
accounts of number, except to observe that, in view of his logiciam, he
would have taken these arguments to bear indirectly on psychologism
with respect to Iogic,
I
Popper's reasona for divorcing epieternology from paychology are very
similar.
 Metaphysical and epistemological questions

a b

But this mystery can be dispelled by concentrating, not on ideas

(which create a problem about objectivity), nor on propositions
(which create a problem about accessibility), but on sentences; for the
verbal behaviour of users of a language is both objective and accessible:

sen ten ce 'p'

a b
(Dewey saw this: see 1929 p. 196.) And this supplies a reason for
preferring, as I urged, the nominalist version of the argument for
weak psychologism.
Logic, I suggested, is prescriptive of reasoning in the Iimited sense
that inference in accordance with logical principies is safe. (Of course,
safety needn't be an overriding consideration; one might, quite
rationally, prefer fruitful but risky to safe but relatively uninteresting
procedures; cf. de Bono's championship, e.g. in 196g, of 'lateral
thinking'.) It is important, however, that on the weak psychologistic
view, though logic is applicable to reasoning, the validity of an argu-
ment consists in its truth-preserving character; it is in no sense a
psychological property. Consequently, weak psychologism avoids the
main difficulty of strong psychologism, the problem of accountíng
for logical error: for, since people surely do, from time to time, argue
invalidly, how could the validity of an argument consist in its con-
fonnity to the way we think? This isn't to say that strong psycholo-
gism is flatly incompatible with logical mistakes; but that the two can
be reconciled only by means of sorne explanation of such mistakes as
the result of sorne irregularity or malfunction of our reasoning powers.
(According to Kant, logical mistakes are the result of the unnoticed
influence of sensibility on judgment.) Nevertheless, its much readier
242 Philosophy of logics

reconcilability with fallibilism speaks, 1 believe, in favour of weak

over strong psychologism.
There are, inevitably, many intriguing questions this leaves un-
answered: for instance, what, exactly, distinguishes logical from
psychological study of reasoning? (lt can't be, as is sometimes
supposed, that psychology, unlike logic, is never normative, nor even
that it is never normative with respect to truth; consider, for instance,
psychological studies of the conditions of reliable/illusory perception.)
What consequences would psychologism about logic have for ques-
tions about the relations between epistemology and psychology? What
has logic to tell us about rationality? What would the consequences
for psychologism be (in view, especially, of Chomsky's claim that
certain grammatical structures are innate) of the conjecture that
logical form can be identified with grammatical form?
lt's good to know (to borrow a phrase of Davidson's) we shan't
run out of work I
 GLOSSARY

• by a term indicares that it has a separate entry, For terminology not

explained here, thc reader may find it useful to consult the Dictionary of
Philosophy (Runes 1966), or rhe entry under Logical terms, glossary of in
Edwards 1967.
Analytic/synthetic
An analytically true judgment is one such that the concept of its predicare
is contained in its subject, or, such that its negation is contradictory
(Kant); an analytically true proposition is either a logical truth •, or else
reducible to a logical truth by means of definitions in purely logical terms
(Frege: and see logicism•); an analytically true statement is true solely in
virtue of the meaning of its terms (logical posirivista"). 'Analytic' is
generally used equivalently with "analytically true'; the negation " of an
analytic truth is analytically false. 'Synthetic' is generally used equiva-
lently with 'neither analytically true nor analytically false'. See discussion
of Quine's critique of analyticity, pp. 172-4 and 236-7.
A priori/a posteriori
A proposition is a priori if it can be known independently of experience,
otherwise a posteriori (an epistemological distinction, by contrast with the
metaphysical analyticlsynthetics distinction). See discussion of fallibilism
with respect to logic, ch. 12 §2.
Atomic An atomic wff• of sentence calculus is a senrence letter (e.g. 'p '), by con-
trast wíth a compound, or 'molecular' wff (e.g. 'p V q '). An atomic wff of
predicate calculus is an n-place• predicare letter followed by n variables+
or singular terms. An atomic statement, analogously, is a statement which
contains no statement as components.
Axiom A wff• A is an axiom of L if A is laid down, its truth unquestioned, in the
system L (rrivially, ali axioms of L are theorems+ of L). An axiomatic
presentation of logic uses axioms as well as rules of inference=. See
ch.2§3.
Bivalence
Every wff• (sentence, statement, proposition) is either true or else false;
see also excluded middle+, See ch. 11 § 3.
Characteristic matrix
A matrix is a set of truth-tables. A matrix M is characteriuic for a system S
iff ali and only the wffs• uniformly deaígnated " (tautologous•) on M are
24-4- Glossary
theorems+ of S. A system is n-valued if it has an n-valued characteristic
matrix and no characteristic matrix with fewer than n values; many-ualued
if it is n-valued for n > 2, infinitely many-oalued if n-oalued for infinite
n. ch. 3 § 1 and ch. 11.
Combinatory logic
A branch of formal logic in which variables+ are eliminated in favour of
function symbols. See discussion of Quine's ontological criterion, ch. 4 §2.
Complete
(i) A formal systern is weakly complete if every wff• which is logically true•
in the system is a theorem+ of the system; or strongly complete if, if any
new independent " axiom • were added, it would be inconsistent•.
Examples: sentence calculus is strongly complete; the usual modal
systerns are weakly complete; set theory, arithmetic are incomplete. See
the discussion of completeness as a criterion for counting a system as a
system of logic, ch. 1 § 2.
(ii) For Junctional completeness, see under truth-functional",
Conditional
The operators ' - ' ' --3 ', etc.
A wff• of the form 'A-+ B', (or statement of the form 'If A then B') is
also called a conditional or hypothetical wff. 'A' is called the antecedem,
'B' the consequent of the conditional. A subjunctiue conditional is one with
a subjunctive verb (as' If income true were to be halved, we should ali be
delighted '); a counterfactual conditional is a subjunctive conditional which
implies that its antecedent is false (as 'Had income tax been halved in the
last budget, we should ali have been delighted'). See ch. J §2; ch. 10 §6.
Conjunction
A wff• (staternent) of the form 'A & B' ('A and B').
Consequence
A wff• (statement) B is a logical consequence of A iff there is a valid•
argument from A to B.
Consistent
A formal system is contistent iff no wff• of the form 'A & -A' is a
theorem •; or, iff not every wff of the system is a theorem ; or (in the sense
of Post, applicable to sentence calculus) iff no single sentence letter is a
theorem.
Constant
A constant is a symbol employed always to stand for the same thing (as,
singular terms such as' a', 'b ', ... etc., or operators such as' & ', ' V' ... etc.)
by contrast with variables+ (as, 'x', 'y', 'z' ... etc.), which range over
a domain • of objects.
Contradiction
Wff• of the form 'A & -A'; statement of the form 'A and not A'.
Principie oJ non-cantradiction: -(A & -A); or: no wff (sentence, state-
ment, proposition) is both true and false.
Contradictory
The contradictory of a wff• (statement) A is a wff• (statement) which
must be false if A is true and true if A is false.
Contrary
Wffs• (staternents) A and B are contrary if they can't be both true but
may be both false.
Decida ble
A system is decidable if there is a mechanical procedure (a 'decision
 Glossary

procedure ') for determiníng, for any wff• of the syatem, whether or not
that wff is a theorems. Examples: sentence calculus is decidable; the full
predicate calculus (including polyadic• as well as monadic " predícates) is
not, Truth-tables supply the decision procedure for sentence calculua;
a truth-table test determines whether a wff is a tautologyé, and, by the
eoundness= and cornpletenessé results, ali and only tautologies'" are
theoremse.
Deduction
A sequenceof wffs• (of L) is a deduction (in L) of BfromA1 ... A,. iff itis
a valid• argument (in L) with A1 ... An as premises and B as conclusion.
Deduction theorem
If, in a formal system L, if
A1 .. , An 1-L B, then 1-L A1->- (A1->- ( ... (A,.--,. B)))
then the deduction theorem bolds for L.
Definite description
Expression of the form 'The so-and-so", written, formally, '(•x) Fx':
See ch. S §3.
Definítíon
An explicit definition defines one expressíon (the definiendum) by means of
another (the definiens) which can replace the first wherever it occurs.
A contextual definítíon supplies a replacernent for certain longer expres-
sions in which the definiendum occurs but not an equívalent for that
expression itself. (If xs can be contextually defined in terms of ys, xs are
sometimes said to be lagical constructions out of ys, and 'x' to be an
incompleto symbal•.) A recursioe definition gives a rule for eliminating the
definiendum in a finite number of steps. A set of axioms'" is sometimes said
to give an implicit definítion of its primítive" terms, See ch. 3 § r for the
interdefinability of connectives: ch. 4 § 3 for Russell's contextual definí-
tion of definite descriptions ; ch. 7 § 5 for Tarski's recursive definition of
satisfaction; pp. 103-4 for formal conditions on definitions.
Designated value
'Truth-like value, such that ali compound wffs• which take a designated
value for ali assignrnents to their componente are tautalcgies•.
Deviation
L1 is a deviation of L1 if it has a different set of theorems/valid inferences
essentially involving vocabulary shared with L1• A deviation of classical
logic is a deoiant lagic. See chs. 9, 11, 12.
Dísjunction
Wff• (statement) of the form 'A V B'. Disjunctive dllemma is the form of
argument: if A 1- C, B 1- C, then A V B 1- C.
Dispositional
A dispositional predicare ascribes a tendency or 'ha bit'; in English man y
such predicares end in '-ble' (as: 'irritable', 'soluble'). Dispositional
statements (' thís sugar lump is soluble') are equivalent to subjimctÍfie
conditionalss (' íf this lump of sugar were to be placed in water, it would
dissolve'). See ch. 10 §3.
Domain
(U ni verse of diacourse) - range of the variables'" of a theory. See ch. 4 f J.
Double negation, principie of
A == - -A. See discussion of Intuitionist logic, ch. 11 f4.
Enthymeme
Argument with a suppressed premise.
246 Glossary

Epistemology
Theory of knowledge.
Equivalence
Two wffs (statements) are lagically equiualent if they necessarily have the
same truth-value, They are material/y equioalent if they have the same
truth-value.
Excluded middle
p V -p (cf. bivalence•). See ch. II § 3.
Extension
L1 is an extension of L2 if it contains new vocabulary, over and above
vocabulary shared with L1, and has new theorems=jvalid ínferences'"
involving the new vocabulary essentially. An extension of classical logic
is an extended logic, See chs. 9, 10, 12.
Extension/intension
Reference (extension) versus sense (intension) of an expression. For a singular
term, the extension is ita referent, for a predicare, the set of things it is
true of, for a sentence, its truth-value. Two expressions with the same
extensión are co-extensiue, Related terminology: Bedeutung ( = extensión)
versus Sinn ( = intension) of an expressíon (Frege); denotation versus
connotation (Mili); extensional versus intemionat• contexts. See discussion
of Frege's theory of sense and reference, ch. 5 § 2; cf. Quine's distinction
between theory of reference and theory of meaning, p. 119.
Extensional/intensional
A context is extensional if co-referential expressions - singular terms with
the same denotation, predicates with the same extension, or sentences
with the same truth-value - are substitutable within it without changing
the truth-value of the whole, 'salva oeritate ', i.e, if Leibni2' law holds for
it; otherwise, it is intensional, Examples: ' lt is not the case that ... ' is
extensional, "Necessarily ... ' or 's believes that ... ' are intensional.
Related terminology: oblique ( = intensional) context (Frege); referentially
transparent ( = extensional) versus referentially opaque ( a intensional)
context, purely referential occurrence (i.e, occurrence in an extensional
context) of a singular term (Quine); trvth-Junctionat• conncctive ( • ex-
tensional sentence-forming operator on sentences), See discusaion of
Davidson's prograrnme, ch. 7 § 5 ¡ cf. Quine's critique of the analytic/
synthetic distinction, ch. 10 § 1.
Finite/infinite
A set is infinite if it has a proper subset such that its members can be put
in one-one correspondence'" with the mernbers of that proper subset.
A set is finite if it is not infinite. A set is denumerably infinite if it can be
put in one-one correspondence with the natural numbers.
Formalism
School in the philosophy of mathematíes (Hilbert, Curry) characterised
by the view that numbers may be identified with marks on paper. See
díscussion of the formalist approach to logíc, p. 224.
Godel's (i ncompleteness) theorem
Arithmetic is incompletc; there is an arithmetical wff which is true but
neither provable nor disprovable (G6del 1931). See p. 139 for comrnents
on the role of self-reference in G6del's proof.
Goldbach'a conjecture
Hypothesis that every even number greater than 2 is the sum of two
primes.
 Glossary

