SPREADING THE WORD

I G E
Groundings · the
Philosophy of Language

by

SIMON BLACKBURN

CLARENDON PRESS · OXFORD

 Oxford University Press, Walton Street, Ox.ford ox2 6nP
Oxford New York Toronto
Delhi Bombay Calcutta Madras Karachi
Petaling Jaya Singapore Hong Kong Tokyo
Nairobi Dar es Salaam Cape Town
Melbourne Auckland
and associated companies in
Berlin Ibadan

Oxford is a trade mark ef Oxford University Press

Published in the United States

by Oxford Universifv Press, New York

© Simon Blackburn 1984

First published 1984

Fifth impression 1992

All rights reserved. No part ef this publication may be reproduced,

stored in a retrieval vstem, or transmitted, in a'!} farm or by arry means,
electronic, mechanical, photocopyin£ recording, or otherwise, without
the prior permission ef Oxford University Press

This book is sold subject to the condition that it shall not, by way
eftrade or otherwise, be lent, re-sold, hired out or otherwise circulated
without the publisher's prior consent in any farm ef binding or cover
other than that in which it is published and without a similar condition
including this condition being imposed on the subsequent purchaser

British Library Cataloguing in Publication Data

Blackbum, Simon
Spreading the word.
1. Languages--Philosophy
I. Title
401 P106
ISBN0-19-824651-X Pbk

Library ef Congress Catalogi.ng in Publication Data

Blackbum, Simon
Spreading the word.
Bibliography: p.
Includes index.
1. Languages--Philosophy. I. Title
PJ06.B47 1984 401 83-17253
ISBN0-19-824651-X Pbk

Printed in Great Britain by

Biddies Ltd
Guil.d.ford and King's I,Jmn
 Preface

Modern philosophy has been dominated by a concern with

language. But modern philosophy oflanguage is highly inacces-
sible. It is very hard for the ordinary student, let alone the
layman, to appreciate the problems it explores, or the methods
it uses. The interest of the results and their relations to other
philosophical or intellectual concerns is thus largely hidden.
Everyone who has any interest in modern philosophy knows the
great names of the subject- Frege, Russell, Tarski, Quine - but
too often even conscientious students know little more. Every
philosophy teacher or examiner will know how fragile is the
ordinary student's grasp of the issues they tackled, the methods
they used, or the interest of their results. This book is an
attempt to introduce the problems, and methods, and some of
the results.
When I formed the intention of writing such a book, it
seemed a modest enough aim. I felt I had at least two qualifica-
tions. One is that I like my philosophy to be clear. I tend to
believe that too many of the complexities of the subject are
really covers for confusions, so that when these are removed the
real beauties can be revealed in dear and striking colours. I
believe that too few philosophers frame the golden words of
Quintilian above their desks: 'do not write so that you can be
understood, but so that you cannot be misunderstood.' The
other qualification I felt was that I believe in the importance of
the subject. I believe that the philosophy oflanguage ought to
be widely studied, and that its results and methods are of more
than merely specialist interest. I think that confusions about
language underlie many other philosophical problems, and
that any serious study of, for instance, the metaphysics of
morals or of persons, of groups and of psychology demands at
least an initial baptism into the issues I want to introduce. In
sum, I believe that the great perennial problems of philosophy
can be felt by any reflective persons. Ifwe practitioners believe
that we have new and better approaches to them, and that
philosophical reflection upon language is a part of those
Vl PREFACE

approaches, then it is up to us to address that interest. We ought

to be able to show why our discussions are worth hearing.
This motivation, however, soon made me realize how
difficult the aim really was. Naturally, it would not do merely to
survey various positions on various issues. For the point of the
book was not to enable a student to go through the hoops; but to
enable him to understand why the hoops are placed where they
are. The keynote was to be appreciation of issues, not mere
acquaintance with them. On the other hand, nearly all the
issues have been studied in immense depth. There is no topic
here on which more cannot be said, or has not already been said
in the immense literature. So I began to feel that I was walking
naked through a landscape where every bush concealed an
army of snipers. Proper academic caution would demand dig-
ging fortified trenches after every step; but the audience I
wanted to reach does not need a trench-eye view, any more than
it just wants a bird's-eye view. I began to understand why the
book I had in mind had not yet been written.
The solution, I felt, was to take the reader into issues by
myself arguing them through: I found I could make no real
distinction between introducing problems and wrestling with
them, but in a way which took the intended audience with me.
And this is what I have tried to do. So this is not so much an
introductory 'text', or something like a survey of its domain. It
is an attempt to show the student what is done by philosophiz-
ing about language. It is my own reaction to philosophical
problems about language which interest me. But it is written
with an audience of beginners in mind. In a sense, it is an
indication of the place of the philosophy of language in many
other intellectual endeavours - in particular, in the philosophy
of understanding, of knowledge, and of truth. It is therefore an
attempt to place some of the investigations that go under the
philosophy of language, and to indicate not only their implica-
tions, but sometimes their limitations as well. Naturally, I hope
that the material serves as the basis for courses on the
philosophy of language. If it does, it will need supplementing,
and I have indicated what I take to be the right supplements in
notes to each chapter.
Because I believe that there is no deep study of language
which is not also concerned with philosophy of mind, and with
 PREFACE vu

the nature of truth and reality, this book contains more general
philosophy than its title might indicate. The philosophical
aspects of language I have selected include the whole inter-
action between thinkers, their language, and the world they
inhabit. It is these large themes, rather than detailed technical
problems, which I have tried to explore. So, for example, in
connection with truth I include quite detailed treatment of
particular domains of truth, such as moral truth, and of some
aspects of the theory of knowledge. My main regret is that space
prevented inclusion of more such examples, for instance on the
theory of conditionals, or of possibilities, or of mathematics.
One of the casualties of the trench-eye view is that not only
students, but thinkers from other countries and traditions, find
much of the philosophy of language incomprehensible. They
can then come to dismiss it as irrelevant to their concerns - the
product of a "linguistic" or "analytical" school within
philosophy, which can be regarded as optional or misguided.
But in so far as these labels suggest some particular body of
doctrines or of techniques, then I could not accept them.
Indeed, in the course of the work I suggest reasons for avoiding
some doctrines associated with these titles (chapters 5 and 6).
The only sense in which they are appropriate is that we are
concerned to think about issues raised by reflecting upon
language, and to do it carefully. But doing that is something
which no self-respecting philosopher, from any school at all, can
hope to avoid.
I have tried to keep footnotes to a minimum, and they men-
tion only works specifically quoted or discussed in the text.
Notes giving fuller sets of references, suggestions for further
reading, and sometimes subsidiary comments, are included at
the end of each chapter. In order not to break the flow I have not
generally included indicators to these notes in the text. I have
included a small glossary of philosophical terms at the end of
the book.
Conversation with many friends and students has helped to
shape the book. I should like to thank especially David Bostock,
Alberto Coffa, Elizabeth Fricker, Martin Davies,John Kenyon,
and Ralph Walker. Edward Craig read the penultimate manu-
script with enormous care and his comments helped me in
many ways. I should also like to thank the Open University for
vm PREFACE

permission to use parts of material which I originally prepared

for them, in the chapter on reference. I owe especial gratitude to
the Radcliffe Trust, for a Fellowship which freed me from my
teaching duties and enabled me to contemplate the work. I owe
thanks as well to the Master and Fellows of Pembroke College
for allowing me to benefit from this release. Finally the text has
been endlessly improved by my wife's editorial and literary
skills; I owe a debt to her greatly over and above that which all
writers must owe to those around them.
Pembroke College S.W.B.
May 1983

ACKNOWLEDGEMENT

I wish to record my thanks to Lady Ethel Wodehouse and

Hutchinsons Ltd. for their kind permission to use the quotation
which appears as the epigraph to chapter 1; and also the
permission of the author's estate and their agents, Scott
Meredith Literary Agency Inc., 845, 3rd Avenue, New York
10022, USA.
 Contents

PART I: OUR LANGUAGE AND OURSELVES

Chapter 1. The Shape ef the Problems 3

1. A Preliminary Map 3
2. Kinds of Question 7
3. Semantic Descriptions 10
4. Further Points about Words and Languages 18
5. Rules and Psychology 26
Notes 37

Chapter 2. How is Meaning Possible?(]) 39

1. Representation and the Dog-leg 39
2. Images 45
3. An Innate Representative Medium? 51
4. Radical Interpretation: Manifesting Meaning 57
Notes 67

Chapter 3. How is Meaning Possible? (2) 69

1. Describing: Three Ways of Being Odd 69
2. Bent Predicates: Wittgenstein and Goodman 74
3. Wooden Communities: Uses and Ways of Life 82
4. Privacy and Practice 92
5. Exercising Mental Concepts 103
Notes 107

Chapter 4. Conventions, Intentions, Thoughts 110

I. Grice's Approach 110
2. Openness and Communication 114
3. Convention 118
4. Force 122
5. The Place of System: Homeric Struggles 127
6. Deferential Conventions 130
7. Thought Again 134
Notes 141
x CONTENTS

PART II: LANGUAGE AND THE WORLD

Chapter 5. Realism and Variations 145

l. Oppositions to Realism 145
2. Rejection 148
3. Reduction 151
4. The Holistic Objection 158
5. The Status Problem 163
6. Expressive Theories: Contrasts with Truth 167
7. Metaphor and Truth l 71
Notes 179

Chapter 6. Evaluations, Projections, and Quasi-Realism 181

1. More Detailed Motivations 181
2. Frege's Argument 189
3. Constructing Truth 197
4. Interlude: Bivalence, Fiction, Law 203
5. Other Anti-Realisms: Causes, Counterfactuals,
Idealism 210
6. Mind-Dependence 217
Notes 220

Chapter 7. Correspondence, Coherence, and Pragmatism 224

1. Vacuity and the Quietist Regress 224
2. Transparency: Dummett and Frege 226
3. Redundancy: Ramsey and Wittgenstein 229
4. Facts andjudgement: Kant 233
5. Coherence: Russell, Bradley, and Peirce 235
6. What is Right about Correspondence? 243
7. Epistemology Regained 248
8. Truth Again 256
Notes 257

Chapter 8. Truth and Semantics 261

l. Truth theories: Tarski's Suggestion 261
2. Truth-in-Land Convention T 264
3. Top-Down and Bottom-Up 273
4. Radical Interpretation Again 277
 CO!\/TENTS XI

5. Truth and Extensionality 281

6. Composition, Logic, Translation 293
Notes 300

Chapter 9. Reference 302

l. Reference versus Description 302
2. Putnam's Strategy: Spinning the Possible Worlds 310
3. Singular Thought Theories 316
4. Singular Thoughts and Method 322
5. Modes of Presentation and Substitutivity Problems 328
6. The Contingent A Priori and Negative Existentials 333
7. The Right Relation 337
8. Perspectival Thinking 340
Notes 345

Glossary 351

Bibliography 355

Index 365
 PART I

OUR LANGUAGE
AND OURSELVES
 CHAPTER l

The Shape of the Problems

"When you come tomorrow, bring my football boots. Also, if

humanly possible, Irish water spaniel. Urgent. Regards. Tuppy."
"What do you make ofthat,Jeeves?"
"As I interpret the document, sir, Mr. Glossop wishes you, when
you come tomorrow, to bring his football boots. Also, if humanly
possible, an Irish water spaniel. He hints that the matter is urgent,
and sends his regards."
"Yes, that's how I read it, too ... "
P. G. Wodehouse, 'The Ordeal ofYoung Tuppy'.

1. A Preliminary Map
A philosophy of language attempts to achieve some under-
standing of a triangle of elements:

speakers
(psychology)

theory of m e a ~ ~ r y of knowledge

language world
(meaning) (metaphysics)
theory of truth

Fig.' 1

The speaker uses the language. With it he can put himself

into various relations with the world. He can describe it, or ask
questions about it, issue commands to change it, put himself
under obligations to act in it in various ways, offer metaphors,
images,jokes, about what it is like. The task of the philosopher
is to obtain some stable conception of this triangle of speaker,
language, and world. This aim will appear somewhat different
to different generations and times. The things which seem
reliable and unpuzzling to one thinker come to seem crucially
4 OUR LANGUAGE AND OURSELVES

problematic to another. One of the difficulties of appreciating

the area is just that of seeing which questions should be framed
first, and which concepts are reliable and legitimate aids to
answering them. Even at this point the choices are contentious.
There is no one proper selection of questions and aids which
philosophers of language unite in respecting. But there are
more or less intelligent guides to choosing, and one,good guide
will also enable one to come to respect the virtues of other good
guides.
At a given time, in a given philosophical tradition, one or
another of the points of this triangle will appear prominent.
That point will represent the primary source of understanding,
so that the natural direction of enquiry is to use that know-
ledge to aim at conclusions about the other elements. Thus in
the European tradition from Descartes until this century the
moving conception has been that of the individual, with his
particular capacities for experience and reasoning. The aim of
metaphysics has been to attain a conception of the world which
would enable this individual to know something about it (or
enable him to put up with scepticism about it). The nature of
his mind determines what kind oflanguage this individual can
intelligibly speak: which ideas the language can express for him.
In Locke, or Kant, the prime investigation is into the kind of
mind the individual has; given this, the nature of language, so
far as it is important, or of the world, in so far as it is intelligible,
follows on.
In a different tradition, for instance that of the Greeks, it
would seem more natural to establish by metaphysical enquiry
some feature of the world, of reality. Thus metaphysical argu-
ment might show that values are real, or that numbers are real,
that they are unchangeable, and so on. It would then appear
that we must have capacities large enough to enable us to
comprehend these established objects, and from this fact
would follow conclusions about our rationality, our minds, and
our epistemic (knowledge-gaining) natures. More familiarly, a
culture may reasonably accord such respect to the science of the
day that the world as depicted by that science becomes the sole
and immediate metaphysical reality. Both the .,nineteenth
century and our own have seen philosophy dominated by
scientific naturalism, in which the results and (alleged) method
 THE SHAPE OF THE PROBLEMS 5

of the natural scientist are regarded as the real philosophical

data. In the nineteenth century this attitude found expression
in the views of Mill in England and the anti-Hegelian, anti-
speculative empiricism of Germany: it led to a vague belief that
a science of psychology would produce the only real advance in
our understanding of logic, language, and thought. The same
attitude has persisted throughout much empiricist philosophy
of language this century, although the science which is to
provide the eventual source of understanding tends to shift:
psychology, formal logic, formal semantics, or structural
linguistics.
But the noteworthy change is towards concentration upon
language itself, rather than the mind of its user or the world he
inhabits. Now it may seem odd to roll the triangle so that the
investigation of language assumes primacy. A language has to
be capable of describing whatever world the metaphysician
allows. It has to be intelligible to whatever creature the psy-
chologist paints. But, it might seem, its interesting properties
would first be discovered by settling the nature of that world or
that creature. For example, if we ask whether our numeral
expressions like '6' or '7' 1 refer to things, and think of that as a
question in the philosophy of language, it might seem that we
have to wait until the metaphysician tells us about the reality of
numbers, and until the psychologist or epistemologist tells us
how we can know about them. If numbers are real and we
can know about them, then presumably we can regard our
numerals as referring to them. Otherwise, we must regard them
in some other light. It is not clear how there could be a self-
governing investigation into language with enough authority to
issue commands to the other philosophical areas. But ideologies
change, and it has become natural to give the nature of
language considerable autonomy, and even sovereignty over
the other elements of the triangle. An individual's psychology
becomes whatever is needed to enable him to understand the
language which stands revealed, and the world becomes what-
ever is necessary to make true the true statements made with
I
In this work I put single quotes around a word or sentence which is being talked
about. Sometimes, where it helps clarity, or a lot of words are being mentioned, I also
use italics. Double quotes are used for direct quotation from other writers, and to
register a deliberate distance from a particular phrasing.
6 OUR LANGUAGE AND OURSELVES

that language. This is the "linguistic tum", for better or worse,

of most of the important philosophy this century: this book will
assemble some of the materials necessary for evaluating it.
Not only can the triangle be rotated, but whole aspects ofit
can wither away. It can diminish to a straight line, or even a
point. This will seem a good or bad thing according to whether
we want to do away with one kind of subject-matter. One
concern can dominate philosophy so entirely that questions in
other comers of the triangle become dismissed. One of the few
things everyone has heard about logical positivism is that it
claimed that metaphysics - the study of the "world" element in
the triangle - was dead; at present many writers believe that
epistemology has just died, and the investigations traditionally
thought of as part of the theory of truth are given regular
obituaries. All this arises because people find it hard to see how
there can be any enquiry except into the relations of speakers
and their language. We see what we are committed to by
surveying the language we speak and the beliefs we express by
it. There is no other philosophical study of our "world" or our
relations to it- although, of course, there are scientific studies of
the nature of the things it contains. It is characteristic of any
dominant philosophy to be hostile to some one of the comers of
the triangle and to shunt all interesting questions into the line
between the other two.
Even if we believe that some one element of the triangle
dominates the rest, we should beware of forgetting other pos-
sible orientations. Amongst other things, this would make it
impossible to appreciate the concerns of people who have a
different perspective, and it thereby causes distorted history of
philosophy. To take a plain example: ifwe are convinced that
all worthwhile philosophy is ultimately the analysis of the
meanings of terms which we use, and ifwe are sure that Hume,
or Kant, or whoever, was doing worthwhile philosophy, then
we must regard them as analysing the meanings of terms.
Unfortunately this corresponds to very little that they appear to
have been doing. So if we persist in this diagnosis, we run the
risk of judging their work by quite inappropriate standards (in
the case of Hume, I show why in chapter 6). But there is in any
case a more general reason for tolerating different attitudes
towards elements of the triangle. This is that the flow ofimplica-
 THE SHAPE OF THE PROBLEMS 7

tions around it is subtle and hard to perceive, and it is some-

thing to which any thinker needs great sensitivity. A theory of
language is likely to affect any metaphysics and epistemology,
but they in turn affect most of our ideas about language. Or so I
shall try to show.

2. Kinds ef Question
What is to be done to approach this stable conception of the
triangle of terms and their relations? Or to put the question
another way, how is there scope for a philosopher to improve
some untutored, common-sense appreciation of how we stand
in relation to our language, and to the world we depict with it?
Suppose we start with the individual. The person who has
mastered a language understands its sentences and the terms used
to make them up. He gives them a meaning, and members of his
linguistic community do the same. What kind of fact is this?
Understanding the sentences of a language is knowing what
they are used to say - which thoughts or questions or com-
mands or wishes they express, in the mouths of speakers of that
language. But what is the difference between having such
knowledge and lacking it? Or is it wrong to think of some one
kind of knowledge; is there only the criss-crossing of ways in
which we do understand each other's sayings, and ways in
which we do not - a continuous ebb and flow of tides of
incomprehension? My experiences, beliefs, ideas, and attitudes
are different from yours, and different again from those of
people in other places and times. How can we give our words
the same significance? But ifwe do not, how is communication
possible at all - how do I pass you information, or tell you what
to do, or learn anything about what is believed by other people?
There is a tension between the rooted, organic, place of
langµage in particular persons and people, at particular places
and times, and the common stock of thoughts which we seem to
express, enabling me to understand you, or understand and
translate what was written five hundred years ago, or even
timeless tru.ths and certainties which do not change. Different
temperaments feel this tension differently. Some, like the later
Wittgenstein, stress the first aspect, the place of any language in
the activities and relations of people at times: "Language has
8 OCR LA'.'iGUAGE AND OURSELVES

grown like any big city: room by room ... house by house, street
by street . . . and all this is boxed together, tied together,
smeared together", wrote Fritz Mauthner. 2 How can I know
my way around your town - how can I rely upon any points of
reference (it's no good relying upon anything you say, since it is
the significance of that which I am looking for)? We need a
philosophy of mutual understanding, protecting shared under-
standing in the face of divergent ways and experiences.
Perhaps the mazes and labyrinths of old towns makes them
unfit places for real seekers after truth, who need accurate and
precise ways of describing things and understanding each other's
descriptions. So there exists a vision of a genuinely scientific
language: a purified, precise instrument of the discovery of
truth. If our words resist accurate analysis, they should be
replaced by ones which permit it; if our own inferences and
modes of argument are messy and unsystematic, they should be
replaced by ones which are precise and computable. The goal of
a logically adequate or even perfect symbolism, pursued through
the progressive refinement of artificially simple languages, has
always been a preoccupation of philosophers, and indeed of
scientists (it was a leading aim of the original Royal Society. See
2.2, footnote 5). When we think about almost anything hard, we
are apt to fear that it is our words that are letting us down. We
need to "remove the mist or veil of words" (Berkeley), or to
avoid being "caught in the nets of language" (Nietzsche); the
remedy is to ensure that our words relate to reality in the proper
way, that our inferences are solid and our distinctions accurate.
Many rationalists have shared Leibniz's dream of a universal
characteristic, a language suitable for science and logic; the great
logical work of Frege made it possible to see more clearly what
such a language might be like, and the vision dominated the
subsequent work of Russell, of the early Wittgenstein, of Carnap
and the Vienna Circle positivists. It still suffuses much research
into the semantics ofvarious parts oflanguage. Of course, the
vision has different components, and I come to distinguish
them in time. The pure vision of a rationally planned new town
for scientists and philosophers, with straight roads leading from
one solid unornamented block of truth to another, a Bauhaus of
2
I owe the quotation to Hans Sluga, Frege.
 THE SHAPE OF THE PROBLEMS 9

the mind, is not now widely shared. Such places prove unin-
habitable. But we learn more about the philosophy oflanguage
by examining them than by dismissing them out of hand in
favour of the old, human jumble in which we feel at home.
Fairly clearly, we will not assess the issues raised by this ideal
without further thinking not only about what it is to under-
stand a language, and to understand other people's use of their
language, but also about what it is to describe the world by
using it. If the virtues of a good language are mainly in the
direction of increased comprehension, of each other and of the
world, then. progress towards seeing those virtues rightly needs
a philosophy of mutual understanding, and also a theory of truth,
taking us onto the other side of the triangle. The understanding
speaker uses his sentences to make judgements, which are true
or false. So we need a solid conception of what it is to do this.
The issues here also root deeply in the history of philosophy.
There are classical theories of truth, and various newcomers.
Truth is something we all respect, and want for our beliefs. But
framing and answering general questions about truth is not
easy. Probably the best way to feel these problems is to start
locally. Particular kinds of judgement, perhaps moral and aes-
thetic judgements, are easily felt not to be capable of truth in the
same way as more homely beliefs: they are not "objective",
mark no features of "reality". The same can be felt for other
things we say, such as mathematical remarks. What makes true
the judgement that 7 + 5 = 12? Even the theses of scientific
theories can be seen as instruments of prediction rather than
descriptions of hidden aspects of reality. Perhaps many of our
commitments should not be regarded as beliefs that something
is true or false (corresponds to reality), but rather should be
seen in some other light. This is the way into the issues of truth,
realism, and the nature of judgement. It has links with the
theory of knowledge, fairly obviously, and back with the theory
of understanding. But finally it connects with a much more
thorough enquiry into the structural workings of our language.
At many points in discussions such as these, use has to be
made of fairly obvious features of words of our languages. Some
are names, some predicates; some are connectives linking dif-
ferent parts of discourse, some turn assertions into questions,
and so on. The goal of systematic, compositional semantics, is to
10 OUR LANGUAGE AND OURSELVES

form a view about how best to describe the functioning of

individual terms in sentences, and how to describe how, given
this functioning of their parts, sentences com!". to have the
meanings they do. Without this kind of knowledge our under-
standing of the nature of our language is exceedingly rudi-
mentary. And the pursuit of such a theory has in one way or
another attracted most major philosophers of language this
century (the notable exception being the later Wittgenstein). In
this book I introduce some issues making semantic description
oflanguages interesting and difficult. But they do not dominate
the philosophy from the outset. This requires some explanation:
the reader may well wonder quite what place issues in compo-
sitional semantics have, in the overall attempt to understand
ourselves and our languages. In some developments there might
seem to be nothing for the philosopher of language to do but
attempt yet more accurate descriptions of the way terms func-
tion in the generation of meaningful sentences and utterances: if
this were understood, it might be thought, then there would be
nothing else for the philosophy oflanguage to worry about. In
another common metaphor, a compositional semantics would
form the "core" of a philosophy oflanguage, with other issues
built entirely around it. In the rest of this chapter I illustrate
why I am not presenting matters quite like this, and introduce
more of the compositional aspect oflanguage.

3. Semantic Descriptions
All theorists of language are impressed by its compositional
nature. Competent users of language are not restrrcted to a
repertoire of previously understood sentences. We possess the
skill to generate new sentences meaning new things and to know
what these new sentences mean. I shall call this the elasticity of
our understanding. It exists because new sentences contain old
words in old patterns. And just as in chess old pieces making old
moves yield new positions, so our familiar devices are capable of
limitless recombination. Our understanding of the words and
the syntax enables us to identify the meaning of new sentences.
There ought to be some way of describing what the words (or
other features) do, so that we can represent the way speakers
and hearers respond to their presence when arriving at inter-
 THE SHAPE OF THE PROBLEMS 11

pretations. So we would understand a language semantically if

we knew exactly how the presence· of words, or any other
meaning-giving features of sentences, work in generating mean-
ings. Finding this out means distinguishing various categories
of expression, such as names or predicates. It will involve a
description of what expressions in those categories do- e.g. that
names refer to things, or predicates are true of sets of things.
And it will involve clauses saying how combinations of items
from these various categories make up sentences meaning
what they do - e.g. that a sentence consisting of a name
followed by a predicate says that the item referred to by the
name has the property expressed by the predicate, or belongs
to the set of things associated with the predicate. Such a
theory faces choices at each stage. There is difficulty over
finding the right categories, over selecting quite the right
description of what items in the categories do, and over
formulating the right clauses stating how items combine.
This is a relatively internal inquiry into the way meanings are
generated in a particular language - English, say, or im-
poverished, artificial fragments of it. But there are external,
surrounding questions, which success in compositional semant-
ics would evidently leave untouched. For instance, if a group
spoke an extremely simple language, with only names and a
fixed vocabulary of predicates, the compositional problem
would be easy. It would require little more than the clause
already indicated. But there would be questions about names,
predicates, and sentences which this success would not answer.
We would still want to know what it is in the behaviour, or
mental life, or whatever, of the community which makes it true
that a particular word is the name of a particular thing. We
would want to know what makes it true that a particular word is
a predicate with a particular assigned meaning, and what kind
of truth it is that sentences in the mouths of the group have the
overall meaning the theory calculates for them. Then there
would be questions about the nature of the judgements made
with this vocabulary: whether they admit of truth and false-
hood, whether they are objective, correspond with the world,
and how they know them.
Questions raised by compositional understanding can be
illustrated by a simple example. Consider our ordinary numeral
12 OUR LANGUAGE AND OURSELVES

notation. This contains the familiar arabic symbols 'O', 'l', ... ,
'9', and sequences of them, such as '437' or '501 '. Any sequence
of the original ten digits counts as a numeral, except, those
beginning with 'O'. An internal enquiry into how this notation
works would notice that there is a system in the way sequences
of numerals refer to numbers. If there were no such system then
it would be a miracle if we all took some previously unfamiliar
numeral, such as '675,896,341', to refer to the same number.
But we understand it readily, because, somehow, we are sensi-
tive to the placing of the individual digits, and understand
which number the expression refers to because of that placing.
The rule we respond to is that which dictates counting to base
ten. We can state it like this. First of all, we know what each
primitive digit does:
'O' refers to the number O
'l' refers to the number 1 ...
'9' refers to the number 9.
The form of these axioms or base clauses is just that first we
mention the symbol talked about. The single quotes show that
we are talking about the enclosed symbol. We then say what the
symbol does, in this case by saying which number it refers to.
We do that by using the symbol. This means that someone
understands these clauses only if they can already use the
notation. So the clauses would be useless for a certain purpose:
they couldn't teach someone to understand the digits. But this
does not unfit them for their role in the compositional theory, as
we shall now see. We could just go on indefinitely:
'10' refers to the number 10
'11' refers to the number 11 ...
But this would defeat the compositional aim. The aim was to
represent how we use the presence of digits in their places to
identify the reference of new numerals. But if each numeral got
its own independent description, as if we continued this list,
that would be no help. It would make it seem as if each piece of
understanding was quite independent of the others, so that a
child would have to be taught each numeral separately, and then
have a skill which wouldn't transfer to unfamiliar numerals. In
other words, we want to explain how it is no accident that, say,
 THE SHAPE OF THE PROBLEMS 13

'219' refers to 219. We therefore want a rule telling us how the

presence of those digits, in that order, determines this fact. A
rule which does it is:
Suppose A, ... , An is a sequence of digits referring to a
number k. Suppose B is an individual digit referring to a
numberj. Then the sequence A, ... , AnB refers to the number
(lOXk)+j.
Thus '219' is the sequence '21' with '9' on the end. It therefore,
by the rule, refers to ( 10 X whatever '21' refers to) + what '9'
refers to - i.e. 9, by the base clause governing that digit. What
does '21' refer to? Using the rule again: ( l O X 2) + 1. Putting the
two calculations together, the original refers to (10 X 21) + 9,
i.e. 219. That's a way of saying how counting to base ten with
the arabic notation works.
The example is trivial, but some of the points it illustrates are
not. Notice that it is not the final result that is news-we already
knew that '219' referred to 219. Everyone who could use our
numerals would know that. The value of the description lies
entirely in showing us why this fact is so. We wouldn't under-
stand how this arithmetical notation works unless we had in
some sense cottoned on to the rule of composition. There are
some examples, such as the Inca use of systems of knots in ropes
as counting devices, which remain uninterpreted precisely be-
cause nobody is sure how complications in the knotting deter-
mined the number being referred to. Some notations would
need quite complex systems of rules - the Roman system of
letters, for instance.
Let us call the notation described by the base clauses and the
rule, the arabic numeral notation: ANN. Then the structure of
ANN has been completely revealed. So in one sense we know
how we refer to numbers. We do so by using that notation. But
we have not provided much of a Christmas present to any
philosopher of mathematics. His question, 'How do we refer to
numbers?', is asking what it is to refer to a number, what
numbers are that they can be objects of reference and what the
ability to refer to o:i.;ie has to do with other skills, such as the
ability to count and add. None of these questions is advanced at
all by the compositional knowledge. This is, of course, particu-
larly blatant since the base clauses simply help themselves to
14 OUR LANGUAGE AND OURSEL YES

the fact that the digits refer to the numbers they do. If this kind
of fact is problematic to a philosopher, then they are merely
examples of the problem. 3
There is one particularly important reason why a phil- ·
osopher of mathematics might pay scant attention to the struc-
ture of ANN. We believe that counting to base ten is just one
amongst an indefinite number of possible ways of counting. We
could have used all sorts of systems, with different numbers of
primitives, different ways of combining them, and different
rules for recovering the number referred to. But these would be
merely alternative ways of doing the same thing - referring to
numbers. The fact that someone uses a differently structured
system does not prevent him from talking of the number 125,
and coming to know the same things as we do: that it does not
divide by 2, has a cube root, and so on. So there seems to be a
clear distinction between what we do - talking about numbers -
and the particular way we do it- using ANN, for instance. But
understanding this distinction in general can be very difficult.
What do different notations have to share if they are to be ways
of talking about the same things, or expressing the same knowl-
edge? Are there limits to the extent to which diverse ways of
expressing things can coincide with sameness of thoughts ex-
pressed?
The compositional nature of language is impressive. A non-
systematic numeral notation, for instance, would only enable us
to count up to some fixed limit. For each new numeral would
have to be learned individually, as the basic ten digits have to be
learned, and we would only have time to master a small number
of them. Similarly, a non-systematic language would enable its
users to say only a fixed, limited, number of things. The mean-
ing of each sentence would have to be learned individually, and
this learning would provide no resources for rearranging words
or other devices to make new messages. Still, counting up to a
limited number is still counting. And saying only a limited
number of things is not the same as saying nothing. It is thus
premature to take the compositional nature of language as the
criterion determining language, so that if a group has no ability
3
For this reason, beware of shorthand titles, 'theory of reference', 'theory of sense',
etc. These titles do not show what has been explained nor what has been used to do it.
See also notes to 1.4.
.
 THE SHAPE OF THE PROBLEMS 15

to separate and recombine elements of messages, it is not using

those elements to communicate. If this ability is to be the litmus
test distinguishing linguistic from non-linguistic skills, we need
an argument that communicating only a fixed, limited, number
of things, and having no ability to do more, either is not
communicating at all, or is at best communicating non-
linguistically. The undoubted essence of ability to use a
language is the ability to express and communicate informa-
tion, desires,·wishes, etc. by its means. Any proposed indicator
of whether this is happening needs to earn its keep. We need
argument that the indicator distinguishes only cases where this
is happening from those where it is not.
Someone might object to the compositional description of
ANN, on the quite reasonable grounds that numbers are not
objects ofreference at all. One might, for instance, suppose that
for us to refer to a thing it has to impinge on us in some way.
Since numbers· are abstract objects, if they are objects at all,
they do not enter into causal relationships, and therefore do not
impinge on us. Hence, we do not refer to them, and the relation-
ship between a numeral and its interpretation would have to be
described differently. We would need a different kind of clause
stating what numerals do, in terms of what digits do. But the
insight we have into the compositional nature of ANN remains.
It simply needs expressing in the preferred style. For example,
noticing that numerals are introduced in counting, when they
describe how numerous are sets of objects, we might try a
semantics couched in terms of their adjectival role. This would
say that 'O' is the digit used to count sets ofO things, ... , '9' is the
digit used to count sets of9 things, and the rule would similarly
just substitute 'used to count sets of ... things' for 'refers to the
number ... '. The compositional insight remains the same. Ever
since the pioneering work of Frege, emphasizing this aspect of
numerical terminology, it has been a matter of some importance
to weigh the merits of each style of description, and this intro-
duces one way in which semantics is of more interest to phil-
osophy than might have so far appeared. Can the adjectival
style be used to compute an interpretation for sentences in
which numerals do not appear to be functioning as adjectives
describing sets, such as '7 is the sum of 5 and 2'; can it be used
to develop interpretations of other sentences which appear to
16 OCR LANGUAGE AND OURSELVES

talk about numbers as general objects of theory, such as 'there is

an infinite number of prime numbers'?
This introduces one kind of way in which semantic theory
influences surrounding philosophical problems. If there are
occurrences of numerals which the adjectival style cannot
cover, they will provide a reason for seeking something more,
and may provide a reason for reverting to thinking of numerals
as genuinely referring to things, such as numbers. A different
example of the same phenomenon might help. Suppose a
theorist notices that 'good' functions somewhat like an adjec-
tive. He might nevertheless have doubts about whether it is
used to describe things or attribute a property to them. He
might be drawn to an expressive theory, which claims that a
sentence 'Xis good' serves some other function than describing
X; its role is to convey the speaker's approval of X, perhaps.
There then arises the question of whether this conception of the
role of that sentence (and hence of the adjective inside it) is
adequate to accounting for all its occurrences. For instance, we
say such things as 'If this is good we should encourage it in the
young', or 'I think that this is good but I may be wrong.' Here it
seems that the embedded sentence is used to do something
other than merely convey the speaker's approval of the topic,
because in the first case it seems to identify some hypothesis,
and in the second case to identify something about which the
speaker has doubts. The question which is internal to the
workings of our language is whether the expressive theory about
the use of the sentence can explain and interpret these occur-
rences of the sentence. But understanding whether it can is no
mere internal, technical issue in semantics. Until we have a
clear view of it we cannot assess the expressive theory, and until
we can do that we cannot form any reliable view about the
metaphysics or the epistemology of ethics. I give the theory of
this issue in 6.2.
These examples show how it is that, to understand the role of
a term in our language, we need some understanding of its
compositional possibilities. A description of.its role which
leaves some occurrences unintelligible, by failing to show how a
word with that role can feature in that context to generate that
meaning, is semantically inadequate. But there remain questions
of what it is about a group of people which makes it true that
 THE SHAPE OF THE PROBLEMS 17

they are speaking and understanding a language which fits a

particular semantic description, of what counts as change in
language, sameness of language, understanding of language in
a certain way. Other questions relate more to the 'world' comer
of the triangle, and can be appreciated ifwe draw a distinction
between needing some semantic description and imposing it. If it
can be shown that we need some semantic description to com-
plete a compositional account, then one aspect of a philosophy
oflanguage is completed. In speaking the language we would be
referring to numbers, or describing things as good, or whatever.
But against philosophical unquiet, yielding epistemological or
metaphysical argument that we cannot be doing these things, a
purely structural argument for saying that we must be doing so
will seem weak. Opposition will continue to look for some other
way of conceiving the interpretation of the contentious terms,
without implying the false conception of how they relate to the
world. We have already seen a small example of how this might
happen, in the different styles of describing ANN, which yet
serve the same compositional aim.
This point can be put another way. A description of the
compositional working of our language is, like anything else, a
description which we offer in the light of our own conception of
ourselves and of the world we inhabit. The compositional aim
controls one aspect of the adequacy of such descriptions. An
adequate compositional theory may describe us as doing all
kinds of things in using the language it talks about: referring to
numbers or possibilities or events or chances or properties of
things; using operators with certain properties, predicates in
certain ways, and so on. But our wider understanding of the
triangle may make us uncomfortable with these descriptions.
There are then a number of courses open. We can seek an
alternative style of description, avoiding saying that numerals
refer, or that evaluative sentences describe things, and so on.
We can criticize the language, saying that it is rightly described
by the given semantics, but that since we do not really refer,
describe, etc., in the way it shows us doing, the language itself
incorporates mistaken philosophical ideas - about numbers or
ethics. In the last ditch we can argue that words like 'refer',
'describe', 'truth', etc., bear different senses in the contested
cases from those which they normally have, so that although it
18 OUR LANGUAGE AND OURSELVES

is correct to allow that numerals refer, or that we refer to

possibilities, chances, or whatever, this is in a rather "different
sense" from that in which we refer to tables and chairs. Discov-
ering how habitable this ditch is, or indeed whether any of these
responses is permissible, involves the wider perspective of the
triangle, and takes us beyond purely compositional concerns.

4. Further Points about Words and Language

A final cluster of points illustrated by ANN is partly terminologi-
cal, but partly more substantive. There could clearly be num-
erical codes, or alternative numerical systems, in which the
written arabic digits,. or their audible counterparts in spoken
English, have different interpretations. The connection between
those signs and their meanings is arbitrary, and, as we shall see,
conventional. The conventions could change. The digit '5'
could come to refer to the number of hands we (presently) have;
the digit '2' could come to refer to the number of fingers we
(presently) have on each hand. (I am told that in modem Arabic
the sign shaped like 'O' refers to the number 5.) The conn&ction
between the actual signs and their interpretations is mutable,
conventional, and contingent - something which could have
been otherwise, had human beings operated differently.
This contingency poses terminological problems and it is
important to be clear about the way we solve them. Do we
identify the notation ANN simply through its lexicon and syntax
( the rule stating that any sequence of digits except those com-
mencing with 'O' counts as a numeral)? If this were so, after the
change described, whereby '5' and '2' change functions, we
would still be using ANN, because we are using exactly the
same terminology. Or should we say that after such a change we
no longer use ANN, but only a related cousin, ANN*: in other
words, should we say that the identity of a language is created
not only by the vocabulary of signs it has, and the way those
signs make up sentences, but also by the meanings attached to
each element? In short, do we define a language syntactically or
through syntax and semantics? It is up to us which decision we
take. But there is no chance at all of clear philosophy oflanguage
unless we remember which decision it was. I regard it as dearer
to adopt the latter convention: ANN is therefore defined as that
 THE SHAPE OF THE PROBLEMS 19

numerical notation which uses those signs in those combinations

to refer to those numbers. A change in this last function would
mean that ANN had been superseded by some related system,
such as ANN*.
The change described results in the sign '2' referring to the
number 5, and the sign '5' referring to the number 2. It is easy to
imagine coming across a population who use this different
system. They would write the numerals in this sequence: 0, 1, 5,
3, 4, 2, 6, ... , and teach children to associate those terms with
the relevantly numbered sets of things ('5: numberofhands you
have, ... ') Such a group would have ANN* as their actual
numerical notation, in the way that we have ANN. So far as I
know, no group does this: ANN and ANN* are equally good
possible numerical notations, but only one is used. In fact there
is an infinity of possible numerical notations. In some the digits
retain the use we give them, but the significance of putting them
into sequences is different (e.g. chains might need reading
backwards, so that '71' refers to 17); some generate the same
interpretations for numerals up to some point that ANN gives
them, then go on to diverge, and so on. Transferring these ideas
to language, we say that there are infinitely many possible
languages, only a few of which are in actual use among human
populations. A misinterpretation of a population happens when
they are assigned the wrong actual language; that is, when they
are interpreted so that sentences and utterances are given one
meanin&Wlren they in fact have another.
It is of course up to us which language we have as our actual
language. It is a matter of our habits, behaviour, ways of taking
the written shapes and spoken sounds which we make. Language
may change as the thoughts, concepts, habits of belief and
attitude, of the population change; in the idiom I am introducing,
this means that one actual language is succeeded. by another.
This highlights the problem of individuation: when are we to say
that a term or cluster of terms has a different sense from that
which it used to have; when can we say that different terms, for
instance from different languages, have the same sense? The
purpose of such questions, in philosophy in general, is not to try
to impose more precision than the shifting phenomena admit of.
The purpose is to find the principles governing the verdict that
terms are being taken in the same or different ways. It doesn't
20 OUR LANGUAGE AND OURSELVES

matter if the principles do not always yield a dear verdict. But it

does matter if we have no grip of what the principles are. For in
that case we do not know what counts as an argument that
groups are or are not speaking the same actual language.
Because the same words could in principle be used to say
different things in different languages, there is an important
difference between merely telling which words someone used
and telling what he said or commanded or thought. To know
the second involves knowing how his words are to be taken -
that is, which interpretation they may bear; or which is his
actual language. To know merely what his words were does not
of itself answer this. For example, Henry may affirm this sen-
tence: 'There are 12 apples on the table', but ifhe uses ANN* he
would thereby say that t~ere are 15 apples on the table. Con-
versely he could have said that there are 12 apples on the table
in any number of ways, since there is no limit to the number of
languages capable of expressing that information. The distinc-
tion is usually known as that between direct reportage, which
tells the words, and indirect, which tells what was said. But this
way of expressing it is slightly unreliable. In normal conversa-
tion when we tell the words someone used we presuppose or
assume that they bear their normal meanings, identified by
whichever language we speak. It is only exceptionally that we
contemplate the possibility that someone is using normal words
in an idiosyncratic way. Indeed, in chapter 4 we meet reasons
why we will not allow that this is happening even when on the
face of it the speaker is indeed using words in a non-standard
way. So normally direct reportage conveys to a hearer not only
which words were used, but also what was said. Nevertheless
the essential point is that these are two different things, regard-
less of how often we convey both together.
It should be noticed that in expressing this clear distinction
in the way that I have, I implicitly treat words in one particular
way. I am treating such things as a sentence, a word, a sign, or a
symbol, as items which can in principle occur in different
notations or languages. Thus I say that the symbol '2' could
occur in ANN* as well as in ANN; that the sentence 'there are 2
apples on the table' could be common to each language, al-
though meaning something different in each. This is a definite
choice, for another way of identifying words and sentences
 THE SHAPE OF THE PROBLEMS 21

tie a word to its meaning. In that sense, ANN* would

contain a different word, '2', from the word looking the same,
which occurs in ANN. In this sense English would contain, say,
four words looking like this: 'ruff'. Whereas in the usage which I
prefer, there is one word, which is capable of four different
meanings (a frill or folds of linen; a male bird of the sandpiper
family; a small freshwater fish; a situation or move in a bridge
game). There are other principles as well for counting words.
An etymologist would distinguish them by history, so that if
the word-in-one-meaning and the word-in-another-meaning
evolved from one common source they would count, for his
purposes, as the same word. And this principle too could be
refined and distinguished.
For the purpose of understanding the original triangle, or in
other words of doing some philosophy of language, there are
advantages, and no disadvantages, in counting or "individuat-
ing" words in the way I favour. This way gives us an easy
description of the difference between identifying the marks or
sounds someone makes and the understanding he has of them,
or the meaning his group gives to them. On the one hand,
words, notation, symbols; on the other hand, their use, inter-
pretation, or meaning, and the connection contingent and
mutable, and laid bare for investigation. An alternative deci-
sion would muddy the way we identify this problem area. If
different languages could not share the same words, or if we
decided that difference of meaning entailed difference of word,
then we would need to understand a man, before we could even
know which words he was using. We would, however, need some
description of what we can identify; the shapes he writes or the
sounds he makes, where these are things which could in prin-
ciple be produced with different meanings. When discussing
such things as the change brought about by revisions of beliefs,
or the relation between ANN and ANN*, it is convenient to
have a simple way of saying what changes and what does not. I
prefer to express things by saying that the words (signs, symbols,
sentences) may remain the same, but that the interpretation or
meaning or understanding of them changes; equally, since a
language is identified by ·which sentences are used with which
meanings, the language changes.
I labour this point, because it is a frequent cause of passion. I
22 OUR LANGUAGE AND OURSELVES

mentioned above the tendency to make features of our language

into the key to understanding conceptual issues about, say, our
minds, the world, necessity, reason, truth. One way in which
this tendency is embodied is by claiming that many issues are
"about words", "matters of definition", "matters of semantics",
"metalinguistic". Issues about these abstract things are brought
nicely down to earth, given a nice concrete subject-matter, if
they are represented as issues about words. But the change of
focus is merely a blurring unless the sense of 'word' which is
intended is made clear. If words are regarded as identified
(individuated, as it is usually put) without regard to their
meanings, then it is not usually true that they are the subject of
these issues, any more than the pieces of paper and metal
actually in circulation are the subject-matter of economics.
Thus, consider a metaphysical thesis, that mind cannot be
matter, or an arithmetical one, that 2 + 2 = 4. These puzzle
thinkers of an empirical bent, since their status and relation to
ordinary experience is unclear. If they are true, they are true not
only of the world that experience reveals to us, but of any
possible world. They have to be true, if they are true. This
inexorable quality is mysterious, and a nice way of removing the
mystery would be to put us in charge of what we take to be
inexorable. Perhaps as pure convention we arrange our mathe-
matical and psychological language so as to give these remarks
their protected status. Necessities emerge as products of our
language, not as impositions to which we have to conform.
Unfortunately, however, this conventionalism looks markedly
less attractive once the necessary distinctions are made. Cer-
tainly, by changing our actual language we could come to
assent to the sentence 'mind can be matter' or '2 + 2 = l O'. We
would get the latter if we adopted ANN* and we would get the
former if we came to use 'mind' to refer to stone or lead. It is
within our ·power to do this. But this is irrelevant, for it does not
show that by speaking a certain kind of language we might
come to believe that 2 + 2 is other than 4, or that mind can be
matter. It simply shows that we have the authority to choose
different ways of expressing things. But this is irrelevant to
discovering whether there are possible or adequate ways of
thinking which deny our cherished doctrines. Certainly a con-
ventionalist can go on to urge that there are, and that it is a
 THE SHAPE OF THE PROBLE'.\1S 23

matter of our own decision that we do not adopt them. But the
difficulty of showing this is quite hidden unless we are clear
about the individuation of words and languages.
So to make it even a half-truth that such issues are "about
words" there is needed a sense in which words are individuated
at least in part by their meanings. In this sense there could not
be a language in which the word 'mind' meant anything other
than mind, or in which the word '2' meant any other number.
Suppose someone then says that the word 'mind' cannot refer to
matter. Does this give the metaphysical thesis some kind of
linguistic or conventional explanation? Not at all. Since, by this
definition, the word 'mind' must refer to mind, it cannot refer to
matter if mind cannot be matter. But there is here no explana-
tion of why this cannot be so. It simply rides on the back of the
definition of the word, and the still untouched metaphysical
thesis. Similarly, we might announce that the word 'claret'
cannot refer to the stuff that comes from the tap, and this is just
because the word must, in this usage, refer to claret, and claret
cannot be the stuff that comes from the tap. And this does
absolutely nothing to suggest that this last truth is a verbal one! 4
Are these distinctions mere nuts-and-bolts, or are more con-
troversial issues in the air? Two points which might prove
worrying should be noticed. One is a matter of common obser-
vation: when we learn another language, there is often no sharp
division between learning which words we are hearing, and
learning which sentences, in which meanings, we are hearing.
In other words, it is only when we have a fair degree of skill at
understanding speech that we can report it directly. The usual
difficulty for the foreigner is not only being unable to interpret
the sentences, but also not knowing what they are - being
unable to repeat what was said. This is disguised by thinking
only of written language. We expect languages from familiar
cultures to be written more or less in our way: we can tell which
German, Spanish, or French words are on the page, even ifwe
do not know what they mean. A word is something which comes
4
This explains why in this work I give no detailed exploration of the problem of
necessary truth, although that is frequently presented as a problem in the philosophy of
language. For the reasons given it-seems to me that there is no theory in the philosophy
oflanguage which would start to explain why no adequate way of thinking could deny
that 7 + 5 = 12, that there are asymmetries between the past and the future, and so on
for other necessary truths.
24 OUR LANGUAGE AND OURSELVES

just after one gap and just before another gap in But we
here rely upon a shared convention of writing. The theoretical
point remains, that if we are working our way into a language
from scratch, in principle there will be no sharp division between
learning to isolate words, and learning the meanings of sen-
tences. The second worry underpins this one. A word, as we see
in chapter 4, is best regarded as a feature of a sentence; or
rather, the presence of a word is a feature which matters because
it is one of the features to which we are sensitive when we
construct the meaning of the sentence. This makes it unsurpris-
ing that in foreign speech we cannot isolate words reliably. To
do so involves becoming sensitive to the features of overall
utterances which determine their meanings, and only a reason-
able acquaintance with many utterances and many meanings
will enable us reliably to abstract out features whose presence
helps to determine meaning. When we can make reliable guesses
about what they are saying, we will also see how they are saying
it. But until we know this we can have no perception of which
words - which meaning-giving features - to identify in their
utterances.
These points deserve respect. But they cast no doubt on the
theoretical possibility marked by allowing that the same word
could have different meanings in different languages. Thinking
of the presence of a word as the presence of a meaning-
determining feature may be right; noticing that we are bad at
identifying such features as long as we are bad at understanding
the meanings of most utterances in a language is also right; it
does not follow that words can only be regarded as intrinsically
connected with their roles. It might take considerable acquaint-
ance with a community to realize that some aspect of a sentence
- say, writing it in red ink- affects its meaning in some determi-
nate way. Before this is realized the feature may simply pass
notice, regarded as part of the " background noise" which we
blot out. We become sensitive to it only upon realizing the
significance of its presence or absence. But this does nothing to
diminish the possibility that two different communities should
use the very same feature to effect different modifications of the
meaning of a communication. Nor does it cast doubt on the
possibility that at a particular stage a hearer may know quite
well that the meaning of a sentence is determined by the pre-
 THE SHAPE OF THE PROBLEMS 25

sence of some feature, but not know what the effect is, and hence
the overall utterance of the sentence is saying. The reflec-
tions of the previous paragraph only suggest that we take care
not to idealize the stages of language learning. In particular a
complete lexicon and syntax of a language cannot be given
before we have any grip on the interpretation its sentences bear.
Common observation shows that this is not a practical division
of our procedures, and the argument I have hinted at, and
develop in chapter 4, shows why not. In this sense semantics
precedes syntax. (There is however the curious phenomenon of
brain injured patients who remain good at determining un-
grammaticality, but are bad at understanding sentences. Pre-
sumably they retain in a different location of the brain kinds of
sensitivity which, on this account, had a partly semantic origin
and significance.)
I earlier mentioned some philosophical questions left un-
touched by the semantic description of ANN. One of them
deserves particular respect. This is the question of what kind of
skill the ability to refer to a number is. We can put the question
in terms of competence. Let us say that a user of the notation is
competent if we can recognize which sequences of symbols
count as numerals, and what their references are if they do.
Similarly, a user of a language is competent ifhe can recognize
which sequences of words (or sounds) are to count as well-
formed and what is meant by any sequence which is so. A
semantic theory of a language, like this one for ANN, would
ideally contain rules enabling anyone following them to derive a
statement of what any particular sentence of the language
means. A fully competent user would know that the sentence
means this. But, again profiting from the example of ANN, it is
obvious that a philosopher will want to probe what this knowl-
edge amounts to. What is it to know what a sentence means? Is it
to be able to tell when it is true? Is it to be able to make sensible
inferences either to its truth, given other facts, or from it sup-
posed truth to other consequences? The bland answer to this
question is that the competent user knows that '219' refers to
219, and so on for any sequence of numerals, or, in the case of a
full language, the competent user knows what any particular
sentence means. What I shall call an external approach to com-
petence probe; further into what this knowledge demands
26 OUR LA:\!GUAGE AND OURSELVES

of someone. The word 'external' merely reminds us that the

internal description of the actual workings of the language does
not itself solve such a question.
To sum up the notions introduced so far. A language is given
a semantics when we can say what each sentence means. We
can do this compositionally by locating which features of a
sentence are responsible for its meaning, and the rules whereby
those features (notably words in various positions) yield the
meaning. A language is thought of as an abstract thing, identified
by its containing given sentences in given meanings. A popula-
tion will use some particular abstract language as their actual
language by indeed using those sentences in those meanings. It
is contingent (something which could have been otherwise)
that words and sentences bear some meaning and not others in
the mouths of any given population. This contingency is equiva-
lent to the fact that the population speaks some language L 1 and
not another one L 2 in which the words and sentences have other
meanings. Finally an external theory of competence hopes to
describe further what has to be true of someone who knows the
meanings of the sentences of his language. This is a distinctively
philosophical query, not settled simply by telling what the
semantic structure of his language is.

5. Rules andPsychology
To end this chapter I shall introduce one difficult question in
the external theory of competence. This is the question of how
the ordinary user of a language stands in relation to its syntactic
and semantic rules. The central question is whether the user
can be said to 'know' the rules; whether they have a 'psychologi-
cal reality'. The question is pressing because it is fairly dear
that any adequate set of such rules (what linguists now tend to
call a grammar for a language) will be horribly complex. So it is
not as if the ordinary user can either himself say what they are,
or even recognize them as reliable once they have been pointed
out. (Consider that someone may be perfectly good at using
ANN but fail to recognize that the rule I wrote for it charac-
terizes the way it works.) So what is the relation between the
user and the rule?
We can distinguish four basic positions. The first two allow
 THE SHAPE OF THE PROBLEMS 27

some "psychological reality" to the rules. They suppose that ifa

correct grammar says that a sentence is to be certified to mean
what it does in some way, then the user who understands it
must have a psychology which somehow reflects this grammar.
The other two deny this: one of them simply denies that any-
thing whatsoever about a user is indicated by the fact that his
language permits some grammatical description, and the other
claims that all that is suggested is neurophysiological, not
psychological.

( l) Chomsky's realism. Chomsky insists that any inference from

grammar to the psychology of the user is empirical in nature. It
will not follow from the fact that a language can be described by
certain rules that the user implicitly or in some other sense
follows these rules. Nevertheless, the user's competence needs
to be explained somehow, and if mention of these rules is to
explain the competence that a user has, then he must in some
sense have mastered them or at least implicitly he must know
them: they must characterize his cognitive system. The natural
analogy is with our perceptual skills. It is a matter of empirical
psychology to discover what are the cues to which we respond in
making utterly straightforward, non-inferential perceptual
judgements. Thus it might be that the cue enabling me to judge
the direction of a sound is the time-lag between arrival of a
signal at one ear and at the other. This is not something I can
discover by introspection: it is my brain, rather than I, that
"computes" the direction from this fact. Similarly, the nature of
the mental rules and representations which underlie a percep-
tion of grammaticalness or the awareness of what a particular
sequence of English sounds means, is the subject of empirical
enqmry.
Two other aspects of Chomsky's realism may be mentioned
here. One is the idea that the ordinary sentence which we have
on the surface is the terminus of a transformational history. It
may be seen as the outcome of processes of change which can
include the deletion of phrases, the insertion of others, the
movement of parts relative to one another. Ifthese processes are
also real then the rules governing our language faculty include
knowledge that they preserve meaning and preserve grammati-
cality - otherwise we would lightheartedly transform sentences
28 OUR LANGUAGE AND OURSELVES

into others which mean something different, or mean nothing.

These rules will be immensely complex. To give a simple ex-
ample, you might think that prepositional phrases can be
moved fairly liberally: 'I get bored in France' means the same as
'In France I get bored.' But 'I didn't have a good time in France
or England' does not mean the same as 'In France or England, I
didn't have a good time.' 5 The rules governing transformations
have to determine why this is so, and which transformations are
permissible to avoid the undesirable consequences. This com-
plexity leads to the second, notorious aspect of Chomsky's
ideas: the large part played by innateness. The central idea is
that of the "poverty of stimulus". This means that after a
relatively small "exposure" to examples, to the vocabulary ofits
native tongue, and to ways of combining the vocabulary, the
child masters an enormous and precise set of rules, adequate to
all the baffling complexity of the syntax and semantics of that
language. According to Chomsky the best explanation of this is
an innate "universal grammar", or biological inheritance of
syntactic and semantic categories and of rules concerning such
things as the movement of phrases, permissible substitutions of
pronouns, and so on. This inheritance encodes the fundamental
features of any possible human language. In learning it is
primed, as it were, with knowledge of the actual exemplars of
such things as nouns, phrases, pronouns, verbs etc. in an actual
language such as English, and the user is then equipped with an
indefinitely elastic range of syntactic and semantic competence.
The innate cognitive structure fills the gap between the im-
poverished, fragmentary input and the rich and extended
output.
The argument from the poverty of stimulus is queried by
philosophers including Putnam, Ryle and Goodman. 6 Ryle for
example urges that we could only talk in Chomsky'.s way of
exposure to a stimulus if we had forgotten what happens in a
normal childhood. An infant is helped, encouraged, guided,
taught, trained; he models sayings on those of others, practises,
impersonates, invents, and so on: "As Aristotle said, what we
have to do when we have learned to do it, we learn to do by
5
Rules and Representations, p. 156.
6
Putnam, Goodman, 'Symposium on Innate Ideas', in The Philosophy of Language,
ed. Searle.
 THE SHAPE OF THE PROBLE'.\1S 29

doing it ... " 7 Ryle castigates Chomsky for representing the

infant as a solitary thinker, unfostered and alone ('Mowgli in
Babel') trying out solitary hypotheses about which noises count
as sentences amongst the strangers around him. If that were his
predicament, unless a kind nature had restricted the number of
possible grammars he could try out, he would have far too many
to choose from. But this is not his task. His task is not to think
and select, but to imitate, to enter wholeheartedly into the ways
of his parents and friends.
Ryle's point is to expand our understanding of what happens
during language learning, to close the gap between what you
might expect the learning to achieve and what it does achieve
( the gap which innatists fill with an endowed structure of
possible grammars - "clouds of biological glory" in Ryle's
view). The gap is also dosed ifwe can see how the apparently
idiosyncratic, shapeless, and arbitrary rules which users appear
to follow may actually have a point. If we can make the rules
involved intelligible, there is so much the less miracle about the
speed with which the learner absorbs them. Making them
intelligible means in particular seeing how the semantic role of
terms explains their syntactic oddities. Consider a simple
example. Ifwe classify phrases such as 'any fish' and 'no fish' as
noun phrases, we might expect them to fill the same grammati-
cal positions. 'You may eat any fish that you catch' and 'you
may eat no fish that you catch' are thus equally well-formed.
Unfortunately, 'if you catch any fish you may eat them' is
well-formed, whereas 'if you catch no fish you may eat them' is
not. How do we master such syntactic quirks? Obviously not by
an innate sense of grammar, but by realizing (roughly) that
given the supposition that I catch no fish, there is no reference
for the pronoun 'them' to pick up, hence no permission to be
given by the sentence. Whereas in the former case there is.
(Precise rules for sentences involving pronouns are still
intensely hard to find.) If there is a point to a rule, the learner's
achievement is easier to understand. By analogy, consider a
theorist trying to find principles of permissible classical com-
position by analysing musical scores, paying attention only to
written shapes. A horrendous task. Yet some musicians
mastered them in early childhood! This is not however because
7
Gilbert Ryle, 'Mowgli in Babel', in On Thinking, pp. 101-2.
 30 OUR LANGUAGE AND OURSELVES

they had an innate grammatical endowment: it is because they

mastered a very simple notational device, and had ears to tell
which of the indefinite kinds of combination were acceptable.
They had no innate propensity to select shapes at all.

(2) This takes us in one direction away from Chomsky's kind of

psychological realism. This is towards insisting that the rules of
semantics and syntax of our language should be conceived not
as hidden and largely innate mechanisms guiding our compe-
tence, but as intelligible, open, means for serving our ends.
They should be discoverable, not by experiment, in the way
that our use of visual and auditory cues in detecting distances
are discovered, but by reflection upon the purposes for which
we use language and the means we have naturally developed to
serve those purposes. The way we use our language is open
before us; reflection on it should enable is to tell how we control
our sayings, and to see the point of the rules which guide us in
doing so.
Chomsky is adamant that his position is the only empirically
respectable stance in this matter. But the second position is
empirical enough, although it may still await development. It
marks a "research programme" of diminishing Chomsky's gap:
of showing how a child's needs and training render it unsurpris-
ing that it conforms to the rules of its native language, since
those rules mark not meaningless and arbitrary restrictions on
patterns of combination of terms, but natural consequences of
the need to communicate. Before coming back to this, I shall
contrast it with the final two positions.

(3) Quine emphasized the important distinction between

behaviour which fits a rule and behaviour which is guided by it.
"Behaviour fits a rule whenever it conforms to it; whenever the
rule truly describes the behaviour. Bl!(Jhe. behayi<:>ur j§_uol_
\guided. by therule unless the _behaver knows the rule and cap.
:\,~t-iM.!;i,fa This beha:'er observes the ~ule. " 8 Thus o~~can
{magine
i 1mmensely comphcated mathematical rules for g1vmg a denota-
\ tion to the sequences of digits in ANN: rules involving
\fxponentiation, rules which take, say, the first hundred numer-
t.8
W. V. Quine, 'Methodological Reflections on Current Linguistic Theory', in
Davidson and Harman ( 1972), p. 442.
 THE SHAPE OF THE PROBLE:\1S 31

als as primitive, instead of only the first ten, and so on. These
rules could end up giving the same numerals the same refer-
ences. But there is no sense in which we qbserve
behavio_l!r: .fitsJfiero;lmfis~~ii-otguided by them. Or consider a
computer chess prpgriJ.mgie. Perhaps it ends up playing the
same way .as a toµniament pl<1.yer,in that the kind of move it
;~Jicfrh:ake in any posit1on is the same as his. The output is the
same. But the rules may be very different. The computer may
get there by brute force: it scans millions of possible outcomes,
and selects according to predetermined features. Whereas the
evidence is that even Grand Masters consider only some thirty
or so positions on their "lookahead tree" (compare the answer
of Richard Reti, asked how many moves ahead he looked in
tournaments: "One ... the right one"). The computer's rules fit
tile COIYIIJ~tence._of Jbe pl.:1.yer, but the playe;
thoS-e ruie~\ - ·- -- -
does
nof follow

. Q1.1in-t_11s_e~ !hi_s_ci_istinction to cast doubt on any notion of

guid3;nselJ.yjm,pli_c~~ruie.s, Even if we had an adequate set of
syntac:tic and sem9-r1tic rules for English,. they could stand to
the ordinary u.ser as the computer's programme stands to the
ordinary player. There is no reason to believe that the user
.emboaies •those-rules, even implicitly. If we imagine two diffe-
rent sets ofrules issuing in a pairing of the same meanings with
the same sentences, Quine would deny that it makes any empiri-
cal sense to ask which of them guides us. Chomsky is well aware
that there is an issue here. 9 He does not simply confuse the idea
of an adequate grammar with the idea that its rules are realized
in the users of the language. But he thinks that empirical
evidence can favour the view that we are implicitly guided in
one way rather than another (after all, it is empirical evidence,
for example from skill at lightning chess, which suggests that
players do not go through millions of computations). Quine's
position is that this is not so. Questions of the psychological
reality of semantic and syntactic rules guiding a user in his
recognition of meaning and well-formedness are not to be
raised. The most that we can do is generate a set of rules such
9
He is especially scornful of philosophers who (a) fail to see that the issue is an
empirical one, and.(b) therefore take him to be giving the wrong answer to the question
whether it follows from the fact that a grammar is adequate that it is also guiding us;
E.g. Rules and Representations, pp. 129 ff.
32 OLR LANGUAGE AND OURSELVES

that if someone follows them, they are competent with the

language as we are; you can never use this and other observa-
tions as evidence that these are rules which do actually guide us.
It will be seen that on this third_J!QS.ltlQ!1§~mantica~synfactic
rules play no part in explaining our competence~ It is sometimes
saiq thaJ they would m_erely 'mo.del' our competence, or give a
'theoretical representation' of it. But this is just an unclearway
of saying that ifwe knew the rules and followed them, we would
end up knowing what we do know about what it is that sen-
tences mean.

(4) The fourth position finds the missing link between us and
the rules in neurophysiology. The idea is that our brains have a
causal structure. Some "bits" are responsible for some aspects
of our competence, and other "bits" are not. This causal
structure could be found to be the same shape as the structure of
some system ofrules, in the following sense. Let us say that a set
of semantic and syntactic rules for English would be crippled if
some particular rule or rules or axioms saying what particular
words or other features do, were removed. The removal will
mean a definite impoverishment: sentences which needed just
those rules for their interpretation will be lost. For instance, if
there is a rule permitting just a certain movement of noun
phrases within sentences, and it is removed, sentences depend-
ing on that operation will no longer be permitted. But others,
not dependent on that rule, will remain. So a particular crip-
pling would have a pattern of effects. Now we might find that
some particular damage to a person's brain produced just the
same pattern of effects: after removal of some bit he cannot
understand the very sentences which a particular crippling also
deletes. This would be empirical evidence that the rule or axiom
is actually embodied in the user's neural processes. The evi-
dence could of course be extended: for example, it might be that
in some semantic set-up a quick computation of what a sentence
means could proceed using some rules, but a circuitous one
could arrive at the result without it; equally a particular injury
might lead to a much slower comprehension of the sentence.
Ultimately a structure underlying the user's competence might
be found which is, as it were, causally the same shape as a
 THE SHAPE OF THE PROBLEMS 33

semantic system. 10 This makes the semantic system a correct

description of the speaker.
The fourth suggestion need not be incompatible with
Chomsky's psychological realism. It may be that we should
allow a "psychological reality" corresponding to the neural
operations, so that talk of implicit knowledge, inherited gram-
mar, representations of grammatical categories, etc. would be
only a way of saying that our brains have the suggested causal
structure. But if so, it is a bad, mysterious, and potentially
misleading way of saying it. For whilst it sounds exciting and
strange to say that an infant inherits a universal grammar, it is
not nearly so strange if this translates into saying that the infant
has a brain which under normal conditions of development will
come to have a causal structure of the kind described.
An analogy due to Quine helps to locate the issues. 11 We can
compare our competence in telling the meanings of sentences
with the visible foliage of a tree. The 'grammar we use' would be
the supporting branches and twigs. Quine's position is that the
shape of the foliage does nothing to indicate the underlying
structure of branches and twigs. Chomsky's position is that it
does: evidence, for example from the pattern of growth of the
foliage can confirm particular structures. The fourth position is
that if we shot bits out we could watch which patterns offoliage
fall, and that this would be evidence that our language is
structured as described by some grammar, although not evi-
dence that we "implicitly follow" the rules of that grammar.
The second position is that we are like squirrels constantly
scampering around the structure. We know what it is because
we know our way around it.
Although the fourth position sounds initially attractive, I
think it is not adequate. The idea is that it is the causal structure
of our brains which creates or constitutes the fact that our
language has some particular semantic structure. But it might
be that our brains are like holographs, which do not encode
information bit by bit. It is simply not true of them that if you

10
Gareth Evans, 'Semantic Theory and Tacit Knowledge' in Holtzman and Leich
(1981), p. 127. The position is close to that described in Martin Davies, Meaning,
Quantification, and Necessity, ch. 4. See also L. Fricker, 'Semantic Structure and
Speaker's Understanding', Proceedings of the Aristotelian Society ( 1982).
11
Also used by Davies, op. cit., p. 82.
34 OUR LANGUAGE AND OURSELVES

destroy a particular area you lose one particular piece of infor-

mation. You only get a somewhat less good overall picture. If
we were like this - and it is not of course a matter of armchair
theorizing to tell whether we are or not - then nothing would
correspond to destroying bits of foliage by knocking out particu-
lar twigs or branches. And then on this account it would not be
true that our language has a particular semantic structure. For
example, suppose it turned out that there is simply no
neurophysiological way of destroying someone's competence
with the numeral '7' (his awareness that it refers to 7) by itself.
No pattern of damage is isomorphic with the 7-crippled num-
eral system which would be left if that numeral were deleted
from ANN. On this account it would follow that the given rule
was not the correct one, the one which describes the actual
structure of the numeral system. And that seems incredible. It
seems right to say that we know that '719' refers to what it does
because of the presence of '7' in its actual place, although we do
not know that any 7-crippling, localized brain damage could be
brought about. Whether it can be brought about is pure specu-
lation, but it is not one on which our knowledge of how the
numeral system works depends. We know that because we can
control the reference -we can substitute digits at will to refer to
what we like - and we can see how we accomplish that control.
If this generalizes to natural languages, it establishes the second
position, that the way we use our language is open to view and
ought to be capable of being known by reflecting on our control
of our devices. This damages the analogy which is sometimes
made between recognizing meaning in a remark, and interpret-
ing a visual presentation. In perception we may be quite ignor-
ant of the cues which we use (or, more accurately, which our
perceptual mechanisms and brains use) in order to recognize
what we are seeing. But just for that reason we are not good
either at controlling those cues. In advance of theory, most ofus
cannot produce perceptual illusions; we do not know how to
arrange the significant fe2.tures in order to make it look to
someone as though he is, for instance, in a tapering room when
in fact he is not. We would need expert help. Whereas in
communication we are extremely skilled at selecting and
organizing exactly the meaning-determining features which
result in our messages communicating what we want.
 THE SHAPE OF THE PROBLEMS 35

The following argument is also relevant. Consider a simple

language containing only twelve names and twelve simple pre-
dicates, any of which can be applied, truly or falsely, to any of
the named things. 12 There can only be 144 sentences. A com-
positional semantics could proceed like this: it could say what
each name refers to ('Fred' refers to Fred ... ) and which feature
each predicate ascribes to a thing ('bald' applies to bald
things ... ) Finally a rule would say that the result of placing a
name next to a predicate is a sentence saying of the thing named
that it has the feature expressed by the predicate. There are
twenty-four axioms, and one rule. But we could have given a
wooden list correlating each of the 144 sentences with a mean-
ing. From Quine's standpoint each is an adequate semantics for
the language; there is no real question of which one is embodied
by speakers. From the first standpoint, learning theory can tell
us that one is better: in particular we can see whether a child
knowing what 'Fred is bald' means, and knowing what 'John is
fat' means, may see without further training what 'John is bald'
means. This confirms the compositional account. It is not true
that each sentence requires a separate, independent, descrip-
tion. The fourth position relies on an underlying neuro-
physiology. It urges that damage would knock out, say, twelve
sentences (the 'Fred' sentences) at a time, or, if it knocked out
one sentence at a time, this would make the wooden view true.
But these views are all lacking any connection between the
structure and the very fact that the sentences mean what they
do. They allow that in principle, if the learning or the
neurophysiology were different, the wooden account might be
true of one part of the population, and the compositional
account of the other, yet they would communicate and under-
stand each other perfectly. Yet what reason could there be for
allowing that a person means that Fred is asleep by some noise,
unless the noise is structured in a way which permits us to see it
as made of a name for Fred, and a predicate for being asleep? To
use a sentence to mean that Fred sleeps is to have a certain
understanding of it: that it refers to Fred, and says that he is
asleep. Now indeed in a code a sign might be dictated to
perform these two functions without itself containing elements

12
Evans, op. cit., pp. 122-3.
36 OUR LANGUAGE AND OURSELVES

which correlate with either. 'Zzz' means that at three minutes

past ten the contact will come out of the second alley ... But the
code is parasitic upon another language where different ele-
ments have the different functions. If a community had no such
language, it is plausible to say that they could not take sen-
tences in the alleged way (see also 4.5, 4.6). In that case the
wooden list for each of the 144 sentences misinterprets any
population which does indeed use them woodenly. It pretends
they are saying complex things when the absence of any ele-
ments corresponding to the complexities proves that they are
not.
The result is that you cannot speak a natural language (a first
language) and mean by a sentence that Fred sleeps unless you
can take an element of it and use it to say other things about
Fred, and take another element and use it to say that other
things than Fred are asleep. This is not a natural fact about us,
but a philosophical truth about the conditions to be satisfied if
we are to mean something by an utterance. 13 But now, how
could you have the right kinds of skill with the elements of such
a sentence without being in a position to know that you have
them (by being made aware of the different powers and
capacities for redeployment the terms have)? You may not have
thought quite which semantic categories are needed to describe
your language. But it could not be wholly surprising to the
competent user, possessing the skill to redeploy 'Fred' and
thereby certified as understanding 'Fred sleeps' to mean that
Fred sleeps, that the terms has something particular to do with
Fred.
If this argument is generalized, it again favours the doctrine
that our control over our language, the way we can deploy
words to generate understood effects, is the fact making it true
that our language is semantically structured in a given way.
Even if we cannot ourselves describe what we do, an adequate
description must answer to our own knowledge of our own
skills: it ought to be recognizable as a description of what we do.
We know that the description of ANN is right, because we can
see it in the way we create and respond to any numeral we wish.
13
Strawson, Individuals, p. 99. Evans, The Varieties ef Reference, 4.3, emphasizes the
same point about thoughts. But the principle which Evans calls the generality con-
straint is stronger than anything I am relying upon.
 THE SHAPE OF THE PROBLEMS 37

Even if this argument is accepted, it certainly does not prove

that the second position is right. The second position, like
Chomsky's, represents an empirical belief: in this case the belief
that we can find pointful rules underlying our grammatical and
semantic competence, and narrow the gap between what hap-
pens during learning, and the competence we end up with. This
is a more optimistic belief than its three competitors. But that
does not ensure that it is true.
This chapter has been a bare introduction to many issues
which are subsequently considered again, and has toured a
large territory. Problems of truth, of speakers and their under-
standings, of semantics, and of psychology all jostle their way
around the triangular path. Crowds have a bad way of shutting
out the light. So I now turn to a more detailed view of what a
language might be, and how we are to think of ourselves as
understanding one.

Notes to Chapter 1
1.3 'not provided much of a Christmas present .. .'
The numerical example is intended to be quite important, in enabling
the reader to avoid confusions of the 'theory of .. .' kind. Nothing is
more discouraging than reading about the theory of force, the theory
of sense, the theory of meaning, the theory of reference, etc. without
having much grasp of the problems these theories set out to explain,
and the devices used to do the explaining. Of course, much phil-
osophical energy goes into discovering good questions to concentrate
upon, but the student can hardly appreciate that unless he knows
accurately which questions are the ones being tackled at any given
time. This section is intended to arm the student against automatic
and parochial assumptions that it is just known, and beyond con-
troversy, that a theory of meaning ought to be doing just this or that.
This is as sensible as supposing that there is just one thing which a
theory of beer or a theory of boots ought to be doing. It all depends on
what we want to explain, about beer or boots or about meaning.

1.4 'to make features of our language into the key ... '
The logical positivists developed a conventionalist approach to neces-
sary truth, and the belief that linguistic arrangements somehow
organize our world-view, and therefore determine what we count as
unalterable and inexorable, is also implicit in Wittgenstein. Students
might try:
38 OUR LANGUAGE AND OURSELVES

A. J. Ayer, Language, Truth and Logic, ch. 4.

W. C. Kneale, 'Are Necessary Truths True by Convention?', Pro-
ceedings of the Aristotelian Society, Supplementary Volume ( 194 7).
E. J. Craig, 'The Problems of Necessary Truth', in Blackburn
(1975).
B. Stroud, 'Wittgenstein and Logical Necessity', Philosophical Re-
view (1965).
C. Wright, Wittgenstein on the Foundations of Mathematics, esp. chs.
22-5.
C. Lewy, Meaning and Modality, chs. 1-5, for a general attack on
conventionalism.
The work of Quine and Wittgenstein (especially the Remarks on the
Foundations of Mathematics) is also important in this connection, but
better read in the light oflater chapters.

1.5 Although the topic of this section is central to our conception of

ourselves as language users, the literature is not encouraging. Many
discussions proceed in the light of particular compositional theories\
and particular views about the indeterminacy of grammar, and may
be better read after chapter 8. Apart from the works mentioned in the
text footnotes, the following might be consulted:
N. Chomsky, 'Knowledge and Language', in Gunderson (1975).
C. Graves et al., 'Tacit Knowledge',Journal of Philosopfg (1973).
S. P. Stich, 'What Every Speaker Knows', Philosophical Review
(1971).
- - 'Grammar, Psychology and Indeterminacy', Journal of Phil-
osophy (1972).
 CHAPTER 2

How is Meaning Possible? (I)

Still, thou are blest, compar'd wi' me

The present only toucheth thee
But och! I backward cast my e'e
On prospects drear!
An' forward, tho' I canna see,
I guess an' fear!
Robert Burns, 'To a Mouse'.

1. Representation and the Dog-leg

The best philosophical problems arise in the following way. We
get a sense of what the world is like, what it must be like. We see
it as made up from some kinds of thing, permitting some kinds
of arrangement, capable of making up some kinds of fact or
states of affairs. Then, we find that there are judgements we
make, commitments we enter into, which seem not to fit at all
into this picture of the world. We are at a loss to "give an
account" of them, or in other words to imagine a kind of fact
which could make them true, or whose absence might make
them false. Thus a scientific world view might picture the world
in terms of an evolving spatial distribution of particles or forces:
how can this dance of atoms make room for consciousness,
agency, causation, or value? How can it even allow my particu-
lar perspective on the world, from one place and one time?
In this chapter and the next I try to introduce the problem of
gaining some conception of what meaning is: how is it even
possible for a world - a natural world of things in space and
time, and made of flesh and blood - to contain some things
which mean other things? As Burns notes, in our thoughts we
represent to ourselves the way the world is, or was, or will be.
We think of absent things, events, and stares of affairs, and
perhaps believe in them or desire or fear them, or hope they do
not exist or will exist. This is the 'intentionality' or directedness
40 OlJR LANGUAGE AND OURSELVES

of thought. With our language we can express judgements,

desires, or hopes about what we thus conceive, communicating
with another. But our words bear no natural connections with
whatever it is that they signify. It is our use of them, our habit of
taking them in one way and not in another, which confers their
powers on them, just as our arrangements confer their powers
on pieces of chess, or tokens which serve as currency. So we
need to say something about the relation between language and
thought, on the one hand, and language and things on the
other: to complete the original triangle, in other words.
The central and obvious truth about words is that we under-
stand them: we confer their powers, know how to use them,
make them work for us. The difficulty, then, is to gain some
appreciation of the kind of fact this might be. What kind of truth
is it that I can use a word to mean something, or express a
thought about something? How can any world make room for a
fact of this sort? I come at this by describing the pressures which
tempt philosophers - and not only outdated, classical philoso-
phers, but also the most up-to-date investigators-into what we
call a dog-legged theory. In this, words are thought of as
reinterpreted into another medium, such as that of Ideas,
whose own powers explain the significance words take on. This
idea, I shall argue, is destroyed by considerations which are by
now quite familiar in modern philosophy of language. But I
shall go on to suggest some essential similarities between this
discredited approach, and others, due to Quine and Dummett,
in which meaning is explored through thinking about interpre-
tation, or the manifestation of meaning to each other. Classi-
cally, we would understand meaning by mapping words back
onto ideas; in this approach we do so by interpreting them back
into a home medium or language which an interpreter brings
with him. And the obstacles and limitations may prove to be
similar in each case. Having explored this, I turn in chapter 3 to
a more direct approach to meaning, which seeks to understand
what it is to use a word in a principle-governed way, without
relying upon any illicit comfort in the way ofreinterpretation of
words back into some other currency.
Aristotle 1 supposed that men confronted by things external
to their minds form mental likenesses of them. Spoken words
I
De lnterpretatione, 16"3.
 HOv\' IS MEANING POSSIBLE? (!) 41

are signs of those mental likenesses, and thence derivatively, of

things in the world. On this view words, as Hobbes put
"signify the cogitations and motion of our minds". The classic
statement of this view in modem philosophy is Book III of
Locke's Essay Concerning Human Understanding. It is treacherous
to suppose that throughout the Essay Locke meant one definite
kind of mental representation of things by his term 'Idea', but at
this point he dearly follows Aristotle and Hobbes. He begins
chapter II of this book:
Man, though he have a great variety of Thoughts, and such, from
which others, as well as himself, might receive Profit and Delight; yet
they are all within his own Breast, invisible, and hidden from others,
nor can of themselves be made appear ... Thus we may conceive how
Words which were by nature so well adapted to that purpose, come to
be made use of by Men, as the Signs of their Ideas; not by any natural
connexion that there is between particular articulate Sounds and
certain Ideas, for then there would be but one Language amongst all
Men; but by a voluntary imposition ... The use men have of these
Marks, being either to record their own Thoughts for the Assistance
of their own Memory; or as it were to bring out their Ideas, and lay
then before the view of others: Words in their primary or immediate
Signification, stand for nothing, but the Ideas in the Mind ef him that uses
them ... [Locke's italics.]
Locke had an argument for giving Ideas this prime position
in the theory of understanding. A man cannot make a word into
a sign, but not a "sign of his Ideas at the same time", for
"Words being voluntary signs, they cannot be voluntary signs
imposed by him on things he knows not. That would make them
Signs of nothing, Sounds without Signification." The premise of
this argument, that we cannot make signs signify something
unless we in some sense know about that thing, surfaces in
many forms, and we shall encounter it again. But Locke's use of
it in this argument depends upon a suppressed premise: that it
is only our ideas which we "know". This comes as a surprise.
We might accept that we cannot make words into signs of things
about which we do not know, but claim that we know a good
deal about chairs and tables, numbers and electrons, rights and
duties: the whole world of common sense, or of scientific,
mathematical, or ethical theory. Why the restriction to our own
Ideas, conceived of as private, hidden, and invisible to others?
42 OUR LANGUAGE AND OURSELVES

Clearly, Locke is arguing within a generally representative

theory of knowledge. We do know about external objects, other
people, their thoughts, and perhaps some more exotic entities
as well. But we know them only derivatively, through the
immediate acquaintance we have with our own Ideas of them.
So perhaps his argument is best restated in terms of what we
in some sense immediately or primarily know - words must
immediately or primarily signify objects of immediate or
primary knowledge; but these are Ideas, secret, within our own
breasts, hidden and invisible. Notoriously, this theory of
knowledge makes it difficult (most philosophers would say,
impossible) to have any reason to suppose that we know any-
thing at all about the world: we cannot tell whether an Idea
adequately represents the world ifwe only have the Idea to go
on, any more than we can tell whether a painting is a good
likeness if we are denied any chance of comparing it with the
sitter. Hence the Idealist response, made in this instance by
Berkeley, that the knowable world must be shrunk to within the
immediate objects of knowledge - the private, hidden, personal
ideas. This is an argument within the theory of knowledge. But
it is essential to see that the parallel move applies with just as
much force in the theory of understanding. For how can I so
much as get the thought that my words can signify not only a
hidden, private, personal item within my own breast, but also
some external object or feature of the world, such as a chair?
Certainly I have the idea of the chair. But that is just an idea. My
words can, let us suppose, stand as marks of such a private thing
- there is still no way of understanding how they point through
such things to a different range of items for which they deriva-
tively stand. Thus the second stage of this dog-leg turns out to
be impassable. Whenever we try to understand our words as
referring beyond the circle of our ideas, we can do no more than
present further ideas to ourselves. It is as if we tried to get
someone to understand that a portrait might represent a sitter,
but just by showing more and more portraits; or as ifwe tried to
get someone confined to a cinema for all his experience to
understand that the films represent objects other than
themselves, but could only attempt this task by showing more
and more films.
At this point it might be suggested that the highly visual
 HOW IS MEANING POSSIBLE? (l) 43

model, according to which Ideas are something like private

reproductions ofreality, is letting us down. The presence of the
reproduction is, it turns out, no explanation of how words can
be understood to refer through it, to the world which it itself
reproduces. The reproduction must need itself to be taken as
representati\;e of the external world, and we so far lack any
account of what it is to so take it- how we can understand it to
point beyond itself. But there is a more general moral to be
drawn, which will apply to theories which do not postulate a
particularly visual intermediary, or one which in any very
natural way reproduces features of the thing signified. The
general moral is that any dog-legged theory needs to avoid what
we might call the "regress-or-elephant" problem. The form ofa
dog-legged theory is that we understand the way in which
words have significance in two stages: they are associated with
elements of an interpreting medium (in this case, Ideas), and
the elements of this medium have their own representative
powers. They have "lines of projection" onto the world
whereby they signify aspects of it. The argument we have just
gone through highlights the problem on this second part of the
leg. We face a regress of interpretations ifwe need to introduce
another medium whose powers explain the powers of any given
medium. And we are in danger of not advancing at all if the
powers of elements of the medium to signify things are left
unexplained. This would be like the theory which explains why
the world stays where it is in space by its being carried on an
elephant, and changes the subject when asked what carries the
elephant. Of course, any explanation must start somewhere.
But it is pointless, a mere shuffie, to introduce an element which
can be seen to require just the same kind of explanation of the
original. So a dog-legged theory must always show how it
avoids the regress-or-elephant dilemma. Notice that this is not
an impossible demand. The intermediary may be of such a kind
that it is in some way easier to see how it can have significance,
and by its presence help to explain how words do.
The line of thought leading to a dog-legged theory is this. The
words of a natural language like English have no intrinsic
connections with what they signify. We could use them to mean
different things, were we to come to speak different languages,
or we could use them like parrots with no understanding at all.
44 OUR LANGUAGE AND OURSELVES

We need to know what they signify, if we are to understand

them. But the argument goes, knowing what they signify
demands that we have some way ef representing to ourselves what
they signify. So consider what we must demand of this represen-
tation. Suppose it were merely the substitution of some different
words for the originals. Then the problem arises again about the
substituted words (the regress). If they are not understood,
their presence is irrelevant. If they are, then there must be some
further way of representing what they stand for. So the condi-
tions on anything which is to serve as the intermediary tighten.
If it is not merely to be a shuffle, introducing just another
demand for a way of representing to oneself what it represents, it
must be essentially representative - that is, its mere presence to the
mind must itself guarantee that we know which item or feature
in the external world is thereby represented. Its connection
with a feature of the world needs to be, as it were, transparent,
so that there is no further act of interpreting it one way or
another. If its own connection with whatever it represents were
mutable and contingent, in the way that the connection of
words with things is, then there would exist the possibility of
taking it one way or another; there would then be a need to show
what determines how it is being taken. But that would intro-
duce another way of representing the candidates, so that one
could ask whether one took the original to mean this one or that
one. This would lead to the regress of interpretations. At every
stage, therefore, if the connection is variable or needs to be
established, it introduces the demand for a manner ofrepresen-
tation which is guaranteed, whose mere presence ensures that it
also represents the right thing. There must be a medium which
carries its own interpretation with it, so there is no possibility of
misunderstanding what is thought of, once it is present.
Furthermore, the possibilities and guarantees talked of here
need to be extremely strong. For suppose it is a merely natural
or contingent matter that once a human being has one of these
essentially representative items in mind, he is thinking of, say, a
chair. There is then the bare logical possibility that thinkers ofa
different, rather inhuman, nature should have the same items in
mind but be thinking of, say, a bed. But what difference can this
be, on the theory? The inhuman character is to take the item
one way, whereas humans take it the other. But the essentially
 HOW IS MEANING POSSIBLE? (1) 45

representative nature of the item is there precisely to plug this

gap: there is to be no further fact that it is taken in one way or
another. I ts presence supplies the way it is taken, for otherwise we
never get an interpretation of the word which we understand by
its means, but only a regress of substitutes, each posing the
same problem.

2. Images
Images were once popular candidates for the intermediaries
now sketched. For the presence of an image in the mind might
well seem to guarantee intrinsically that a particular object, or
feature of the world, is represented by its presence. Images have
another nice property, in that our ability to create them, as faint
copies or reproductions of experiences we have had, just as
photographs are copies of scenes which the camera points at,
seems to be a hice, relatively unmysterious kind of ability. It
promises an easy theory of how we come to understand words,
as well as a simple story about what that understanding consists
in. Unfortunately it is inadequate. For the presence of an image
far from guarantees understanding, even in any weak sense,
and its absence far from guarantees lack of understanding.
We can sense that something must be wrong. For how is a
world in which some things mirror others, or possess mirror
images of others, also a world in which anything means any-
thing? If we imagine a landscape with only unthinking,
unmeaning things in it (including perhaps creatures which
move and interact with things), it is impossible to see why
adding mirrors and pictures, even placing them in the internal
parts of the things that move, introduces significance. The
landscape and its pictures may reflect each other perfectly well,
but this in itself gives us no ground for saying that the pictures
or their possessors are thinking offeatures of the landscape, any
more than that the features of the landscape signify the features
of the picture. This general inadequacy is usually held to be the
problem with the very static picture or diagram theory of
meaning which Wittgenstein espoused in his early work and
later rejected. We can understand the general inadequacy by
working through some arguments.
Firstly, there is bound to be a gap between having an image
46 OUR LANGUAGE AND OURSELVES

on one and taking it to represent one or another

feature of the world on the other. Wittgenstein makes the point
with a pithy example: imagine an old man climbing up a hill.
Concentrate upon your image carefully. Now, might it not be
an image of an old man sliding backwards down a hill? That it
was an image of one and not the other is not a matter of
anything intrinsic to it, guaranteed by the image, but is a matter
of the way it is taken. So its presence as the final explanation of
what it is to take words (say, the words 'an old man climbing up
a hill') one way or another is redundant. It itself requires
interpretation as much as the words do. The point is that
however naturally it comes to regard a picture in one way or
another, as a picture of this, or a sample of that, or as represent-
ing one kind of thing and not another, this is in effect adopting
one "method of projection" rather than another. But then the
presence of the image is not by itself an explanation of what it is
to do that, and the dog-legged theory using images collapses.
This kind of argument was anticipated by Berkeley in his
famous attack on Locke's account of "abstract ideas". The
problem concerned the general significance of words. Consider
my understanding of what it is for a shape to be triangular. If
this is a matter of associating the term with an image standing
as an intermediary through which I understand the term, what
properties must the image have? Since it is in some sense to
represent any triangle, perhaps it had better not be too specific,
having some particular geometry. If it were, for instance, right-
angled, then I might take it to represent only that kind of
triangle, in which case I would misunderstand the general term
'triangle' by using it. Perhaps then I should in some sense blur
it, making its particular geometry indistinct, so that it is neither
definitely one kind of triangle nor another. But now its rep-
resentative capacity is threatened in another way. Perhaps it
represents only blurred triangles. Or, suppose it is like a photo-
graph sufficiently indistinct to be plausibly taken to be a picture
of any triangle. Then it would also plausibly be taken to be a
picture of something which isn't a triangle at all, such as the
segment of a circle, or a trapezoid. Berkeley saw the problem.
He also saw the right thing to say, which is that in so far as one's
thinking about a triangle involves an image at all, the image is
to be taken to "equally stand for and represent all rectilinear
 HOW JS MEANING POSSIBLE? (1) 47

triangles whatsoever, and is in that sense universal". Berkeley

innocently concludes, "All of which seems very plain and not to
include any difficulty in it. " 2 This is fine so long as we forget that
the whole point of the theory is to give an account of what it is to
take a sign in one way or another, so that once the proposed
intermediary, in this case images, is shown to need this sup-
plementary comment, there is no theory left. It is as though I
accept a challenge to make a film illustrating the truth that all
triangles have some property or other. I can show long, fat,
right-angled, obtuse, isosceles triangles; I can blur my images, I
can superimpose some upon others. But I cannot do anything
which suggests that these exhibitions are to be taken to be
representative of all triangles whatsoever. I can only rely upon
the audience to "catch on". Or, I can add the words 'what is
shown here applies to all triangles'. But images were introduced
as the very intermediary to explain how we understand such
words in the first place. So they cannot now rely on words for
help. And if we rely on the audience to "catch on" we still lack
any explanation of what it is to do that. It can't be merely
rehearsing the film just shown, and it cannot be just repeating
some set of words, since words need understanding in one way
or another.
Probably it is obvious, once pointed out, that gazing at a
public picture of a thing is not the same as thinking of the object
or features of the world pictured. A picture can prompt one to
think of many matters, and ones which resemble or fail to
resemble aspects of the picture itself in any variety of ways. The
same is true of mentahmagery. This is not to deny that it may
be useful, as well as pleasurable, to turn over images of things as
one does one's thinking. An image can indeed be taken to
represent any triangle, or any man, or any unemployed con-
temporary teenager, or whatever, and the presence ofa compel-
ling image, in the head or on paper, can help all kinds of
thinking. It is just that the presence of the image is not sufficient
for the thinking to be going on. Nor does it logically determine
the character of any thinking that is going on. 3
2
Berkeley, Introduction to the Principles of Human Knowledge,§ 15.
3 I do not want to claim that imagining is only a matter of the presence in the mind of
representations. Images may themselves be intrinsically representational. They are
directed at the world. But then we cannot use the fact that they are like this to explain
how anything can be like this. In the tradition I am attacking, the pictorial quality of
images was held to explain this.
48 OCR LANGUAGE AND OURSELVES

The insufficiency of images is evident again if we turn from

thinking about things and features of things - thinking about
triangles or men or climbing hills - and thinking that something
is the case. A scene is not itself an assertion or judgement to the
effect that something is so. Neither is a picture, however natur-
ally is might come to us to make some judgements in the
presence of some pictures. We need not even accept that any
state of affairs obtains which in any way resembles what is
depicted. And to judge that one does obtain which in some way
resembles what is depicted demands interpreting the picture:
selecting features, generalizing, abstracting, conceptualizing.
Finally, an inability to image some scene is not the same as the
inability to make a related judgement: ifl ask you to imagine a
bed without an attractive member of the opposite sex in it you
may, temporarily, be unable to do so. But you do not temporar-
ily lose your ability to understand, and believe, that there are
such beds.
How general is this argument? We started only from the need
to take words in one way or another, if we are to understand
them. This became the need to have a way of representing to
ourselves what it is that they stand for. This became the need for
the intrinsically representational medium, in order to halt the
regress which develops if there is the possibility of taking the
representations in one way or another. Images are suggested as
a plausible kind of candidate, for their connection with the
appearance of things, which they literally re-present, is especially
natural. But even they are not good enough, for a picture too
needs definite interpretation, and its presence is not sufficient to
guarantee any particular way of taking it. Perhaps then nothing
can fill the role: no thing can halt the regress ef interpretations. So
where has the argument gone wrong? We can come at this by
seeing why images, at least, cannot be necessary for thinking
either.
Wittgenstein puts the case against the role of images in a
theory of understanding like this:
If I give someone the order "fetch me a red flower from that
meadow", how is he to know what sort of flower to bring, as I have
only given him a worrP.
He imagines someone using the word to determine a mental
 HOW IS MEANING POSSIBLE? (!) 49

picture of a red flower, or even a red square on a physical colour

chart carries about, and then finding a flower which
matches it:
... But this is not the only way of searching and it isn't the usual way.
We go, look about us, walk up to a flower and pick it, without
comparing it to anything. To see that the process of obeying the order
can be of this kind, consider the order "imagine a red patch". You are
not tempted in this case to think that before obeying you must have
imagined a red patch to serve you as a pattern for the red patch which
you were ordered to imagine. 4
The point is not just that we do not feel any such second-order
pattern. It is stronger than that. The image cannot be necessary
to understanding the order, because it can only figure in an
explanation if we understand that it is the right image. But that is
just as blankly mysterious as understanding that you have the
right flower. Knowing which is the right kind of flower is not
explained by knowing that the flower matches some given
pattern, unless we also know that it is the right kind of pattern.
But recognizing patterns to be the right ones is no easier than,
or different from, recognizing flowers. The point is the more
vivid if we take a colour term which is slightly unfamiliar:
magenta, or puce. HI ask you to pick me a puce flower, you may
imagine various shades. But if you do not know which of them is
puce, you cannot obey me, except by luck, by matching a flower
to a particular shade. On the other hand, if you do know which
of them corresponds to puce, then you know which flower does
in the same way - whatever way that is. So the detour via
intermediary images is useless. And this point has wider appli-
cations. For any dog-legged theory faces the objection. We need
to know that we are interpreting the words via the right
intermediary. But knowing that we have used the right inter-
mediary for a word is no easier than knowing that we are using
the word itself to apply to the right kind of thing.
What then was wrong with the original argument? The little
transition from 'we need to take words in a particular way, if we
understand them' to 'we need a way of representing to ourselves
what they stand for' must have played us false. A representation
is too much of a re-presentation - a mental presence which maps
4
The Blue' and Brown Books, p. 3.
50 OUR LANGUAGE AND OURSELVES

onto the real world according to some method of projection.

And as we have seen, introducing any such thing is apt to be just
a shuffle. The one problem, that of interpreting words to stand
for aspects of the world, becomes replaced by two: that of
interpreting words to connect to the right aspects of one's
mental scene, and that of interpreting aspects of one's mental
scene to stand for aspects of the world. The flaw. is easy to see:
indeed, there is nothing very controversial in the arguments
which have just demolished the type of theory. But the convic-
tion that there must be an intrinsically representative medium,
for otherwise there is no account of what it is to understand
words, is very strong. It is as though we can envisage no other
model of what it could be to understand words, except one that
more or less surreptitiously regards such understanding as a
matter of mapping the words into some other representative
medium, whose powers are conveniently left unexamined.
In Book III of Gulliver's Travels Swift describes the language
school of Lagado, which is occupied in making men use in
language only the very things they are to talk about: " ... since
words are only names for things, it would be more convenient for
all men to carry about them such things as were necessary to
express the particular business they are to discourse on." 5 Swift
imagines the Professors of Lagado as pedlars, carrying the
objects of their conversations on their backs and just showing
them to each other in place of speaking. In effect a dog-legged
theory applauds this ideal, and only avoids the absurdity first
by making the things mental and secondly by magically endow-
ing them with a power of pointing beyond themselves.

5
Gulliver's Travels, p. 151. Swift's thrust is directed against the linguistic pretensions
of the Royal Society. Late seventeenth-century thinkers were convinced that stylistic
faults and faults of language had impeded all scientific progress. The Society
maintained, in the words of its historian Thomas Sprat, "a constant Resolution, to
reject all the amplifications, digressions, and swellings of style: to return back to the
primitive purity, and shortness, when men deliver'd so many things, almost in an equal
number of words" (History of the Royal Society, p. 113). Of course, in so far as this is just a
plea for plain speaking or purity and clarity in prose style, it is perfectly just. But Swift
must have sensed an underlying delusion that the presence of things themselves, or
even pictures of them, made the best kind of judgement, whereas of course in reality it
makes no judgement at all. I owe the reference to Sprat to Professor Hermann Real of
Munster: a description of the stylistic preoccupations of the early Royal Society is given
in 'Science and English Prose Style in the Third Quarter of the Seventeenth Century'
by R. F.Jones (reprinted in The Seventeenth Century by R. F . .Jones).
 HOVV IS MEANING POSSIBLE? (I) 51

3. An Innate Representative Medium?

It might be thought that all this affects only some old mistake,
committed perhaps by Aristotle and Locke, Hobbes and Hume,
but transcended in the twentieth century. But here is Jerry
Fodor, a founder of one of the most influential recent
approaches to language and language learning:
Learning a language (including, of course, a first language) involves
learning what the predicates of the language mean. Learning what
the predicates of a language mean involves learning a determination
of the extension of these predicates. Learning a determination of the
extension of the predicates involves learning that they fall under
certain rules ... But one cannot learn that P falls under R unless one
has a language in which P and R can be represented. So one cannot
learn a language unless one has a language. 6
Instead of taking this as a reductio ad absurdum Fodor takes it to
prove the existence of an innate language, a "language of
thought", in which we initially represent (that word again)
what might be being signified by the word of the language we
are trying to learn. Thus ifleaming the word 'carrot' in English,
the baby's "cognitive system" forms various hypotheses,
expressed in this innate language, about what the term might
mean: he thinks, unconsciously, to himself that perhaps 'carrot'
applies to all and only things which are R. Eventually he hits
upon the right rule, R. This is what Fodor means by rules
determining the extension of a predicate - i.e. rules which
determine the things to which it applies. The language of
thought in which these infant hypotheses are framed cannot of
course itself be learned - or at least if it is, there must be a
preceding language in which its predicates and possible mean-
ings for them were represented, and to stop the regress we might
as well stick at one language of thought. But although
unlearned, it can be part of our innate equipment.
It is a pretty impressive part. For by the argument we can
never come to understand a term unless we can as infants
express its rule of use to ourselves in our innate language.
Unless the infant has innately the resources to describe what a
Romantic sonata is, or a chivalrous gesture, or a carrot, or a
6
The Language oJThought, p. 64.
52 OUR LANGUAGE AND OURSELVES

bonfire, he will never learn. For his only strategy is to map

the English terms, which express these notions, onto his old
vocabulary: he represents to himself what the terms mean in his
innate system. The re-presentation could go in stages: he might
define a Romantic sonata or carrot in different words of English,
and then use a previous representation of those in his innate
language. To make this seem remotely plausible one needs
startling confidence in "decompositional semantics": in other
words, in analyses and definitions which reduce such complex
notions as that of a chivalrous gesture or a carrot, into concepts
and ideas which might plausibly be thought of as an innate
endowment, and common to all mankind, regardless of varia-
tions in experience, knowledge, and culture.
The analogy which inspires this school of cognitive psychol-
ogy is that between the processing mechanisms whereby a
person comes to understand and use the vocabulary he does,
and the inner workings of a computer. Fodor's view is that the
internal system of representations is like the "machine
language" of a computer, the physically realized bits which
represent inputs of data and instructions, and upon which the
machine effects its transformations, before retranslating some
end stage of these internal processes back into a surface display.
Part of the surface instruction to a machine - say, typing '7' at
some point - effects a modification of the subsequent process,
until an answer emerges which contains some feature which is
itself a function of the original input: e.g. '14' if the subsequent
instruction was 'X 2'. What is present to consciousness is like
the computer's visual display. The underlying operations go on
in a different medium, inside. And this medium has to be rich
enough, structured enough, to be able to 'represent' any of the
data and instructions upon which the computer operates.
It is no objection to this analogy that the underlying proces-
ses are supposedly complex, and yet hidden from conscious-
ness. Psychologists are very familiar with the need to postulate
just such kinds of processes. For instance, it is quite clear that
the visual cues which cause us to interpret a visual scene in a
particular way - such as containing objects of a given size at
certain distances - are complicated, and largely hidden from
our consciousness. It takes a good deal of experiment to ev~n
begin to separate them and to generate a theory of how the eye
 HOW IS MEANING POSSIBLE? (l) 53

and brain respond to the stimuli impinging on the eye. As we

saw in connection with Chomsky ( 1.5), thinking of the causal
process as analogous to the transformations from one state to
another of a computing mechanism is not something which can
be fruitfully criticized from the armchair: it is a matter of
scientific judgement whether causal theories of our abilities
inspired by the computational analogy prove their worth.
But it is a matter of philosophical argument to see whether
the elements of a computing system can be regarded as items
whose own representative powers actually explain, or even play
some part in explaining, what it is for us to understand our own
terms in particular ways. One thing is certain. There is no sense
in which elements of a chemical or electrical system intrinsically
represent aspects of dogs or carrots. The elements of computa-
tional mechanisms are just that: elements ofa causally complex
structure which determines various outputs given various
inputs. The elements play a role in this "horizontal" transition
from state to state. But they do not also have a God-given
"vertical" connection with the world outside the mechanism, in
virtue of which their mere presence suffices to explain how we
understand our words. As we have already seen, this was not
even true of images, which at least appear to have some natural
connection with the states of affairs they represent. It is quite
impossible to see how one element of a chemical or electrical
system could in and of itselfrepresent an aspect ofa carrot, any
more than the fall of a box of paper tissues could in and of itself
represent the sinking of the Titanic. Thinking otherwise would
be like explaining how numerals refer to numbers by unscrew-
ing a computer, finding some modification in the circuits
which nurses an input of a given numeral through the computa-
tions, and concluding that the numeral '7' represents the
number 7 because the numeral is translated into that modifica-
tion of the circuits and that intrinsically represents the
number 7.
However, although the internal elements have no intrinsic
powers which enable them to explain, by their mere presence,
our understanding of terms, they may have some other role to
play. We can speak of a modification of some mechanical or
biological system representing one thing or another. We can
impose the vertical connections, just as I can impose the
54 OUR LANGUAGE AND OURSELVES

instruction that the movement of the box of tissues is to repre-

sent the sinking of the Titanic, e.g. for the purpose of some
game, or some description of events. We can give some modi-
fication of the machinery the significance we want. Since we
take the visual display which is the output of some device in a
certain way-e.g. as referring to the number 7-we can interpret
the modifications of the system which underlie this production
in various ways as well. However, if that is all that could be said,
we wouldn't be much further advanced in understanding what
it is for us ourselves to take terms in one way or another. We
would just know that since we do so, we can also impose a
representative function on other things as well. That is hardly
news. Our own understanding is left, like a frog at the bottom
of a mug, simply staring. It powers our ability to see other
things, such as elements of internal systems, as having a rep-
resentative role. But the whole point ofa dog-legged theory is to
reverse this direction, so that the representative nature of the
intermediary elements plays some part in explaining our own
understandings.
The.re is, however, one further possibility. Might we discover
that as a matter of contingent, scientific, fact, there are such
elements which have representative powers? The problem then
is still to get some conception of the kind of fact this might be.
Perhaps inside our heads there are structures which because of
some fact or other actually do play a representative role for us;
perhaps words have their significance because they are mapped
into these structures, in some natural and unconscious trans-
formation. This scientific version of a dog:legged theory cannot
be ruled out of court. But its immediate appearanre is not all
that promising. For it shares with the absolute starting-point of
the enquiry the need to see the kind offact that is missing: the
element of organization, or function, or whatever it might be
which makes it tru"' that a given internal structure has a given
significance, which 1t transmits to words which are transformed
or mapped into it. But why shouldn't the missing kind of fact
which endows it with its representative powers directly infuse
ordinary straightforward language with its significance?
For example, suppose we are attracted to a kind of func-
tionalist account. Our brains steer us around the world whose
properties impinge upon them. An element represents some-
 HO\\' IS MEANING POSSIBLE? (I) 55

thing, such as a carrot, if it plays a certain part in this system.

Very crudely, we could imagine the system activated by certain
stimuli, and itself mediating various reactions, from simple
salivation to other things more oblique, themselves dependent
upon other states of the system which in a similar way cor-
respond to other beliefs and desires we have. The missing fact is
here identified as functional role. It is dearly an attractive avenue
to follow. But then, why not follow it by thinking of the func-
tional role of the good old words of our ordinary language? We
can ask whether it is their function in steering us around the
world which makes it true that they have the significance they
do for us. The transformation into elements of underlying
machinery is then part of a causal explanation of how this
functioning takes place, but not part of a philosophical account of
what the functioning actually is. The causal explanation is of
course highly interesting. But in principle it could be very
different while the functioning remained the same (for instance
a biological bit of the mechanism which starts to malfunction
might be replaced by a silicon chip, with entirely different
microstructure, but capable of yielding the same outputs from
the same inputs). In the current jargon, systems may function
in the same way, but have "variable realizations". My pocket
computer may have quite different underlying circuitry from
yours. My head might, too. So in the first instance we would do
better just to explore the functional role of ordinary, overt uses
of words and language. The bits inside are interesting but, as far
as understanding goes, mere machinery.
What then was wrong with Fodor's argument? The mistake
came in the one misleading sentence: "One cannot learn that P
falls under R unless one has a language in which P and R can be
represented.'' Fodor sees this as following from the need to learn
rules which determine the kinds of thing which fall under
predicates. Learning what things can be truly described as
carrots or dogs or Romantic sonatas can be described as learn-
ing a rule of correct application of the terms. Now the infant
might, like the man with the colour card who uses it to search
for a red flower, or like a man learning a second language,
already have a wa,y ofrepresenting (that word again) to himself
the kind of thing a carrot of dog or Romantic sonata is, and have
a way of translating the English word into his own idiom. I have
56 OUR LANGUAGE AND OURSELVES

already criticized the powers this account attributes to the

innate idiom. But even if the infant might have done this, there is
no reason at all to suppose that he did. William] ames described
the infant's world as one of a "blooming, buzzing confusion".
Out of that confusion patterns. begin to separate and order
begins to form: we begin our acquaintance with sounds, tastes,
colours, shapes, and with the reliable recurrent things which
make up the landscape we live in. With joy or sorrow, we make a
first acquaintance with dogs and carrots, and grow into an
appreciation of what a Romantic sonata, or a smile, or an
electron, is. Growing into such an appreciation needs an
acquaintance with things. Without this acquaintance it is not
even possible to think of anything as a carrot, or as a Romantic
sonata. With the acquaintance comes not only an appreciation
of what words actually mean, but of the kind of thing they could
mean. Acquiring the concept needs learning that there are
things which persist in time, which have various dusters of
properties, and it requires becoming able to distinguish those
things from others on the basis of these properties. Without this
knowledge terms cannot be understood. Hence, they cannot be
understood by translating them into a medium which itself
represented things in advance of the knowledge.
To meet this obvious point the language of thought theorist
must think of the language as itself growing, in its semantic
powers, as the infant increases his appreciation of the world,
and his ability to cope with it. But this is then a matter of some
inner states acquiring representative powers. And this then is not
a matter of them being mapped onto some other states which
already have those powers. Because if it were, the argument
would repeat itself at this level. So Fodor cannot maintain that
learning requires a previous ability to represent what is learned.
And if this cannot be maintained in full generality, there is no
reason to postulate an "inner code" at all: we do just as well by
trying to think of what it is for ordinary words of a public
language to acquire their representative roles, and to abandon
any attempt to understand this by thinking of them as trans-
lated back into conveniently powerful innate media.
The account of learning which I have been attacking has
affinities with the famous episode in Plato's Meno, 7 when
7
Jfeno 80d ff.
 HOW IS MEANING POSSIBLE? (I) 57

Socrates takes a slave-boy through a geometric proof, and

concludes that since he has come to appreciate something new
he must in some sense have understood it all along. The infant is
a "little linguist~' or, as we might call it, a cognitive imperialist:
he brings a complete stock of concepts with him, and cannot
comprehend any influx of experience or learning which would
expand it: anything he learns he does so by translating it back
into a home medium. Learning becomes just a matter of
associating new terms with ones already playing the same
representative role. To avoid this seductive idea it is essential to
remember the argument just given. Suppose we accept some
provisional equation between a word's having a meaning for
some subject, and it playing some causal or functional role in
the way he behaves. Then the reason for avoiding an innate
language of thought becomes stronger. For a person functions
differently in a world of carrots and Romantic sonatas than he
would in a world without them: functional organization itself
grows, adapts, and changes with an appreciation of the world.
Before the infant adapts to the world, nothing can play the role:
hence, coming to understand a language cannot be understood
as translation back into a home medium, with elements repre-
senting things in advance.

4. Radical Interpretation: Manifesting Meaning

So the attempt to understand meaning by a dog-legged
approach fails. But what other ideas can we tise? Perhaps the
most promising is to stand back and think about the way we
attribute meaning to the utterances of other people. The idea is
to think about what meaning is by thinking about how we
detect it. This is the approach dramatized in the person of the
radical translator, or radical interpreter. 8
Next, perhaps, to the works of Wittgenstein, the most
influential book in modem philosophy of language has been
W. V. Quine's Word and Object. In chapter II of this great work
Quine introduces the figure of the radical translator. This charac-
ter has the task of discovering which language is spoken by some
group of speakers amongst whom he finds himself Hts position
8
There is a distinction between interpretation and translation, but that is for later.
See 3.1.
58 OUR LANGUAGE AND OURSELVES

is radical in the sense that he is to have nothing by way of

collateral information to go on. He has no information from
persons who have already interpreted the native language, or
from similarities of native vocabulary or syntax with other
known languages. In the radical position it is simply and purely
the use the natives make of their terms which is to provide the
data for the interpretation or translation. The point of intro-
ducing someone in this position is to isolate the nature of the
evidence available to him and the nature of the theorizing
which he must base on that evidence. Such a "rational recon-
struction" of the procedures of the outsider is then intended to
illuminate the concept of meaning in a classic way: by seeing
how we come to know that a group means such-and-such by their
utterances, we improve our knowledge of what it is for them to
mean such-and-such by their utterances.
The concepts we have already met enable us to ask some
questions about this idea. Firstly, we have learned to be scepti-
cal about the value of some dog-legged approaches to meaning.
Is this just another such approach, but one in which the native
speech is understood to be meaningful just in so far as it is
interpreted back into some other medium (the one the inter-
preter brings with him) about whose meaningfulness we don't
ask? Secondly, notice that the goal of assigning meanings to the
sentences of a group is the goal of coming to understand them. To
call someone who is doing this from scratch a radical translator
or interpreter already seems to imply a distinct conception of
the task. It implies that understanding is gained by translation
back into a home medium which the outsider brings with him.
Like Fodor's infant an interpreter is a cognitive imperialist, who
imports his own compound of categories and concepts, beliefs
and principles, into which he retreats while he ponders the
significance of native carryings-on. But since we have already
found that not all acquisitions of understanding can be
regarded as reinterpretation back into an antecedent medium,
will it be legitimate to consider questions about meaning by
considering a situation where this is the only method allowed?
Thirdly, the radical translator first flourished when the
philosophy of science was relatively confident of "rational
reconstructions" of scientific method. These would provide
some ideal description of the data and the use made of them in
 HOW IS MEANING POSSIBLE? (I) 59

proper evaluation of scientific theories. But this confidence

has largely gone. Writers now are much more inclined to stress
the creative, non-rational elements in our reaction to things, in
our selections of "data" and our ways of using what we select.
So we will need to ask whether there is any prospect of an
objective and adequate scientific description of the procedures
of interpretation.
Our immediate problem in this chapter is the possibility of
meaning: how are we to conceive of the way in which a physical,
natural world contains anything as strange as the fact that one
thing can think of others? How do the facts (about physical
systems, or natural, behaving animals) determine or permit the
facts about thought and meaning? This is a rather different
question from how we, in the guise of radical interpreters,
determine the facts (about some group's meanings). In fact, it is
crucially different, inasmuch as ifwe take our own intentional
powers for granted, the second question might have a relatively
simple answer, while the first question was untouched. I, think-
ing of Henry VIII, can interpret you as doing so too provided
that you ... ; I might fill out such a claim with a plausible
description of the kinds of thing which would lead me to so
interpret you. Yet my own understanding would remain quite
inexplicable. In short, we might know how we determine the
facts without, as David Lewis put it, knowing how the facts
determine the facts. 9 This would be through taking our own
powers as simply given.
Quine's problem was not the possibility of meaning, and it
need be no criticism of the use he makes of his radical translator
that he would mark only a dog-legged approach to this prob-
lem. (It will be a matter of judgement to see how far this
incompleteness worries w,.
Part of the point of my original
metaphor of a triangular l~ndscape is that philosophers differ
not only over which landmarks to climb to, but also over how
good is the view from any given one.) Quine was concerned
with the extent to which dispositions to verbal behaviour
uniquely fix the meaning of native remarks: his concern was to
argue that they do not, so that speakers sharing identical dis-
positions to say things might nevertheless not mean the same by
their remarks. Quine used this conclusion to destroy the
9
David Lewis, 'Radical Interpretation', Synthese (1974).
60 OUR LANGUAGE AND OURSELVES

scientific respectability of any concept of meaning. He intended

to establish the conclusion by considering the way in which two
different translators could properly come to assign quite diffe-
rent meanings to the sentences of the natives they observe. My
second worry, about the cognitive imperialism implied by con-
flating coming to understand a language with interpreting it into a
home medium, is relevant to this strategy. For if two different
radical translators bring with them different home media - if
they have different conceptual schemes to some significant
d'egree - then we might expect that they will end up regarding
their natives as saying different things. But this will not make
unscientific the idea that there is a definite single thing which
the natives mean. There could be a fixed and definite fact about
what a native sentence means, even if differently reacting
interpreters miss it in various degrees. An interpreter may be
unable to understand native speech without first dropping his
own preconceptions, and then learning to appreciate things
their way.
The distinction between interpretation back into a home
medium, and understanding, affects uses which other philoso-
phers than Quine have made of the radical translator. Donald
Davidson has influentially argued that we can make no sense of
fundamentally divergent conceptual schemes. Davidson knows,
of course, that there can be languages of different expressive
powers in particular areas (Eskimos with fine and untranslat-
able distinctions amongst kinds of snow; Arabs and camels;
scientists and the language of theories which they alone under-
stand). But he believes that:
We can be clear about breakdowns in translation when they are local
enough, for a background of generally successful translation provides
what is needed to make the failure intelligible. 10

The idea is that we can tell when a group is expressing ideas

which are new to us only after we have become good at translat-
ing them in general. Davidson wants this point to sustain the
conclusion that we can have no conception of what it might be
for a group to have a language which expresses concepts and
beliefs radically different from ours- too different to provide the
10
'On The Very Idea of a Conceptual Scheme', Proceedings of the American Philosophi-
cal Association (I 973).
 HO\,\' IS MEANING POSSIBLE? (l) 61

backdrop of shared thoughts which make mutual translation

generally possible. Such an argument relies on a principle of
verification - an inference from the fact that we cannot verify
that something is the case to the conclusion that we cannot
conceive as a genuine possibility that it might be the case. By this
principle an hypothesis which camiot be verified marks no
intelligible possibility. But although verificationism surfaces in
many forms, the current consensus would be that it is danger-
ous to rely upon the principle. This is especially so when a
general theory or way of looking at things allows unverifiable
possibilities, but itself contains a way of explaining just why they
are unverifiable. And it is of course explicable - trivially so -
why we cannot verify that a group possesses a radically different
conceptual scheme from ours by interpreting them in our own
terms. Such an interpretation would be given by saying, in
English: 'they think that ... ', where the dots are filled out by an
English sentence. And this form of report can only describe
thoughts which English speakers can have. But this does not
suffice to show that the existence of a divergent scheme cannot
be verified by any means at all. Couldn't we verify the existence
of such a scheme not by translating it, but by coming to under-
stand it?
Consider the following analogy. People enjoy some pretty
dreadful sounds, but not any old production of sound counts as
making music. To interpret a group as making music we try to
recognize our notes, intervals, harmonies, rhythms, and so forth
in the noises produced. So if anything counts as making music it
must use notes, etc. recognizable to anybody else: there can
only be one musical scheme!
This argument is flawed because it ignores the possibility of
working our way into a different musical scheme: going native,
and learning to appreciate quite different structures of sound
as genuinely musical. This is what is done by people who,
for instance, embrace the Scottish bagpipes. But there is an
interesting point about the sense in which they verify that these
things do indeed provide music. For an outsider who refuses to
learn can consistently regard the learner both as losing his
power to tell whether an activity counts as making music, and
also as regrettably spending his time doing this other thing.
People who go native are always suspect: to any good, old~
62 OUR LANGUAGE AND OURSELVES

school imperialist, the fact that such a fellow finds untranslat-

able concepts in native thoughts just confirms what a crackpot
he is. The verification is itselfrelative to an understanding.
This is inevitable and applies to any family of concepts.
There is, for example, no telling what someone means by tenses,
which does not demand appreciating the passage of time; no
telling what someone means by moral language, which does not
demand a moral understanding; and so on. There is no way of
"manifesting" what is meant by terms from these families to
anyone incapable of the kinds of judgement they make. The
situation is extremely clear ifwe consider an advanced theory,
such as high-level physics. A physicist can only show what he
means by spin, charge, field, and so on by having the learner
come to understand the overall theory. There is no likelihood
that the learner has any terms into which such notions can be
translated before doing this. So it is the verification of the
learner, not the translator, which counts, and Davidson's argu-
ment fails.
The third query which I raised about the device of the radical
translator concerns the prospects for a rational reconstruction
of his method. Pessimism would predict only very bland results:
that the translator must use his observation of native disposi-
tions, and his own understanding of things, in order to come to
see the natives as sharing similar beliefs, thoughts, and mean-
ings. More detailed accounts of the scientific way of doing this
might be misguided even in principle: rather like manuals for
telling how to use purely physical descriptions of facial muscles
and geometry in order to verify whether someone is smiling.
The trouble with such a manual would be that it is unusable: if
someone's unconscious "cognitive system" does not deliver an
immediate judgement of whether an observed person is smiling,
a conscious calculation is unlikely to help. This is not however
to deny that the unconscious cognitive system (the same thing
that enables us to make immediate, non-inferential, unargued
judgements of spatial size and distance, direction of sounds,
etc.) may react to geometrical and muscular features in coming
to a verdict. So there is room for a science of which features do
determine such a verdict: similarly, is there not room for a
science of the features of behaviour, with words and with the
world, which determine verdicts of what words mean? In effect
 HO\\' IS MEANING POSSIBLE> (l) 63

this is the science of what in chapter 1 we called the external

problem of competence ( 1.4).
Philosophical preconceptions very quickly dominate this
embryo science. The basic preconception which Quine brings
to the issue is behaviouristic: the understanding of a sentence
should at bottom be a matter of a disposition to respond to kinds
of stimulus. So the initial procedure for the radical translator is
to note the kinds of stimulus which prompt an affirmative
reaction to a sentence, and the kinds of stimulus which do not.
It is the actual presence of a situation in which, say, there is a
rabbit visible, and the disposition ofspeakers in that situation to
affirm 'rabbit!', which affords the prime datum-the prime way
in which a sentence connects with its interpretation. The capa-
city to use a sentence in full understanding is basically a disposi-
tion to respond appropriately to situations which do, and to
ones which do not, exhibit whether it is true. The basic image is
one of confrontation with situations in which the correctness of
some response can be immediately recognized; the interpreter
then starts his task by himself correlating the situations and the
sentences whose truth they guarantee. Difficulties in this model
fall under two headings: first of all, whether there are any
sentences at all to which it is appropriate, and secondly, how we
extend it to sentences to which it is not appropriate. These last
clearly include sentences about the past (we cannot exhibit our
understanding of past-tense statements by making the right
reaction when confronted with past circumstances, since we are
never so confronted), future, spatially distant states of affairs,
states described in theoretical terms, and indeed any state of
affairs whose "recognition" involves using networks of other
beliefs, principles, and concepts. Many philosophers would say
that no sentence can be thought to have its own individual
footing in experience: even recognizing a "stimulus" as indicat-
ing a rabbit implies grasp of a framework of linked concepts
( that of an individual animal, persisting through time, in a
world of things spatially distant from the observer). So there
arises a general bias towards emphasizing the "holistic", anti-
foundationalist nature of all verification.
This fundamental issue in epistemology is pursued a little
further in chapters 5 and 7. The present point is that it makes a
crucial difference to the way we have to regard translation. In so
64 OUR LANGUAGE AND OURSELVES

far as we sympathize with the holistic theme, we will deny that

our grasp of any sentence is a matter of a disposition to affirm it
in any set range of situations. All that will be true of someone
who understands even such a basic sentence as 'there are
rabbits about' is that he should be disposed to assent to it when
he believes that there are rabbits about; but there is no limit to
the number of other beliefs, ways of confirming this, contrary
interpretations of things, and so on, which he might have, and
which would make a difference to the actual range of circum-
stances in which he does believe this. Thus if I hold the dotty
belief that the Russians have little rabbit-formed spies every-
where, I will not assent to 'rabbit!' just when you do. My
"stimulus meaning" - the class of situations prompting assent
to the utterance or dissent from it - is different. But I could
nevertheless mean the same as you do by the term. It is indeed
because I mean what is normally meant that I say that there are
so very few rabbits about. As epistemologists move away from
the idea of a purely ostensive guarantee of the truth of anything,
so philosophers oflanguage move from the idea that it is in such
a confrontation that we find the fundamental word-world rela-
tionships that bestow meaning. The shift is away from a model
in which the radical translator starts with a privileged range of
affirmations which relate directly to a range of stimuli, towards
one in which he works his way into a whole block of meanings,
beliefs, thoughts, and principles by following the general policy
of attempting to see native thoughts as overall intelligible and
rational.
This shift marks the basic division between Quine's attitude
towards radical translation, and the later conception ofit in the
hands of Davidson. 11 I return to this concept in the context of
systematic descriptions of natural languages in chapter 8.
Another highly influential recent philosopher whose work takes
a slightly different complexion in the light of this shift, and of the
earlier points in this section, is Michael Dummett. Dummett
has emphasized that we must see meaning as something which
is capable of being manifested in the use we make of an expression.
What he says here of mathematical expressions is intended to
apply quite generally:

" :\otably in 'Radical Interpretation', Dialectica (1974).

 HOW IS MEANING POSSIBLE? (I) 65

The meaning of a mathematical statement determines and is

exhausted by its use. The meaning of such a statement cannot be, or
contain as an ingredient, anything which is not manifest in the use
made ofit, lying solely in the mind of the individual who apprehends
that meaning: if two individuals agree completely about the use to be
made of the statement, then they agree about its meaning. The reason
is that the meaning of a statement consists solely in its role as an
instrument of communication between individuals,just as the powers
of a chess-piece consist solely in its role in the game according to
the rules. An individual cannot communicate what he cannot be
observed to communicate ... 12
Manifestation is primarily a relational matter. A man may
manifest marvellous musicianship by blowing the 'Pibroch of
Donald Dubh' if the hearers are capable ofappreciating it, but
he may not manifest anything to me by his performance. So
what kind of manifestee is appropriate to Dummett's require-
ment? The last sentence suggests one who is capable of making
observations, but no more. But let us suppose that some things
lie outside observation: the past, or other peoples' sensations, or
sub-atomic particles. Then it is clearly not a sensible require-
ment that a man should manifest his understanding of these
things to someone who himself is capable of only making obser-
vations. Since this audience is incapable of thinking beyond the
present, or beyond other peoples' behaviour, or the world of
macro-objects, it is incapable of appreciating that anybody else
is either. Such limited observers make poor audiences. The only
reason for respecting them is the same as Quine's: the idea that
the understanding of a sentence has to be the disposition to do
the right thing in the right range of empirical circumstances, for
then nothing can count as manifesting understanding unless it is
visible to a passive observer. But the very word 'manifest'
reveals the doubtful nature of this requirement. Like 'display'
or 'reveal' it has highly visual overtones: I cannot display or
make visible the past events I talk about, the future ones, my
own pains and thoughts, let alone electrons or numbers: taken
literally the requirement that you should be able to manifest
meanings would suggest Swift's image of people only able to
communicate about whatever they can carry with them. 13 A
12
'The Philosophical Basis of Intuitionistic Logic', in Truth and Other Eni.f!.mas,
p. 217. u n. 5. above.
66 OUR LANGUAGE AND OURSELVES

better expression ofDummett's idea would be that the meaning

of a statement cannot be, or contain as an ingredient, anything
that cannot be understood or known from the use made of it. And
then, for instance, the question of whether the meaning may be
a function of something lying beyond observation will hinge
upon what we say about our knowledge of such matters. There
will be no swift argument from lack of manifestation which will
settle that. 14 This recasting makes it plain how if we pursue
meaning through either the device of the radical interpreter, or
(which is essentially the same thing) through considering how
we make meanings known to each other, it takes a whole theory
of knowledge to sustain any conclusions.
I have sketched three reservations about the persona of the
radical translator, considered as a device for improving our
understanding of meaning. Firstly, ifit is the very possibility of
meaning which puzzles us, he suffers from the same objections
as other dog-legged approaches: the interpreter's own under-
standing underwrites his interpretation of theirs - but what
funds it itself? Secondly, since we must draw a distinction
between translation (or interpretation back into a home
medium) and understanding, it is what is involved in the
acquisition of understanding which should really interest us.
Otherwise we risk importing unjustifiable restrictions on what
the interpreter himself brings with him. Thirdly, the prospects
for a scientific reconstruction of the procedure of interpretation
are not wholly good, and the form we impose on this procedure
is likely to reflect controversial doctrines in epistemology. None
of these reservations deny some use to the device: indeed we
shall see in chapter 8 how it can be used to further some aspects
of our understanding.
The attack on dog-legged theories, both overt and disguised,
was one of the central achievements of the l.ater Wittgenstein.
But he wrestled constantly with the problem of the kind of thing

14
Although I have taken this passage from Dummett as an example of a dangerous
mistake, it must not be supposed that the mistake infects his whole philosophy of
language. But it is significant that Dummett is a staunch advocate of the priority of the
philosophy of language over the theory of knowledge, whereas at this crucial point, it is
an error in the theory of knowledge which needs correction, and only a better theory of
knowledge which could provide it. An excellent discussion of this is E. J. Craig,
'Meaning, Use and Privacy', Mind ( 1982).
 HOW IS MEANING POSSIBLE? (I) 67

meaning is, and how it is possible. It is his approach to the

problem that I now introduce.

Notes to Chapter 2
2.1 Some of the problems surrounding Locke's use of the term
'Idea' can be gleaned from the collection Locke on Human Understanding,
ed. I. C. Tipton, especially section III.

What is in effect the regress or elephant problem is presented in

Jonathan Bennett, Locke, Berkelry, Hume: Central Themes, ch. 1.2.

2.2 The whole phenomenon of imagining needs very careful and

extended attention. As well as the Wittgenstein quoted in the text, the
two volumes Remarks on the Philosophy of Psychology are excellent. But
the work of Wittgenstein in this connection is better studied alongside
chapter 3, where his own puzzles with meaning are developed. The
reader might also try:
Gilbert Ryle, 'Imagination' in Gustafson (1964).
H. Ishiguro, 'Imagination' in Williams and Montefiore (1966).
R. Scruton, Art and Imagination, chs. 7 and 8.
The place of imagination in perception is mentioned in chapter 7
below, and the role of imagination as an adjunct to passive observa-
tion, if we are to form any conception at all ofour world, is developed a
little there.

2.3 The idea of an internal language of thought, "mentalese", has

recently been very popular, notwithstanding the kind of argument
given in the text. Readers might consult:
G. Harman, Thought, ch. 6.
D. Dennett, 'A Cure for the Common Code' in Brainstorms.
H. Field, 'Mental Representation', Erkenntnis ( 1978).
J. Heil, 'Does Cognitive Psychology Rest on a Mistake?', Mind
( 1980).
The text here does not consider all the arguments that might be used
in favour of giving inner states some representative power. But I
believe that similar considerations stand in the way of any such
theory.

2.4 The methodology of the radical interpreter, and the conclusions

to be drawn by considering him, have generated an enormous litera-
68 OUR LANGUAGE AND OURSELVES

ture. This is partly because of the startling nature of the conclusion

that Quine himself drew: that translation is indeterminate (and, since
there is no truth about meaning which is beyond an interpreter, so is
meaning). As well as references in footnotes, the reader might consult:
N. Chomsky, 'Quine's Empirical Assumptions', in Davidson and
Hin tikka (1969).
W. V. Quine, 'Reply to Chomsky' (ibid.)
- 'On The Reasons for Indeterminacy ofTranslation',Jouma/ ef
Philosophy ( 1970).
R. Kirk, 'Underdetermination of Theory and Indeterminacy of
Translation', Anarysis (1973).
C. Boorse, 'The Origins of the Indeterminacy Thesis', Journal ef
Philosophy ( 19 75).
S. Blackburn, 'The Identity of Propositions', in Blackburn ( 1975).
The requirement which Dummett has insisted upon, that meaning be
something which is manifested in the use made of an expression, has
given rise to a large but difficult literature. As well as the article by
Craig (chapter 2, n. 14 above), the following might be useful:
C. Wright, Wittgenstein on the Foundations ef Mathematics, esp. Pt. 2.
P. Strawson, 'Scruton and Wright on Anti-Realism', Proceedings of
the Aristotelian Society ( 1976).
C. McGinn, 'Truth and Use', in Platts (1980).
There are references to Dummett's revisionism about logic under 6.4
below. The next chapter, which relates meaning to dispositions, is
also relevant.
 CHAPTER 3

How is Meaning Possible?

Frankly, it is not my words I mistrust, but your minds.

Joseph Conrad, Lord}im.

1. Describing: Three Ways of Being Odd

In the last chapter I went through some arguments against
supposing that we understand words by connecting them with a
directly representative "presence", carrying its own lines of
projection onto aspects of the world. We discovered the nega-
tive point, as I shall refer to it, that no thing could haj! the
regress of interpretation. Our conception of what it could be to
have a thing, including a diagram, model, or picture, present to
the mind, allows for different ways in which the thing could be
taken or understood, and this defeats the purpose of the theory,
which is itself to explain what it is to take signs one way or
another. This negative point means that we must cast around
for some other way to come at the phenomenon of meaning.
So far the discussion has been extremely general. Now,
however, we specify a little more the kinds of word whose
understanding concerns us. This chapter centres upon
straightforward, simple applications of predicates to things: the
description of things as red, blue, buses, heavy, expensive, and
so on. We can apply these terms all right! And we know what we
mean when we do so. Our ability to follow principles ofapplica-
tion for predicates is our ability to use universals: to classify,
think, and judge at all. The difficulties philosophers have found
in understanding this kind of ability are not entirely easy to feel.
Fortunately however, there are modern ways of approaching
these problems, developed more or less simultaneously, and
independently, by Wittgenstein, Russell, and Nelson Goodman.•
Together they put immense pressure upon our understanding
of what it could be to assign a meaning to a predicate.

I
References in notes to this chapter.
70 OUR LANGUAGE AND OURSELVES

Imagine, then, a man going about applying some predicate....,

say 'round' - to some things, and withholding it from others. He
gives the appearance ofjudging that things have some property
or other, which he expresses by means of this term. Perhaps he
means that these things are round - he might seem to go through
the right procedure to make this his judgement: he assesses
distances from the edge of the thing to what we should deem the
centre; sees whether such things roll evenly, and perhaps does
other tests. He seems liks a man judging whether things are
round. Now we imagine that he comes upon a new thing - a
thing which is quite obviously square. He considers, and he
says, 'It's· round.' He applies the term! Naturally, our first
thought is that he has made a mistake - he has said something
which he ought not to have said, something false or incorrect.
But two other hypotheses are possible. Perhaps he meant some-
thing different by the word- some property which was posses-
sed by the initial set of things to which he applied it, but is als_o
possessed by this square thing. In the terms of chapter 1, his ·
actual language is one in which 'round' means something other
than - round. Finally, perhaps he means nothing by his term. He
is just going through a parade of making a genuin~ judgement
as children may sometimes parrot adult judgements without
really understanding the terms used in them. So we have three
kinds of hypothesis: that he makes a mistake, that he means
something different, or that he means nothing at all. What
determines which hypothesis is true? What kind of fact is it,
which is ultimately the fact in virtue of which one of these
hypotheses is correct, and the other two not? For we cannot
doubt that on many occasions one is correct and the other two
are not, although which one varies from occasion to occasion.
Mostly, of course, we think that the first is true, because we are
used to believing that people mean something by what they say,
and that mistakes are more common than different usages,
particularly of a common word like 'round'.
It will be convenient to have titles for the three hypotheses. In
the first, the man means what we do by his term, but makes a
mistake. He uses it in accordance with the same rule, or, to put
the same thing another way, expresses the same concept or
same judgement. That indeed is why he is wrong, and says
something incorrect, false. Call this the right-rule view. In the
 HO'W IS MEANING POSSIBLE? (2) 71

second, the man takes the term in a surprising way: he uses it in

accordance with some deviant rule, meaning that some of the
things which he calls 'round' are what we would also call round,
but that others are not. Call this the bent-rule view. The last case
is that of someone who gives an appearance of applying words
to things in accordance with some meaning given to them, but is
not really doing so. His sounds are just sounds, capable neither
of correctness not incorrectness, because no rule exists to
determine whether a particular "application" of the term is
right or wrong. Since an utterance cannot be incorrect, neither
can it be correct: no judgement is qiade. Call this wooden
utterance of a term, in which no rule exists to determine signifi-
cance, the no-rule view.
It comes naturally to say: the man himself knows which of
these hypotheses is true. So whatever fact it is, which makes one
of them true and the others not, it is accessible to him. Perhaps
he can introspect it, and make us aware of it in the same way
that he can make us aware of other mental facts about himself.
He knows whether he is making a judgement, and which judge-
ment it is. Whereas we, perhaps, might be less good at judging
which of these hypotheses is true, and might get it wrong. On
occasion this is how it is. For example, a child might show
enormous concentration in writing down an apparently ran-
dom series of numbers. We might suppose that his production
of a numeral is a random event and that the series is not being
determined by any rule-just as a lunatic might cover pages and
pages with formulae in the belief that he is a mathematician
doing great calculations, but signify nothing. The child on this
hypothesis has in mind no rule determining a series: his writing
one numeral after another is wooden.Nothing counts as correct
or incorrect. But the child (like the infant Gauss is said to have
done) might surprise us. He might explain which series he is
expanding, and show is that he is doing it rightly. We change
our. mind, and admit that there was method in the madness
after all. On a given occasion we might not be sure whether this
was going to happen or not. When we don't know, the irresist-
ible image is of something in the child's mind, accessible to him,
but only guessed at, perhaps fallibly, by us. We imagine that if
we could, as it were, lift off the mental lid- iflike God we could ·
look into the glassy essence of the mind-we would then know,
72 OUR LANGUAGE AND OURSELVES

just as the child does, which ofthe,three hypotheses is true.

A great deal of the work of the later Wittgenstein is devoted to
showing that this picture of the situation is false. One of his
main arguments is the negative point of the last chapter. How
do we envisage the subject's own knowledge of his meaning?
What does the introspective candidate find as he considers his
own mind, which tells him determinately which hypothesis is
true? Perhaps pictures, or formulae, or definitions of terms. But
the presence of any such thing cannot be the fact which determines
which hypothesis is true. For any such thing can be taken in
diffetent ways. Of course, the presence of an image, or of words
framing a definition, or some other presence, might give the
candidate confidence that he knows what he means, that one
hypothesis out of the first two is right. But it doesn't make any
particular hypothesis right. It cannot of itself constitute the
missing fact, because of the problems of the last chapter, sum-
med up in the negative point. No thing can halt the regress of
interpretation, for any thing can be taken in different ways, or in
no way at all. Images or words may flit through the candidate's
mind, but leave him using the word meaninglessly.
It does not follow that the subject himself is not an authority
on whether he means anything, or if so, what. All that follows is
that we need some different approach to this kind of self-
knowledge. It must not be conceived of, as knowledge so often
is, as an acquaintance with any kind of presence, mental or
otherwise. But that leaves other possibilities. The child in the
above example may rightly have perfect confidence that he is
genuinely calculating, and that his placing of one number after
another in his series is not wooden; he may know this at a point
at which an outside observer would not know it. This is not in
question: it is the introspective picture of how it can be true
which the negative point attacks. Perhaps the candidate knows
what he is doing in whichever way we know what we intend or
what would please us. The case of the lunatic shows that such
confidence can be mistaken, but of course often it is not
mistaken.
One way in which Wittgenstein pursues the negative point is
particularly compelling. He considers someone who under-
stands correctly a simple numerical operation, such as develop-
ing a series by adding 2 to the preceding number. We could all
 HO\\' IS MEANING POSSIBLE?. (2) 73

do this, and a learner, after a little instruction with a small

sample of such sums, might "catch on" - perhaps in a flash he
might come to see what is meant, and then know that he can
continue the series correctly, indefinitely. Again, the fact is not
in question. And it might seem particularly tempting to think.of
it in terms of the sudden presence to the mind of a display, a
revelation of what is required. But further thought shows that
this cannot be right. For consider a later application of the
understanding: when the learner writes, ·say, 188 after 186. If
the right-rule view is true, the learner means something which
makes it incorrect for him to write anything but this. Ifhe put
down 193, he would be wrong. But suppose the second view was
true, and the learner had taken our instruction in an unin-
tended, queer way: perhaps he caught on to the bent rule 'add 2
up to 186, and then add 7'. This is a perfectly good instruction~
we could programme a computer to follow it, and we might
have a purpose in doing so. Now ask: what display in the mental
life of someone determined the fact that he took the instructions
one way or the other? Not a display of all the numbers, because
there are too many of them. It would be a pure accident if, in
considering the instruction, someone actually thought of this
particular application. Perhaps a display of some other words:
'Do what was done in this initial sample, whenever mry number
is proposed.' Such a display might occur, of course. But suppose
the learner had, in some bent and remarkable fashion, takett the
instruction to introduce the bent function 'add 2 up to 186 and
then add 7'. Why shouldn't just these words also go th:r:ough his
mind? He could think of himself as "doing what was done in the
initial sample whenever any number is proposed". That is,
compute this function whenever any number is proposed. So the
presence of these words does not seem to separate the right-rule
learner from a bent-rule learner.
At this point a great variety of issues start to clamour for
attention, and it is difficult to preserve a sense of direction.
In particular many philosophers see here an opening into a
relativistic, conventionalist, view of our own classifications. To
us it seems absurd and almost incomprehensible that someone
would actually take the instructions and the initial sample in
this bent way. Why should he get the idea that there is such a
gross singularity just at 186? Why didn't he enquire about it?
74 OUR LANGUAGE AND OURSELVES

But might it be that we are here imposing our own, accidental

perspective? From the point of view of this bent learner, going
on to add 7 after 186 is going on in the same way. It is our
, "similarity space" which, allegedly, he finds bent. Since it came
naturally to him to think of the relation between 186 and 193 as
like that between 184 and 186, he finds our tendency to insist on
188 as the right successor highly deviant. This conventionalism
is sufficiently important to deserve a section to itself. But that is
in part a digression from the main issue about meaning. The
main issue is not whether there is an element of conventionality
in taking the instruction one way or another. The main issue is
to obtain some ·conception of what it is to take the instruction in
any way at all -in other words, to find out what makes true the
right-rule view, rather than the bent-rule view, or the no-rule
view. Still, it may help with that problem to think a little further
about bent classifications.

2. Bent Predicates: Wittgenstein and Goodman

What are we to make of the possibility, if it is one, that some-
body takes our instructions, and an initial sample of cases, to
introduce a rule of application of some term, but a bent one? It
will help to have some examples in mind.

'Add2' Bent-rule: Add 2 up until 186, then add 7.

'Red' Bent-rule: A thing is to be called 'red' just

ifit is observed before l January
1986 and is red, or is not so
observed, and is yellow'.

'round' Bent-rule: A thing is to be called 'round'

just ifit is one ofan initial
sample, and has the shape
defined by a point, travelling
equidistantly from a fixed
point, or does not belong to the
sample, and has four straight
sides of equal length, at right-
angles to one another.
 HOW IS MEANING POSSIBLE? (2) 75

There is nothing intrinsically wrong with the bent rules.

They define perfectly good uses for terms. As I have already
remarked, a computer programmed to recognize shapes, if it
can determine which things are in the original sample, can give
verdicts according to the last rule. Things like canaries which
are only observed after 1986 has begun will be properly called
'red' if that term is used in accordance with the given rule -
red BR (red-bent-rule), as I shall index it. The bent rules describe
meanings which words can take: indeed the words used to
express them together define them. The curiosity is that terms
such as red BR and round 8 R apply quite properly to the sample of
objects which a learner of our ordinary vocabulary will have
been shown. So it seems that there could be nothing improper in
the learner taking our instructions to introduce these bent
meanings. But if he does so, one fine day he will apply the terms
in accordance with his understanding to quite astonishing
objects -yellow or square objects. In the numerical case he will
continue the series in a dramatically devious way.
The bent learner can be thought of graphically in the follow-
ing way. We can imagine the dimension of colour, shape, or
arithmetical addition functions (add 2, add 3, ... ), arranged
vertically on the side of a graph. Increase in time (or in number)
is plotted along the horizontal axis. We would plot continuities
like the line C, and kinks and changes with a line like Kin Fig. 2.

.------K
ordinary t
dimension of
II-----_.__ _ _ _ _ C
description
~--------~
- - new applications

Fig. 2

The bent learner has got hold of a "dimension" of properties

which reverses the picture. The state of affairs represented to us
by C appears to him as a kink; what appears to him as a simple
straightforward continuity is the state of affairs which appears
bent to us. So his graph is like Fig. 3. Thus, our dimension of
arithmetical plus-functions represents someone who adds 2 up
to 186 and starts to add 7 after 186 as bent; but the BR
76 OUR LANGUAGE AND OURSELVES

dimension represents him as going on the same way, and some-

body who continues to add 2 after that point is represented as
kinked. Of course, there is an indefinitely large variety of pos-
sible bent dimensions in addition to the examples chosen. They
can be created quite automatically.