Implication
(i) 'P' material/y implies 'q' (' p----+ q') if it is not the case that p and not q;
'p' strictly implies 's' (' p � q') if it is ímpossible thatp and not q (p � q
= L(p-+ q)). See ch. 3 §2 on '----+' and 'if, then'; ch. 10 §6 on relations
between material, strict, and relevant conditionals, and the intuitive idea
of entailment.
(ii) 'Implies' is also used in another way, as 's implied that p' (where it is
a relation between speakers and proposítions, rather than, as above,
between propositions). In this use it means something Iike 's hinted,
though he didn't exactly say, that p'. Compare discussion of Grice's
'conversational implicature ', p. 36,
lncomplete symbol
Contextually defined• expression. See ch. 5 §3, ch. 7 §7.
lndependent
The axíomse of a formal system are independent of each other if none is
a logical consequenceé of the others.
lndexical
Expression the reference of which dependa on the time, place or speaker,
e.g, 'now', 'I', 'here'. See ch. 7 §6(c).
Induction
(i) An argument is inductiuely strong if the truth of its premises makes the
truth ofita conclusion probable. See ch. 2 §2.
(ii) Mathematical induction: a form of (deductively valid) argument used in
mathematics, to show that ali numbers have a property by showing that
o has that property, and that if a number has that property, ita successor
also has it.
Inference
A person infers q from p if he comes to accept q on the strength of p, or comes
to accept that if p were the case, then q would be the case. See ch. 12 § 3
on the relevance of logic to inference; and ch. 2 § 3 on rules oj inferenc«.
Interpretation (of a formal system)
A set (the domain'", D) and a function assigning elements of D to
singular terms=, n-tuples of elements of D to n-place predicares, and
functions with n-tuples of elements of D as argurnent and elements of
D as val u e to function symbols. See interpreted and uninterpreted systems,
pp. 3ff; chs. 4 and 5; ch. 10§4 on 'pure' versus 'depraved' semantics.
Intuitionism
School in the philosophy of mathematics (Brouwer, Heyting), charac-
terised by the view that nwnbers are mental constructions: leads to a
restricted arithmetic and to a non-standard logic. See ch. 11 §4.
Logical atomism
School of philosophy (early Wittgenstein, Rusaell) seeking logically
to analyse the structure of the world into ita most fundamental com-
ponents (the' logical atoms '). See discussion of the correspondence theory
of truth, ch. 7 §2; of affinities with Davidson's programme, p. l24fl,
Logical positivism
School of philosophy centred on the Vienna Circl« (Schlick, Camap);
characterised by the oerification principie, according to which the meaning
of a statement is given by ita verification-conditiona, and unverifiable
statementa are meaningless. See discussion of Quine'a attack on the
analytic/,ynthetic• distinction, ch. 10 § 1.
248 Glossary

Logical truth
A wff• is logically true in L iff it is true in ali interpretations of L. See
ch. :z §:z.
Logicism
School in philosophy of mathematics, characterised by the thesis (Frege,
Russell) that the truths of arithmetic are reducible to logic (or, analytic»
in the Fregean sense); numbers are reducible to sets. See díscussion of the
logícist programme and the question of the scope of logic, ch. 1 §:z; of the
effect of Russell's paradox, p. 137.
Mass term
Expression denoting a kind of sruff or material (as: 'water', 'snow',
"grass ') rather than, like a 'singular term ', an individual object (as: 'glass
of water'). See ch. 7 § 6(c).
Metalogic
Study of formal properties - e.g. ccnsistency", completeness'", decida-
bility•, - of formal logical systems. See discussion of relations between
philosophy of logic and metalogic, ch. 1 § 1; of modal logic conceived of
as a metalogical calculus, p. 18:z.
Metaphysics
Traditionally, 'the science of being as such '. 1 use "rnetaphysical '
primarily to emphasise the distinction between questions about the way
things are (e.g. ' Is there one correct logic? ') from epistemological ques-
tions, questions about our knowledge of how things are (e.g. 'Might the
laws of logic be other than we take them to be?'). See ch. 12.
Modus ponens (MPP)
The rule of inference ", from 'A' and 'A--+ B' to infer 'B'. See discussion
of the failure of MPP in relevance logic, ch. 10 §7.
Monadic/dyadic/polyadic
An open sentence/connective is monadic (r-place) if it has one, dyadic
(a-place) if it has two, polyadic (many-plaee) if it has more than two
argument(s); e.g. ' ... is red' is a monadic, • . , . is larger than - ' a dyadic
open sentence, See the discussion of the role of sequences of objects in
Tarski's definitions of satisfaction/truth, pp. 105ff,
Monism/pluralism/instrumentalism
(i) In metaphyslcs'", monism is the thesís that there is only one ultímate kind
of thing, dualism the thesis that there are two, pluralism the thesís that
there are more than two.
(ii) Monism about logic is the thesis that there is only one correct logical
system, pluralism the thesis that there is more than one correct logical
systern, instrumentalism the theais that the notion of' correcmess ' doesn 't
apply to logical systems. See ch, 12 § 1.
Natural deduction
A natural deduction presentation of a logical system relies on rules of
inference= rather than axioma•, See ch. 2 §3.
Necessary/sufficient conditions
A is a necessary condition for B, if B can't be the case unless A is; A is
a sufficient condition for B, if, if A is the case, B is.
Negation
The negation of 'A' is' -A'.
Nominalism/Platonism/ conceptualísm
The nominalist denies, the Platonüt uaerta, that there are real UfflfJtr1al1
(e.g. redneu squareness etc.r: the conceptualist claims that universals are
 Glossary :249
mental entities. Related terminology: reism, materialism, pansomatism
(forms of nominalism) versus realism (form of Platonism). See the discus-
sion of second-order quantifiers=, ch, 4 § 3; of Davidson's extensionalism•
and Kotarbif>.ski's nominalism, p, 124n; of the status of possible worlds,
ch, IO §4,
Object language/metalanguage
In talking about systems, the systern being talked about is known as .the
object language, the system being used to talk about it, the metalanguage.
(N.B. this is a telative rather than an absolute distinction; e.g, one
might use French (the metalanguage) to talk about English (the object
language) or English to talk about French.) Thus, metalogic=, the study
of logical systerns. See discussion of Tarski's use of the distinction in the
definition of truth, ch, 7 §6; cf. its relevance to the semantic paradoxes+,
ch. 8 §2.
One-one eorrespondence
Two sets• x and y are in one-one correspondence if there is a one-one
relations, R, by which each member of x is related to exactly one member
of y, and each rnember of y to exactly one mernber of x,
Ontology
Part of metaphysics concerned with the question, what kinds of thing
exist. See ch. 4 §2 for discussion of relations between logic and ontology.
Oratio obliqua
lndirect (reported) speech, as: "s said thatp'. See ch. 7 §6(c).
Paradoxes
(i) (Also known as 'antinomies ',) Contradictions derivable in semantica"
and set• theory; they include the Liar ('This sentence is false') and
Russell' s parado» (' The set of ali sets which are not members of themselves
is a mernber of itself iff it is nota member of itself'). See ch. 8.
(ii) The "paradoxes ' of material and strict Implicaeíons are theoremss of
classical, a-valued and modal logic (' p-+ ( - p ...... q) ', 'L-p-+ (p -3 q))'
which seern rather counterintuitive with '-+' or ' -3 ' when read 'if ... '.
I use scare quotes beca use these • paradoxes ' involve no contradiction.
See ch. 10.
Peano postulates
Set of axioms for the theory of natural numbers:
1. o is a nurnber,
a. The successor of any number is a number.
3, No two numbers have the same successor.
4. o is not the successor of any number.
5, If o has a property, and, if any number has that property, then the suc-
cessor of that number has that property, then ali numbers have that
property, (Inductíone axiom.)
Pragmatisrn
American school of philosophy initiated by Peirce and James (other
pragmatista include Dewey and F. C. S. Schiller); characterised by the
• pragmatic maxim ', according to which the meaning of a concept is to be
sought in the empirical or practica! consequences (Kant - pragmatische -
empirically conditioned; Greek praxis - action) of its application. See
diacussion of pragmatist theory of truth, ch. 7 §4.
Presupposition
'A' preaupposea • B' if 'A' is neither true nor falee unleu • B' is true. See
ch . .5 §3.
250 Glossary

Primitive
Undefined tenn (see definitíon'").
Proof A proof (in L) of A is a deduction• (in L) of A from no premises except
the axioms'" (of L), if any. A wff• A is provable (in L) if there is a proof
(in L) of A; it is disprovable if its negation • is provable.
Propositional attitude
Verbs such as 'knows', 'believes', 'hopes' etc., which take the eonstruc-
tion ', <lls that p ', are known as verbs of propositional attitude (Russell).
Quantifier
Expression ('(3 ... )'- the existential quantifier- or '( ... )' - the universal
quantifier) binding variables=, See ch. 4.
Quantum rnechanics
A physical theory concerned with atomic structure, emission and absorb-
tion of light by matter. See discussion of 'quantum logic ', ch. 11 §2.
Refute Show a thesis (or theory etc.) to be false. N.B. to deny that p is not to
refute 'p'.
Relation
A 2- or more-place predicate is called a relation symbol; its extension• -
the set• of ordered pairs (triples ... n-tuples) of which it holds - is known
as a relation-in-extension. A relation R is transitiue if, if (x) (y) (e) Rxy
and Ry«, then Rxz; symmetric if (x) (y), if Rxy then Ryx; reflexive if
(x) Rxx,
Salva ceritate
Without change of truth-value.
Satisfaction
(i) In Tarski's definition of truth (ch. 7 § .5): relation between open sentences
and sequencea" of objects (as e.g. (Edinburgh, London, ... } satisfies
'x is north of y').
(ii) In imperative logic (p. 8.5): analogue of rruth-value, assigned to irn-
perative sentences (as e.g, 'Shut the door I' is satisfied iff the door is
shut),
Sequence
Ordered pair, triple ... n-tuple of objects (i,e, like a aet•, except that the
order matten; while (a, b) • (b, a), (a, b) +(b, a)). See the role oí
sequences of objects in Tanki's definition of satisfaction•, ch. 7 § .5.
Set • any collection into a whole ... of definite, distinguishable objects'
(Cantor); however, set-theory includes the null set, which has no mem-
bers. See Russell's paradox•,p. 136. '(a, b, e}' means'thesetconsisting
of a, b, e'; '(xlFx}' mesns 'the set of things which are F'; 'a e {x(Fx}'
rneans 'a is a member of the set of things which are F'. (In GlSdel-
von Neumann-Bemays set-theory a distinction is drawn between sets,
which can both have members and themselves be mernbers, and classes,
which have members but can't themselves be members.)
Skolem-Lowenheim theorem
Every theory that has a model (is consistent•) has a denumerable (see entry
under finite/infinite•) model. See p . .51 for ita bearing on substitutional
quantification.
Sound
(i) An argument is sound if (i) it is valid • and (ii) ita prerniae1, and hence, ita
conclusión, are true.
(ii) A lagical ,ystem is s0tmd iff all ita theorem1 are lo¡ically true•; 10undneu
is the converse of completenesa'",
 Glossary

Syntax/semantics/pragmatics
Syntax is the study of formal relationa between expressions; thus, the
vocabulary, fonnation rules and axioms•/rulea of inference s of a system
are called the syntax of the systern. Semantics is the study of relatiom
between linguistic expressions and the non-linguistic objects to which
they apply ; thus, the interpretation= of a system is called the sernantics of
the system. (Roughly, the distinction between syntax and semantics could
be cornpared to that between grammar and meaning.) Pragmaties is the
study of relations between expreasions and the use or users of theae
expressions. See syntactie and semantic accounts of validity, ch. 2 §2;
syntactic, semantic and pragmatic approaches to propositions etc., ch. 6 § 1;
pragmatic relations of com·ersational implicature, p. 36; and pre-
supposition •, pp. 67ff.
Tautology
Technical sense: a wff• is a tautology if it takes the value 'true' for ali
assignments of truth-values to its atomic • components (extended, in the
case of many-valued logics, to: if it takes a designated'" value for ali
assignments to its atomic cornponents). The soundneu• proof for
sentence calculus shows that only tautologies• are theorernsw; the com-
pleteness • proof that ali tautologies are theorems.
Non-technical sense: a statement is tautologous if it says the same thing
twice, and so is trivíally true. See the discussion of the pre-systematic idea
corresponding to the technical notion of logícal truth ", pp. 14-15.
Theorem
A wff• A is a theorern of L iff A follows from the axioms• of L, if any,
by the rules of inference " of L. See ch. 2 §2, ch. 12 § 1.
Theory of types
Russell's formal solution to the paradoxes+; the simple theory o/ types
avoids the set•-theoretical, the ,amified theory o/ types avoids the
semantic+ paradoxes. See ch. 8 §2.
Truth-functional
A connectiue (sentence-formíng operator on sentences) is trutJr-funetional
if the truth-value of a compound of which it is the main connective
dependa solely upon the truth-values of its componente; in which case
a truth-table can be given for that eonnectíve. A logical systmr is truth-
[unctional if ali its constants are truth-functional. An n-valued system is
junctionally complete - has an adequate set o.f connectioes - if it has enough
connectives to expresa ali n-valued truth-functions. Examples: the con-
nectives of classical and finitely many-valued sentence calculi are truth-
functional; modal operators and episternic operators are not. See
discussion of logicians' preference for truth-functional connectivee, ch. 3
§2; of many-valued and non-truth-functional calculi, ch. 11.
(T) acherna
Tarski's material adequacy condition requirea that any acceptable definí·
tion of truth have as consequence ali instances of the (T) acherna:
u
S true iff p
where ' S' names the sentence on the right-hand side, See ch. 7 ff S and 6.
Valid A formal argument is:
syntachºcally valid-in-L iff its conclusion follows from its premíses and
the axioms• of L, if any, by means of the rules of inference• of L,
snnantical/y valid-in-L iff its conclusion is true in ali interptttations of
L in which ali its premises are true.
252 Glossary