~i~~t :;men-t I-------'----_-_-_-_-_-_-_- :

description .
- new applications

Fig. 3

Goodman used the possibility of these bent dimensions to

cast a new light on the classical problem of induction. 2 This
problem queries our right to take observed regularities in things
as representative - as likely to continue or to have continued in
regions of space and time beyond our actual acquaintance.
Thus we suppose that things tomorrow will be pretty much the
same as things today, and that where they are not the operation
of underlying similarities is responsible for any particular
changes. We do not expect objects to gratuitously change
shape, colour, size, and weight; we do not expect physical
constants to suddenly vary, forces to spring up and die down.
Our whole lives are premised on the stability of the natural
world in myriads of respects. But now take a dimension in
which, in given circumstances, we expect stability. We can then
mechanically define a bent dimension of predicates &"overned
by bent rules, with the feature that if things stay the same in our
respect they change in the bent respect, and vice versa. For
example, take the bent predicate red 8 R mentioned above; we
would expect new rubies, mined after 1985, to be red; we would
expect new wounds to give red blood. If so, they are not, like
rubies and drops of blood observed before 1986, red 8 R. Since
they issue after the crucial date, they would be red 8 Ronly if they
were yellow. Now why do we prefer redness to red 8 Rness? What
makes us think we have got hold of a similarity which nature

2
Fact, Fiction and Forecast, ch. IV.
 HOW IS MEANING POSSIBLE? (2) 77

will herself protect, whereas some unfortunate who catches on

to a bent rule will get a nasty shock in the dawn of 1986?3
Goodman himself, and many philosophers influenced by
him, saw nothing but linguistic convention to hold onto. We do
speak a language which sees redness as simple and redBRness as
bent; we could have spoken a language which placed them the
other way round. But since we do and have used the former, we
are in the habit of expecting things to continue red and of
regarding any other prediction as irrational. We ourselves
would look just as irrational from the perspective ofa commun-
ity which "entrenched" a predicate following the bent rule. It
just so happens there are no such communities. This answer of
course presupposes that there is some fact making it true that
we mean the one thing and not the other by our terms. In other
words, it brings us no closer to solving the problem of how it is
possible to meil,n a determinate thing by a term, which loomed
in the last section. It still leaves it possible that we form a
wooden community, there being no truth that any oLus mean
one thing and not another by our predicates. Or perhaps we
form what we call a mutually bent community, in which indi-
viduals have taken their training and the initial samples in
different ways, formulating different rules which may suddenly
dictate divergent applications, to our mutual consternation.
I But let us put that on one side. It is still true that our preference
for our rules and not their bent counterparts is hard to regard as
a product of purely conventional arrangements. For it is incohe-
rent, in a way which I shall later elaborate (chapter 7), both to
be confident that, say, future blood samples will be red, but also
to regard that confidence as the outcome of an arrangement
which we merely happen to have hit upon, out of a selection of
equally attractive arrangements. If the opinion had such an
insubstantial ancestry, so disconnected from the way things are,
it had better not be trusted. In other words, if Goodman's were
the ultimate answer, we would have no defence against total
scepticism about whether the world is stable in any of the
respects which we rely upon. Butter might be red tomorrow,
-' There are sciences where this kind of question matters. Econometrics is bedevilled
by the fact that many equations or models of what varies with what in an economy may
fit initial segments of data, but diverge in their predictions of what will happen when
things change.
78 OUR LANGUAGE AND OURSELVES

and blood yellow; either might sing to us or cause golden eggs to

come out of thin air.
A slightly more substantial answer would be that nature
looks after our groupings, ensuring that our natural dimensions
of classification are ones which are stable. Perhaps an evolutio-
nary explanation works, so that time and selection weed out
groups with a tendency not to latch onto the right groupings,
not to "carve nature at the joints". This answer restores the
coherence which seems lacking in a pure conventionalism: we
might reasonably retain confidence in a prediction ifwe felt that
a history of pressures which selectively favour the successful
had brought it about that we have that confidence. But the
answer has its problems: a group using red 8 Rwould have done
just as well as us until now. It is only in 1986 that they will get
their come-uppance. The evolutionary story might actually
predict that there would be a multiplicity of predictions at any
time, since only the pressure of the future redness (or red 8 Rness,
as the case may be) of things defeats bent-rule-followers. Surely
liberal nature would have grown some?
At this point we might begin to suspect the way we are
looking at the issue. How credible is the possibility of these
bent-rule-followers? Perhaps we can't ultimately make sense of
the stories. For instance, consider again the deviant interpreta-
tion of the arithmetical rule: add 2 up to 186, then add 7. What
else is true of the learner who takes this rule in the bent way?
Does he think that he is going on in the same way when he adds
seven bricks to a pile to which he was previously adding two at a
time? What ifhe can carry two bricks but not seven? Does he not
notice the difference: is he not aware of the shambles he gets
into? Does the roundBR operator not notice that wheels "like"
the ones he was initially shown do not roll, but meet flat planes
sometimes at a point and sometimes along a line, and so on?
Worse still, these people normally notice what we would con-
sider to be similar deviations. If the foreman tries to load the
bent bricklayer with seven bricks after he has added 184 bricks
he protests: this is not going on in the same way. But two bricks
later, it is. Again, the bent interpreter of'red' is normally aware
of the sudden emerging of a yellow member of a class whose
members were hitherto red. It is only if the yellow arrival times
its entrance for midnight 1985 that he regards everything as
 HOW IS MEANING POSSIBLE? (:2) 79

"going on in the same way". I do not myself believe that we can

really conceive of this kind of sensibility. Certainly we would in
practice cast around for other interpretations of these people:
perhaps they are blind to certain changes; perhaps they suffer
from mysterious lapses of memory; perhaps they aren't really
talking about numbers, shapes, and colours at all, but have
other features in their minds, features of which we are not
aware, and cannot define even in a bent way. In this respect, we
do not inhabit the same world.
ff this remains just a point about what we would do, it
suggests cognitive imperialism. We would find it difficult to
imagine someone who takes the bent rule to be straight; we
would be mystified by his words, would be inclined to
reinterpret him, and charge him with blindness to various
differences of things. But perhaps that is just us, locked into our
own capacities and "similarity spaces", with our own particu-
lar imaginations, natures, or conventions. If this is all that can
be said, then from a more objective standpoint, the bent-rule-
follower is not only possible, but there is no sense in which we
are doing something properly, which he is doing badly. We
couldn't regard ourselves as having locked onto the real
similarities amongst things, or the right way to classify things,
about which he is mistaken. There would only be correctness
relative to a scheme of classification. This is the position in the
theory of universals (that is, of rules governing predicates) we
know as nominalism.
Can we really see these problems with the bent-rule-followers
as the outcome of some contingent, parochial, fact about
ourselves? This is not as simple as it looks. We can press the
difficult questions. For example, the bricklayer responds very
differently if the foreman loads him with seven bricks after he
has added 184 bricks from how he responds after he has added
186. h follows that he must know how many bricks he has
added. Equally if this bricklayer is watching a film about such
an episode he doesn't know whether the man is going on in the
same way unless he knows how many bricks he has already
added. There seem to be three possibilities. Perhaps the bent-
rule-follower knows these things in some mysterious, innate,
way. But then he is not separated from us by,mere conventions,
or even by normal differences of receptivity: he is separated by
80 OCR LANGUAGE AND OURSELVES

total mystery! The second possibility is that he knows these

things normally: when he is loading bricks he keeps track of the
number he has added, and ifhe loses track he doesn't know how
to keep on the same way: coming in the middle of a film he
cannot describe whether the colleague adds two bricks at a
time, since he cannot tell when he passes 186. But that just
shows that he is using this number to mark a difference: he
needs to know whether the different thing which happens after it
happens at the right point. This destroys the intended
symmetry with us, that he should think of everything as going
on the· same way.We don't need to be aware of how many bricks
have already gone to know whether the man continues to add
two at a time. Finally, perhaps there are no signs of normal or
abnormal knowledge. It is just that after 186 he struggles with
seven bricks and shows no awareness that anything is different.
To him it is as though everything is the same. But how can this
be? This bent-rule-follower fails to perceive what we regard as
differences, if they happen at the right point. But that's just
failure. How does it transmute into the picture of someone who
is genuinely making different judgements, following a different
rule? It builds an image only of someone who forgets what he
was doing at a particular time, or is unaware of much of what
is going on: if we generalized this alternative over a good
number of normal predicates (so that the man shows the same
insensitivity to changes of shape, weight, colour, number, ... )
we simply end up with someone who does not know the world
about him.
All these are things which we say! Well, they are of course.
But before this rekindles the nominalist flame, we might reflect
that since they are things which we say, we ought also to say
that we can make no sense of the possibility of a sensibility, a
way of perceiving and classifying things in the world, which
naturally operates in terms of the bent dimensions. Bent predi-
cates are important in the philosophy of language (and, of
course, in the philosophy of induction, and for that matter in
any branch of philosophy where similarities and differences are
important, such as ethics or the philosophy of mind) because
they offer a way of filling out the vision that our descriptions are
in some way arbitrary, conventional, parochial to us. If the
filling does not help the vision, then something else might in
 HOW IS MEANING POSSIBLE? (2) 81

principle do so, but it remains true that the introduction of bent

predicates does not do the work which was expected ofit.
Let us return to the bent learner. All this does not mean that
he is impossible. It only means that if he caught on to the bent
rules he would have to do more than simply go astray at 186 or
the beginning of 1985, or wherever. He would need to show
some awareness of the bend in the function he takes '+ 2' to
express or in the class of things properly described as 'red' or
'round'. If he did catch on in the bent way he might, for
instance, argue with a foreman who asks him to add bricks two
at a time. He would point out how unreasonable this would be
after he has added 186 bricks. Society's amazement at his
difficulty would bring to light the odd way he took the explana-
tions, and we would expect some simple account of what went
wrong, and expect to have a reasonably dear remedy, explain-
ing again what the foreman intends. What the learner cannot
do is both take the initial samples and explanations in the bent
way, and show no awareness of the bend, for by the above
arguments we only deceive ourselves if we think we can make
sense of that. It follows too that we can make no real sense of the
possibility that we, now, might form a mutually bent commun-
ity, so that having latched onto quite different rules of use of
terms, we are poised to diverge in mutually unintelligible ways
when presented with new things to describe. I can certainly eye
my fellows askance in a number of ways. I can mistrust their
judgements, their memories, their real grasp of any principle for
applying terms. And on occasion someone may get hold of the
wrong end of the stick; ifhe is brighter or less bright than others
he may perceive similarities between things compared with
which other differences are unimportant, or perceive differ-
ences where we see only similarity. This can lead to scientific
reforms and advances, for although these mistakes and dis-
satisfactions may at a time be undetected, they are in principle
detectable. They involve different dispositions and ways of
talking and acting, in advance of different occasions of applica-
tion of a term. 4 So since my companions show no undue interest
4
This is a point on which I diverge from Saul Kripke. In his superb discussion of this
( Wittgenstein on Rules and Private Language) Kripke denies that dispositions uniquely fix
the question of whether a person is faithful to the right rule or a bent rule; in other
words, a bent-rule-follower·need not have different dispositions. For a discussion see
my 'The Individual Strikes Back', Synthese (1984).
82 OUR LANGUAGE AND OURSELVES

in 186, no undue fear lest next year's motor cars will have round
wheels, no concern to lay in a stock of observed, yellow,
bandages to match 1986 blood, I know them not to interpret the
words in the bent ways defined. But are we entitled to talk of
intended meanings at all? Does the wooden, no-rule picture, the
third hypothesis, now come into its own?

3. Wooden Communities: Uses and Ways ef Life

This is the central problem of the later world of Wittgenstein.
Until paragraph * 198 in the Investigations he develops what we
have termed the negative point: 'any interpretation still hangs
in the air along with what it interpreted, and cannot give it any
support.' The existence of a rule governing the application of a
term is not created by the existence offurther substitutes for the
term, for they pose the same problem. Wittgenstein then
squares up to the threatening paradox, that the wooden picture
is inescapable, and that nothing can create the existence of this
rule. A man who calls a taxi a bus or a banana red would be no
more wrong than us; just different. His training gives him one
kind of disposition, ours gives us another. In neither case can
the contents of our minds provide a rule determining whether
what we are saying is correct or incorrect. Any course of action
(i.e. any application or withholding of terms) can be made out
to "accord" with what went before, as the example of the bent
learner shows, and this just means that there is no right or
wrong in our sayings, and therefore no judgemerits are made. 5
Wittgenstein's answer to this wooden picture is well known. He
wants to connect the fact of a term being governed by a genuine
rule, determining correctness and incorrectness of application,
with its use, with a custom, a technique, a practice, with the fact that
the word is embedded in the "language games" and "form of
life" of a community of people. So how do practices give rise to
meaning, when it looked impossible that anything would?
One suggestion is this. There is indeed nothing but the
5
The example of a pocket calculator may help. Suppose such a machine with the
oddity that although it adds 2 correctly to numbers before 186, when instructed to do so
for numbers above that, it adds 7. It computes the bent function correctly, the straight
one incorrectly. But in itself it just shows numerals, and there is no correctness or
incorrectness about it. Only from some outside point of view is there a truth that it is
functioning well or badly.
 HOW IS MEANING POSSIBLE? (2) 83

continuous flow of utterance, and the continuous play of dispos-

itions, which seemed to allow nothing better than the no-rule
view. But after all the dispositions include tendencies to correct,
criticize, and adjust deviations. So in a community a deviant who
calls a taxi a bus, or a banana red, is criticized. His behaviour
will give rise to inopportune actions on the part of other people.
Taking his utterance as they naturally would, they will go
wrong. If the man appears normal in most respects then they
say he spoke falsely; if enough of his classifications are out of
step with the rest of the group's they find him unintelligible.
(They may speculate about the possibility of bent-rules, sug-
gesting that he takes his terms in some genuinely different way.
But for the reasons I developed in the last section, they are
unlikely to make much of that.) The point is that the normative
aspect of meaning- the fact that some applications of terms are
incorrect, and that the rules prescribe what kind of thing is
correctly described as what - emerges from mutual pressures
towards conformity.
On this view a community, in its language-using practice, is
like an orchestra without either conductor or score, but with a
tendency to turn on players whose notes are discordant with a
democratic attempt at harmony. The negative point is sup-
posed to prevent us from believing that each individual player
has his own score - his own private instruction how to describe
things, which gives him a standard for applying terms. 6 If we
take this orchestral metaphor seriously it implies that the lone
individual, considered quite apart from any surrounding group,
could not mean anything by his terms. The wooden, no-rule
view would be inevitable, for nothing exists to give the solitary
speaker standards of correctness. Suppose he faces a square and
describes it as round. This seems right to him, and what else is
there to create a standard by which he has made a mistake?
"Whatever is going to seem right to me is right. And that only
means that here we can't talk about 'right'. " 7
This is the heart of the famous anti-private language argu-
6
Idealists who were sensitive to these considerations (including Kant and
T. H. Green) tended to see the standard for description as laid down by the infinite or
absolute mind in which we each to some extent participate - corresponding to the
antecedent instruction ofa composer. This theory is not attractive: what voice tells me
how to take the instructions of this mind? \Vhy should I listen to it? And how did it get
off the hook? 7
Philosophical lm•estigations, § 258.
84 OUR LANGUAGE AND OURSELVES

ment in the later Wittgenstein, which I explore further in the

next section. But the orchestral metaphor looks dangerously
unconvincing if it is supposed to have such a strong conse-
quence. Remember that the initial suggestion was that the
existence of a technique or practice in which a term is embed-
ded is the magic ingredient which creates a meaning for it. It
needs to be argued that the technique or practice which is
necessary to create meaning has to be public. Why shouldn't the
solitary individual embed terms in instructions to himself:
instructions which contrive to give him a technique, to cope in
some way or another with his world, although there is no
pressure from surrounding speakers creating standards of cor-
rect ways of taking those instructions? It is easy to go through
the thought-experiment of coming across such an individual. A
solitary individual growing up in perfect isolation - a born
Robinson Crusoe - might give all the appearances of following
rules, including linguistic rules, and of having a practice which
embodies a distinction between correct and incorrect perform-
ance. Indeed we can imagine cases in which we simply have to
say this. An example due to Michael Dummett is that of a born
Crusoe who over the years evolves a technique for solving a
Rubik's cube washed onto his island. There is no way of regu-
larly doing that by chance. You have to follow rules. Perhaps to
help himself he creates symbols reminding him of what to do at
various points, and appeals to these on the way through the
cube; with these symbols he can do it, and deprived of them he
cannot. Clearly he has the practice or technique which entitled him
to be regarded as meaning various determinate things by these
symbols.
Wittgenstein's followers have tended to divide on the issue of
whether his solution to the problem of meaning denies meaning
for born Crusoes or not. The difficulty is that if the argument
does deny it then, as I have just suggested, it seems unconvinc-
ing ( consider what else he might do to show us that he really
means things by his signs and symbols, or follows countless
rules in his practices); on the other hand, ifhe is allowed to be a
rule-follower, the orchestral metaphor ceases to embody any
solution to the problem. A compromise is suggested by Saul
Kripke. He points out that we might indeed think of Crusoe as
following rules, but all that follows is that if we do so, we are
 HOV', IS MEANING POSSIBLE> (2) 85

"taking him into our community and applying our criteria for
rule following to him". 8 But it is not clear what this means, nor
whether it gives the community any particular prominence in
the creation of meaning. An orchestra coming across a solitary
player, concentrating hard and making noises, might well say
that if he were with them, he would be doing well or badly
making these noises. But on the "democratic harmony" theory,
they could not say that he is doing well or badly since in
isolation there is nothing for him to do well or badly. The
problem Crusoe poses is that he does have a practice (follows a
tune) regardless of how we or anybody else think of him. Of
course, Kripke is right that when we say this we apply our own
criteria for rule-following to him; it is our judgement that he is
. following a rule. But this does not bring our community or any
community far enough into the picture. It would be our judge-
ment that an island has a tree on it. But whether an island has a
tree on it is quite independent of how we or any community
describe it, or even of whether any community exists to describe
it. On the face of it, the situation is the same with the solitary
intelligent Crusoe, in which case he has rules, meanings,
standards for applying terms in his own solitary state, and with
no reference to any community.
The problem with Crusoe shows that we must not fall into the
common trap of simply equating practice with public practice,
if the notion is to give us the heartland of meaning. It will need
arguing that, contrary to appearance, the practice of isolated
individuals cannot count. In any case, if the practice of an
individual in isolation is not enough to create the fact that his
words have meaning, how is the practice of a lot of us together to
create the fact that our words have meaning? We talked earlier
of the norms which arise from mutual pressures towards con-
formity in description. But how exactly does group conformity
relate to understanding a predicate? In the orchestral analogy,
a player knows whether he is wrong by listening for concord-
ance-with the group, and nothing else matters. But when I
judge something to be red I am certainly not offering a
shorthand for the more elaborate judgement: 'this is what most
members of my group wo·uld call 'red'.' This can be no general
solution to the problem of the meaning of predicates. For to
• Kripke, op. cit., p. 110.
86 OCR LANGUAGE AND OURSELVES

make this judgement I need to understand the more compli-

cated predicate 'what most members of my group would call
'red'.' This predicate allows just as much for bent rules and no
rules. If we apply the equation once more we get: 'This is what
most members of my group would call 'what mos.t members of
my group would call 'red'''; and we are off on a regress. Any
predicate 'P' transforms into 'P*': 'what most members ofmy
group would call 'P' ': this transforms into 'P**', and so on.
There is no solution to our problem here. The more complex
predicates are not even synonymous with the bases from which
they derive, for it need not be true that most members of my
group call things which are X, 'X'. They may make systematic
mistakes. And in any event, it is no easier to conceive of the
fugitive fact, that we are genuinely guided by principle, when
we think of the more complex predicates, than when we con-
sider the simplest ones. So the orchestral analogy cannot be
taken this way.
The point also damages another analogy in which the
practice of a community gives rise to certain kinds of fact. This
is the analogy with the conventions which underlie money. Pieces
of paper can be of no value to an individual in isolation. But the
practice of a group of individuals can create the fact that their
pieces of paper have value to them individually. (Paper money
astonished Marco Polo when he visited China.) The analogy is
short-winded: the value to me of a banknote is directly a matter of
what other people will do for it. But in applying a word I am not
directly concerned with the reactions of other people. I do not
generally consider their assent or dissent to be the final court of
appeal on whether I am right. I am directly concerned with
whether a thing is red or round or whatever, which is a quite
separate issue from whether people describe it as such.
The democratic harmony view is responsible for the rela-
tivism which is frequently associated with the later work of
Wittgenstein (quite against his own intentions). An individual
player in the orchestra may go wrong. But how can the orchestra
itself do so? There seems to be no external standard by which it
can be deemed to be doing well or badly. The standard for
correctness in description seems to have shifted, as it were, from
conformity with how things are, to conformity with each other.
If I live in a community which calls·the earth flat, is not this just
 HO\\' IS MEANING POSSIBLE? (:2) 87

one more element of the dance, on which I ought not to get out
of step? There are actual illustrations of the way in which the
emphasis on practices and customs introduces this danger.
Suppose a group has a religion. Part of the religious practice will
be to say certain things - that the God or Gods are thus-and-so,
that various actions need doing, various doctrines are true. The
practice is to say these things; perhaps the practice stands the
group in good stead practically or emotionally. Saying the
things is an action: if it works, how can it be criticized? The
natural opposing thought is that ifin these sayings they intend to
describe what the world is like, then they may be wrong. But
remember that it is the nature of the practice which is being
held to determine what the intentions are, not the other way
round. It is not their mental lives which determine the correct-
ness or incorrectness of saying that God is thus-and-so, but the
nature of their customs, techniques, ways oflife. So they derive
the powers of these sayings from their role in their customs. And
then there seems to be no room for an ingredient of meaning
which makes it possible for the sayings to be false.
If this conclusion were right, it would best be taken to show
that the notion of a practice is an insufficient source of
standards of correctness, that is, of rule-following and of mean-
ing. If practices do not lift sayings into a normative dimension 9 in
which they are susceptible of falsity, and hence of truth too, they
are not filling the role which is demanded of them. The example
illustrates that we cannot glibly announce that the concept of a
custom or practice obviously has this power: it is going to be
difficult to picture the emergence of truth and falsity out of
customs and practices,just as it is difficult to picture the emerg-
ence of meaning out of any amalgam of mental and physical
facts. But the descent into relativism can be avoided. If the only
ingredient in the practice were to say the words, then it is indeed
hard to see why they should be taken as expressions of belief,
and susceptible of truth and falsity. And we have already
learned to doubt the authority of the person using the words: it
is not clear that he will have privileged access to whether they
express a belief, or serve some other role. 10 However the
• This normative aspects of things is stressed in Kripke.
10
Suppose you say to yourself'! believe in life after death'. Do you know that this
expresses a belief? \\"hy not suppose that it expresses an attitude, or vague emotion?
88 OL"R LA>!GUAGE AND OURSELVES

practices may have many strands: the words used in religious

ceremonies also occur elsewhere; the sayings are subject to the
same kinds of criticisms and doubts as others, perhaps the
commitments they express influence people in the same kind of
way as other beliefs, and so on. The idea will be that a saying
expresses a judgement if the practice involves taking it as
expressing a judgement, which itself involves procedures of
assessment, acceptance, and rejection. This, at least, ought to
be V\'ittgenstein's answer. Ironically it follows that the major
industry of taking his later work to release religious faiths from
arguments concerning their likely falsity, is misdirected. The
price of the release is that the sayings no longer express beliefs,
for without the controls which are part ofa genuine judgemen-
tal practice, language is just on holiday.
Crusoe showed that it is not clear what the word 'public' is
doing if 'public practice' is regarded as the source of standards
of application of terms. The religious example shows that it is
not very obvious what counts as a practice either. In particular,
we must be careful over what counts as identity of practice. In the
last section I introduced the possibility of a mutually bent
community, in which each individual had taken his initial
exposure to terms in a different way. I urged, against
nominalism, that although an individual could catch onto a
bent rule, we could make no sense of his both doing this and
failing to appreciate the bend. And this appreciation would
display itself in different dispositions, or practices (the brick-
layer who is asked to carry bricks up two at a time). Suppose,
however, that I were wrong about this. In that case a concealed
mutually bent community would appear to be a possibility, in
which the apparent identity of practice at a time is a cover for
different individual ways of taking terms, each supporting
its own standard of future application. Each member of the
orchestra would be resolutely following his own conception of
how the theme should go, and the harmony in the first few bars
would be a matter of luck. If a concealed mutually bent com-
munity is a possibility, then their common practice at a point
seems to be no source at all for standards of future correctness.
At a point of divergence the individual carries on one way, and
others carry on other ways. There is just nothing to say who is
"right", since their preceding practice allows for this diversity:
 HO\\" IS MEANING POSSIBLE? (2) 39

their tunes can be continued whichever way they see fit. The
orchestra is in no position to criticize any individual player,
since the democracy is no longer speaking with one voice.
Of course, we do not believe ourselves to form a concealed
mutually bent community, partly because we believe in the
common nature of mankind, and partly perhaps for the anti-
nominalist reasons I developed in the last section. Wittgenstein
might be seen too as denying the mere possibility of such a
community. It gives each individual a conception of the right
way the tune should go, or in other words his own previous
intention to use a term in some specific, determinate way, and
this intention exists and determines a standard for truth in his
judgements entirely without reference to other people. This is in
many commentators' eyes exactly the idea which Wittgenstein
opposes, substituting instead either the democratic harmony
view, or some close cousin which we come to later. 11 But it is
still at this stage unclear why he can oppose it. Certainly the
negative point warns us off one particular conception of this
individual intention. If the individual has this determinate
intention, and knows what it is, this is not made true by the
presence to his mind of a particular display. But on the face ofit
that leaves other possibilities. Crusoe may know how he intends
to use the symbolism which determines the way to solve Rubik's
cube not just because particular pictures come into his mind, or
other symbols, but because he does something which counts, for
him, as according with the rule, and something which does not,
if he makes a mistake. He has his own practice. Similarly we
naturally think of the child of the last section, who has a rule for
developing a series, as aware of how he should go on by his own
lights. It is this determinate intention which gives him standards
of correctness and incorrectness, and generates the truth that he
means something, and is not, like a wooden child, merely
writing numbers one after the other. This natural picture is not
destroyed by the negative point. That only attacks one concep-
tion of what it is to have a determinate intention (it is to have
some presence in the mind) and how we know of it when we
have one (by introspective awareness). It is not by itself strong
enough to suggest that no conception of the difference between
11
See Crispin Wright, Wittgenstein on the Foundations ef Mathematics, pp. 20 ff. for a
good discussion of this.
90 OCR L\'.\GL'AGE A'.\TD OURSEL\'ES

the solitary understander, and the wooden individual, is

possible.
The upshot is that it is no easy matter to make public practice
or custom into the magic ingredient which would tum the
wooden picture into the full one. The negative point offers
nothing strong enough to oppose Crusoe's claim to be using and
understanding a term, and to know its meaning through first-
person knowledge of his own rules, intentions, and procedures.
The negative point only shows that this knowledge is not to be
thought of as a simple matter of the presence of some display in
the mind.
My Crusoe would have invented a fragment of a personal
language- an "idiolect". Now the fact that he is possible should
not be taken to imply that a linguistic community may be
regarded as a group ofindividuals, each with their own idiolect,
but amongst whom, because of their need to communicate,
there arises a pronounced similarity ofidiolect. This conformity
would be in a sense accidental: the fact making a word mean
what it does in an individual mouth would be entirely a fact
about the speaker, and a fact upon which he would be the
authority. Whereas in an actual linguistic community we recog-
nize independent authority. What a word means, and what a
person has said by using words, is a socially fixed matter, and
often does not accord with a speaker's own understanding, or
lack ofit. If a man says that he has an elm in his garden, or that
his father has arthritis, he may have a very poor understanding
of his own saying: he may understand no more than that he has
some kind of tree in his garden, or that his father has some kind
of ache in his joints. He may himself be quite unable to tell an
elm from a beech, or arthritis from rheumatism: nevertheless
we will not hold him to have spoken truly ifhe has a beech in his
garden, or if his father has rheumatism. We enforce what in the
next chapter I call deferential conventions, meaning that we
recognize community authority, and expert authority, in pro-
viding the actual sense of words (4.6). We emphatically do not
allow that someone has spoken truly because in his private
idiolect 'elm' covers beeches as well. The reason why we do not
allow this is that it threatens the social utility of language: we
need social norms towards confirmity of usage if we are to rely
upon the messages made with words. Otherwise we could not
 HO\\' IS MEANING POSSIBLE? (2) 91

reliably tell what is to be understood by the utterances of an

individual.
Although this point is undoubtedly correct, it must not be
overstated. It does not deny that an idiolect could be a self-
standing language. It leaves it open whether an isolated indi-
vidual might have the determinate intentions or procedures to
afford a sense to his terms quite without any question of defer-
ring to the authority ofanyone else. For example, if the meaning
of a term were thought of along verificationist lines, as the kind
of procedure or experience which determines whether a term
applies, then an individual might himself invent and fix such a
procedure or such an experience (but see next section). It is just
that such an individual can never rely upon there being a
determinate meaning to a term when he does not himself know
what it is. He thereby differs from members oflinguistic groups,
who defer to others, as in the elm and arthritis example. I am
here dissenting from Dummett's treatment of this issue. 12
Dummett correctly takes the way we bind ourselves by deferen-
tial conventions to show that "there is no describing any indi-
vidual's employment of his words without account being taken
of his willingness to subordinate his use to that generally agreed
as correct". This is true. But he continues: "That is, one cannot
so much as explain what an idiolect is without invoking the
notion ofa language considered as a social phenomenon." This
is not true, or, if it is it needs a different support. For it is not
equivalent to the first claim: it implies that an individual cannot
do for himself what a society can do together (provide meanings
for terms), whereas the first claim says only that an individual
will not actually have done that, but will be bound by social
facts which he will recognize, or ought to recognize, and which
can split his actual meaning from his own understanding. The
stronger claim needs the idea that meaning has to be social, and
this goes beyond saying that it actually is social. 13
The essence of the matter is that there seems to be no impossi-
bility in an individual creating and abiding by his own rule of
use of a term. But if an individual cannot do this, the fact that
he is surrounded by others seems a doubtful source of help. The
12
'The Social Character of Meaning', in Truth and Other Enigmas, pp. 424-5. Also
'\Vhat is a Theory of Meaning?', in Guttenplan ed., Mind and Language, p. 135.
" I discuss deferential conventions further in chapter 4.
92 OL'R LA'.\!GCAGE AND OURSELVES

fact that there are lots of individually wooden people, forming a

public group, is not itself calculated to suddenly transform
them into a non-wooden community, sharing genuine identity
of concepts, with real rules of application.
Crusoe's idiolect is not a private language, in the s'ense which
that phrase bears in Wittgenstein's most famous development
of these thoughts about meaning: the anti-private language
argument. Crusoe's practice, for instance with his signs which
help to solve the Rubik's cube, is only accidentally private. If
Man Friday arrives, there is no reason why he should not be
taught the same procedures and rules. But the private language
which Wittgenstein opposes is private in a stronger sense. Its
terms are given their meanings by reference to private episodes,
such as sensations. We must now consider Wittgenstein's argu-
ments against such a language, and the place of those argu-
ments in philosophy.

4. Privacy and Practice

So far in this chapter we have tried to find what makes it true
that either (i) a person is using a word in one definite sense, or
(ii) he is using it with a different, bent interpretation, or (iii) he
is not using it with a sense at all, but is merely parading the term
under the impression that he is doing so. Some fact must
determine which of these is true. We have accepted the negative
point, that it is not a display to the mind which does it. We have
cast doubt upon the idea that the actual presence of a com-
municating group is essential: whatever fact it is which marks
the difference, it seems possible that an individual should satisfy
it by himself. In this way his earlier self can transmit informa-
tion for his later self to profit from,just as different members ofa
group can. In effect we are left with a pair of suggestions (not
necessarily exclusive): it is the existence of a practice or technique
which makes the difference, or it is the existence of a determi-
nate intention, known to the speaker in whichever way we know
about our own intentions, which makes the difference.
As is often the case in philosophy, a good way of exploring
these ideas is to see what they rule out. The anti-private
language considerations aim at this conclusion: no language
can contain a term whose meaning (or sense) is constituted by a
 HO\\' IS MEANING POSSIBLE" (2) 93

connection the term has with a private item, which lies, or lay,
solely in the mind of the individual who understands the term.
We can call the doctrine that rules out such a term, "semantic
externalism". 14 It is a doctrine about the terms of any language
at all, including ours. It tells us that no term of any language,
including terms like 'pain' 'tickle' 'experience as of seeing red',
have their meanings fixed by a certain kind of connection. The
importance of this claim is that the reverse doctrine is so tempt-
ing. It is tempting to say that I know what a term like 'pain' or
'burnt taste' means from my own case. Under some circum-
stances I have a certain kind of experience. Others, supposing
that I have it because ofmy situation or my reactions, teach me
to use some word to apply to it. I absorb their teaching by giving
myself a private ostensive definition: I focus upon the sensation,
say a pain, and promise to call just that kind of sensation 'pain'
in the future. It is the fact that I use the term in conformity to
that rule which identifies its meaning. The rule fixes the connec-
tion between the term and the private sensation, lying solely in
my mind. Semantic extemalism opposes this model. By oppos-
ing it, the doctrine threatens whole clusters of ideas in the
philosophy of knowledge and the philosophy of mind. It stands
against the thought that our best or fundamental knowledge is
of the contents of our own minds. It stands against the whole
Lockean model of language, whereby the immediate signifi-
cance of a word is an Idea in the mind of the person who
apprehends it: what Locke believes to be true of all words,
semantic externalism believes to be true of none. It eventually
alters the whole conception of the privacy of our own experi-
ences and sensations, although the consequences here are indi-
rect, and need a little explanation.
Semantic externalism says that terms like 'pain' do not have
meanings which are constituted by their connection with a
private item. Now we might decide that pains, experiences,
sensations, are precisely items of this proscribed private kind.
In that case the doctrine forces us to revise the idea that it is
through a connection with such things that any words have
their meanings. Alternatively, we might suppose that it is quite
certainly through that connection that the words have their
meanings. What could be more certain than that the word
14
I borrow the term from E.J. Craig.
94 OCR LANGUAGE A:\TD OURSELVES

'pain' means what it does because ofits connection with pain? In

that case we are forced to revise our philosophy of mind. Pains,
sensations, and experiences then cannot be items of the
proscribed kind, and we must seek an account on which they
are not items of private acquaintance, but are thought of in
some other way. On the first alternative the private language
argument, and the semantic externalism which is its conclu-
sion, remain doctrines about meaning. We can think of minds
in a traditional ("Cartesian") way as private repositories of
experience, sensations, perhaps intentions, of which the subject
has a privileged acquaintance. But our ideas of what gives the
meaning to terms like 'pain', 'experience of seeing red', and so
on, have to be altered. On the second alternative it is the
conception of privacy itself which is threatened.
The first alternative seeks to reconcile Cartesian privacy with
semantic externalism. Its strategy is to distinguish between the
sense or meaning of a term like 'pain', which is not given by its
connection with the private experience, and its reference, which
may yet be the private content. But although the sense/refer-
ence distinction is quite legitimate (see chapter 9) it cannot
effect this marriage. The distinction is at its most visible when
we take phrases which can be fully understood (i.e. whose sense
can be fully apprehended) when it is not known to whom or
what they refer. Thus I perfectly understand many sentences
containing definite descriptions ('the person who committed
this crime'; 'the richest man in the world') although quite
ignorant of who or what it is that they refer to (see 9 .1). This is
why I can understand and obey instructions like 'Look for the
person who committed this crime', or statements like
'Economists do not know who is the richest man in the world'.
The sense of these sentences is a function of the sense of the
individual words occurring in the descriptions, and this is quite
independent of whether x ory is the person who committed the
crime, or the richest man in the world. But 'pain' does not
function like this. There is no understanding of the term by
people who do not know what pain is. It is through knowing what
pain is that we come to understand pain ascriptions, whereas it
is not through knowing who is the richest man in the world that
we come to understand sentences using the description.
Because of this we cannot separate out two processes: learning
 HO\\' IS '.IAEA"ll"IG POSSIBLE" (2) 95

the sense of the term 'pain', perhaps compatibly with semantic

externalism, and then learning to what it refers ( the Cartesian
private item). Rather, it is by knowing what pain is that we
come to understand the term, and what it means to apply it. If
this knowledge is essentially knowledge of a privately shown
item, lying solely in the mind of the subject, then semantic
externahsm is refuted.
For this reason, the second alternative is more promising: if
we accept semantic externalism, we must revise our whole
conception of the privacy of the mental, and our knowledge of
the contents of our own minds. But what is there to be said for
semantic externalism? What is the force of the anti-private
language considerations?
Let us consider a proposed case of private ostensive defini-
tion. A man has a certain kind of sensation. This sensation has a
"phenomenal quality" which is known to him alone: he is aware
of it, just by having it. He can attend to it, like it or dislike it,
relish it, and, let us suppose, christen it. By this christening he
(purportedly) provides himself with an intended rule: in the
future call only this kind of sensation, 'S'. This rule would
determine what is correct application of the term 'S' and what is
incorrect. A later sensation with its own definite phenomenal
quality would be rightly called 'S' ifit falls within the intended
range of the term, and wrongly called 'S' ifit does not, but the
subject mistakenly takes it to do so - perhaps by forgetting
the actual nature of the original example. To use one of
vVittgenstein's metaphors, the intended rule provides a mea-
sure or yardstick to lay alongside a new sensation, which will
then conform with it or not, as the case may be. Let us say that a
man isfaithful to the original christening if there was an original
episode of ostensive definition of this kind, which gave him a
definite intention to call only sensations of a certain kind of
quality 'S', and if the man later uses that intention as a rule
which allows some sensations to be 'S', and disallows others.
Wittgenstein's endeavour is to show that there can be no such
truth as this: no truth that a man is (or is not) being faithful to
the original episode of this kind. The appearance must be a
sham, and hence the idea that a man is later judging a new
sensation to be 'S', or not to be 'S', is also a sham. For genuine
judgement demands faithfulness to a pre-existing rule.
96 Ol"R L\:'\GC.\GE A'.\iD OURSELVES

Otherwise the later occasion shows nothing but the man mak-
ing a new decision (I'll call this 'S' /I'll not call this 'S'). Such a
decision would not be responsible to anything that happened
previously. Hence it would not be correct or incorrect, and it
cannot be regarded as the making of a judgement. For judging
is something which is essentially capable of being correct or
incorrect. In this case, according to Wittgenstein, nothing
previous created a standard whereby this can be so. Once
again: "whatever is going to seem right to me is right. And that
only means that here we can't talk about 'right'."
\\'hy can there be no truth that a man is being faithful to an
intended rule, whose content was fixed by the first sensation?
\Vhen the later case arises he might say, "Ah, here is an 'S'
sensation again, and he has the impression of the term 'S' having
a definite sense, so that this remark makes a judgement. But
being under the impression that you are following a rule is not
sufficient to be truly following a rule. We have already met the
no-rule hypothesis, or possibility of a subject who thinks that he
is following a rule when he applies or withholds some term, but
who is like a lunatic covering pages with "sums", or like the
man whom Wittgenstein considers in§ 237 of the Investigations,
who intently follows a line with a pair of compasses, with one leg
on the line, and the other following at a distance, but at a
distance which he constantly alters by opening and shutting the
compasses as he draws the points along. This man may think he
is tracing a path defined by the first line, but not be doing so.
For since nothing would be a violation of this rule, the
hypothesis that there is a rule is mere show. A rule must allow
some procedures and disallow others.
In the public case the "wooden" individual, whose use of a
term is not rule-governed, can be detected, because his practice
is eventually different from that of someone making genuine
judgement with the term. The lunatic's "sums" form no part of
the practice of an applied mathematics (if they do, we might
revise the opinion that there is no method in them). But in the
private case, only the subject himself is an authority on whether
his applications of the term conform to an intended rule. So
Wittgenstein can ask what, in the private case, is the distinction
between (a) someone who is genuinely faithful to a pre-
established rule, which determines correct and incorrect appli-
 HOW IS MEANING POSSIBLE? (2) 97

cation of 'S', and (b) someone who is disposed to use the term
under the illusion that he is following a rule determining its
application?
At this point one is inclined to concentrate upon the
phenomenology of the matter. Give yourself a sensation,
remember it, and ask whether a later sensation is the same or
different. It seems a well-formed, well-understood question, at
least if you take care to specify various respects of sameness, like
intensity, or felt location. But there is a possibility, even if it
is one you are likely to dismiss, that you later misremember
what the original sensation was, and hence misapprehend the
intended rule it was used to introduce. You have then the new,
candidate sensation, and a memory of the intended rule, fixed
by the old exemplar. But the memory would be deceiving you.
It would lead you to think that the new example is very like the
old, and deserves the same name, when in fact it is quite
different. Let us call this possibility (c).15
If Wittgenstein is allowed to use the verification principle, he
is well placed to attack the idea that there is a real distinction
betwen (a) and (b) and (c). For anything the subject does or
experiences at a moment, or himself says, is compatible with
each hypothesis. And the public is in no position to tell which is
true either. No third person can tell whether the later sensations
are really like the first, or really different, or whether the subject
is really following no rule at all in what he calls 'S'. If there is no
verifiable difference between the three hypotheses, then by the
verification principle there is no real difference between them.
But for the term 'S' to be meaningful there must be a difference
between them, for it must be rule-governed and permit of
incorrect application.
What is much more doubtful is whether Wittgenstein can
reach this conclusion without relying upon a verificationist
step. Many writers suppose he can. 16 They think that the
challenge to say what makes the difference, in the private case,
has its own force. It is not just a question of how we might tell
15
"Always get rid of the idea of the private object in this way: assume that it
constantly changes, but that you do not notice the change because your memory
constantly deceives you". Philosophical Investigations, II, p. 207.
16
e.g. A. Kenny, 'The Verification Principle and the Private Language Argument',
in 0. R. Jones (1971); C. Peacocke, 'Rule Following: the Nature of Wittgenstein's
Arguments', in Holtzman and Leich (1981).
98 OCR LA'.'IGUAGE /\.ND OURSELVES

which one is true, but of whether we have any conception of

what would make one of them true. The challenge is to state this,
and it is alleged that the would-be private linguist has no
answer.
The challenge needs quite delicate handling, or it threatens
to destroy public language as well. That is, in so far as the
individual has difficulty in meeting it, it is also possible that a
group does. We have already seen how hard it is for a group to
defend themselves as being genuine rule-followers, sharing an
identity of concept, rather than a mere number of wooden
individuals, or a mutually bent community holding quite diver-
gent interpretations of shared predicates. But even if we waive
this problem, perhaps because we draw on the idea of a
practice, or technique with a term, which saves the public, the
prospects for the challenge are not all that bright. This is
because it is rather hazy whether the challenge, to show what
makes the difference, can legitimately be met just by repeating
the description of the three cases. Suppose the private linguist
defends himself by saying: "We already know what makes the
difference. In the one case there is a rule, and it is determinate
whether an application conforms to it; in the second case there
is merely illusion; in the third case there is a misremembering of
which rule was established. If there is a challenge to verify these
hypotheses, then unfortunately it cannot be met. But that is
often the way with sceptical challenges, and does not disturb
the genuineness of the distinction." The challenger will impa-
tiently reply that this is not good enough: he wants to be shown
what the distinctions consist in, in the private case. But what
does this mean? Perhaps only that the distinction should be
drawn in other ways, themselves making no mention of inten-
tions, or rules, or fidelity to a pre-established sense. But why
should this request be legitimate? A distinction made with one
kind of vocabulary often cannot be captured except by using
that vocabulary: the distinction between red and green is essen-
tially a distinction of colours, and cannot be shown to "consist
in" some difference which does not refer to colours. The distinc-
tion between happiness and pleasure is a psychological distinc-
tion and cannot be made except in terms from that theoretical
vocabulary, and so on. Notice too that by urging the negative
point, Wittgenstein has already led us to think of intentions as
 HOW IS MEANING POSSIBLE? (2) 99

irreducible, in sense that the issue of the intentions with

which a person uses a term is never just the issue of which
display he has before his mind.
So Wittgenstein's challenge is objectionable if it presupposes
that the distinctions between (a), (b), and (c) must be capable
of being drawn in other terms. It is also objectionable if it
threatens public language as much as private language. Follow-
ing Kripke we can put Wittgenstein's challenge in the garb of
scepticism: this sceptic denies that anyone can know whether,
after an attempted private ostensive definition, and attempted
further use of the term, it is really (a), or (b), or (c) which is in
force. 17 But then we have already discovered the scope for
equivalent scepticism in the public case: the sceptic who wins
against the private linguist looks well set to win against a public
group when, corresponding to the three hypotheses in the pri-
vate case, he asks whether the right-rule view, or the no-rule
view, or the bent-rule view, is the true one. If the moral of the
rule-following considerations is that we, the public, cannot
meet this challenge except by insisting that we do know what we
mean, and that we mean the same, and know this by knowing
our intentions, then the would-be private linguist can avail
himself of the same liberty.
If the would-be private linguist sits tight on his claim to have
a determinate intention in calling a new sensation 'S', the only
way forward is to concentrate upon the notion of a practice, or
technique. Suppose we accept that understanding a term is
possessing a skill with it, and that this skill is to be thought ofas
a kind of technique or ability. All terms ofa language must be
associated with such a technique. Then perhaps there is argu-
ment to show that the term 'S', introduced by private ostension,
equips its user with no genuine technique at all. It forms no part
of a practice whose proper pursuit stands the user in good stead,
and whose improper pursuit leads to errors and disappoint-

17
One must be careful of framing these issues around the figure of a sceptic.
\\"ittgenstein's point is never to arrive at a conclusion of the form 'so we don't know
whether ... 'His aim is to alter our conception of the facts we take ourselves to know: the
aim is metaphysical. But his means to such conclusions may use sceptical dialogues as
an integral part: if this conception of the facts were the right one, then we wouldn't know
such and such, but we do, so we need this other conception of the facts. Again, I discuss
this further in 'The Individual Strikes Back'.
100 OUR LANGUAGE AND OURSELVES

ments. At first sight this idea seems promising. For precisely

because there is no verifiable difference between (a) and (b) and
( c), it seems that a "mistake" in applying 'S' is utterly inert. Let
us adopt the standpoint which Wittgenstein is attacking for a
moment. Imagine two prospective private linguists, each giving
themselves what is in fact the same ostension, and coming away
with the same intention. One has a better memory or better luck
than the other, and only applies 'S' thereafter to cases which do
fall within the range of the original intention. The other is fickle
and faithless and often applies 'S' to cases which should have
been excluded. How is it that one does well and the other does
badly? What cost does the errant linguist incur? Apparently,
none whatsoever. And if this is so it suggests that there is no real
failure of a technique or practice here. A technique is essentially
something which has consequences, and whose failure can let
us down. So the hypothesis that the would-be private linguist is
really operating a technique seems to be pure show. The normal
surroundings and stage-setting of the technique of judgement
are missing: he is like a man driving an imaginary motor car, or
playing an imaginary piano. Consequently it cannot matter
whether we regard him as a case of (a) or (b) or ( c).
The obstacle to this range of thought is that it is we who
introduced the S-classification as an isolated and inert incident,
with no consequences for good or ill. But suppose on the con-
trary that the private linguist's performance is part of a tech-
nique which he is forming, testing, trying to render reliable. The
technique is to bring order into his life. By correlating the
recurrence of one experience with the recurrence of others of
related kinds (warmth - pain! visual experience x - tactile
experience y, etc.), the private linguist can begin to find order in
his subjective world, and an ordered subjectiv,e world is a nice
thing to have, since without it we can have no understanding of
ourselves as conscious of an objective or spatially extended
world. So the enterprise to which the classification of private
experience belongs can be a serious one, and it can go well or
badly, and it can matter to the subject why it is going badly. If,
for instance, S which is usually followed by R is on some
occasion not so followed, the private linguist may doubt their
general correlation, or doubt either or both of his classifications.
And it can be a serious matter to decide where the failure lies,
 HOW IS MEANING POSSIBLE' (2) 101

and to accommodate future classifications and expectations

around it.
In this circumstance there is no reason at all for the private
linguist to take the attitude that whatever seems right is right.
He may do better to take the attitude that his memory is not
totally reliable, that it is easy to fail to notice genuine differences
between 'S' and sensations like 'S' but importantly different in
what surrounds them, and so on.Judgements ofrecurrence take
their place as corrigible in the light of subsequent experience.
The moral will be that there is a point to discriminating the
private linguist's performance as genuine judgement, capable
of truth or falsity in the light of pre-established intentions, only
when the performance is part of some general technique of
belief-formation. It would then follow that beliefs have to come
in populations, but not that believers do. The would-be private
linguist's title to think of himself as a believer would be derived
from his title to think ofhimselfas a theorist, attempting a whole
set of views about the order of his mental life.
A philosopher impressed by the challenge to make the dis-
tinction between (a) and (b) and (c) will still complain. Perhaps
we can suggest why the private linguist may pointfully take the
attitude to himself, that he is making genuine judgement. He
need not take the attitude that anything that seems right will be
right, nor that his dispositions to call things 'S' answer to no
previous intentions. They have the crucial normative dimen-
sions of correctness or incorrectness. But, the opponent will ask,
how is this attitude justified? If there is no fact of correct or
incorrect applications of the term 'S' to a new sensation, then
surely the attitude itself involves a delusion. It is not enough to
say that a private linguist (or public group) may dignify or
compliment himself on being a rule-follower or on making
judgements which have a genuine dimension of correctness: the
compliment must not be empty. There must be a fact of the
matter whether he is one or not. And we still have no conception
of how the original episode reaches out, as it were, to constrain
the proper use of 'S' on any subsequent occasion.
I do not think that it is true to the later Wittgenstein to pursue
the challenge this way. 18 Firstly, Wittgenstdn is on the side of
those philosophers who query the borderline between genuine
18
This is another point of difference from Kripke.
102 OCR LANGUAGE AND OURSELVES

beliefs (in facts, with truth-conditions) and other kinds of commit-

ments (such as possession of attitudes, or acceptance of rules).
The error of his earlier philosophy was to exalt a simple, single
conception of a fact (as in effect a spatial array of objects), and
to make no room for truths expressed in terms which do not
refer to such spatial arrays: for instance, truths about causal
relations, about psychology, about the will, or about ethics. The
later work is acutely conscious of the way in which our difficulty
in conceiving of psychological facts arises from a spatial or
physical model of what a fact must consist in. This is why we
find our ability to think of absent things or to form intentions
which cover cases which we have not thought of, so mysterious.
But the characteristic tone of the later work is one of toleration
towards different vocabularies, even when we "have no model"
of the truths they describe. 19 A pertinent example is
Wittgenstein's famous reaction to the talk of the belief that
another person is conscious: "My attitude towards him is an
attitude towards a soul. I am not of the opinion that he has a
soul. " 20 The belief or attitude is not a belief in a certain fact (a
spiritual interior to the animal, or a ghost in the machine), but is
something whose content is given by my reactions to the person,
ways of behaving and dealing with him. Against this back-
ground it would not be appropriate for Wittgenstein to insist
that the private linguist can only take up an attitude towards his
own classifications ( the attitude that they answer to a previous
rule or intention), but that there can be no fact about whether
they do: the justifiable attitude is just the kind of thing to give
content to this fact. (I offer my own exploration of the attitude/
belief distinction in chapter 6.)
When the would-be private linguist classifies a new sensation
as 'S', he thinks of himself as laying it alongside his rule. The
rule came into his possession after the private ostension; its
content is given in his intention which was formed on that
occasion, to call only things like the original example by the
term. Wittgenstein's brilliant strategy was to shrink this alleged

• Philosophical Investigations, § 192. Cf also the discussion in Remarks on the Philosophy of

1

Psychology, vol. I, when Wittgenstein says of expressions of intention: "Yes, and such
use of language is remarkable, peculiar, when one is adjusted only to consider the
description of physical objects"(§ 1137).
20
Philosophical Investigations, p. 178.
 HO\\' IS '.\IEA:\ll'.\JG POSSIBLE? (~) 103

comparison down to a point, leaving nothing but the new

sensation, and a bare disposition to say 'S' or not. The previous
history drops out from the use of the term, for everything about
that use could be the same whether or not your memory is
utterly deceiving you about the character of the original exam-
ple (case ( c) ) , or whether you are only under the impression
that you are really guided by the nature of the original example
(case (b) ) . The difficulty is to destroy our stubborn conviction
that we are right to believe ourselves to be in one category and
not in either of the others. Broadly speaking there are two
possibilities. Wittgenstein can put on the garb of a sceptic, and
allege that we do not know the nature of the previous episode,
and the intention which it was used to form. Or, he can raise the
metaphysical or ontological charge that we have no conception
of the fact that we are genuinely guided by the previously
formed intention. Either way, the issue is likely to remain
inconclusive. The sceptical charge is too near to blockbuster
scepticism, which would destroy our knowledge of anything,
especially of public meanings. The ontological or metaphysical
charge is too near to insisting that there should be a fact of our
being guided by a previous intention which can, as it were, be
laid out to view in the form of an image in the head or other
guide. This is just what the negative point attacks. And then, if
the private linguist refuses to try to force his fact into this kind of
shape, it is hard for the later Wittgenstein to deny his right to do
it.

5. Exercising Mental Concepts

Even if the anti-private language argument is inconclusive, it
does a tremendous service in the theory of knowledge. It
entirely subverts the idea that our knowledge of our own mean-
ings, derived from the acquaintances we have with our own
mental lives, is a privileged, immediate, knowledge, beyond
which lies only sceptic-ridden insecurity. This Cartesian
picture, according to which my knowledge that my present
sensation is green, or hunger, or a headache, is peculiarly
incorrigible, is, I believe, overthrown by the realization that the
only "incorrigible" element is the single point, the present
sensation; any enterprise ofjudging it to be one thing or another
104 OCR L\:\'GUAGE AND OURSELVES

involves bringing it into contact with a rule or previous inten-

tion, and hence brings the possibility of misidentification of
what that was. But this does not entail that there could be no
such judgement. It only entails that it has no unique immunity
from error.
There is one further aspect of this difficult area which needs
mentioning. I characterized semantic externalism as the denial
that the meaning of any term can be "constituted by its connec-
tion" with a private exemplar. The same thought is sometimes
put by denying that it is "from our own.case" that we learn the
meaning of sensation terms. The alternative account of the
meaning of such terms is likely to stress public criteria of their
applicability. To understand a sensation term requires knowing
what kind of situation or what kind of display would make it
appropriate to attribute the sensation to a third person: it might
also demand knowing that people can sham, or be hypnotized,
or whatever, into appearing to have sensations when they do
not. But this, it is suggested, gives no ground to Cartesian
privacy. It only shows that our public evidence is inconclusive
or "defeasible" by further public evidence, which could arise if
peoples' motives for shamming, or queer state of hypnosis, were
removed.
Accuracy demands a little suspicion of some of this. There is
at least room for a theory which admits that it is not wholly from
one's own case that one understands a sensation term, or that
the meaning of such a term is not wholly a question of its
connection with the private example, but also insists that it is
partly from one's own case that one understands the term. That
is, a full grasp of sensation terms would require knowing what
the sensation is like (privately), and realizing that other sub-
jects might or sometimes do possess it too. It might also require
understanding the kind of evidence which justifies the belief
that they do. It is an awkwardness of the anti-private language
argument that it seems to rule out even this diluted semantic
internalism. It is so strong that it leaves no role for the internal
examplar. For the subject can only be under the impression that
his own example of the sensation plays any role at all in identify-
ing a rule of application, and this impression is, supposedly, not
enough to make it true that it plays a role. And nothing else can
make that true either.
 HO\\' IS '.\IEANING POSSIBLE? (2) 105

Even dilute semantic internalism faces or raises difficulties.

So far I have concentrated upon what we can call a "vertical"
version of the private language argument. This considers one
agent and his relation to his past states. But there is a "horizon-
tal" aspect of the argument too. This questions whether dilute
semantic internalism can allow any sense to the thought that
some other subject has the same kind of sensation as oneself. To
make the problem vivid, imagine someone arguing that he does
indeed get the concept of pain from his own case; his own case
enables him to tell when there is pain about, which is to say,
when his body is injured or he is affected in some unfortunate
way; hence no other events are associated with pain at all. Your
injury is just not the sort of thing which causes this sensation. He
never feels it when you are injured. The concept of sensation is
exercised "vertically" as it were: in the one dimension of my
own feelings.
The challenge is to explain how, if our basic use of the
concept is in our own case, we can ever come to exercise it in full
generality. How do we understand that we are each just one of
the many creatures which equally have sensations (even ifwe
then go on to wonder how like our own those of our fellows
might be)? This kind of challenge is of great importance in
philosophy. For example, Berkeley anticipated Hume in find-
ing it hard to understand how we gained a concept of causation
by acquaintance with the ("passive") flow of events in the
world which we sense. He proposed instead that the origin of
the idea lies in our knowledge of our own agency or exercises of
will. 21 But (as he realized) that makes it impossible to see how
we can properly describe non-mental things as causing any-
thing. Again Hume describes the origin of our idea ofjustice in
the need for a scheme or system of rules whereby we can gain
reciprocal advantages from one another: there is a problem
then whether he can make sense of our idea that we have a duty
of justice to future generations, or animals, who cannot
reciprocate. 22 We could put the challenge by asking why, on
these theories, the term in question is not ambiguous, meaning
one thing in the home case (from which we get the idea) and
another thing in the further cases. How can it be the same
21
Principles ef Human Knowledge,§ 25.
22
Treatise ef Human Nature, Bk. III, Pt. II, Sect. I.
106 Ol'R LA'.\iGL'AGE .\'.\iD OURSELVES

concept, which is exercised in the home case and in the further

cases?
If the challenge were just one of explaining the genesis of our
understanding, it would seem feeble.We might reply that it just
comes naturally to us to extend the concept from the one kind of
case to the other. But the real challenge is to say why it is the
same concept in each case. Is it responsive to the same kind of
evidence or argument; do we appreciate its consequences in the
same way? In the current example a dilute semantic intemalist
cannot just say, for instance, that in my case I recognize and
remember my private exemplars, but in your case I exercise the
concept of pain by, for example, caring about you or reacting
with emotion when I judge that you are in pain. 23 He must go on
to connect these two exercises: to show why it is appropriate to
talk of one concept which I apply equally to you and to me.
However there are steps towards meeting this demand which
the semantic intemalist can take. He can point out that many of
the things I know (or believe) about my own pains I also believe
about yours: that it is no accident that this kind of sensation
makes me behave in this kind of way; that there are ways in
which I can conceal sensations, that sometimes I cannot, that
there are ways for you to behave towards me if you believe me to
have a sensation and also have various attitudes towards me,
and that these are ways I behave towards you in the light of the
same beliefs. In other words, I can judge so that my sensations
and yours are subjects of similar predicates and claim that the
same concept is involved just because of that. It would, I
believe, be extremely hard to phrase the demand or challenge so
that this kind of answer does not meet it. Consider this parallel:
people think it makes sense to ask whether numbers are objects,
but how can the question even make sense when the notion ofan
object is at home in talking of ordinary spatio-temporally
located and bounded, solid and visible things? Answer: because
(perhaps) we say enough about both numbers and more ordi-
nary objects to explain the common term (they are equally
referred to, counted, known about, independent ofus ... ) 24
So semantic internalism can attach a meaning to the view
21
· Although curiously this is at least a part of\Vittgenstein's own thought.
24
So we do have identity of meaning, or only similies or metaphors? Numbers are
like objects in that ... ; unlike them in that ... See more on this in 5. 7.
 HO\\' IS MEANING POSSIBLE? (2) 107

that other people have sensations like ours. This leaves the
question of whether we know that they do, and how we rebut
scepticism about their similarity to us. But that takes us too far
from the philosophy oflanguage. The present verdict is that the
private-language considerations seem at best inconclusive.
There is no compelling reason why there cannot be a practice of
judging that our own private sensations are thus-and-so. And
the intention with which we apply the classification may, so far
as the argument goes, be identified by private ostension. There
is equally no compelling reason why such a practice should not
also serve to identify (part of) the meaning of our public sensa-
tion terms. The two elements of meaning which we have been
forced to make prominent in coming to this conclusion are,
firstly, the relation between meaning and intention, and sec-
ondly the relation between meaning and a whole practice of
coping with the world. I propose to pursue the first of these in
the next chapter, and the second infuses the next part of the
book, in which we consider ways in which we judge the world,
and the kinds of truth they deliver.

Notes to Chapter 3
3.1 The central refi;rences for this chapter are:
L. Wittgenstein, Philosophical Investigations §§ 134-230.
N. Goodman, Fact, Fiction and Forecast, ch. IV.
B. Russell, Human Knowledge, its Scope and Limits, pp. 422 ff.
Russell makes only passing reference to bent rules. Goodman
uses them to shed new light on the problem of induction, but bent
rules and no rules are used to cast their shadow over meaning by
Wittgenstein. Easily the best commentary to date is Saul Kripke's,
referred to in the text (n. 4 above).
"It does not follow that the subject himself is not an authority ... "
It is vital to see that Wittgenstein is not denying that there is such a
phenomenon as rule-following (nor that there are facts about which
rule is in force, not that it is true or false that there are rules in force).
The whole debate is about the conception of these facts that we can
obtain. Even the best modern commentators (including Wright and
Kripke) do not bring this point out fully enough; they leave it uncer-
tain whether we can properly allow facts of this kind. They have a
. good excuse, because we have here a classic philosophical problem: a
108 OUR LA;\;GUAGE AND OURSELVES

critique of one conception of a kind of fact is so powerful that it leaves

people unsettled whether we can any longer go on saying the things
we used to say about the area. This predicament is discussed further
in 5.1 and 5.2.

3.2 "A slightly more substantial answer ... "

Although I say enough in the text to register my disagreement with
both conventionalist and naturalistic approaches to Goodman's
problem, a fuller discussion would bring in much more. I discuss the
paradox with reference to induction in Reason and Prediction, ch. 4.
Important discussions of Goodman's paradox include:
S. Barker and P. Achinstein, 'On the New Riddle of Induction',
Philosophical Review ( 1960).
J. J. Thomson, 'Grue' ,Journal of Philosophy ( 1966).
P. Teller, 'Goodman's Theory of Projection', British journal for the
Philosophy of Science ( 1969).
S. Shoemaker, 'On Projecting the Unprojectible', Philosophical
Review ( 1975).
See also 7. 7

3.3 "A compromise is suggested by Saul Kripke ... "

I give a more detailed discussion ofKripke's views in 'The Individual
Strikes Back', Synthese (1984) (this volume, edited by Wright, also
contains relevant papers by J. McDowell and C. Wright).
" ... taking his later work to release religious faiths ... "
There is a trenchant discussion of post-Wittgensteinian views of
religion inJ. Mackie, The Miracle of Theism, ch. 12. It is, in my view,
very uncertain whether the stress of religion as a human phenome-
non, giving vent to human needs and feelings, actually conflicts with
the view that religious beliefs are real beliefs, capable of truth or
(usually) falsity.

3.4 Any selection out of the huge exegetical and critical literature on
the private language argument is bound to be fairly arbitrary. Treat-
ments which should profit students, in addition to those already
mentioned, include:
A. J. Ayer, 'Could Languge be Invented by a Robinson Crusoe', in
0. R.Jones (1971).
R. Fogelin, Wittgenstein (The Arguments of the Philosophers), chs.
XII and XIII.
J. J. Thomson, 'The Verification Principle and the Private
Language Argument', in 0. R.Jones ( 1971 ).
 HO\\' IS '.\1EANING POSSIBLE? (2) 109

C. Peacocke, 'Rule-Following: The Nature of Wittgenstein's Argu-

ments', in Holtzman and Leich (1981).
But the best modern discussions, triggered by Kripke, seem to me to
be nearer to the real heart of Wittgenstein's problems.

"I do not think it is true to the later ... "

I here agree with two other writers who have explored Wittgenstein's
later conception of fact:
P. Winch, 'ImAnfangwardieTat',and
B. McGuiness, 'The So-Called Realism of Wittgenstein's
Tractatus', both in Block, ed. ( 1981).

3.5 "horizontal" and "vertical"

This adverts to the extensive literature which believes Wittgenstein to
have subverted the familiar argument from analogy for other minds.
The idea is that we cannot learn what pain is, in our own case, and
then as much as. understand what it could be for someone else to be in
pain; hence we cannot argue by analogy with our own case that other,
similarly behaving bodies belong to subjects with similar mental
experiences: we cannot understand what this means. A good collec-
tion on this problem is The Philosophy of Mind, ed. V. C. Chappell,
especially the editor's introduction, and the papers by Malcolm and
Strawson.
I am very conscious that this section suggests more profound
problems than it manages to treat. It is probably fair to say that the
philosophical community, at present, is involved in a general shift
away from supposing that anything is ever learned in our own case-
this being supposed to involve a "Cartesian" conception of mind,
whereby the contents of our own minds are immediately present,
private showings, whose nature wholly determines our thought. The
whole difficulty is to separate what is right about the direction, from
what is wrong, or questionable, about the individual theses which
people have taken to support the direction. Although I am sceptical of
some of the arguments from the philosophy of language which have
been used to support the shift, I have no settled opinion on the shift
itself (the time will come when people begin to ask what was right
about Cartesian intuitions in the philosophy of mind). In this work I
am only concerned with the arguments as they emerge in connection
with understanding and meaning. This is also evident in chapter 9,
where in order to avoid a Cartesian, or even solipsistic view about the
nature of thought, philosophers have been led to defend unduly
implausible views about reference and thought.
 CHAPTER 4

Conventions, Intentions, Thoughts

"I saw you take his kiss!"" 'Tis true"

"Oh Modesty!" " 'Twas strictly kept:
He thought me asleep; at least I knew
He thought I thought he thought I slept."
Coventry Patmore,
"The Kiss".

1. Grice's Approach
What change is involved when someone comes to possess a
language? What is the difference between linguistic and non-
linguistic creatures - what difference are we looking for if we
debate whether whales or chimpanzees really use language?
The last two chapters may leave us dissatisfied with an answer·
which merely cites thoughts, and describes language mastery as
an ability to display those thoughts to others. We might
sympathize with the idea that somehow the practice of trans-
mission, and the grasp of symbolism with which to do it, actu-
ally creates our ability to have the thoughts. But this remains
mere speculation, unless we get a better focus on what it is to
have a language anyway.
The classic paper introducing modern attempts to analyse
communication was written by H. P. Grice. 1 Grice's approach
has been much debated, rejected, altered, and improved upon.
The basic ideas can be appreciated without following many of
the elaborations which have emerged, and I shall in any case be
attempting to show why many of the complexities which
abound in the literature are misconceived.
Grice presented his analysis as an attempt to locate a notion
of"non-natural meaning" -the kind of meaning a sign or action
or utterance may have, not because it is naturally a symptom of
something else (in the sense that red spots mean measles, or
1
H.P. Grice, 'Meaning', Philosophical Review ( 1957).
 CONVENTIONS, !NTENTIO:\IS, THOCGHTS 111

clouds mean rain), but because it is somehow intended to signify

something. Thus I might present you with the head ofjohn the
Baptist, intending to show you that he is dead, or draw a picture
of Mr. X being intimate with Mrs. Y, intending to show you
that this is how they are; or I might throw a bucket of water over
you, intending to suggest that you leave; or open a window,
intending to have you look out and see that it is raining.
Suppose then an action done by one party ( the utterer, although
the action need not at this stage involve language) and directed
to another ( the audience). The primitive basis, then, is an
utterer doing x with the intention that the audience comes to
believe something, or comes to do something. Call such an
action an AIIB-an Action Intended to Induce Belief. (Follow-
ing most writers, I shall concentrate upon transmission of belief
for the moment, leaving commands, etc. until later.)
Now an AIIB may aim at inducing belief that p without
signifying or meaning that p. It seems stretched to say that by
opening the window I meant that it was raining. But it seems all
right to say that by drawing Mr. X and Mrs. Y, I meant or
communicated that they were on certain terms. So what is the
difference? Grice effectively located it in the way the transmis-
sion of belief works. In the window case the audience may take
the fact that the window is open, or the fact that I evidently
want him to look out, as a reason for looking out, and having
done that he sees the rain and acquires the belief. But it is not
important to this process that at any point he should realize that
I intend him to believe that it is raining. By contrast, when I
draw the picture, unless the audience realizes that I do so with
the intention of getting him to think that the couple stand in a
certain relation, there is no likelihood that he will then come to
believe that they do. In other words, the mechanism of getting
the audience to have the belief involves in this case the audience
recognizing that this is what is intended.
We can appreciate the difference in this way. Suppose you
and I are in what I shall refer to as the "one-off predicament": I
need to get you to believe something, p, and I can rely upon no
shared language. (The point is to build an account of what it is
to have a shared language, so we obviously cannot help
ourselves to its properties at any stage.) In the one-off predica-
ment I must come up with an AIIB. One such would be
112 Ol'R L\:'\GCAGE A:\ID OURSEL\'ES

engineering a natural sign that p. For instance, if there is

quicksand about, I might rig up a device for dropping a stone in
front of any traveller; the stone will sink and the traveller realize
that there is quicksand ahead. If I am a brain surgeon I might
be able to directly modify the audience's brain, so that they then
believe that p. But the scope for this direct action is limited.
Now in the one-off predicament there is only one other way of
proceeding. I have to do something which you are likely to take
in one specific way. The natural way to do this, given the ways
we interact with other people, is to perform some action with
the intention that you realize why I am performing it. I might,
for instance, draw a skull on a post, hoping that you come along,
see the skull, ask yourself why someone should draw that, and
become alerted to the quicksand. I would rely upon your
perception of my intention, for there is nothing else to rely upon.
And we can imagine ways in which that reliance would be well
placed, although they involve rather primitive messages. This is
because unless we have shared habits of ways of taking utter-
ances or actions, it is not so easy to do anything which you are
likely to take in any one specific way.
Let us call an action intended to induce the beliefthatp, and
relying for its effect upon the audience appreciating that this is
the intention behind it, a Gricean AIIB that p (GAIIBp). Then
we must certainly applaud Grice for suggesting a mechanism to
be used in one-off predicaments. But what is the next step? The
line which Grice himself inaugurated is to ask whether a GAi IBP
means thatp, or whether by such an action the utterer means that
p; or even says that p, or communicates (if successful) that p.
The promise is this. If we can wrap up a nice condition for an
action in a one-off predicament meaning that p, then we can
expect it to be a short step to saying what it is for an action or
utterance to mean in general that p (in the mouths of some
group, say). It would be, roughly, for them to have it as a habit
or convention that the action is performed with the intention of
inducing the belief that p, relying upon the Gricean mechanism.
Regular meaning would be fossilized one-off meaning.
This direction of explanation is called "linguistic nomin-
alism" by Jonathan Bennett although the title is a trifle mis-
leading. (It has nothing to do with nominalism in the theory of
predication.) We should notice one very odd feature ofit. In the
 CONVENTIONS, INTENTIONS, THOCGHTS 113

one-off predicament I have to rely upon Grice's mechanism -

that is, your recognition of my intention in uttering - because
there in nothing else for you to go on. But evidently this is no
longer true once we have habits of taking utterances one way or
another. If I have to gesture or mime or otherwise invent a
performance I must hope tnat you appreciate why I am doing
it. But once we have methods of communicating into which we
have been trained, perhaps I need not care at all if you recog-
nize my intention in uttering. It would be enough if you heard
my words, because you will have been trained to take them in a
certain way, and so taking them, you will understand me. Since
this is so, linguistic nominalism has a powerful rival. This rival
would avoid importing conditions on one-off communication
into the full, general case of shared communication through
language. We could thank Grice for his insight into the one-off
predicament, without in any way supposing that the complex-
ities which enable people to resolve it must always be present,
even when they have shared habits to rely upon, enabling them
to bypass the complexities. In other words, conventions or
habits would not need to fossilize complex Gricean conditions -
they would supplant the Gricean mechanism, which is only
needed in their absence. Where xis the AIIBP the difference is
between:
fossilized
one-off AIIBP(x)- one-off GAIIBP----- regular meaning
of x that p

which is linguistic nominalism, and

fossilized
one-off AIIBp(x) regular meaning of x that p

Lone-off GAIIBP

The second route represents the idea that if a way of inducing

belief becomes fixed in a community, that might be enough for
it to be said to mean that p, even without considering the
complex intentions which would be needed to communicate
that p, were the regular habit not already extant.
From the second standpoint, linguistic nominalism is like
imagining the psychological complexity of a situation in which
two non-communicants manage something - say to pass each
other on an appropriate side of the road. This would involve,
114 OL"R LA:',GUAGE ..\'.'JD OURSEL\"ES

perhaps, a guess at other's intentions, or ,,this beliefs about

one's own intentions, or even at his beliefs about one's beliefs
about his own intentions. And then the total psychology of this
situation is supposed to be fixed when there is a convention or
regular habit of taking one side of the road rather than another.
Whereas it is far more plausible to say that the habit or conven-
tion supplants the need for such beliefs and such intentions.

2. Openness and Communication

However, there quickly arose another reason for being
interested in Grice's higher-order intentions (intentions that
intentions be recognized). Although this has a place as a sheer
mechanism, being the only hope of people in the one-off predi-
cament, it quickly came to seem more integral to meaning. For
philosophers soon pointed to cases where Grice's conditions
may be met, yet where various kinds of deception and conceal-
ment were involved, with the uHerer intending to bring about a
misapprehension on the part of an audience. These seemed to
destroy some ideal of full openness in communication. And they
also seemed best avoided by demanding more and more high-
order intentions - intentions that other intentions be recog-
nized. So there is a coincidence of reasons for adding higher-
order intentions to an analysis of communication. But the
matter soon became extravagantly complex.
For instance, Strawson described a case in which someone
intends by a certain action to induce in someone the beliefthatp:
He arranges convincing-looking 'evidence' that p, in a place where A
is bound to see it. He does this knowing that A is watching him at
work, but knowing also that A does not know that S knows that A is
watching him at work. He realizes that A will not take the arranged
'evidence' as genuine or natural evidence that p, but realizes, and
indeed intends, that A will take his arranging of it as grounds for
thinking that he, S, intends to induce in A the belief that p. That is, he
intends A to recognize his (first) intention ... 2
A, the audience, is intended to argue that S would not be trying
to arrange apparent evidence that p for him, unless S intended
to get him to believe that p, and, if he trusts Son this matter, he
2
"Intention and Convention in Speech Acts", Philosophical Review, 1964.
 CONVENTIONS, INTENTIONS, THOUGHTS 115

will then come to believe that p. But, Strawson urges, this is not
a case of attempting to communicate that p by an action. The
action does not mean that p.
Strawson believed that what was missing was a further inten-
tion. The speaker or utterer should, ifhe is to mean that p by his
action, intend the audience to recognize his intention to get the
audience to recognize his intention to think that p. This is not
easy to keep tabs on. Furthermore, there is no end to the
possibility of more and more complex kinds of deception. I may
want you to believe that p, but also want you to think that I
want . . . something else. There is then some deception or
exploitation of a mistake involved, and this, perhaps, destroys
some idea of openness in communication. In good cases of
communication we want everything to be above board. It
began to be speculated that we need limitless strings of
intentions to really communicate.
But linear strings of higher-order intentions are not the best
way of ensuring full openness. Such strings can exist all right.
Take a case of Nagel's: 3 in a restaurant I try to attract the
attention of a pretty woman. I want to look at her. I want her to
see that I want to look at her. I may want her to see that I want
her to see that I want to look at her. Or, take a blacker case. You
do something which slightly offends me. But you in turn may be
substantially offended by my offence. When I realize this, I may
become grossly offended-you have no business judging me like
that. And this, to you, may be the last straw ... A common way
in which trivial causes generate desperate hostilities.
But in the restaurant is it really plausible to say that when our
eyes meet in full mutual awareness, I (and she) have an endless
stock of wants? The ideal of full openness is more simply
captured if we just add the want that nothing about my wants
be concealed. This rolls the rest of the linear regress into one
want, which is a good deal more economical. Thus I want to
look at her, want her to look back, and want nothing of my
wants to be unknown to her. I may have more complex wants as
well, but the openness is captured in this last want.
If we cannot shut the regress off with such a clause it becomes
quite unclear whether there could be a "natural kind" which it

-' T. Nagel, Mortal Questions, p. 45.

116 OCR LANGUAGE AND OURSELVES

defines. Take, for example, the seventh or eighth order in 'I

intend that the hearer recognize that I intend that he recognize
that I intend .. .' How could it have mattered to us that just this
condition enter as part of a definition of meaning? Why should
nature have built us to regard this as part of an important
condition, so that were just it absent we should no longer allow
the action as a case of meaning something by an action? Most
people cannot understand the condition (try to think of a case
where just it fails, although the sixth-level intentions are pre-
sent). The verse at the beginning of the chapter probably
stretches most people quite far enough. In fact, the only division
which matters to us is that between cases where an utterer
wants everything to be open, and ones where he is intending
some kind of concealment. 4
We can appreciate what I am proposing like this. Suppose
some bunch of intentions, I, is produced as an account of what it
a
is for an action to mean something in one-off case. The linear
regress gets under way because someone then imagines a case
where the utterer has I, but is indulging some higher-order
concealment or deception. To meet this someone suggests
adding further intentions (e.g. that each of I be known to the
audience). This gives I+. The argument repeats itself, yielding
I++; I+++ ... But soon the bunch loses all contact with any
normal interests or capacities for forming intention, and the
analysis grinds to a halt. Suppose instead that we include in the
initial bunch the intention that all intentions be recognized. Then
anyone possessing this bunch has no opportunity for further
concealment: he has an intention which itself closes offhigher-
order chicanery. So we could offer a simple concept of an open
Gricean AIIB. This would be an action (i) done with the
intention of inducing in an audience the belief that p, (ii) relying
for success upon the audience's recognition of this intention,
and (iii) performed by a speaker who wants all his intentions in
so acting to be recognized.
It is hard to find in the literature, but I suspect that
philosophers have avoided this simple concept in favour of the
linear complexities because they are morbidly afraid of
paradoxes of self-reference. And in condition (iii) there is an
4
To avoid misunderstanding I should point out that in this chapter I intend no
difference between 'wants' and 'intends'; they are stylistic variants.
 CONVENTIONS, INTENTIONS, THOUGHTS 117

element of self-reference: in (iii) there is a want which falls

within its own scope. The totality of wants which the speaker
wants recognized will include that very want. To understand
this, imagine a certain kind of love affair. I want you to know
everything about me. And everything includes, especially, the
fact that I have this want. If you didn't know that about me, you
might suspect me of concealment, and I wouldn't want that.
There is no paradox here, and no regress either. One might
think: well, isn't there an element of vacuousness: imagine
someone whose only intention is that all his intentions get ful-
filled? What does he do? But his problem arises because there is
nothing else in the totality. When there is something else, the
self-respecting want or intention is quite intelligible and, as the
· love affair shows, quite common.
Do open GAIIBs form a natural kind? We can see how an
open Gricean performance would be what we should expect in a
one-off case: it is the honest informant's natural action. But I
am not convinced that it is a good idea to pursue whether such
actions have meaning. For that notion is only at home in systems
of communication, and our intuitions or decisions about
whether it applies are apt to be hazy in one-off cases. Suppose,
for instance, I want to communicate to you that Sam and Janet
are arriving in our restaurant: I do this by whistling the first few
bars of "Some Enchanted Evening", knowing that you know
the joke (Sam and Janet evening ... ) relying upon you under-
standing why I am whistling, and desiring to conceal nothing.
Does my whistling then mean that Sam and Janet are entering
the restaurant or is this not a case of meaning at all? Personally,
I have no strong attachment to either answer. I believe that it is
a bad question to concentrate upon. By whistling I intended to
communicate that they are, and perhaps succeeded. Why go on
to query whether, since that is so, my action meant that they
are? (We shall see further at the end of the section 4 why this is a
bad question.)
The success of an action in a one-off case can become
entrenched, so that a group comes to possess a reliable and
regular way of signalling and transmitting information. If lin-
guistic nominalism is right it is an open GAIIB which becomes
regularized. For an action x to mean that p would be for it to
regularly, or habitually, or conventionally (i) be made with the
118 OL'R LA'.\JGUAGE AND OURSELVES

intention of inducing beliefthatp, (ii) rely upon the audience's

recognition of the intention for success, and (iii) be performed
by a speaker who wants all his intentions in so acting to be
recognized. But I have already suggested that clause (ii), so
important to the one-off case, loses its point when the speaker,
instead of relying upon Grice's mechanism, can rely directly
upon the habits of his audience. Furthermore the third condi-
tion, useful as it is in defining the honest communicant, seems
curiously irrelevant here. For on the face of it a community
might entrench a meaning, communicating that p when they
say or utter x, although habitually and regularly when they do so
they have devious intentions, concealments, and stratagems up
their sleeves, falling short of openness. For example, there is no
regularity or habit or convention that when a young man says
"Of course I'll respect you in the morning" he wants all his
intentions in saying it to be open to view; in fact (as a matter of
statistics) the sentence may be used mostly by people with
various deceptions in hand; yet for all that his utterance means
that he will respect the girl in the morning, and this is what he
has said. We must not confuse questions of morals with ques-
tions oflinguistic convention.
So I prefer a version of the second route, according to which
all that needs to be regular, habitual, or conventional, is that a
particular action be usable to induce the belief that p. The
detour through Grice's mechanism and through openness does
not, in my view, yield anything which belongs to the essence of
linguistic meaning. But to see how such an analysis might
proceed we need first to understand the place of convention.

3. Convention
Convention is a concept which has suffered varying fortunes. In
the positivist high summer, it seemed plausible to explain many
features of our intellectual lives as the outcome of convention:
this was a convenient way of removing the mystery from commit-
ments which seemed to escape reduction to brute experience. 5
But then under the influence of Quine the idea that we could
separate out any particular truths about language as due to
convention became doubtful. Certainly, groups have the habit
5
This is explained further in chapter 5.
 CO'.\JVENTIONS, INTENTIONS, THOUGHTS 119

of taking utterances in particular ways. But calling such habits

conventional seemed to imply a ludicrous historical claim.
Imagine a primitive group sitting around and agreeing to take
various words to refer to various things. To do this they must
have a way of referring both to the new words, and to the things.
So forming a language by agreement presupposes having a
previous language, of equal or greater power, and this puts a
vicious regress in front of any such explanation either of the
origin oflanguage or of its essential nature. It is exactly like the
regress in imagining a historical meeting in which people con-
tract into the recognition of promises: what happens at a meet-
ing can only be contracting into something if the institution of
promising is already current.
Convention was rehabilitated in a beautiful study by David
Lewis. 6 Lewis separated out the claim that a regularity is
conventional in a group, from the claim that it is explained by
historical agreement. The core of his analysis is the notion of a
regularity (such as driving on one particular side of the road, or
meeting at some particular spot, or meaning some definite thing
by a particular word) which benefits each member of a group.
This happens when a group has a problem of co-ordination: it is
best for each of them if they follow the same regularity, and they
each have an interest in co-ordinating themselves so that they
do. But compared with the need to co-ordinate somehow, it
may not matter much, or at all, which regularity they hit upon.
It may not matter where we meet, so long as we meet, or which
side of the road we drive on, so long as we all drive on the same
side. Now any number of things might explain how some
particular regularity, out of a number of possible candidates,
becomes a group's preferred solution to a problem of co-
ordination. It may strike them as obvious, or be handed to them
historically, or be a solution they shuffied towards in a number
of attempts to get their actsJogether. But once a regularity has
become the selected one, it will be conventional if people con-
form to it, and expect others to, and if this mutual expectation is
at least part of the reason why they do so as opposed to doing
something else. Were I to cease to expect other people to drive
on the left (in England) I- would lose my reason for driving on
the left. Were I. to cease to expect you to meet at our usual place,
6
David K. Lewis, Convention: A Philosophical Study.
120 OUR LA'.'iGUAGE AND OURSELVES

I would lose my reason for going there. The same is true of you.
A convention would break down if these expectations began to
be disappointed, and we lost any way of re-establishing co-
ordination. So Lewis's preliminary definition is:
A regularity R in the behaviour of members of a population P
when they are agents in a recurrent situation Sis a conven-
tion if and only if, in any instance of S among members of P,
(I) everyone conforms to R;
( 2) everyone expects everyone else to conform to R; ,
(3) everyone prefers to conform to Ron condition that the
others do, since Sis a co-ordination problem and uniform
conformity to R is a solution to the problem. (Lewis calls
this a co-ordination equilibrium.) 7
This analysis captures the idea that a regularity may be
conventional regardless of how it emerged. What is important is
the explanation of why it continues. The continuation is to be
dependent upon people's preferences, in particular their prefer-
ence for some co-ordination with others, and secondly their
expectation that the others will do a particular thing- drive on
the left, wear certain kinds of clothes, row together if they are
jointly managing a boat, and so on. I shall rephrase Lewis's own
definition, to bring out this place that explanation has:
(CON) A regularity is maintained as a convention
among P if and only if all or most members of P
conform to R, and (at least an important part
of) the explanation of why they do so, as
opposed to conforming to any equally service-
able rival, is that they each expect the others to
do so, and each prefers to do so if the others do.
The definition CON has a number of nice properties. It allows
the conservative idea that conventions are essential and
respectable, for some co-ordination problems (driving on the
same side of the road) simply have to be solved. It allows the
romantic idea that there can be too many conventions in a
society - people may wish to co-ordinate (e.g. over dress)
because they have been brought up so to wish, and things might
have been better if they hadn't. The definition allows us to meet
7
Ibid., p. 42.
 CONVENTIONS, INTENTIONS, THOVGHTS 121

a desideratum first mentioned by Tyler Burge (in criticism ofa

later complication which Lewis introduced), which is that the
status of a regularity as conventional or otherwise should be
something about which discoveries can be made and about
which people can be ignorant or mistaken. 8 People might think,
for example, that English word-order is not conventional, but is
explained by the way in which thoughts arrive. More seriously
there is controversy as to how far the grammar of languages is
conventional, and how far it should be regarded as "wired in",
so that we have no option but to adopt languages which obey it.
There is controversy too over whether some features of our
descriptions of the world- such as the imposition of a particular
geometry on space - are explained by the facts or are due to
conventions of description. The definition explains well what is
at stake in such disputes, and why they might be difficult to
resolve.
However, this very feature caused Lewis to complicate
things. Suppose you and I have the habit of meeting outside the
Town Hall for lunch. I have to travel to get there, but do not
mind because I believe that you work there and cannot very
well travel anywhere else. You believe the same about me: we
are each under a misapprehension. Nevertheless, we are each
there because we expect the other to be there, and prefer to be so
if he is. Is it fair to say that the meeting at the Town Hall is
maintained as a matter of convention? Notice that we would
each deny that it is. If the truth came out we would also be likely
to say that the arrangement was due to mistake rather than
convention. We don't have the convention ofso meeting. Lewis
responds to this kind of case by adding the imposition that the
population P should know the status of the regularity-not only
that, indeed, but know that the others know it as well, and know
that the others know that the others know it ... In fact, there is
the same kind of regress that Grice and his followers met, and
again the basic idea is that everything should be above board.
We can say, when nobody is mistaken or ignorant about the
status of a regularity, or about anyone else's awareness of that
status (and of the fact that everything about it is recognized by
everyone), that we have an overt convention. Lewis restricted
the term conven6on to overt conventions. But I think that
8
'On Knowledge and Convention', Philosophical Review (1975).
122 OCR LANGUAGE AND OURSELVES

Burge is right, and that it is correct to call things like word-order

conventional even if their status is something the population
does not understand. To analyse language we probably need
the concept of a non-overt convention: there will be conventions
in our use of terms which require complicated and even techni-
cal description, which ordinary speakers will not be aware of.
Asserting that a regularity is conventional means showing
that we need to co-ordinate on some feature out of a choice of
equally serviceable ones, and that the reason we adhere to one
is, at least in part, that we expect others to do the same. In
chapter 7 I shall use this analysis to address some recent
proposals in the theory of knowledge; other problems which
benefit include that of how conventions can give rise to norms-
duties to conform, or laws establishing conformity. As far as
language goes, Lewis's work enables us to see how attaching a
particular meaning to a particular sentence or feature of a
sentence can be conventional. This does not, for instance,
involve claiming that anyone ever decided or consciously
arranged that it should be so. Nevertheless, caution is still in
place. For even if we decide that language actually is a system of
regularities which each has a conventional status, this may not
tell us much about whether it has to be so. For instance, it would
not immediately entitle us to rule out a private language
because one individual has no problem of co-ordinating with
himself. And it would leave the possibility of groups whose
linguistic regularities are "wired in", so that they naturally take
certain sounds to have certain meanings; the regularity would
not have the conventional status, but it might be arbitrary to
deny that they have a language and genuinely attribute mean-
ing to the features. In other words, Lewis's work does more to
show how semantic regularities can be conventional than it does
to show that they must be. Yet the former achievement was
remarkable. And, inevitably, it gives rise to difficulties.

4. Force
Which kinds of habits or regularities in our use of signs might be
regarded as conventional if CON is right? The simplest sugges-
tion is that we might take an utterance of a certain sentence, and
regard it as a regularity with the conventional status, in a group,
 CONVENTIONS, INTENTIONS, THOUGHTS 123

that the sentence is uttered only with the intention of inducing

belief that p. But this will not do, because sentences may be
uttered with all sorts of intentions without flouting any lin-
guistic conventions. This is because in two ways speech is too
flexible an instrument for such connections to be rigid. Firstly,
confining ourselves to straightforward assertion, I can properly
assert that p although I know that my audience already believes
that p (I want to associate myself with them, perhaps), or
although I know that they will never come to believe that p (I
want to define my difference from them, perhaps). Secondly, a
sentence can be uttered in other ways or other contexts than
those of assertion. In particular, sentences may be put into
indirect contexts, where their presence indicates no kind of
commitment to their truth. I may assert 'the cat is in the
garden', but I may also say 'if the cat is in the garden you will
not see any birds' or 'when the cat is in the garden it sleeps over
there'. In each of these the subordinate utterance of'the cat is in
the garden' indicates no commitment - nobody is saying that
the cat is in the garden, or trying to induce belief that it is. Yet
the sentence means what it always means. 9
The problem is this. Suppose we aim for an analysis of what it
is for a sentence Sin the mouths of some group to mean or say
that p. We want to profit from Lewis's work, represented by
CON, and see this as a matter ofa conventional regularity. But
then we need to say quite what is the regularity. We cannot
hope for anything as strong as 'regularly and conventionally,
when they utter S they intend to induce belief that p', because of
these two kinds of flexibility. On the other hand, we cannot put
a term like 'means' or 'understands' inside the description of the
regularity. Something such as 'regularly and conventionally,
when members of a group utter S they mean/are understood to
be saying ... that p' leaves untouched the problem of what it is
for a term or sentence to mean something.
The natural way forward is to identify some basic functions of
utterance, and to identify the conventional regularity in terms
of those. Perhaps it is plausible to see language as a develop-
ment of animal signalling systems, whose prime evolutionary
purpose is the transmission of information, enabling one animal
to use a display by another as itself a sign of something. So
9
This phenomenon is of great importance later: see 6.2.
124 OUR LANGUAGE AND OURSELVES

imagine an action habitually taken by a group as a signal that p,

and let us say that someone performing it is displaying that p.
What is displayed need not be a belief of the utterer. An
appropriate utterance will display that a house is on fire or that
the fish are rising. Usually it will also be a sign that the utterer
believes it, because in normal circumstances we can only reli-
ably transmit information we believe to be veridical. But we
shall see in 4.6 that speakers may be taken to display informa-
tion which they do not properly understand. Now among self-
conscious human beings (perhaps unlike animals) an agent will
be able to go through a display for all sorts of reasons: to
transmit information, to mislead, with no hope of transmitting
belief, as a ritual, and so on: if he does so he may be liable to the
penalties attending carelessness or deception. But for the two
reasons mentioned, utterance of a sentence will not count, in
and of itself, as a display. Only the utterance of a sentence with
a definite force, which itself will need some indication, would
count. Thus 'the cat is in the garden', said solemnly, in the right
context, may betray signs that it is intended to be taken as a sign
that the cat is in the garden, and that the utterer believes that it
is. Said with other intonation, or in other contexts, it would not
be so taken.
An utterance usually works as a sign of the truth only via the
fact that it is a sign that the utterer believes that it is the truth,
and is in some kind of position to know, or to have his belief
taken seriously. So it is initially tempting to concentrate upon
the central case, in which an agent is taken to display his own
belief. The kind of conventional regularity which might then
exist would be identified in terms of making an utterance with
signs of basic assertive force. The conventions will be twofold;
firstly, that the signs (such as tone of voice, or presence of the
indicative mood) are signs of this assertive force; secondly, that
the particular utterance, say of a sentence S, correlates with the
belief that p:

It is a conventional regularity in group G that one utters S

with indicators of assertive force, only if one intends to dis-
play beliefthatp.

Do such conventions exist? Someone uttering something will

 CO:\"\'E'.\'TIONS, INTENTIONS, THOUGHTS 125

standardly include an indication of the way his utterance is to

be taken - as a command, or a sincere expression of belief, as a
question, ironically, and so on. But it is not at all clear that such
indicators are standardly indicators of his intention in uttering.
For there is still that first kind of flexibility. I can assert some-
thing, or command it, with all kinds of intentions: I can
command something with no intention of being obeyed, for
example (an authority may issue something with all the con-
ventional signs of command but with no intention of being
obeyed, because it has to, or it's the done thing to do: "students
shall not drink in public after examinations", etc.). These are
not cases in which convention is flouted. Things are properly
said or commanded, with the most diverse variety ofintentions.
There has been a tendency to think of this as the end of the
road for a convention-belief approach to meaning. By contrast,
some writers have urged the "autonomy" of semantics, mean-
ing that the only reliable truth about utterances is that they
have meaning (words refer, or sentences have truth-
conditions), and that there is no way of seeing such facts as
reducible to the conventions and intentions current in a popula-
tion. But this is defeatist: something about a population makes it
true that their sentences and terms have the semantic proper-
ties they do - that they speak one language (in the sense of 1.4)
and not another. And there is still plenty of room for finding
what it is. For even ifwe abandon conventions concerning the
intention with which a kind of performance is to be made, there
are others.
I believe that the basic convention with which a linguistic
group binds itself is that someone who makes an utter.ance with
an appropriate indication of its force may be taken as having
displayed that something is so. This does not mean that it is
regularly sensible to take people to actually believe what they
say. But it means that they become liable to being treated as
having displayed that p. An utterance with an indication of
assertive force is then an act which renders a speaker liable to a
certain kind of consequence: it gives others in the group a right
to regard him as a displayer of a fact; his intentions in so acting
do not themselves influence this social liability. He may have
spoken foolishly or carelessly or dishonestly or with the most
devious and complicated intentions, but still be open to the
126 OUR LANGUAGE AND OURSEL\'ES

same attitude from others. We can frame this kind of conven-

tion this:
It is a conventional regularity in group G that someone
uttering S with an indication of assertive force, may be
regarded as having displayed that p. 10
I shall say that when this is true a group conforms to an Sip
regard-display convention. Such conventions are defeasible;
they can go out of force (for instance, if the context is one of
play-acting, or the utterer is a child), but this is no argument
that they do not normally hold.
There is no principled reason, that I can see, why a group
should not conform to such a convention. 11 If they do, is it right
to see them as speaking a language in which an utterance (or
sentence) S says thatp or means thatp? My answer is: nearly.
To appreciate how nearly, we need to draw in one more strand,
which is the place of system and composition in language.
Meanwhile, there is one comment to make about the difference
between Sip regard-display conventions, and ones which try to
link utterances with intentions, or beliefs.
SIp regard-display conventions are social. The regularity in a
group defines a way in which they may regard someone who
makes an utterance. This is not in itself a defect; on the con-
trary, it avoids the problem over the flexibility of intention, and
accords with many features of the social nature oflanguage, and
particularly of the way in which a group is an authority over
what has been said (see the end of the next section). Neverthe-
less, it opens up a problem. For suppose that with certain
provisions we came to regard a system of Sip regard-display
conventions as sufficient for a language. We would naturally be
eager to see if it is necessary as well to lay claims to having
defined just what it is for a group to have a language in which S
means that p. But this would be presumptious. For there has
been no serious argument that a person could not (systemati-
10
I have here identified utterances by the sentences they involve. This is a simpli-
fication. The rule for sentences containing indexical expressions like 'I', 'here', 'now'
require recovering what was said not from the sentence, plus rules of language, but
from the sentence, plus context, plus rules oflanguage. So S/p regard-display con\'en-
tions are an approximation, but one which does not affect the points of this chapter. See
9.1.
11
And to similar ones for utterances with other forces.
 CO'.\IVENTIONS, INTENTIONS, THOCGHTS 127

cally) voice his thoughts without being a party to any such

system of conventions. Still less has there been any argument
that unless someone is party to such a system he cannot have
thoughts, or thoughts of a certain complexity, at all. In other
words, although membership of a group which abides by such
conventions might be enough for speaking a language, it may also
be more than is required if language use is also to be seen as
somehow itself creating thought. I return to this at the end of
the chapter. Meanwhile, what of system in language?

5. The Place of System: Homeric Struggles

We have moved some way from linguistic nominalism, and
from the cycles of Grice's mechanism and higher-order inten-
tions. But Grice's work has been subject to a fairly persistent
kind of criticism which might embrace this approach too, and
indeed any attempt to understand meaning in terms of some
amalgam of intentions, conventions, or beliefs. The criticism I
have in mind has two connected prongs. The first is that any
such approach ignores or distorts the compositional character
oflanguage. The second is that because of this it fails to give any
theory of how we come to understand a language - fails to
address the question, which is dramatized in the radical
interpreter, of how it is established that a sentence or word
means any particular thing.
The first kind of objection is voiced by Mark Platts. He is
talking about Grice's own theory, but the point, if good, would
apply to any convention-belief theory of meaning:
On Grice's theory, sentence-meaning is defined in terms of the inten-
tions with which the sentence is uttered, along perhaps with the
response standardly secured in an audience by that utterance. Now,
as an account of the meanings of sentences in natural languages this
will not do for a simple reason: the majority of such sentences ... will
never be uttered ... what then can Grice say about these unuttered
sentences?
He goes on to point out that if Grice simply mentions the
hypothetical intentions, with which these sentences would be
uttered, he must admit some constraint upon what these would
be, and continues:
128 OCR LA:\GUAGE AND OURSELVES

Generally the constraint upon the hypothetical intentions with which

a sentence can be uttered, and upon the audience's response to such
an utterance, is precisely the meaning of the sentence ... If this is
correct, the attempt to define the meanings of unuttered sentences in
terms of hypothetical intentions and responses is hopeless: for it
presupposes a prior notion of sentence meaning. 12
This is a forceful expression of the autonomy of semantics.
But it is confused. It is certainly true that although a speaker
will at any time only have understood a certain set of sentences
which he has been exposed to, he is equipped to go on and
understand new ones, and will standardly do so in just the way
other speakers would as well. That is fortunate. But it is not
some mysterious thing, the meaning of the new sentence which
"constrains" the speakers and explains this identity in psychol-
ogy. By analogy, consider numerical codes which interpret
digits and sequences of them making up numerals in the
standard arabic way as far as enormous numerals, and then go
on to diverge (e.g. if there are more than 53 digits in a sequence,
then it refers ton + 7, where n is the number assigned to it in the
ordinary arabic interpretation). Something has brought it
about that I conform to the ordinary interpretation of these
numerals, rather than those they get in these bizarre codes. But
it is not possible to explain that by saying, "well, that's what the
numeral refers to". The point is precisely that it could refer to all
kinds of different numbers. There is no such thing as its refer-
ence, in advance of a common habit of taking it to refer to one
number rather than another. Nothing constrains a group to fall
into one such habit rather than another except their training
and the way they find it natural to take that training. This
results in us, for example, using ANN (see chapter l) rather
than a possible Goodman-style bent competitor, which
diverges at advanced points. The same is true of a fully-fledged
language. It is our habit of taking new English sentences in
definite ways which itself creates the fact that they have definite
meanings. Every theorist recognizes that it is our trainings and
our psychologies which explain those dispositions. And there is
no reason why the dispositions should not themselves be Sip
" :\lark Platts, The Ways qf Jleaning, pp. 89-90. Chomsky charges Lewis with the
same problem: conn·ntions are finite; language runs beyond them (Rules and Representa-
tions, p. 83).
 CONVENTIONS, INTENTIONS, THOUGHTS 129

regard-display habits with the status of conventions, or be

describable in any other way a convention-belief theorist
proposes.
How then should we understand the fact that our conven-
tions apply to unusual or unspoken sentences, even when there
is no common habit of taking them one way or another, since
they are not commonly put forward at all? Obviously, we
perceive in new sentences old words (or subsentential elements,
such as inflexions) in familiar grammatical patterns. Now one
of the merits of a convention-belief approach is that it concen-
trates upon the total act of communication- the whole desire or
belief communicated by a whole sentence. (This need not
necessarily be a grammatical sentence. An infant's single word
may be intended as and be taken to communicate a belief or a
command.) The presence of a word is subsidiary - a word is
something whose presence is a meaning-determining feature of
a sentence. If we like we can illustrate this in the figure of the
radical interpreter. His initial hypothesis is that some native
utterance communicates some whole judgement or command-
that p. From a number of such hypotheses he can start to extract
the features which recur and whose presence seems to be
determining the interpretation to be given to any sentence; once
this is done he can predict the way in which new sentences will
be taken. A convention-belief approach will see the second
stage as one of correlating recurrent features of utterances with
recurrent features of beliefs which natives seem to be displaying.
The presence of the word 'fish' may indicate, for instance, that
the speaker is displaying a belief about fish. Once there is
regularity in the features of sentences which indicate features of
beliefs, there is the possibility of using those features in new
combinations to display new beliefs, and we have the elasticity
oflanguage. 13
Here, then, we have the place of system according to one
13
See Jonathan Bennett, Linguistic Behaviour, §§65-7. Bennett's proposal is mis-
understood by B. Harrison, Introduction to the Philosophy of Language, who thinks that it
fails to protect a proper distinction between a complete utterance, and a meaning-
determing feature ofoI)e. But the only argument that supports this is that the interpre-
·ter may go wrong on occasion. He may mistake one utterance with two words for two
utterances of one word, each conveying a separate message, or vice versa. Of course,
there is the possibility of such mistakes, but increased acquaintance with a language
irons them out. There would be trouble for any approach to language ifit could not say
what the mistake amounts to, but the convention-intention approach does this nicely.
130 OUR LANGUAGE AND OURSELVES

natural view of language. It too has a number of nice conse-

quences. As foreshadowed in chapter l, it suggests a strong
interdependence between syntax and semantics. That is,
identifying the words in a sentence and the recurrent patterns of
composition is secondary to understanding, in the sense that
isolating particular words and particular grammatical
structures in a sentence is itself a matter of charting the features
whereby its meaning is generated. When we say, for instance,
that the sentence 'the police apprehended him' contains neither
the word 'lice' not the word 'hen' we are denying that its
meaning is in any way a function of the difference in beliefs
(making their topic lice or hens, respectively) which these
words conventionally make. This is easier to see ifwe think of
spoken language, where a skill at hearing which words are said
only comes after practice at understanding total utterances,
and learning to hear recurrent features of them. This is why
foreign speech is initially mere noise.

6. Deferential Conventions
If a convention-belief approach thinks of the systematic nature
of language like this, does it suggest any elaboration or modifi-
cation of the idea that meaning is a matter of Sip regard-display
conventions holding in a group? It gives us scope to play the
individual sentence against the system. That is, an individual
sentence might start to be used and regarded as though it
displayed that p, when its composition and the effect of its
elements on other sentences would lead us to predict that it
would be used to say that q. Sometimes this divergence can
become institutionalized, and then we might talk of different
meanings of the terms involved, or metaphorical or other uses of
expressions. Sometimes too the convention or habit in play in a
group seems less to relate to the actual use of a term, and more to
relate to the kind of authority they defer to, in order to determine
what was said. For instance, if I use an unfamiliar term there
may be no direct evidence of a convention that someone using it
is regarded as saying such-and-such; the convention is that
someone using it is taken to say whatever the Oxford Dictionary
or some other authority describes him as saying. If I tell you
that Fred went punting with a quant the only habit ofinterpre-
 CONVENTIONS, INTENTIONS, THOUGHTS 131

tation in my group that I can rely upon is that they use the
dictionary to identify what I can be held to have said (that Fred
went punting with a pole suitable for use in mud).
Such deferential conventions have come into prominence
recently under the heading of the "linguistic division oflabour"
(Putnam's phrase). This means that a speaker may use a word
in substantial ignorance of its real meaning. Nevertheless, he
counts as saying that ... , where the content is given by that
meaning. To use the example mentioned earlier (3.3), ifl make
an assertion by saying the sentence 'Fred has arthritis', I count
as saying something quite specific about Fred and the degen-
eration of his joints. And this is so even ifl myself think vaguely
that arthritis is any old rheumatic feeling, or pain in the leg, or
whatever. As a group we defer to medical authority in defining
what does and does not count as arthritis, and I will be held to
have said whatever it tells me I said - thus, I spoke falsely if
Fred has a vague rheumatism, but does not have the disease
doctors know as arthritis. And this is so regardless of my own
understanding of what I was doing. Using such a term is rather
like picking up one of those government-stamped pieces of
paper that some countries go in for. By writing on it, you might
suddenly find that you had done something which you had little
or no intention of doing, or did not understand yourself to be
doing. 14
Deferential conventions open up interesting possibilities. For
it could be that a group institutionalizes two different ways of
expressing beliefs. You may intend to communicate that p, and
choose one form of words, with the consequence that you speak
truly if one kind of state obtains. But you might choose another
form of words - more official notepaper, as it were- and count
as speaking truly only if some rather different decision proce-
dure (like the experts' agreement) goes your way. The content
of your remark, so judged, and the content of your understand-
ing of it, will then be divergent. This needs careful applica-
tion to problems surrounding our understanding of names
(chapter 9).
Naturally there is room for all the different elements to clash.
14
I intend this analogy quite closely. Sip regarc:1-display conventions have a nonna-
tive element. They define how a group may regard someone. Saying something is like
performing an act with legal consequences.
132 OCR LANGUAGE AND OURSELVES

The intention of the utterer, the meaning of his remark as

predictable from its methods of composition coupled with the
way the group regards the utterances with the same elements,
and the meaning as it is or would be certified by conventionally
deferred-to authority, can all be different. Hence, in particular,
the juristic art of interpreting statutes, whose words so often
belie the intentions of those framing them. Those who framed
the rule prohibiting spilling of blood on the streets of Bologna
were embarrassed to find a surgeon charged with the offence; a
patriotic British statute prohibiting residents in an alien
country from inheriting estates, passed during the Second
World War, was not intended to cover British prisoners-of-war
temporarily residing in Germany. One advantage of the
convention-belief approach to meaning is that it enables us to
appreciate such difficulties as both understandable, and inherent
in the use of a systematic language. That is, the difficulties do
not only arise because of some primitive faith in a fixed "real"
meaning of a term or utterance standing opposed to ordinary
usage of it; they do not only arise if we forget that we are the
masters of our own language. They arise because whatever we
intend to display, we cannot always remember the deposit of
usage which will allow our words to be taken to display some-
thing else. And then too there is scope for different attitudes to
different elements of the jungle (whether etymology is relevant
to identifying what a term should now be taken to mean, for
instance): the stuff of law-courts and correspondence columns.
The philosopher's contribution to all this is to show that we all
stand on the same raft of conventions and higher-order c~nven-
tions; that these count as in force only ultimately because we
have habits of taking each other in definite ways; and that there
is unlikely to be just one "right" way of regarding the construc-
tion of the raft. (The other contribution is that there is no
principled way for meaning to stand fixed when beliefs change,
and we see more of this in the next chapter.)
The natural way to incorporate this into the results of the last
section is to talk of a suitably vague system of conventions:

(RD) A sentence S means that pin the language of group

G ifit is a regularity, or the consequence ofa
system ofregularities, with the status ofa conven-
 CONVENTIONS, INTENTIONS, THOUGHTS 133

tion that one who utters S with basic assertive

force may be regarded as having displayed that p.
A language becomes in effect, a .rystem of SIp regard-display
conventions. Of course, although for simplicity I have con-
centrated upon display of belief, a similar approach would also
cover such things as commands and questions. A sentence
would be a command to bring it about that p if there were the
same type of conventional regularity that someone uttering it
with basic imperative force might be regarded as having com-
manded that p, and so on for other forces.
In his famous inaugural lecture in 1969 Professor Strawson
identified two sides in what he called a Homeric struggle in the
theory of meaning. 15 One side sought to develop a convention-
belief approach to language. The other side believed that this
was unnecessary or even misguided, and hoped instead to
illuminate meaning by developing formal semantics for natural
languages and by showing ( through the person of the radical
interpreter) how a formally described language can be
attributed to a group. Strawson himself hinted that the cause of
hostilities was obscure, although many writers over the last
decade have been eager to join one side and belittle the other.
By now I hope to have deflected most of the popular objections
to a convention-belief approach: these have included the
alleged difficulty over composition and the unuttered sentence,
the unattractive complexities of Grice's own regress, the reali-
zation that meaning sits some distance apart from the inten-
tions of an utterer, the belief that all conventions have to be
overt, and the charge that the approach yields no theory of how
we come to know about the meanings of remarks-no procedure
for the radical understander. The answer to this last charge
emerges from the place of system: to understand a language
demands coming to know what whole utterances mean, and to
be sensitive enough to how the meaning arises to be able to
predict the way new ones should be taken.
In so far as the other side in Strawson's Homeric struggle
have offered any account of what makes it true that utterances
mean what they do, or that words refer and sentences are true or
false in various circumstances, they have relied upon the right
15
P. F. Strawson, 'Meaning and Truth', in his Logico-Linguistic Papers.
134 OCR LANGUAGE AND OURSELVES

method for the radical interpreter. The idea is that if his ideal
way (W*) of attributing a language to a population results in his
regarding them as speaking one in which S means that p, then it
is true that S means that p. The meat of the proposal (which
is otherwise purely programmatic) comes when W* is better
described, and in particular if the relation between S and p
which it certifies can be described in terms which avoid the
notion of meaning. Thus, following Davidson, many theorists
favour a formaja which substitutes truth for meaning:
(W*L) A sentence S means that pin the language of
group G ifW* would result in a radical
interpreter assigning to Gan abstract language
whose systematic semantic description yields a
theorem stating that Sis true if and only ifp. 16
This is just a rough outline, and the approach demands many
subtleties. I discuss it much further in chapter 8. Meanwhile it
is important to point out that as a focus for a Homeric struggle
the two offerings RD and W*L suffer from a major defect. This
is that they are perfectly compatible. Both will be true provided
that W* and the systematic description mentioned in W*L pick
out those Sip pairings of which the regard-display conventions
described in RD hold. And why should they not? Indeed, why
mustn't it be regarded as an essential condition on the ·
adequacy of W* that it should do just this? The upshot is that
although there is much more to say about particular proposals
for filling out each analysis, there is no reason to see them as
essentially in opposition. They offer no invitation to phil-
osophers to form rival schools (philosophers, however, need few
such invitations).

7. Thought Again
It is pleasant to be a member of a group with an established
system ofregard-display conventions; pleasant no doubt to be a
member of a group which a radical interpreter (2.4) regards as
making utterances with definite content (at least ifhe is right).
16
Davidson himself does not favour the idea that any such formula analyses the
notion of meaning ('Radical Interpretation', pp. 324-5). But it clarifies or elucidates or
somehow makes progress in understanding the notion.
 CONVENTIONS, INTENTIONS, THOCGHTS 135

But unless we can get a fuller view than this, it makes it seem as
though mastery of a language is merely a matter of being able to
transmit information, and to prompt action, and to become
receptive to the transmissions of others. And this does not touch
the question of whether there is any more intimate connection
between language and thought - in the limit, whether having
the mastery of a language creates the capacity to think, or at
least to think at a certain degree of complexity.
In chapter 2 we met the difficulty of understanding what it is
to take words, or any other symbolic elements such as images, in
a given way. The vague answer is that it is to use them. So
suppose someone argued: a creature is not thinking of (say) an
absent state of affairs unless it has something present to mind
and takes that thing in a definite way; its so taking the thing is a
matter of how it uses it; e.g. what it is disposed to do in the light
of its other desires or beliefs. For example, a non-linguistic
being might start to think of food. Its doing this would demand
its having some counter - an image, perhaps - present to mind,
and taking that image in a definite way. This might be, for
instance, starting to hunt for food because ofit, or even starting
to salivate or change physiologically because ofit. If this counts
as "using" its mental modification in the right way, then there
seems to be no reason why only elements of a language, with a
conventionally cemented regularity of use, should be given uses
by individual animals.
In fact, our ordinary unphilosophical opinion allows
thoughts to pre-linguistic and non-linguistic creatures. Babies
and animals plan, believe, and anticipate; they may think of
various events and states of affairs, and have a variety of
attitudes or emotions towards what they think of. At least our
best, and often our only, way of describing their doing is to take
up this "intentional stance" towards them - that is, describe
them in terms of their thoughts and purposes. Now it is a
mistake to infer, from the fact that we properly and perhaps
inevitably invest their doings with such significance, the con-
clusion that they actually invest some elements with signifi-
cance, taking it to represent absent states of affairs, for
example. At least, this is not a mistake only if the fact that we
find it appropriate, useful, or even inevitable to describe a thing
as thinking, planning, regretting, etc. is sufficient to show that it
136 OCR LANGUAGE A'.\fD OURSELVES

is true that it is doing these things. And that move requires a

particular theory of truth (a form of pragmatic theory - see
chapter 7). As far as the presence of thought or understanding
goes, the danger in the approach is obvious. Even ifwe decided
that the well-tuned interpreter may legitimately describe an
animal or a system which is doing various things as thinking
about (say) absent states of affairs, he can only do this if he
himself can think about those absent states ofaffairs (in order to
relate the original subject to them). What makes it true that he
can do this? If the only kind of answer is that a second-order
interpreter could legitimately see him as doing so we have a
useless postponement or regress - a disappearance theory of
mind. Problems of mental power are endlessly shuffled back
onto what would or would not be said by an equally or more
powerful mind, itself capable of doing all the thinking which is
puzzling. This is the dog-leg of chapter 2, again.
In recent years this problem has flared in disputes over the
thinking of machines. One school, headed by Daniel Dennett, 17
argues that it is legitimate to take up the intentional stance
which describes such machines as thinking (planning, believ-
ing) where the content of the thoughts, plans, etc. might be 'I
must get my queen out early' etc. Often such machines are only
readily describable as the embodiment of a programme which
itself needs intentional description (it believes that the best way
to achieve a knight and rook ending is ... ). Because a computer
is legitimately described in these terms, it is in fact thinking, and
conclusions can be drawn about other thinkers, such as
ourselves. Another view, taken for instance by John Searle, 18 is
that the transition is illegitimate. For any sufficiently complex
system yielding some output from a given input can be
described, by us, as thinking, ifwe interpret the output (noises,
visual displays, whatever) in various ways, but it will not follow
that it is thinking. Searle's favourite examples include massive
arrays of beer cans arranged so that some input (e.g. striking
some of them in a certain way) brings about some definite
output (e.g. some system of noises or rearrangement or what-
ever). If the array is delicate and complex, we might describe
17
"Intentional Svstems', in Brainstorms.
'" ''.\linds, Brain~ and Programs', Dennett and Hofstadter ( 1981 ). 'The !\lyths ol'the
Computer", New York Review of Books ( 1982).
 CONVENTIONS, INTENTIONS, THOUGHTS 137

the (I asked it what to do with the pawn on queen 4) and

the output (it said that I should sacrifice it) as though the field full
of beer cans is thinking about a chess problem. But that is no
argument that it is doing so. Its elements have 'horizontal'
connections with each other, but no 'vertical' connections to
chess. We impose a semantics, taking its elements to represent
chess situations, bits of the world, states of affairs, etc. But it
takes them in no way at all (compare 2.3 above).
Searle is, in my view, right about this. It leaves it most
unclear, however, what does make the difference. I suspect that
most of us would think that it is not only legitimate in Dennett's
sense, but actually true, that a pre-linguistic baby can think of
absent states of affairs, and similarly perhaps for some animals.
And we ought to remember that there is no coherent story of
how mastery of a language should enable a being to pull off the
trick of conceiving of things which without it he could not. For
example, if a being lacked the capacity to take a present image
or other element to signify a past state of affairs, then this would
seem to be a simple restriction on what he can take words to
mean, and the sort of language he can learn. Learning to talk,
for such a being, would not include learning to talk about the
past. For there is a pronounced pull towards seeing a language
as secondary to our intentional powers, and not itself explaining
them. Of course, if that is right, it is no objection to the
convention-belief description of language that it fails to show
how language can explain such capacities.
This is one side of the argument. But there is another.
Perhaps we have been concentrating upon too big a quarry- no
less than a total understanding of our powers of thought. If we
take local examples, the interplay between learning to under-
stand and think in various ways, and learning a language, is
inextricably close. There is no distinction between mastering
physics, and mastering some language in which to express
physical laws and concepts. There is no distinction between
learning to discriminate finely between various modes of taste,
colour, or feeling, and learning a vocabulary with which to
express the differences. People who can understand mathemat-
ics can use the notations; people who cannot say (if only to
themselves) what the answers are do not know what they are. A
man may be a mute inglorious Milton, or Einstein,just because
138 OCR LANGUAGE AND OURSELVES

he never gets around to writing it down or saying what he

knows; but he cannot be one, if, were he to try to express what
he knows, nothing would happen.
A useful example might be the appreciation of number.
Perhaps it is plausible to suggest that a creature with no capac-
ity to use a numeral notation simply could not believe or
understand that there are 25 matches on the table (as opposed
to 26 or 17). The notation is necessary for the thought. Why
might this be? Well, counting is a procedure which requires some
kind of tally - a process of ticking off something against each
element of the set counted. And we might suggest that you
cannot appreciate the result of the procedure unless you can
appreciate the procedure itself by operating it, or being able to
operate some equivalent system. So the argument is that under-
standing the number of the set requires understanding the kind
of procedure which certifies it; understanding this requires
ability to operate or appreciate the operation of counting proce-
dures; doing that requires being able to take some action (the
saying of' 5', making another notch in a stick, tying a knot in a
rope, etc.) as having a certain significance (it is made to locate
how many things we have found so far). And that is, in essence,
to be using a numeral notation.
If this sketchy argument were generalized, it might suggest
an interdependence of the following kind between language and
thought. To think and understand that something is the case
often involves an awareness of procedures for certifying that it is
the case, or more generally for placing its truth in a context of
other states of affairs. But these procedures and connections
need things to mark where we are - what has been done or
remains to be done, what has been established and what has
not. It is as though there is no thought without the possibility of
movement of thought, and movement of thought needs not so
much a linguistic vehicle (inviting the question: why not some
other sort?) as a structured map, telling us where we are. But
even if this is right, there is no reason to say that the counters
telling us where we are must be socially shared elements of a
system of regard-display conventions. A man might invent his
own way of counting, and use it in complete isolation. And in
fact there are cases of people with severe handicaps who cannot
hear or use elements of a public language, but who are dis-
 CONVENTIONS, INTENTIONS, THOUGHTS 139

covered to have their own private way of expressing themselves.

The relevance of language to thought may be tested by
considering the possibility of a totally unstructured language.
Suppose a long list of unstructured sounds, and a long list of
beliefs or thoughts of some complexity- that Fred is asleep, that
there are 25 matches on the table ... If a group could take the
one as a.first language to express the others, then it will be very
tempting to suppose that the competence with the language is
simply irrelevant to the mental power enabling them to do the
thinking. J3ut is this a possibility? A radical interpreter is
unlikely to certify that this is the language of any group, since
his favourite device of looking for the systematic effects of
subsentential elements draws a blank. He may find, for instance,
that a certain noise is made and apparently approved ofin some
circumstance. But it is a long step from that to supposing that
the noise expressed the thought that this circumstance obtains.
For circumstances exhibit many things with many features, and
it needs work to find out which ones matter to the speakers (is it
Fred being asleep, or Fred not being on the farm, or Fred
snoring ... which interests them?) The difficulty is that since
their utterances have no structure, there is nothing to work on -
no feature which can reliably be correlated with the presence of
Fred or the feature that something is asleep.
Still, perhaps this is just an interpreter's problem. Can we
find a principled reason why speakers, if they can have pre-
linguistic beliefs of the right type, could not generate habits and
conventions whereby these atomistic sounds display them? It is
tempting to suppose that a learner would face the same problem
of selection as the interpreter, so that he would never have
reason to suppose a given sound to be displaying the belief that
p as opposed to any of an indefinite number of possible sur-
rounding beliefs. But perhaps the population share the same
feature space: they naturally carve things in a given way, notice
and wish to communicate one aspect of things rather than
another. If a particular situation naturally causes a member of
this group to form a particular belief and to wish to display it
and to interpret others as so wishing, then an utterance made
with apparent assertive force will be taken by the learner to be
displaying that belief and no other.
However, such a language, if not an a priori impossibility, is at
!-l:O OCR LA;\JGUAGE AND OURSELVES

least markedly unstable, in the sense that the capacities the

population must have, in order to use it as suggested, are the
very ones which would naturally bring it about that they do
something better. For they must be capable of selecting particu-
lar features of situations - say the fact that there are 25 things in
a group, or that something is asleep - for attention and com-
ment. But selecting such a feature means being capable ofrecog-
nizing when it recurs, and recognizing the similarity between
two different situations in which it recurs (Fred being asleep
and John being asleep; 25 matches or 25 children). So the
speakers are, as it were, already equipped to produce a feature
of speech to correlate with the recurrent feature which they can
notice. It would be inexplicable if they did not do this- so much
so that we would naturally deny that a system of sounds which
had not done it actually expressed beliefs referring to that very
feature. If they think in terms of particular things and features,
they would naturally need elements of utterances to indicate
which things and which features make up their topic.
So a genuinely expressive but unstructured language is
intrinsically unstable. Does this settle whether thoughts and
beliefs can be understood in the absence of a structured means
of expression? Certainly I see no argument that the means must
be a fully-fledged language, in anything like the sense of a
conventionally fixed system of regard-display habits. And
unless we can broaden the tentative suggestion I sketched, over
what might be necessary to appreciate the truths arrived at by
counting, the assertion that thoughts require language seems to
be a promise rather than a result. The way forward would
require a convincing description of ways in which acquiring a
language creates understanding, and this philosophers have not
so far provided.
There is of course very much more to be said about this. But
provided the reader appreciates the delicacy of the issues, it
may be left in order to tum to the third corner of the original
triangle: the world, and the way our language, and we ourselves
in thought, relate to it.
 CONVENTIONS, INTENTIONS, THOUGHTS 141

Notes to Chapter 4
4.1 " ... the complexities which abound in the literature ... "
A basic set ofreadings would include:
H. P. Grice, 'Meaning', Philosophical Review ( 195 7).
- 'Utterer's Meaning and Intentions', Philosophical Review ( 1969).
P. Ziff, 'On H.P. Grice's Account of Meaning', Ana(ysis ( 1967).
J. Searle, 'Meaning and Speech Acts', Philosophical Review (1962).
- Speech Acts, ch. 2.
P. F. Strawson, 'Intention and Convention in Speech Acts',
Philosophical Review ( 1964).
S. Schiffer, Meaning.
J. F. Bennett, Linguistic Behaviour.
But my experience is that the literature is unnecessarily difficult, for
the reasons stated in the text.

4.3 " ... the autonomy of semantics ... "

This is another of those issues where strong passions soon conceal the
issues. Indeed, the autonomy of meaning has become one of the great
issues in structuralist and post-structuralist theories of literary criti-
cism. Roughly, orthodox literary critics are supposed to refer back to
the intentions of an author, as revealed by their writings; noticing that
language is a conventional system, their successors urge that the text
itself is an object of study, with no reference at all to the intentions or
lack of them which the author may have had. This is not a new idea in
literary criticism, but it is a fashionable one. It is expressed in the view
that writing (where there is no "authorial presence") is as fundamen-
tal a phenomenon in the use of language as is speech - which most
philosophers of language have taken as primary. But the debate has
nothing essential to do with auditory or written modes of communica-
tion, nor with the relative permanence of signal involved in writing
versus the transitory signal of speech. It is really a debate about the
authority of the speaker versus the independence which his signs
have, given their place in public and conventional systems of com-
munication. Unfortunately the issue seems to be taken, by writers
following Jacques Derrida, as a licence to take the text in a purely
syntactic way, and neglect the conventionally fixed semantic proper-
ties of the words - the properties which, whatever else is true of the
author and his intentions are ones which explain why he put them
down. A rather bemusing debate on this is:
J. Derrida, 'Signature, Event, Context', G(yph (1977).
J. Searle, 'Reiterating the Differences', G(yph ( 1977).
142 OUR LANGUAGE AND OURSELVES

Rather more austere treatments of force and convention include:

M. Davies, Meaning, Quantification, Necessity, ch. 1.
C. Peacocke, 'Truth Definitions and Actual Languages', in Evans
and McDowell (1975).
J. McDowell, 'Meaning, Communication, and Knowledge', in van
Straaten ( 1980).
 PART II

LANGUAGE AND E WORLD

 CHAPTER 5

Realism and Variations

The one discovers objects as they really stand in nature, without

addition or diminution: the other has a productive faculty, and gilding
or staining all natural objects with the colours, borrowed from internal
sentiment, raises in a manner a new creation.
David Hume, Appendix I of the Enquiry Concerning
the Principles ef Morals.

1. Oppositions to Realism
Hume's description of the different offices of reason and taste sets
a challenge. How do we tell when we are discovering objects "as
they really stand in nature", and when we are doing some other
thing, such as projecting onto them our own subjective senti-
ments? Which side of this divide do we fall on when we describe
objects as good or bad, nice or nasty, hot or cold, red or blue,
square or round? Echoing Hume the scientist Heinrich Hertz
says that "the rigour of science requires that we distinguish well
the undraped figure of nature itselffrom the gay-coloured vesture
with which we clothe it at our pleasure" . 1 How do we know
where to draw this distinction, or what counts as an argument for
putting a given saying on one side or the other? This is the issue
between realists and their opponents. It takes a bewildering
variety of forms, for philosophers have seen it very differently.
Realists are contrasted with a variety of alleged opponents:
reductionists, idealists, instrumentalists, pragmatists, verifica-
tionists, internalists, neo-Wittgensteinian neutralists, and no
doubt others. They also form obscure alliances with, and hos-
tilities towards, various views about truth: correspondence,
coherence, pragmatic, redundancy, semantic, theories. To
swim at all in this swirl of cross-currents we need a better lifeline
than any which these mysterious labels provide.
My strategy will be to follow through a variety of possible
' Quoted in van Fraassen, The Scientific Image, p. 6.
146 LA:'IIGUAGE AND THE WORLD

attitudes to some particular area of commitme1 its. This will give

us a sense of the options locally. But by com;ng to appreciate
local issues, arising, say, in the philosophy of value, or of mathe-
matics, we can work our way into more general problems of
realism and truth, and get a sense of why the global issues fall out
as they do. I prefer this to a top-down strategy, which would
approach highly general issues of the nature of truth first, and
then apply the results to particular cases, because the general
issues are more or less unintelligible unless given particular
applications.
When discussing meaning in 2.1 I introduced the possibility
of a perspective from which facts about meaning appeared
utterly mysterious - a perspective from which the world could
not contain any such facts. Similar doubts pepper the whole of
philosophy. Hume could not see what kind of thing a causal
connection between distinct events could be; mathematical,
moral, and aesthetic facts seem suspicious to many people;
semantic, psychological, conditional facts invite scepticism, and
so on. Once such doubts are felt- motivated in whatever way-
a number of attitudes are possible. We might reject the whole
area, advocating that people no longer think or speak in the
terms which seem problematic. Or, we might seek to give a
reductive analysis of it, advocating that the problematic commit-
ments be put in other terms, and claiming that when this is done
the problems disappear. We can try to see the commitments not
as beliefs with truth-conditions but as expressions of other sorts.
We can query whether the commitments are mind-dependent -
not really describing a mind-independent reality at all, but as in
some sense creating the reality they describe. And at each
choice-point there will be a jumble of issues and of suggestions
about what the debate hinges upon. In particular there can be
the attitude which I christen quietism or dismissive neutralism,
which urges that at some particular point the debate is not a
real one, and that we are only offered, for instance, metaphors
and images from which we can profit as we please. Quietism is a
relative newcomer to the philosophical world, owing much ofits
inspiration to the positivist mistrust of metaphysics, and to the
belief of the later Wittgenstein that such problems required
therapy rather than solution.
Views about truth will be particularly relevant where we
 RE ..\LIS:\1 AND VARIATIONS 147

discuss whether some commitment is best regarded as a belief

with a truth-condition, or in some other light. For this contrast
seems to hinge upon our view of what it is for a commitment to
have a truth-condition, and it is here that the various lights in
which truth is put - correspondence, coherence, and so on -
affect the issue. Thus, to take a simple example, moral commit-
ments are often thought of as not really beliefs, but as more like
attitudes, emotions, or prescriptions: this contrast in turn may
look very different ifwe think of beliefs in pragmatic or instru-
mental terms rather than in terms of correspondence with facts.
Putting together the positions I have suggested we then get
the map illustrated in Fig. 4. But, as I have said, the issue need
not be clear-cut at any of these choice-points: there will be
different suggestions as to what the decision hinges upon, and,
from the quietist, the general disenchantment with discussing
the issue at all. Quietism is currently expressed by denials that
there is a "god's-eye view" or an "external" or "Archimedean"
point from which we can discover whether some commitment
is, as it were, describing the undraped figure of nature or
imposing clothing. Perhaps we either accept the sayings in a
given area, and "from that standpoint" suppose that they
describe reality, or we do not accept them, and of course think
that they do not. In other words, an overall quietist is only
interested in the first choice-point, and ignores the rest. We
shall see that this is unduly short-winded. I shall now introduce
the various options at greater length.

"
Area of commitments

/
Accept area Reject area

In its own terms Reduce it

• i.e. according to Attribute genuine Treat them as offerings

truth-conditions· with some other point
Correspondence or and other virtues
Coherence or to the commitments
Semantic or
Redundancy or Regard these truth- Regard them as mind-
Pragmatic Theory of conditions as mind- dependent
Truth independent

Fig. 4
148 LA'.\'GUAGE AND THE WORLD

2. Rejection
We might want to reject a given commitment within an area
without rejecting the whole area. We do not accept that 2 + 2 =
5, but we do, of course, accept many beliefs about numbers.
Rejecting a whole kind of commitments is more sweeping.Until
relatively recently it seemed to raise few problems. If a remark
in some area (e.g. a moral remark) is meaningful it must have a
definite content. If it does, either things are as it says they are,
and it is true, or they are not, and it is false. Ifit does not have a
definite content it is meaningless, or at least vague and defective
on that score. How is there any room for subtlety here?
The only subtleties arise because the neat division between
attributing a definite, false, content to a remark and rejecting it
in some other way is, in practice, impossible to rely upon. I can
introduce the problems this causes with a simple example. 2
Consider the use of the term 'Kraut' as a term of contempt for
Germans. If someone describes Franz as a Kraut, I want to
reject his remark. Do I say it is meaningless? Hardly. I know what
I am being asked to think, and it is because I know this that I
find the remark offensive. Is the remark false? We usually accept
an equivalence between 'it is false that p' and 'it is true that
not-p'. But I do not want to say that it is true that Franz is not a
Kraut. That is the remark made by someone who has the
contemptuous attitude towards Germans, but believes that ·
Franz is not a German. So I do not want to say that on two
counts. So should I say that the remark is true or even half-true?
Uncomfortable, again.
One theory would be that the remark is a conjunction: 'Franz
is a German and on that account he is a fit object of derision.' A
conjunction is false if either part is. Since I regard the second
conjunct as false I should maintain that the whole remark is
false. In this spirit we would say things like 'there are no
Krauts', or 'nobody is a Kraut'. This is a way of construing the
remark, and thence of disowning the attitude. But a different
option is to regard the remark as true, but to disown the
phrasing. A parallel would be 'Franz is a German' said with a
derisive intonation on the last word. Here we suppose that what
2
In the nature of the case, the following example employs offensive terms with
which I do not associate myself.
 REALIS'.\I AND VARIATIONS 149

was strictly said was true but reject the overtone. The convention
would be that you only put beliefs about Germans using that
overtone (or the derisory word) if you have the contemptuous
attitude. If this is the right account then it is in fact true that
there are lots of Krauts (although I would not put it that way
myself).
Each proposal is "semantically coherent": in other words, a
population could properly speak a language in which the first
analysis is right, and the remark is false, or L 2 , in which the
second is right, and the remark is rejected, although what it
strictly says is true. Is there bound to be a fact determining
whether English operates like L 1 or L/ If not, then there is no
uniquely right way of expressing rejection of the utterance .
. Since we are left with a slight sense of discomfort with either of
the sharp options, perhaps there is this indeterminacy. The
vocabulary belongs to people who accept a certain attitude -
that being a German is enough to make someone a fit object of
derision. Rejecting the attitude we reject the vocabulary. But if
the way the attitude is expressed is indeterminate, it will also be
indeterminate whether remarks made by people using it are
true or false. In other words, people using a certain vocabulary
tend to have clusters of belief, some true and some not, or
clusters of attitudes, some acceptable and some not, and ten-
dencies to favour certain inferences, some reliable and some
not. But there may be no conventions determining how many of
these habits we are endorsing by accepting some remark as
true, nor how many we are rejecting by regarding it as false. The
overall rejected theory may have distributed its content in no
very secure way over the various sentences involved. It will then
be unsettled how to express the rejection.
The problem arises in more serious areas. In writing the
history of science (or any other body of thought) we want to
express what was right and wrong about particular doctrines.
But the vocabulary people had may offer no particular way of
doing this. Terms will have been used in the context of clusters
of beliefs, attitudes, habits of reasoning and inference. Some of
these we will accept, and others we will reject. But how could
the people have formed- conventions determining that some
particular thing is the right thing to say, in the face of unforeseen
disruptions of beliefs, attitudes, and habits? There is no cause to
150 LANGUAGE AND THE WORLD

worry whether we are speaking L 1 or L 2 just so long as the views

and attitudes remain intact. But when they are disrupted, then
from the later standpoint it may be indeterminate which sayings,
expressed in the earlier vocabulary, can be thought of as true,
and which as false. For instance, Newtonian mass was a concept
central to classical physical theory, and integral to the enormous
success and near-truth (as we now see it) of that theory. Unfortu-
nately amongst the things believed in that theory is that the
mass of an object is quite independent of its velocity - its rest
mass and moving mass are always identical. When this thesis is
abandoned, and we consider other sayings of classical physics,
such as 'force applied is proportional to mass times change in
velocity', it is simplistic to regard them either as definitely true,
or as definitely false. They were part of a theory or overall web of
belief which involved false views; how much of the falsity is to be
read into an individual saying is often indeterminate. The
reason we can put up with this indeterminacy is not that in any
sense we cannot express what the old theory held. We can
express it very easily, in our own terms, just as we can say what
people think when they describe Germans as Krauts. We do not
have to share the views and attitudes in order to know what they
are. What we cannot easily or determinately do is describe
things in their terms (are there or are there not Krauts?) because
there is no unique way in which the rejected views can be pinned
onto individual sayings, and insulated from others. We cannot
neatly partition truth and falsity, sentence by sentence, across
their individual sayings.
The point deserves attention whenever we consider how to
describe our distance from other theories, or webs of views and
attitudes. Was there such a stuff as phlogiston? Can a Marxist
believe in human rights? Using our terms we can distinguish
what was true and false in the beliefs held by phlogiston theorists.
Using his scheme of values the Marxist can say what is accept-
able and what not so in the (bourgeois) conception of rights.
But it is quite hopeless to try to express that distance by
allowing that individual theses of the rejected theories express
uncontaminated truths, or uncontaminated falsities. That
requires using the old vocabulary (there are or are not Krauts; are
or are not human rights), and this is something which, from the
opposed point of view, it will be undesirable to do.
 REALIS'.\1 AND VARIATIONS 151

There is a temperament which would dismiss this as an

unfortunate product of the vagueness of ordinary thought, con-
trasted with some ideal language with the precision to give
every remark definite, indubitable conditions of truth and falsity.
If we met this ideal, then whatever the changes in belief and
attitude, we could always look back and attribute definite truth
or falsity to each individual saying in the old terms. But how
could such an ideal language be speakable? The significance of
remarks has to be one which we are capable of giving them.
That capacity is a function ofour beliefs, attitudes, of the things
we have been exposed to and the habits of thought, reasoning,
and inference in which we indulge. We cannot possibly foresee
how these may be improved, or what strange possibilities will
.disrupt them. When they are changed, previous usages will pull
in different ways, and there is no reason to expect previous
sentences to express individual propositions, isolatable as truths
or falsities in the new scheme of thought. To paraphrase Witt-
genstein, when we start to abandon a way of thought, the lights
do not go out one by one, but darkness falls gradually over the
whole.

3. Reduction
A different attitude to a theory, which yet shades into outright
rejection of it, is that its theses can be accepted, but their
content can be expressed in other ways, using a different kind of
vocabulary. Thus we might find some particular set of terms
awkward or puzzling in various ways; the demand becomes
that we can give an "account" - an analysis, or reduction, or
reinterpretation - of the things said using them. The analysis
supposedly reveals the true or proper content of remarks in the
area. For example, if moral commitments appear particularly
puzzling, attempts might be made to reinterpret them in some
other terms: perhaps 'Xis something which ought to be done'
means the same as 'X will produce more happiness than any
alternative'. The moral vocabulary would then turn out to be
just a different way of putting ordinary, natural, or psychological
truths. In that case it would import no particular problems ofits
own - such as ones of what kind of thing moral facts can be, of
how we can know about them, or how they relate to underlying
152 LANGUAGE AND THE .WORLD

natural facts, and so on. If the moral vocabulary demanded a

distinct type of moral fact these questions would arise, but
since, according to the analysis, it does not, they resolve them-
selves. We know about what ought to be done in whatever way
we know about the creation of happiness, since it is a proposition
about this which gives the real content of the moral statement.
Famous positions in philosophy which are based on the claim
that analyses and reductions are possible include:

Phenomenalism: the analysis of propositions about

external reality into ones about actual
and possible experiences;
Positivism: the analysis of propositions about
theoretical entities into ones about
regularities in experience;
Behaviourism: the analysis of propositions about
mental states into ones about disposi-
tions to behaviour;
Logicism: the analysis of propositions about
numbers into truths of (elementary)
logic, thought of as describing no
especial, abstract, subject-matter;
Naturalism: the position in ethics already described.

And there are many other local analyses - of propositions about

causation into ones merely describing successions of events,.of
ones about the meanings of terms into ones about peoples'
intentions when they use the terms, and so on. Indeed, so many
philosophical problems centre around the possibility and desira-
bility of providing analyses of various kinds that in the early
part of this century the philosopher's role became identified
with this (analytical philosophy). Naturally in an empiricist and
verificationist time attempts are made to reveal in terms especi-
ally close to experience the true content of statements which
appear to concern esoteric subjects. The idea behind the veri-
fication principle is that the real meaning of any assertion can
be found by analysing the difference in sense-experience which
its truth would make.
The motivation for reductive analyses is based on a contrast.
The commitments expressed in some original vocabulary (the
 REALISM AND VARIATIO'.\iS 153

A-vocabulary, or A-commitments) must be felt to introduce

some apparent puzzle, either of meaning, or of epistemology, or
of metaphysics, not introduced by statements made in the
analysing, B-vocabulary. An analysis represents a piece of im-
perialism on behalf of the concepts expressed in B-terms, and
will be motivated if the A-statements are in some particular way
puzzling, or if there is a background view that everything which
it is possible to say truly can be said in B-terms. Ifonly we did
not have the additional A-commitments, the puzzles would go:
the solution is to propose that in the relevant sense we do not
have them - they are not additional, because their true content
can be revealed entirely in B-terms. Examples on a local scale,
suggesting the kind ofrelief which reductive analysis can bring,
are quite persuasive. Suppose I tell you that Henry's prestige is
enormous. Suppose that you are attached to an ontological
doctrine - roughly that everything which exists has a place in
space and time, and has scientifically measurable properties of
weight, charge, velocity, etc. Then Henry's prestige seems an
odd kind of object: you cannot put it into a bucket or weigh it or
measure it-what kind of thing is it?Your problem is removed by
analysing the original A-remark so as to remove reference to
this mysterious thing: it means nothing different from this
B-remark: other people admire Henry enormously. If your
world-view allows for this kind of fact, then there is nothing
further to jib at in talking of prestige. Notice of course that other
problems might still remain - problems with the kind of state-
ment to which the original was reduced. Your world-view
might make it difficult to accommodate people admiring one
another; in which case you have to continue the analysis until
you find statements which you can put up with. B-statements
might need reduction to C ... W which is respectable.
In itself the claim that one statement has the same content as
another is quite symmetrical: this suggests that the problems
affecting the first must also affect the second. But, as the
example about prestige shows, this is not inevitable. The
problem might arise only because of a feature of the original
which is not shared by the analysing statement. In this example
the feature was the apparent reference to a non-physical thing.
Once it is accepted that this feature is not essential, since the
same content can be expressed without it, the problem resolves
154 L\'.\'GUAGE A'.'ID THE \\'ORLD

itself. Usually, however, it is not so obvious whether the

analysis does more than draw other statements into the
problematic class. For instance, a counteifactual analysis of
causal remarks is quite attractive:. to say that X causes Y is,
perhaps, to say something like 'if X had not happened, and
other things had been the same, Y would not have happened'.
But even if this equation works, its significance is debatable.
What is the contrast which motivates such a reduction? What
problem of meaning, epistemology, or metaphysics, does the
initial statement have, which is removed if we can see its
content as the same as that of the second? Or does the
counterfactual come to seem just as problematic, in terms of
what makes it true or what shows it to be true, as the original
causal statement? A prominent example of this difficulty is the
equation of arithmetical with set-theoretical remarks.
Propositions about numbers may be reduced to propositions
about sets in various ways. But even if one of these is taken to
really give us the meaning of the original arithmetical
statements, the question is whether sets then take on a good
deal of the mystery of numbers, posing substantially the same
problems of existence, mind-independence, and of knowledge
and logic, which originally make arithmetical statements into
desirable objects of analysis.
So for a successful reduction we need an A-discourse which
poses some problem, a B-vocabulary used to say things which
do not pose this problem, and an equation between any
propositions expressed in the first and some set of judgements
expressed in the second. Now, however, we can see that there
can be considerable tension between the disappearance of the
problem, and the equation of meaning. If the original problem
is a substantive or persistent one, then the fact that A-
statements are subject to it by itselfsuggests that we take them to
mean something different from the B-statements, so that the
equation fails. This is a kind of Catch 22: if the reduction is
really well motivated, then it cannot be true. For example, we
have given our moral vocabulary a meaning which results in
moral statements posing characteristic problems of proof or
verification, of truth and objectivity. For this very reason, we
might urge, they cannot be identical in meaning with other
statements which do not pose these problems. Moralizing is a
 REALISM AND VARIATIONS 155

specific activity, so how could these other statements give us a

non-moral way of doing it? The reply has to be that the
problems prompting the reductions are in some sense only
apparent - in other words, they arise only because we
misapprehend the true content of A-statements. When these
are seen for what they are - in B-terms - the temptation to
puzzle over the problems disappears (as with prestige).
However, this reply in turn raises a query: what right has
anyone to separate out the reduced content as the only true
content of the original, if our practices, including our doubts,
problems of verification and proof, and so on, all suggest that we
take those statements to have a further or different content?
A sharp formulation of this problem comes with the classical
"paradox of analysis". Consider as an example the proposal
that 'X ought to be done' means the same as 'X would bring
about more happiness than any alternative'. Now it is quite
intelligible to describe someone as wondering, or doubting, or
thinking, whether all and only things which ought to be done
would bring about more happiness than any alternative.
Indeed, faced with the proposal the first thing to do is exactly to
try the thought-experiment of finding a counterexample - a
thing which satisfies the one concept, but not the other. But
how is this possible, if the analysis is correct? If the two notions
just mean the same, then it would seem that the doubt or
wonderment can have no content - it would be the same as
doubting whether all and only things which ought to be done
ought to be done, or whether all and only things which bring
about more happiness than any alternative bring about more
happiness than any alternative. The paradox (which has other
forms) gives rise to quite difficult technical problems, some of
which I discuss under the general heading of difficulties over
substitution of synonyms for one another, in 9.5. Here it
illustrates the general point, that frequently one set of terms is
given a sufficiently different role by us, as witnessed by our
beliefs, doubts, queries, etc., for the claim that it has the same
role as some reducing set of terms to seem implausible from the
beginning. A common aspect of this is an asymmetry in the way
we regard A-truths and B-truths, in respect of explanation and
evidence. Frequently the B-truths (e.g. about sense experience,
or behaviour, or non-moral states of affairs) are regarded as
 156 LA'.\rGUAGE AND THE WORLD

evidence for the A's but not vice versa. This asymmetry is
incompatible with the view that the content of the A-statements
is identical with that of some suitable set of B-statements. Now
reductionism often has just this kind of asymmetry as its target:
it is sceptical whether there is a legitimate inference from B-
truths to a different set of A-truths. This is why it prefers to
reduce the A's down, meaning that the inference is no longer
vulnerable. It leads nowhere different, outside the B-range. But
since this revises the natural belief, it hardly gives the meaning of
the original concepts, but suggests substitutes.
Because of this problem, there is a tendency for reductionist
programmes to take on a revisionist air. It becomes tempting to
shelve the question of whether the reductions mean the same as
the original statements. Suppose our doubts, puzzles, or other
practices illustrate a way in which we take A-statements as
having more content than B-statements. Still, perhaps the B-
statements exhaust the legitimate content of the A's. In that
case the extra can be dismissed as the product of muddle, of
"prehistoric metaphysics" and failure to see the A-vocabulary's
only legitimate role. The extra that we add is to be pruned
away, in a programme of reconstruction. Reductionists from
Berkeley right to Russell have tended to uncertainty over the
relation of their analyses to the original discourse. One part of
them wants to accept the original discourse - the A-statements -
because, of course, they have identical content with quite legiti-
mate B-statements. But another part wants to voice suspicion of
the A-statements, because along with the pure content, there is
the intruding illegitimate element which disguises it. Thus
Berkeley presents himself as siding in all things with the mob:
his analysis of the world as a community of spirits and ideas
allows us to think that there exist tables, chairs etc. However, in
another mood he will insist that it is a vulgar error to suppose, for
example that anything is ever both touched and seen (since the
ideas of touch and those of sight are quite different from each
other, and his idealism disallows any common object). Yet chairs
. and tables are ordinarily thought to be both seen and touched.
So Berkeley havers over whether the reduced, legitimate content
exhausts the actual meaning we can give to statements, or
whether in the actual meaning there is additional, false material,
arising because of our misunderstanding of the idealist truth,
 REALISM AND VARIATIONS 157

and resulting in statt;!Tlents which ought to be rejected. Russell

too presents an uncertain attitude towards statements of the
A-kind: a phenomenalist analysis means, on his view, that
chairs and tables become 'logical fictions', but whether this
allows us to say that there really are tables and chairs (surely
something the mob, the vulgar, want to say) is not certain. 3 The
one line would be that we can say this (meaning by it something
susceptible of expression in terms of actual or possible sense-
experience), the other line would be that we cannot, because
it involves views which are themselves impugned by the phe-
nomenalist reduction.
Not that it is too discreditable to sit on this particular fence.
From the reductionist point of view the legitimate content is
given by the B-statements. Whether the A-statements have
a meaning which supplies them with more, but illegitimate,
content, may be left relatively indeterminate. Suppose, for
example, that false views about the relation between A-sayings
and B-states of affairs have prevailed - prehistoric metaphysics
which supply us with images and metaphors, ways of taking
A-statements which serve to separate them from B's. Then it
may be indeterminate to what extent we accept these particular
views when we accept A-statements. If we do, then the B-
statements should be seen as replacements. Ifwe do not, they
may be analyses revealing the real content. A phenomenalist
may happily admit that ordinary people are so infected by false
views about the independence of objects from our sense-
experience, that there is something of a shock in his analyses.
And, according to me, given that the shock is one of realizing
that some of our thoughts about objects and ourselves were
wrong; he need not try to settle whether these thoughts were
part of the content of our ordinary remarks, so that the proposal
is to replace these, or whether on the other hand these thoughts
were mere accidental sideshows, not part of what we ever said,
so that the full content of this is revealed in the reducing
B-s ta temen ts.
A nice example comes from moral theory. Suppose people are
naturally drawn to the realist options when they think about
moral remarks. Then the realist thoughts might become so
3
e.g. Russell, 'The Philosophy of Logical Atomism', in Logic and Knowledge, pp.
272-3. Berkeley, Principles ef Human Knowledge,§§ 38-51.
158 LA:\iGUAGE AJ\'D THE v,;oRLD

associated with the vocabulary that anti-realist theory is no

longer perceived as giving an analysis or reduction of the original
content (or as treating it in some other anti-realist way) but as
proposing to replace that content by some purged substitute.
An anti-realist who. accepts that realist views were 'part of the
meaning' of the original saying will have an 'error view' of these,
and urge replacement not reduction: someone who denies this
may regard himself as merely revealing the meaning. As we
have already seen, ifwe grow up to use a given vocabulary and
have various views about the subject matter, there is no likeli-
hood of definite conventions determining how many of those
views are properly part of the content of statements made in the
vocabulary. This does not take the interest out of reductionist
claims. The interest remains in whether sufficient approxima-
tions to A-statements can be made in the B-vocabulary, for us to
regard any differences as in effect detachable -the product of
erroneous ways of thinking of the area. Serious questions can
still remain about the adequacy of a B-vocabulary, as either a
reconstruction or substitute for an A-vocabulary, even if we
remain agnostic about which of these it is best taken to be.