An informal argument is :
valid iff its premises could not be true and its conclusion false.
See ch. 2 §2; ch. 10 §6.
Variable Expression, as:' x', 'y' ... (in first-order predicate calculus), ranging ouer e
domains of objects; by contrast with conetanta", as: 'a', 'b' ... , each of
which denotes a specific element of the domain. An expression which can
be substituted for a variable is called a substituend for the variable; the
elements over which it ranges, its oalues. A bound variable is one within,
a free variable one without, the acope of a quantifier ". See ch. 4.
Verisimilítude
Nearness to the truth (Popper); see ch. 7 §6(b).
Vicious circle principie
Poincaré and Russell diagnose the paradoxes'" as resultíng from violations
of the vicious circle principie (V.C.P.): 'whatever involves all of a
collection cannot be a member of that collection '. See ch. 8 § 2.
Wff Well-formed formula, i.e, string of symbols of a formal language correctly
constructed with respeet to its [ormation rules. A formula is any string of
symbols of a formal language.
 ADVICE ON READING

I have given full references in the text, to enable the reader to Jocate
relevant literature on specific issues. The point of the present section is to
give those new to the subject sorne suggestions about where to begin
reading.
I have taken for granted an acquaintance with elementary formal logic, as
presented in, say, Lemmon 1()65, or Quine 1950, which is somewhat
harder but a good dea) richer, A compact presentation of metalogical
results can be found in Hunter 1971 or Boolos and Jeffrey 1974. For the
history of logic, consult Kneale and Knea)e 1()62.
Although there are severa) 'introductions • to the philcsophy of logic, they
are generally harder, and require more sophistication in the reader, than
their titles suggest: Strawson's Introduction to Logical Theory (1952) pre-
sents a sustained critique of formal logic from the point of view of
ordinary language philosophy, and should be read in conjunction with
Quine's review (1953c); Quine's Philosophy oJ Logic (1970) is, though
short, rich and wide-ranging, but it takes for granted a good deal of dis-
tinctively 'Quinean' philosophy, and is more suitable for the advanced
than the beginning student; Putnam's Philcsophy oJ Logic (1971) is
devoted to a single issue, the need for abstract entities in logic.
There are severa) valuable collections of papera, Van Heijenhoort 1()67a
contains the classic papers from Frege' s initiation of modern logic with the
Begriffs1chrift (1879) to Godel's incompleteness theorem (1931). Other
useful collections of more recent philosophical papers include Copi and
Gould 1()67, Strawson 1()67, Iserninger 1()68.
If you want to find reading on a specific subject, but don't know where
to begin Jooking, you may find the articles on Jogic and philosophy of
Jogic in the Encyclopaedia oJ Philcsophy (Edwards 1967) uaeful; they are
generally infonnative, and have helpful bibliographies. The revíews in the
journal oJ Symholic Logic of (philosophícal as well as formal) anides frorn
other journals may also be found valuable. 1 recommend, in general, that
you begin with primary rather than secondary material, - that you read
Frege's own papera before commentators on Frege, for cumple; you will
find that aecondary material is generally much more useful if you've
already sorne acquaintance with the work on which it's baaed.
254 Advice on reading

Sorne suggestions about where to begin your reading on topics díscussed

in the present book:
Chapter
I On the airns of formalism: Frege 1882a, b.
On the scope of logic: Kneale 1956; Quine 1970 ch. 5.
2 On induction and deduction: Skyrms 1966 ch. l.
On logica utens and logica docens : Peirce, 'Why study logic?' in 1930-58
vol. 2, especially 2, 185ff.
On validityand logical form: Cargile 1970; Davidson 1970; Harman 1970.
J On 'tonk': Prior 1960, 1964; Belnap 1961; Stevenson 1961.
On 'if' and '-+': Faris 1962.
4 On the development of quantifiers: Frege 1891,
On substitutional versus objectual interpretations: Belnap and Dunn 1968.
On non-standard treatments of quantifiers: Montague 1973.
5 Frege 1892a; Russell 1905; Strawson 1950; Quine, 'On what there is' in
1953a; Kripke 1972,,
6 Frege 1918 (and cf. Popper,' Epistemology without a knowing subject' in
1972); Quine 1970 ch. 1; Putnam 1971; Lemmon 1966.
7 Definitions versus criteria: Rescher 1973 chs. 1 and 2.
Correspondence theories: Russell 1918; Austin 1950; Prior in Edwards
1967,
Coherence theories: Bradley 1914; Hempel 1935; Rescher 1973.
Pragmatist theories: Peirce 1877; James 1907; Dewey 1901 ; Rescher 1977
ch. 4.
The semantic theory: Tarski 1944 (and cf. Quine 1970 ch. 3, Rogers
1963); Popper, 'Truth, rationality and the growth of scientific know-
ledge' (in 1963); 'Philosophical comments on Tarski's theory of truth'
(in 1972); Davidson 1967.
The redundancy theory: Ramsey 1927; Prior 1971; Grover et al. 1975.
8 Russell 19o8a; Mackie 1973 ch. 7; Kripke 1975.
9 On temporal logic: Quine 196oa §36; Prior 1957, 1967; Lacey 1971;
Geach 1965,
On fuzzy logic: Zadeh 1975; Gaines 1976.
10 On necessary truth; Quine 1951.
Formal presentation of modal logícs: Hughes and Cresswell 1968.
Philosophical issues: Quine 1953b; Linsky 1971; Plantinga 1974.
u Rescher 1969; Haack 1974.
12 On metaphysical questions: van Heijenhoort 1967b; Rescher 1977
chs. 13, 14.
On epistemological questions: Quine 1951; Putnam 1969; Popper 1970.
On psychologism: Mili 1843 book n; Frege (on mathematics) 1884
eapecially §26; (on logic) 1918; Peirce 1930-58, 3, 161ff.; Russell 1938.
 BIBLIOGRAPHY

• by an entry indicates to which edition page numbers in the text refer.