4. The Holistic Objection

The most pressing motive for reductive analysis will be given
from one of the perspectives already introduced - from where
moral, mathematical, mental, or whatever facts become impos-
sible, since the world is perceived as capable of providing
nothing but some one kind of states. So the A-states of affairs
have to be brought down to the B-earth, and the.natural way to
set about this is to equate A-statements with particular B-
statements. But is it the only way? To put it the other way
round, if we find that we cannot equate the content of some
particular A-statement with anything claimed in purely B-
terms, does this show that in committing ourselves to A-
statements we are claiming there to be other realms offact than
B-facts?
Most twentieth-century philosophy, and ways of writing the
history of philosophy, suppose that the answer is obviously that
we are. The option is either to attempt a reduction, or to admit
more facts, more and different objects of knowledge, into our
 RL\LIS:.\I A'.\lD VARL\TlO:,,./S 159

ontology epistemology. So philosophers such as

have ambitions to diminish our epistemological and meta-
physical commitments, must be seen as proposing, in a primitive
and unsophisticated way, reductive analyses: attempting to
say, in acceptable terms, exactly what we say when we talk
about morals, or about causal connections, or numbers,
or whatever. But there is another option. What philosophers
certainly want is a way of explaining our propensity to use and
understand the problematic vocabulary given only that we live
in a world in which we are sensitive to no more than the
underlying truths. The problem they set themselves is to explain
the thoughts without invoking the enlarged reality, with moral
facts or numbers, or with causal powers between particular
events to which we are responding. But must any such explana-
tion proceed by reducing the content of the original vocabulary,
showing that exactly the same can be said in the underlying
B-terms? We can appreciate that it need not do so, by consider-
ing some of the reasons why reductions are rarely obtainable.
For the objections may themselves leave it open whether the
metaphysics is as the reductionist thought, even if meanings do
not drop down in the way he hoped.
Currently reductions are mostly attacked on "holistic"
grounds. These argue that the individual statement is the
wrong unit of analysis. Any individual statement in a given area
will be given its meaning partly by its connections with a
multitude of other statements and concepts. These multifarious
connections, these interanimations in a web of belief, mean that
no one individual statement, considered in isolation, will be
wholly identified through any purely downward anchoring in a
different (B) vocabulary. It is a body of belief, a network of
doctrines, which arises as a whole when we start to appreciate
the world in any given terms - be they moral, or causal, or
involving independent objects in space and time, or numbers,
or whatever. As Wittgenstein put it: "When we first begin to
believe anything, what we believe is not a single proposition, it is.
a whole system of propositions. (Light dawns gradually over the
whole.)" 4
This image has a destructive impact on reductionist hopes.
We can illustrate it by considering the positivist ideal of a
On Certainty, * 141.
4
160 L\:\GCAGE AND THE WORLD

language of pure sense-experience. Suppose an individual could

make a complete record of how he experienced things at each
moment, having at his disposal an extremely untheoretical,
inactive medium for recording this - a sense-datum language;
the "fancifully fanciless medium of unvarnished news"' as Quine
puts it. 5 Suppose now that the individual conceptualizes things
further, beginning to regard his experience as experience of
objects in space and time, which continue independently of his
experience, and which have aspects which at given times he
does not experience, and further which have their own causal
powers and laws of behaviour. How do these new thoughts
relate to the descriptions in the old language? The verificationist
ideal was that they should each have.a determinate footing in the
language of experience: each should correspond to some large
but definite bunch of theses expressed in the bare reportage of
sense experience. The interanimation of sentences destroys this
ideal: no single statement using the new, theoretical, concepts,
will have a "fund of implications to call its own". 6 To put the
point the other way round: if things go wrong, so that experience
does not fall out as expected, there will be no one theoretical
statement which must alone be regarded as falsified, as ifjust its
footings have been knocked away. There is the possibility of
spreading the blame across different theses of any theory, creat-
ing saving hypotheses, adjusting the web of belief in different
ways.
The holistic objection to reductionism was impressed on
philosophers by Quine. 7 But Quine maintained an odd penchant
for putting these sophisticated thoughts about theoreticalinter-
animations alongside primitive verificationist views about
meaning- in particular the idea expressed by C. S. Peirce;· that
the meaning of a statement actually consists in the difference its
truth would make to possible experience. 8 The result is explosive. If
this is what you say about meaning, and you then prove that no
individual statement has such a footing in experience, you have
a proof that no individual statement means anything! But the

5
Word and Object, p. 2.
6
W. V. Quine, 'Epistemology Naturalized'.
7
Initially in 'Two Dogmas of Empiricism'. See also the discussion of the British
Idealists in 7.5 below.
8
W. V. Quine, 'Epistemology Naturalized', p. 68.
 REALIS'.\1 A:\:D \';\RIATJO;.;S 161

dramatic result comes from marrying the sophisticated view of

verification with a definition of meaning which presupposes the
crude view. It is exactly as though you have only seen Stone-
henge, and so define the function of a piece of architecture as
that bit of the roef which it supports. You then find Sydney Opera
House, and realize that it is so structured that you can never
attribute to any one part of the structure one definite bit of roof
which it supports. The webs of engineering interanimate each
other. Perhaps any one part could be taken out without any of
the roof falling down. If we cling to the Stonehenge definition,
this would mean that no bit of the structure has any function. But
this is not the best reaction. The path of wisdom is to revise the
crude definition to cope with the sophisticated reality. The
function of an element in the structure is not a matter purely of
its "vertical" connection with a part of the roof, but also of its
"horizontal" connections of support with other members of the
structure. Quine's reluctance to modify Peirce's definition is a
striking example of the power of verificationist ideas. The idea
that a proper belief should have just one determinate footing in
experience here survives realization that it cannot, to deliver
the odd result that there is no such thing as a single belief, or the
meaning of any single sentence.
If there were a B-vocabulary which stuck especially closely to
the stream of sensations, it would be plausible to suggest that an
ordinarily theoretical A-vocabulary would fit over it holistically.
The tribunal of experience would deliver, in the first instance,
B-verdicts, and the A-theory would face this tribunal en bloc.
But it is not as though there actually is such a B-vocabulary. In
Chapter 7 we see what difference this makes to the way we must
think of experience controlling belief. But the holistic point
remains valid in a conditional form: if there is a B-scheme
devoid of certain concepts, and a richer more theoretical A-
scheme, the judgements of the latter will have their own hori-
zontal connections, meaning that they fit only loosely over the
former. And this destroys the enterprise of isolating distinct
footings for the richer beliefs, identifying their content in the
poorer way.
How is it then that in spite of their general acceptance of these
thoughts, philosophers still pursue relatively local attempts at
analysis: semantics to psychology, actions to events, rights to
162 LA:s;GCAGE AND THE WORLD

duties, causes to regularities, and so on? Perhaps they are not

yet comfortable with other ways of explaining an A-vocabulary
in terms of our reactions to a B-describable world. But the
better defence is that it is often impossible to tell in advance
whether the prestige/admiration model is the right one. In
other words, only after attempting the reduction can we tell
whether it is right to talk of a different scheme of concepts, or
whether instead we just have different ways of putting the
B-thoughts. It is only with hindsight that we can say that
the conceptual scheme of meanings, or causes, or whatever,
is different, and could only relate holistically, to the related
descriptions couched in weaker terms.
Suppose that in a particular area we become convinced that a
reduction will not be found, because the A-vocabulary fits only
holistically over the B-terms. How does this affect the original
impulse to attempt a reduction - the metaphysical, or epistemo-
logical, or logical motivations? Remember that the key concept
is that of explanation. If we can explain why it should come about
that we have the A-concepts, with their horizontal connections,
loosely fitting the B-descriptions, and in this explanation rely
only upon our exposure to a B-describable world, then at least
the metaphysical and epistemological motivations will be
answered. We will be able to do away with a distinct area of
A-facts, which troubled the metaphysician. And we will be able
to explain the reliability ofour A-beliefs without demanding a
queer mechanism enabling us to know about them, which
troubled the epistemologist. Let me give an example. Consider
a system of arithmetic, with its mysterious reference to abstract
objects. One attitude might be that we should do away with it,
reject the vocabulary, and try to do "science without numbers".
Another is that we should reduce the problematic sayings: 'To
say that 2 + 2 = 4 is to say that ... ', where the paraphrase
avoids the abstract reference. The third attitude is to see the
whole arithmetical system (the thing over which light dawns as
a whole) as an explicable and legitimate instrument for pursuing
our concerns in the world - a world which does not contain any
abstract objects. The hope would be that we can explain why
we should think as if there are numbers, taking arithmetical
sayings quite literally and seriously. But this would be done
without postulating that there are numbers and that we can
 RL\LIS:\l A'.\JD VARIATIONS 163

fortunately intuit them and their properties, and without pre-

tending that thinking as if there were numbers is really just
thinking in another, reduced, way.
It is as though nature permits portraits in richer or more
austere styles. The portraits are different, but it does not follow
that we have to imagine a different sitter for each of them.
To work out these thoughts will require different detail in
different areas. It is not likely that the explanation of why we
think arithmetically could share much with an explanation of
why we think morally or causally or in terms of space and time;
nor is it evident that there will always be a legitimate project of
attempting such an explanation on the thinner base. For
example, if we can have no conception of a non-spatial or
non-temporal world, nor of our exposure to it, we cannot really
mount an explanation of the spatial and temporal aspects of our
thinking by mentioning it (see further on Kant and this, in 6.5).
But the problem to which I now turn is not this, but is the
puzzling question of the status of reductions, and whether,
when we reflect upon their status, we can always accept the
benefits they appear to bring.

5. The Status Problem

There is not much problem about the equation between having
high prestige and being much admired. The fact that the one
concept is identical with the other is not, as it were, a mysterious
fact about a realm of concepts and thoughts which exists inde-
pendently of our understandings and our words. We would
rather say that there is just one concept, and we have slightly
different ways of expressing it. This indeed raises serious ques-
tions - most pressingly, the question of how a person can
understand each term in an equation such as this, yet fail to
realize that the equation is true. This is a version of the paradox
of a ~ , and I discuss it briefly later (9.5 and notes). But
there are other reductions, or attempted reductions, where the
problem of status becomes very pressing indeed. I shall illustrate
this with reference to the problem of personal identity.
In 2.1 I described how the best philosophical problems arise
because we have a metaphysic, a view of the kinds of fact which
make up the world, and discover that some particular kind of
164 L\:'\GUAGE A:'JD THE WORLD

judgement does not seem to be made true by those facts. We are

at a loss to see what makes it true. And reduction is a hopeful
way of answering this problem. One such judgement is that of
personal identity through time. We might have a metaphysic of
physical and psychological atoms: physical atoms are the self-
subsistent basic building blocks of physical stuff, and psycho-
logical atoms are the particular mental states whose succession
makes up the conscious life of any person. These atoms have
relations to one another: through time they group together,
enter into causal relations, and these groupings can come and
go; through long enough time, they will all go. We draw a line
around some groups: they make up the life of one person. But
what kind of fact is it, that some previous mental state, say, and
some present one, both belong to one person, to me? Which
relation do they bear to each other for this to be so?
There are many possible answers. We can select, for instance,
the causal facts, and look for personal identity when the earlier
stage of the person is causally in some favoured relation to the
later one. We can ask for identity of underlying physical stuff.
We can ask for sufficient mental continuity, and we can ask for
an amalgam of any of these. The literature is full of thought
experiments, or puzzle cases, which tell stories of rearrange-
ments of the physical and mental atoms, and ask whether
various proposed criteria give the right answers in these bizarre
possibilities: could a person survive teletransportation, in which
a physical replica is created elsewhere; does a person survive
gross amnesia and personality changes; does he survive replace-
ment of enough cells by others; does he survive fusion with
another, in some massive graft, or fission into separate people in
some massive break-up?
Suppose that after considering these examples, we hit upon
one solution. We would say: given a history of physical and
psychological atoms, in such-and-such groupings, we should
group them into persons by principle X. (X might allow for
borderlines and questions of degree, of course.) This would
allow a reductionist claim: there are no facts in the world
beyond the onward flow of physical and psychological atoms in
different groupings, and the fact that some such groupings are
united under principle X; those that are we call persons. The
question of status is then this: what kind of truth is it that if two
 RL\LIS:\1 AND VARIA TlONS 165

mental states belong to one X group they belong to the same

person, and if they do not, then they do not belong to the same
person? There seem to be three possibilities:

(I) Such a claim is justified by convention. It reflects the way we

have chosen to group atoms into persons: it represents our
solution to a problem which could equally well have been
solved in other ways. The trouble with this is that it is hard to
believe. If two experiences are both mine, it is hard to believe
that an alternative linguistic arrangement could have led a
group to say with equal propriety that one was mine and one
was not.

(2) Such a claim is contingently true. In the actual world, X

groupings make up persons, but God could have chosen to sort
things so that some other groupings did. The trouble with this is
that it is the end ofreductionism. For it admits that there is this
further fact of the matter - the one which actually obtains or
which, as it were, God selected - and it offers by itself no
account of what kind of fact that is. The fact of personal identity
through time remains unexplained.

(3) Such a claim is necessarily true and represents a deep truth

about the way we have to think of persons. It is not a question of
choice, but neither is it a question of a contingent actual truth.
This is rather like the status that Kant sought for truths about
time and arithmetic- the synthetic a priori. But then it is one thing
to say that we do think of X groupings as making up persons; it is
quite another thing to say why we have to do so; to say what
kinds of (conceptual? logical? moral?) mess would arise if we did
not choose just X, but some other unifying relation X*.

The problem of status was insufficiently remarked by twentieth-

century reductionists, because they were happy with a rough
concept of synonymy which had one foot in ( 1) and the other in
(3). In other words, reductionist claims would be logically true,
representing the right way to think about the problematic
judgements and their relation to the underlying facts, but also,
like other logical truths, somehow protected only by convention.
But this fudges the issue. The problem of status is quite crucial
166 LANGUAGE AND THE WORLD

whenever we start out by believing that there is a serious issue,

an issue to be got right, when we are selecting a favoured class of
underlying B-truths to represent an A-truth. It thus arises when
naturalistic analyses are proposed in ethics (what is the status of
the claim that just these features are good?) when possible
worlds analyses are proposed for counterfactuals (what is the
status of the claim that just this relation between possible worlds
makes it true that if he had jumped off the cliff he would have
fallen?), when set-theoretical reductions are proposed for
numbers (what makes it true that just this set is the number 7?),
and indeed whenever serious metaphysical doubts prompt the
reductive search.
I think the problem of status raises profound difficulties for
the enterprise of doing metaphysics by searching for reductions.
I do not think it is an insurmountable problem: we would
expect the doings of nations to be reducible to the inter-related
doings of people in them; the identity of bicycles to reduce to the
identities of groups of bicycle parts; and we would expect it to be
at least possible for categories of A-fact to evaporate into B-facts.
But this can only be claimed to have happened when we have
both selected a plausible bunch ofB-facts, and solved the status
problem.
I now return to other ways of pursuing the debate between
realism and anti-realism. I shall illustrate this by following out
the detail in one case, that of evaluations and of ethical or
broadly moral language. Apart from its intrinsic interest this
also gives us some leverage against quietism. For always in the
wings there is the doubt that we cannot finally make any of
these explanatory attempts stick. The quietist will argue that
there is no independent access to the true face of nature, to
show us which portraits have· added the least realistic stylistic
flourishes. We only have our own descriptions, our own collec-
tion of portraits to rely upon. Quietism argues that this blocks
anti-realist explanations of any commitments. But it does not:
we can get further insights into the sitter by walking around
the collection of portraits, and thereby determining which are
adding gay-coloured vesture, and which are nearer the undraped
figure of nature herself.
 REALIS:\I A'.\JD VARIATIONS 167

6. Expressive Theories: Contrasts with Truth

This brings us to the third point of departure for anti-realism:
the attempt to explain the practice of judging in a certain way,
by regarding the commitments as expressive rather than descriptive.
The commitments in question are contrasted with others - call
them judgements, beliefs, assertions, or propositions - which
have genuine truth-conditions.
Two classic examples of such theories are instrumentalism,
as a philosophy of science, and emotivism in ethics. According
to this latter, the commitment that a thing is good or bad, right
or wrong, permissible or impermissible, is not a judgement with
truth-conditions of its own (probably irreducible to other terms,
by the argument of the last section, and therefore highly
mysterious). It is a commitment ofa different sort, maintained
not by believing something but by having an attitude towards it.
The theory was expressed with characteristic vigour by A. J.
Ayer:
The presence of an ethical symbol in a proposition adds nothing to its
factual content. Thus if I say to someone, "You acted wrongly in
stealing that money", I am not stating anything more than if I had
simply said, "You stole that money". In adding that this action is
wrong I am not making any further statement about it. I am simply
evincing my moral disapproval ofit. It is as if I had said "You stole
that money," in a peculiar tone of horror, or written it with the
addition of some special exclamation marks. 9
Emotivism is sometimes dubbed the "boo-hooray" theory of
ethics. Ayer here presents it by a contrast between what is stated
and what it evinced. The same contrast can be put in terms of
what it is to accept a moral remark; it is to concur in an attitude
to its subject, rather than in a belief. Alternatively we might say
that the speech-act of putting forward a moral opinion is not
one of asserting that some state of affairs obtains, but one of
evincing or expressing an attitude, or perhaps of exhorting or
encouraging others to share an attitude.
In the heyday of linguistic philosophy similar suggestions
were applied to a wide range of commitments:.

9
A.J. Ayer, Language, Truth and Logic, ch. 6, p. 107.
168 LANGUAGE AND THE WORLD

Saying that X causes Y = Offering X as an instrument

or recipe for obtaining Y
Saying that action Xis Expressing willingness to
voluntary blame or commend the agent
Saying that you know X Allowing other people to take
your word for it
Saying that a statement is Endorsing it
true
Saying that a statement is Expressing guarded assent
probable to it

Saying that a statement is Refusing to endorse ruling it

possible out of consideration.
Expressive theories contrast with reductions. On an expres-
sive account there is a considerable difference between saying,
for instance, that X causes Y, and saying anything which might
be a plausible candidate for a reductive analysis - e.g. that
events similar to X are always followed by events similar to Y.
You might only offer the recipe if you believed this latter thing.
You might be open to criticism if you offered it when this latter
thing was false. In other words, the regularity might provide a
standard for endorsing the recipe. Nevertheless the standard
might, in principle, be variable or displaced by something more
subtle, but this would not be a change in the meaning of the
causal saying. This is particularly important in the moral case.
Naturalism finds it impossible to say how disputants with
different standards mean the same by moral remarks. On the
account according to which the meaning of such a remark is
given by the standard for saying it which the speaker has,
people with different standards mean different things. But that
makes it impossible to see how they are expressing conflicting
opinions. If a utilitarian says that contraception is an excellent
thing, he would mean that it promotes happiness; ifa priest says
that it is an awful thing, he might mean that it is against the
wishes of the creator of the universe. But in that case each
remark could be true, and there would be no contradiction
between them. The expressive theory avoids this undesirable
consequence. It locates the disagreement where it should be, in
the dash of attitudes towards contraception.
 RE.-\LIS:\1 A'.\/D VARIATIO'.\iS 169

Expressive theories must be sharply distinguished from more

naive kinds of subjectivism. An expressive theory does not give
a moral utterance a truth-condition which concerns the speaker.
A man saying 'Hitler was a good thing' is expressing or evincing
an appalling attitude. But he is not saying that he has got this
attitude. If he were, what he said would be true, provided that
he is sincere. But we do not regard his remark as true; we allow
that he is sincere, if he is, without accepting his remark (which,
on the expressive theory, would mean endorsing the attitude it
expresses).
The point of expressive theories is to avoid the metaphysical
and epistemological problems which realist theories of ethics,
and of the other commitments in the list, are supposed to bring
with them. Again it is important to remember the overall
motivation, This is to explain the practice of moralizing, using
causal language, and so on, in terms only of our exposure to a
thinner reality- a world which contains only some lesser states
of affairs, to which we respond and in which we have to conduct
our lives. Unless this is borne in mind, it is easy to charge
expressive theories with irrelevant mistakes. For instance, it is
frequently pointed out that a term may occur in an utterance
which both is a description of how things are, and expresses an
attitude. If I say that there is a bull in the next field I may be
threatening you, or warning you, or expressing timidity, or
challenging you to cross, or doing any of a range of other things,
and expressing any of a range of subtle attitudes and emotions.
But none of these doings has any bearing on the meaning or
content of my remark, which is true or false in a determinate
range of circumstances, and is a paradigm of a saying with a
truth-condition. If I say'that someone is a Kraut, or blotto, I
may express an attitude of contempt towards Germans, or of
wry amusement at drunkenness, but I also say something true
or false about their nationality or sobriety. In the bull example
the attitudes expressed are incidental to the conventional
meaning of the remark. In these other examples they attach to
the vocabulary as a matter of convention. You should not use
those terms unless you sympathize with those attitudes. But in
each case it would be wrong to infer that no description is given
from the fact that an attitude is also expressed. Similarly, critics
have pointed out, it is wrong to infer that there is no strict and
170 L\'.\'GUAGE AND THE \\'ORLD

literal content, capable oftmth and falsity, in remark thatX

causes Y, or X knows the truth about Y, or X ought to be done,
from the fact that when these things are said, attitudes are
expressed. This "speech-act fallacy" has been widely accepted
as the root mistake of expressive theories. 10
However, the fallacy need not be committed (and I rather
doubt ifit ever was). There are two reasons why not. First of all,
an expressive theory should not infer that the attitude gives the
role of the saying, simply from the fact that it is expressed when
the saying is made. So long as the attitude may give the role, the
argument for saying that it does is the superior explanation of
the commitments which we then arrive at. There is no inference
of the form 'this attitude is expressed, so these remarks have no
truth-conditions', but only 'this attitude is expressed; if we
see the remark as having no truth-conditions the philosophy
improves; so let us see the remark as expressive rather than
descriptive'. There is no fallacy there. And there is a second
point. Remembering the anti-realist motivation, we can see
that it does not matter at all if an utterance is descriptive as well
as expressive, provided that its distinctive meaning - the aspect
which separates it from any underlying B-descriptions - is
expressive. It is obviously a useless argument against anti-
realism about values to point out that the word 'Kraut' has a
descriptive element. That is quite acceptable. It is the extra
import making the term evaluative as well as descriptive, which
must be given an expressive role. It is only if that involves an
extra truth-condition that expressive anti-realism about values
is impugned. But perhaps there is no good reason for supposing
that it does: the natural thing to say about such terms is that the
extra ingredient is emotive, expressive.
There are, however, much more respectable arguments sur-
rounding expressive theories. These concern the extent to which
they can explain the appearance that we are making judgements
with genuine truth-conditions. Ultimately it is the attempt to
explain this which introduces the need for a wider theory of
truth, and enables us to appreciate the point of the different
contenders. We should realize that expressive theories, like
reductive theories, may be uncertain about how much they
need to explain. Suppose that we say we project an attitude or
10
J. Searle, Speech Acts, p. 139. H. Putnam, Reason, Truth and History, pp. 208-10.
 RE.\L!S:\I .\:s;D \'.\RL\T!O:s;s 171

habit or other commitment which is not descriptive onto the

world, when we speak and think as though there were a property
of things which our sayings describe, which we can reason
about, know about, be wrong about, and so on. Projecting is
what Hume referred to when he talks of "gilding and staining
all natural objects with the colours borrowed from internal
sentiment", or of the mind "spreading itself on the world".
Then expressive theorists often tend to the view that this projec-
tion is a mistake - that itself it involves flirting with a false
realism. (Ayer entitled his chapter from which I quoted, "A
Critique of Ethics and Theology".) John Mackie believed that
our ordinary use of moral predicates involved an error, because
the underlying reality was as the expressive view claims, whilst
in using those concepts we claim more. 11 For this reason he
denied that emotivism gives the actual meaning of moral terms,
and this claim is frequently put forward as almost self-evident.
But it is not. The issue is whether the projection is only explicable
if we mistake the origins of our evaluative practices. The idea
would be that were we aware of these origins we would give up
some or all of our tendency to practice as if evaluative commit-
ments had truth-conditions, and were not expressive in origin
and in their essential nature. But perhaps there is no mistake.
I call the enterprise of showing that there is none - that even
on anti-realist grounds there is nothing improper, nothing
"diseased" in projected predicates - the enterprise of quasi-
realism. The point is that it tries to earn, on the slender basis, the
features of moral language (or of the other commitments to
which a projective theory might apply) which tempt people to
realism. The issues are complicated enough to deserve a chapter
to themselves. It is only when we have understood them that we
can properly assess the final divisions between realists and their
opponents, over mind-dependence and over truth.

7. Metaphor and Truth

There is one other overtly expressive phenomenon of interest to
the philosophy oflanguage: metaphor. Metaphor has not com-
manded the respect of many mainstream theorists of meaning:
it can appear to be merely a poor relation of proper judgement,
J. :\lackie, Ethics: Inventing Right and lrrong, pp. 30-5.
11
172 L\:\"GCAGE A'.\i"D THE \\'ORLD

arising only when language is on holiday, and the appropriate

attitude is that voiced by Hobbes:
In Demonstration, in Councell, and all rigorous search of Truth,
Judgement does all; except sometimes the understanding have need
to be opened by som.e apt similitude; and then there is so much use of
Fancy. But for Metaphors, they are in this case utterly excluded. For
seeing they openly professe deceipt; to admit them into Councell, or
Reasoning, were manifest folly. 12
By contrast with this stark hostility we can mention Ni.etszche's
attitude that truth itselfis merely good dead metaphor. 13 Those
who appreciate a good metaphor certainly seem to acquire a
gain of some sort. So we should at least ask how this gain
compares with that of acquiring a true belief. I can best
organize this issue by comparing four different positions, in
ascending order of the importance they attach to metaphor.
( 1) The prosaic end of things. Some metaphors scarcely
deserve the name: they are close to being dead metaphors, or
idioms. If I tell you not to cross the path of someone who is
prickly, steaming, or up in arms, you understand my language,
for all its figurative nature, as immediately and certainly as if I
had chosen more literal ways of expressing myself. The words I
use are customarily associated with features of things which I
could have expressed directly. Let us say that the figurative
language yields an interpretation if the interpretation is
customarily given, and customarily expected. If the custom has
hardened into a convention we have a case of an idiom, and it
will be right to say that we have a new or extended literal
meaning. Perhaps 'prickly' in this context means 'easily
aggravated'. In that case a dictionary ought to report the fact,
and a learner ought to be taught it. And someone could fully
understand sentences in which people are described as prickly,
without knowing that the word has a use in which it applies to
cacti and hedgehogs. But a term may yield an interpretation
reliably enough without this having happened: the yield is
given firstly by the literal meanings of the terms, and secondly
because some feature of the original thing meant is 'salient', or
an associated commonplace. In other words it is a feature
12
Hobbes, Leviathan, p. 44.
1., See start of chapter 6 below.
 REALIS:.\! A'.\iD VARIATIONS l 7:l

people think of when they think of the original, and which in the
context is a likely feature to apply to the subject-matter. 'Bert is
a real gorilla' yields that Bert is strong, rough, and fierce. It
does this because the word 'gorilla' has its normal meaning,
and because people (wrongly) associate these features with
gorillas. 14 Any hearer aware of this can follow the metaphor to
its intended interpretation.
In these first-level cases the metaphor has an intended
interpretation which is reliably given. But the mechanism goes
via the ordinary meaning of the terms, to the suggested mean-
ing: we have only an indirect way of suggesting some definite
truth by saying some definite falsehood (for Bert is not really a
gorilla). There can be some debate about how we fit these facts
into other theoretical categories. Is it right to describe the
speaker as having asserted falsely that Bert is a gorilla? Is
it right to describe him as having asserted truly the yielded
propositions, that Bert is strong and rough and fierce? Because
of our analysis of convention and tone we need not find these
questions too hard. The speaker said that Bert is a gorilla, but
did not assert it: he did not intend anyone to believe that this
was the truth, and would not normally be taken to have dis-
played that it is. He did, on the other hand, intend people to
believe that Bert is strong, rough, and fierce, and chose a
reliable method of transmitting this belief, and of being taken to
do so. But the method was one of reliable suggestion, and we do
not allow that people assert everything that they reliably sug-
gest, and are known to be reliably suggesting. 15 However, the
responsibility the speaker bears is much the same as ifhe had
said straight out what he transmits only indirectly. He has
chosen a customary and certain way of representing to his
audience that Bert is strong, rough, and fierce: ifhe is not, Bert
has an equal right to feel aggrieved (ifhe is really like a gorilla,
perhaps he will not do much about it).

14
People may think of these features although they do not believe gorillas to have
them. But in a population which comes to realize this, the metaphor will begin to die
out, or degenerate into an idiom. For consider children learning. They will only realize
the intended yield by learning that it is a distinct custom to use 'gorilla' as if fierceness
were a feature of the animal, when natural history has told them it is not. This is not
readily distinguishable from learning a distinct convention, or meaning.
" See also 9.1.
174 L-\:\'GUAGE AND THE WORLD

(2) The next cases differ from the previous ones in that they
maintain an open-ended or creative element. The range of
features indicated remains indefinite: both speaker and listener
are able to explore the comparison or image suggested, and find
new features of the subject matter as a result. The critic I. A.
Richards talked of poetic language as a "movement among
meanings": there is no single literal truth which the figure or
metaphor yields, and no single class of such truths. Thus when
Romeo says that Juliet is the sun we can profit from the
metaphor indefinitely: we can move among respects in which
someone's lover is like the sun: warm, sustaining, comforting,
perhaps awesome, something on which we are utterly depen-
dent ... This process is quite open-ended. Shakespeare need
have had no definite range of comparisons which he intended, '
and it is quite wrong to substitute some definite list and suppose
that the exploration is complete. The metaphor is in effect an
invitation to explore comparisons. But it is not associated with
any belief or intention, let alone any set of rules, determining
when the exploration is finished.
This is the first sense in which metaphor is both valuable and
ineliminable. It is valuable because it directs our attention
towards aspects of things which we might not otherwise have
thought of. It is ineliminable because there is no single list of
literal thoughts which cashes it in. In this respect the metaphor
may work like a picture (we talk in the same breath of
metaphorical and figurative uses of language, or uses of
imagery). Possessing a picture I may think of the subject
pictured, and the picture may lead me to think of all kinds of
things. But no list of these things substitutes for the picture,just
because the picture has an unlimited potential for directing me
to further aspects of the subject, whereas the list does not. As
Davidson puts it, "Joke or dream or metaphor can, like a
picture or a bump on the head, make us appreciate some fact-
but not by standing for, or expressing, the fact." 15

(3) The preceding level is enough to defend and in a way to

explain the value of metaphor. It gets us beyond the crass
thought that using metaphor is always an indirect and inferior
way of putting what would best be put directly (although it can
16
'\\'hat Metaphors Mean', in Platts (!980), p. 253.
 REALISM AND VARIATIONS 175

be this, and at the first level often is). at the second level
the value of a metaphor is essentially that of a means to an end.
The end product is appreciation of a literal truth or several
literal truths. The metaphor suggests how to go about finding
some such truth or truths. Its success is dependent upon their
value to us, and perhaps too upon providing us pleasure in the
exploration. (There is pleasure in exploring the metaphor of the
Church as a hippopotamus even if we do not believe anything
about the Church at the end that we did not believe at the
beginning.) So far, however, the only way that a metaphor can
provide a gain in understanding is by provoking a quest which
may end up in our grasping some new strict and literal truths.
The third level of description queries this. It alleges that there is
a distinct, intrinsically metaphorical, way of understanding.
The appreciation of the metaphor constitutes a different, dis-
tinctive success of its own: the success of seeing one thing as
another. Seeing history as a tidal wave, or architecture as frozen
music, is doing something different from believing, and diffe-
rent from accepting an invitation to search for a range of beliefs.
But it is a gain in understanding, a success, of its own kind.
It is absolutely vital to see that this kind of description is not
forced on us by level (2) facts. And it is a defect of some of the
best literature on the subject that it does not dearly separate the
two ideas. Some even explain the ineliminability of metaphor
via its alleged level (3) powers. Davidson, at the end ofhis highly
illuminating paper on the subject, says: "Since in most cases
what the metaphor prompts or inspires is not entirely, or even
at all, recognition of some truth or fact, the attempt to give
literal expression to the content of the metaphor is simply
misguided." But the attempt to give literal expression would be
misguided even if we accepted only the second level of descrip-
tion, where the metaphor does prompt or inspire a search for
literal truths or facts. It is quite another question whether it can
prompt or inspire a mode of insight all of its own, so that the
appreciation of one thing as another or in the light of another is a
distinctive metaphorical way of understanding. It is this idea
which Hobbes was opposing.
The sheer psychology of coming to appreciate a good
metaphor, like seeing a joke or seeing the point of a comparison,
may suggest this way of describing things: If I suddenly see
176 L\'.\TGUAGE AND THE WORLD

architecture as frozen music, or history as a tidal wave, it feels

like a gain which is quite akin to acquiring a new piece of
knowledge. It seems like possession ofa self-sufficient thought. It
doesn't seem just like accepting an invitation to explore the
comparison, because I may relish the metaphor and yet have no
sense at all that anything remains to be done, which is the
distinctive aspect according to (2) descriptions. In support of
this we can notice how often we assent to a remark without
realizing that there is metaphor buried in it. A large amount of
philosophy consists in unravelling how much is metaphorical,
and how much literal, when we talk of: the foundations of
knowledge; bodies acting upon one another; the flow of time;
propositions corresponding to facts; the mental realm, and so on.
Suppose we decide that there is a large amount of metaphor
involved when we talk, say, of the foundations of knowledge.
What then do we say of someone who passionately believes that
knowledge has foundations, and in no sense realizes the degree
of metaphor? He does not realize how much exploration needs
to be done to arrive at some literal truth about knowledge, so he
does not see himself as someone who has just accepted an
invitation to look for such truths. And he is not someone who
has come to a literal belief about knowledge, because he may
have no non-metaphorical way of expressing himself. Are we to
credit him with a metaphorical piece of understanding (or
misunderstanding, if knowledge needs no foundations)?
The opposing view would be that there is no distinctive
understanding here. The man says the words supposing that
they mean something true, and that he knows what it is. More
thought would disillusion him. So he simply believes he has a
belief, when in fact he has none. He thinks he is rich when he has
a cheque which he cannot actually cash. But he has not there-
fore come into possession of a different kind of currency. Con-
sider too the exercises of imagination involved in seeing history
as a tidal wave, or architecture as frozen music. We present
ourselves with images; we put a sense of the inexorable sweep of
a tidal wave alongside our thoughts of the progress of historical
events, or our sense of the structure of a cathedral alongside a
sense of the structure of a piece of music, suddenly stopped in
time. Such exercises are pleasurable, even marvellous. But the
ability to conduct them is not itself the exercise of a distinctive
 REALISM AND VARIATIONS 177

piece of understanding. A historian who performs the exercise

every so often may enjoy the idea; he may even find that in some
way it guides his writings or his attitudes to events. But it is
what he then says which means that he understands history
well or badly. In short, it is not necessary to postulate a distinc-
tive success which the metaphor provides. Its virtue lies in
prompting insights, but the insights themselves are strict and
literal truths.

(4) This rebuttal of level (3) descriptions of metaphor may

leave some dissatisfaction. I would express it like this: the
rebuttal takes for granted the kind of success involved in acquir-
ing a (strict and literal) belief But what right have we got to
contrast this with seeing one thing as another? We have seen
how ordinary belief involves the transference of a term from one
set of things to another, and it was the business of chapter 3 to
search for some conception of the rules and intentions which
govern this process. In that chapter we met the idea that only
natural propensities to find things similar underlie any kind of
description of things. Why shouldn't a natural propensity to see
one thing as another or in the light of another, which yet falls
short of applying the same predicate to them in the same
meaning, not serve to found a self-standing piece of metaphori-
cal insight? I take it that this is why Nietzsche describes literal
truth as worn-out metaphor (see the start of chapter 6). Is belief
perhaps just one end of a spectrum of attitudes which can go
with the new application of a term: at this end, pure prose, and
at the other end, pure metaphor?
To think of this, compare two populations. Each of them is
taught a certain term initially in connection with a certain
range of cases or range of procedures. In one population they
find it natural to describe some new cases by the same term. In
the other they find it natural to see the new cases in the light of
the old, or see them as possessing the feature, but they view this
as irreducibly a case of'seeing as'. These think of the extension
as metaphorical, whereas the ohers think ofit as a simple case of
belief. What is the difference?
A plausible example might help. Consider the way in which
we use terms which also describe physical affairs to describe
psychological affairs. We talk of matches in boxes, and thoughts
178 LANGUAGE AND THE WORLD

in the head; of being pulled by ropes, or being Julled by desires;

ofjumping to attention, or jumping to a conclusion. Metaphori-
cal, of course. Or is it? We most probably cannot cash the
metaphors, either by giving a single literal way of saying some-
thing which they yield or even by indicating a range of compari-
sons which they suggest. On the other hand we should not be
happy with just postulating two different senses of the terms: it
is not pure accident that we talk of thoughts in the heads, for
example. If it were simply a case of the word 'in' being ambigu-
ous we might just as well have planted the ambiguity on some
other term, and talked of thoughts being on the head, for
instance. Yet we should also be unhappy with the idea that
there is a smooth or natural extension of the terms from the
physical to the mental. Someone who understands what it is for
one thing to be in another physically has a definite range of
capacities: he can find things, obey instructions, perform a
range of tasks which exhibit or exercise that understanding.
These tasks are not the same as any (whichever they might be!)
which exercise an understanding of thoughts being in the head.
It is profitable here to recall the bent learners of chapter 3.
Imagine a bent learner who responds perfectly well to some
range of cases, but whose capacity to exercise his concept
involves odd bends and blind spots. From our point of view his
capacity is flawed or partial. From the point of view of the
population which finds it natural to spread a term over a larger
range, people who do not find it natural, but who regard the
extension in the spirit ofa metaphor, would similarly be flawed
or partial. Now if we sympathize with either conventionalist or
even naturalist responses to the bent learner it will seem that
such reactions are all that there is to go on: it is then a matter
merely of our reactions which determine whether an affirma-
tion expresses a belief or only a metaphor. But if the opposition
to nominalism which I tried to force in 3.2 is roughly right, we
do not get this result. According to me there is a distinctive fact
whether the extension over new cases is, or is not, an exercise of
the same capacities that are involved in applying the term to the
original kind of case. This ought to give determinancy to the
issue of whether a new kind of application is metaphorical or
tlOt.
But even if the issue is determinate, it is often not easy.
 REALISM AND VARIATIONS 179

Remember the examples of philosophical descriptions/meta-

phors, and the difficulty of disentangling the literal from the
metaphorical which they illustrate. I suggested that the man
affirming that knowledge has foundations may only think he is
expressing a literal belief, but not be doing so; similarly for
someone saying that thoughts are in the head, or that we are
often pulled by desire. If such a person recognizes no obligation
to find a better way of putting it and insists on resting content
with his way, what can we urge? We show the way the practice
differs: what verifies whether a building has foundations,
whether one thing is inside another, whether one thing is
pulling another. We show how in the ordinary cases these
truths influence things which, for example, we want; beliefs in
them therefore couple with our desires to have consequences for
action. Since the practice is different in the disputed case,
however much we sympathize with the passive acceptance of
the same terminology, the obligation to look for the points of
similarity and difference remains.
Sadly, then, I incline to Hobbes's view that understanding
things metaphorically is not understanding them at all,
although it may often immediately yield understanding, and
guide it and increase it. On this account a good metaphor at the
open-ended level is expressed by an utterance which does not
say that such-and-such is the case, but rather expresses an
invitation or suggestion that a certain comparison be followed
up. In this respect such an utterance is like the other speech-
acts listed in the last section. It does not have truth conditions,
but is successful or not in a different dimension. In the next
chapter I explore this contrast a little more, and consider
objections which such accounts tend to meet.

Notes to Chapter 5
5.3 Reductionism in the Philosophy of Science deserves much more
discussion than I have been able to give it here. A good literature
would include:
A. J. Ayer, The Foundations of Empirical Knowledge.
I. Scheffler, The Anatomy of Inquiry, Pt. II.
R.Carnap, 'The Methodological Status of Theoretical Concepts',
Minnesota Studies in the Philosophy of Science, vol. i ( 1956).
B. van Fraassen, The Scientific Image, ch. 2.
180 L\:\GL'AGE AND THE WORLD

5.4 The status problem is not well highlighted in most of the litera-
ture. This is probably because undue optimism about the conven-
tional status of!ogic, and the logical status ofreductions, made it seem
enough to find a good correlation between A-facts and B-facts, with-
out worrying too much about the status of the correlation.

5.5 It is important to be clear about the distinction between pro-

jectivism and quasi-realism. Projectivism is the philosophy of evalua-
tion which says that evaluative properties are projections of our own
sentiments (emotions, reactions, attitudes, commendations). Quasi-
realism is the enterprise of explaining why our discourse has the shape
it does, in particular by way of treating evaluative predicates like
others, ifprojectivis,m is true. It thus seeks to explain, and justify, the
realistic-seeming nature of our talk of evaluations - the way we think
we can be wrong about them, that there is a truth to be found, and so
on. One might believe that quasi-re~Jis.mis~~\lccessful, yet still dislike
pro~ectivism, and one might lik€ot-:cti,~~)but st~ll believe th~t
ordmary features of our thougrtt="'a:re not explicable, quas1-
realistically, but indeed ir1xpl¥e~cc!".!:ro!::, This was the position John
Mackie took in his influential book Ethics: Inventing Right and Wrong.
But quasi-realism at least removes the most important range of
objections to projectivism - namely, that it cannot account for the
phenomena of ordinary moral thinking. However, a full defence of
projectivism would need to discuss moral motivation more than I
have done.

5 .6 The most comprehensive recent collection on metaphor is

A. Ortony (ed.), Metaphor and Thought.
However, I think nobody would claim that the study of metaphor has
been one of analytical philosophy's brighter achievements. In
particular, a number of discussions seem to presuppose that the
problem is one of how we "compute", according to rules and princi-
ples, a non-literal meaning from a literal one. I hope that the stress in
the text on the dynamic and open-ended exploration of metaphor
removes the idea that we do any such thing.
 CHAPTER 6

Evaluations, Projections, and

Quasi-Realism

What then is truth? A movable host of metaphors, metonymies,

anthropomorphisms: in short, a sum of human relations which have
been poetically and rhetorically tempered by, transferred, and embel-
lished, and which, after long usage, seem to be fixed, canonical and
· binding. Truths are illusions which we have forgotten are illusions;
they are metaphors that have become worn out and have been
drained of sensuous force, coins which have lost their embossings, and
are now considered as metal and no longer as coins.
Nietzsche, 'On Truth and Lies in a Normal Sense'.

l. More Detailed Motivations

A projective account of some commitments puts them at the top
of the following picture:_

spread
properties projected
tj habits, emotions,
sentiments, attitudes

genuine, observed
properties

explanatory
properties
impinge

real beliefs
0
Fig. 5

This is contrasted with the two other. places for them. In the
middle one, the features talked about (e.g. the possession of
value by things, or the existence ofrights, duties, and so on) are
 182 L\:\"GUAGE AND THE WORLD

themselves part of the genesis of our beliefs. It would be because

values, etc. are distributed in some w_<iy arnund the world, and
because we are capable ofreacting to them~·of~nowing them by
some acquaintance or intuition, that we mora:lize as we do. In
the oflie): lower place there would. be no cognizance of these·'
featur~s, nothing to be compared with a more ordinary mode of
perception. All the information about the world which we take
in would be describable in natural terms. But just as we can use
that information to construct theories involving higher-order 1

concepts, such as those of physics, so we can use it to constructf

the moral concepts. But when we have done so, we have a I
further description of the world, and are regarding it as contain-
ing further, moral, states of affairs.
To ensure that a projective theory starts at reasonable odds, I
shall briefly mention three motives for preferring it to either of
its rivals. The first of these is economy. The projective theory
intends to a~k no 1119re fro111 the world than wliaT we know i~c-,.· _
th.e.ce - the ordinary features of things on the basis of \Vhich we
make decisions about them, like or dislike them, fear them ancl
avoid· them, .desire them and seek them. out. It asks no.more
than this: a natural world, and patter~r~lre<tction to it~)By
contrast a theory assimila!,ir1g111oral ur1~~E~ta~c'ii'ngto percep-
_ti()r,i ~~!l}l!l}cis.~or.e of.the~~1J!:.J~erception IS. a-causaI
proces1:;··
we perceive t.Qpse features of things which are_resp~msible for
9ur.,exp~rien~ey-rrisuneconomical to postulate both.a.feature"
."o:(tliipg~_J the v~fues they;:J:1_a ve) and a mechanism (intuition} by
;.C~hi.c;J),~e",9;1"e 4~in~ily 3:wc:i:xegfit. , .. · ·
The second argument for preferring projective theories is
metaphysical. It concerns the relation between the values we
find in the world, and the other properties of things of value. 1
The argument arises from the common claim in philosophy that
one kind of state supervenes upon another. The idea is that some
properties, the A-properties, are consequential upon some
other base properties, the underlying B-properties. This claim
is supposed to mean that in some sense of necessary, it is neces-
sary that if an A-truth changes, some B-truth changes; or if two
situations are identical in their B-properties they are identical
I
The argument of the next few pages is quite hard, and self-contained. It can quite
easily be skipped, especially if the reader already sympathizes with projectivism. The
main thread is resumed in 6.2.
 EVALl:ATIONS, PROJECTIONS 183

in their A-properties. A-properties cannot (in this same sense)

vary regardless of B-properties. Now there are very different
strengths of 'necessity' and of 'cannot'. Most people would say
that colour properties cannot vary independently of underlying
physical properties, such as those of surfaces, the kind of wave-
length oflight reflected, and so on. This is a physical, empirical,
claim. It is easy to imagine finding that the colours of things have
started to vary independently of their physical properties, so
that two things which are reflecting the same wavelengths
nevertheless have different colours. No doubt scientists would
either disbelieve this, or start to search for other, more obscure
physical differences which cause this variation. But there is no
conceptual inevitability that they succeed: the supervenience of
colour on other (primary) properties is a matter of physical
rather than conceptual necessity. By contrast moral properties
seem to have to supervene upon natural ones in some much
stronger sense. It seems conceptually impossible to suppose
that if two things are identical in every other respect, one is
better than the other. Such a difference could only arise if there
were other differences between them. So suppose that we have a
complete, base description of a thing, B*, telling us everything
that could be relevant to determining its A-state. The superveni-
ence claim is that necessarily if there is a thing which is B* and
A, then anything else which is like it in being B* is like it in
being A as well. There is no possible world in which one thing is
B* and A, but other things are B* and not A (in this it is
important to remember that B* is some complete specification
of the B-states of a thing, whereas A is some particular A-
property - being good to a certain degree, being of a certain
colour, or whatever). Call this B* I A supervenience.
This property now needs contrasting with a stronger prop-
erty, that necessarily, if a thing is B*, it is A. This links B* to A
rigidly - in all possible worlds, if a thing is B*, it is A. It is a
stronger property, for B* I A supervenience does not rule out
possible worlds in which there are things which are B* but not
A. It merely rules out any in which there are both things which
are B* and A, and things which are B* and not A. Thus even if
the supervenience claim holds, it does not enable you to infer
that a thing is A from the fact that it is B*. It just means that
once you know of something which is A and has this underlying
184 LA:\lGlJAGE AND THE vVORLD

state, you can be sure that anything else just like it is A as well.
Call the stronger property B* I A necessity. 2
The point of introducing these two properties is that
philosophers sometimes find it plausible to claim B* I A
supervenience without going so far as B*/A necessity. In
particular in the moral case it seems conceptually or logically
necessary that if two things share a total basis of natural proper-
ties, then they have the same moral qualities. But it does not
seem a matter of conceptual or logical necessity that any given
total natural state of a thing gives it some particular moral
quality. For to tell which moral quality results from a given
natural state means using standards whose correctness cannot
be shown by conceptual means alone. It means moralizing, and
bad people moralize badly, but need not be confused.
How does the argument proceed that this point favours
projectivism? The argument is best thought of by imagining
possible worlds - complete states of affairs corresponding to the
various possibilities involved. If a truth is necessary, then it
obtains in all possible worlds. Otherwise there is one in which it
does not. A contingent proposition, which merely happens to be
so, is true in some possible worlds and not in others. Then the
structure is that possible worlds divide into two sorts. There are
those in which something B* is A, and in those everything else B*
is A also. But there are those in which things are B* without
being A. Now this distribution of possible worlds needs expla-
nation. For at first sight there should be a further mixed kind
allowed- in which some things are B* and A; but in which some
things are like those whose possibility is already allowed - B*
and not A. So we need to explain the ban on mixed worlds, and the
argument goes that anti-realism does this better than realism.
Consider a different example from the moral one. It has been
influentially argued that there can be no 'psychophysical laws',
meaning no necessity that a given physical state B* produces a
given mental state, A'. The reason why there cannot be such
laws is, in Davidson's view, that mental predicates, expressing
2
The contrast is neatly expressed formally:
N( (::Ix) (B*x & A'x)--,> (Vy) (B*y- Ay) ), as opposed to:
N(Vy)(B*y__,. Ay),
the latter being the necessitation of the consequent, when the former only necessitates
the conditional.
 E\' ..'\LCATIO'.'JS, PROJECTIONS 185

as they do a different set of concepts from the physical, answer

to different constraints, so that we would never be in a position
to insist upon the existence of a given mental state, on the basis
of a given physical state. 3 We can leave on one side whether this
argument is forceful (at first sight the same thought rules out
e.g. laws connecting wave length of light and colour, or forces
and motions, or indeed any interesting laws). But Davidson
also believes in the supervenience of the mental upon the physi-
cal. So we have the same structure: we should allow the possi-
bility of B* which might be, for instance, a given pattern of
neurones or brain stuff without A', but given that one B* thing is
A', the others in any particular world must also be. It is as
though God had it in his power to make a given physical state
underlie one mental state - e.g. thinking of my aunt - or
underlie another - e.g. thinking of my dog. But once he had
decided to let it underlie the one state in me, say, then he has to
do the same for you. Whereas at first sight if we allow that he
could associate the states as he likes, we would expect him to be
able to associate the one state with B* in me, and the other one
in you, breaking supervenience. So why is God thus con-
strained? The whole point of talking of possible worlds is to
allow as many as we can conceive of. Equally, ifit could be true,
as far as conceptual constraints go, that Fred has some set of
dispositions and is vicious, but also that he has the very same
set, and differs in no other way except that he is not vicious, then
why could it not also be true that Fred has the set, and is vicious,
whereas Bill has the set and is not? The matter is especially
obvious with physical necessity. If there is no necessity that a
given wavelength underlies a given colour, how could there be a
necessity that if it does in one case, then it does in all the others?
Why the ban on mixed possible worlds?
These questions are especially hard for a realist. For he has
the conception of an actual A-state of affairs, which might or
might not distribute in a particular way across the B-states.
Supervenience then becomes a mysterious fact, and one which
he will have no explanation of (or no right to rely upon). It
would be as though some people are B* and thinking of dogs,
and others are B* and thinking of their aunts, but there is a

3
Davidson, 'Mental Events'.
186 L.\:\GC.\GE AND THE \\'ORLD

ban on them travelling to inhabit the same place: completely

inexplicable. From the anti-realist point of view things are a
little easier. When we announce the A-commitments we are
projecting, we are neither reacting to a given distribution of
A-properties, nor speculating about one. So the supervenience
can be explained in terms of the constraints upon proper projec-
tion. Our purpose in projecting value predicates may demand
that we respect supervenience. If we allowed ourselves a system
(shmoralizing) which was like ordinary evaluative practice, but
subject to no such constraint, then it would allow us to treat
naturally identical cases in morally different ways. This could
be good shmoralizing. But that would unfit shmoralizing from
being any kind of guide to practical decision-making (a thing
could be properly deemed shbetter than another although it
shared with it all the features relevant to choice or desirability).
Supervenience claims are very popular in philosophy, because
they promise some of the advantages of reduction without the
cost of defending B* I A necessity claims. But the promise is
slightly hollow: supervenience is usually quite uninteresting by
itself. What is interesting is the reason why it holds. Thus in the
philosophy of mind, although many writers claim the
supervenience of the mental upon the physical, not so many are
dear about why we ought to accept it, nor about the strength of
the necessity involved. From a Cartesian perspective, according
to which mental properties are logically quite distinct from any
physical ones, and as it were are only accidentally found in
conjunction with any given physical set-up (brains rather than
wood fibres) there would be no right to rely upon super-
venience. So when we rely upon it we are in effect promising an
alternative to the Cartesian picture of mind and body. But the
supervenience claim does not by itself fulfil that promise, and
the pair of properties, supervenience and lack of necessity are
awkward bedfellows in the theory of mind just as they are in
moral theory. If there is no necessity that a given physical
set-up generates a particular mental property, it is hard to force
the ban on mixed worlds, in which it sometimes does and
sometimes does not. And it is not nearly so plausible to take the
anti-realist way out, thinking of our attribution of mental states
to one another in other terms than descriptions (although
Wittgenstein was tempted to an expressive theory, at some
 E\'ALL\Tlo:,,.;s, PROJECTIO'.\"S 187

points in the later philosophy). 4 All in all, then, philosophers of

mind who want to oppose Cartesianism are better off trying to
defend the stronger necessity claims. Merely relying upon
supervenience, and disowning the stronger thesis, seems likely
to leave an unstable position.
It should also be noticed that the relation between theoretical
facts in a science, and the empirical facts which afford evidence
for them, is not like that in the moral case. Suppose we imagine
a total phenomenal description of a world, giving an answer
to every possible question about what is observed or would
be observed under any circumstances. Suppose that this B*
description does not as a matter of conceptual necessity entail
the truth of any particular scientific theory, A: it merely affords
evidence for it, whilst leaving open the possibility of other and
perhaps conflicting theoretical truths, A'. Just because of this
there is no conceptual pressure to suppose that wherever we
have B* we must have the same truth A. Nor is there any reason
to suppose that if one part of a world is phenomenally just like
another, there must be the same underlying A-truths. At least
for a scientific realist, the real underlying A-facts, the theoreti-
cal states of affairs which explain the appearances, might be
different in one case from another. It would be only considera-
tions of simplicity and economy which would make it natural to
hope that they are the same from case to case. Whereas in the
moral case it is not just simpler and more economical to believe
that naturally identical states of affairs compel the same moral
description. It is absurd, contradictory, a failure to understand
the nature of evaluation, to believe otherwise. This damages the
approach to moral language according to which we infer the
existence of moral states of affairs in the same kind of way that
we infer the nature of scientific truths.
This argument for a projective moral theory is in effect a
development of the simple thought that moral properties must
be given an intelligible connection to the natural ones upon
which they somehow depend. It generates a metaphysical motive
for projectivism. The third and last motive I shall mention
comes from the philosophy of action. Evaluative commitments
are being contrasted with other, truth-conditional judgements
or beliefs. This contrast means that to have a commitment of
4
Philosophical Investigations, p. 178.
188 L\:\GCAGE A'.'iD THE \\'ORLD

this sort is to hold an attitude, not a belief, and that in

turn should have implications for the explanation of people's
behaviour. The standard model of explanation of why someone
does something attributes both a belief and a desire to the
agent. The belief that a bottle contains poison does not by itself
explain why someone avoids it; the belief coupled with the
normal desire to avoid harm does. So if moral commitments
express attitudes, they should function to supplement beliefs in
the explanation of action. If they express beliefs, they should
themselves need supplementing by mention of desires in a fully
displayed explanation of action (fully displayed because, of
course, we often do not bother to mention obvious desires and
beliefs, which people will presume each other to have). It can
then be urged that moral commitments fall in the right way of
the active, desire, side of this fence. If someone feels moral
distaste or indignation at cruelty to animals, he only needs to
believe that he is faced with a case of it to act or be pulled
towards acting. It seems to be a conceptual truth that to regard
something as good is to feel a pull towards promoting or choos-
ing it, or towards wanting other people to feel the pull towards
promoting or choosing it. Whereas if moral commitments
express beliefs that certain truth-conditions are met, then they
could apparently co-exist with any kind of attitude to things
meeting the truth-conditions. Someone might be indifferent to
things which he regards as good, or actively hostile to them.
Being moral would need two stages: firstly discerning whether
the truth-conditions are met, and secondly forming desires to
which those truth-conditions matter. But if that were the right
account, morality could be short-circuited. You could hope to
obtain the same actual dispositions, the same choices for or
against things possessing given natural properties, the same
behaviour, the same indignations and passions, by reacting to
the natural features of things. It is an unnecessary loop to use
those natural features to determine a belief in further moral
features, and then to hope for a particular attitude to the
revealed moral features.
Unfortunately this argument is not quite as compelling as it
looks. The relationship between evaluative commitment and
action is subtle. For there is undoubtedly an attitude of "not
caring about morality". As Philippa Foot has persistently
 E\'ALlJATIO'.\iS, PROJECTIONS 189

argued, there is no straight inference from moral commitment

to desire. People often care more about etiquette, or reputa-
tions, or selfish advantage, than they do about morality. So the
moral attitude, if it is an attitude, cannot be distinguished from
belief through an inevitable connection with choice and action.
I ts influence on these things is indirect. But in turn this does not
completely block this third argument. For other attitudes also
have indirect and variable influences on action. Ifl find it wryly
amusing that a colleague is drunk I may channel my sentiment
into a variety of actions or inaction. But finding it amusing is
surely a matter of having an attitude rather than a further
belief, true or false, about my colleague.
To establish projectivism would need a close exploration of
the nature of the attitude which is spread on the world. This
involves locating what it is about an attitude which makes it a
moral one, whether we would be better off without having such
attitudes, what their best replacement might be, and so on. It
would involve discussing whether moralizing is a relatively
parochial habit of people with particular cultural and theologi-
cal traditions, or whether any society whatsoever can be
regarded as holding distinctively moral attitudes. But the
philosophy oflanguage can remain relatively quiet about these
juicy issues. Our concern is whether one range of argument,
starting from the theory of the meaning of moral remarks,
blocks projectivism. The issue is whether quasi-realism (see 5.6
above) is successful in explaining why we can permit ourselves
the linguistic expressions, and the thoughts they enable us to
express, if projectivism is true. For my part I would say that the
success of quasi-realism, as I shall try to present it, leaves a
projective account of morality far the most attractive, on
grounds of economy, of metaphysics, and of the theory of desire
and action. It is, in fact, the only progressive research pro-
gramme in moral philosophy. But why does the philosophy of
language provide any kind of obstacle to it?

2. Frege's Argument
In a very influential article, P. T. Geach used a point ofFrege's
to block expressive theories. 5 The "Frege point" is very simple.
5
'Assertion', Philosophical Review (l 964).
190 LA:\'GUAGE AND THE WORLD

Sentences containing given predicates may occur in utterances

by which we are claiming the predicates to apply, as when I call
something good, true, probable, a cause of something else, and
so on. But such sentences may also occur unasserted, inside the
context provided by other words, making up larger sentences. I
may assert: 'It is wrong to tell lies.' But I may also assert: 'If it is
wrong to tell lies, then it is wrong to get your little brother to tell
lies.' In this latter occurrence the italicized sentence is not
asserted. It is the antecedent of a conditional - in other words, it
is put forward to introduce an hypothesis or supposition. The
Frege point is that nevertheless the sentence means the same on
each occurrence. The proof of this is simple and decisive. The
two sentences mate together to make up the premises of a valid
argument:
It is wrong to tell lies.
If it is wrong to tell lies, it is wrong to get your little
brother to tell lies.
So It is wrong to get your little brother to tell lies.
This is a valid argument, illustrating the general form: P; if P
then Q; so Q. But the argument is only of this form because the
sentence 'It is wrong to tell lies' means the same on each
occurrence. If it did not there would be a fallacy of equivoca-
tion, as in: 'He is working at the bank; if he is working at the
bank he must have his feet in the river; so he must have his feet in
the river.' Here the second premise is true only if'bank' is taken
in a different sense from that in which, we might imagine, it
makes the first premise true, and if so the argument does not
illustrate the valid form. The question now is: how does an
expressive theory explain the identity of meaning? For anyone
asserting the second, hypothetical premise is not expressing an
attitude of condemnation towards telling lies. He commits him-
self to no attitude towards it at all. He just says: 'If telling lies is
wrong ... ' without offering any indication of whether he thinks
it is. Let is call contexts in which a sentence occurs like this, an
unasserted context. Then the question is whether expressive
theories can cope with unasserted contexts in such a way as to
allow sentences the same meaning within them, as they have
when they are asserted.
It is a nice sharp problem. It might seem to provide a swift
 EVALCATIONS, PROJECTIONS 191

refutation of expressive theories. In unasserted contexts no

attitude, etc. is evinced when the sentence is uttered; the mean-
ing is the same as in direct contexts when such an attitude is
evinced; therefore this (variable) feature does not give the
( constant) meaning. But before quasi-realism surrenders it
needs to see whether expressive theories can give any account at
all of these contexts.
There are in fact two distinct aspects to this problem. Firstly,
can we explain what we are up to when we make these remarks?
U nasserted contexts show us treating moral predicates like
others, as though by their means we can introduce objects of
doubt, belief, knowledge, things which can be supposed,
queried, pondered. Can the projectivist say why we do this?
Here he faces two questions. Consider the fact that we can
conjoin evaluations and ordinary expressions of belief: 'It is
wrong to tell lies and your mother is going to be annoyed'. Now
it is surely not surprising that we might link together two
commitments, even if one expresses an attitude, and the other a
belief. The one sentence conjoins the two disparate commit-
ments, and since we often want to communicate that we have
both, it is hardly surprising that we have a way of doing it. That
gives us an idea of what we are up to in offering the conjunction.
But it does not fully answer the second question: why do we
have this particular sentence to serve that purpose. For we might
want to say other things about 'and', which make it difficult to
see why it is serving this function. For instance, we might
explain the semantic function of 'and' like this: it stands
between two sentences to make one large sentence out of them;
the large sentence is true if and only if each smaller one is true.
Otherwise it is false. Now this little semantic theory fits badly,
initially at any rate, with the occurrence of the evaluation as a
conjunct. For suppose, according to the expressive theory, the
evaluation is not susceptible of truth or falsity. Then it should
not mingle with an operator which needs truths and falsities to
work on. But there are ways of easing around this obstacle. One
is to expand the way we think of 'and'. We have to do this
anyway, for it can link utterances when they certainly do not
express beliefs which are genuinely susceptible of truth-value -
e.g. commands: 'hump that barge and tote that bale.' We would
instead say something like this: 'and' links commitments to give
 192 U,:\'GUAGE AND THE WORLD

an overall commitment which is accepted only if each compo-

nent is accepted. The notion ofa commitment is then capacious
enough to include both ordinary beliefs, and these other atti-
tudes, habits, and prescriptions.
So to tackle Frege's problem the first thing we need is a view
of what we are up to in putting commitments into conditionals.
Working out their implications, naturally. But how can attitudes
as opposed to beliefs have implications? At this point we must
turn again to the projective picture. A moral sensibilit.J, on that
picture, is defined by a function. frorri inJmt of beliefto output of·
6
~ , ~ · Now not all suchsensil>iliti_~~-~~~c!B!L~§:~l~:.,§g.me are
.'.:coari?if)insensitive, some are plain horrendous, some are con-
·~rv1'.tive and inflexible, others fickle and unreliable; some are
too quick to form strict and passionately held attitudes, some
too sluggish to care about anything. But it is extremely
important to us to rank sensibilities, and to endorse some and to
reject others. For one of the main features affecting the desira-
bility of the world we live in is the way other people behave, and
the way other people behave is largely a function of their
sensibility. So much is obvious enough. And amongst the fea-
tures of sensibilities which matter are, of course, not only the
actual attitudes which are the output, but the interactions
between them. For instance, a sensibility which pairs an attitude
· of disapproval towards telling lies, and an attitude of calm or
approval towards getting your little brother to tell lies, would
not meet my endorsement. I can only admire people who would
reject the second action as strongly as they reject the first. It
matters to me that people should have only this pairing because
its absence opens a dangerous weakness in a sensibility. Its
owner would have the wrong attitude to indirect ways of getting
lies told (and for that matter the wrong attitude to his little
brother).
The conditional form shows me expressing this endorsement.
Of course, it is ~P endorsement which is itself the expression2(.1
lllOrc1J p9int of view. Some casuistry might lead peopl.e to the
other commitment, that there is a great difference between
telling lies and getting your little brother to do so (am I my
6
Or, more generally, an input of awareness rather than belief. A man may respond to
perceived features without realizing that they are the ones responsible for his reactions.
For example, we olien do not know what we find funny in a situation.
 E\'.\LL\TIO'.\iS, PROJECT!o:s.;s 193

brother's keeper?). 7 Such people would reject the conditional.

But it is quite satisfactorytliaf tfie<::onditional expresses a moral
point of view. The task was. not to show that it does not, but to
-explain what it does at all. Other conditionals have the same
generalrble: - - --
!flying makes you feel good, then it's all right.
If you ought to give him £10, then you ought to give him
something.
I
The latter is held on logical grounds. I can only endorse a
sensibility which, in the presence of the antecedent attitude,
also has the consequent one, because it is logically impossible
that the action specified in th~_ilnte_cedent can be done without
giving ~he man ~omet_l}i11g") _couldnot endorse at all an illogical
sensihili ty, which its elf paired approval of an action with disap~
provaf of a logical implication of the performance of the action:_
The former conditional is the expression of a repuh;ive standard:
it endorses a function from an inputof~~owledge that a lie has
made you feel good to an ou_tput of satisfaction with it. Finding
better descriptions of admirabieinput/outptitTunctions is the
task of moral philosophy.
This account of what we are up to when we use the condi-
tional form with evaluative components now needs supple-
menting by a semantic theory. We can put the need this way.
Imagine a language unlike English in containing no evaluative
predicates. It wears the expressive nature of value-judgements
on its sleeve. Call it Eex· It might contain a 'hooray!' operator
and a 'boo!' operator (H!, B!) which attach to descriptions of
things to result in expressions of attitude. H!(the playing of
Tottenham Hotspur) would express the attitude towards the
playing. B!(lying) would express the contrary attitude towards
lying, and so on. For the reasons I have developed, we would
expect the speakers ofEex to want another device, enabling them
to express views on the structure of sensibilities. They would
need a notation with which to endorse or reject various coupl-
ings of attitudes, or couplings of beliefs and attitudes. Suppose

7
Examples of genuinely controversial commitments of the form may help: 'if
something ought to be done, any means to it ought to be allowed'; 'if a group has been
discriminated against, it is now right to give it better treatment.'
 194 LA:\'GUAGE AND THE WORLD

we talk about an attitude or belief by putting its expression inside

bars: /H! (X) I refers to approval of (X). And suppose we use the
semi-colon to denote the view that one attitude or belief
involves or is coupled with another. Then the speakers of E,x
will need to express themselves thus:
H! (jH! (Tottenham)/;/H! (Arsenal)/)
H! (/B! (lying)/;/B! (getting little brother to lie)/)
The first endorses only sensibilities which, if they endorse
Tottenham, also endorse Arsenal, and this is what we express
by saying that if Tottenham is a good team, so is Arsenal. The
second is our old friend.
Eex will naturally want further constructions. We want to say
things like 'X used to be a good thing, but now it is not', in which
evaluations connect with tenses. Notice that this does not mean
the same as 'I used to approve of Xbut now I do not': it implies
that X has changed, not that I have. The favourable evaluation
attaches to X as it was, so E,, will need a device to express this:
perhaps an index indicating that the past state is the object of
evaluation. Again, consider our different attitudes to our own
attitudes. Since I have the concept of improvement and_J!t~rjg_:
--ratioain a sen~~ibility, Ikno.w that I amvulr1erable to argument
-that in forming ; :i;;i=tic~lar attitud~ tam myself talliiiii!ii:~
Jo ~ fl.,!W_t;9iripiif(ol!t:pt1t functi,C>,tJ:. I may be exhibiting disposi~
tions wllis.h I ~o not endorse, cir committing myself.to p~rings
ofattitu.de q.nd attitude, or attitude and belief,,-whicb, Ialso
cannot endorse, So in some cases I can be uncertain not urrly-of-
\: the fa,c:ts_ of the case, .but of how to react to them. I will need to
.·explore. the~ other aspects of my- moral Commitments, anq_§.e1:-
whf'.ther, when they are brought to bear,one attitudeor"another
begins tQ setde }tself. And when I have taken up an attit~,-L
might be uiicomtortably aware that.it may tµm.~ouJ to be
viilne-rable to criticism. So E,~- will need a way ofsignatling
different degredofrobus'.tn;~siin ourattH~!i~~:~<:Iifterent ways in
which they can be regarded asHkeiy or ~nlikely to succ_t1mb to
an,in1pr(:>Ved perspective. H!(X) can co-exist with~s~mething
like ? H! (jH! (X)/): uncertainty that one's own attitude of
approval can itself be endorsed.
E,, will be spoken by people who need to signal and respect
consistencies and inconsistencies. Consider:
 E\'.\LL\TIO'.';S, PROJECTIO:'\S 195

B! (lying)
H! (IB! (lying)l;IB! (getting little brother to lie)I)

Disapproval of lying, and approval of making (disapproval of

getting little brother to lie), follow upon (disapproval oflying).
Anyone holding this pair must hold the consequential disap-
proval: he is committed to disapproving of getting little brother
to lie, for if he does not his attitudes clash. He has a fractured
sensibility which cannot itself be an object of approval. The
'cannot' here follows not (as a realist explanation would have it)
because such a sensibility must be out of line with the moral
facts it is trying to describe, but because such a sensibility
cannot fulfil the practical purposes for which we evaluate
things. Ee, will want to signal this. It will want a way of
expressing the thought that it is a logical mistake that is made, if
someone holds the first two commitments, and not the commit-
ment to disapproval of getting your little brother to lie.
In short, Ee, needs to become an instrument of serious,
reflective, evaluative practice, able to express concern for
improvements, clashes, implications, and coherence of atti-
tudes. Now one way of doing this is to become like ordinary
English. That is,it_ would invent a_predii:ctte an~weriogJg die
.:t~tth!~:::<U!il tr~.<t.L~Oplllil££1~~~: as 1f they.iere judgeIDents,
ang. tnen .llS_~.<1jlJ:J:i_~_na!ll,i:_ajc:lsyic1:~faf cleb<1Jing_trnJh_,Jf this is
right, then our useofjt1cli.rect contexts does nQt proyg th<tt an
.. expressive ihe~ry ofrnorality i_s wrong;_it merely proves uido
have adopted a form of expression adequate to our needs... Inis.is... .•
w/tat.is1JJ.rn11Lby_'p..rojecting'pJ.tit11:dJLon(Q_J/te,.wo.rJd.
What I have done here is to explain how conditionals can be
regarded as ways of following out implications, although it is
not imperative that the commitments whose implications they
trace have 'truth-conditions'. Now you might say: even if this
can be done, hasn't the quasi-realist a very dreary task in front
of him? For remember that the Frege point was entirely general:
it could cite any unasserted context. So mightn't others arise
which require separate and ingenious explanations, and is the
quasi-realist faced with an endless task? Isn't he like Ptolemaic
astronomers, having to bolster his theory with ever more com-
plex or ad hoc epicycles, whereas by comparison there is a
simple, common-sense view that moral predicates are just the
196 L.\:\"GCAGE A:'\iD THE WORLD

same as more ordinary ones, so that there is nothing to explain

about the way they function in unasserted contexts? 8
Questions of what does or does not require explanation
involve delicate matters of philosophical judgement, but the
objection here is surely overdrawn. For what plays the role of
Copernicus to the allegedly Ptolemaic complexities? What was
wrong with Ptolemaic astronomy (by the end of its reign) was
that there was a better way of explaining the sam~ things. But
this better way did not just take those things for granted. It was
not the stultifying position that everything is just in order
without our bothering to explain it. We have seen enough of
why projectivism is a plausible moral philosophy. And this
being so it is extremely important to tell whether it is blocked by
arguments from the philosophy of language. Nobody denies
that the surface phenomena of language - the fact that we use
moral predicates, and apply truth or falsity to the judgements
we make when we use them - pose a problem for projectivism.
This is why they tempt people into realism. But by overcoming
the problem projectivism also steals a march on its rivals. For it
removes the temptation to think that our surface forms of
expression embody a mistake, that they are "fraudulent" or
"diseased": it protects our ordinary thinking, in a way that mere
reminders of the way we do actually proceed cannot do. It
solves Kant's question of the right to our concepts, as well as the
question of what they are actually like.
But now a new and rather surprising vista is opened. For if
this is right, might 'it is true that . . . ' also be given this
quasi-realist explanation? Initially an expressive theory stands
in stark contrast to one giving moral remarks truth-conditions.
But if we sympathize with the pressures I have described, we
come to appreciate why it should be natural to treat expressions
of attitude as if they were similar to ordinary judgements. We
come to need a predicate, whose behaviour is like that of others.
Why not regard ourselves as having constructed a notion of moral
truth? If we have done so, then we can happily say that moral
judgements are true or false, only not think that we have sold
out to realism when we do so.

• ·!'his response was made to me by Professor Geach.

 E\'.\LL\TIO:\'S, PROJECTIO'.':S 197

3. Constructing Truth
The arguments of the last section may give us a right to a notion
of an improved set of attitudes; they give us some right to a notion
of the coherence and consistency of such a set. But do they suffice to
build all that we need from a conception of truth, applicable to
moral judgements?
The root disquiet here runs very deep. In effect, quasi-
realism is trying to earn our right to talk of moral truth, while
recognizing fully the subjective sources of our judgements,
inside our own attitudes, needs, desires, and natures. The sense
of subjectivity triggers all kinds of wild reactions. Can the
projectivist take such things as obligations, duties, the "stern
daughter of the voice of God", seriously? How can he if he
denies that these represent external, independent, authoritative
requirements? Mustn't he in some sense have a schizoid atti-
tude to his own moral commitments - holding them, but also
holding that they are ungrounded? And when the tension comes
out, shouldn't he become frivolous, amoral? A recent influential
book even believes that an emotivist should approve of manipu-
lating people, bullying and lying and brainwashing as we
please, rather than respecting their independence. 9 Words like
'relativism' and 'subjective' focus these fears; books and
sermons alike pronounce that the projectivist sp.ould, if consis-
tent, end up with the morals of a French gangster.
Fortunately, all this is ridiculously beside the point. Just as
the senses constrain what we can believe about the empirical
world, so our natures and desires, geeds and pJ~asures,. co11~
st:rain'muc;h ofwhatwt:san admire ang commend,-tole:X:atea;;.d
work for.•·· There are not so many livable, unfragmented,
cf~\/elpp~Q., con~istent, ang i:;pl;i~rent systems ofattitude. A pro-
-~jecifvist,.likttanybtrease;may:be~s~ri~tffve'fo-the features which
make our lives go well or badly; to the need for order, contracts,
sources of stabil1ty. If his reflection on these things leads him to
endorse a high Victorian love of promises, rectitude, contracts,
conventional sexual behaviour, wen·and good: there is nothing
in his meta-ethic to suggest otherwise. For instance, a proper
respect for promises, the kind of respect which sees them as
9
A. Macintyre, After Virtue, p. 22, and throughout.
 I
198 L.\:\GUAGE A'.\'D THE WORLD

making requirements, as bounds on conduct, is certainly a good

attitude to foster. But it may, for all that, be just that: an
attitude.
The problem is not with a subjective source for value in itself,
but with people's inability to come to terms with it, and their
consequent need for a picture in which values imprint
themselves on a pure passive, receptive witness, who has no
responsibility in the matter. To show that these fears have no
intellectual justification means developing a concept of moral
truth out of the mater_ials to hand: seeing how, given attitudes,
given consrfafrits- upon them, given a notion of improvement
a~d-ofpossib_leJault in any~e!!_sibili_ty including our own:-we
can-co9stn£c;-t a notion oftrnth. The exercise is importarrcFc5r-
""'onemoral of the brus}1 '\V1tnGoodman's paradox and the rule-
following e<;msiderations (chapter 3) is that judgement never
involves quite the pure passivity which is supposed to be an
untainted source of objectivity and truth. We have to see our
concepts as the product of our own intellectual stances: how
then ar<':_they suitable means for framing obj~ctivel)'- corr-ect,
true_,jt1_dg,ement, describing the mind-independent world as it
in fact is? It is not only moral truth which starts to quake. But
we can learn how to approach the general problems of truth by
starting with it.
The simplest suggestion is that we define a 'best possible set
of attitudes', thought of as the limiting set which would result
from taking all possible opportunities for improvement of atti-
tude. Saying that an evaluative remark is true would be saying
that it is a member of such a set, or is implied by such a set. 10
Call the set M*. Then if m is a particular commitment,
expressing an attitude U;
mis true = U is a member of M*
To test this suggestion we must find conditions which truth
obeys, and see whether they square with it. In particular, does
the definition justify the constraint of consistency (m cannot be
true and false)? The first hurdle is to define the idea of a unique
best possible sensibility. Certainly there is improvement and
deterioration. But why should not improving sensibilities
0
' Although this is the simplest projectivist account of truth, and is one used by many
anti-realists, I do not myself think it is the best. It is only a first approximation, but
serves to make the immediate points. See below (7. 7).
 EVALGATIO\;S, PROJECTtOl'liS 199

diverge in various ways? An imperfect sensibility might take

any of several different trajectories as it evolves into something
better. We might imagine a tree (see Fig. 6). Here each node
(point at which there is branching) marks a place where equally
admirable but diverging opinion is possible. And there is no
unique M* on which the progress ofopinion is sighted. So there
is no truth, since the definition lapses. More precisely, truth
would shrink to only those commitments which are shared by
all the diverging systems: truth belongs to the trunk.