Ackennan, R. 1967. Introduction to Many-Valued Logics (Routledge and

Kegan Paul)
Alston, W. P. 1958. Ontological commitments, Philosophical Studies 9; and in
lseminger 1968
Altham, J. E. J. 1971. The Logic of Plurality (Methuen)
Anderson, A. R. 1970. St Paul's epistle to T'itus, in The Paradox of the Liar,
ed. Martin (Yale U.P.)
Anderson, A. R. and Belnap, N. D., Jr 1962a. Tautological entailments,
Pbilosophical Studies 13
- 1962b. The pure calculus of entailment, Journal of Symbolic Logic, 27 (1962a
and b are partly reprinted in Iseminger 1968)
- 1975. Entailment, vol. 1 (Princeton U.P.)
Anscombe, G. E.M. 1957. Analym puzzle, 10, Analysis 17
- 1959. An Introduction to Wittgemtein's Tractatus (Hutchinson)
Au1tin, J. L. 1950. Truth, Proceeding, of the Aristotelian Society, Supplnnent 24;
and in Urmson and Wamock (eds.) 1961; and Pitcher 1964
Ayer, A. J. 19.58. The Problem of KnOflJledge (Macmillan)
Baker, A. J. 1967. 'If' and '::> ', Mind 76
Bar-Hillel, Y. 1957. New light on the Liar, Analym 18
Barker, S. F. 1965. Must every inference be either inductive or deductive?, in
Philosophy in America, ed. Black (Afien and Unwin)
Belnap, N. D., Jr 1961. Tonk, plonk and plink, Analym 22; and in Strawaon 1967
- 1974- Grammatical Propaedeutic, in The Logical Enterprúe, ed. Anderson,
Marcus and Martin (Yale U.P.); and in Anderson and Belnap 1975
Belnap, N. D., Jr and Dunn, M. 1968. The substítution interpretation of the
quantifiers, Noib 2
Belnap, N. D., Jr and Grover, D. L. 1973. Quantifying in and out of quotea, in
Truth, SyntWl and Modality, ed. Leblanc (North Holland)
Bennett, J. 1954. Meaning and implication, Mind 63
Bergmann, G. 1g6o. The philosophícal significance of modal logic, Mind 69
Binkley, R. 1970. Quantifying, quotation, anda paradox, No(b 4
Birkhoff, G. and van Neumann, J. 1936. The logic of quantum mechanicl,
Annals of Mathematics 37
256 Bibliography
Black, M. 1937. 'Vagueness', Philosophy of Science 4; and in Language and
Philosophy (Cornell U.P. 1949)
- 1948. The semantic definition of truth, Analysis 8; and in MacDonald 1954. •
Blanché, R. 1962. Axiomatics, trans. Keene (Routledge and Kegan Paul)
Blanshard, B. 1939. The Nature of Thought (Allen and Unwin)
Blumberg, A. 1967. Logíc, modern, in Edwards 1967
Bochvar, D. A. 1939. On a three-valued logical calculus and its app}ication to the
analysis of contradictories, Matematiceskij sbornik 4
de Bono, E. 1969. The Mechanism of Mind (Cape; Pelican 1971)
Boole, G. 1854. An Inuestigation of the Lauis of Thought (Dover 1951)
Boolos, G. S. and [effrey, R. C. 1974. Computabilitv and Logic (Cambridge U.P.)
Bradley, F. H. 1914. Essays on Truth and Reality (Oxford U,P.)
Brandom, R. 1976. Truth and assertibiliry, Journal of Phílosophy 73
Brouwer, L. E. J. 1952. Historical background, principles and methods of
Intuitionism, South African Journal of Science 49
Burge, T. 1973. Reference and proper names, Journal of Philosophy 70
Bynurn, T. Ward (ed.) 1972. Conceptual Notation (Oxford U.P.)
Cargile, J. 1970. Davidson's notion of logical form, Inquiry 13
- 1972. On the interpretation of T, S4 and S5, Philosophia 2
Carnap, R. 1931. The logicist foundations of mathematics, in Phílosophy of
Mathematics, ed, Putnam and Benacerraf (Blackwell 1964)
- 1934, Logische Syntax der Sprache (Julius Springer); English translation,
The Logical Syntax of Language (Kegan Paul 1937)
- 1942. Introduction to Semantics and Formalisation of Logic (Harvard U.P.)
- 1947. Meaníng and Necessity (Chicago U.P.)
- 1950. The Logícal Foundations of Probability (Chicago U.P.)
Cartwright, R. 1954. Ontology and the theory of meaning, Phílosophy of Srience 4
- 1960, Negative existentials, Journal of Philosophy 56; and in Caton 1963
- 1962. Propositions, in Analytical Philosophy, vol. 1, ed. Butler (Blackwell)
Caton, C. (ed.) 1963. Philosophy and Ürdinary Languag« (Illinois U.P.)
Chihara, C. 1972, Russell's theory oftypea, in Bertrand Russell, ed. Pean (Anchor)
- 1973, Onto/ogy and the Vicious-Circle Principie (Cornell U.P.)
Chisholrn, R. 1967. ldentity through possible worlds, Noib I
Chomsky, N. 1957. Syntactic Structures (Mouton)
Church, A. 1943. Review of Quine, W. V. O., 1943, Journal of Symbo/ic Logíc 8
Clark, M. 1971. Ifs and hooks, Analysis 30
Cohen, L. J. 1962. The Diversíty of Meaníng (Methuen)
Copi, l. 1971. The Theory of Logica/ Types (Routledge and Kegan Paul)
Copi, l. and Gould, J. A. (eds.) 1967. Contemporary Readings in Logícal Theory
(Macmillan)
Curry, H. 1951. Outline of a Formalíst Philosophy of Mathematics (North Holland)
Dauer, F. 1974. In defence of the coherence theory of truth, Journal of
Phílosophy 71
Davidson, D. 1967. T'ruth and meaning, Synthese 17; and in Davis, Hockney
and Wilson 1969
- 1968a. The logical form of action sentences, in The Logíc of Decision and
Action, ed. Rescher (Pittsburgh U.P.)
- 1g68b. On saying that, Synthese 19
- 1970. Action and reaction, Inquiry 13
- 1973, In defense of convention T, in Truth, Syntax and Modality, ed. Leblanc
(North Holland)
 Bibliography
257
- 1974. Belief and the basis of meaning, Synthese 27
Davidson, D, and Hintikka, J. (eds.), 1969, Words and Objectiom: Essay, on the
Work of W. V. Quine (Reidel)
Davis, J. W., Hockney, D. J. and Wilson, W. K. (eds.) 1969, Philosophical I.ogic
(Reidel)
Derrida, J. 1973. Speech and Phenomena, ed, Garver (Northweatern U.P.)
Destouches-Février, P. 1951. La structure des théories physiques (Presees
Universitaires de France)
Dewey, J. 1901. A short catechism conceming truth, in The Influence of Darunn
on Philosophy (Henry Holt)
- 1929. Experience and Nature (Open Court; and Dover 1958•)
- 1938. Logic, the Theo,y of Inqui,y (Henry Holt)
Donellan, K. 1966. Reference and definite description1, Philosophical Review 75;
reprinted in Steinberg and Iakobovíts 1971
Duhem, P. 1904. La théorie phynque: son objet, sa structure; page referenees to
The Aim and Structure of Physical Theory, trans. Wiener from the and edition
(1914) (Atbeneum 1962)
Dummett, M. A. E, 1959. Truth, Proceedings of the Aristotelian Society 59; and
in Pitcher 1964
- 1973. Frege, Philosophy of Language (Duckworth)
Duncan-Jones, A. E. 1935. Is strict implication the same aa entailment? Analysis 2
Edwards, P. (ed.) 1967. Encyclopaedia of Philcsophy (Collier-Macmillan)
Faris, J. A. 1962. Truth-Functional Logic (Routledge and Kegan Paul); appendix
on 'if' reprinted in Iserninger 1968
Feigl, H. and Sellars, W. 1949. Readings in Philosophical Analysis (Appleton-
Century-Crofts)
Feyerabend, P. K. 1963. How to be a good ernpirícist, in Delauor« Seminar in
the Philosophy of Science, vol. 2 (lnterscience); and in Nidditch 1968
Field, H. 1972. Tarski's theory of truth, Journal of Philosophy 69
Fitch, F. 1949, The Problern of the Moming Star and the Evening Star,
Philosophy of Science 17; and in Copi and Gould 1967
Fitting, M. C. 1969. Intuitionistic Logic, Motkl Theory and Forcing (North Holland)
Flew, A. G. N. (ed.) 1956. Essay« in Conceptual Analysis (Macmillan)
- 1975. Thinking About Thinki,rg (Fontana)
Fellesdal, D. 1965. Quantification into causal contexts, Boston Studies in the
Phi/o,ophy of Science, vol. 2, ed, Cohen and Wartofsky (Reidel); and in Linsky
1971
- 1969. Quine on modality, in Davidson and Hintikka 1969
van Frassen, B. C. 1966. Singular terma, truth-value gap1, and free logic, Journal
of Philosophy 63
- 1968. Presupposition, implication and self-reference, Journal of Philc,ophy 65
- 1969. Presuppositiona, aupervaluationa and free logic, in The Logical Way of
Doi,rg Tlnº,rgs, ed. Lambert (Vale U.P.)
Frege, G. 1879, Begn.ffsschrift (Nebert); English translationa in Heijenhoort
1967a, and in Bynum 1972
- 1882&. Uber den wiasenschaftliche Berechtigung einer Begriffuchrift (On the
scientific juatification of a conceptual notation), Zeitlchrift Jfir Philo,ophú und
plu'/osoplrüche Kritik, 81; Engliah tranalationa by Bartlett, Mind 73, 11}64; and
Bynum 1972
1882b. Uber den Zweck der Begrifmchrift (On the auna of the conceptual
notation), Sitnngberiehte dn JttlllUchen Gestllschaft Jür Mtdism und
258 Bibliography
Naturunssenschaft 16, 1882-3; English translations by Dudman, Australasian
Journal of Philo1ophy 46, 1968; and Bynum 1972
- 1884. Die Grundlagen der Arithmetik (Koebner 1884); English translation by
Austin, The Foundations of Arithmetic (Blackwell 1950)
- 1891. Function and Concept, Jenaische Gesel/1chaftfür Medizin und
Naturunssenschaft; 1891; trans. Geach in Geach and Black 1952
1892a. Über Sinn und Bedeutung, Zeiuchrift Jür Philosophie und phi/01ophische
Kritik, 100; trans. Feigl in Feigl and Sellara 1949; and reprinted in Copí and
Gould 1967; trans, Black in Geach and Black 1952
- 1892b. Concept and object, Vierteljahruchri/t fiir Wiuenschaftliche Philosophie
16; trans. Geach in Geach and Black 1952
- 1893. Grundgeutze der Arithmetik 1 (H. Pole); partía} English translation in
Furth The Bum: Lauu of Arithmetic (California U.P. 1964)
- 1903. Grundge1etze der Arithmetik 2 (H. Pole); partial English translation in
Geach and Black 1952; the Geach and Black selection includes Frege's
response to Russell's paradox
- 1918. The thought: a logical inquiry, Beitrage sur Philosophie der Deutschen
Idealismus; trans. Quinton and Quinton Mind 65, 1956; reprinted in Klemke
1968•
Gaines, B. 1976. Foundations of fuzzy reasoning, International Journal of
Man-Machine Studje, 8
Garver, N. 1970. The range of truth and falsity, in The Parados of the Liar,
ed. Martín (Yale U.P.)
Geach, P. T. 1956. On Frege's way out, Mind 65
- 1962. Reference and Generality (Cornell U.P.)
- 1965. Sorne problema about time, Proceedings of the Britisb Academy 51
- 1967. lntentional identity, Journal of Philosophy 64
- 1976. Reason and Argument (Blackwell)
Geach, P. T. and Black, M. 1952. Translations [rom the Philosophícal Writing, of
Gottlob Frege (Blackwell)
Gentzen, G. 1934. Untersuchungen Uber das logische Schliessen, Mathematische
Zeit1chrift 39; English translation by Szabo, American Philosophical ()uarterly 1,
1964 and 2, 1965
Gochet, P. 1972. E1qui11e d'une théorie nominaliste de la propo1ition (Librairíe
Armand Colín)
Goodman, N. 1955. Fact, Fiction and Foreeast (Harvard U.P.)
- 1970. Seven stríctures on simílarity, in Experience and Theory, ed, Foster and
Swanson (Massachussets U.P.)
Grice, P. and Strawson, P. F. 1956. In defense of a dogma, Phi/o,ophical
RetJiew 65
Grover, D. L. 1972. Proposítional quantiñers, Journal of Philosophical Logic I
- 1973. Propositional quantification and quotation contexts, in Truth, Syntax
and Modality, ed. Leblanc (North Holland)
Grover, D. L., Camp, J. and Belnap, N. D., Jr 1975. A prosentential theory of
rruth, Philosophical Studies 27
Haack, R. J. 1971. On Davidson's paratactic theory of oblíque contexts, No{u s
Haack, R. J. and S. 1970. Token sentences, translation, and truth-value, Mind 79
Haack, S. 1974. Deviant Logic (Cambridge U.P.)
- 1975. Mentíoning expreselons, Logique et Analyu 67-8
- 1976a. The justification of deduction, Mind 85
- 1976b. Critica! Notíce of Wooda, J., 1974, Canadian Journal of Philo,oplty 6
 Bibliography
259
- I 9?6c, The pragmatist theory of truth, British Journal for the Philosophy
oj
Science 27
- 1976d, Is it_true what they say about Tarski?, Philosophy 5
1
- 1977a. Lew1s's ontological slum, Reuiew of Mttaphysics 30
- 1977b. Two fallibilists in search of the truth, Proceedings of the Aristotelian
Society Supplement 51
- 1977c, Analytic(ty and logical truth in The Roots of Reference, Theoria 43
- I 97 + , Pragmarism and ontology: Peirce and James, forthcoming in Reoue
Internationale de Philosophie
Hacking, l. M. I 97 +, What is logic? forthcoming in Journa/ of Philosophy
Halldén, S. 1949. The Logic of Nonsense (Uppsala Universitets Arsskrift)
Harman, G. 1970. Deep structure as logieal form, Synthes« 21
- 1971. Substitutional quantification and quotation, Noüs 5
- 1975. Moral relativism defended, Philosophical Reuiew 84
Harris, J. H. 1974. Popper's definitions of verisimilitude, British Journa/ [or the
Phi/osophy of Science 25
Heidelberger, H. 1968. The indispensability of truth, American Philosophica/
Quarter/y 5
van Heijenhoort, J. (ed.) 1967a. From Frege to Góde/ (Harvard U.P.)
- 1967b. Logic as calculus and logic as language, Syntheu 17
Hernpel, C. G. 1935. On the logical positivists' theory of truth, Ana/ysis 2
Herzberger, H. 1970. Paradoxes of grounding in semantics, Journa/ of
Phi/osophy 17
Heyting, A. 1966. Intuitionism (North Holland)
Hintikka, J. 1962. Knowledge and fü/ief (Comell U.P.)
- 1969. Model» for Moda/ities (Reidel)
- 1973. Logic, Language Games and Information (Oxford U.P.)
- 1976. Quantifiers in logic and quantifiers in natural language, in Komer 1976
Hughes, G. and Cresswell, M. 1968. Introduction to Modal Logic (Methuen)
Hunter, G. 1971. Metalogic (Macmillan)
Iseminger, G. (ed.) 1968. Logic and Philosophy (Appleton-Century-Crofts)
James, W. 1907. Pragmatism (Longman's, Green)
- 1909. The Meaning of Trutb (Longman's, Green; and Miclúgan U.P. 1970•)
Jdkowski, S. 1934. On the rules of supposition in formal logic, Studia Logica I
Johannson, l. 1936. Der Mininmalkalkül, ein reduzierter intuitionistischer
Formalismus, Compositio Mathematica 4
Johnson, W. E. 1921. Logic (Cambridge U.P.)
Kanger, S. 1957a. The Moming Star paradox, Theoria 23
- 1957b. A note on quantification and modalities, Theoria 23
Kant, l. 1800, Logik; trans. Abbott Kant's Introduction to Logic (Longman's,
Green 1885•)
Kleene, S. C. 1952. Introduction to Metamathm,atics (North Holland)
Klemke, E. D. (ed.) 1968. Essays on Frege (Illinois U.P.)
Kneale, W. C. 1945-6. Truths of logic, Proceeding« of the Aristotelian Socitty 46
- 1956. The province of logic, in Contemporary Britisb Philosophy, ed. Lewis
3rd Series (Allen and Unwin); page references to and, 1961 edition
- 1962a. Universality and necessity, British fournal for the Philosophy of Scienc« 12
- 1962b. Modality, de dicto and de re, in Logic, Methotwlogy and Phiú,,ophy of
Science: Proceeding« of the 196o Intemational Congres« (Stanford U.P.)
- 1971. Russell's paradox and sorne others, British Journal for the Philosoplry of
Science 22
Kneale, W. C. and Kneale, M. 1962. The Dtveloptnent of Logic (Oxford U.P.)
260 Bibliography
Korner, S. 1966. Experience and Theory (Routledgc and Kegan Paul)
- (ed.) 1976. Philosophy of Logic (Blackwell)
Korarbiñaki, T. 1955. The fundamental ideas of pansomatism, Mind 64
Kripke, S. 1963. Semantical considerations on modal logic, Acta Philosophica
Fennica 16; and in Linsky 1971.
- 1972. Naming and necessity, in Semantics of Natural Language, ed, Harman
and Davidson (Reidel)
- 1975. Outlíne of a theory of truth, Journal of Philosophy 72
- 1976. Is there a problem about substitutional quantification?, in Truth and
Meaning; ed, Evans and McDowell (Oxford U.P.)
Lacey, H. 1971. Quine on the logic and ontology of time, Australasian Journal
of Philosophy 49
Lakatos, l. 1963-4. Proofs and refurations, Britisb Journal for the Philosophy of
Science 14; and edited by Worrall and Zahar (Cambridge U.P. 1976)
Lemmon, E. J. 1959. Is there only one correct system of modal logic?, Proceedings
of the Aristotelian Society Supplernent 23
- 1965. Beginning Logic (Nelson)
- 1966. Sentences, statements and propoaitíons, in British Analytical Philosophy,
ed, Williams and Montefíore (Routledge and Kegan Paul)
Lewis, C. l. 1912. Implication and the algebra of logic, Mind 21
- 1918. A Survey of Symbolic Logic (California U.P.)
Lewis, D. K. 1968. Counterpart theory and quantified modal logic, Journal of
Philosophy 6 5
- 1973. Counterfactuals (Blackwell)
Lewy, C. 1946. How are the calculuses of logic and mathernatics applicable to
reality?, Proceedings of the Aristotelian Society Supplernent 20
Linsky, L. (ed.) 1952. Semantics and the Philosophy of Language (Illinois U.P.)
- (ed.) 1971. Reference and Modality (Oxford U.P.)
- 1972. Two concepta of quantification, Noés 6
Lukasiewicz, J. 1920. On j-valued logic, in McCall, Polisb Logic (Oxford U.P. 1967)
- 1930. Many-valued systems of propositional logic, in McCall, Polisb Logic
(Oxford U.P. 1967)
McCall, S. 1970. A non-classical theory of truth, with an application to
Intuitionism, American Philosophical Quarterly 7
MacColl, H. 188off. Symbolic reasoning, 1-8, Mind S, new series 6, 9, l 1, 12,
14 (two papers), 15
- 1906. Symbolic Logic and Its Applicatio,u (Longman's, Green)
MacDonald, M. (ed.) 1954. Philosophy and Analysis (Blackwell)
McDowell, J. 1977. On the sense and reference of a proper name, Mind 86
Mackie, J. L. 1973. Truth, Probability and Parado» (Oxford U.P.)
McKinsey, J. C. C. and Tarski, A. 1948. Sorne theorems about the sentential
calculus of Lewis and Heyting, Journal of Syml,o/ic Logic 13
MacTaggart, J. M. E. 1908. The unreality of time, Mind 17; and in Philosophical
Studies (Amold 1934)
Marcus, R. Barcan 1946. A functional calculus of first order based on strict
implication, Journal of Symbolic Logic 11
- 1962. Modalities and intensional languages, Synthese 27; and in Copi and
Gould 1967
- 1967. Essentialism in modal logic, Notü 1
- 1972. Quantification and ontology, Notü 6
Martín, R. L. (ed.) 1970. The Paradox of the Liar (Yale U.P.)
 Bibliography

Massey, G. 1974. Are there any good argumenta why bad argumenta are bad]
Philosophy in Context 4 '
Meinong, A. 1904. The theory of objecta, in Unter,uchungen sur Gegenstandt-
theorie und Psychologie; trans. Levi, Terrel and Chisholm in Iseminger 1968
Mellor, D. H. 1974. In defense of dispositions, Philosophical Rf!f!itW 82
Mendelson, E. 1964. Introduction to Mathematical Logic (van Nostrand)
Michalski, R. S., Chilansky, R. and Jacobsen, B. 1976. An application of
variable-valued logic to inductive learning of plant disease diagnostic rules,
Proceedings of the Sixth International Symposium on Multiple-Valued Logic
(Utah State University)
Mili, J. S. 1843. A System of Logic (Longman's)
Miller, D. W. 1974. Popper's qualitative theory of verisimilitude, British Journal
for the Phi/osophy of Science 25
Molnar, G. 1969. Kneale's argument revisited, Phi/osophica/ Reoie» 77
Montague, R. 1963. Syntactical treatments of modslity, Acta Philosophica
Fennica 16; and in Thomason 1974
- 1973. The proper treatment of quantification in ordinary English, in
Approaches to Natural Language, ed. Hintikka, Moravcsik and Suppes (Reidel);
and in Thomason 1974
Moore, G. E. 1908. Professor James' 'Pragmatism', Proceedings of the Aristotelian
Society 8; and in Moore 1922
- 1922. Philosophica/ Studies (Routledge 1922, 1900)
- 1952. Commonplace Book, ed. Lewy (Allen and Unwin)
- 1953. Some Main Problems of Philosophy (Macmillan)
Nagel, E. and Newman, J. R. 1959. Gódel's Proof (Routledge and Kegan Paul)
Nelson, E. J. 1933. On three logical principies in intensión, Monist 43
- 1946. Contradiction and the presupposition of existence, Mind 55
Nerlich, G. 196 5. Presupposition and entailment, American Philosophical Quarterly 2
Ness, A. 1938. 'Truth' as conceived by those who are not professional
philosophers, Skrifter utgitt av Der Norske Videnkaps-Akademi; Os/o 11:
Hist-Filos Klasse, 4
Neurath, O. 1932. Protocol Sentences, in Logical Positivism, ed. Ayer (Free
Press 1959)
Nidditch, P. H. (ed.) 1968. Philosophy of Science (Oxford U.P.)
O'Connor, D. J. 1975. The Carrespondence Theory of Trutñ (Hutchinson)
Parsons, T. 1969. Essentialism and quantified modal logic, Philosophical Review
77; and in Linsky 1971
- 1974. A prolegomenon to Meinongian semantics, Journal of Phi/osophy 71
Pears, D. F. (ed.) 1972. Bertrand Russell (Anchor)
Peirce, C. S. 1868. Questions conceming certain faculties claimed forman,
Journal of Speculatioe Philosophy 2; and in Peirce 1930-58, 5, 213ff.
- 1877. The fixation of belief, Popular Science Monthly 12; and in Peirce
1930-58, 5, 358ff.
- 1885. On the algebra of logic, American Jo11rnal of Mathematics 7; and in
Peirce 1930-58, 3, 210-38
- 1930-58. Collected Papers, ed, Hartahome, Weiss and Burla (Harvard U.P.)
(Reference to the collected papers are by volume and paragraph no., a•,
e.g. 3, 117 = volume 3, paragraph 117.)
Pitcher, G. (ed.) 1964. Truth (Prentice-Hall)
Plantinga, A. 1974. The Nature of N«essity (Oxford U.P.)
Popper, K. R. 1947. New foundatiom for logic, Mind 56
- 1954. Self-reference and meaning in ordinary language, Mmd 63
262 Bibliography
- 1960. On the sources of knowledge and ignorance, Proceedings of the British
Academy 46; and in Popper 1963•
- 1961. The Open Society and Its Enemies (1945), vol. 2, znd edition (Routledge
and Kegan Paul)
- 1963. Conjectures and Refutations (Routledge and Kegan Paul)
- 1970. A realist view of physics, logic and history, in Physics, Logic and History,
ed. Yourgrau and Breck (Plenum)
- 1972. Objectioe Knowledge (Oxford U.P.)
- 1976. Unended Quest (Fontana)
Post, E. 1921. Introduction to the general theory of elementary propositions,
American Journal of Mathematics 43; and in Heijenoort 1g67
Prawitz, D. 1965. Natural Deduction (Almqvist and Wiksell)
Priest, G. G. 1976. Modalíty as a rneta-concept, Notre Dame Journal of Formal
Logic 17
Prior, A. N. 1955. Formal Logi'c (Oxford U.P.)
- 1957. Time and Modality (Oxford U.P.)
- 1958. Epimenides the Cretan, Journal of Symbolic Logíc 23
- 1960. The runabout inference ticket, Analysis 21; and in Strawson 1967•
- 1964. Conjunction and contonktion revisited, Analysis 24
- 1967. Past, Present and Future (Oxford U.P.)
- 1968. Papers on Time and Tense (Oxford U.P.)
- 1971. Objects of Tbought, ed. Geach and Kenny (Oxford U.P.)
Putnarn, H. 1957. Three-valued logic, Philosophical Studies 8
- 1g69. Is logic empirical?, in Boston Studies in the Philosophy of Science 5,
ed. Cohen and Wartofsky (Reidel)
- 1971. Philosophy of Logic (Harper Torchbooks)
Quine, W. V. O. 1934. Ontological remarks on the propositional calculus,
Mind 43; and in Quine 1966a
- 1940. Mathematical Logic (Harper and Row)
- 1943. Notes on existence and necessity, Journal of Philosophy 40; and in
Linsky 1952
- 1947. The problern of interpreting modal logic, Jo11rnal of Symbolic Logic u;
and in Copi and Gould I g67
- 1950. Method, qf Logic (Holt, Rinehart and Winaton); third edition 1974
(Routledge and Kegan Paul)
- 1951. Two dogmas of empirícism, Philosophical Reví� 60; and in Quine 1953a•
- 1953a. From a Logical Point of Vi� (Harper Torchbooks)
- 1953b. Three grades of modal involvement, Proceedings of the Xlth
International Congress of Philosophy 14; and in Quine 1966a
1953c. Review of Strawson, P. F., 1952, Mind 62; and in Copi and Gould 1967
1955. On Frege's way out, Mind 64
- 196oa. Word and Object (Wíley)
- 196ob. Variables explained away, Proceedings of American Philosophical Society;
and in Selected Logic Papers (Random House 1g66)
- 1966a. Ways of Paradox (Random House)
- 1g66b. Russell's ontological development, Journal of Philosophy 63; and in
Pears (ed.) 1972
- 1g68. Ontological relativíty, Journal of Philosophy 65; and in Ontological
Relativity (Columbia U.P. 1g69)
- 1969. Replies, in Wordt and Objections, ed. Davidson and Hintikka (Reidel)
- 1970. Ph,1osophy of Logic (Prentice-Hall)
- 1973. The Roots of Reference (Open Coun)
 Bibliography 263
- 1976. Worlds away, Journal of Philosophy 73
Quine, W. V. O. and Ullian, J. 1970. The Web of Belief (Random House)
Ramsey, F. P. 1925. The foundations of mathematica, Proceedings of th« London
Mathematical Society, Series 2, 25; and in Ramsey 1931
- 1927. Facts and propositions, Proceedings of the Aristotelian Society
Supplement 7; and in Ramsey 1931•; and, in part, in Pitcher 1964.
- 1931. The Foundations of Mathematics (Routledge and Kegan Paul)
Reichenbach, H. 1944. Philosophic Foundation of Quantum M«hanics
(California U.P.)
Rescher, N. 1969. Many-Valued Logic (McGraw-Hill)
- 1973. The Coherence Theory of Truth (Oxford U.P.)
- 1974. Bertrand Russell and modal logic, in Studies in Modality (American
Philosophical Quarterly Monograph)
- 1975. A Theory of Possibility (Blackwell)
- 1977. Methodological Pragmatism (Blackwell)
Richards, T. 1975. The worlds of David Lewis, Australasian Journal of
Plu"losophy 53
Rogers, R. 1963. A survey of formal semantics, Synthese 15
Ross, A. 1968. Directioes and Norms (Routledge and Kegan Paul)
Rosser, J. B. and Turquette, A. R. 1952. Many-Valued Logias (North Holland)
Routley, R. 1963. Sorne things do not exist, Notre DameJournal of Formal �e 7
- 1966. On a significance theory, Australasian Journal of Philosophy 44
- 1969. The need for nonsense, Australasian Journal of Philosophy 47
Runes, D. B. 1966. Dictionary of Philosophy (Littlefield, Adams)
Russell, B. 1903. The Principies of Mathematics (Allen and Unwin)
- 1905. On Denoting, Mind 14
- 1906. Review of Mc'Coll, 19()6, Mind 15
- 1908a. Mathematical logic as based on the theory of types, American Journal
of Mathematics 30; and in Russell 1956•; and in Heijenhoort 1967a
- 1908b. James's conception of truth, Albany Revino; and in Ruasen 1910
- 1910. Philosophical Essays (Allen and Unwin 1910, 1916)
- 1918. The philosophy of logical atomism, in Ruasen 1956
- 1919. On propositions: what they are and how they mean, in Ruasen 1956
- 1923. Vagueness, Australasian Journal of Philosophy and P,ychology I
- 1938. The relevance of psychology to logic, Proceedings of th« Aristotelüm
Society Supplement 17
- 1956. Logic and Knowledge, ed. Marsh (Allen and Unwin)
- 1959. My Philosophical Dl!fJtlopment (Allen and Unwin); § on Strawaon's
criticisms of the theory of descriptions reprinted in Copi and Gould 1967
Russell, B. and Whitehead, A. N. 1910. Principia. Mathematica (Cambridge U.P.)
Russen, L. J. 1970. 'If' and '=> ', Mind 79
Ryle, G. 1949. The Concept of Mind (Hutchinson)
- 1952. Heterologicality, Analysis 11; and in MacDonald 1954
- 1954. Formal and informal logic, in Dilemmas (Cambridge U.P.)
Salmon, W. 1967. Foundations of Scientific Inference (Pittaburgh U.P.)
Scheffier, I. 1954. An inscriptionalíat approach to indirect quotation, Analyns r4
- 1967. Science and SubjectffJity (Bobbs-Merrill)
Scheffier, I. and Chonuky, N. 1958. What is aaid to be, Proceerim,1 of tire
Aristotelian Socitty 59
Schiller, F. C. S. 1912. Formal Logic: a Scientific and Social Problem (Macnullan)
- 1930. Logic far Use (Harcourt, Brace)
264 Bib/iography
Schlick, M. 1934. The foundation of knowledge, trans. Rynin in Logical
Positioism, ed. Ayer (Free Press 1959)
Schock, R. 1968. Logics Without Existence Asmmptions (Almqvist and Wiksell)
Searle, J. 1969. Speech Acts: An Essay on the Philosophy of Language (Cambridge
U.P.); pp. 162-'74 reprinted in Steinberg and Jakobovits 1971•
- 1975. The logical status of fictional discourse, New Literary History 6
Sellara, W. 1967. Science, Perception and Reality (Routledge and Kegan Paul)
Skyrms, B. 1966. Choice and Chance (Dickenson)
- 197oa. Return of the Liar: three-valued logic and the concept of truth,
American Philosophical Quarterly 7
- 197ob. Notes on quantíficatíon and self-reference, in Martin 1970
Smíley, T. J. 1959. Entailment and deducibility, Proceedings of the Aristotelian
Society 59
- 1g60. Sense without denotation, Analysis 20
- 1g63. The logical basis of ethics, Acta Philosophica Fennica 16
Smullyan, A. 1948, Modality and description, Journal of Symbolic Logic 13
Smullyan, R.M. 1957. Languages in which selt-reference is possible, Journal of
Symbo/ic Logic 22
Stalnaker, R. 1968. A theory of conditionals, in Studies in Logical Theory,
ed, Rescher (Blackwell)
Stebbíng, S. 1939. Thinking to Some Purpose (Penguin)
Steinberg, D. D. and Jakobovits, L. A. (eds.) 1971, Semantics (Cambridge U.P.)
Stevenscn, J. T. 1g61. Roundabout the runabout inference ticket, Analysis 22
Strawson, P. F. 1949. Truth, Analysis 9; and in MacDonald 1954
- 1950, On referríng, Mind 59; reprinted in Flew 1956; in Copi and Gould
1g67; and in Strawson 1971
- 1952. Introduction to Logical Theory (Methuen)
- 1954. Reply to Mr. Sellara, Philosophical Review 63
- 1957. Propositions, concepta and logical truth, Philosophical Q11arurly 7;
in Strawson 1g67; and in Strawaon 1971
- 1959. Indioiduals (Methuen)
- 1961. Singular terms and predication, Journal of Phi/osophy 57; and in
Strawson 1971
- 1964. ldentifying reference and truth-valuea, Theoria 30; reprinted in
Steinberg and Jakobovíts 1971
- (ed.) 1967. Philosopluca/ Logic (Oxford U.P.)
- 1971. Logico-Linguistic Papers (Methuen)
Suppes, P. 1957. Introduction to Logic (van Nostrand)
Tarskí, A. 1931. The concept of truth in formalised languages, in Tarski 1956
- 1936. The establishment of scientific semantics, in Tarski 1956
- 1944. The semantic conception of truth, Philosophy and Phenomenological
Research 4, and in Feigl and Sellars 1949•
- 1956. Logic, Semontics and Metamathemauc«, trans. Woodger (Oxford U.P.)
Tarski A. and Vaught, R. L. 1957. Arithmetical extensions of relational systems,
Compositio Mathematica 13
Thomason, R. (ed.) 1974. Formal Phi/osophy (papera of Richard Montague)
(Yale U.P.)
Thouless, R. H. 1930, Straight and Croohed Thinking (Hodder and Stoughton,
1930; Pan, 1953)
'Tiehy, P. 1974. On Popper's definitíons of verisimílitude, British Journal for
the Philosophy of Science 25
Toulrnin, S. 1953. Philosophy of Science (Hutchinson}
 Bibliography
Urmson, J. O. and Wamock, G. J. (eds.] 1961. Philosophical Paj,n'J of
J. L. Austin (Oxford U.P.)
Wallace, J. 1970. On the frame of reference, Synthese zz
- 1971. Convention T and substitutional quantification, Nods 5
- 1972. Positive, comparative, superlatíve, Journa/ of Phüosophy 69
Weinstein, S. 1974. Truth and dernonstratives, Noés 8
Weston, T. S. 1974. Theories whose quantification cannot be subtltiturional,
Nolh 8
White, A. R. 1970. Truth (Anchor)
White, M. G. 1950. The analytic and the synthetíc, in Hook, S. (ed.), John
Dewey, Philosopher of Science and Freedom (Dial)
- 1956. Touiard Reunion in Phi/osophy (Harvard U.P.)
Whitehead, A. N. 1919. The Concept of Nature (Cambridge U.P.)
Williams, C. J. F. 1976. What is Truth? (Cambridge U.P.)
Wittgenstein, L. 1922. Tractatus Logico-Philosophicus, trans, Ogden (Routledge
and Kegan Paul); and trans. Pears and McGuinness (Routledge and Kegan
Paul, 1961)
- 1953. Philosophica/ Inuestigations, trans. Anscombe (Blackwell)
Wolf, R. 1977. Are relevant logics deviant?, Philosophia 7
Woodruff, P. 1970. Logic and truth-value gaps, in Philosophical Problems in Lcgic,
ed. Lambert (Reidel)
Woods, J. 1974. The Logic of Fiction (Mouton)
von Wright, G. H. 1957. Logica/ Studies (Routledge and Kegan Paul)
- 1963. Norm and Action (Routledge and Kegan Paul)
Zadeh, L. A. 1963. On the definition of adaptivíty, Proceedings of the Institute of
E/ectrica/ and Electronic Engineer, 51
- 1964. The concept of state in system theory, in Trendt in General Syltems
Theory; ed. Mesarovic (Wlley)
- 1965. Fuzzy sets, lnformation and Control 8
- 1972.. Fuszy languages and theír relation to human intelligence ; Proceedings
of the Intemational Co,iftrence, Man and ComputtT (S. Karger)
- 1975. Fuzzy logic and approximate reasoning, Synthe1e 30
- 1976. Semantic inference from ÍU%Zy premíses, Proctedings of the Sixth
Inttrnational Sympo1ium on Multiplt-Valued Logic (Utah State University)
Zadeh, L. A. and Bellman, R. E. 1976. Local and fU%Zy logic, Memorandum
ERL-M584, Electronics Research Laboratory (Univeraity of California, Berkeley)
Ziff, P. 196o. Semantic Analysis (Comell U.P.)
 INDEX