----
----
----
-----
----
----
Fig. 6
---
This is the deep problem of relativism. It is not the vague and
unfounded disquiet that I have no right to judge unfavourably
people with any other opinion - those who practice human
sacrifice, or murder Jews, for instance. Of course I have. My
attitudes, and those involved in any system r could conceive of
which rriighihe_superior to mine, alike corid~mh them. Thr
deep problem is the suspicion. thatother, equally_.adrriirable ..
sensibilitiq,, over \Vtli~li I c<1;.ri.cliiirn no superiority of my own, .
lead to div~rgeliTj:g~g~~nts. This does take away my right to
tfiink-of mine as true, which is equi\!aJent to uii§ettling my
COJ!):mitm~tn-. --
The classic introduction of the problem is Hume's superb
and neglected essay 'Of the Standard of Taste'. He introduces
"such diversity in the internal frame or external situation as is
entirely blameless on both sides". He illustrates it with the
difference between "a young man, whose passions are warm"
and "a man more advanced in years, who takes pleasure in
wise, philosophical reflections concerning the conduct of life
and the moderation of the passions". The former prefers the
amorous and tender images of Ovid, the latter the wisdom of
'.200 L\:\GUAGE AND THE WORLD

Tacitus; in the twentieth century no doubt the one reads Playboy

and the other The Economist. Now we imagine the young man's
literary sensibility improved and refined into a system M\ and
the old man's into a system M\, and one containing the com-
mitment that Ovid is a better writer than Tacitus, whilst the
other contains the reverse. How can we recover a notion of truth
for either judgement? Hume's answer is subtle: it consists in
expanding the difference between a (mere) matter of predilec-
tion, and a fit object of literary comparison: "it is plainly an
error in a critic, to confine his approbation to one species or style
of writing, and condemn all the rest." The point of this emerges
if we ask the question: who is to adopt the following three views:
(I) M\ - it is true that Ovid is better than Tacitus
(2) M\ - it is true that Tacitus is better than Ovid
(3) There is no possible improvement on either M\ or M*,
(where 'M* - it is true that m' means that M* contains an
attitude which is expressed by the sentence m).
How can you have each of these? Think of the detail. Is
someone who has either of the two sensibilities aware of the
other? If not, it would surely be an improvement ifhe were. But
if so, then quite what is the combination of attitudes required?
It surely begins to seem wrong to hold, straight out, that Ovid
is bettec than Tac1tus; or vice versa. If we admit aJ~~Ily--
develop_ed._Jnr~T--'YiY. of looking .at. it, are we not required to
soften the opposition - to say, for instance, that Ovid and
Tacitus are of equal merit, although each has features which
appeal to different people? In short, as soon as I hold that a case
begins to look as though the tree structure applies, I also hold
that there is a truth about the subject on which the divergent
attitudes are held, and, holding that, I would also judge that
one or both of the rival sensibilities is capable of improvement,
until it yields my own attitude. Hume's case depends upon the
audience accepting that Ovid and Tacitus actually are of equal
merit - otherwise it would not be just the different tastes of
different ages which result in the different rankings. But if we
think this, then each of the two systems is flawed. The young
but able literary critic is insensitive to the virtues which appear
to the older man, and which in truth result in the two writers
coming out equally; similarly for the older literary critic.
 E\',\LL\TIO'.\S, PROJECTIO'.\lS 201

What does this mean? It means that an evaluative system

should contain the resources to transcend the tree structure: evi-
dence that there is a node itself implies that it is wrong to
maintain either of the conflicting commitmen(s. It is itself a
signal that the right attitude - the truth abo4t the relative
merits of Ovid and Tacitus - is not that expressed by either of
these partial perspectives. The better perspective may judge the
merits equal, or it may award the prize to just one view, or it
may regret and change the terms of the discussion, by losing
interest in the simplistic question of whether Ovid is a better
writer than Tacitus, and concentrating upon different merits of
each, with no intention of finding a summary comparison. In
that case the system of each of these literary critics is defective,
by containing too many crude comparisons. So in practice
evidence that there~JLnode-is·just treated as a signal that the
truth is not yet fi11ally argued, and it goes into discussions as
part of the evidence: We are constrained to argue and practise
. as tlio"tigh the trufh"is single, and this constraint is defensible iri
spite of the apparent possibility of the tree-structure.
It is as though the trunk - the core of opinion to which there is
no admirable alternative - contains the power to grow through
I any of the choices of opinion which lead to branching: the
·. choices become themselves part of the knowledge which the
progress of attitude must use to form its course. 11 In so far as
acquaintanc~--~jJhanother value,-systen1 makes me respect it,
di.en itproper:ly.rn.akes me rethink both systellls,transcending ·
the tree-structure.\ · ··· -·· .
--onecoriimon element of "relativism" is the thought that if
we can conceive of different, equally admirable systems, then
that must weaken our own in favour of toleration. It seems to be
particularly the attitudes associated with obligation which are
vulnerable to worries about rival systems. The explanation of
this is quite pleasant, although perhaps slightly peripheral to
language.
Suppose we symbolize 'it is obligatory to do A' by OA, and 'it
is permissible to avoid A' by 'P1A'. P1A contradicts OA. Now
1magme:
11
Readers may be reminded of the Hegelian three-step (look left, look right, and
then cross over) of thesis, antithesis, synthesis, by which we lurch towards the
.\bsolute.
202 LA:\'GUAGE AND THE WORLD

(l) M\----,. OA
(2) M* 2 ____,. P1A

(3) There is no possible improvement on either M*, or M*z.

This is the tree structure. One way of transcending it goes:
(4) ( 3) implies that it is permissible to hold M* 2 ; this implies
that it is permissible to hold that I A is permissible,
which in turn implies that 1A is permissible;
(5) So any view, such as M* 1 which implies the reverse, is
wrong, and ipso facto capable of improvement.
This transcends the tree by showing that ifwe suppose that the
choice leading to M* 2 is permissible, then we must dislike the
other. We cannot maintain the even-handed, Olympian stance
which finds them each blameless. But now notice that if we have
a primafacie case of ( 1), (2), and (3), it is likely to be the obligation
which is the more threatened. Suppose we try to transcend the
tree by the reverse argument:
(4') (3) implies that it is permissible to hold M\; this
implies that it is permissible to hold that A is obligatory,
which in turn implies that A is obligatory;
we need the different reduction principle that ifit is permissible
to hold that A is obligatory, then A is obligatory, and that is not
nearly so intuitive. 12 Evidence of a permissible system which
permits , A is evidence that I A is permissible; evidence of a
permissible system which obligates A is not so plausibly
thought of as evidence that A is obligatory. So here, once we
have an initial inclination to each of the three propositions, we
are more likely to escape the inconsistency by seeing M* 1 as
capable of improvement.
So on this account it is correct, as well as natural, to find faith
in particular obligations shaken, if what seems to be admirable
sensibilities do not recognize them, This is why travel
broadens the mind.
12
· Formally the reduction principles difler in a way analogous to the diflerence
between the modal system 54 and the stronger system SS. lf'it is permitted to permit A'
implies 'A is permitted', we have PPx--,. Px, which corresponds to S+. This does not
demand that 'it is permitted to obligate x' implies 'xis obligatory', for which SS is
needed. The comparison is suggestive, and indicates a respectable side of worries about
relativism.
 E \'.\LL' .\Tl O:\'S, PROJ ECT!O:\'S 203

4. Interlude: Bivalence, Fiction, Law

Because of the writings of Michael Dummett it has become
common for people to think that an anti-realist will not accept
that every commitment in the relevant area is either true, or
false. The idea is that only a realist can adopt such a logical
principle. For the construction of truth suggested might evi-
dently leave it that M* neither implies that m, nor that mis false,
in which case on the definition m is neither true, nor false. It is
worth exploring this connection a little. Dummett associates his
views about the justification ofbivalence with a quite different
problem which he believes he has discovered for realism - the
problem qf what it is to 'manifest a grasp of' sentences whose
truth-codditions in some sense outrun our capacity to verify
whether they obtain. I mentioned this in chapter 2: it all
depends on the manifestee. But the connection ofbivalence and
realism is independently interesting. In particular it introduces
an intriguing possibility - that a conception of truth underlies
logical principles, so that ifit is abandoned, they must be so too;
converse-ly, adopting the logic is in itself a commitment to the
conception.
We must first distinguish two possible views about falsity:
mis weakly false = it's not the case that mis a member of M*
mis strongly false = the attitude expressed by 'not m' is a
member of M*
To see the difference between these, remember the remarks
about the ways in which improved theories may relate to their
predecessors, not by containing direct denials of theses previ-
ously held, but by transcending those terms of discussion. For
example, suppose then that m is a proposition that there is a
right to strike. And suppose that we believe M* to be likely to
have abandoned the concept of a right altogether - perhaps for
Marxist reasons, it sees the concept as part ofa way of thought
about human affairs which is itself defective and due to be
transcended. Under the first definition it is then false that there
is a right to strike: this commitment is no part of a true system.
But under the second definition it is not false that there is a right
to strike. The true system does not contain 'there is no right to
'.204 LA'.\IGUAGE A'.\ID THE WORLD

strike' either. It simply does not speak in those terms. This is

actually quite a common reaction, on a small scale, in the face of
some kinds of evaluation. It's not that we take up the attitude
expressed by directly contradicting someone's commitment,
but we reject the terms of the discussion. If someone asserts
thatBeethoven is a better composer than Brahms I might not
respond by asserting that Brahms is a better or equal composer,
but by regretting these simplistic comparisons. Certain kinds of
tendency to see everything in moral terms are best met by
de-moralizing issues, rather than taking up opposing moral
attitudes (although one of the subtleties here is that a wish not
to discuss a choice in moral terms is itself describable as a moral
attitude, and as argued above, usually one in favour of tolerance
and permissions).
A useful example to distinguish the first and second defini-
tions of falsity is that of fiction. We can assert that Hamlet had
this or that feature: he had a murdered father, a difficult family
life. He did not have a peg-leg or a harem. What do we mean by
regarding the first two remarks as correct, and the last as
incorrect? They are grounded in Shakespeare's text in some such
way as this: we imagine a character conforming to the way
Shakespeare describes or shows Hamlet, and if we are compel-
led to regard him as cp, then it is true that Hamlet is cp; ifwe are
compelled to regard him as not cp, then it is wrong to say that he
is cp, and false, under each definition, that he is cp. The notion of
compulsion here is not purely mechanical or formal: it can be a
matter of sensibility to realize that the texts compel a particular
reading of Hamlet's character, meaning that seeing him in
accord with that reading best fits or explains the course of the
drama. If we call the grounding text G, and the relevant notion
of compulsion - , then we can have

G- p = It is true thatp
G- ,p = It is false thatp (on each definition)
"\
But we also expect cases where neither obtains. The texts do not
compel us· to regard Hamlet as having a nice light baritone, nor
as not having one: neither proposition appears in any admirable
list of truths about Hamlet. So is it false that Hamlet had a nice
light baritone? It is weakly false. But it is not strongly false: the
 E\',\LC.\TIONS, PROJECTIONS 205

proposition that Hamlet had some other kind of voice appears

in no such list either. The distinction between weak and strong
falsity matters whenever the grounding, in virtue of which
propositions are true or false, 1is potentially incomplete, failing
to yield just one verdict on ev€ry question.
Consider now truth in law. A contract is valid if the law
makes it so. But law is a matter of human arrangements,
actions, and decisions. The contract is valid if these - the
grounding - compel the judgement that it is. Jurisprudence is
weighed down by wrangles over quite how much may be added
to the grounding, and quite, how much can be read into the
notion of compulsion. But this does not matter, for in any event
we have the same structure as in fiction. Ifp = 'this contract is
valid', and 1p = 'this contract is not \:alid':
(1) Tp = G (p) compelsp
(2) Tip = G (p) compels 1p
where G(p) represents the sum total of circumstances, prece-
dents, statutes, pertinent to the issue, and compulsion is
interpreted according to the best view of how those things
should be used. But it now seems possible that neither p nor 1p
is compelled by the grounding. Perhaps the arrangements of a
society have simply not "got up to" a certain issue, either by
providing directly for it, or providing indirectly, by laying down
what is to be decided when direct provision fails. Is a defamat-
ory naval signal libel (written) or slander (spoken)? Is an
ice-cream van a place of sale? Is a prisoner-of-war a resident in
an alien country? We might well suppose that even if these
examples are determinate, there will be others where grounding
fails, and the verdict is, as we should say, a matter of discretion.
In the terms so far suggested, if p is in this way a matter of
discretion, each of p and 1p is weakly false. But one might just
as well call each of them weakly true. Weak truth wou1d cor-
respond to the idea that judgement in favour ofpis at least not
incorrect:

F"p =Twp= 1(G(p) compelsp) & 1(G(p) compels 1p)

In these circumstances 1p is weakly true as, well; and each is

weakly, but not strongly false. The natural way of symbolizing
206 L:\'.\iGUAGE AND THE WORLD

this state would be Dp-discretionary whether p-meaning that

the issue is one to be settled by a decision of a court which could,
so far as the proper use of the grounding goes, fall either way.
The question now asked is: if all this is right, why is any legal
system bivalent or polar, admitting just two verdicts (it is true
that the contract was valid, it is false that the contract was
valid)? Why don~ we characteristically find three sides in each
dispute: one arguing for the judgement Tp; one arguing for Fp;
one arguing that Dp? If this question has no answer, then the
fact that systems are polar should be surprising. Why, then, do
we find so little flicker in the smooth determination oflawyers to
regard every contract as either valid or not valid? Literary
critics put up with incompleteness and the absence of strong
truth or falsity about Hamlet's voice. And if two of them
mounted a Mad-Hatter debate about it, the most respectable
view would be a third advocating that the matter is un-
grounded. Why not at law?
Professor Dworkin supposes that these questions force us to
review the view that law is grounded in human arrangements in
such a way as to invite incompleteness. 13 The ingredients which
Dworkin brings to plug the gaps are twofold. Firstly, as well as
"black letter" statute and precedent, we must consider the
political theory which best justifies settled law. If that theory
suggests an answer one way, then even when hard legal facts
give out, then that answer is correct. The other ingredient is
determinate moral truth. If on each issue just one side is morally
in the right, the verdict should go to him. Neither ingredient
escapes the problem. The shape which M* is likely to take will
leave it quite silent on many issues at law (do Right and Truth
attach only to the cause which sees an arrangement of flags as
libellous?). And in addition there is often no such thing
as the political theory best justifying settled law. In many
jurisdictions the political theories of people starting the
patchwork would be virtually unintelligible to us, as ours would
be to them. And the particular conceptions of the values, rights,
etc. enshrined in any theory, even when it is explicitly part of a
constitution, shift and bend and change: the tree-structure
threatens again. But if we cannot regain completeness, how do
we answer the questions?
13
R. Dworkin, 'No Right Answer', in Law, lll!orali!J, and Socie!J, ed. Hacker and Raz.
 EVALUATIONS, PROJEGTIOJ 207

arrange things so that the law is to deliver one of two

verdicts: p or 1p. Why do we not allow a judge to shelter behind
incompleteness, able to deliver the answer 'Dp'? Because Dp
answers nothing: we need to know (for instance) whether the
contract is valid, and one party to pay the otlfer, or invalid, and
vice versa. For this pragmatic reason a judge must think and
argue as though there is one proper verdict, even if the overall
theory of grounding licenses no such optimism. So some con-
stitutions explicitly rule Dp out; the Swiss legal code, for exam-
ple, dictates that a judge shall not shelter behind insufficiencies
of the law. 14
Disallowing Dp as a verdict is therefore explicable. We
wouldn't expect this crack in bivalent practice. We might
expect others. A system could make place for a third answer
(the contract was middling: neither party wins or loses quite as
much as if the case had gone a definite way). But having three
answers, or five, or indeed any number, is not the same as
allowing Dp to stand as a verdict. Consider a case hinging on
whether a man was of sound mind when he made his will, in
which case I benefit, or of unsound mind, in which case the
Royal Corgi Society does. A court may have only two verdicts at
its disposal. Given the spectrum of soundnesses of mind, it
might be better if it had many, and was empowered to appor-
tion the money according to the degree of lunacy. But things
would not be at all better, under either system, ifit was entitled
to announce that the issue is discretionary, and then walk off the
site.
It is then explicable that a court should need to practise as
though there is one truth, one correct verdict, out of however many
it is empowered to give. It is therefore no argument against the
equations ( l) and (2) that it so practises. We have here a
pragmatic argument for practising as though there is a legal
truth. But shouldn't this induce schizophrenia in a judge, ifhe
has to practise as though this is so, although his background
philosophy allows failure of grounding? Perhaps if he was
operating a really incomplete, poor system, it should. But two
features, also relevant to the moral case, apply to diminish the
difficulty. The first is that legal and moral reasoning are
sufficiently open-ended for it never to be a n;iatter of certainty
14
I owe this information to H. LA. Hart.
208 LA'.\TGUAGE AND THE WORLD

that we are at a node in the tree-that a proper decision could go

either way ( this is why I have talked only in terms of evidence that
we are at a node). So it seldom becomes rational to stop arguing
as though there were just one right answer, which progressively
more generous and engaged views would embrace. The second
is that we often stack things so failure of grounding cannot
apply. We can do this by putting the "burden of proof" on one
side. If that side fails to get a grounding, then the negation
automatically does, and completeness is restored. This is paral-
lel to the moral structure about obligation and permissions. If
M* does not oblige us to refrain from A, then it permits A. There
is no way in which it can remain silent on whether A is
permitted.
These two points are merely suggestive. There will doubtless
remain cases where a judge will regard his verdict as discretion-
ary, as one on which reasonable men may have differed, and
where the controlling idea of one legal truth is a fiction. It may
be that if he so regards a verdict, he ought to feel self-conscious
about delivering it, as though he is claiming there to be one
right answer, when there is none or are several. But this degree
of discomfort is one that judges manage to tolerate. Notice that
moralists do not: in this situation, like Hume's literary critics,
they have the option of ceasing to regard the issue as a moral
one.
We have been dealing with cases where one type of proposi-
tion has its truth grounded in other facts. Legal and fictional
judgements are good examples; on the projective theory, truth
grounded in membership of M* is another. If there can arise
incompleteness in grounding, there will be propositions which
are not true, and not strongly false either. There are other
subjects where the same structure suggests itself. Arithmetical
truths gain their status as something more than mere
formalisms because numbers matter: they are what we count
and measure with, and because of this we are not free to invent
propositions about them as we wish. But it is quite unclear: how
far this constraint goes. It seems to compel just one ordinary
arithmetic. But mathematicians can devise systems which con-
tain conflicting accounts of how many infinite numbers there
are, and how they are packed. Is one of these true, and the others
not? Perhaps the correct account of how mathematical truth is
 E\'.\LL\TIO'.\iS, PROJECTIO:\"S 209

grounded in practice shows that the grounding does not extend

so far, and that we have here only a matter for choice. Or
perhaps it does not. Again counterfactuals are types of proposi-
tion, about what would have been the case if things had been
otherwise, which seem to have their truth grounded in facts
about the real world (for more on this see the next section). But
this grounding too might fail: there may be no fact of the matter
whether if I had thrown this coin yesterday, it would have
landed heads, or would have landed tails. Perhaps many such
assertions are only weakly false (or weakly true). But again we
may need to practise as though there is one genuine truth to be
had, if we can find how to argue it.
It is tempting, because of these structures, to say that only a
realist can commit himself to strong bivalence (every proposi-
tion is strongly true or strongly false). He imagines (and
perhaps it really is just an image) a determinate legal, moral,
arithmetical, counterfactual, etc. fact existing quite indepen-
dently of any incompleteness in grounding, filling up the gaps.
But the matter is more subtle than that, on two counts. First, we
have seen how a system can be organized so as to transcend the
tree-structure by putting the onus on one side: unless there is a
proof of p, the proof of an absence of one counts as a proof 9Clp.
This is the likely structure with obligations and permissions, as
I have argued. If M* spoke only about obligations and permis-
sions, we could presume it to obey strong bivalence, in spite of a
projective or anti-realist account of what we mean when we so
talk. However, this point by itself will not force strong biva-
lence. For (as with Hamlet's voice) p and 1p are often
symmetrical, so that there is no reason to describe the fact that
the ground does not compel one, as itself a compulsion of the
other. The second point is the pragmatic need to practise as
though there is just one truth, as in the case of law. A judge
should regulate his practice like this: in every case that comes
up he must proceed as though the grounding compels p, or the
grounding compels 1p. This bivalent practice is earned on an
anti-realist basis, and itself no sign that legal reality is any more
complete than the nature of the grounding suggests. This
means that it is wrong simply to identify bivalent logical practice
with acceptance ofrealism, for it can be explained and justified
on a more economical basis. When we judge, with the aim of
210 LANGUAGE AND THE WORLD

g1vmg assent to p or assent to its negation, we will practice

under the umbrella that p is true, or p is fals~, just as a court
must do. But this practice is quite consistent with realizing that
each might be a half-truth, or that perhaps no facts exist to
ground just one option.

5. Other Anti-Realisms: Causes, Counterfactuals, Idealism

Let us get back to projectivism.
Apart from its own interest as an account of moralizing, there
is one supremely important point about a projective theory.
Suppose the quasi-realist earns us the right to put our commit-
ments in the terms we do. It is just that the explanation of this
right starts from the subjective source, in attitude and senti-
ment. It follows that the issue of whether projectivism is correct
is not readily decidable. That is, you cannot rely upon first
thoughts, or immediate armchair reactions, or your unproces-
sed knowledge of what you mean by some commitments, to
determine whether the theory is true. This alters the terms in
which anti-realism needs discussing: in particular it alters it
from those in which the analytic tradition has usually tried to
discuss it. That tradition has persistently tried to make the
anti-realist a reductionist, bent upon giving an account ofall we
mean by various commitments; from the armchair we can
announce that we really mean more (and this can be sup-
plemented by arguments that we do), and then this is seen as a
victory for realism. Whereas now there need be no attempt to
deny the distinctive nature of the commitments, and the unique
meaning of the various vocabularies, and this still leaves open a
projective theory of what is true ofus when we use them. This
implies that the pure philosophy oflanguage has less to offer to
such problems than most recent discussions assume. Neverthe-
less such a return to an explanatory mode, away from an analytic
mode, is essential, especially if we are to understand the history
of philosophy rightly.
Consider for instance Hume's treatment of causal necessity,
perhapsthe cJas§icprojeqive theory inphil()sophy. The central
-thought is thatdignifyirig-a''relationship'5etween events as
casual ~s spreading ,()LP,r:()j~~-ting,, a ,reaction whic:h., "",e-~<1._y_e_!o
somethmg else.~ wt: are awar_e of about the events - Hume
 E\'.\LL\TIO:\S, PROJECTIO:\S 211

thought of this input in terms of the regular succession ofs~milar

such events, one upon the other. Exposed Jo such regularity,
our minds ( cannot help but) form fiabits of expectation; which
they then project by describing the one event as causing the
other. We are more subtle than Hume suggests: we take more
into account than regular succession of similar events. 15 But
this does not matter to the overall nature of the theory. The
essential anti-realist core is quite missed by counterexamples
where we think that A causes B, although B's do not regularly
follow A's (e.g. because of countervailing circumstances), or
where B's do but we do not think that A's cause them (e.g.
because some third thing causes them both). We may both be
subtle and active in choosing to dignify various relations as
causal: it is still possible than when we do we give expression to
a habit of reliance, a habit which we depend upon in inference
and in other practice, and which we project when we insist upon
the existence of causal connection. Again, since we have a
purpose in so projecting we will have standards by which to
assess the evidence we use for the existence of causal connec-
tions, and the quasi-realist can again earn a right to the notion
of truth, and a notion of the true causal structure of things.
Since this is the basic structure of Hume's theory, it follows
that he has been sqamefully abused by many commentators
and their victims. He is not denying that there exist causes; he is
not inconsistent when he says that there are unknown causes of
things; he is not concerned to say that causal propositions can
be analysed into ones about the regular successions of events,
which then capture their entire content. He is merely explain-
ing our normal sayings, our normal operations with the con-
cept, in terms of the reactions we have, after exposure to a
reality which exhibits no such feature. The explanation may be
defective; even in principle it may not work, but at least it needs
tackling for what it is. We fail to do this if we insist on seeing it as
a meaning equation ( consider again my strictures against
naturalism in ethics). We also fail to do it if we merely insist, as
many thinkers do, that we properly describe the perceived states
of affairs in causal terms - see bricks splashing in water, balls

15
Although Hume is more careful than often realized: Treatise, Bk. I, pt. II I, sect.
xv.
212 LANGUAGE AND THE WORLD

breaking windows, things pushing and pulling. Certainly we

do: but what is the best explanation of our so seeing them?
Amongst the questions wh_ic:h g~ri,Yi!!~ly;u:ist the most promi-
nent will be the kind of conception of a non-causal reality whose
features. explain our. reactiC>r1s/ Hume thought in terms of a
0

regular~succession:of event{ desc;ibable in purely phenomenal

terms. But is this even a possible conception of·the world?
Perh.aps we have to use the concept of a causal connection to
describe the world at all (ordinary descriptions involve things
with all their powers). In that case there is no way to explain our
causal sayings as projections generated by something else, for
there will be no stripped, B-vocabulary in which to identify the
something else.
These questions loom still larger when we get to Kant, and
the whole description of the world as spatially extended and
temporally ordered is seen as the work of the mind, whose only
exposure is to a noumenal, inaccessible reality of things-we-
know-not-what. For the picture at the beginning of this chapter
defines one of the central problems of philosophy. How far does
the projected reality extend: what is our creation and what is
given? When the spread world drips down entirely over the real
world, whose nature is blotted out and left completely
"noumenal", we have idealism. Before rushing to embrace it,
there are two points to remember. One is that respectable
projectivism is an explanatory theory. But in so far as we lack a
conception of the noumenal reality on which our reactions
depend, we also fail to explain how its features are responsible
for our reactions. A particularly gross instance of this general
problem is Kant's removal of the noumenal reality from space
and time. For there is no prospect at all of an explanatory
account of how our reactions depend on the features (events?
changes?) of such a reality: depends is a causal notion, and at
home only within the familiar world. 16 Secondly, in my develop-
ment, projectivism needs a sense of the role of a saying
contrasted with simple expression of belief - in the case of
evaluations, a role of expressions of attitude, or in the case of
cause, a role as dignifying a regularity for some theoretical or
practical purpose - and this is the foundation for the belief-
'" Kant is well aware that this is only a form of explanation and that we'can;Em:-IJ;-
own grounds, do nothing to fill it out: Critique qf Pure Reason, Bl 46, Bl 66, footnote.
 E\',\LL\TIO'.\JS, PROJECTIO:'!S 213

apViar,aris:~., . But describing a world in spatial and

term~ 1.s JUSt that - describing. These two points
suggest that each class of commitments faces the decision on its
own: quasi-realism needs detail, not pictures.
Am(;mgst commitments which seem particularly likely to
benefit from such treatment are those involving modal idioms -
possibility and necessity. As already noticed, these come in
different strengths. We talk of physical possibility and neces-
sity; metaphysical possibility (is it metaphysically possible that
there should be minds without body, experience without time,
etc.?); logical possibility (the bare possibiUty that persons
should fly or horses talk is not ruled out by logical considera-
tions); sometimes epistemic possibility - situations left open by
some body of knowledge. How are we to understand claims of
· necessity and possibility? Naturally, the sources of these judge-
ments, in our connections of logic, or of metaphysics, or of
natural laws, attract detailed philosophical treatment, and
have their own problems. But as far as the realist/anti-realist
debate goes, there are some general points to be made. Consider
this famous passage from the chapter on 'Foundations', in
David Lewis's book, Counteifactuals:
I believe that there are possible worlds other than the one we happen
to inhabit. If an argument is wanted, it is this. It is uncontroversially
true that things might be otherwise than they are. I believe, and so do
you, that things could have been different in countless ways. But what
does this mean? Ordinary language permits the paraphrase: there are
many ways things could have been besides the way they actually are.
On the face ofit, this sentence is an existential quantification. It says
that there exist many entities of a certain description, to wit 'ways
things could have been'. I believe that things could have been diffe-
rent in countless ways; I believe permissible paraphrases of what I
believe; taking the paraphrase at its face value I therefore believe in
the existence of entities that might be called 'ways things could have
been'. I prefer to call them 'possible worlds'. (p. 84)
Lewis goes on to make two further comments: that talking
about possible worlds leads to no decisive problems, and that
the modal idioms are irreducible. You cannot express things
that we say using 'necessary', 'possible', and closely connected
notions, such as 'ways things might have been', in other terms.
Given this, he wants to take the idioms "at face value", which
214 LANGUAGE AND THE \\'ORLD

means, in his view, having a realistic theory of possible worlds.

There really are such things, and we really describe them, and
can know what they contain and what they do not. Now it will
be evident that this ignores the resources of anti-realism, and in
particular the possibility of a projective view. For does the fact
that we have commitments expressed in modal terms, and that
those commitments are irreducible, support realism? Only if
rejecting an idiom or reducing it are the only anti-realist
options, and they are not. (Of course, an identical mistake is
made by those who dislike realism about possible worlds, and
conclude that we must either reject or reduce the idiom.)
Thinking about necessities and possibilities in terms ofrelated
possible worlds is helpful for many purposes. It provides a
picture or image which aids thinking; it enables logicians to
approach modal arguments by mapping them onto more familiar
arguments. For instance, the inference from 'necessarily p' to
'p', is mapped into 'in all possible worlds p' to 'in this possible
world p', and becomes just the instantiation of a generalization.
This is a familiar inference form of basic 'first order' logic. But
realism? In present terms the issue must be one about how we
explain our making of modal commitments. We certainly do not
explain it by thinking that we are made sensitive to possibilities
because of some quasi-sensory capacity which responds to the
presence or absence of possible worlds. Firstly, since the.only
possible world that is actual is the actual world, others cannot
actually influence us: we cannot receive information about them,
because there is nothing to actually influence any receptors.
Ways things might have been cannot be seen or heard. Secondly,
the position that we are simply describing different aspects of
reality needs a supplementation which it finds hard to give. If
the possible worlds, like moral properties, represent a new
realm of fact, why should we be interested in it? Talking of
possibilities would be as optional to us, interested in the actual
world, as talk ofrieighbouring countries or different times
would be to us, ifwe are interested in the here and now. But it is
not. We want to know whether to allow possibilities or insist upon
necessities because we want to know how to conduct our think-
ing about the actual world. The possibilities we allow or which
we rule out, determine how we conduct our inferences, and
eventually our practices in the actual world. It is quite inexplic-
 r--
E\'ALcAT10Ns, PROJECTIONS 215

able how they should do that, if we relied upon the image of

different spheres of real facts or states of affairs. Why should it
interest me ifin such spheres something holds, or in some it does
not, or if in ones quite similar to the actual world things do or do
not hold, if I want to discover and use truths about the ways
things actually are?
Thus consider tfie claim that if you had fallen off the edge of
some cliff, you would have hurt yourself. Think of this as Lewis
wishes: it means that in any possible world as closely similar to
ours as may be, but in which you have fallen off the cliff, you
have hurt yourself. 17 But then there is debate over what makes
for similarity. After all, a world just like ours, except that your fall
off the cliff was followed by a minor miracle restoring you to the
top (perhaps with amnesia), continues just like our world-isn't
· that similarity for you? There has been considerable debate
over how to select the right similarity ranking, for otherwise we
would endorse the wrong counterfactuals (on this occasion: if
you had fallen over the cliff, you would have been quite un-
harmed). There is then a question over whether rightness can
be defined without circularity. 18 But suppose it can be. There is
still something crucial to explain. Why are we interested in just
that "right" kind of similarity? If our interest lay exclusively
in possible worlds, it would seem legitimate to roam around
describing the various dimensions of similarity and difference
with equal concern and respect. We only select one kind of
similarity as the "right" one because we use counterfactuals in
the real world - for the purpose of arguing, modifying practice,
avoiding falling over cliffs, and so on. In other words, the person
who thinks that if you had fallen over the cliff you would have
been unharmed is not just someone who describes possible
worlds with idiosyncratic standards of similarity. He is not just
like a tourist who makes oddjudgements of similarity between
two places bcause he is more concerned with the botany than
with the food. He is a menace. Lewis's realism leaves this use
unexplained. It would need supplementing by a theory of why
17
I abstract here from the complications of counterpart theory, which do not aflect
the present point.
18
The kind of circularity which threatens is this. Perhaps a minor miracle makes a
world very dissimilar from ours because it involves breaking a law ofnature, but perhaps
laws of nature are identified by thinking of which counterfactuals we hold true. Myself;
I doubt whether this problem is decisive against Lewis.
216 LANGUAGE AND THE WORLD

one kind of similarity between possible worlds should be the one

which matters. But then the whole package becomes extrava-
gent, like the moral theory which sees us as responding to a real
world of duties, obligations, and so forth, and then has to add an
explanation of why we bother about these things, as well as the
pains, pleasures, harms, and desires which make up the actual
world.
By contrast the quasi-realist will start by seeing the remark
(whether formulated in terms of possible worlds or not) as
expressing a commitment- one which endorses the structuring
of our beliefs about the world in a certain way. Notably, it
endorses an inference from supposing that someone has fallen
off a cliff, to supposing that they have got hurt. It dignifies this
connection, allowing it to be a reliable guide to the conduct of
affairs. The quasi-realist must then pursue the way we can
come to treat such a commitment as if it represented a judge-
ment with genuine truth-conditions. Similar problems arise as
in the moral case. Consider a simple example. We can regard
even our most reasonable counterfactual commitments as poten-
tially false. We know Henry's habits: we know that the best bet
is that if he had come to the party he would have got drunk. But
for all that it may be true ifhe had come he would have stayed
sober. Once again, this contrast must be explained, essentially
with the same materials as in the case of evaluations. We have a
sense of the potential for improvement in any argument: our
reliable estimate of Henry's behaviour might be supplemented
by information unknown to us (he had a cold, has just got a
girl-friend in the Salvation Army, etc.). We need a proper
notion of the improvement or deterioration in the bases for argument,
and proceed to construct a notion of truth. The eventual theory
( counterfactuals are intricate, and this is not the place to pursue
all the detail) would be one which maintains the benefits of
possible-world imagery, but disallows the metaphysical
extravagance. 19
Projectivism is also a promising option in the theory oflogical
necessity. We not only believe it to be true that 7 + 5 = 12, but
we also find the truth inexorable: it could not have been
19
The treatment of counterfactuals as condensed arguments was begun by John
Mackie, 'Counterfactuals and Causal Law', in Butler, ed., Analytical Philosophy, !st
Series.
 E\"ALCATIO'.\fS, PROJECTI.6.,.1\TS 217

otherwise. We cannot imagine it otherwise; we could make

nothing of a way of thought which denied it. But this may be
just a fact ~b?ut us and the limit~tions of ~ur present im~gi~a-
tions, and 1t 1s natural to complam that this fact cannot Justify
us in saying that the proposition has to be true, or is true in all
logically possible worlds. One reaction would be to avoid saying
this, and eschew the category of logically necessary truth: this is
Quine's position. But another is to face the fact that such truths
do occupy a special category, for we can easily imagine
otherwise many of our most cherished beliefs, and to say that
when we dignify a truth as necessary we are expressing our own
mental attitude - in this case our own inability to make any-
thing of a possible way of thinking which denies it. It is this
blank unimaginability which we voice when we use the modal
vocabulary. It is then natural to fear that thi9 has nothing to do
with the real modal status of propositions. (Compare: what have
our sentiments to do with the real moral truths about things?)
But the quasi-realist will fight this contrast: he will deny that
anything more can be meant by the real modal status of a
proposition, than can be understood by seeing it as a projection
of our (best) attitude of comprehension or imagination towards
it. Once again, the advantage of such a theory is that it avoids
the mystery of a necessity-detecting faculty and it avoids the
strained scepticism which tries to avoid admitting that any
truths are necessary at all.

6. Mind-Dependence
is one more aspect of such theories I need to mention. It is
tempting to think that a projective theory must leave the truth
of commitments on which it works "mind-dependent". And
this prompts hostility. It is not because of the way we think that
if kangaroos had no tails they would topple over. We discover
such facts, we do not invent them. I tj,§AOt l:>ecause qf the way
we form sentiments that kicking dogs is\VroI1g. It would be
· wrong wh-1-teY~!:~"Y~Jhoughtaboutit.Fluctuat:ions in our senti-
. ments only make us"l;>,etter or>worse ablt tq.appreciate how1
wrong it is:Tns.nod)e~ause of ihe way we conduct our argu-
ments that trees cause shade. The "mind-independence" of
such facts is part of our ordinary way oflooking at things. Must
 218 LANGUAGE AND THE WORLD

a projectivist deny it? On his construction 1s truth mind-

dependent?
Forf.ynately the quasi-realist treatment of indir~£!~J1texts
shows that it is not. Suppose someone said 'if we-had different
sentiftlenTs; it would b~ right_t:o k.(ck dogs', what could he be-i:ip
-io?-Apparentty;-.rreendorses a certainsen"sibility: one which lets
information about what people feel dictate its attitude to kick-
ing dogs. But nice people do not endorse such a sensibility.
What makes it wrongto kick dogs is the cruelty ()~pa.into _t_}le
_animal. That input should yielcldisapprqvaJand indignation~
tneoutput. Similarly, if someone so organizes his beliefs and t~ \)
way he makes inferences that he cannot let the presence of a tree!, i
in suitable sunshine suffice to give him confidence that there is j \I
shadow, but needs information about whether people think one\!
way or another,-he is in a mess. He will fail to be confident inf
truths about the actual world when he should be. ··
Of course there are moral truths and counterfactuals which-:.
are "mind-dependent". Behaviour which we call rude is often
wrong only because people think that it is wrong; sermons cause
people to switch off because people have those kinds of minds.
The point is that when a commitment is unlike these in being
independent of our minds and their properties, the projectivist
can conform to ordinary claims that it is. He does not need to
deny any of the common-sense commitments or views about the
way in which their truth arises. This is extremely important.
Idealists always face the problem of finding an acceptable way
of putting what they want to say about the involvement of the
mind in the world. Some fudge it: it is quite common to find
people writing that 'objects' do not exist outside our conceptual
schemes, or that we 'create' objects (values, numbers) rather
than discover them. 20 This is not a good way to put anything.
With the inverted commas off, such remarks are false. (We do
not create trees and galaxies, nor the wrongness of cruelty, nor
the evenness of the number 2. Nor can we destroy them either,
except perhaps for the trees.) But what can they mean
otherwise: what is meant by saying that 'trees' are mind-
dependent, if trees are not? Perhaps just the platitude that ifwe
did not have minds of a certain kind, we could not possess the
concept of a tree. The problem for the idealist, or the anti-realist
20
e.g. Putnam, Reason, Truth and History, p. 52.
 E\'.\LL\TIO:'\S, PROJ

in general, is to steer a course between the platitude and the

paradox.
The quasi-realist way of approaching indirect contexts offers
;i,a better approach. The utterance 'whatever I or we or anyone
lelse ever thought about it, there would still have been (causes,
Icounterfactual truths, numbers, duties)' can be endorsed even
iif we accept the projective picture, and work in terms of an
, explanation of the sayings which gives them a subjective source.
The correct opinion about these things is not necessarily the one
we happen to have, nor is our having an opinion or not the kind
of thing \\Thic:b: m,1!{es for correctness. The standards governing
projection make-it irrelevant, in the way that opinion is irrelev-
ant to the wrongness of kicking dogs. The temptation to think
otherwise arises only if a projective theory i~-Q1is1akeri for a
reductionist one, giving the propositions involved a content,
but one which makes them about _ll~ or out!:niQd~l:iey areilot
- they ha,ve a quff~_cHfferentrole, and
one whlch. gi~es. them_ no
,_ such trnth:ca"nditiori: When I say that Hitler was e~il or that
·. trees caiis~ shade I am not talking about myself. 21
It may now appear that a projective approach is too good to
be true. Initially, the contrast between expressive theories and
ones giving the same commitments genuine truth-conditions
seemed reliable enough. But the subtleties, the earning power of
quasi-realisl·d·evices, have tended to blur it. We can hear the
philosopher who gives a projective explanation say highly
realistic-sounding things. Why not regard him as giving the

21
\\'ittgenstein may ha\'e come close to the kind of theory here explained. He
certainly seems to want an anti-realist theory of arithmetical necessity without in any
way regarding truths of arithmetic as truths about us or as truths of natural history . .\nd
my projecti\'e way with mind-dependence oflers a model for doing this.
It should be noticed that because of the twist in construing these counterfactuals
this way, it comes outfalse that ifwe had thought or felt otherwise, it would have been
permissible to kick dogs. This means that the metaphor of 'projection' needs a little
care. \'alues ar~ the children of our sentiments in the sense that the full explanation of
what we do when we moralize cites only the natural properties of things and natural
reactions to them. But they are not the children of our sentiments in the sense that were
our sentiments to vanish, moral truths would alter as well. The way in which we gild or
stain the world with the colours borrowed from internal sentiment gi\'es our creation its
own lite, and its own dependence on facts. So we should not say or think that were our
sentiments to alter or disappear, moral facts would do so as well. This would be
endorsing the detective counterfactuals, i.e. endorsing the wrong kinds of sensibility,
and it will be part of good moralizing not to do that. Similarly, it would ha,·e been true
that 7 + :i = 12, whatever we had thought about it.
220 LA:\iGUAGE AND THE \\'ORLD

commitments in question genuine (mind-independent) truth-

conditions? He practises soberly and responsibly, as though
there were one truth which it is his business to find. So perhaps
the right thing to say is that the commitments are true or false~-,~
in a straightforward way, after all. Perhaps expressive theories, ,/
properly developed, pull themselves into truth-conditional ones Iv
by their own resources. It is just that by disavowing the direc- !
tion of explanation associated with the idea of a truthJ
condition, they do this without entering troubl~s of metaphysics
and epistemology, and the resulting scepticism which plagues
other approaches. To assess whether this is right, we need
turn to more general considerations about truth.

Notes to Chapter 6
6.1 Supervenience, in the sense of this section, should not be con-
fused with the stronger requirement often called 'universalizability',
following the usage ofR. M. Hare. There are actually three importantly
different notions:
Consistency: the requirement that you do not contradict your-
self, both judging that something is the case and that it is not.
Supervenience: the requirement that moral judgements supervene,
or are consequential upon natural facts.
Universalizability: the requirement that moral judgements are
somehow dependent only upon universal facts; facts specified
without reference to particular individuals or groups.
The last of these is an attempt to build into the very definition of
morality a requirement of impartiality, or of treating like cases alike,
or of abstracting from any personal position or interest in achieving a
moral point of view. Little is gained by building this into the definition
of morality, or into the "logic" of the term moral and its dependent
vocabulary: this just invites the question: why be moral? why not be
"shmoral" - something a degree or two less strenuous than being
moral, allowing one to pay some attention to other people's interests,
but discounting for their distance from oneself? A clearer approach is
to admit that it is a substantive question whether we conduct our
practical reasoning by abstracting away from particular interests,
and to stress the advantages (to themselves or to the community) of
bringing up people to do so. This would be a variety of what is called
'motive utilitarianism': universalizing is defended as being a good
 E\'ALL\TIO'.\S, PROJEC1lONS 221

thing to do, because if we can get people to respect the results of doing
so, things go better. The relevant literature is, in my view, rather
muddy on this. It includes:
R M. Hare, Moral Thinking, esp. ch. 6.
J. Mackie, Ethics: Inventing Right and Wrong.
and specifically on supervenience:
J. Kim, 'Supervenience and Nomological Incommensurables',
American Philosophical Quarterly ( 1978).
J. Dancy, 'On Moral Properties', Mind (1981).
The argument from the modal relations between moral and natural
judgements, to anti-realism, can be challenged. If we pay careful
attention to the different possible ways of taking the necessity
involved, there is a possible counter. This would involve distinguish-
ing metaphysical necessity from logical or conceptual necessity. The
realist might say that if we are talking of metaphysical necessity, then
once a total natural grounding for a moral judgement is located, it will
be metaphysically necessary, i.e. true in all possible worlds, that when
that grounding is present, the relevant judgement is also true. In
short, he would accept B* I A necessity. If we are talking of conceptual
or logical necessity, then he does better to deny this, although he must
still accept the conceptual necessity of supervenience. But then he
could try accepting the .conclusion: saying that it is a conceptual or
logical constraint on the moral vocabulary that it supervenes on the
natural, whereas no particular B* I A connection is logically forced
upon us. The request for explanation still arises, however. Where
does the logical constraint come from if realism is true? Logic is
interested in what can be true, and as far as realism can show us, it
could be true that the moral floats quite free of the natural.

6.2 The theory of conditionals with evaluative components, which I

develop in this section, is just part of the larger programme of quasi-
realism, the attempt to show how an economical, anti-realist, and
expressive theory of such things as ethics can account for the
phenomena which lead people to realism. Papers in which I pursue
the same theme include:
'Truth, Realism and The Regulation of Theory', in French et al.
( 1980).
'Opinions and Chances' in Mellor ( 1,980).
'Rule Following and Moral Realism', in Holtzman and Leich
( 1981).
222 LANGUAGE AND THE WORLD

6.3 ' ... a schizoid attitude to his own moral commitments ... '
\Vhy is it that people want more than the projectivist gives them?
Thomas N age! talks ( The Possibility ofAltruism) of philosophers such as
Kant, "driven by the demand for an ethical system whose motiva-
tional grip is not dependent on desires which must simply be taken for
granted" (p. 11). This is the permanent chimaera, the holy grail of
moral philosophy, the knock-down argument that people who are
nasty and unpleasant and motivated by the wrong things are above
all unreasonable: then they can be proved to be wrong by the pure sword
of reason. They aren't just selfish or thoughtless or malignant or
imprudent, but are reasoning badly, or out of touch with the facts. It
must be an occupational hazard of professional thinkers to want to
reduce all the vices to this one. In reality the motivational grip of
moral considerations is bound to depend upon desires which must
simply be taken for granted, although they can also be encouraged
and fostered. Notice that this is consistent with saying that there are
values which we come to recognize or discover, just as there are
rewards and satisfactions which we come to recognize and discover.

' ... we are constrained to argue and practise as though the truth is
single ... '
This needs some care. People may be wrongly tempted to relativism
by this thought. There are obligations which we feel although we are
also aware that other equally admirable systems do not recognize
them. The best examples are those of ceremonies and rituals which
arise, we suppose, because there is some deep need in us which they
serve, although how this need is served can then be highly variable.
For example, I might feel the strictest obligation to dispose of the
body of a relative in some prescribed way, even when I know that
other societies would do it differently. In this case I do not assent to
'all human beings ought to bury (say) their dead'; I do assent to 'I
ought to bury my dead'. But this is not relativism in the sense of the
text, for I would also regard each judgement (that I ought to bury my
dead, that not everyone ought to bury their dead) as true, and there is
no equally admirable conflicting alternative to either of them. It is
just that what creates such obligations are parochial facts about
people and their societies and their customs.

6.4 A basic reading on bivalence would include:

M. Dummett, Preface to Truth and Other Enigmas, and the papers
'Truth', and 'The Philosophical Basis of Intuitionistic Logic',
collected therein.
 EVALL'ATIO'.'iS, PROJECTIONS
)
223
C. Wright, Wittgenstein's Remarks on the Foundations of Mathematics, pt.
especially ch. 7.
D. Prawitz, 'Meaning and Proofs', Theoria (1977).
D. Edgington, 'Meaning, Bivalence and Realism', Proceedings ef the
Aristotelian Society ( 1980).

6.5 ' ... Hume's treatment ... '

The two classic places are:
Treatise ef Human Nature, Bk. 1, Pt. HI, sect. XIV.
Enquiry Concerning Human Understanding, sect. VII.
For possible worlds, and some of the disputes over their standing, see:
A. Plantinga, The Nature of Necessity, chs. VI-VIII.
S. Kripke, Naming and Necessity, passim.

6.6 The relationship between post-Wittgensteinian attempts to

avoid realism and the actual work of Wittgenstein is hard to unravel.
Two modern metaphysicians tempted towards recognizably idealist
positions include Hilary Putnam and Richard Rorty (see Rorty,
Philosophy and the Mirror ef Nature and more recently, The Consequences of
Pragmatism). I discuss some of these things further in the next chapter.
An excellent paper is 'Wittgenstein and lpealism' by Bernard Williams,
reprinted in his collection Moral Luck.
In this chapter I have not discussed directly the 'anti-realism'
about sciei1ce which allows theoretical sentences to have an intelli-
gible truth-condition, and allows us to accept them and to use them
when they are empirically adequate, but counsels us not to believe
them. This is the position of van Fraassen, and perhaps Popper. It
needs faith in some distinction between accepting a statement with a
truth-condition, and believing it, and I see no such distinction. I
would urge that the right path for instrumentalism is to deny that the
commitments have a truth-condition, at least until the quasi-realist
does his work.
 CHAPTER 7

Correspondence, Coherence,
and Pragmatism

Her terrible tale you can't assail

With Truth it quite agrees;
Her taste exact for faultless fact
Amounts to a disease.
W. S. Gilbert, The Mikado.

1. Vacuiry and the Quietist Regress

We have come to the theory of truth through trying to under-
stand a contrast. Some commitments do not have truth-
conditions, do not mark genuine judgements, but are somehow
expressive. This allowed the quasi-realist supplement, which in
turn gave us the right to treat the commitments as if they had
truth-conditions, and justified our common practice. But is this
anything other than the real thing? Our conception of truth, as
the point ofjudgement, or as success in judgement, must tell us
whether truth is ever more than these hopefuls earn. So what is
there to say about truth in general?
The first things are platitudes. A true judgement gets things
right; says that things stand in a certain way when they do so
stand, tells it like it is. The classic expresson is Aristotle's: "To
say of what is that it is, or of what is not that it is not, is true." 1
The trouble is that these platitudes are only a prelude to serious
attempts to understand what truth is. If we worry about the
nature and point of the division between judgements which are
true and those which are not, or between utterances which
express judgements and those which express other commit-
ments, these truisms about truth are no help.
The first suggestion which looks as if it might be more sub-
stantial is that a judgement is true ifand only ifit corresponds to
the facts. A taste exact for faultless fact is just the same thing as a
' Metaphysics, I' 7.27.
 CORRESPO:\'DENCE, COHERENCE, AN}lYRAGMATISM 225

taste for truth. And it is, presumably, a suspicion that on the

theories of the last chapter there is no such thing as moral,
counterfactual, arithmetical, etc. fact which still prompts dis-
quiet about genuine truth in those areas.
The platitudes are harmless. And we do not seem to raise the
theoretical temperature very far if we substitute 'corresponds to
the facts' for 'says of what is that it is' and the like. So it must
surprise students to find how hostile most writers are to what
they call 'the correspondence theory of truth'. Some think it
vacuous; others false. First of all it is necessary to state the
theory. Th~e is nothing objectionable in introducing
'corresponds-to-the-facts' as a grammatical predicate having
just the same use as 'is true'. Someone trained in the Pentagon
might introduce any number of such paraphrases: 'is-currently-
operative', 'is-operationally-valid', and the like. It is not predict-
able from the normal meaning of the words in such predicates
that the phrase is to be a synonym for 'is true'. But people
manage to learn them. Suppose, then, that 'corresponds-to-the-
facts' is in fact synonymous with 'is true'. It does not follow that
it provides a "theory" of truth, any more than the Pentagon's
imaginative synonyms make it a theorist of truth. It just pro-
vides one more way of saying that things are true. So to deserve
the title a correspondence theorist must say more than that a
judgement is true if and only if it corresponds to the facts. He
must say that this is not a mere rephrasing, but sets afoot a
genuine elucidation of the idea. He must say that we have some
genuine conception of what facts are, and what correspondence
is, so that thinking in terms of correspondence with the facts will
help us to answer the kind of question left at the end of the last
chapter. I define a correspondence theory as any theory holding
that there is this content, enabling correspondence and facts to
play some role in elucidating truth. This means that there is not
one such theory. There is room for different conceptions of
correspondence and of fact. All we have is an invitation to
explore in a certain direction. Should we accept it?
The first objection is that 'corresponds with' and 'facts' have
no independent substance. They trade entirely with each other.
Neither has an identity outside the other's company, and in that
company they gain a use only by being allowed as a substitute for
'is true'. The charge is that the equation is vacuous. It is
226 LANGUAGE AND THE WORLD

harmless, only so long as we do not think ofit as providing any

key to the nature of truth.
To meet this kind of criticism means showing that there is a
kind of content to 'corresponds with the facts', which is
predictable from a prior understanding of its component
notions, and not wholly derivative from its synonymy with truth.
'Corresponds', after all, is a word with other subject matter.
Pictures correspond with scenes, music corresponds with
moods, memories with events, outcomes with intentions.
Doesn't this suggest a general conception of correspondence,
carrying a content from other contexts to this one? Perhaps facts
too live their own lives: facts cause things, they can be witnes-
sed, they matter. So can't we have two conceptions locking
together to provide us with our understanding of truth? Of
course, it is possible that particular philosophers have held
conceptions of fact and of correspondence which are inade-
quate. Debates supposedly on the correspondence theory often
rush to introduce, say, Wittgenstein's early picture theory of
language (as though picturing it were the only kind of way one
thing could correspond to another). Or, they point to unappeal-
ing things people have said about our acquaintance with facts
(see below, 7.4). But it is much harder to pin the weaknesses of
particular developments onto the invitation itself. Indeed, how
could it be proved that there can be no such conceptions as the
invitation demands?

2. Transparency: Dummett and Frege

Surprisingly there exists argument that there cannot be any.
Consider what is sometimes called the "equivalence thesis": a
judgement pis true iffp. 2 It is true that there is a cat on the mat
iffthereisacatonthemat;itistruethat2 + 2 = 4iff2 + 2= 4;
and so on. We can take this equivalence to be as strong as
possible: to judge that pis nothing different from judging that it
is true that p. In fact let us suppose that, for any judgement and
any relation we may hold to it, it makes no difference whatever if
we preface the judgement by 'it is true that'. To ask, prove,
enquire, deny, wonder, imply ... that pis always the very same

2 Dummett, Frege, ch. 13. 'lff means 'if and only if.
 CORRESPO:\'DE'.',CE, COHERE'.\ICE, A'.',D PRAG:\UTIS:\I 227

thing as asking, proving, etc. that it is true that p. This is the

strong equivalence thesis. It might be helpful to call the pro-
perty of'it is true that ... ' which it illustrates, the 'transparency
property'. It is as though you can always look through 'it is true
that' to identify the content judged, inquired after, and so on if
the reference to truth was not there. Now the transparency
property is a peculiar one. In particular it puts a dilemma in the
way of any analysis of 'it is true that'. Either the proposed
synonym is no real analysis, but a vacuous, Pentagon-type
paraphrase; or it is a genuine analysis. But in the latter case
there is a content to the notions it introduces. And in that case
'it is true thatp' contains that content, whereas mere 'p' does not
(does not for instance m~ntion correspondence and facts). And
in that case the two propositions are different. Adding 'it is true
that' alters things, and in that case the transparency property is
lost. There will be no general presumption that wondering,
doubting, proving, etc. that pis the same as wondering, etc. that
it is true that p.
To make this problem vivid, imagine that on some theory
there is always a difference of content between 'p' and 'it is true
that p'. Then call this latter judgement 'q'. We can then make a
yet different judgement, that it is true that q: call this 'r'. We are
off on an ascending hierarchy of different judgements, each
claiming that its predecessor is true. But it is impossible to
believe in the distinctness of each member of this series. We can
get a kind of visual sense of the awkwardness, ifwe imagine the
correspondence theory developed by some genuine under-
standing of what a fact might be, as a kind of entity of good
standing, fit to be corresponded with. Then starting with a
judgement thatp, proceed to the judgement thatp corresponds
(C) to some fact, say W: p C W. The judgement that p C Wis
itself capable of truth: perhaps it corresponds to W*:
pC WC W*. For this to be true it must correspond to W**, and
so on. If you imagine a judgement at one point, and above it a
fact to which it corresponds, then the fact that it so corresponds
seems to be somewhere else-out at a tangent- and the fact that
the judgement that the original judgement corresponds to a
fact, corresponds to this fact, is out at another tangent, and so
on.
The image is worrying, but does the argument actually work?
228 LANGUAGE AND THE vVORLD

We can lay it out like this. Consider some proposed analysis of

truth, in terms of some concept cp. Then consider the two series:
p Tp TTp TTTp .. .
p cpp cpcpp cpcpcpp . . .
then:
( l) In the first series, each member has the same content as
any other; none says anything more than or different
from the first. (Transparency property.)
(2) If cp is an elucidation, analysis, or definition of truth, then
it must share the transparency property.
(3) Either cp is a genuine predicate, containing concepts ofits
own, or a manufactured synonym ofT.
(4) If the latter, then it offers no elucidation, etc.
(5) If the former, then it cannot share the transparency
property, and again (by (2) ) is no elucidation, etc.
(6) So there is no possible elucidation of truth.
The argument is accepted in this strong form by Frege, who
regarded it as refuting not only the correspondence theory, but
any other attempt to define truth as well. Dummett does not
allow it this much power; he thinks that it refutes the kind of
definition embodied in the correspondence theory, but leaves
the possibility of others. 3 But he does not say how it might be
evaded - what it is about some proposals for cp and not others
which would make them targets. Apparently any cp unaffected
by the argument would have to be a genuine elucidation, but
one which shares the transparency property. But if that is a
possible combination, it must be something peculiar about
'corresponds to the facts' which disqualifies it from being so
regarded. Perhaps it too can slip between importing more than
the transparency thesis tolerates, being taken as a manufactured
synonym.
In parenthesis I should mention that the issue is particularly
delicate for Frege. He held the curious doctrine that a sentence
is true ifit refers to a particular abstract object- the truth-value
3
Frege, 'The Thought: A Logical Inquiry' in Strawson, ed. (1967) p. 19. Dummett,
Frege, pp. H3 ff
 CORRESPO:'\iDENCE, ~:OHERENCE, AND PRAGMATISM 229

'True'. This will be so only if the thought the sentence expresses

relates in some suitable way to this object - say 'determines' it.
Now this doctrine may not have been intended as an elucida-
tion or analysis of the concept of truth. But it is still pertinent to
ask how it fares in face of the transparency property. The issue
whether p, generates the issue whether the thought that p determines
True; this in turn leads to the issue whether the thought that (the
thought that p determines True) itself determines True, and so on. If
'determines True' does not have the transparency property,
then these issues may be different from each other. If it does
have the transparency property, in spite of being intended to
have a definite content, describing the relation of a judgement
to an abstract object, then why cannot 'corresponds to the facts'
have the same combination?

3. Redundan9: Ramsey and Wittgenstein

The issue would be clearer if we knew why truth has the
transparency property in the first place. At first sight the pro-
perty robs the word of any content at all, making it vacuous or
redundant. Some writers, notably Ramsey and Wittgenstein,
have so concluded. They are often described as holding a
'redundancy theory of truth'. This is the view that truth is not a
contentful notion, and not a fit object of analysis. The role of the
word is not one of introducing a property which some utterances
or judgements have and others lack. I prefer to call this view
'quietism', for it is not really the view that the word is redundant
on all occurrences. It is the view that it never marks a property
of judgements. But does the regress really suggest that we have
no general conception of success in judgement, or of what
correctness is, and how it might relate to other concepts, such as
those of rationality, evidence, proof? Someone who lacks any
notion of assessing sayings for truth or falsity lacks the idea that
they express judgements at all-words remain wooden to him in
the sense of chapter 3. How can the concept be redundant ifit
plays such a central role in our reaction to utterances? Is there
to be no way at all of describing something that all true judge-
ments share, and all false judgements lack? In the view of a
redundancy theorist there is no such property. But what then of
issues such as those of the last chapter, which result in the query
230 LANGUAGE AND THE WORLD

whether different classes of judgement are true in the same

sense as others? It seems ill-judged to abandon any such discus-
sion merely because of the transparency property.
Tp.e best way for the student to appreciate the attractions ofa
redundancy theory is this. We can say equally that judgements
(statements, thoughts, beliefs, propositions) are true, and that
sentences (or sentences in contexts, or utterances) are true.
(There are differences between these, as 1.4 described, but for
the moment they do not matter.) Now compare 'is true', as a
predicate of either judgements or of sentences, with a genuine
target of philosophical analysis: 'is conscious', or 'has rights',
for example. We investigate these by looking for the principles
which determine whether something is conscious, or has rights.
These principles are intended to govern any such judgement, so
that we get a unified class: the class of conscious things, or
things that have rights. Each item in such a class is there
because it satisfies the same condition, which the analysis has
uncovered. Or, if this is slightly idealized, we find only a
"family" of related conditions or "criteria" for application of
the terms. Still there is then a family relationship between the
members of the class. But now contrast 'is true'. We know
individually what makes this predicate applicable to the
judgements or sentences ofan understood language. 'Penguins
waddle' is a sentence true, in English, if and only if penguins
waddle. It is true that snow is white if and only if snow is white.
The reason the first sentence deserves the predicate is that
penguins waddle, and the reason why the judgement that snow
is white deserves the predicate is that snow is white. But these .
reasons are entirely different. There is no single account, or
even little family of accounts, in virtue of which each deserves
the predicate, for deciding whether penguins waddle has
nothing much in common with deciding whether snow is white.
There are as many different things to do, to decide whether the
predicate applies, as there are judgements to make. So how can
there be a unified, common account of the "property" which
these quite different decision procedures supposedly deter-
mine? We might say: give us any sentence about whose truth
you are interested, and simple by "disquoting" and removing
the reference to truth, we can tell you what you have to judge in
order to determine its truth. Since we can do this without any
 CORRESPONDENCE, COHERENCE, AND PRAGMATISM• 231

analysis or understanding of a common property of the

idea that there is such a thing is an illusion.
Perhaps the conjuring trick was pulled before we were look-
ing. Truth is internal to judgement in the sense that to ma~e or
accept a judgement is to have it as an aim. So how could there
be a difference between making a judgement on the one hand,
and describing the judgement as meeting the aim, on the other?
Consider some parallels. In a game there is an equivalence
between making a move and judging that the move is a good
one. Or, there is an equivalence between choosing a key and
supposing that the key fits. Discounting cases of fraud, acting,
and so ori, a normal choice of move, or choice ofkey, is governed
by these aims. It would therefore be redundant to add, when
moving or choosing, that the move is one which promotes
winning, or the key is one which fits. This is already implicit in
the choice. But this redundancy does not suggest that we have
no conception of what it is to win or what it is for a key to fit.
Even more closely, consider commands. Since the internal
point of command is to get something done, it is redundant to
add to a command a 'do this' indicator. 'Shut the door' and 'Do
this: shut the door' come to the same thing. But this trans-
parency does not stop there from being a contentful notion of
obeying an order. The analogy is that since truth counts as
success in judgement, making a judgement and describing it as
true are evidently equivalent. But for all that we may have a
substantive conception of what that success is. Discussing,
proving, querying, etc. whether p is of course the same as
discussing, etc. whether p is true. But that is not, on this
account, because 'is true' is somehow vacuous. It is because its
content has already gone into the bag: ifwe are discussing, etc. a
judgement then we are already governed by a conception of
success in judgement, and one which philosophers should try to
explain. Judgements are those amongst our commitments
which admit of truth or falsity.
A theory would deprive the equivalence thesis ofjustification
if it appeared to show that one might correctly judge that p
without judging that it is true that p, or vice versa. A view will
not do this provided its favoured conception of truth allows for
truth being the point ofjudgement, the goal which we suppose a
judgement to have achieved when we assent to it. This point is
232 LANGUAGE AND THE WORLD

to be internal or essential to judgement, in the sense that

nothing could count as judging unless it had this aim. So the
question boils down to whether a correspondence, or other,
theory fails because it is impossible to reconcile it with the
demand that truth be this internal point ofjudgement. And at
first sight, a correspondence theory should do well at just that
point. The reason we make judgements is that we want the
facts.
It seems then that an elucidation of truth can have content,
and not only permit the transparency property, but in a sense
explain it. Which step in the above argument should be denied
if this is right? There is room for some choice. The most direct
line would be to deny (5). Provided the concepts involved do
define the internal point of judgement, they should themselves
have the transparency property. A more subtle line would
expand our notion of an elucidation. So long as we are confined
to seeking a synonym, we need a substitute with the same
capacity for vacuous repetition as 'it is true that ... '. But
suppose our elucidation took a different form. Consider again
the difference between crude naturalism in moral theory, and
the more subtle approach to moral language suggested in the
last chapter. In such developments we seek to understand a
concept not by reduction, but by explaining its properties; the
explanation is not aimed at providing a substitute (a non-moral
way of moralizing, as it were) but at making the phenomenon of
moralizing unproblematic. There is no principled reason why a
conception of what truth is might not play the same role,
although it provides no analysing phrase which shares just the
same repetitive powers as 'it is true that ... '. It would explain
what we are doing when we make judgements, display a con-
ception of success or correctness which governs judgement, and
make it unsurprising that we have a notion of truth, and the
consequent behaviour of predicates announcing it. On such an
account we escape the Frege argument by denying (2).
For good measure we might notice that (I), the strong
equivalence thesis itself, is not entirely unproblematic.
Remember that we may use a vocabulary in some area where
we are uncomfortably aware that there might be an improved
way of looking at things-one of those descendant theories from
whose standpoint our classifications and judgements appear
 CORRESPONDENCE, COHERENCE, AND PRAGMATISM 233

not actually false, but best avoided (see 6.3 above). But we may
have no sense of how to improve ourselves, and therefore go on
judging in the best terms we have. So I may assent to the
opinion that Schubert was a romantic, that Macbeth was
ambitious, or that America is democratic, whilst being at the
same time aware that superior opinion might make qualifica-
tions, distinctions, tum things in a different light, and perhaps
end up avoiding these terms altogether, leaving my present
commitments at best half-truths. In that case I accept the
opinion more readily than I accept its truth. The natural similes
and metaphors for such worries chime in well with the cor-
respondence image: it is as though we paint an approximate or
distorted picture, or an inferior picture, or a partial sketch.
The blockbuster argument against searching for an elucida-
tion of truth fails, then. Is correspondence to the facts an answer
to that search? There are other, more traditional obstacles.

4. Facts and judgement: Kant

When people first start to sympathize with the correspondence
theory, they are apt to seize upon some homely, immediate,
truth, a truth which "leaps to the eye". ·I consider, say, the fact
that my typewriter is on the table, and say 'well, that's a fact, that
is what my judgement that the typewriter is on the table is made
true by'. This seems as pure a case of sheer acquaintance with a
fact as can be got (nobody with this kind of experience is going
to deny that my typewriter is on the table). The instinct behind
such a choice of case is sound. The critic is pressing for a
conception of 'the facts' which is not wholly derivative from an
antecedent conception of true judgement. So it is right to turn to
a good case where we might talk of acquaintance with the facts,
where the facts leap to the eye. For then that acquaintance
might give us an understanding of one end of the relationship
( the facts), and then we could look forward to building cor-
respondence in the light of that.
This is why the correspondence invitation to truth has often
seemed tied to a view which emphasizes an element of 'the
given': some bare unvarnished presentation of fact to a purely
receptive mind. Philosophers are virtually unanimous in con-
demning any such notion, and its ill-fame (coupled, perhaps,
234 LANGUAGE AND THE WORLD

with the thought that nothing else even looks like a candidate
for elucidating truth) is the main cause of quietism. The critics
of "the given" have a simple, and strong, case. Experience
cannot be regarded as an independent source ofa conception of
a fact - independent, that is, of the operation ofjudgement-for
two reasons. The first is obvious: to see a situation as one
containing or illustrating or displaying a fact is just to judge and
interpret. Even such a low-grade judgement as that my
typewriter is on the table involves recognizing that the elements
of the situation are spatially external to me, that they are objects
with a temporal history, that they have various physical proper-
ties such as solidity, and so on.Judgementjust is the isolation of
facts. Now, faced with only this point, we could retain an
intuitive distinction between experience - the "raw" uninter-
preted presentation - and the judgements we make in the light
of it. But the second point is that our conceptual powers
themselves infuse and condition the experience. Interpretation
goes into the making of the experience. And since that is so,
experience can no more be a judgement-independent source of
acquaintance with facts than dough can be a flour-independent
source of bread.
This interweaving of the categories of thought with experi-
ence is a central element in Kant.
Without sensibility no object would be given to use, without under-
standing no object would be thought. Thoughts without content are
empty, intuitions without concepts are blind. 4
Kant carefully goes on to say that this is no excuse for confusing
the two contributions. But the point is that their separation
is not given to us. When we see the typewriter on the table, we
see the situation as involving extended and independent
objects. Kant, like Hume before him, and Wittgenstein after
him, understands this "seeing as" as a product of imagination;
it is essentially a matter of linking the present impressions to
actual and possible perceptions of similar things. By imagina-
tion we fill in the continuities and stabilities which turn our
experience from that of a disconnected sequence of independent
4
Kant, Critique of Pure Reason, A 51-2; B 75-6. The content of thought is here its
connection with the sensible world; intuitions are what we receive; they are the
materials delivered by the senses.
 CORRESPO'.\"DE'.\'CE, COHERE>iCE, A'.\'D PRAG'.\L-\TIS'.\I 235

events, into that of a stable world of spatially extended objects

with enduring histories. Our experience would not be what it is
if it were not infused with this interpretation. This general fact is
most apparent when we consider cases of "gestalt switch" or
double-aspect figures. With these, when we change from one
interpretation to another (seeing a figure as a black shape on a
white background or vice versa; seeing the ambiguous figure
as a duck or as a rabbit), we start to have a quite different
visual experience. But these cases just illustrate a universal
interdependence of experience and judgement.
The theme is commonplace in writers on perception. What is
not so well agreed is the implications for our conception of a fact
and of the control of beliefs by facts. Clearly an initial burst of
seeing something as a duck, or a smile, or a circle, may be
followed by painful realization that it is actually a rabbit, or a
snarl, or an ellipse. As more experience comes in, initial judge-
ments may have to give way. But if no judgement or experience
has a pure footing in the way things are, uncontaminated by our
powers of thought and imagination, how are we to think of this
control? Is it even possible to think of our beliefs as responding to
facts, let alone corresponding to them? Or are we left with a kind
of majority tyranny in which any odd judgement, formed on an
occasion, but surrounded by occasions which prompt conflict-
ingjudgements, is deemed false, whilst the majority are deemed
true (and said to correspond to the facts)? It is this problem that
leads to the coherence theory of truth.

5. Coherence: Russell, Bradley, and Peirce

If the innocent student thinks that nothing much can be wrong
with the idea that truth consists in correspondence with the
facts, he probably also thinks that nothing much can be right
with the idea that it consists in some kind of coherence between
a judgement and other judgements. And the idiom of the
philosophers usually credited with this direction of theory is
scarcely calculated to overcome this initial scepticism. Bradley,
Joachim, Bosanquet, and other "British Idealists" of the late
nineteenth and early twentieth centuries now get a poor press,
or no press at all. Yet it is easy to present their views in a
favourable light.
236 LA'.'/GlJAGE AXD THE \\'ORLD

The first attractive doctrine of the Idealists was the "holistic"

nature of meaning (see 5.4). No belief is an island; to under-
stand it entails being able to use its concepts in other contexts,
to make other related judgements: it involves mastery ofa web
or system. In chapter 5 I mentioned this doctrine in connection
with Wittgenstein and Quine, but they were not the first. Here
is Joachim, writing in 1906:
No universal judgement of science, then expresses in and by itself
a determinate meaning. For every such judgement is really the
abbreviated statement of a meaning which would require a whole
system of knowledge for its adequate expression. It is this larger
meaning, embodied more or less fully in such a system, which, so to
say, animates the single judgements and gives them determinate
significance. 5
The second point of contact with modern work is the same
dis belief in passive acquaintance with facts. Bradley writes that
"the merely given facts are the imaginary creatures of false
theory" and that the given facts "show already in their nature
the work of truth-making". He emphasizes that even if there
ever existed a datum of sensation or feeling wholly unmodified
by thinking and judgement, then we would not know ofit, fat in
order to become an object of knowledge it would need description
and interpretation, and this reintroduces the work of judge-
ment. This is the Kantian theme of the last section. It means
that there is no way in which any mind can step back from its
own system of belief, survey without its benefit a reality the
system attempts to depict, and discover whether it is doing well
or badly. Any such survey produces only more beliefs. to blend
in with the original system. If we think in terms of the ship of
knowledge controlled by tides of experience, it is as though all
the tide does is cast up more planks - more elements to take into
the system.
At this point we are liable to suffer a sharp sense ofloss. If the
senses are "impregnated with the work of truth-making" what
is their title to control the way in which we form systems of
belief? Bradley, like everyone else, believes that they should do
so:

5
H. H.Joachim, The Nature of Truth, p. 96.
 CORRESPONDENCE, COHERENCE, AND PRAGMATISM 237

I do not believe in any knowledge which is independent offeeling and

sensation ... our intelligence cannot construct the world of percep-
tions and feelings, and it depends on what is given - to so much I
assent. 6
The difficulty is that the assimilation of observation to judge-
ment makes it seem as though our faculties are simply ready,
cheap, and fertile sources of belief. When we open our eyes
experience floods in, but so far as truth goes, we might just as
well say that judgements do. 7 The tribunal of sense experience
delivers only further beliefs. So it seems that it offers no inde-
pendent test of truth. To test the truth of a belief, to find its
truth, must be to try to fit it to others (no doubt gained from the
senses). So we arrive at the suggestion that perhaps this coher-
ence is, in effect, what truth is. To develop the suggestion means
exploring the virtues which systems of belief might have, and
trying to erect a concept of truth on their basis. This copies the
procedure the quasi-realist adopted in the case of morals, where
a notion of the virtues in different sensibilities produced enough
to justify a workable concept of moral truth. The root idea is that
the virtue of truth is constructedfrom the virtues of method.
Examples of the attractions of this idea offset its initial
strangeness. Consider the common marks of merit in scientific
theories: simplicity, responsiveness to experiment, utility,
theoretical elegance and strength, fertility, association with
familiar models rendering processes intelligible, and so on.
There is an inclination to think: but what have all these virtues
got to do with truth? A coherence theorist cuts this knot, most
simply by proposing that scientific truth is membership of some
system which, like M* (see 6.3), would be the ideal possessor of
all these virtues. For example, consider the "holistic" attribu-
tion of beliefs and desires to people, as ways of rationalizing
their sayings and doings. There are better and worse ways of
doing this. If we can define the virtues of some ideal method of
6
F. H. Bradley, 'On Truth and Copying', Essays on Truth and Reality, p. 108.
7
I am not here assenting to the view that having an experience just is being disposed
to make a judgement. I do not believe that it is. There are experiences, such as hearing
indescribable sequences of sound, whose nature is not identifiable by any judgement
the subject could make about them. vVe can see this must be so, because the experience
could change, although the subject cannot judge how it has changed, nor say anything
about how the new experience differs from the old one. And any experience will have
aspects of which this is true. The point in the text is only that judgements have to be
distilled from experience before questions of truth arise.
'.238 LANGUAGE AND THE \VORLD

doing it, someone might still ask why we suppose that this
method gives the truth about what people believe and desire:
the coherence answer is that this is what psychological truth is.
It is true that someone believes that p or desires that p if and
only if the use of this method would rationalize his sayings and
doings in such a way that this comes out true. Evidently the
advantage of this approach to the theory of knowledge is that it
ensures that our preferred methods of forming theories are
adequate to the truth. It ensures it by so defining truth. This is
why the early British Idealists thought of themselves as good
empiricists with a sane appreciation of what observation is, of
the way theory is grounded in experience, and of the way to
answer scepticism. This too is why the Logical Positivists also
come to favour a coherence theory of truth, in the 1930s (see
notes).
The Idealists were led to the doctrine mainly by reflection
upon method in history, but later on reflection on method in
science performed the same office. Bosanquet, for instance
wrote:
The facts, in history, at any rate, are not simply there, so that they can
act as a given standard correspondence to which is truth. The prim-
ary working standard is critical system, or, what is the same thing,
scientific investigation. 8
But the passage is misleading in presenting the virtues of
systems of well-judged belief as mere standards of truth. A coher-
ence theory does not simply describe the common-sense virtues
which good methods of conducting enquiry must exhibit. What
is distinctive is that it sees truth not as an independent property,
which these virtues are hopefully signalling, but as a construct
out of them.
To assess the idea, we must start with the classic objection.
This is that we can make up coherent stories ad lib entirely
without regard to the way the world is. By following our fancies,
and paying attention only to consistency, we can generate
comprehensive descriptions of possible but non-actual worlds.
For any such description to be true of the actual world requires
more than its mere presence in any such set. Russell made this
objection both to the early coherence theorists, and to their
8
B. Bosanquet, Logic, vol. ii, p. 287.
 CORRESPO:'iDENCE, COHERENCE, AND PRAGMATISM 239

positivist successors. 9 His argument can be put like this.