• indicates glossary entry

abduction, un assertibility, 98-9, 119n

absolute versus model-theoretíc asaurnptions, rule of, 19
semantics, see semantics 'at least one' and '3 ', 34; see a/10
Ackermann, R., 205 quantifiers
acquaintance, 92; see also Russell's atornic versus molecular, 91-2, 243•
theory of descriptions artributive adjectives, 123
adverbs, 25-6, 122, 123; applyíng to attributíve versus referential use, 69-70
'possible', 195; applying to 'true', Austin, J. L., 82, 86, 93-4, 113-14,
168-9 254
airns of logic, xiii, 1, 3, 5, 32-5, 38, axiom, 13-14, 18, 19-21 passim, 243•,
227ff. passim Z45, 248; logical versus proper, 19;
Alston, W. P., 49 schemata, 19
Altham, J. E. J., 35, 41 axiomatic aystems, 18, 19-21, 243•
ambiguity, z3, 75, 121, 186 axiom schemata, see axiom
ampliative venus non-arnpliative Ayer, A. J., 98
argumenta, 1 z; se« also inductive
logic Baker, A. J., 36
analytic versus synthetic, 171-5, 178, Bar-Hillel, J., 79, 140
182, 193, 236-7, 243•, 247; see also Barker, S., 12
meaning, theory of; necessity; belief, 8, 11, 35, 74, 123, 125, 127-8,
synonymy 233; see also propositional attributes,
'and' and '&', 1, 31, 34, 35; 'and verba of
then ', 34; see aíro conjunction; Bellman, R. E., 167
sentence connectives Belnap, N. D., Jr, vi, 16, 31, 37, 38,
Anderson, A. R., 16, 37, 38, 136, 139, 42, 88, 15on, 179, 197-203 pa,11"m,
197-203 pas,im ZS4
Anscombe, G. E.M., 24, 124n, 15on Bennett, J., 197
a priori venus aposteriori, 170-1, 243•; Bergmann, G., 178
see also necessity Berry's paradox, see paradoxes
argumenta, 12-13, z39; see a/so Binkley, R., 149
validity Birkhoff, l., 211, 228
bivslence, xii, 69, 7z, So-2, 85, 101-a,
Aristotle, 4, 5, 87, 100, 110, 176, zo4,
zo8, zz7 110, 130, 304, zo8-10, Zl2, ZIJ-15,
arithmeric, s. 9-10, 137, llff, a46, 2.43•, -2.46; se« also excluded middle,
z,.S; ,ee abo lo,icism; mathematics law of
268 Index

Black, M., 110, 164 245 (see also dispositional

Blanché, R., 19n properties); see also entailment;
Blanshard, D., 89, 91 irnplication
Dlumberg, A., 20 conjunction, 244•; and universal
Bochvar, D. A., 140, 145, 207, 211, quantifier, 34, 41, I 13; see also
212, 215 'and' and '&'
Boole, G., 153, 223n, 238 connotation, 58-60 passim, 246; see
Boolos, G., 253 also meaning, theory of; Sinn versus
borderline cases, see vagueness Bedeutung
bound versus free, 39, 252; see also consequence, 244 •, 247; see also
quantifiers; variables deducibility
Bradley, F. H., 66, 86, 89, 91, 94-7 consistency, 1, 139, 189, 193-4, 244•,
passim, 254 248, 251
Brandom, R., I I 8n constant, 23, 244 •
Brouwer, L. E. J., xi, 216-20 passim, constructivism, 217-20 passim; see also
247; see also Intuitionist logic Intuionist logic
Burali-Fortl's paradox, see paradoxes contextual definition, see definition
Burge, T., 58, 64, 70, 127 contradiction, 22, 202, 224•; principie
of non-, 244
Camp, J., 88 contradictory, 244•; see also negation
Cantor's paradox, see paradoxes contrary, 244•; see also negation
Cargile, J., 122, 178, 254 conventíon, 93-4 passim
Camap, R., xi, ion, 20, 67, 94, 125, conversational implicature, 36, 247,
126, 131, 163, 173, 247 250
Cartwright, R., 45, 71, 77 Copi, I., 141, 253
category, dífference between space and correctness of logical system, 22 1-32
time, 162 ; rnistake, I 63n correspondence theory of truth, see
causal staternents, 123 truth
causal theory of narning, 59-60 counterfactual conditional see
change, 160-1 conditional
characteristic rnatrix, 29, 218, 243•; counterpart theory, 19:z-3; ltt abo
srt a/so designated value; truth-table possible individuals
charity, principie of, 119 Cresswell, M., 177, 254
Chihara, C., 142 criteria of truth, stt truth
Chisholrn, R., 191 Curry, H. B., 47, 189, 246
Chomsky, N., 26, 46, 121, 242
Church, A., xi, 55, 143, 184fl Darwin, C., 232
Clark, M., 36 Dauer, F., 86
classical logic, 4, 5, 7, 152, 155, 163, Davidson, D., 26, 42, 58, 64, 94, 102,
201, 204, 216-19 passim, 226-31 103, 108, no, 114n, 116, 118-27,
passim, 245, 251 161, 242, 246, 247, 249, 254
coherence theory of truth, see truth decídability, 1, 7, 29, 212-13, 244•,
combinatory logic, 47-8, 244• 248
comparatives, 127 declarative (sentences), 76, 83
completeness, 1, 6-?, 9, 244•, 245, de dicto versus de re modalíty, 182,
246, 248, 250, 25 I 185; see also essentialism; necessity
conceptualism, 19 I, 248 •; see also deducibility, I 98-203 passim; see also
Intuitioníst logic valídity
conclusion, 13ff. passim deduction, 245 •
conditional, 244•; material, xii, 230; deduction theorem, 199, 245•
counterfactual, 33, 123, 244; definition, 171-5 passim, 245•, 250;
subjunctive, 38, 123, 179-81, 244, of the connectives, 28-g, 218-19;
 Index

definition (cont.) E, 200-3

conditions on, 103-4; contextual, Edwards, P., 243, 253
6 5; recursive, 1o6; enumerative, Einstein, A., 158, 232
111 ; complete versus partial, 147; eliminability of singular terma, see
implicit, 245; see also analytic singular terms
versus synthetic entailment, xii, 152, 198-203 passim,
degrees of truth, see truth 247; see also implication
denotation, 57�5 passim, 246; enthymeme, 245•
distinguished from reference, 70, enumerative definítion, see definitíon
212; see also reference; Sinn versus Epimenides paradox, see paradoxes
Bedeutung epistemic logics, xií, 4-8 passim, 175,
deontic logics, xii, 4, 5, 17 5; see also 183, 228, 251; see also propositional
moral discourse ; 'ought' atti tude, verbs of
Derrida, J., 189 epistemology, xiií, xiv, 8, 91, 94--6,
Descartes, R., 236n 110-11, 115-17, 119, 16o-1, 190,
descriptions, definite, 46, 65---'70, 245; 193, 213, 217, 225, 226, 232-42,
names assimilated to, 60-5; in 246•
modal logic, 185-7; see also Russell's equivalence, 246•
theory of descriptions erotetic (interrogatíve) logics, 4, 5, 82,
designated value, 29, 243, 245•, 251; 84
see also characteristic matrix; essentialism, 184---'7 passim, 194; see
truth-ta ble alto de dicto versus de re modality
Destouches-Février, P., 226 eternal sentences, 81--2, 158; see also
detenninism, 159, 196; see also tense
fatalism Euclid, 19, 228
deviant logics, xiii, 4, 5, 29, 30, 154, excluded middle, law of, 21, 30, 31,
201-3 passim, 204, 218-20, 222-32 130, 155, 162, 204, 217, 232, 234,
passim, 245• 243, 246•; see also bivalence
Dewey, J., xi, 86, 97--9 pa11im, 249, explicative versus non-explicative
254 argumenta, 12; see also inductive
disjunction, 245•; and existencial logica
quantifier, 34, 41, 113; inclusive expressíve power versus doctrinal
versus exclusive, 36; intensional content, 162; see also deviant logics;
versus extensional, 37, 200, 230; extended logica
Kleene's matrix for, 215; see alto extended logics, ziii, 4, 5, 154, 17 5,
'or' and 'V' 179, 201, 218-20, 222-32 paslim,
disjunctive dilemma, 216, 245; see also 246•
vel-elimination, rule of eztensional versus intensional, 43,
disjunctive syllogísm, 201 45--6, érn, 119, 246•
disposítional properties, 38, 18o-1, extensionalism versus intensionalism,
245 •; see also conditional 77, 1 ra, 124---'7 pauim, 158-62
distributive law, 162; see also quantum passim, 173-4, 193, :z.t6
logica
domain, 41, 245•, 252; see also facts, 91-4 passim, 114; see also truth,
quantifiers correspondence theory of
Donellan, K., 69 fallibilism, 90'1, 9-4---'7 passim, 115-17
double negation, law of, 21, 245•; passim, 232-8, 241-:z, 243
see aJso Intuitioniat logic Faris, J. A., 36, 254
Duhem, P., 119 fataliam, 209""10; see also determiniam
Durnmett, M. A. E., 7-8, 35, 86, C)8-<) tiction, 56, 70-3, 172
Field, H., 111-12, 122
Dunn, J. finite venus infinite, z.tb•
270 Index