Suppose we have a coherent and reasonably comprehensive
description of some historical period. Call the system, S. Then
we might maintain or increase coherence, and increase com-
prehensiveness, by adding the proposition that some late pre-
late died on the gallows (S + E). But for all its membership of
the system, E might be false. So the ideals of comprehension
and coherence give us no reason to control admission to the
system in sane accepted ways. As a consequence they not only
admit falsehood, but also allow increasingly divergent systems,
each of which meets the ideals.
Bradley's response to this devastating objection is interest-
ing. Reminding us of the requirement that a system be made as
comprehensive as possible, he says:
But imagine my world made on the principle of in such a case
accepting mere fancy as fact. Could such a world be more comprehen-
sive and coherent than the world as now arranged? Would it be
coherent at all? ... The idea of system demands the inclusion of all
possible material. Not only must you include everything to be gained
from immediate experience and perception, but you must also be
ready to act on the same principle with regard to fancy ... 10
He then points out that this requires the admission of con-
tradictory fancies, and so is ruled out by the requirement of
coherence.
At first sight this is an evasion. Russell's charge is that if Sis
comprehensive and coherent - I abbreviate this to CC - or
approaches being so to some degree, then so is S + E, to a
greater degree, in spite of the falsity of E. Bradley's retort is that
a different set is not CC. This different set includes what I shall
call a pedigree, telling how E got into the first set. So the set he is
considering is S + E + F, where F tells us that Eis the product
of fancy. In fact Bradley argues that this set is only CC if the
principle on which E is allowed in is accepted - in other words,
only if S + E + F + G is CC, where G generalizes the principle
that you can add E on the basis of F, i.e. it endorses the
admission of any old fancy. And this set is not CC, because the
principle enjoins the admission of inconsistencies. But Russell
wanted to consider only S + E. So Bradley missed his point.
9
B. Russell, An Inquiry into Meaning and Truth, ch. l 0.
1
° F. H. Bradley, 'On Truth and Coherence', Essays on Truth and Reality, p. 213.
240 LANGUAGE AND THE WORLD

Or perhaps Russell missed Bradley's. A coherence theorist is

quite within his rights to describe the virtues of his system of
beliefs how he likes, provided that he avoids smuggling in an
independent reference to the truth of any of the members. So
Bradley may insist that his preferred system is, as far as pos-
sible, what we can call controlled: it should not contain elements
without also containing a pedigree for them, and once it con-
tains an element on the basis of a pedigree, it should allow any
others with equally good pedigrees. So whenever an addition to
a system is proposed, the test may be run on the element and the
principle on which it is proposed. And then Russell's counter-
example fails, because as Bradley points out, a fanciful pedigree
could be given to any belief at all. Bradley seems to regard the
ideal that a system be controlled as implicit in the idea that it be
comprehensive. But it is worth separating it out. The proposal
then refines into regarding truth as membership of a CCC -
controlled, comprehensive, coherent system. The essence of the
defence is that the wrong type of control, applied comprehen-
sively, lets in inconsistent pairs ofbelief(you mightjust as well
believe anything you first think of, as that the prelate died on
the gallows), destroying coherence.
The defence is subtle. But does it work? From a common-
sense standpoint the trouble with fanciful origins for a judge-
ment is precisely that fancy might just as well have issued the
opposite judgement, regardless of which is true. That is the point
of Russell's challenge. Bradley retorts that he can discriminate
against fancy, and in favour of observation, memory, induction,
and sober practices of enquiry and judgement, just as much as
Russell. The ground of his discrimination is that a comprehen-
sive use of fancy destroys coherence. A CCC system must not
use a method of control with this property. Well, suppose we
followed a policy of admitting fancies-tailored-for-consistency?
I might inflate my belief system by adding fancies, only paying
judicious attention to their coherence with anything else also
believed, like any good story-teller. What is wrong with that?
From the common-sense standpoint it is uncontrolled in the
sense that it has an even chance of delivering a judgement or its
negation regardless of which is true. But Bradley has to earn this
concept of truth. He has to describe the virtues of proper
control, comprehension, and coherence, without smuggling in the
 CORRESPO:\TDE'.\i'CE, COHERE~CE, AND PRAGMATISM 241

idea that they are midwives to truth. Story-telling is a poor policy for
delivering the truth. But it might be a good policy for delivering
CCC systems ofjudgements. 11 If Bradley then claims that it is
not, because it involves the wrong sort of control, we need to
know where he gets this standard from.
It will help to remember here the relationship between moral
truth and progressively more admirable systems of attitude.
The development of the last chapter was possible just because
we have, in our own normal attitudes, the basis for comparing
and preferring one system to another. This gives a content to
the judgement that one system is "nearer the truth" than the
other. In the ordinary case it is harder to say what is wrong
about belief systems controlled in bizarre ways, except that they
are likely to contain falsehood. And if this is so, there is no
prospect of a genuine elucidation of truth in this direction.
The same result affects philosophies which, out of admiration
for scientific method, construct a notion of truth as membership
of an ideal scientific theory. There are many difficulties with
this view, including that of saying what counts as progress, and,
as in the moral case, saying how we avoid the threat of the tree
structure. But the present problem is that of saying what is so
good about progress, except that it leads to the truth. Consider
this analogy. It would be no merit in a house to be one on which
all modern architectural design would eventually converge,
because there is nothing particularly superior about later
houses as against earlier ones. A limit definition of truth is only
possible if there is a firm sense of the virtues of the processes
leading to the limit. Bradley or Peirce can select features, like a
given kind of control and coherence and comprehensiveness.
But they need to say what is good about systems with those
features, without mentioning the idea that they deliver the
truth.
Pragmatism offers the utility of beliefs as the virtue needed at
this point. An improved system will be more useful than its
ancestor: the virtue of scientific progress, from which truth can
be constructed, is that of achieving ever more useful systems.
This need not be the crude idea that all and only beliefs which
it is useful to have are true. That is open to familiar counter-
11
And if only we could believe in stories, it would be a g-ood policy for removing- the
anxieties of doubt, which Peirce regarded as one of the main incentives to inquiry.
242 LANGUAGE AND THE WORLD

examples. There are man·y things useful to believe, although

false, and vice versa. A little credulity here and a little ignorance
there makes for success in life. No, the virtue should be attached
to the kinds of control admitted: the judicious use of the senses,
for instance, as opposed to fancy and invention. It is hard to see
how it could be generally useful to form beliefs without respect-
ing the judgements delivered by the senses, by memory, or by
sober processes of enquiry. Beliefs direct our actions, and
systems of belief direct habits and courses of action which may
or may not be useful. Surely there is a virtue in those controls
which make for useful systems?
There is, of course, but it seems a doubtful place to build
truth upon. The difficulty is that so much of what we normally
think about utility presupposes a notion of truth. The reason
why it is useful to believe that there is a typewriter here, in the
face of this morning's experience, is that there is one. My utility
is increased by my capacity to move around in, control, and
react with a world which my senses and my judgement tell me
about. That they tell me about it truly seems to explain their
utility, rather than be in any sense a construct out ef their utility.
Epistemologically we have the same direction. Knowing how
well my ends are being served is just an aspect ofknowing about
the world. It needs the 'proper' use of the senses, memory, etc.
A different, and in some ways appealing, suggestion is that
nature has solved Bradley's problem for him. Bradley is talking
about systems of belief, not of propositions in the abstract. Now
there is nothing difficult about manufacturing oddly controlled
systems of propositions, but they will not be systems of belief if
nature commands us not to believe them. And she often does. If
we open our eyes we have little or no control over what we see
and believe as a result. Conversely we cannot believe what we
know ourselves to have just invented. (Thus one thing amiss
with Russell's example is that we cannot add the belief to our
historical system while we know ourselves.to have just invented
it. We would have to go through some process of attaining
conviction.) Perhaps nature has built us to be knowers: truth
might attach to members of naturally controlled, coherent, and
comprehensive systems of beliefs. When we alert our senses
nature forces us to the beliefs which then flood in. The most we
can do is to use those, and our best ways of forming CCC
 CORRESPONDENCE, COHERE'.\i"CE, AND PRAGMATISM 243

systems. If truth is anything mor~ than th~s, how co_uld we

possibly regard ourselves as knowmg anythmg about 1t? And
how would it matter, if it were so remote from the human world
we find ourselves in?
6. What is Right about Correspondence?
I hope that at this point a coherence theory begins to look
attractive. But it tends to keep bad company, particularly in
contemporary developments. In particular is it just accidental
that a coherence theory of truth associates with Idealism - a
view of the ordinary world as in some sense a mental construc-
tion? The loss of the "barely given" fact, and the difficulty of
describing the control of a system of beliefs by experience, seems
to leave an image of such a system as entirely self-absorbed.
Beliefs are fitted with other beliefs, everything adjusted to make
for a CCC system, but the image is of a system disconnected
from the world. Neurath's famous picture of the system of
science as a boat - a structure with no foundations, but with a
strength given by the mutual support of members - left us
searching for the control of the shape of the boat. If the tides of
experience just throw up further planks then they are merely
our most fertile source of boat-building material: so what can
we do but walk around inside the boat, testing and adjusting for
coherence? At this point we go back to starting at the typewriter
- that's a fact, that's what makes true the belief that there is a
typewriter on the table. Surely it is just common sense, not the
higher reaches of theory, which sees us as responsive to mind-
independent facts? In the Mikado (see chapter epigraph) Pitti-
Sing's tale might have been unassailable, but it certainly wasn't
true.
What makes true the belief that there is a cat in the garden, is
there being a cat in the garden. There being a cat in the garden
is, on the face ofit, a very different kind of thing from there being
a system of beliefs, controlled in whatever way we like, which
would have this belief as a member. It is the cat which we
respond to, not anything mental. We are good instruments for
detecting cats, and that is why, if we are careful, our beliefs
about them tend to be true. Suppose, then, we have, as part of
this common sense, a rough notion of what counts as a sane,
sober practice of enquiry, enabling us to find out whether there
244 LANGUAGE AND THE WORLD

is a cat in the garden. We can now formulate what I shall call a

correspondence conditional:
( 1) If I form only beliefs with a proper pedigree, and end up
believing that there is a cat in the garden, then there is a
cat in the garden; if the same is true and I believe there is
no cat, then there is no cat.
In other words, I am a good signaller of the presence or absence
of cats. My beliefresponds to the cat. IfI am i:o assert this, then
such correspondence conditionals as ( 1) need to be assertible. If
they are not, then our conception of the pedigree, the certificate
given by the sane sober practices of enquiry, leaves it discon-
nected from truth. Of course, we know we may be deceived on
occasion. The conditionals represent a general truth, not an
unbreakable one. It is the task of epistemology to make sure we
can believe such conditionals, and to explain why.
Now the reason why we are good at cats is nothing like the
reason why we are good signallers of duties, numbers, possible
worlds, and the like. We are good at cats because our senses
make us causally receptive to their presence or absence. This
may not be the whole story about why we can assert (1), but it is
certainly part of it. It is part of our conception of the world -
part of our system of beliefs - that it is only because we are in
this way responsive that we have any right to the correspond-
ence conditional.
However, if a coherence theorist is unwise enough to present
membership of a CCC system as just the same thing as truth, his
grip on these correspondence conditionals becomes very
strange. Consider by the side of ( 1):
( l *) lfI form only beliefs with a proper pedigree, and end up
believing that there is a cat in the garden, then it is true
that there is a cat in the garden; ...
This differs only in adding 'it is true that' to the consequent of
the conditional, which is legitimate because of the transparency
property. Now substitute a coherence conception of truth:
( 1 **) If I form only beliefs with a proper pedigree, and end
up believing that there is a cat in the garden, then the
belief that there is a cat in the garden is a member of
some best CCC system of beliefs; ...
 CORRESPONDENCE, COHERENCE, AND PRAGMATISM 245

(I**) seems quite different from (1). It threatens to be a taut-'

ology, because the notion of proper pedigree and of the best
CCC system are simply interdefi,nable: any notion of proper
pedigree can yield a corresponding notion of the maximal
system of beliefs with such pedigrees! But common sense does
not regard us as good signallers of cats because of any such
tautology. It demands that we look to the cat and our causal
responses to it. It does not want to see us as good signallers of
whether our beliefs would enter into a maximal coherent and
comprehensive set of beliefs controlled in the ways we find
natural.
The point is obvious if we imagine two different views of the
virtues of systems of belief A hardline empiricist ("Carnap")
may choose one kind of pedigree as the only worthwhile one
· (Ve), whereas a hardline Catholic ("Newman"), say, may
choose a somewhat different set, ( V:J. Suppose that each set of
virtues allows us to define a maximal set of beliefs possessing
them: V*c and V\, respectively. Then consider:
If a belief has ~ then it belongs to V*c
If a belief was V" then it belongs to V\
Each of these is harmless - an uncontroversial tautology. Each
of Carnap and Newman can believe each of them. But this just
shows that they fail to capture the real point of difference, which
is that Carnap thinks that if a belief has Ve and not any other
kinds of feature, then it is true, whereas Newman disbelieves
this, but believes that if a belief has V", and not any other kind of
feature, then it is true. As in the case of ethical disagreement,
where there is divergence over the right standards, we cannot
see the disputants as each with his own, standard-given notion
of truth. For that makes the dispute disappear.
In sum, there is a distortion in any definition of truth in terms
of membership of a system of belief, where the members need
some correct pedigree to enter. The distortion is rather like that
which assailed the "democratic harmony" theory of universals
(3.3). Our judgement that a cat is in the garden is made true, if
it is true, by the cat being in the garden. The issue of how other
people would judge it is BO part of this truth-condition. Nor is
the question of whether the belief that it is would enter into any
proposed system of belief. We don't, as it were, look sideways,
246 LANGUAGE AND THE WORLD

either to other people or to systems of belief. We look at the cat

and look round the garden. 12
It is a general maxim in science that a good new theory
should be able to explain what was attractive about a
superseded old theory. In view of the rooted and stubborn place
that correspondence with the facts has in normal thinking
about truth, it would be surprising if there was nothing right
about it. This is now becoming apparent. The notion of cor-
respondence at least registers a half-truth: whether or not true
beliefs correspond with the facts, true believers must certainly
respond to the facts. The world has cats in gardens, and it has
believers: the believers must be able to see themselves as prop-
erly responsive to the facts they take to obtain. Our causal
theory of the world insists that the fact that there is a cat in the
garden (or the state ofaffairs of there being a cat in the garden)
can bring about many things: the shadow on the parsley, the
absence of the birds, or my realization that it is there. 13 If my
belief does not respond to the fact then I am a poor register of
cats being in gardens, and my confidences in these matters are
disconnected from their truth. Neurath's boat, the ship of our
interlocking confidences, may need no foundations. But it does
need to be built in response to the states of affairs which obtain
in the world in which it sails. In the next section I show how this
demand makes room for a proper theory of knowledge. Mean-
while the question is whether these virtues in the concept of
correspondence undermine the coherence theory entirely. Does
any such theory founder on the distinction between ( 1**) and
(l)? Must it misrepresent the ways in which we are forced to
12
This is not the objection (see Hartry Field, 'Realism and Relativism',joumal ef
Philosophy ( 1982) p. 556) that some propositions are true but not members of any CCC
system - e.g. a proposition of the form 'there have been exactly n dinosaurs', which
nobody could ever know or even reasonably believe. There are such propositions but it
is a mistake to urge them against a proper coherence theory, which will reply that our
conception of the world permits truths which would enter an idealized set of beliefs,
created in accordance with our virtues - in this case the virtues attending proper use of
observation, memory, and counting, in situations to which we don't actually have
access. The objection in the text is not that a coherence theory delivers the wrong class
of truths, but that it delivers the wrong conception of truth.
13
It is common to find contemporary philosophers speaking as though only events,
not states of affairs, enter into causal relations. This is just wrong. It makes nonsense of
the notion of a stable equilibrium in which a thing's being at a place or being thus-and-so
causes nothing to change. Events are changes in states of affairs, and lack of change can
need causal explanation just as much as change.
 CORRESPONDENCE, COHERENCE, AND PRAGMATISM 247

think of beliefs being sensitive to facts? We must remember

here the subtle route whereby an elucidation of truth might
generally escape from Frege's regress. This meant giving up
ambitions to provide a synonym for truth, but insisting that a
substantive conception of success in judgement is nevertheless
possible. Now our causal theory of the world is undoubtedly a
central component of our system of beliefs. Surely a coherence
theorist should not deny these elements of common sense. His
role must be to incorporate them (to speak with the vulgar but
think with the learned, again). In other words, he must try to
respect the real content of conditionals like ( 1). He must recog-
nize the causal element underlying them. He must avoid the
mistaken equation giving ( 1**). But he must still maintain that
he has a distinct conception of success in judgement, and that it
sees success in terms of membership of a CCC system.
Is this combination of attitudes possible? The idea would
have to be that internally, we talk of states of affairs, or of
correspondence, and of the causal theory allowing ( l); but that
this permits an external reflection that all this is part of our own
system of beliefs, and that truth accrues to any such system in
virtue of its coherence, comprehensiveness, and control. The
model would be the quasi-realist's approach to the mind-
independence of moral truth. He faced the similar obstacle that
we think as though the wrongness of doing something were
quite different from any question of the sentiments we happen
to feel about it. He respected this claim, seeing it as in effect
internal to any admirable system of attitude, but not seeing it as
settling the issue of the construction of moral truth. The parallel
coherence theorist respects the claim that as believers we are
sensitive to the presence of the cat, the very state of affairs which
makes our judgement true, ifit obtains, and false ifit does not,
but rolls this into the overall CCC system, whose virtues he
hopes to describe, and out of which he supposes himself to
construct a notion of truth.
These are deep waters, and nobody, as far as I know, has ever
swum very far in them. It is as though the CCC conception of
truth is, like Idealism, a kind of optional gestalt-switch: you can
see things, wilfully and in the study, as if the virtues of systems
of beliefs are all that we have to build a notion of truth out of,
just as you can think as though the world is a mental construe-
248 LANGUAGE AND THE WORLD

tion. But out of the study the vision goes; objects, facts, re-
emerge and demand their independence of us and our
believings.

7. Epistemology Regained
There are writers who associate epistemology, as the systematic
account of ourselves as knowers of the .world, with the cor-
respondence theory of truth. Since they regard that as refuted,
they regard epistemology as misdirected as well. It is important
to understand that this is not so.
As I have tried to stress, there is not really a correspondence
"theory" of truth: there is rather an invitation to think of the
relation between true belief and whatever it is in the world
which makes it true. This invitation may lead to bad develop-
ments: to the idea of the mind's awareness of fact as something
which, favourably, is uncontaminated by judgement, and
purely passive; or to the idea of thoughts as pictures in the mind
copying the world, or to the idea that each individual judge-
ment has its own identity regardless of its associations with any
others in a body of belief, and is in tum made true by one
isolated, self-subsistent state of affairs. Any of the bad develop-
ments could prompt a bad epistemological tradition. For
example, the idea of vision as the paradigm of knowledge, and
yet itself different from judgement, undoubtedly dominated the
theory of knowledge throughout the seventeenth and eight-
eenth centuries. (God just has to look to know.) 14 The goal of the
theory of knowledge was to show how our ideas represented
reality - a correspondence which forever seemed fugitive,
because we only had our ideas to go on. It is then tempting to
jump to th& view that if this concept of correspondence, this
accent on copying and re-presentation, is jettisoned, all the
problems of knowledge disappear with it. If we substitute a
pragmatic or coherence conception of success in judgement,
that success seems more assured, and scepticism easier to avoid.
The truth in this is that epistemology may become easier.
It may no longer be possible to mount a wholesale contrast
between our total CCC system of belief, and The Truth. We
may have no concept of truth, on the coherence development,
14
I owe this point to Edward Craig.
 CORRESPONDENCE, COHERENCE, AND PRAGMATISM 249

which entitles us to make any sense of the idea that all of our
best, most careful, refined, useful, simple, virtuous ways of
building belief are leading to falsity. There would then be
no problem of answering wholesale scepticism. Descartes's
demon, who allegedly systematically feeds us misinformation to
delude us into believing that we are in a world which does not
exist, would fail in his plan: he would put us in a world which is
truly as we take it to be (so would Berkeley's God). Nor, of
course, does the ship with which a system of belief is compared
need foundations - the bare, incorrigible, passive reception of
fact, 'imprinted without the intervention ofjudgement. But the
ship does need constant inspection. As we walk around it, we
may find it hard to preserve coherence, and may also find that
doing so demands giving a good deal of time to traditional
epistemological problems. Otherwise a system breaks down
from the inside.
I have cautiously said that if we sympathize with the coher-
ence theory wholesale worries of a sceptical nature may disap-
pear, not that they must do so. This is because I mistrust one
more proposal for marking the realist off from the anti-realist
(here represented by a coherence theory). This proposal is that
the realist, for better or worse, can make sense of the idea that
even an idealized successor of our own system of belief might be
false, whereas the anti-realist can make no sense of this possibil-
ity. If the anti-realist adopts the Peircean equation of'Tp' with
'pis a member of M*' then, of course, he cannot understand the
suggestion that the members of M* might be false: membership
of M* is defining truth. Hence, if this equation is accepted, there
is no space for wholesale scepticism. But looking at it the other
way round, we can ask why it should be a virtue in a system that
this space should disappear? What is so good about a system
which does not permit its adherents to formulate a contentful
doubt about its own overall adequacy? The undergraduate
student who is impressed by the possibility of Cartesian
demons, hypotheses that we are all brains in a vat, and so on,
may be using legitimate aspects of our own conceptual scheme-
ones which would be present in any idealized successor as well.
If this is right, the coherence theorist or anti-realist would
allow sayings like 'perhaps all our experience is a dream;
perhaps our best procedures still fail to give us the truth'. He
250 LANGUAGE AND THE WORLD

would see the concept of truth (explained, remember, in terms

of other virtues) as elastic enough to allow this, because
amongst the virtues he ranks highly expressions of fallibility
and consequent attempts to secure the boat. Although many
writers, most notably in recent years Hilary Putnam, believe
the contrary, it would not be true that only a crude "God's eye
view" or "copy" theory of truth permits wholesale worry. The
coherence theorist could share it. So he would avoid any defini-
tion such as Peirce's, for it closes the space which I want the
coherence theorist to keep open. Naturally, it does not trouble
me that this leaves no precise definition of truth. For we have
already seen good reason why the investigation into truth
should be more oblique than this. Already in the last section we
found a difference between ( l) and ( l **) which is itself
sufficient to force the coherence theorist to avoid direct analysis.
In effect I am allowing him to do this, and then to exploit the
possibility of incorporating into his idealized or imaginary set of
virtues the virtue of insisting upon the fallibility of all
judgement.
However, I am fully aware that this is a delicate and surpris-
ing position. In what follows I shall not rely upon it. I return
instead to the relatively local and internal problems of
epistemology which arise as we walk around our boat, attempt-
ing to secure its internal coherence.
Thus, consider the incoherence from which a belief-system
will suffer if its account of the correspondence conditionals
leaves them unassertible. These conditionals allow us to see a
favoured method offorming opinion as a signal ofits truth. Ifwe
find we cannot assert this, then the system fractures: we will
have confidence in some element, but no confidence that this
confidence is placed in a way which leads to the truth. Our
naturalized epistemology - our best description of ourselves as
knowers of the world - must avoid this extended incoherence:
extended because there is no actual contradiction in both
believing that p, and believing that the belief that p is formed in
a way disconnected from the truth ofp. I take i.t as obvious that
extended incoherence is a defect. It would be like trusting the
readings of an instrument whilst supposing it to be discon-
nected from whatever it is allegedly registering. And incon-
sistency can quickly develop. For if the system is rich enough to
 CORRESPONDENCE, COHERENCE, A'.\iD PRAGM,\TlS:\I 251

speak about chances, we can express our confidence by saying

that there is a good chance that there is a cat in the garden, and
the sceptical disconnection by saying that the observation
leaves no better chance that there is a cat in the garden than
there would have been had it not been made.
Now it is very easy to say things in epistemology which
introduce failure of extended coherence - in other words, which
leave the correspondence conditionals unassertible. The thrust
of traditional sceptical problems in philosophy is to challenge
our conceptions of ourselves as good signallers of various kinds
of truth. So a coherence theorist must pay as much attention to
those problems as anyone else: it is a mistake, and unfortunately
a common one, to suppose that they affect only some crude
foundationalism or crude conception of correspondence with
the facts. On the contrary, since virtues falling short of extended
coherence will certainly not serve to define truth in a system, a
coherence theorist needs to make sure he can retain it. I shall
illustrate this point, and the difficulty of meeting it, with refer-
ence to the bent-predicates, Goodman's curious dimensions of
similarity and dissimilarity, introduced in 3.2.
The traditional problem ofinduction is that of showing why it
is reasonable to put belief in the straight continuation of some
regularity: reasonable to expect future blood samples to be red
rather than red 8 R, to use the example of that section. We can
define a "space" of alternative things which could happen to
the colour of blood throughout a certain period. Staying red is
just one of many conceivable happenings. Goodman's predi-
cates take some alternative - changing yellow at some particu-
lar time, for instance - and define a property so that retention of
it means following that alternative. A thing remains red HR ifit
turns yellow at the end of 1985. We expect presently red things
to remain red; a red 8 R user is supposed to expect them to stay
red 8 R -i.e. to turn yellow. Our prediction that blood will remain
red means that we expect him to be wrong. He expects us to be
wrong. Why have we any reason to trust our opinion against
his?
The answer which introduces extended incoherence sees
nothing but convention or accidental linguistic arrangement to
which to point. We speak and have always spoken a language
which classifies and predicts in terms ofredness. We could have
252 LANGUAGE AND THE WORLD

spoken one, on this account, which classified and predicted in

terms of red 8 Rness. Had we done so we would now be sitting in
equal confidence that blood will turn yellow at the end of 1985.
But the fact that we use English and not this rival, English 8 R, is
as conventional, and as much subject to our habits and arrange-
ments, as any other aspect of vocabulary and syntax.
This account of the pedigree of our belief that blood will be
red after 1985 will simply not cohere with the confidence we
have in that modest prediction. Mention of convention is relev-
ant only if our needs, natures, or causal interactions with col-
ours and other features of things, might equally well have left us
holding the inconsistent prediction. Convention fills the gap
between such things and our actual arrangements (see chapter
4). But then it is an explanation of why we hit upon one
arrangement and one prediction, which is quite disconnected
from anything which affects the likelihood that the beliefis true.
In this it is like conceding a fanciful origin for a belief; indeed
fancy, or the toss of a coin, or any random or accidental happen-
ing, is just the kind of thing to determine which convention to
adopt out of a space of competing, equally serviceable ones. But
you cannot coherently regard yourself as a good signaller of the
future colour of blood, if you also think that what determines
the colour you predict is itself quite disconnected from anything
affecting or affected by that colour. The correspondence condi-
tional becomes unassertible, and extended coherence fails. The
general point is that unless there can be a causal story connect-
ing the likely future colour with the factors making us select one
commitment, it will be incoherent to maintain the commit-
ment. But if there can be such a story, showing us why it is
natural or inevitable for us to speak the one language and not
the other, the point of mentioning linguistic arrangements and
conventions is lost. Our predicting that the blood will be red
and not red 8 R will be explained not by our linguistic conven-
tions, but by whichever story best explains why we must make
our actual classifications, and not others.
Fortunately, if the arguments of3.2 are right, it is wrong in
any case to see our preference for redness rather than red 8 Rness
as in any sense arbitrary or conventional. For we can form no
real understanding of how these other dimensions can really be
perceived as free from change. Using this point to found a better
 CORRESPONDENCE, COHERENCE, AND PRAGMATISM 253

theory of ourselves as good signallers of the future needs a

head-on attack on the classical problem of induction - of why
the future or the unobserved is reliably indicated by the present
or the observed - and this is therefore a traditional problem of
epistemology which a coherence theory of truth must face. The
extended incoherence which must be avoided is not imposed
only by some exalted conception of truth. Nor is it the product
of a delusive desire for certainty or proof where none can be had,
nor of a desire for "foundations" of systems of belief, when all
that can be had is the mutual support of elements of such
systems. It is just that when a perfectly natural account of
ourselves as judgers and signallers involves a conventional
explanation of our commitments, it imposes a definite chance of
there having been other commitments, and it is this chance
which produces the incoherence. It emerges from something
presented as a piece of straightforward human psychology.
It has been extremely common in epistemology this century
for philosophers to think they reach bedrock in mention of our
ways, our conventions, arrangements, or "linguistically
imposed" practices (the "linguistic turn" again). The idea that
a particular claim to knowledge can be justified ifit is part ofa
"language game" permitting us to assert it, has become
extremely widespread since Wittgenstein introduced the term.
But what status can this give to any belief? That depends on the
way we regard alternative arrangements. If it is just a matter of
convention that we have the one game and not another, then
extended incoherence threatens, and we do nothing to certify
the claim as likely to be true. 15 The conventionalist geist blows
philosophers who are quite different in spirit from Wittgenstein
dose to this peril.
Consider, for instance, Quine's attempt to describe the way
in which experience controls systems of belief. Quine is acutely
sensitive to the problems of the holistic nature of systems and of
the difficulties with "the given". But he wants some conception
of an observation sentence, or verdict properly prompted by
experience, and with genuine authority over what beliefs can be
truly held. He describes it like this: "a sentence is observational
15
\\"ittgenstein himself, in spite of careless use of notions like 'game', seems largely
to think that nature rather than convention forces aspects of our intellectual lives onto
us.
254 LANGUAGE AND THE WORLD

insofar as its truth value, on any occasion, would be agreed to by

just about any member of the speech community witnessing the
occasion." 16 The laudable intention behind this definition is to
exclude highly theoretical remarks, dependent upon specific
background knowledge, from being regarded as observational.
But Quine also intended his definition to "accord perfectly with
the traditional role of the observation sentence as the court of
appeal of scientific theories". 17 It is not clear that his definition
does that. For what is the authority which resides in the speech
community and what is its source? Just about any English
speaker, looking out of the window, would agree that the grass is
green. But is it the fact that they would so agree which gives this
belief its authority as an observation? Surely that depends upon
our account of why the agreement arises, or, in other words, of
what is so good about this feature of the English-speaking
community. Then if at this point we simply cite conventions, or
what our speech community has us do or say, implying that
there could equally well be other arrangements leading to
verdicts inconsistent with ours, then we lose any grip on the
authority of the report. Once we suppose that equally meritori-
ous languages could lead their users to look out of the window
and dispute whether the grass is green, we have lost any right to
see ourselves as good signallers of colour. The correspondence
conditionals become unassertible. and extended incoherence
sets in. Quine did not intend his definition in this con-
ventionalist way, but others, notably Richard Rorty, have so
construed it. 18 The difficulty is that if the speech community is
mentioned at all, then unless we have some story about why the
community's reaction to experience is reliable or authoritative,
the road to conventionalism is wide open. Rorty takes Quine to
imply that it is the consent of the people which makes it true
that in saying any particular thing we are making an obse.rva-
tion. But then, unless we can hope for some theory according to
which that consent is itself a reliable sign of the fact reported, we
have incoherence, and conventional origins of the consent block
any such theory. They do this because convention explains a
16
W. V. Quine, The Roots of Reference, p. 39.
17
\\'. V. Quine, 'Epistemology Naturalized', in Ontological Relativity and Other Essays,

p. 87.
18
Richard Rorty, Philosophy and the Mirror of Nature, esp. pp. 174 ff.
 CORRESPONDENCE, COHERENCE, AND PRAGMATISM 255

feature only if any of a space of other different features would

have served us equally well.
The theory of knowledge has always struggled to form some
conception of the control of belief by experience and observa-
tion. It is not my purpose to enter that struggle here. The point
is that the conventionalist stance, either towards Goodman's
problem or towards the definition of observation sentences, is
incoherent. The search for something better is a quite proper
part of naturalized epistemology, even if a coherence theory of truth is
favoured. For the incoherence of a false step at these points is an
internal, structured fault in any system of belief, noticeable as
we inspect our boat. It is not a fault, or alleged fault, which only
exists on a correspondence vision, whereby we can, as it were,
detach ourselves entirely from any of our actual beliefs, and
adopting a "God's eye view", or standing on an "Archimedean
point" outside the whole system, discuss whether it depicts
Reality well or badly, draped or undraped. We can attempt to
understand ourselves as good signallers of the truth without
worrying about any such idealized perspective. The con-
ventionalist turn represents a false step in any such attempt.
And reference to our ways, our customs (games, conventions,
languages, consensuses), cannot be bedrock when we make the
attempt, for, as I have tried to stress, unless such reference is
supplemented by an account of why these arrangements lead us
to opinion which is well connected with how things are, it is no
better than mentioning fancy or any other random pedigree for
a belief, and leaves the correspondence conditionals quite
unprotected.
It is only when we believe that the correspondence condition-
als are properly assertible that we can be happy with ourselves
as registering the truth, or happy that our procedures for form-
ing belief are reliable, or scientific. It is because the proponents,
say, of some wilder psychoanalytic theories, or of some
dogmatic versions of Marxism or other social theories, do not
appear to form beliefs in a way which enables the belief formed
to vary with the truth about the issue, that they are deemed
unscientific: as soon as we can say that someone would have
believed what he does regardless of what the facts are, we
remove any respect for his opinion. I think it was this demand
which was approached, but not quite accurately, by Sir Karl
256 LANGUAGE AND THE WORLD

Popper's famous demarcation criterion, according to which to

behave scientifically is to be open to falsification, whereas to
behave unscientifically is to hold beliefs as immune to falsifica-
tion. The difficulty with this, as Lakatos showed, 19 was that the
most respectable, paradigm scientists often hold theories in
ways which the criterion would condemn. An adherent of
Newton in the nineteenth century might have been quite
unable to conceive of circumstances which would lead him to
declare the general principles of mechanics false; any disturbing
results would be deemed to be mere anomalies, and met with
relatively ad hoc adjustments to the overall theory. And why
should we demand higher standards from psychoanalysts or
social theorists than from physicists? According to the view I
give the real difference is that the physicist has a coherent story,
showing why his belief in Newton's laws depends on the facts in
the way that beliefs ought to; ifhe is genuinely unscientific the
psychoanalyst or social scientist will have no parallel story to
offer. Closing one's mind to falsification is not at all the primary
criterion of demarcation. It would serve, at best, as a secondary
symptom, that the theorist is not concerned about the need to
protect the correspondence conditionals, and it is here that his
unscientific vice lies.

8. Truth Again
We are now armed with some sense of what is right about a
correspondence theory, and of what is distinctive and attractive
about a coherence theory. But we have also learned to be wary
of attempts at direct analysis, for such attempts tend to distort
the simple platitudes, such as that what makes it true that there
is a cat in the garden, is there being a cat in the garden. The
exploration so far conducted does, however, cast some light on
the question hanging at the end of the last chapter. A system of
attitude, admirably controlled, is in many ways like a system of
belief. It permits argument, inference, improvement, deteriora-
tion, assessments of acceptability, and a quasi-realist construc-
tion of truth. Is it then right to say that a moral commitment is
true or false in just the same sense in which any other belief is?
19
'Falsification and the Methodology of Scientific Research Programmes',
Philosophical Papers, vol. i.
 CORRESPONDENCE, COHERENCE, AND PRAGMATISM 257

The difference will come, of course, in the theory of the cor-

respondence conditionals. There is no causal story, parallel to
that which must be given to justify ourselves as good signallers
of cats, to justify ourselves as good signallers of virtues, duties,
obligations, and goods.Just because values and the rest are the
children, and not the parents, of our sentiments, there is no
need for any such theory, and its absence is no obstacle to the
regulative use of truth in moral contexts. Does this make moral
commitments true in the same sense as others, or only in a
different sense? I do not greatly commend the question. What is
important is our right to practise, think, worry, assert, and
argue as though they are. The extent to which we see this as a
valuable fiction will depend in the end on how much we
sympathize with the coherence gestalt-switch which permits us
to see all truth as constructed out of membership of virtuous
systems. If we sympathize with this in general, then moral truth
will be just a kind of real truth; if not, then we will regard it is a
legitimate, but imaginary, focus, upon which the progress of
opinion is sighted. But at present I know of no way offorcing the
issue.
There has been one notable absentee from this discussion of
truth. Many philosophers believe that the work of Tarski either
augmented or even superseded more traditional problems, such
as the debates I have related. The so-called "semantic concep-
tion of truth" might seem a promising, modem, approach to the
area. To see whether this is so, and what Tarski's achievement
actually was, demands a separate chapter.

Notes to Chapter 7
7.1 The most useful single collection on truth is still Truth, edited by
George Pitcher. But substantive theories of truth have not been in
philosophical fashion recently. Partly this is because of the arguments
of the next section, partly because of a vague feeling that Tarski and
Wittgenstein showed that there could be nothing to say about truth
itself, but only about the way truth-conditions are attached to
particular sentences in particular languages. And a great deal of the
literature is involved with technical problems, arising from paradoxes
of the liar family, or arising from the attempt to avoid admitting a
general concept of truth (see below, under notes to 7.3).
258 LANGUAGE AND THE WORLD

7.2 There is a careful discussion of the transparency argument in

Peter Carruthers, 'Frege's Regress', Proceedings ef the Aristotelian Society
( 1981).

7.3 The redundancy theory is initially found in Frege. In 'My Basic

Logical Insights' (1915) Frege says: "If I assert 'It is true that
sea-water is salt', I assert the same thing as if I assert 'sea-water is
salt' ... This may lead us to think that the word 'true' has no sense at
all. But in that case a sentence in which 'true' occurred as a predicate
would have no sense either. All one can say is: the word 'true' has a
sense which contributes nothing to the sense of the whole sentence in
which it occurs as a predicate." (Posthumous Writings, p. 251.)
Similarly F. P. Ramsey: "there is really no separate problem of truth,
but merely a linguistic muddle" (The Foundations ef Mathematics,
p. 142). Like Frege, Ramsey saw that the transparency property
enabled us to paraphrase away direct ascriptions of truth to judge-
ments: it is true that p reduces down to straightforward: p. But he
realized that this does not by itself cope with less direct contexts, such
as: 'everything he says is true'. The natural suggestion would be: (Vp)
ifhe says that p, then p. But this is not uncontroversially well-formed,
and philosophical logicians have disputed the matter ever since. The
problem is whether the variable 'p' is functioning as a variable
normally does, in which case ' ... then p' is not well-formed, since the
terms needs completion by a predicate: (Vp) ifhe says thatp, thenp is
... ; and the only way of filling in the dots would be to reintroduce
'true'. The literature on this knotty point includes:
C.]. F. Williams, What is Truth?
J. Mackie, 'Simple Truth', Philos1Jphical Quarterly (1970).
D. Grover et al., 'A Prosentential Theory of Truth', Philosophical
Studies ( 1974).
P. Geach, Reference and Generality, *69 ff.
But as the discussion in my chapter shows, I do not believe that the
central issue over the redundancy theory is its formal ability to do
away with mention of truth from all contexts. The real question is the
possibility of a substantial theory of what success in judgement
amounts to.

7.4 "The critics of the given have a simple and strong case ... "
I hope it is clear from this section that this is only so when they are
attacking a particular conception of the footing that judgements have
in experience - the conception that fails to realize the way in which
judgement conditions experience. It should not be inferred from this
 CORRESPONDE:\i"CE, COHERENCE, AND PRAGMATIS'.\1 259

that there is no sense in which things are "simply given" to us. There
is no avoiding that: if you are run over by a bus it is simply given that
you are. There is no control for you, no exercise of judgement or
choice or interpretation of facts. Perhaps one thing which makes it
hard for students in this area is that so obviously many facts are
simply borne in on us whether we like them or not. In this sense the
world presents us with so much, and what it presents is given. This is
true, but it does not affect the point that to take what it presents in any
given way is to exercise our understanding.

7.5 I have taken the British Idealists as the best source for the ideas
leading to a coherence theory of truth. There is a fascinating further
history in the way the logical positivists became embroiled in such a
theory: the theory is a permanent temptation for rigorous empiricists.
See:
C. Hempel, 'On the Logical Positivists' Theory of Truth', Analysis
(1935).
0. N eurath, 'Protocol Sentences', in Hanfling ( 1981).
I. Scheffier, Science and Subjectivity ( 1967).
J. A. Coffa, 'Carnap's Sprachanschauung Circa 1932', Philosophy of
Science Association ( 1976).
B. Russell, An Inquiry into Meaning and Truth.
In the last of these (ch. 10) Russell repeats the arguments he used
against Joachim and Bradley, but against the positivists' version of
the coherence theory, represented by Neurath and Carnap. (Russell
explicitly recognizes that the new, positivist doctrine is a repeat of the
Idealist doctrine, and formed in answer to the same pressures - see
e.g. p. 133). As in the earlier work, Russell refuses to allow a form of
coherence theory in which language still maintains a relation to
non-linguistic occurrences. He insists on seeing the coherence theory
as making the world of words a self-enclosed world, and this makes it
easy for him to ridicule it.

" ... pragmatism offers the utility ... "

It may seem strange that I present pragmatism here as a version ofa
coherence theory, whose distinctive stress is on utility as the virtue of
a system. It is more common to find it presented as though a
pragmatic theory of truth is a quite different animal altogether. But I
do not accept that it can be. (At present the title seems to be conferred
on almost anything from a vague sympathy with empiricism, to a
redundancy theory or disbelief in the prospects for any serious theory
of truth at all. See R. Rorty, Introduction to Consequences of Pragmatism:
260 LANGUAGE AND THE WORLD

Rorty sees the rejection of questions as the distinctive theme of what he

calls pragmatism.)

7.6 It is not common to find writers with anything at all good to say
about a correspondence theory of truth. So the introduction of the
correspondence conditionals represents a concession to common
sense rather than to any specific philosophical position of which I am
aware. But Putnam notices the need for such a theory (Reason, Truth
and History, p. 132) although generally hostile to the kind ofepistemq-
logy I want to protect. The line I try to sketch at the end of this section
is supposed to be reminiscent of Hume's famous division between the
attitude to the external world which we find forced upon us when we
think about it, and that which is forced upon us by nature. There is a
tantalizing amount of philosophy which consists in balancing such
images or visions of how things are against one another.

7.7 This attack on conventionalism can be followed up by consult-

ing 3 .2 and the notes to that section.
 CHAPTER 8

Truth and Semantics

GENERAL. I don't think we quite understand one another. I ask

you, have you ever known what it is to be an orphan, and you say
'orphan'. As I understand you, you are merely repeating the word
'orphan' to show that you understand me.
KING. I didn't repeat the word often
GENERAL. Pardon me, you did indeed
W. S. Gilbert, The Pirates ef Penzance.
For what does a proposition's 'being true' mean? 'p; is true = p. (That
is the answer.)
Wittgenstein, Remarks on the Foundations of Mathematics, I,
Appendix I, §6, p. 50.

1. Truth Theories: Tarski's Suggestion

In view of the struggles with predication, rule-following, and
judgement in the later work, it might be surprising that Witt-
genstein expresses such a dismissive attitude to truth. We
would expect him to be well aware of the difference between
offering a remark as a metaphor, invitation, prescription,
expression of attitude, or rule, and offering it as a judgement; to
isolate judgements it also requires some account of their dimen-
sions of success - truth and falsity. But we also found, in the last
chapter, some faint reason to respect this quietism. The trans-
parency property of truth provides an obstacle to head-on
attempts to analyse what truth is. And we found no sharp way
of developing either the coherence or the correspondence sug-
gestions into plausible definitions of truth. In fact, the only
equations which seem cast-iron are those of the equivalence
thesis - it is true that snow is white if, and only if, snow is white;
true that penguins waddle if and only if penguins waddle.
Turning to sentences we can say that the English sentence
'snow is white' is true if and only if snow is white; 'penguins
waddle' is true, in English, if and only if penguins waddle.
262 LANGUAGE AND THE \VORLD

It was the suggestion of Alfred Tarski that these equations

enable us to propound what he variously called a definition of
truth, or a "materially adequate and formally correct definition
of the term 'true sentence' ", or a semantic conception of truth
which makes precise and improves upon preceding notions.
Tarski's work inspired virtually all subsequent formal and logi-
cal approaches to language. But from the beginning it provoked
the most divergent reactions. To some philosophers it finally
solved the problem of the nature of truth. To others it was quite
irrelevant. Attempting to explain its real significance is treading
on sacred ground. I shall therefore give rather more textual
backing to my discussion than I have tended to do. The three
papers of Tarski which are important to the philosophy are the
classic first paper, 'The Concept of Truth in Formalized
Languages' (CTFL); 'The Establishment of Scientific Semantics'
(ESS); and 'The Semantic Conception ofTruth' (SCT). 1
CTFL begins: "The present article is almost wholly devoted
to a single problem - the defi-nition eftruth. The task is to construct
- with reference to a given language - a materially adequate and
formally correct defi-nition ef the term 'true sentence'." (Tarski' s
italics.) Tarski's papers make constant use of the ideas of set
theory, and even elementary modern expositions presuppose an
understanding of quantification theory which, at this stage, I do
not want to rely on. So the nature of the enterprise is hidden
from all except advanced students. I shall introduce it by
reference to an exceedingly simple language- so simple that the
difficulties Tarski faced and which required his masterly tech-
nical achievements do not arise. To logicians this will seem like
discussing Hamlet without the Prince, but it is the other
characters who claim philosophical princedom.
Imagine, then, an abstract language like that of 1.3 which
has a few names and a few predicates, enabling it to make
altogether a small number of sentences. The syntax oflanguage
tells us that if 'a' is one of the names, and 'P' is one of the
predicates, the result of putting a next to P (Pa) is a well-formed
sentence; nothing else is a sentence. How are such sentences to
be understood? What is needed is a way of saying what the
I
Page references for the first two are to the volume Logic, Semantics, Metamathematics;
to the third, to Readings in Philosophical Ana/ysis, eds. Feig! and Sellars. CTFL was first
made public in Polish in I 931.
 TRUTH AND SEMA;\;TICS 263

different categories of component do - in this case, what names

do and what predicates do. This semantic role or semantic
value is that they contribute to the overall meaning. The obvi-
ous choice is that names refer, and predicates introduce proper-
ties, or rules or principles of grouping things. Any sentence is
interpreted as saying, of the thing named, that it satisfies the
property, or satisfies the rule or principle of grouping intro-
duced by the predicate. This is what has to obtain for the
sentence to be true. It can be put as a "compositional axiom"
saying what makes the sentence true in terms of the semantic
roles of its constituents:

(D 1 ) Any sentence Pa in any language of this sort will be true

if, and only if, the predicate applies to or is satisfied by
whatever it is that the name refers to.

What we have here is a framework with which to describe any

of.a whole family of abstract languages - those containing only
names, which refer, and predicates, which we can say apply, to
things or are satisfied by things. D 1 reveals the issue of whether
any such sentence is true as a twofold issue-one of the reference
of the name and the other of the satisfaction of a predicate. This
reassuring thought does not touch, of course, any of the
philosophical issues we have been involved in. By itself it does
not tell us when a sign in a community is being used as a name;
what determines the particular thing it does name ifit is; when a
sign is being used as a predicate; what determines the 'applica-
tion' of the predicate (remember the difficulty with rules) ifit is.
Thus when in chapter 6 we struggled with truth of evaluations,
it would hardly help to be told D 1 : puzzling whether evaluations
are true can be thought of as puzzling whether evaluative
predicates apply to their subjects (or do so in the same way as
other predicates do), but it doesn't advance the philosophy to
put it that way.
Because D 1 yields only a framework of description, to
describe semantically any particular abstract language within
the framework, we must specify what the names refer to and
what the predicates apply to. To illustrate thfa stage I shall run
two abstract languages in parallel. Each has two names (a, b)
and two predicates (P, P').
264 LANGUAGE AND THE WORLD

a in L 1 refers to Lenin a in L 2 refers to Paris

bin L 1 refers to Marx b in L 2 refers to Rome
Pin L 1 applies to bald things Pin L 2 applies to French things
P' in L 1 applies to pink things P' in L 2 applies to warm things.
Each language can express a modest four sentences. Let us call
the framework (D 1 ) and these interpretations, the applied
framework. Then using the applied framework we can work out,
for instance that the sentence 'Pa' is true in L 1 if and only if
(hereafter 'iff') Lenin is bald, and in L 2 iff Paris is French.
It is now important to consider the status of the clauses.
What is meant by saying that 'a' in L 1 refers to Lenin? We must
distinguish between taking such a proposition as (to use
Carnap's terms) a piece of descriptive semantics and a piece of
pure or abstract semantics. As a piece of descriptive semantics
we imagine a term 'a' used in the mouths of some people, whose
language we call L 1; it is an empirical fact, established by
observation or experiment, due to their behaviour, thoughts, or
conventions, that this sign refers to Lenin. As a piece of pure
semantics we are not interested in any actual people. We are
merely specifying what I called an abstract language - a pos-
sible, but not necessarily spoken, language. Since that is all that
we are doing, nothing empirical or contingent enters into it.
That will all arise when we fit the abstract language we have
specified to any particular people, for it is then that the question
arises whether they actually use any signs in the ways
stipulated. Pure semantics, in this sense, is entirely stipulative-
merely a matter of armchair invention. So the clauses saying
what the terms of L 1 and L 2 do are definitional - part of the
stipulation of which abstract language, .or semantical system,
we are considering. Since such a system is created by fiat
nothing yet touches the question of what would make it true
that a sign functions in any of these various ways, or of how we
could discover it.

2. Truth-in-Land Convention T
We can think of the applied framework as a manual. Using it,
and supplied with a sentence of the object language (L 1 or Li)
we can compute what has to obtain for the sentence to be true.
'Pa' is true in L 1 iffLenin is bald, and so on for each of the four
 TRUTH AND SEMANTICS 265

sentences. The manual is complete, for no sentence that can be

formed in either of these languages escapes it. It would be
natural to say that such a manual characterizes the abstract
languages. However Tarski favoured the slightly surprising
description that it characterizes truth, or reference, or satisfac-
tion for those languages. What lies behind this shift of
emphasis? At first sight, all that the manual tells us about truth
for these languages is what was contained in DP that it arises
out of the reference of names and the application of predicates.
But Tarski describes himself as having constructed, for the
language he treats, definitions of the semantic concepts, which
include reference, and the satisfaction of predicates by things,
as well as truth (e.g. CTFL, p. 252; ESS, p. 406; SCT, p. 63).
He thought of the definitions as establishing semantics on a
scientific footing, by showing how to replace semantic terms
altogether, for suitably simple languages. What is going on?
The essential change in perspective has been well described
in a classic paper by Hartry Field. 2 The key is that Tarski means
what he says when he describes the clauses in the applied
framework- the clauses stating that 'a' in L 1 refers to Lenin,
and so on, as partial definitions of the semantic concepts, whose
full definitions are given by the total set of these partial defini-
tions, and their consequences, (e.g. CTFL, p. 192; ESS, p. 404;
SCT, p. 55). How can such things as the clauses for L 1 and L 2
be regarded as "partial definitions" of any semantic concepts?
Initially they seemed best regarded as partial definitions of
the abstract languages L 1 and Li- But suppose we decided to
view them as partial definitions of six hyphenated concepts:
reference-in-L 1 ; satisfaction-in-L 1 ; truth-in-L 1; and similarly
for L 2 • What are these hyphenated concepts? Here is the
answer:
X refers-in-L 1 to Yiff: Xis a and Yis Lenin, or Xis band Yis
Marx
Y satisfies-in-L 1 X iff: Xis P and Y is bald, or Xis P' and Y is
pink
Sis true-in-L 1 iff: Sis Pa, and Lenin is bald; Sis P' a, and
Lenin is pink; Sis Pb and Marx is bald; S
is P' b and Marx is pink.
2
H. Field, 'Tarski's Theory o[Truth',joumal ef Philosophy (1972).
266 LA'.\'GL'AGE AND THE WORLD

The reader can construct the corresponding clauses for L 2 • It is

important now to try to think of these lists as full and final
accounts, definitions, of the hyphenated terms. They are not mere
lists; they are definitions, and each item on the list is part of the
definition. So we have been given a definition of truth-in-L 1 ;
furthermore, a definition which makes no mention of any semantic
terms. Using these definitions you can tell whether any sentence
in true-in-L 1 without making any semantic judgements at all.
You only have to be able to tell which sentence you are faced
with, and such things as whether Lenin is pink. 3 I shall call
these definitions list-accounts of the hyphenated concepts.
The little languages L 1 and L 2 are unusual in that a list-
account of true-in- ... can be provided. This can only be done
because there is a finite number of sentences (CTFL, p. 188). If
the devices of the language enabled us to construct sentences ad
lib, we would need a recursive clause, that is, one which shows
how to add to the true-in-L list an arbitrarily complicated
sentence of L, in terms of the semantic properties of its parts.
This would be like the complex rule governing ANN. 4 It is easy
to see how such clauses go. For instance, supposing we enriched
L 1 to L + by adding a connective ' & ', meaning conjunction. At a
stroke we can create an infinite number of sentences, for any
pair of sentences can be conjoined, the result conjoined with
any other, or with itself again, and so on. But we can also see
how the truth-in-L + of any such sentences is constructed: 'P &
Q' will be true-in-L + if and only if P is true-in-L + and Q is
true-in-L +. By using this rule on any conjunction you will
eventually reduce the question of its truth-in-L + down to a
sequence of questions of the truth-in-L + of some of the four
original sentences. Tarski's great technical achievement was to
see how to give a recursive clause for truth-in-L for languages
containing quantifiers: I describe this in the notes to 9.1. (List-
accounts with rules like this added are called 'recursive
accounts' of the hyphenated concepts.) But the philosophical
question is most easily appreciated by returning to L 1 and L 2 •
Can the lists really be taken as definitions ofanything? Notice
3
"To tell which sentence you are faced with": if the semantics is fully formalized, a
sentence is identified by a structural description, representing it as a sequence ofletters.
This does not affect the philosophy.
4
l .3 above.
 TRUTH AND SEMANTICS 267

that if they are, then 'refers-in-Li' and 'refers-in-Lz' are entire!:J

distinct relations. In fact, the appearance of a common relation
(reference) is misleading. For Xrefers-in-L 1 to YiffXis 'a' and Y
is Lenin, or Xis 'b' and Yis Marx. ButXrefers-in-L 2 to YiffXis
'a' and Y is Paris, or Xis 'b' and Y is Rome. These are utterly
different notions, for they are defined by totally distinct lists.
Consider some analogies. Suppose I am again troubled by
jurisprudence, and wonder what it is for a legal verdict to be
properly grounded at some time. Someone advances a proposal.
Seeing which verdicts are properly grounded on Wednesday
(say, a, b, and c), and which verdicts are so on Thursday (x,y,
and z), he urges the following definition:
A verdict is properly-grounded-on-Wednesday
iff it is one of a, b, or c;
A verdict is properly-grounded-on-Thursday
iff it is one of x,y, or z.
What is the point of this manoeuvre? Its most immediate point
is to stop the hyphenated concepts from having anything to do
with one another. In fact, we would do better to drop the
hyphens, and regard the lists as defining two new terms -
properlygroundedonWednesday, and properlygroundedon-
Thursday. But these new terms no more contain the concept of
proper legal grounding than the word 'concatenation' contains
the notion of a cat. That is, in each case the term can be used in
entire ignorance of any meaning attaching, in other contexts, to
the embedded strings of letters. 'Cat' is not a semantic con-
tributor to 'concatenation', and 'properly grounded' is not a
contributor to 'properlygroundedonWednesday'; the proof in
each case is that to tell if the term applies nobody need know
anything connected with the (apparently) embedded terms. To
tell if a word is a concatenation ofletters (how it is spelled) you
need know nothing about cats; the ultimate proof whether a
verdict is properlygroundedonWednesday is whether it is one of
a, b, or c. This has nothing to do with jurisprudence (jurispru-
dential judgements might have gone into making up the list).
Or take this parallel. I am interested in reference to numbers.
I define my systems ANN and ANN* (1.3), and give the base
clauses and the recursive rule determining the reference or
arbitrarily long numerals. I have done something, certainly,
268 L.\:\GUAGE AND THE vVORLD

but what I have done and left undone needs careful identifica-
tion. Suppose I describe myself as having defined numerical
reference for those languages. That sounds good: what is the
definition? Well, a numeral refers-in-ANN-to-n iff the numeral
is 1 and the number is l, or the numeral is 2 and the number is
2 ... Because of the recursive rule I can know whether any
numeral/number pairing is on this list. On the other hand, a
numeral refers-in-ANN*-to-n iff-and then follows the different
list; so on for any system I care to invent. To know whether a
numeral 'n' refers-in-ANN-to-n you need know nothing ofrefer-
ence to numbers - you just look at the list. Does this put the
semantic relations between numerals and numbers onto a
sound scientific footing? Is it a definition of reference to
number, or a theory of it?
Suppose firstly that we consider it as a piece of pure seman-
tics. Then I am just inventing systems. I am not concerned with
the question of what it would be for a population to actually use
one of these systems; or what kind of truth it would be, and how
it would be known, that a population uses 7 to refer to 7. Since I
offer nothing to that question my account does not put reference
to numbers on a scientific footing, or indeed on any footing at
all. For instance as far as it tells us, numbers may be objects in
the mind of God and reference to them a kind of telepathy. The
issue is simply not addressed.
Suppose on the other hand that we regard one of the systems
as a piece of descriptive semantics. Then I pass to you a piece of
paper saying that in the system, ANN of some population, 1
refers-in-ANN to I ... ; since I give you the rule, you have a
complete manual for reference-in-ANN. You ask what use that
is; I tell you that it enables you to determine reference-in-ANN.
You ask me what that is, and I repeat that reference-in-ANN is
defined by the list. This gets us nowhere. To connect with
something you might want to know - such as which number
someone in the population was referring to by some sign - I
have to tell you that the reference-in-ANN of a sign is also the
number they use it to refer to. That is fine for you: you can now
interpret them. But it's only fine for you because I myself could
detect which numbers they referred to by which signs! And once
again we are given no further understanding of what it was that
I discovered; how I could discover it, or what makes it true. I
 TRUTH AND SEMANTICS 269

had to apply the contentious notion in drawing up the list. Compare

my telling you that a, b, care properlygroundedonWednesday;
you say 'What is that?'; ifl say it is defined as applying to one of
a, b, or c I have got nowhere; ifl tell you that when something is
properlygroundedon Wednesday it is also a properly grounded
legal verdict in the system of some population I do tell you
something, but I don't advance any understanding of what
would make that true, or of what I did to discover it, or how you
might detect proper legal grounding in any new system.
The obvious moral is that the idea that the lists (or lists plus
recursive rules) defi,ne any semantic property is completely mis-
guided. How was this disguised? Two large fig-leaves hide the
enormity. The first is that a common proposal for ensuring that
we know what we are talking about suggests that the word
'semantic' is reserved for the notions used in the construction of
abstract semantic systems, and the title 'pragmatics' is used for
any science or art of interpreting a population as using any one
such system. As a terminological proposal this would not mat-
ter, except that it suggests that you can, for instance, tell what
people refer to without raising any semantic problems unsettled
by list-accounts. It sounds an advance to say that I have a
complete account of numerical reference, because I have a
recursive definition of reference-in-ANN, so the only problem
left with reference to numbers is one of pragmatics, of fitting an
abstract language to a population. But since the pragmatic
problem is just that of telling whether a group uses 'n' to refer to
k, there is a genuinely semantic concept left to use, and whose
problems are still untouched.
The second fig-leaf is the ambiguity of such phrases as
"relativizing truth to a language". It is common to find Tarski's
definitions defended on the ground that this is all that he is
doing, and a good thing it is too, since the truth of a sentence is
indeed relative to the language in which it is used. This is
because the same sentence may be used in different languages
to say different things. Tarski himself stresses this point (e.g.
SCT, p. 53) and Quine uses it to introduce the hyphenated
concepts, in the course of approving ofTarski's method. 5 But it
is quite insufficient to motivate list-accounts. Of course a horse
5
W. V. Quine, 'Notes on the Theory of Reference', in Fr~m a Logical Point of View,
p. 134.
270 L\:\"GUAGE AND THE WORLD

may win one race and not another, or a verdict be well-

grounded in one system oflaw and not another; '2' may refer to
2 in ANN and 5 in ANN*. But it does not follow that there is
nothing in common to winning different races, or being well-
grounded in French law and well-grounded in English, or
referring to a number in one numerical system and in another,
or for that matter to truth as expressed in English sentences,
and as expressed in those of any other language whatsoever.
Reflection upon the application of an abstract semantic system
to any actual population shows that there must be.
The lists may not themselves serve as definitions of semantic
concepts, but it might still be suggested that they have a
philosophical role. The line would be to turn Tarski to the
service of a redundancy theory of truth. This was the approach
considered above (7.3). This points out the fact that application
of the predicate to a sentence is determined, for each sentence,
by a different matter (which the truth-theory identifies for us).
This is then used to block attempts to find a unified concept or
property of either sentences or judgements, which the predicate
expresses. I rejected that line on the grounds that it fails to put
out of court the search for a (correspondence, coherence, etc.)
conception of success in judgement. Tarski's way of connecting
truth, reference, and predication in the semantic description of
complex languages does not actually affect this. It leaves quite
open the dazzling but elusive prospect of a substantive theory of
truth.
Let us call the connection of truth, reference, and satisfaction
the 'neutral core' ofTarski's work. For the simple languages we
are considering, and for the more complex ones which Tarski
succeeded in describing semantically, the neutral core connects
together truth, reference, and satisfaction. But it gives us no
theory of how to break into this circle; that is, of how to describe
what it is about a population which makes it true that any of
their words or sentences deserve such semantic descriptions. As
far as the neutral core goes, we might prefer a "bottom-up"
approach, thinking of the semantic properties of words first,
and believing that sentences are secondary to them. This would
run contrary to the ideas of chapter 4; influenced by them we
might prefer the other "top-down" direction, in which the
semantic properties of names and predicates are thought of as
 TRUTH AND SEMANTICS 271

abstracted out from the features of whole sentences in which

they occur, rather as the fact that a few dots in a newspaper
photograph represent Sophia Loren's ear is secondary to the
fact that the whole photograph represents her face. But before
considering the ideas philosophers have added to this neutral
core, I want to introduce the famous 'Conventiol) T' and the
semantic conception of truth which Tarski endorsed.
A semantic account of ANN is complete if it enables us to
determine the reference of any numeral of the system. If we can
do this, we can write something of the form 'n in ANN refers to k'
for any numeral n. Tarski laid down a similar desideratum for
languages. Corresponding to the sentences we want for ANN he
took the form: 's' is true in L if.!p. He called these, equivalences of
the form T, and they are referred to as T sentences. (SCT, p. 55.
Tarski does not explicitly put in the reference to L, but it is clear
that this is simply for ease of exposition.) A semantic descrip-
tion of a language might be thought of as incomplete ifit cannot
deliver an equivalence of the form T for each sentence of the
language. Putting the matter in the way criticized, a definition
of truth-in-Lis incomplete unless it enables us to do this. That is
apparent, given that each particular T-sentence is a partial
definition of the notion, and their totality is the whole defini-
tion. Naturally,just as 'n' in ANN refers to k will only be true, for a
given fixed notation ANN, if the sign 'k' refers to the same
number as the numeral 'n' does, so 's' is true in L if!Jlwill only be
true, for a fixed language L, if 'p' expresses something bearing
the right relation to whatever 's' means; Tarski requires that it is
the same sentence as 's' if the metalanguage includes the object
language, or is a translation of it into the metalanguage if it is
not. 6 (Don't be surprised that it can be the very same sentence.
Remember that the merit of the rule for ANN is that it enables
you to see why '219' refers to 219.) Therefore, a definition of
truth-in-L requires a complete list, or way of showing how to
calculate a T-sentence for each sentence of the language con-
sidered. Such a definition meets Tarski's material adequacy condi-
tion for a definition of truth-in-L; the notorious convention T.
For the reasons brought forward in this section we ought to
be reluctant to see a definition of any semantic concept at all in
• If we are just stipulating then we make 'P' the interpretation of 's' in the
metalanguage, and it is wrong to describe this as a requirement.
272 L\'.\IGUAGE AND THE WORLD

any such list or rule for producing one for a language. If I

produce Convention N, requiring a semantics for a numeral
system to enable a referent to be determined for any numeral
the system can form, I could describe it as a 'material adequacy
condition' on a definition ofreference to a number, or (better) of
reference-in-Stoa number, where Sis an arbitrary system. But
that suggests, misleadingly, that when we have a list, or list and
recursive rules, meeting the material adequacy condition, we
have thereby achieved an advance in understanding reference
to numbers, and we have seen that this is not true.
Tarski did not see it like this. He is insistent that there is a
distinct and genuine conception of truth somehow implicit in
his work-a semantic conception of truth. He regards it as just a
proper, modern, way of "grasping the intentions which are
contained in the so-called classical conception of truth ( truth -
corresponding with reality)" (CTFL, p. 153; the same claim is
made in SCT, pp. 54-6). He himself cheerfully admitted that
his conception of truth might be merely one amongst others,
"including pragmatic conceptions, coherence theory, etc." But
he goes on to doubt whether these have been put in an intel-
ligible and unequivocal form. And in true positivist spirit he
allows himself considerable scorn for "the philosophical prob-
lem of truth";
I have heard it remarked that the formal definition of truth has
nothing to do with 'the philosophical problem of truth'. However
nobody has ever pointed out to me in an intelligible way what this
problem is. (SCT, p. 66.)
But as we have seen, list-accounts, or recursive definitions, of
truth-in-L or reference-in-L or satisfaction-in-L, involve no
particular conceptions of semantic relations. To repeat, this is
because they address no question of what it would be for a sign
to be used to refer, or for a sentence to be used to say something
true; they address no question of what it would be for a popula-
tion to use a language L for which the adequate truth-
definitions afe given - in other words to refer, apply predicates,
or assert the things paired with their signs in the clauses govern-
ing L. A 'recursive definition of reference-in-ANN' is simply
silent over what numbers are and what it is to refer to one; it can
be accepted by people who believe that they are sets, who
 TRUTH AND SEMANTICS 273

believe that they are not, who think that they are mental, who
think they are ideas in God's mind, even who think that they are
nothing. And silence is not the same as introducing a distinct
conception of reference. The same is true of truth. To get rid of
any temptation to think otherwise, ask which part of a semantic
description, such as that given of LP or that governing ANN, is
unacceptable to people with distinct and differing conceptions of
truth - non-classical ones, if Tarski is elaborating a distinct
classical viewpoint. Take pragmatists, coherence theorists, or
correspondence theorists (Tarski is sometimes taken to have
superseded a correspondence theory, rather than as he himself
thought, merely laid it out properly). Do they deny that a
subject-predicate sentence is true as the framework principle
D 1 says? Do they deny any of the clauses in the lists and
recursions - for instance, that 'London' refers in English to
London, that 'red' applies to things that are red, and 'good' to
things that are good? Of course not. Their contributions and
differences relate to what it is for these things to be so, not to
whether they are so.

3. Top-Down and Bottom-Up

It is not now very controversial that the neutral core ofTarski's
work leaves us with reference, satisfaction, and truth in a tight
little circle, nor that it would be acceptable to philosophers of
any bent: realist, anti-realist, correspondence, coherence, dis-
missive neutralist, or whatever. The question is how to add to
that core to generate philosophical body. One possibility is that
more is added when we come to more complex languages.
Tarski's technical achievement lay in describing the structure
of languages with expressions of generality. To do this he
regarded sentences as having a relationship to things analogous
to that which a predicate bears to the things which satisfy it.
Truth of a sentence becomes the "satisfaction" of a sentence by
everything. It is possible to suggest that it is just here that the
work gains philosophical importance. But the suggestion is not
attractive, for two reasons. The first is that when we understand
why Tarski was led to this move, we can see that it should not be
a move which offends any kind of anti-realist, coherence
theorist etc. The second is that the move in fact represents a
274 LA'.'l;GUAGE AND THE WORLD

kind of trick. It is made in answer to a technical problem, and

the technical problem can be solved in other ways. In those
other ways, sentences are not regarded as kinds of predicate,
but rather predicates are thought of as in effect kinds of
sentence. But the compositional insights remain the same. The
reader can follow up these allusive comments in the notes to 9.1.
For present purposes I am going to put this way of adding to the
neutral core aside. The other way would be to find a separate
range of ideas which can be added to it.
I described a "top-down" direction of explanation as one
which takes the functioning of sentences as primary, and
believes it can abstract out reference and satisfaction. A
"bottom-up" direction works the other way round. Because in a
truth-theory the thing we end up saying about a sentence (that
it is Tiff ... ) is deduced from what we say about its components,
it is natural at first blush to associate truth-theories with the
bottom-up direction.
Thus in 1969 Donald Davidson wrote:
Statements are true or false because of the words used in making
them, and it is words that have interesting, detailed, conventional
connections with the world. Any serious theory of truth must there-
fore deal with these connections, and it is here if anywhere that the
notion of correspondence can find some purchase. 7

The idea appears to be that the truth of a sentence is explained

in the bottom-up direction, from the self-standing, "detailed,
conventional" semantic properties of its constituents, notably
the satisfaction conditions of predicates. But do predicates have
these self-standing, detailed or conventional connections with
the world? Is it not from understanding what is judged, when it
is judged that a thing is yellow, that we come to understand the
idea that the predicate stands for a certain property, or is
satisfied by various things (yellow things)? In chapter 4 we
learned to think of a word as a recurring feature of sentences
which share some similarity of content. One persuasive reason
for that direction was the way in which radical interpretation
seems to be based upon attributing a content firstly to whole
messages, and only derivatively to features of them. Another

7
'True to the Facts',Joumal qf Philosophy (1969), p. 754.
 TRUTH AND SEMANTICS 275

reason would be the impossibility of imagining a society which

has succeeded in giving detailed conventional semantic proper-
ties to words, but without ever putting them together in
sentences. The incoherence of this is rapidly apparent. Even
if they say 'rabbit' just when there are rabbits about, or
'Napoleon' just when they have a mental image ofa little man in
a cocked hat, there can be no reason to allow the words to refer
(to the rabbit? to the image?) unless something is then done with
it. Something might indeed be done: for instance, the group
might take someone uttering the word as a signal of the
presence of rabbits or whatever. But then it is treated in effect as
a sentence meaning that there is a rabbit or are rabbits present.
There is simply no way in which a word can be given a detailed
conventional role as referring or applying to things, in the
absence of a habit of conveying information by using it.
In the paper cited Davidson was not only defending a
bottom-up direction, but also claiming that the particular thing
Tarski would have" us say about predicates - that they are
satisfied by things - gives us a legitimate sense in which truth
is explained as correspondence: "the notion of truth can be
explained by appeal to a relation between language and the
world." But even if the notion of truth is regarded as explained
via a prior understanding of what it is for a predicate to be
satisfied, it is not at all clear that Tarski weds us to a cor-
respondence theory of this. The clause for a predicate can take a
variety of forms, compatibly with working in a smooth recursive
theory of a language: 'banana' in English is satisfied by a thing
iff: it is a banana/is a member of the set of bananas/has the
feature of being a banana/is one ofx,y, Z, ... where these are all
the bananas that there are/is a yellow curved fruit of the banana
tree ... Some of these appear more concerned with a relation
between the predicate and the world than others - for instance,
the ones which mention things x,y, z, or the ones which mention
features and sets. But once again it is difficult to believe that this
appearance is reliable. Nobody is going to deny that the predi-
cate will apply to a thing x iff xis a banana; people may dispute
over the kind of thing this judgement is, and the way coherence
and pragmatic virtues form it.
Davidson took a very different view in later work. The whole
idea that the semantic properties of sentences depend upon
276 L\:\'GUAGE AND THE \\'ORLD

interior relations such as reference of words and application of

predicates is swept aside:
Having explained directly the semantic features of proper names and
simple predicates, we could go on to explain the reference of complex
singular terms and complex predicates, we could characterize
satisfaction (as a derivative concept), and finally truth. This picture
of how to do semantics is (aside from details) an old one and a natural
one. It is often called the building-block theory. 'It has often been
tried. And it is hopeless. 8
The switch he advocates is to a top-down theory, but it remains
to be seen how such a theory looks.
Top-down theories seem unnatural because of two things.
Firstly the semantic description of a language will have a
bottom-up appearance: the properties of sentences are deduced
from the properties of their parts. Secondly we have a feeling
that this is psychologically right: it is because I know which
words come in a sentence, and what they do, that I can under-
stand it. But we can avoid worries on the first score by distin-
guishing the abstract description of the language from the
question of what makes it true that a population is speaking that
described language. The description may move from small
words to large sentences. But it may be in their use of the large
sentences that we get the basic fact that makes it true that the
group is using the language. Davidson makes the same distinc-
tion in terms of explanation (of the semantic properties of bigger
units of speech) within the theory and explanation ofthe theory-
showing why it is not a mere abstraction but connected with
actual human populations, their ends and interests. 9 It is when
we interpret a population, by seeing how they use their
language to express their beliefs, needs, and wishes, that the
primacy of the sentence is apparent. The second point, t.llat
psychologically there is something to the bottom-up direction,
need not be denied. An explanation of how I understand a new
sentence as I do ought indeed to cite the familiar words in their
familiar roles. But this does not at all work against the idea that
those roles are essentially sentential, so that the word has no
detailed or conventional function of its own, isolated indepen-
8
'Reality without Refer.ence', in Platts (1980), pp. 134-5.
9
Ibid., p. 137.
 TRUTH AND SEMANTICS 277

dently of what it does in whole sentences. Consider that a

chess-player can appreciate a new position in a game because
he knows the power of the pieces; but the powers of pieces in
chess are entirely given by how they affect the game.