first versus seeond order quantification, 82, 83, 97n, QS, 112n, u7n, 118n,
see quantifiers 130, 15on, 154, 172n, 208, 210, 214,
Fitch, F., 185 254
Fitting, M.C., 219 Hacking, l. M., 19n
Fitzpatrick, P., 64 Halldén, S., 163, 211
fixed point, 147 Hanson, N. R., 12n
Flew, A. G. N., 11 Hannan, G., 26, 127, 150, 254
Fogelin, R. J., 40, 199 Harris, J. H., 117n
Fellesdal, D., 182n "hedges ', 168; see also adverbs
formalism, 189, 241', • Hegel, G., 94
fonn versus content of arguments, Heidelberger, H., 131
5-'7, 22-7, 122, 153, 231, 242 van Heijenhoort, J., 223n, 253, 254
van Fraassen, B. C., 67, 130 Hernpel, C. G., 95, 254
free logics, 4, 71-2 Hernsworth, P., vi
Frege, G., xi, 2, 9-10, 22, 32, 39, 40, Henkin, L., xi
58, 65-8 passim, 72, So, 121, 125, Herzberger, H., 145
127, 139, 152, 153, 162, 171, 176, 'heterological' paradox, see paradoxes
212, 215, 227, 235, 238-42 passim, Heyting, A., 216-zo passim, 247; see
243, 246, 248, 253, 254 a/so Intuitionist logic
functional completeness, 28, 244, 251 Hilbert, D., 246
future contingenta, 81, 208-10; see Hintikka, J., 8, 40, 121n, 156, 172,
a/so fatalism 187, 191
fuzzy logic, xii, 35, 162-9; see also holism, 119, 121
vagueness Holmes, Sherlock, 71- 2; see alto
fiction
Gaines, B., 165, 254 Hughes, G., 177, 254
Garver, N., 140 Hunter, G., 253
Geach, P. T., II, 40, 139, 160-2
passim, 254 ídealism, 86, 94-6
Gentzen, G., 19n ideu, 61n, 240-2
geometry, 19, 228 identity, 4n, 60-1, 184-7 pnssim;
global veraua local, 223-32 criteri1 oí, 43, 77-8, 80-1; oí
Gochet, P., vi, 74 poselble Individuals, 191-3
GOdel, K., xi 'ií' and • -.. ', :11, 3 6-38; 111 a/10
GOdel's Incompletenese theorem, :z:z, conditionel; implication
139, 246•, 253 imperative logics, 4, 5, 81, 83-5
Goldbach's conjecture, 61, 246• pasrim, 250
Goodman, N., 38, 78, 172n implication, xii, 36-7, 152, 178-g,
Gould, J. A., 253 247•; material, 201, 230, 247,
grades of modal involvement, 181-7; paradoxes of, 37, 135n, 176, 197-
see a/so modal logic 203 passim, 249•; strict, 37-8, 152,
grammar, 26, 82, 132, 17�, 242; see 178-9, 230, 247, ps radoxes of, 37,
also Chomsky, N. 135n, 197-203 passim, 249•;
Grelling's paradox, see paradoxes relevant, 200, 230, 247; see also
Grice, H. P., 36, 194, 247 entailment
groundedness, 145-8; see also implicit definition, see definition
paradoxes incomplete symbols, 65--6, u7n, 245,
Grover, D. L., vi, 88, 129, 130, 132-5 247•
J)turim, 15on, 151, 254 inconsiatent 1yatem1, C}-IO, :z:z; ,u a/so
paradox
Haack, R. J., vi, 8o incorri¡ibility, 94-7 pmrim; set a/,o
Haack, S., xiv, 8, un, 17n, 7a, So, fallibiliam
 Index

independence (of axiorns), 19, 247• Lacey, H., 157, 161, 254
indexicals, 93, II3-14, 123, 156, 247• Lakatos, l., 235
inductive logics, 4, 5, 12-14 passim, Leibniz, G., 9, 192, 223n; -'• Law,
17, 247• 246; see also extensional versus
inference, 238-42 passim, 247•; see intensional; substitutability salva
also reasoning oeritate
informal argument, xiii, 1, 2, 14-15, Lemmon, E. J., 18, So, 178, 253, 254
33-4, 74, 83-4, 162, 196-7, 201-z, Leáníewski, S., 8
222-32 passim Lewis, C. l., xi, 37, 152, 159n, 178,
inquiry, theory of, 97-g pasnm 197, 198, 200
instrumentalism, 20, 221-32 passim, Lewis, D. K., 33, 35, 37, 49, 180,
248• 18g-g3 passim, 192, 194
intensional isomorphism, 126 Lewy, c., 79, 83, 214
interpretation, theory of, u9-20 Liar paradox, see paradoxes
interpretation, of formal systems, 3, Linsky, L., 42, 254
8, 247•, 251; of sentence con- lagica utens and lagica docens, 15-16,
nectives, tee sentence connectives; 25; see also informal argument
of quantifiers, see quantifiers; see logical atomism, 26, 86, 91, 94, 122,
also semantics 247•
interrogatives, see erotetic logics logical constructions, 66, 142, 159,
Intuitionist logic, xi, 4, 21, 29, 99, 245; see also Russell's theory of
155, 191n, 216--zo, 224, 247• descriptíons
Iseminger, G., 253 logical necessity, see necessity
logical positivism, 86, 237, 243, 247•
James, W., 86, 95, 97-<) passim, 249, logical systems, identity conditions
254 for, 17-22
Jaákowski, S., 205 logical truth, xi, xiii, 1, 13-14, 113,
Jeffrey, R., 253 171-5 passim, 194, 202, 221-32
Johansson, l., 218 passim, 243, 244, 248•, 251
Johnson, W. E., 37 logical versus proper axioms, se« axiom
Jones, A., 1241\ logicism, 9-10, 41n, 137, 171, 19m,
240n, 243, 248•
Kanger, S., 187 Lotze, R. H., 94
Kant, l., 97n, 152, 171, a35, 238, 241, LOwenheim, L., xi
243, 249 Lukasiewicz, J., xi, 165, 205, 2o8-II
Kaplan, D., 194 passim
Kepler, 232
King of France, 65-70 passim McCall, S., 213
Kleene, S. C., 146, 1+8, 2o6, 215; MacColl, H., xíí, 152, 176, 204
-'s 3-valued logic, 2o6--7, 212-13, McDowell, J., 70
215-16 Mackie, J., 37, 38, 88, 91, 110, 113,
Kneale, W. C., 6-7, 19, 20, 64, 79, 137, 143, 146, 151, 254
So, 1+o, 171, 197, 254 McKinsey, J. J. C., 219
Kneale, W. C. and Kneale, M., 7, 79, MacTaggart, J. M. E., 16on
214, 253 many-valued logics, xi, xii, 2, 4, 8,
KOrner, S., 165 29-30, 140, 145, 1+6, 1+8, 164-5,
Kotarbínski, T., 12411, 249 167, 204-20, 24+, 251; ,�� abo
Kripke, S., 1, 42, s8--6o, 64, 79, 83, characteriltic matra:
88, 144-8, 171, 185, 187, 188, 192, Marcus, R. Barcan, 42, 51, 52, 151,
213, a54 176, 185-7 pam,n
IJl8l8 terma, 1a3, 248•
M->', G., 23
272 Index

material conditional, see conditional; natural language, 12o-? passim; see

implication also informal argument
material implication, see implication necessiry, logical, gen, 133n, 159n,
matcrialism, see physicalism 172-5, 181-2; physical, gon, 171;
Mates, B., 42 and fallibilism, 233-5; see also
mathematics, 216-19pauim, 234, 235; analytic versus synthetic; meaning,
see also arithmetic; logicism theory of; synonymy
mathematical induction, 247•, 249 negation, 245; interna) versus externa),
matrices, see characteristic matrix; 35; of wffs containing definite
designated value; truth-tables descriptions, 65; and negative facts,
meaninglessness, 81, 163, 207, 211 92; redundancy theory and, 129;
meaning, theory of, 77; Davidson's see also 'not' and '-'
theory of, 118-27; versus theory negative facts, see facts: negation
of reference, 119, 246; see also Nelson, E. J., 67, 198
analytic versus synthetic; extension- Nerlich, G., 68
alism venus intensionalism; Ness, A., 112
necesairy; synonymy von Neurnann, J., 211, 228
meaning-variance, 30-1, 218-19, Neurath, O., 86, 94-5
223-4, 229-31 Newman, J. R., 139
Meinong, A., 66, 71-2 Newton, l., 160, 232
Mellor, D. H., 180 No Funny Business, rule of, 199
Mendelson, E., 21 nominalism, 55, 77, 98, 239ff., 248•
Meredith, C. A., 21 'nominalistic Platonism ', 5 5
metalogic, 1, 2, 155, 231, 248, • 249 non-contradiction, principie of, see
metaphysics, 221-32, 248•, 249 contradiction
Michalski, R., 214 non-existents, 70-5; see also fiction ;
Mili, J. S., 58, 246, 254 ontology
Miller, D. W., 117n non-standard logics, tee deviant logics,
mistakes in reasoning, :z41-:z; see also extended Jogics
fallibilism, psychologism normal, 208
Mitchell, O., 40 normative venus descriptive, 238-4:z
modal logics, xi, xii, 4, 6, :z3, 34, 43, passim
sz, 56, 77, 159n, 170-203, :zo4, 'not' and ' - ', 3 s; see also negation
:zos, :zo9, :z:z:z, :z:z4, :z48, zs 1; se« notational varian ts, 18; ,ee a/10
also necessity meaning-variance
modus ponens, rule of, 201, 229, 248• numerical sraternents, 40; se, a/10
Molnar, G., 171 arithrnetic
monadic/dyadic/polyadic, 40, 248•
monism venus pluralism, xiv, 1, 2, 6, objective versus subjective, 114-16
95, 178, 221-32 passim, 248• passim, 24o-2
Montague, R., ziíi, 40, 121n, 193, 254 object language venus metalanguage,
mood, 76, 195-'7 passim 103, 129-30, 143-4, 249•
Moore, G. E., 37, 80, g8, 228 objectual interpretation of the
moral discourse, 179; see also deontic quantifiers, see quantifiers
logic; 'ought • oblique contexts, 61n, 246; see also
Moming Star paradox, 185; see aho extensionaliam venus intension-
modal Jogic; quantifiers ; singular alism; propositional attitudes,
terms verbs of
O'Connor, D. J., 94
Nagel, E., 139 Ockham, William of, ss; -'s razor, 66
natural deduction syatema, 18-zi, 247, one-one correapondence, :z49•
34,11• ontological commitment, Quine's
 Index

ontological commitment (cont.) Pitcher, G., 79

criterion of, 9, 43-<J, 244; see also Plantinga, A., 30, 7:z, 170, 18:z, 185,
Quine, W. V. O. 187, 189, 19:z, a54
ontological relativity, Quine's thesis Platonism, 55, 77, a39ff., a4B•
of, 45n; see also Quine, W. V. O. Poincaré, H., 141
ontology, xiv, 9, 43-55, 71-5 passim, Polish notation, 18
129, 249• Popper, K. R., 82, 88, 90n, no,
open versus closed sentence, 39, 112-17 passim, 139, 164, 197, a33,
106-8 passim 23�,24on, 254
"or ' and 'V', 35; see also disjunction possible individuals, 77, 188, 189,
oratio obliqua, 124-'7, 249•; see also 191-3
propositional attitudes, verbs of possible worlds, 37, 6o, 77, 108, 159n,
'ought', 127; see also deontic logic; 170, 188-<)5, 249
moral discourse Post, E. L., xi, 152, 205, 244; -'s
many-valued logic, 2o8, 214-15
paradoxes, 88, 102-3, 114, 1 20--1, 130, postcard paradox, see paradoxes
134, 135-51, 202, 207, 211, 249•, Potts, T., 40
251; "set-theoretical ' versus pragmatic maxim, 97
'semantic', 1 37-8; solutions to, pragrnatícs, xi, xiií, 72-3, 74, 251 •
138-48; Russell's paradox, 10, 22, pragrnatism, 89, 97n, 249•
136-51 passim, 248; Liar paradox, pragrnatist theory of truth, see truth
135, 138, 139, 2II; postcard Prawitz, D., 19n
paradox, 135, 139-40; Berry's precisification, 164-g passim
paradox, 136, 138; Cantor's paradox, precision, see vagueness
136, 138; Epimenides paradox, 136; predicare calculus, xi, 4, 39-55, 56-73,
"heterological ' (Grelling's) paradox, 108-10, 245
136, 138; Richard's paradox, 136, preference logics, 4--6 passim
138; 'truth-teller' paradox, 136; prernise, 13ff. passim
Burali-Forti's paradox, 137, 138; presupposition, 68-g, 72, 249•, 251;
Strengthened Liar paradox, 140, see also descriptions, definite
145-6, :ZII Príest, G. G., vi, 182
"paradoxes ' of material implication, prímitive notation, 18, 20, 28, 45, 49,
see implication 245, 250•; see also paraphrase
"paradoxes ' of strict implication, tee Prior, A. N., 19, 21, 31, 32, 88, 129,
implication 132, 140, 156--62 passim, 213, 254
paraphrase, 49, 153, 161 probability, 17, 195; see also inductive
Parsons, T., 71, 194 logics
partial definition, see definition proforms, 132
partial truth, see truth pronouns, 39, 132
Peano, G., 138 proof, 250•; see also theorem
Peano postulares for arithmetic, 9n, proper names, 39, 57-65, 127;
249•; see also logicism logically proper names, 66-7; non-
Peirce, C. S., 12n, 16, 40, 55, 86, denoting, 70--5; see also singular
97-<J passim, 153, 23�, 236n, terma
238-42 passim, 249, 254 properties, 43, 5 3
perceptual judgments, 94-5, 242 propositional attitudes, verbs of, 75,
performative .theory of truth, su truth 123-7, 250•; su also belief, oratio
personal identity, 43, 158 obliqua
'philosophical logic ', 2 propositions, 43, 53; su also sentences,
'philosophy of logic ', 1, a, 248 statements, propositiona
physicalism, 1 u-u, ra4n, 184-5 prosentences, 13:z-4; see also truth,
physical necessity, su necessíty proeential theory of
 274 Index

prosentential theory of truth, see extensionalism verus intensionalism;

truth substitutability salva oeritate
protocol statements, 95 refute, 250•
psychologism, 238-42 Reichenbach, H., 210
Ptolemy, 232 relations, 40, 21-9, 250•
'pure' versus 'depraved' semantics, relativism, 226
see semantics relatívity theory, 158-62 passim
Putnarn, H., 77, 82, 210, 211, 226, 253, relevance logics, xii, 16-17, 37, 197-
254 203, 248
Rescher, N., 3, 8, 86, 88-91, 96-97,
quantifiers, xi, xii, xiii, xiv, 1-fl, 23, 153, 205, 208, 226, 228, 254
34, 39--56, 250•; first versus second revisability of logic, xiv, 225, 232,
order, 1-fl, 6, 9, 40, 52-5, 128, 130-3, 237-8; see also fallibilism
249; non-standard, 34-5; and names, rhetoríc, 11, 17, 201-3 passim, zz7,
40, 12m; numerical, 40-1; objectual 229
versus substitutional, 42-3, 79, 131, Richard's paradox, see paradoxes
151, 186-7; and quotation marks, Richards, T., 190
104-5; in modal logíc, 182-7 passim, rigid designator, 58-6o, 192; see also
194-5; Intuitionist, 219; see also possible individuals; singular terms
variables Rogers, R., 108, 254
quantum logics, 4, 29, 162, 210-11, Ross, A., 85
228 Rosser, J. B., 205
quantum mechanics, 250• Routley, R., 71, 163
Quine, W. V. O., xi, 3, 9, 23, 24, 38, rule of inference, 13-14, 19--21 passim,
42, 43, 51, 52, 55, 58, 66-7, 77-9, 31-2 pasnm, 224-5, 243, 247•, 248;
81, 82, 95, 97, 98, 104, 108, 119, see a/so natural deduction
125, 139, 15�2 passim, 171, 202-3, Runes, D., 243
225, 230, 236-7, 246, 247, 253, 254; Russell, B., xi, 17, 18, 26, 40, 42, 58
attack on analytic/synthetic 65-6, 68, 70-1, 86, 88, 89, 91-2,
distinction, 172-5, 182, 243; critique 113, 121n, 122, 137, 143, 146, 151,
of modal logic, 178-87, 193-5; on 152, 153, 156, 162, 164, 176, 178,
revisability of logíc, 232-8 passim; 198n, 228, 245, 247, 248, 250, 254
- 's criterion of ontological comrnit- Russell, L. J., 36
mene, set ontological commitment Rusaell's paradox, set paradoxes
quotation, 104-5, r27, 149--50, 183 Russell's theory of descriptiona, 26,
47, 65-6, 79, 127n, 247; see a/so
R, 200-1 descriptions, definite
Ramsey, F. P., 83, 88, 127-34 passim, Russell's theory of types, 141-3, 147-
142, 21-7, 254 8, 251•; see a/so paradoxes
ratíonality, 242 Ryle, G., 5, 23, 142-3, 146, 151, 163
realism, 98, 116, 141
reasoning, xiii, 239; set also inference So·5, 177, 178
recursive definition, see definition S 1, 193
reductionism, 111-12; see abo S4, 177, 178, 190, 202
physicalism S5, 177
redundancy theory of truth, see truth Saccheri, G., 19
reference, 61-5 passim, 246; dis- Salmon, W., 12
tinguíshed from denotation, 70; salva oeritate, 250•; set also
see abo denotation ; Sinn versus substitutability salva oeritate
Bedeutung samesaying, 124-7 pastim; set a/so
referentially transparent versus intensional isomorphiam; oratio
referentially opaque, 246; set also obliqr1a
 Index

satisfaction (in imperative logic), 85, eliminability of, 46--7; and ontolo¡y,
250• so; and substitutional quantification,
satisfaction (in Tarski's definition of 51-3; in modal logíc, 182-7 pa11im
truth), 85n, 88, 105-10 passim, Sinn versus Beckutu"(f, 61-2, 212,
113-14, 250• 246; se« also denotation; reference
Scheffier, l., 46, 95, 125 Skolem-Lewenheim theorem, 250•
schemata, 20, 78 Skyrms, B., 12, 17, 140, 254
Schiller, F. C. S., xi, xii, 73, 98, 202, Stupecki, J., 30
249 Smiley, T. J., vi, 67, 179, 198, 212,
Schlick, M., 91--6 passim, 247 213, 215
Schock, R, 71 Smullyan, A., 185-7 passim
Schonfínkel, M., 47 Smullyan, R.M., 139
science (relevance to logic), 161-2, soundness (of a logical system), 14,
179-80 245, 250•
scope, 18, 65, 186 soundness (of an argurnent), 14, 250•
scope of logic, 1, 3-rn, 19n, 153, 155, Stalnaker, R., 37, 18o
202, 216-17, 248 Stebbing, S., 11
acope of science, 3 Stevenson, J. T., 32, 254
Searle, J., 58, 63, 72 Stoic logicians, 152
self-evídence, 10, 235-6; see also Stout, G. F., g6
fallibilism Strawaon, P. F., xi, xii, 35, 47, 51,
self-reference, 139; see also paradoxes 67--<;1, 73, 79, 81, 88, 11411, 156,
Sellars, W., 112n 16on, 175n, 194, 253, 254
semantic closure, 102-3, 120-1 ; see Strengthened Liar paradox, see
a/so paradoxes paradoxes
semantic primítives, 105-6, 111-12 striet implication, see implication
semantics, xi, 13-14 passim, 41, 74, subjunctive conditional, ,ee conditional
154, 251; "pure ' versus 'depraved ', substituend, 42, 252•
30, 188-90, 205; absolute versus substitutability salva vmtate, 173,
model-theoretic, 43, 108, 122; 183-7 passim; see also extensionalism
semantic paradoxes, 135-51 passim; versus intensionalism; Leibniz', )aw
for modal logic, 187--<;13 substitutional interpretation of the
semantic theory of truth, set truth quantifiers, ,ee quantifiers
sense-data, 92 supervaluations, xii, 216
sentence (propositonal) calculus, xi, 4, Suppes, P., 104
189, 244, 245 syllogistic, 4
sentence eonnectives, xi, xiii, 1, 28-38, synonymy, 76-7, 126, 173-5 ptum,r,
223-6 passim, 245; interna! versus 193, 237; ,ee also analytic versus
extemal, 2o8; see also conjunction; synthetic; meaning, theory of;
disjunction; implication ¡ ncgation necessity
sentence letters (propositional syntax, xi, 13-14 panim, 30, 74, 104,
variables), xiii, 1, 78-9, 214 113, 251•
sentence, statementa, propositions,
xii, 74-85, 86n, 114, 126n, 140, T, 177
239ff., 251 j set also trurh-bearers (T) scheme, ,ee Tarski's (T} Khml•
sequences, 1o6-10 pan;,,,, 250• Tarski, A., a6, 82, 88, 90, 99-127
set theory, 6, 7, 9, 135-51 pan;,,,, :144, pasmn, 128n, 130, 145""'9 pan;,,,,
250•; (uzzy, 165 15on, 219, 245, 247, 254; -'• (T}
set-theoretical paradoxes, se« paradoxes IIChema, 100-5 panirn, 112-13, 118,
similarity, 75, 78 119, 251 •; ,ee abo truth, IICffllllltic
aimplicity, 167 theory of
singular terma, xii, 39, 56--73; tautolotY, 15, 178, 243, 245, 251•
 276 Index

tense, 81-2, 156-62, 195-7 passim ; 'truth-teller' paradox, su paradoxes

and substitutional quantification, 51 truth-values, 78, 207, 212, 213-15
tense logics, xii, 4, 5, 151)-62 passim, Turquette, A. R., 205
176 type versus token, 75
'thatt', 132-5 passim
theorem, 2, 13-14 passim, 21, 182, use versus mentíon, 178
243, 244, 245, 251•
thoughts (Geqanken), 240-2 vagueness, xii, 7, 34-5, 81, 121, 162--9
Thouless, R. H., 11 passim ; a/so see fuzzy logic
Tichy, R., 117n validiry, xii, xiii, 1, 5, 12-27, 84-.5,
time, 158-62 passim; see also tense 163, 174-.5, 200--3 passim, 221-32
'tonk', 19, 31-2 passim, 251•; su also deducibility;
topie neutraliry, 5-6, 7, 23; see also relevance logics
form versus content of arguments value (of variable), 42, 252•
Toulmin, S., 225n variables, xiii, 244, 252•; free versus
transworld identity, 191-3; see a/so bound, 39; values of versus
identity ; possible individuals substituends for, 42; eliminability
truth, xi, xii, xiii, 2, 8, 86-134; and of, 47; see a/so quantifiers
substítutíonal quantificatíon, 50; Vaught, R. L., 108n
definitions versus criteria of, 88--91, vel-elimination, rule of, 19, 24•
95--'7, 115; adequacy conditions on verisimilitude, 88, 116-17, 252•
definitions of, 99-rn5, 120--1, 215 vicious circle principie, 138, 146, 148,
(see a/so Tarski's (T) schema); 149-51 passim, 252•; see a/so
Intuitionist view of, 15 5; degrees of, paradoxes ; Ruesell's theory of types
166-9; partial, 169, 214, in virtue Vienna Circle, 247; see a/so logical
of meaning, see analytic versus positivísrn
synthetíc; redundancy theory of,
55, 83, 88, 91, 113, 127-34, 247, Wallace, J., .51, 127
249, and the paradoxes, 149-.51; Weinstein, S., u3, 12.5, 126
correspondence theory of, 82, 86, Weston, T. S., .51
89, 91-4, 11-17 pa11ím; semantic wff, 13, 109, 2.52•
theery of, 82, 88, 97, 99-127; White, A., 82
coherence theory of, 86, 89, 91, White, Morton G., 49, 17an
94-7, 102, 11 .5; pra¡matiat theory Whitehead, A. N., xi, 17, 18, 66, 1.52,
of, 86, 89, 91, 97"""9, 102, 11.5; 16on
perfonnative theory of, 88; pro- Williams, C. J. F., 88
sentential theory of, 88, 135-5 Wittgenstein, L., xi, 26, .58, 63, 82,
truth-bearers, 8, 74, 79-84, 86n, 115 86, 91-2, 99, 113, 122, 124, 1.52,
128--9, 140; see also sentences, 205, 247
statements, propositions Wolf, R., 152
truth-conditions, 118-27 passim Woodruff, P., 212
truth-functional logic, 15, 23, 34, 246, Woods, J., 72
251• (see a/so many-valued logics); von Wright, G. H., 23, 34
non-truth-functional logic, 215-20
(see a/so Intuitionist logic; modal Zadeh, L. A., 35, 164-9, 254
logics; supervaluations) zero-prernise conclusion, 14, 15; see
truth-tables, xii, 29, 205; see also o/so theorern
characteristic matrix Ziff, P., 58