4. Radical Interpretation Again

So far I have talked airily about abstracting the semantic
properties of words from those of whole sentences. But what
kind of detail should we add to this idea? This can be focused by
considering again the familiar figure of the radical interpreter. I
want to contrast two ways in which he can be added to the
picture: a bleak way and a homely way; each label may cover a
spectrum of real procedures, but it is the families which are
interesting.
A homely radical interpreter starts with hypotheses about
what members of the group are likely to be interested in, saying,
and communicating. His initial hypotheses are therefore highly
specific: he wonders whether by S said on such-and-such an
occasion they mean that it is raining; whether by S* they mean
that it has not been a good year for water-melons, and so on.
He guesses at the determinate intentions and meanings lying
behind utterances. What controls him in these guesses? Mainly
a desire to see the natives as intelligible - as having certain
intelligible aims, and performing actions, such as saying things,
as sensible means of furthering those ends. Another desire will
be to see their language as systematic: for reasons mentioned at
the end of chapter 4 this is not too far distant from seeing them
as having a language at all. In any case, it will certainly be
necessary if their language is elastic enough to allow them to
comprehend new sentences. The desire to see the natives as
intelligible may include quite specific constraints on how to
interpret them - for instance, if they all appear to have their
attention focused on one place or thing, whose changes are
exciting them, not to regard them as talking about a different
thing which they cannot see. I suppose that most of us, faced
with the radical interpreter's predicament, would be homely.
A bleak radical interpreter, by contrast, dislikes this
armchair psychologizing. He likes his evidence cut and dried,
which in this case might mean a bare description of what was
278 L-\'.\IGUAGE AND THE WORLD

said, and of what purely natural, non-p~chological features of the

native's situation surrounded the saying ofit. His evidence is to
be presented in non-psychological terms, and of course, in
non-semantic terms, for it is evidence which he is going to use to
construct a scientific semantics of the native language. What
controls the way he does this? Mainly a principle of charity, or
desire to see as many of the native utterances as true as he
possibly can (by contrast, I will call the principle of intelligibil-
ity used by the homely interpreter a principle of humanity). The
bleak interpreter is to be regarded as embodying the scientific
or physicalistic truth about the natives. The homely interpreter
enters straight away into their psychology.
The interesting thing is that the position efthe bleak interpreter is
hopeless. There are a variety of results which show this. For
workers in formal semantics have long known that sentences
can be "interpreted" in very different ways. That is, a given
result at the macro-level (what to say about the whole sentence)
may be compatible with very different properties of words at the
micro-level (different interpretations of individual words). The
grandfather of these results is the Lowenheim-Skolem theorem.
They show that any theory of a standard form, and considered
purely as a set of affirmed sentences, if it is interpretable as
speaking truly about some set of things, is equally interpretable
as speaking truly about any other set of things of equal or
greater number. In other words, if your relationship with me is
confined to knowing which sentences I hold true, then as far as
that goes I could be talking about cows or cabbages or numbers
or sets or stars. You could, in principle, provide an interpreta-
tion cif my words which has me saying just as many true things
about any of these domains. This should not be surprising. It is
as though you have a list of sentences I hold to be true, and
several lists, each of many propositions, but about entirely
different things. You then play the game of mapping my sen-
tences onto theses first from one list and then from another. It is
no wonder that different solutions are possible. Now of course
we only get this wild indeterminacy by putting the macro-
knowledge so sparsely. (By contrast, the homely interpreter will
control his interpretations by the principles of common-sense
psychology. It will not be sensible, for instance, to interpret
football fans evidently intent upon a game as talking about
 TRUTH AND SEMANTICS 279

transcendental set theory.) The sparseness here is taking as

data only the list of sentences to which assent is given. Can the
bleak interpreter do more?
He can look at any natural (physical, or causal, for example)
relations speakers have with things. For instance, if the football
game seems to be causing the comments, he may take that as a
reason for supposing the comments to refer to the game. But we
have a multiplicity of causal relationships to things. When I
refer to Napoleon, I may be prompted to do so by a text
containing words; the words were written via a printing system,
via an author himself relying on other authors, and so on: But it
is Napoleon I am referring to, not any of the intermediate
elements of the causal chain. Two questions arise. First of all, is
there any prospect of a purely natural description of just the
relations to things I have when I refer to them? Secondly, even if
there is, what is it that makes just that causal or natural relation-
ship the relationship of reference? (A similar question to that
asking what makes just one particular similarity ranking across
possible worlds the right one for counterfactuals - 6.5).
If we think in terms of the homely radical interpreter, we are
asking how we determine the facts about meaning, reference,
and satisfaction conditions of predicates. Ifwe think in terms of
the bleak radical interpreter, we are asking, as David Lewis
puts it, how the facts determine the facts - how semantic
interpretations sit on top of physical facts about things. And the
trouble seems to be that there is no unique way in which the
physical facts seem to determine the semantic ones. Two
supernatural beings, each acquainted with all the physical facts
. about us and our world, each aware therefore of all our causal
relationships with things and their features, might yet choose
different ways of co-ordinating these facts with semantic
descriptions, and end up regarding us as saying different things,
referring to different things, and selecting different features of
them for comment. At ileast, :so many 1philosophers have
argued.
If we accept this line of thought, then it can suggest a number
of different conclusions. The most radical, but also the most
partisan, would be to confine truth to physical truth (the per-
spective from which meaning becomes impossible, again).
Then there is no literal truth in interpretations of people, saying
280 LA::".GUAGE AND THE WORLD

what they said and what they referred to. Reports saying these
things are not, as Quine put it "limning the true and ultimate
structure of reality", but are indulging a second-rate "essen-
tially dramatic idiom"; useful in the market-place, but no good
in the study. This is partisan because of the high-handed con-
finement of truth to physical truth: anyone who has
sympathized with the general difficulties we have had with
truth, and in particular anyone who has seen the resources of
other anti-realist attempts to earn the notion, will hardly be
confident that physical descriptions are the only kinds of saying
that merit truth. And apart from general considerations, there
is the point often made against Quine that the parallel move can
deprive even physical descriptions of truth. Two omniscient
beings each acquainted, for instance, with the whole story of all
the experiences of sentient beings in our world, might construct
very different physical theories on that basis.
The state of play, then, is that the bleak radical interpreter
will not arrive at a determinate reference or satisfaction condi-
tion for the names and predicates in sentences, nor at a determi-
nate interpretation of overall sentences. The homely radical
interpreter, by exercising a principle of humanity and entering
into the likely needs, desires, thoughts, and beliefs of his sub-
jects, may well do so. Is it right to see his method as exclusively
top-down? The principle of humanity, enjoining him to make
them reasonable and intelligible, certainly suggests that he first
consider the likely intentions behind their whole utterances,
and perform the abstraction of subsentential elements on that.
But it may enjoin other things as well. In particular it may offer
general constraints at a subsentential level. Two principles of
common-sense interpretation would plausibly be: don't inter-
pret people as referring to a thing if there is no means by which
they could have been brought to be aware of it; don't see them
as ascribing a feature to a thing if the feature is undetectable by
their sensory modalities (don't interpret blind men as talking
about shades of colour). And, as the football match example
shows, there will be a natural inclination to see reference as
often determined by attention: the first thing that may come to
mind on hearing utterances made in the crowd is that the likely
reference is the match, and only secondarily do we think of what
might be being said about it.
 TRUTH AND SEMANTICS 281

Let us call the preferred method of interpretation of the

homely radi!=al interpreter, W* h, and that of the bleak one, W* b·
Then, provisionally, it would seem that neither justifies an
exclusively top-down direction. W* b indeed excludes it, in that
what it allows us to say about sentences is quite insufficient to
enable anyone to abstract out determinate semantics, either for
subsentential elements, or for whole sentences. By contrast W* h
may enable us to do this, but also may incorporate direct
constraints on what we are allowed to say about reference and
predication.
It would be nice if the indeterminacy which affiicts W* b did
not exist. For then we would have a proper sense of how the
physical universe makes room for meaning - how the referen-
tial, intentional powers of our minds and words arise in the
order of nature. But if we accept the indeterminacy, then even if
we do not allow the high-handed confinement of truth to physi-
cal truth, still unease remains. Are we to conclude, with the
philosopher Brentano, that the semantic properties of our
words and sentences refute physicalism - that they prove the
existence of a realm of fact which is quite additional to and
separate from the physical universe? Or are we to conclude,
with Quine, that there are no such determinate semantic facts
to warrant such a conclusion? But after all, I know what I mean
when I say that rabbits are good to eat - I know what I refer
to and what I say. Indeterminacy may affiict the bleak
physicalistic outsider looking at me, but to me and to my
fellow-speakers there is no shadow of indeterminacy to be seen.
Again the brute problem of incorporating meaning into a
physicalist's universe lies heavily across the landscape.

5. Truth and Extensionality

The observant reader will have noticed an asymmetry between
the semantic description of ANN, and the truth-theoretical way
of characterizing the sentences of a language. In the case of
ANN we want to know what complex numerals refer to, and
this is what we are told. In the case of language, we want to
know what complex sentences mean, but this is not what we are
told. We are not told that sentence S means that ... , but only
that sentence Sis Tiff ... And this is a very different thing. For
282 L\>iGUAGE AND THE WORLD

as logicians use it, 'iff' is an extremely weak relation. Any

sentence of the form 'p iff q' holds provided p and q have the
same truth-value - i.e. provided both are true or both are false.
If you take any two true propositions,p 1 and qi' you are allowed
p1 iff qi' and similarly if both are false. The moon is a star iff
penguins fly; rivers flow downhill iff money circulates. Similarly
rivers flow downhill is true-in-English iff money circulates. But
this does not tell us what it means. Why then put it at the centre
of an attempt to construct a semantics? Why not obey the
natural course, and go directly for an explanation of how it is
thats in L means that p? On the formal approach, this would be
a formal proof of sin L meaning that p.
It is important to understand the size of the gap. 's' is true if!p
is as weak as this: 'you would be as right or wrong if you said
that 's' is true, as if you said that p.' This allows for any true
sentence to be paired with 's', if's' is true, and any false one to be
paired with it, if it is false. The subject-matter can be different,
the structure, the concepts, anything. You would be as right to
say that men have landed on the moon is true-in-English, as if you
said that bees make honey, or as if you said that penguins
waddle. Some authors misleadingly suggest that the use of the
T-form is justified because it incorporates the insight that giv-
ing the meaning of a sentence is doing something called 'giving
its truth-condition' . 10 This is not so: there is no legitimate sense
of 'truth-condition' in which the sentence men have landed on the
moon has the truth-condition that penguins waddle. If there
were, then all true sentences would have the same truth-
condition, and all false ones too.
The advance due to Davidson, which hopes to extract more
from the T sentences, comes as follows. Imagine an abstract
language, English* characterized by a truth theory. Suppose a
T-sentence for English* yields a clause stating that 'penguins
waddle' is true in English* iffmen have landed on the moon. As
far as that goes English* is like English. But now ask: How does
it get this clause? There are two cases. It might treat the
sentence as unstructured - i.e. just give a one-off axiom coupl-
ing this interpretation to it. But then there is good evidence that
10
Mark Platts, Ways ef Meaning, ch. I, cites many writers who seemed to think this.
Platts himself approves of the idea partly because he supposes that the lists or recur-
sions do define reference and truth.
 TRUTH AND SEMA'.\!TICS 283

we do not speak English*. For we are sensitive to the presence of

words in understanding the sentence: we understand it as we do
on the basis of the systematic effect the words 'penguin' and
'waddle' have on the content of what is said. So a decent W* 11
should not attribute English* to us. Suppose on the other hand
that the theorem is derived from some clauses governing the
components. Perhaps in English* penguins refers to men, and
waddles is satisfied by things which land on the moon.But then
things are liable to go wrong elsewhere. Suppose the predicate
covered in black and white feathers is satisfied in English* (as it is in
English) by things which are covered in black and white feath-
ers. Then consider further sentences. Penguins are covered in black
and white feathers is true in English* iffmen are so covered. And
now W* comes in. Most of us disbelieve that men are so
covered, but believe that the sentence is true in the language we
actually speak- English. IfW* is sensitive to this (for instance,
it finds that people hold the sentence to be true, but behave in
no way which suggests that they believe that men are covered in
black and white feathers), then it is likely to disqualify English*
from being thought of as the language we actually speak.
So the idea is that the fine-grained difference between sen-
tences - differences of meaning - can be constructed on the
basis of a coarse-grained relation - sameness of truth-value - if
we put enough constraints, in the form of W*, on the way the
coarse-grained truth is certified. To take another example,
consider the two predicates has physical shape and has physical size.
They apply to exactly the same things. So it might appear that
we could write, indifferently, that has physical shape is satisfied in
English by things which have physical shape, or has physical
shape is satisfied in English by things which have physical size.
Yet intuitively they mean something different. Can an approach
which deals only in such clauses salvage this intuition? The idea
is that it can, if the right description of the predicate - the
meaning-giving one -is the only one that can occur as part of an
overall compositional account of English which ends up truly
describing each sentence. Thus x has physical shape is true in
English iff x has physical size. But now take, for instance, the
English predicate modifier 'nice'. A compositional semantics
11
In either a homely or a bleak form.
284 L.\:\GUAGE AND THE \\'ORLD

will want to write down a clause saying how the meaning of'is a
nice cp' relates to that of '.is a cp'. However this is done, it has to
avoid the result that ifwe apply the rule to compute the mean-
ing of'x has a nice physical shape' we get the result: 'x has a nice
physical shape' is true in English iff x has a nice physical size.
For this last remark isfalse: there are things and people with
nice shapes but not nice sizes, and vice versa. If we fed in the
wrong 'intuitive' meaning at one point, we get false T-sentences
at another.
It would thus appear that the discipline of finding a true
T-sentence for every sentence of a population's language rules
out false entries. If the wrong role is assigned to a meaning-
giving feature, then, although in an isolated case (penguins
waddle/men have landed on the moon) a true T-sentence can
arise, over the whole spectrum ofT-sentences which the feature
helps to compute, there will be false ones. The interpreter
adopting this discipline then forms hypotheses about what a
population is likely to believe and say to one another (seeing
them as reasonable, by our lights); he observes which sentences
they use to communicate, and which they assent to and dissent
from, and he works his way into identifying the fine-grained
meaning of each sentence. If, for instance, he finds Englishmen
assenting to 'penguins have black and white feathers', and his
manual tells him that this sentence is true iff men are so
covered, then since we are unlikely to believe this, he should
suspect that he has gone wrong.
Of course, at any stage he may make a wrong turn, but the
beauty of the idea is that error eventually reveals itsel( A
good illustration of this comes from an example ofQuine's. In
chapter II of Word and Object Quine pointed out the size of the
step from observing when a speaker assents to or dissents from
an announcement, to assigning any precise semantic role to the
terms of the announcement. The example he took was native
assent to 'gavagai!', supposed to be made when and only when
rabbits are about. Quine pointed out that although it might
come naturally to us to interpret the word as meaning 'rabbits!',
other hypotheses are possible. In particular the natives might
carve up the world differently; they might be more interested in
parts ofrabbits, or perhaps by not thinking in terms of continu-
ing objects as we do, think in terms ofrabbitform appearances,
 TRUTH AND SEMANTICS 285

or time-slices of rabbits. Nevertheless, as interpretation con-

tinues, these possibilities can be ruled out. For example, if the
people are sensitive to number, and we incline to interpret some
signs as numerals, we may enter the hypothesis that 'zonk'
refers to 2; if they then assent to 'zonk gavagai' just when there
are two rabbits arnund this confirms that they are thinking in
terms of rabbits and not, for instance, undetached rabbit parts,
since two rabbits ·presumably involve more than two unde-
tached parts. Or a black and white rabbit may have black
rabbit parts and white rabbit parts, so if the word 'gavagai'
referred to undetached parts, we should expect the natives to
infer from 'black and white gavagai' both 'black gavagai' and
'white gavagai', whereas given a black and white rabbit we
would not allow that we have both a black rabbit and a white
rabbit. 12 So we can test an entry by matching it up with other
conjectures until we get a theory which best fits their inferential
and descriptive behaviour.
Quine inclined to the alternative view that a wrong entry at a
first stage could be compensated by other wrong entries (e.g.
over numerals, or which words function as conjunction, or the
syntax of adjectives) together conspiring to give a different
manual of interpretation. It would be wrong to infer this possi-
bility from the example, for on the face of it the discipline of
system enforces the right interpretation, just as the discipline of
the crossword puzzle can enforce one solution, even if on any
one clue or on the initial clue various answers seemed possible.
Much depends on whether we are surreptitiously imagining the
procedure of the bleak or the homely interpreter. The homely
procedure will undoubtedly use such methods to enforce
unique interpretation. The bleak procedure will have a whole
universe of bizarre, irrational, and indeed unintelligible
interpretations to select from, and indeterminacy will remain.
The advantage of concentrating upon T-sentences is that the
fine-grained property of meaning emerges from the discipline of
finding a coarse-grained property: a good way of putting it is
that the truth theory, if it characterizes a language which can
properly be attributed to a population (by W*), serves as a
theory of meaning for the sentences of the language. Yet in spite

12
Evans, 'Identity and Predication',Joumal ef Philosophy ( 1975).
286 L.\:\"GUAGE AND THE \\'ORLD

of the attraction of the idea, it brings difficulties which ulti-

mately demand modifications. These arise if we contrast the
intensionality of our own language, with the extensionality of
the theory. I now tum to introducing these notions.
The root idea of the extension of a term is that it is the thing to
which the term applies. The extension of a name is thus its
reference; the extension of a predicate the set of things which
satisfy it. A highly doubtful argument of Frege's extends the
notion to sentences, and leads theorists to say that two sen-
tences have the same extension if they are each true, or each
false. For present purposes we do not need to worry about the
argument: the notion of extensionality relevant to this discus-
sion is one which accepts it. Because ofit, two sentences sands'
have the same extension just when it is also true thats iff s'. Two
predicates may have the same extension, but, intuitively,
determine it in quite different ways and mean quite different
things. ('x has a heart' versus 'x has a liver' is the classic pair.
They apply to just the same creatures, but apparently say
different things about them.) Still more, any two true sentences
or any two false ones have the same extension although obvi-
ously there·are differences of meaning between them.
Now we can think ofa language as a way of putting names,
predicates, and sentences into contexts. A context might, for
instance, be 'the father of ... ', which filled by a name makes up
a term referring to the father of the person named. Of course a
language provides for the embedding of names, predicates,
sentences, and any other components within progressively
more complex contexts, making up larger sentences. For
instance, in English, the name 'Napoleon' might occur in the
sentence 'Napoleon is squat', or in the sentence 'Because ofhis
genes, Napoleon is squat'. The predicate 'is squat' also occurs
embedded in each sentence, and the first sentence is embedded
in the second. Any sentence can therefore be thought of as
providing a context upon its component predicates, names, or
embedded sentences.
A context is extensional ifand only if the extension of the whole
which it makes up, is a function of the extension of the parts.
Thus the context 'the father of ... ' is extensional, because its
reference is determined by the reference of any name intro-
duced. The context 'there are seventeen ... 's' is extensional,
 TRUTH AND SEMANTICS

because when it is completed by a predicate to make a sentence

(e.g. there are seventeen people in this room who have hearts)
the extension (truth or falsity) of the whole sentence is depen-
dent upon the extension of the predicate. Because of this, if any
other predicate with the same extension were substituted, the
truth or falsity ( truth-value) of the whole sentence will be
unaffected: if there are seventeen people who have hearts, and
all and only people who have hearts have livers, then it follows
that there are seventeen who have livers. If we consider only
contexts which make up whole sentences, then the result is,
respectively for names, predicates, and sentences, that:
C(a) is extensional iffC(a) & a= b together entail C(b)
C(F) is extensional iffC(F) & all and only Fs are G's together
entail C( G)
C (s) is extensional iff C (s) & s iff s' together entail C(s')
A language is extensional if all its contexts are - in other
words, if the extension of large parts is determined by the
extension of small parts. The language of, say, elementary
arithmetic is extensional, for the reference of an expression such
as (5 X (3 + ... ) ) depends purely on what is referred to by
any expression substituted. The way the expression refers to the
number is immaterial-for instance, the value is the same if you
putin'7',orifyouputin(l/3 x 21).
English, however, is not on the face of it extensional. The
sentence 'penguins waddle' has the same extension as the
sentence 'men have landed on the moon', but the referring
expressions 'a description ofhow ... ', or 'an explanation of why
... ' is wildly different depending on which is substituted. The
extension of these expressions is therefore not a function of the
extension of components. Similarly 'John believes that ... ' may
be true if one sentence is substituted, and false if the other is.
Quite simple contexts are not extensional. Suppose Fred dived
into the pool, and also thatjohn did. The two sentences 'Fred
dived into the pool' and 'John dived into the pool' are both true.
But the sentence 'Fred dived into the pool and then John dived
into the pool' may be true, while the sentence 'John dived into
the pool and then Fred dived into the pool' is false. In other
words, whether you get a truth by saying' ... and then ... '
depends not merely on the truth-value of sentences you feed in,
288 L\.'\GUAGE AND THE WORLD

but on something else as well - the order of the events they

relate.
The best understood logics are extensional. That is, they
consider the inferential relations amongst sentences, where the
sentences are built up by extensional constructions. Because of
this it seemed formally sensible to require that the language in
which the truth theory is couched, and in which the deduction
of T-sentences from base clauses is done, be extensional - in
other words, that it should be sensitive to no more than the
extensions of elements of the inferences. Hence the need to
operate in terms of the T sentence, which has an extensional
form. By contrast 's' means that pis heavily intensional., for it can
obviously change its own truth-value when different sentences
are substituted for p, even if those sentences have the same
truth-value themselves.
There now arises tremendous tension between the evident
fact that natural languages permit intensional constructions,
and the avowed aim of describing them in an extensional way.
If a language permits intensional constructions, this is because,
as speakers and hearers, we are sensitive to more than the
extensions of the terms or sentences which are the input in those
contexts. Consider for example that the two sentences 'Bert
came' and 'Fred came' may have the same truth-value. But
'Ethel was glad that Bert came' and 'Ethel was glad that Fred
came' may differ: the first could be true and the second false. So
the embedded sentences are not merely contributing their
extensions (truth-values) as their part in the overall process of
arriving at a truth-value for the total sentence. In turn this
means that there can be no truth-theoretic description of'Ethel
was glad that s' which contents itself with an extensional
description of s. That is, it is futile to search for a rule which
goes:

Ifs is Tiff p then 'Ethel was glad thats' is Tiff cpp.

In the first occurrence you can substitute any two sentences

with the same truth-value. But if 'cpp' gives the sense of 'Ethel
was glad that p', then in the second you cannot. The only rules
of this form that you could get would need to use a much
stronger Sip relationship:
 TRUTH AND SEMANTICS 289

If S means that p, then 'Ethel was glad that S' is Tiff ({)P,

for example.
This does not immediately destroy the prospect of giving an
extensional semantics for such a construction. The escape
would come by redescribing the 'logical form' of the sentence. If
'Ethel was glad that ... 'and' ... and then ... ' are sentential
operators, we have the need for intensional descriptions. But
suppose we parse them differently. Perhaps for example, there
is concealed reference to events in them. 'Ethel was glad that
Fred came' relates Ethel to an event- Fred's coming. Provided
that this is a different event from Bert's coming (which it is)
Ethel may be glad at one and not the other, just as she can kick
one chair and not another. So the project of giving an exten-
sional construction brings with it a heavy commitment to con-
cealed logical forms: sentences are reparsed so that their real
("logical") form enables the inferences they permit to be laid
out in an extensional form. But the need for other theoretical
categories rapidly arises: 'Ethel was glad that she owned
Broadacres' does not relate Ethel to an event, but at best to a
state ofaffairs (and Ethel is not ambiguously glad).
There are many excellent and ingenious suggestions for
forcing recalcitrant contexts into an underlying extensional
form. But is the motivation so very convincing? I suggest that
it is not. For remember that the promise is to build the fine.
grain of meaning upon the coarse grain of purely extensional
description. And then there is a kind of schizophrenia in the
requirement that semantic structure be exhibited in a purely
extensional way. For this dissociates the radical interpreter,
operating vV* (in one version or another), from the ordinary user
of the language. The promise is that by using vV*, the radical
interpreter can sift out which among the enormous number of
true T-sentences is the interpretational, meaning-giving one:
the one that serves as a theory of sense. This gives him a stronger
s/p relation, empirically and properly discovered, than merely
thats is Tiffp. But if he can do such a thing, then so can users of
the language. And if they can do it then they can also build
constructions which are sensitive to the difference - in other
words, which do not get their interpretation on the basis of the
extensions of the inputs, but on the basis of the stronger proper-
290 LA'.\TGUAGE AND THE WORLD

ties. For instance, it can matter to the truth of what we say that
an embedded sentence means one thing or another, not just
whether it is true or false. It can matter to the truth of what we
say whether a predicate picks out a set of things by one feature
or by others, and not just (or even not at all) what the actual
extension of the predicate is.
Let me put the point this way. Suppose a population is one of
what we can call Hopeless Extensionalists. They just cannot
make themselves aware of anything except the extension of
terms: the reference of names, the set of things satisfying predi-
cates, and the truth-value of sentences. Then they cannot build
contexts like ' ... and then ... ', or 'is a nice ... ', whose extension
is a function of anything finer grained than the extension of
components. Since they cannot do this, liV* will have nothing to
do: any extensionally correct description of their sentences will
do as well as any other. There will be no advanced contexts
where a wrong input lower down, by finer-grained standards,
gives a wrong extension to a term or sentence higher up. So
conversely if this is to happen, it is because the population is not
one of Hopeless Extensionalists in which case it is absurd to
treat their language as though they are.
The classical intensional contexts are those which tell what
someone thinks, believes, desires, etc. A sentence starting 'x
thinks (etc) that ... ' and completed by an embedded sentence
identifying what x thinks, naturally puts the embedded sentence
into an intensional context, for it is its content and not merely its
truth-value which determines whether the report of x is true.
Obviously a person may think that s and not think that s*
although s and s* share the same truth-value. So the kind of
clause which gives the semantic form of 'x thinks thats' cannot
start off 'ifs is true iffp then 'x thinks thats' is true ... ' because
the information the antecedent gives us about s is too thin to
provide an input (intuitively: even ifs is true iffp, there is no way
of connecting the truth of the saying that x believes thats top,
just because of that). Why not avoid schizophrenia, and use the
fact that a good 'interpretational' truth theory contains a clause
stating thats is true iff p? The clause would be of the form 'ifa
good, interpretational truth theory contains a clause saying that
sis true iffp, then 'x believes thats' is true iff x believes that p'.
This uses an in tensional input - a stronger s/p connection than
 TRUTH AND SEMANTICS 291

merely that s is T iff p. But it is an input which Davidson's

programme need not eschew as unscientific. Its respectable
status comes from the background theory, telling us of the
virtues which a truth-theory would have and which would
make it fit to serve as an identification of the sense of
sentences. 13
The detailed semantics of belief contexts is complex, and
there are many suggestions in the field. Many try to sidestep the
problem, by in effect denying that the sentence 'x believes thats'
has semantic value which is a function of any feature (such as
the sense or content, or even truth-value) of the embedded
sentences. The crudest version of such a theory would equate 'x
believes thats' with a direct quotation: 'x believes this: "s" .' The
embedded sentence is mentioned, not used, and since this is so
the semantic value of the overall sentence is no function at all of
the meaning of the embedded one - just as the sentence 'x
spelled this word: "dog" 'means what it does regardless of what
the word 'dog' means. You could understand it and tell that it is
true without knowing what the English word means. U nfortu-
nately exactly this separates it from 'x believes that s'. To
evaluate whether 'Galileo believed that Venus is a planet' is
true you need to understand (at least) what 'Venus' refers to,
and what is meant by being a planet. Someone, for example,
who does not possess enough astronomy to understand the
concept of a planet cannot understand the report of Galileo's
belief either. Since this is so, a semantics of these contexts
(called 'propositional attitude contexts') cannot escape the
problem of showing how we use our underst<1.nding of the
embedded sentence to arrive at an understanding of the whole.
And this requires finding some property of it beyond its exten-
sion, and properties of embedded predicates beyond their
extensions (since Tom may believe that ifa thing is a mammal it
has a heart, but not believe that if a thing is a mammal it has a
liver).
There are two broad motives for suspicion of anything
stronger than extensional semantics. One is that if we allow that
the truth value of 'x believes (etc.) that s' if a function of some
stronger property of the embedded sentence (its sense), we
'-' I have been encouraged to urge this by finding the point well developed in T.
Baldwin, 'Prior and Davidson on Indirect Speech', Philosophical Studies ( 1982).
292 L\:'\GCAGE AND THE \VORLD

must ask when we allow two sentences to have the same sense.
The difficulty arises that we are fairly unprincipled, in practice
at any rate, about the substitutions we allow and disallow in
propositional attitude contexts. The very ideal of sameness of
sense (defined as the property which permits this substitution)
has seemed spurious to many. For instance, consider the two
predicates ' ... is a widow' and ' ... had a husband, who died
while still married to her, and has not since remarried'. These
are surely excellent candidates for sameness of sense. They not
only apply to the same objects, but do so in virtue of the same
feature: they represent the same principle of classification.
They are synonyms. And yet there are sentences which appear
to change in truth-value depending upon which we substitute:
we may incline to describe someone as realizing that Sheila is a
widow, but not realizing that she had a husband, etc. (because,
for instance, we fail to draw one of the obvious consequences).
We may certainly describe someone as puzzling whether all and
only widows are people who, etc., yet it seems wrong to describe
such people as wondering whether all and only widows are
widows. Everyone knows that. The intuitive idea we want is that
although two predicates may have the same sense they may
somehow display that sense in a different way. The difficulty for
the theory of understanding is that of trying to make sense of
this idea. The difficulty for semantics is to connect any such
feature of predicates to their lack of substitutivity in proposi-
tional attitude contexts. Why isn't the pure fact that the predi-
cates represent the same principle of classification, and are
understood in the same way, enough to show that anyone whose
thoughts, beliefs, doubts, etc. are truly described using one of
them, is equally truly described using the other?
By themselves problems of substitutivity could not show that
there is no coherent ideal of sameness of sense. They only show
the need for a theory connecting sense and substitutivity. (Every
kind of term suffers from substitutivity problems - names,
demonstratives, predicates, adverbs, etc. See 9.5.) To show that
the ideal of sameness of sense is itself delusive another range of
argument is needed. One I have already touched upon in 5.4:
the Quinean move of insisting both that the sense ofa sentence
must be identified by its method of verification, or its correct
anchoring in experience, and that sentences (and predicates)
 TRUTH AND SEMANTICS 293

do not have such individual anchorings. But the bias towards

extensional semantics is fed from another source.
This is the idea that the cautious or scientific radical inter-
preter takes only evidence of the extensions of terms, and makes
only conjectures about them. The image is that of the scientific
observer noting when something is said, but not indulging
speculation about why it is said or the principles which make it
correct in the group to say it. However, if we are speaking of
W*1, it is not as though the interpreter even gets as far as
determinate attributions of extensional properties (attributing
references to names, or attributing sets of objects satisfying
them to predicates). So talk of extensional properties of terms is
as unscientific as anything else. On the other hand if we are
speaking of W* h then the whole bias towards extensional
descriptions is quite unrealistic. It is not as though the human
infant or the sensible foreigner actually take in data or make
hypotheses about extensions in the complete absence of evi-
dence and hypotheses about how those extensions are deter-
mined. It is not as though nature builds us with a preference for
the form of hypothesis: 'someone sayings is true in this language
is as right or wrong as he would be ifhe said that p' in advance of
restricting p to give a plausible meaning to s. Initial conjectures
concern objects of attention, the determinate needs and wishes
and beliefs and projects of those with whom we communicate,
and the determinate features of things and responses they
cause. It is in fact conjectures about the meaning of parts and of
wholes which solidify into views about the meaning of wholes
and of parts.

6. Composition, Logic, Translation

There is one pay-off which would arise if we translated a
construction into an extensional language: we would come to
know its logical capacities - why inferences involving it are
valid, if they are, and invalid, if they are not. For extensional
languages are logically beautifully understood. But it is not at
all likely that explanations sufficient to remove the miracle of
a
understanding all the new sentences which connective, say,
can form, is also by the same token sufficient to provide a logic
for it. To illustrate the difference, consider the English condi-
294 LANGUAGE AND THE WORLD

tional: the form of sentence arrived at when two component

sentences are embedded in 'if .. then ... '. This has a systematic
effect. F.or any sentence 'ifs thens*' we can say that if the small
sentence s means that p, and the small sentence s* means that
p*, then the sentence 'ifs thens*' means that ifp then p*. This is
of course hardly news, any more than it was news that '219'
refers to 219. But anyone who can appreciate that point of the
inductive description of ANN can also appreciate that this
description of'if. .. then ... ' might perfectly well take part in an
inductive description of the build-up of meaning of English
sentences which meets a similar need. The need it meets is to
exhibit system. Ifwe had to think of each sentence of the form 'if
... then .. .' as learned afresh, so that its meaning bore no
systematic relationship to the meaning of its components, then
it would be a miracle how speakers could use the connective
outside cases in which they learned it. But they can do this,just
as we can interpret new numerals, and the informal rule I gave
describes the meaning which they arrive at.
But 'if ... then ... 'is logically extremely puzzling. Principles
of inference involving it are notoriously unreliable. For ex-
ample, one might expect this to be a valid form ofargument: 'ifp
then q, if q then r, so if p then r'. This is the property of
transitivity, and we rely on it quite happily: 'if he comes then
she'll come; if she comes things will hot up. So ifhe comes things
will hot up.' Encouraged by such examples we might write
transitivity into our logic for the connective. But then suppose
Smith and Jones are two candidates for an election. We might
happily suppose that if Jones wins Smith will be envious of
Jones. And obviously if Smith dies, Jones will win. But by
transitivity the two together yield that if Smith dies, Smith will
be extremely envious ofJones. This isn't true, or at least doesn't
follow, or we would have a quick proof of a particular kind of
afterlife. 14 A logician needs a theory of the connective which
explains these conflicting appearances. Obviously the descrip-
tion given above does not explain them, because it simply uses
the connective itself to describe its semantic place in English.
Logical puzzles about that use are simply not addressed, any
more than puzzles about the whole notion of referring to num-
bers were addressed by the structural account of ANN.
14
Ernest Adams, The Logic of Conditionals.
 TRUTH AND SEMANTICS 295

Fairly obviously, we could describe 'if ... then ... ' in this
logically unhelpful way because we are allowed English, which
itself contains the connective in all its obscurity, as our
metalanguage. But if we had some logically perspicuous
language, containing only connectives of which we do have a
logical theory, then describing what 'if. .. then ... ' does in this
logically favoured language would reveal its logic at the same
time as describing what it does. For example, ifwe have a full
logical theory of 'sands*' 'it is not the case thats', 's ors*' we can try
to describe the role of'if. .. then ... ' using only these devices. If
we succeed, we reveal its logical powers at the same time. Thus
we might try saying that ifs means that p, ands* means that p*,
then 'ifs thens*' means that either it is not the case that p, or is the
case that p and p*. This interpretation uses only the logically
perspicuous connectives.
In fact, these connectives are extensional, or 'truth-
functional'. From not p and p if! q you can deduce not q. If you
know that either p or q is true, or that r has the same truth-value
as p, you can validly infer that either r or q is true. These
extensional operators are studied in the propositional calculus:
it would be nice for logic if the English conditional could be
expressed in their terms. But in fact it cannot be: the evidence is
overwhelming that conditionals are intensional; that is, it
makes a difference to the truth of a conditional whether the
components mean one thing or another, and substitution of
differently meaning but extensionally equivalent sentences,
turns an acceptable conditional into an unacceptable one. For
instance, it is likely false that nuclear war will break out in the
next ten minutes, and also false that the whole humen race will
become marvellously wise and happy in the next ten minutes.
But the conditional 'if ... then we shall all enjoy ourselves more'
is false or unacceptable if one of these is substituted, but true
and acceptable if the other is. Once again we face the fact that it
is the fine-grained fact about the input sentences, not the fact
certified by any T-sentence, which is needed to determine
whether the overall sentence is true or false. Meaning matters.
Of course, the logic of conditionals is studied successfully by
non-extensional means. But it remains that we can meet the
compositional aim of removing the miracle from our systematic
understanding of conditional sentences, before we have a
296 LANGUAGE AND THE WORLD

satisfactory logic. The compositional aim is one thing; logical

understanding is quite another. Notice, then, that we can say
that we 'give the semantics of' a connective or other part of
speech in two quite different circumstances. One would be
when we had a clause sufficient to remove the "miracle" of our
indefinitely elastic use of the construction, to build any of an
indefinite number of sentences. The very bland description of'if
... then ... ' does that. The other would be when we had a logic
of the connective: when we knew how to describe the inputs and
outputs that matter, so that we could determine which infer-
ences were valid and which were not. One reason why formal
theorists dislike the use of meaning in describing the function of
a connective like 'if ... then ... ' is that it provides no help at all
with this latter task. But this just means that we must separate
the two tasks.
Our modes of composition need not be miraculous or un-
systematic just because they do not operate in a way which
makes the T-form an adequate characterization of sentences, fit
to feed into theories of how they contribute to the meaning of
bigger parts of speech. But there is another feature of the
T-form which raises at least some doubt about the role it is to
play in a theory of composition. This can be illustrated with the
example of demonstrative expressions, such as 'this' or 'now',
which have the trick of having different references on different
occasions of use. Similar problems arise with ambiguous
names.
We can see the problem these raise by contrasting L 1 with a
minutely enriched language L 1 +, just like it except that there is
one further name, 'Jones'. This name is however ambiguous: on
some occasions it is used by speakers to refer to Jim Jones, and
on others to refer to John Jones, in the same way as ordinary
English proper names are apt to be. Now there was a list
"defining" true-in-LP and which itself contained no semantic
terms. But for L 1 + we can give no such list. Consider the
sentence P Jones. We might try any of:

P Jones is T iff: Jimjones is bald

John Jones is bald
Jim Jones is bald and John Jones is bald
Jim Jones is bald or John Jones is bald
 TRUTH AND SEMANTICS 297

But none is these is right. An ambiguous name may occur in a

true sentence provided the intended reference or the reference
on that occasion possesses the relevant property, and none of
these clauses captures that. Neither of the first two allows for
the opposite reference; the third does not allow the sentence to
be true regardless of what the other Jones (the one not talked
about) is like; the fourth allows it to be true because of what the
other Jones is like.
The kind of clause really wanted is something like:
P Jones is true on an occasion of use iff whichever Jones is
referred to on that occasion of use is bald.
Similar considerations lead to similar clauses for sentences
involving demonstrative expressions:
This is bald said by a speaker s in a context C is true iff the
thing which s indicates (demonstrates, refers to) in the con-
text C is bald.
I am pink said by speakers is true iff s is pink.
There is nothing at all wrong with these clauses. But they have
two noticeable properties. They use undefined semantic terms-
notions of something being referred to on an occasion or in a
context. And they do not provide translations of the input
sentence: the clauses describe when the input sentences are
true, but not by giving interpretations with the same content.
Let us consider each point in turn.
Tarski writes that "in particular we desire semantic terms
'referring to the object language' to be introduced into the
meta-language only by definition". He says that if this is so the
definitions will fulfil what we intuitively expect of them, and if
not, then presumably they are at fault through describing truth-
in-L only by using some undefined notion ofreference. Now the
clauses given for the ambiguous name and the demonstrative
do just that. For there is no list definition of reference-on-an-
occasion in L 1 +. So if he uses such a clause, Tarski fails to
provide a list definition of truth in L 1 +. To put it another way,
the manual for L 1 could be used by someone who is quite
incapable of making semantic judgements. But the correspond-
ing manual for L 1 + can only be used by someone who can tell
298 LANGUAGE AND THE WORLD

what is referred to or demonstrated on an occasion. Tarski

himself was well aware that his method could not define truth in
L, in non-semantic terms, for languages containing demonstra-
tives and ambiguous terms. And if the aim is to give a non-
semantic, list account of truth he is quite right. On the other
hand, if the aim is simply to exhibit the structure ofa language,
it is not at all crippled by the presence of these clauses. For there
is no sense in which L 1 + is semantically less well understood
than L 1 • There is nothing peculiarly difficult to the philosopher
of language in the phenomenon of ambiguity: if we have any
idea at all of what it is to use a sign as a name referring to a thing,
we will ipso facto understand what it is to use it sometimes to
refer to one thing and sometimes to refer to others. Ifwe know at
all what would make L 1 the actual language of a population, we
would also know what would make L 1 + their actual language.
(This is not an argument that we should not study the
phenomenon of ambiguity. It is a good source of problems - in
other words, a good way of discovering that we do not under-
stand as much as we would like to about reference or meaning.
All I am asserting is that the mere presence of clauses using the
semantic terms does not make the language structurally less
well understood.)
Now, however, we face the problem that because of the form
they take, these clauses do not provide translations of the object
language sentences. A speaker using a demonstrative or an
ambiguous name does not refer to himself and say that he is
referring to something. He just does it. Thus suppose I say to
you that this is pink, indicating as I do so a scarf in rpy hand,
and you understand me, and agree with me. I am a particular
person, Simon Blackbum, talking, say, at 10.47 on 13 April
1983, in my rooms at Pembroke College. Nevertheless, I have
not said that the object referred to by Simon Blackbum at 10.47
on 13 April 1983 in his rooms at Pembroke College is pink, and
this is not what you understand or agree to. You might not have
the least idea whether this latter proposition is true: you might
not know the time, nor the place, nor for that matter who is
talking to you. So the description of the utterance given by such
clauses does not proceed by producing an interpretation which
means the same, or serves to give the real sense of the original.
This point is a familiar one. Davidson, for example, wrote in
 TRCTH A'.'iD SE'.\1A'.'iTICS 299

1969 that "when there are indexical terms (demonstrative,

tenses), what goes for 'p' cannot be in general whats names or a
translation of it ... " What is not so clear is how it affects the
enterprise of revealing the semantic structure of a language by
providing formal derivations of T-sentences for each sentence.
If the T-sentence can claim to pairs with its translation into the
metalanguage, then we can see why we can fairly suppose
ourselves to have given the semantics of s. But ifs is not paired
with a translation the achievement seems slightly different. It
seems that we have given a perfectly satisfactory discursive
account of when the ambiguous utterance, or the
demonstrative-containing sentence, will be true - it is not as
though there is any mystery of structure or meaning lurking in
the sentences - but not by providing anything which identifies
the sense of the original. So we need to look sceptically both at
the idea that a satisfactory truth theory provides a theorem
which gives the sense of the original sentence (or serves as a
theory of sense), and at the idea that a discursive description of
what words do is somehow inferior to the formal provision of an
interpretation into a metalanguage.
The purpose of finding semantic structure is to remove the
miracle from the elasticity of language. To do this we need a
description of how terms function which explain how we come
to understand new occurrences in unfamiliar sentences. This
explanation would have to take the form of a formal derivation
delivering the content we give to the new sentence only if the
fresh understanding were itself an instance of a translation of the
new sentence into an antecedently understood medium. But
that is the mistake of chapter 2: it imposes a regressive theory of
understanding.
This section has not been much concerned with truth, but
rather with the problems and issues which arise when formal
truth theories are used to elicit the structure of natural
language. None of these problems suggests that the enterprise
of formal semantics is in any way misguided; that would be
absurd. They only show that we must be very careful of
arbitrary impositions of form on such theories, and of supposing
that the connection between the theories and meaning is closer
than it actually needs to be. But the chapter started with the
aim. of relating Tarski and his successors to the problems of
300 LA'.\iGUAGE AND THE WORLD

realism and ofrival classical theories of truth. And the upshot of

that is that the relation is highly indirect. I have argued that not
only is there no distinct rival conception of truth implicit in
Tarski's work, but also that to use it at all in connection with
issues of truth and reference means adding to the neutral core.
The additions can be controversial and subtle, as the top-down/
bottom-up division is; they can suggest various conclusions if
the persona of the radical interpreter is presented in different
ways. But overall the impression is one ofphilos~phical neutral-
ity, of investigations which stand to the philosophy oflanguage
as the description of ANN stands to philosophy of arithmetic, or
as good book-keeping does to the theory of the economy.

Notes to Chapter 8
8.1 Other elementary expositions ofTarski's work include
M. Platts, The Way of Meaning, chs. 1 and 2.
S. Haack, Philosophy of Logics, ch. 7.
As may be clear, I believe that Platts mislocates the real significance
of Tarski's work, both for the philosophy of truth and for the
philosophy of meaning. Haack is a useful guide to modern theorizing
about truth. Carnap was the first writer to be really clear about the
difference between taking a semantics as a stipulative characteriza-
tion of an abstract language, and as a description of the way we
actually use a language: see his Introduction to Semantics.

8.4 Homely and bleak radical interpreters, principles of charity etc.

As well as the literature cited under 2.4, readers might like to,consult:
D. Lewis, 'Radical Interpretation', Synthese (1974).
M. Friedman, 'Physicalism and the Indeterminacy ofTranslation',
Nous (1975).
C. McGinn, 'Charity, Interpretation and Belief, Journal ef
Philosophy (1977).
Hilary Putnam stresses the lack of a backward road from certain
macro facts about sentences to determinacy of reference and meaning
in their components, in his recent book, Reason, Truth and History, ch. 2.
He shows that even ifwe knew the truth-value ofa sentence a speaker
uses, and knew the truth-value it would have in any possible world,
still we could assign references to terms, and meanings to predicates,
 TRl'TH AND SEMANTICS 301

in non-standard, highly divergent ways. If such truth-values had to

be the facts determining references and meanings, then this would
show that they are not up to the job. But this is a highly implausible
version of the 'top-down' route. It certainly distorts the position of
any actual interpreter of the speech of another, to see him first
acquiring this highly esoteric knowledge of truth-values, and their
distribution under different possibilities, and then using that to find
subsentential interpretations. Consider, for instance, how this simply
ignores evidence of causal influences - of the effect upon utterances of
what a speaker is looking at or evidently responding to, of the place of
the utterance in a whole psychology. Putnam goes on to use his
indeterminacy result to argue for an 'internal' view of interpretation,
which somehow puts up with the idea that there are no objective
external facts, discernible from a 'God's eye point of view', about
what terms refer to and what predicates mean. It is hard to be sure
that this helps: the problem Putnam finds arises within a perfectly
ordinary view of the world -one which sees it as containing words and
things, and a multiplicity of possible relationships between them. It is
more plausible to suggest that the facts which do determine the
interpretation of subsentential components do not lie where Putnam
looked for them. They lie in the psychology of the speakers and the
history of their terms (see also 3.1 and 9. 7). See also the interchange
between Field and Putnam, in journal of Philosophy ( 1982).

8.5 A difficult paper using compositional arguments against

Quine's argument 'from below' is
G. Evans, 'Identity and Predication',Joumal of Philosophy (1975).

8.6 The problem of why the disquotational axioms and theorems of

a compositional theory should have an extensional form, rather than
going straight for meaning, is addressed by Barry Taylor in:
'On the Need for a Meaning Theory in a Theory of Meaning', Mind
( 1982).
A famous paper urging caution about supposing that semantic theory
must take one particular form, such as Tarski's, and in particular
urging that the philosophy of meaning should not be bludgeoned by
obscurely motivated formal constraints (such as ones leading to the
dismissal of intensional contexts) is Saul Kripke's:
'Is There a Problem about Substitutional Quantification', in Evans
and McDowell (1976).
 CHAPTER 9

Reference
It was Gatsby's mansion. Or, rather, as I didn't know Mr Gatsby, it
was a mansion inhabited by a gentleman of that name.
F. Scott Fitzgerald, The Great Gatsby.

1. Reference versus Description

Two preoccupations have constantly surfaced in previous
chapters. First we need to make our theory of our understand-
ing cohere with our views of the reality which surrounds us.
Secondly, we need to improve our conception of the way our
thoughts relate to the things they concern. Both preoccupations
get full expression in the philosophy of reference. Referential
expressions ('that man over there'; 'this chair'; 'Napoleon';
'today'; 'I' ... ) identify the things about which we think and
talk, and we understand them to do so. The philosopher's task
is to improve our grasp of each of these facts.
Some of the topics of previous chapters have been relatively
neglected in recent philosophy oflanguage, although others are
hotly debated. The philosophy of reference is the very reverse of
neglected. Indeed if analytical philosophy has one great glory to
its name, it must be the tradition which goes from Frege
through Russell, to modern writers such as Saul Kripke and the
late Gareth Evans, and which wrestles with the problems of
classifying and understanding the various ways we have of
referring to the world around us. The story is a long and
satisfying one, and in this work I can only indicate what seem to
me to be the most interesting aspects of contemporary posi-
tions. Like many writers I locate the interest in a certain ten-
sion, which arises because we can easily come to feel both that
our understanding ofreferential expressions must be intimately
connected with the object referred to, and also that it cannot be.
This tension threads its way through the tradition I have
mentioned. Also the different strategies for resolving the tension
illustrate the delicate balance of power between mind and
language, foreshadowed in chapter 1. It is no easy matter to
 REFERENCE 303

discover whether we are to understand reference first, and

thought second, or to approach it the other way around.
To set the scene we need some terminology. We begin by
considering reference to the familiar spatially located and
bounded things which surround us, and which we can perceive
by the unaided senses - houses, animals, people, tables and
chairs. We think of particular things, and utter our thoughts.
Such utterances characteristically contain a term whose func-
tion is to identify the object thought of - the singular term. Prime
examples include demonstratives (this cup; that cat; ... ) and
ordinary proper names (Napoleon; Regan; ... ). A demonstra-
tive needs a context to give it a reference: it can refer to different
things on different occasions, and a competent user of the
language will know this. The rule for recovering the reference of
a demonstrative from the context gives its character. For exam-
ple, the term 'I' refers to whoever uses it: this rule is known by
competent users, who can use that feature of an utterance (who
produced it) to yield its reference in that utterance. 'That
cat ... 'would normally require some accompanying indication
which would direct the audience's attention.
Now consider the position of someone who understands all
this, but on an occasion does not know which is the relevant
feature of the context. For example he hears someone in the next
room say 'This one is a real diamond', but cannot see the
accompanying indication of which thing is referred to. It is
natural to say that he knows the meaning of the sentence, but
does not know which particular truth or falsity it is used to
communicate. He may assume that something is being said to
be a real diamond. But not knowing which thing it is he does not
know which information is expressed. This can be put by saying
that although he knows the character of the utterance, he does
not on this occasion know its content. But 'content' is a tricky
word, and has been used for other things. So I shall say that
someone not knowing the identity of the thing referred to does
not know the particular information expressed by the utter-
ance. I shall say that in such a case the information is identity-
dependent. An utterance may be described as identity-dependent
if the information it expresses is so. Knowing the information or
misinformation given requires knowing which thing was refer-
red to.
304 LANGUAGE AND THE WORLD

An utterance may be identity-dependent with respect to

some thing, but not with respect to others. For instance, con-
sider the sentence 'the richest man in Germany is hard-
working'. This can be understood by someone who does not
know who the richest man in Germany is (in case that seems
odd, I explain it below). But it cannot be understood by some-
one who does not know which country Germany is. Similarly
'the man who murdered this poor girl must be a maniac' can be
said, understood, and believed when we do not know who the
murderer is; but it would be necessary to know which poor girl
is being referred to.
Russell discovered the crucial importance of the fact that
many sentences which appear to be identity-dependent in some
respects are not really so. To understand this we have to con-
sider various ways of "deconstructing" and reconstructing
sentences. Start with a simple sentence with one singular term,
and a predicate: 'that cat is mangy'. Deleting the singular term
leaves an incomplete expression - a predicate expression or
'open sentence':'- is mangy'. A singular term is a device for
turning such an expression into a full sentence expressing
identity-dependent information. But there are other ways of
reconstructing a full (closed) sentence. Suppose that we are
restricting attention to some range or domain of things (the
animals in some zoo, perhaps). I might then say 'some tiger is
mangy'. This is information, but it is not identity-dependent. It
can be understood and believed without knowing which tiger is
mangy, and even when it is not true that any tiger is mangy, or
indeed when it is not true that there is any tiger in the zoo. The
expression 'some tiger' is therefore not functioning as a singular
term. So how does the sentence work? Its truth depends upon
whether a certain procedure yields a certain result. The proce-
dure is that of taking the predicate around the domain, and
asking of each animal in turn 'is this mangy?' The sentence
reports that if you do that, then at some point, when the
question is asked of a tiger, the answer is affirmative. This does
not inform us which tiger is mangy, and the information may be
perfectly apprehended although no tiger is in fact mangy. This
information is identity-independent. 1
I
In respect of tigers. It can only be understood, as I have presented it, by someone
who knows which domain (which zoo) is in question.
 REFERENCE 305

These ideas are put slightly more formally like this. Starting
with a closed sentence with a singular term, we may delete the
term. I ts position may be marked by a letter: usually x,y, z (with
indices if more are needed). The letter may be treated as a
variable, meaning that the open sentence may be taken into a
given domain, and the letter can be treated as an 'instant name'
of items of the domain in turn. We can then construct the
information that if you do this then at no point do you get a
truth, or at some point you do, or always you do, or mostly you
do, or on exactly n occasions you do ... Such reports of the
quantity ofinstances on which the predicate is satisfied are made
using quantifiers. The two most familiar are ('.fa) Fx, saying that
there is at least one instance of F, and (Vx) Fx, saying that there
is no instance which is not-F, or that everything is F. You can
obviously understand such a report about a domain without
having the first idea which things are in it.
The power of quantificational analyses arises when we con-
sider step-by-step constructions. Consider a particular domain:
say, a philosophy class. Perhaps Tom loves Amanda. Perhaps
someone loves everyone, or everyone loves someone. We can
represent the way these last two differ by a different order of
construction. Deleting both names from 'Tom loves Amanda'
and putting in different letters gives the open sentence 'x loves
y'. Notice then that putting one quantifier in - e.g. '(3:x) x loves
y' still leaves an open sentence. It can be hawked around the
domain, and various people may satisfy it as the variable y
serves as an instant name for them. They will do so if someone
loves them. I can tell you the quantity of times this is true. Ifit is
always true I can say '(Vy) (3:x) x lovesy', meaning that of
everyone it is true that someone loves them. Notice how diffe-
rent other sequences of reconstructing full (closed) sentences
would be. From 'x lovesy' I could quantify over they position-
e.g. '(Vy) x lovesy'. I can take that predicate into the domain,
now letting the x variable light on individuals in turn. Perhaps
there is an example which satisfies the predicate: (3:x) (Vy) x
loves y. Someone loves everyone. In a really jolly class, (Vx)
(Vy) x lovesy. 2
Quantification theory is the heart of modern logic. For our
2
Or, perhaps not. It might be better if (Vx) (3: 1y) (x lovesy &y loves x) where
(3:,y) ... means that there is exactly oney ...
306 LANGUAGE AND THE WORLD

philosophical purposes, it depends on tw.) essential ideas.

Firstly, the information conveyed by a whole sentence may be
reconstructed in a sequence of operations on an open sentence
or predicate expression, itself understood as the result of
deleting actual names from a full sentence. Secondly, the infor-
mation so reconstructed may be identity-independent: under-
standing it need involve no knowledge of which items, if any,
satisfy the predicate of the sentence. Perhaps this last point is at
its most visible when the sentence has the explicit form: 'not
('.fa) Fx', saying that nothing is F.
I mentioned in chapter 8 that Tarski's achievement was to
produce a truth-theory for quantificatiorial languages. This
involved turning the informal introduction of the role of
quantifiers into something more precise. In particular it meant
unifying the role they have in turning open sentences with some
number of free variables into open sentences with one less (the
change from 'x lovesy' to '(3:x) x lovesy') and the role they have
in turning an open sentence, such as this last one, which is still a
predicate, into a full sentence saying something true or false,
when the last free variable is mopped up. The same duality is
seen with other simpler terms, notably connectives. 'And' can
connect sentences ('Tom was ill and Harry was ill as well'), and
predicates ('someone was drunk and ill' - which does not mean
'someone was drunk and someone was ill'). A good semantic
theory should provide a unified account of what the word does
when it conjoins predicates, and when it conjoins sentences. In
the case of conjunction it is not hard to see the underlying
explanation: from conjoined sentences with the same subject we
can delete the singular term, and get 'x was drunk and x was ill'
where repetition of the variable corresponds to the cross-
reference. Something satisfies this if it was drunk and it was ill,
or, as we put it, it was drunk and ill. The predicate forming role of
'and' is explained as a natural extension of its operation on
sentences. I sketch further in the notes how the same thing is to
be tackled for quantifiers.
A particularly important kind of identity-independent infor-
mation tells us that just one thing has a property. 'x is a
much-loved philosopher who writes poetry' may be true ofjust
one person. When l suppose this is so I can introduce a definite
description - and talk of the much-loved philosopher who writes
 REFERENCE. 307

poetry. It was Russell who first realized that sentences involv-

ing definite descriptions may be taken to express identity-
independent information. They can be understood without
knowing who satisfies the description, and indeed when
nothing does. This is because the information expressed can be
represented quantificationally. This can be done in several
ways. 'The much-loved philosopher who writes poetry was here
last night' may be represented:
('.3:x) (xis a MLPWWP) and
not (3:x) (3:y) (xis a MLPWWP andy is a MLPWWP and x¥
y) and
(V x) (if xis a MLPWWP then x was here last night).
The second conjunct is the uniqueness clause. It says that there
are not two separate MLPWWPs. Intuitively, the second
clause says that a certain question is always answered no: the
question 'is this a MLPWWP and separate from another
MLPWWP?' Since there is no value for the x variable of which
this is true, we get the second clause. Another quantificational
representation of the original is:
(3:x) (xis a MLPWWP and (Vy) (ify is a MLPWWP theny =
x) and x was here last night).
Here the quantifier tells of the satisfaction of a complex
condition: being a MLPWWP, and uniquely so, and being here
last night. This is logically equivalent to the tripartite conjunc-
tion above. Russell's famous theory of definite descriptions is
the assertion that the information expressed by many English
sentences using definite descriptions may be represented in one
of these ways. This has the consequence that the information is
identity-independent. From this in turn it follows that such
sentences can be understood although nothing actually satisfies
the description. There may be no much-loved philosopher who
writes poetry, but we can understand the assertion that there is
one, since we understand the predicates involved; understand-
ing that, we can appreciate the assertion that there is exactly
one, and that he was here last night.
It can be seen that Russell's is an empirical theory about
the kind of information actually expressed by many English
sentences. As such, it has been hotly debated. Some moves in
308 LANGUAGE AND THE WORLD

the debate are peripheral to the main topic of thought and

things, which is the heart of Russell's contribution, but I shall
mention them here, since they throw up other points of general
interest. The main value of this part of the debate has been that
it has taught philosophers to respect a number of needed dis-
tinctions - for instance, between semantics and pragmatics,
and between literal meaning and speaker's intention.
Keith Donnellan produced a famous counterexample to Russell
along these lines. 3 At a party I point out to you a man, with the
words 'the man in the corner with the Martini is a lord'.
According to Russell, I have said that there is at least and at
most one man in the corner with a Martini, and he is a lord.
This would be false if, for instance, nobody had a Martini.
Perhaps the man whom I am trying to indicate is holding a glass
of white wine. But, he might be a lord, and this is what I
intended to communicate. My purpose would be to tell you that
this particular man is a lord, and mentioning the Martini is just
a means to that end: my purpose is identity-dependent. Yet the
sentence I choose is a perfectly natural way of trying to fulfil it.
So how can it be right to analyse the sentence as if it merely
talked of the satisfaction of a complex predicate, and bore no
essential reference to the actual object of my interest?
This attack pays too little attention to the distinction between
what is strictly and literally said by an utterance on an occasion,
and what is conveyed, and intended to be conveyed, by it.
Compare this case: we are sitting together, and a vain young
man walks by. I say 'some people from Australia are too big for
their boots'. My saying this on such an occasion strongly con-
veys that I think that that particular man is an example of what I
have in mind. And this may be what I primarily wish to
communicate. But I do not actually say it. What I say - that
some people from Australia are too big for their boots - may be
false, since they are all modest and shy, although that young
man, not being Australian, is indeed appalling; conversely what
I say may be true, although that young man is modest and shy,
so that what I convey is false. This is an example of the distinc-
tion between saying and conveying, or between an implication
of the content of an utterance on the one hand, and on the other
hand, what is often called an 'implicature', or conclusion which
3
'Reference and Definite Descriptions', Philosophical Review ( 1966).
 REFERENCE 309

the hearer will naturally draw from the fact that the utterance
was made, or made in just that way, or with just that intonation,
or instead of something else. (In some situations,just remaining
silent can have such implicatures.) If an implication arises out
of the content of what was said, it is part of the semantics of the
utterance; if it arises in one of these other ways, it is a pragmatic
implication. 'I see that the Professor is sober today' implies
semantically that the Professor is not drunk today, for it cannot
be true unless this is true. But it implies pragmatically, or has as
a "conversational implicature" that he is normally drunk. It is
the difference between:
It cannot be true that p unless it is true that q, and
'He wouldn't have said that 'p' (in just that way, in those
words, etc.) unless he believed that q'
Russell may use this distinction to deal with Donnellan's
case. What is strictly and literally said is identity-independent,
and analysed quantificationally. What is conveyed because of
the surrounding context, and manifest intention of the speaker,
will be that some particular man is a lord. But that is no more
reason for complicating the semantics or descriptions than
the parallel case is for complicating the semantics of 'some
A us tralians ... '.
Another attack which founders on this distinction is one
which points out that sometimes, faced with a definite descrip-
tion which we know to be empty, we refuse to categorize the
overall utterance as false, when Russell says that it is. If some-
one tells me 'the divorce between the Queen and the Duke of
Edinburgh was great fun' I am not likely to reply 'that's false',
but rather I am likely to boggle or ask what he means or deny
that there has been any such event. If I just say that it's false I
seem to imply that there has been such a divorce, but that it
wasn't fun. Russell's reply here is that we often avoid saying
things, or just saying things, which may be perfectly true and
correct to say, but may generate unfavourable implicatures.
Suppose that mostly a certain kind of remark is false for one
reason, but that on occasion it is false for a different kind of
reason altogether. Then just saying that it is false is likely to be
highly misleading - to leave the audience supposing that we
310 LANGUAGE AND THE WORLD

have a normal case. The classic here is the lawyer's question:

'have you stopped beating your wife?' The hapless witness does
not want to answer either 'yes', or 'no'. But in fact 'no' is the
correct answer for an innocent witness to give. Stopping doing
something entails having at some time done it, and at a later
time to not be doing it. If you never beat your wife, then you
never stopped doing so. Normally, however, we only discuss the
question of whether someone has stopped doing something
when we know they once did it: this is why the answer 'no' by
itself is likely to mislead the jury. It needs signalling that the
case is the unusual one, and in the absence of such a signal ('no
- because I never started') the wrong information may be
conveyed to the jury. Russell can make exactly the same point.
We normally talk of objects and events whose existence is not in
question. If someone makes a mistake about existence, such as
supposing that there was such an event as the Queen's divorce,
which he goes on to say something about, we naturally want to
signal that the remark is false in an unusual way.
The strategy is to keep the semantics of definite descriptions
as simple as possible, which means regarding them as Russell
did, and to use ordinary pragmatic principles, governing what
it is natural or unnatural, or helpful or misleading to say, in
order to explain apparent surface divergences from the account.
Looking at it the other way round, we can stipulate that a
population should be speaking an abstract language in which
descriptions are Russellian, and inquire whether we would
expect just the same reactions to utterances as we show. My
own belief is that we would, so that there is nothing in our
linguistic behaviour which shows that we take descriptions in
any way o'ther than Russell's.

2. Putnam's Strategy: Spinning the Possible Worlds

I have been defending Russell's theory as an empirical theory of
English descriptions. But its impact does not depend upon that.
I ts importance is that it raises a possibility. For it gives us an
account of how utterances using definite descriptions can be
identity-independent in respect of whichever item satisfies the
description. That is, we can see how 'the richest man in
Germany is hard-working' can be understood, and believed,
 REFERENCE 311

and express information, even amongst people who do not

know who the richest man in Germany is. The information is
composed from understood operations on understood predi-
cates. It is general or quantificational in form. Now let us return
to the contrasting, genuinely singular proposition.
Suppose I met someone, or hear a musical instrument, or see
a planet, and on a later occasion talk of that person, instrument,
planet. Following Evans 4 I shall call the original occasion the
informational episode; the later thought is based on the original
episode. The intention is to refer to the person, instrument, or
planet involved in the original informational episode. Of
course, talk of one episode is an idealization: it is important that
most of the things we commonly talk about have been seen or
heard by us on numerous occasions. But concentrating upon
one informational episode and one subsequent saying simplifies
the exposition. The information expressed in the saying is
identity-dependent. The utterance is understood only if it is
known which person, instrument, or planet was involved, and,
as already said, it is this which the speaker intends to talk of.
The tensions which dominate modern problems with refer-
ence can now be seen like this. In the original episode the
speaker was affected by the object. He was put into a state of
seeing or hearing something, and as a result he talks ef the
object which affected him. But now we ask a familiar
philosophical question. What might have been the same, about
the subject, even had it been a different object causing the very
same difference to him? The question is of a familiar form,
because it dominates the philosophy of perception. Any explo-
ration of the relations between our experieqce and the objects of
the world which we experience, must consider the way in which
our experiences might have been the same, even had other
objects caused them. Some philosophers might like to deny that
different objects could have caused the very same experiences.
They fear the split between experience and object, and conse-
quent epistemological worries which this might usher in. But
since, apparently, other things might have presented the same
appearances, by having the same features, and impinging on us
in the same way, this cannot be an attractive way to exorcise the
fear.
4
The Varieties of Reference, pp. 121 If.
312 LANGUAGE AND THE WORLD

The importance of this kind of question in the theory of

meaning was first stressed by Hilary Putnam, notably in his
paper, 'The Meaning of Meaning'. Putnam invites us to
imagine different worlds: earth and twin-earth. On twin-earth
the substance that plays the part of water is not chemically
water, but is something else, XYZ rather than H 2 0. However
we are to imagine that XYZ has just the same phenomenologi-
cal properties - the same appearance - as water. It looks the
same, tastes and gurgles and splashes and washes the same
way. It is called 'water' in the language of twin-earth, which
indeed is like earth in every single other aspect, so that if one
were transported instantaneously from one to the other, one
would notice nothing different at all. Now most philosophers
accept that the natural kind term 'water' is a singular term
referring to the particular chemical kind of stuff, H 2 0. From
this it follows that we ought to describe twin-earth by saying
that they don't have water there, although they have this
perfect substitute. It is not true of twin-earth, for instance, that
the seas contain salty water. They contain this other stuff, no
doubt with salt in it. The problem which now causes tension is
what to say about the thoughts and meanings of terms in the
language of twin-earth. What is the same, and what is different?
Is it right, for instance, to say that a person similar to me on
twin-earth- my doppelganger- has the same thoughts that I do,
when I reflect on the wonderful properties of water, or attaches
the same meanings to his words? Or are we to say that because
the reference is different, everything else is too?
Before starting to answer the question, it might help if we
have some other examples in mind. As in the case of earth and
twin-earth, we "spin the possible worlds", or try to keep things
as much as possible the same from the subject's point of view,
while imagining different external causes of his being in the
state he is in. I shall imagine a couple of episodes staying the
same from the subjects' points of view, although caused in three
different ways. Firstly, suppose you and I are musicians, and we
hear the most magnificent violin playing. We think: 'we must
purchase that instrument', and set about tracking it down,
amassing money to buy it, and so on. In the first possible world
( w.) the violin was one particular instrument, Stradivarius
number 1. In the next or substitute possible world (w,) it was a
 REFERENCE 313

different instrument we heard- Stradivarius number 2. Finally

in the third situation there was no instrument and nothing
resembling one. There was a computer simulation of the sound
of an early Stradivarius which was good enough (as computer
simulations now are) to completely deceive us. I call this the
empty possibility (we). Again, consider two astronomers looking
through their instrument, and suddenly seeing a new planet
where there should be no planet. In w. they have indeed spotted
a particular planet, in w, they have spotted a different heavenly
body (a star perhaps), and in we they are victims ofa defect in
the optics of their telescope which made it look as though there
was a planet. But in each story we are imagining the situations
to be the same from the standpoint of the subjects. And in each
case there is a subsequent intention to refer, voiced by some
utterance such as 'we must buy that violin' or 'let's get the radio
telescope onto that planet - it'll make our fame and fortune'.
Since these demonstratives are functioning as singular terms,
in w. there is reference to some particular planet or violin, and in
w, to a different one. In we the speakers do not succeed in
referring to anything at all. But we have said that in each case
the situation is the same from the subjects' point of view, and so
it is natural to say that their understanding of what they have
said is the same in each situation. Ifwe say that, then we need
another ingredient than reference to locate their understand-
ing. There is need for a notion, traditionally called that of sense
as opposed to reference, which can remain constant across these
differences of reference, and whose identity gives the thought,
or object of understanding. This sense would be identified by
the way in which the objects appeared, or the "mode of presen-
tation", in Frege's phrase. And it is this which determines what
the speakers manage to understand by their use of the singular
term. This way in which the objects appeared, or mode of
presentation, is universal in just this sense: it can remain con-
stant across changes of possibility which result in different
objects having been presented in the same way. In our cases it
remains constant across earth and twin-earth, or between the
three possible situations w., w,, and we. I shall call the view that
all thoughts are essentially identified or individuated by these
universal features, the universalist view of thoughts.
This view of thoughts would accept the following argument:
314 LANGUAGE AND THE WORLD

(I) In each possible world, the subject's experience or

mental states, and the features of the environment to
which he is responding, are all the same;
(2) Ascribing thoughts is identifying the subject's psychol-
ogy: it is ascribing mental states;
(3) So in each possible world all the thoughts are the same,
i.e. thoughts are universal.
But there is an apparently inconsistent argument. This one
goes:
( l ') In the different possible worlds the subject apprehends
different truths or facts (in wa the truth that Stradivarius
number l is mellow; in w, the truth that number 2 is, and
in we there is no such truth);
(2') A thought should be individuated by its truth-
conditions;
( 3 ') So some at least of the though ts differ from world to
world, and are not universal.
These thoughts, individuated by reference, could be called singu-
lar thoughts. One possible reaction to a pair of arguments
pulling in different directions is to postulate ambiguity. Perhaps
in one sense thoughts are psychologically real, and then univer-
sal, and in another sense they are tied to a common notion of a
truth or a fact, and then, since there are some different truths or
facts in each possible world (that is how we draw them up, after
all) in this second sense thoughts are singular. The prospects for
saying that just one of these is the right way to classify thoughts
might seem to be fairly bleak. But before adjudicating this, I
wish to explore the idea of the essentially universal character of
thoughts a little more.
On the universalist view, thoughts are identity-independent
even in favoured cases like the ones I have described, where the
thought directly answers to an informational episode. The
thought is identity-independent, or universal, just because
thinkers could think the same even were they in the presence of
a different cause of their thinking. It might then seem that we
should represent the psychology of the thinker best by referring
directly only to the mode of presentation itself, rather than to
the object which it latches onto. In other words, the subject's
 REFERENCE 315

thought would be that the object determined by this mode of

presentation should be bought however much it costs (or will
make our fame and fortune). Let us symbolize such thoughts
as 'the determinant of this mode of presentation is cp' by
'(D(MP) is cp)'. As a definite description 'D(MP)' is identity-
independent, and can be understood in the same way in any of
the possible worlds described. This development is motivated
by the idea that when I think of a thing I do so because it has
features which have impinged upon me. Objects are the sources
of sounds, visual effects, or stories and descriptions, in the light
of which we are aware of them. But the looks, sounds, and so
forth which we have to go on are themselves universal - in
principle other things could have been responsible for them.
Russell was the pioneer of reference who was most acutely
aware of this line of thought. Because of it he made one move
which is now generally regarded as mistaken. He thought that
ordinary names and demonstratives functioned to introduce
the universal features, or in other words that they are equiva-
lent to the definite descriptions 'D(MP)'. But this is not so, for
the very reason that it is true, for instance, in both of our
possible worlds wa and w, that D(MP) is making a beautiful
sound ( the violin example) or is visible through the telescope
(astronomical example). The very same thought or proposition
is true in each situation. But it is only true in wa that
Stradivarius number I is making a beautiful sound, or that
some particular planet is visible. This is not a truth about w, or
we. This is because when we use the singular term we go into a
different semantic register, as it were. We intend to refer to
a particular definite and individual thing or substance, and
nothing we say could have been true had that not existed, but
had been substituted for by other things. This is why the second
argument ( l ')-(3') was able to insist that the different possible
worlds contained different truths. But even if Russell was mista-
ken about the semantics of singular terms, and the way in which
they function differently from definite descriptions, his instincts
about thoughts may remain plausible.
Russell believed that only items which did not allow for
substitute and empty possibilities could be referred to directly.
Only names referring to such a privileged class would function,
for him, as genuinely singular terms. For only when we refer to
316 LA'.',GUAGE AND THE WORLD

such an item will it be true that we couldn't have been thinking as

we are had that thing not existed, and Russell thought that this
condition on thinking was the hallmark of an understood use of
a singular term. Other candidates for singular terms would be
demoted to identity-independence - that is, seen as definite
descriptions. Which things do not allow the possibilities?
Evidently only ones which could not be replaced by doppel-
gangers or could not be absent altogether whilst leaving our
thinking the same. Russell shrank the field down to thoughts
about my own self, to the present time and place, to my own
experiences and properties of things (which themselves define
modes of presentation). Nothing else, in his view, could have
the right essential connection with the very identity of my
thoughts. For others we must peel reference away from sense: in
other words, other singular terms are understood by their con-
nection with a sense or mode of presentation, rather than
directly by their connection with an object.
One consideration which might seem to count against
Russell can be ignored. Notice that there should be no require-
ment that a subject should be able to put words to all the
features which made up the way an object presented itself to
him. Characteristic qualities of appearances - shades of colour,
shapes, qualities of sounds -are just the kinds of thing we find it
impossible to describe adequately. So a subject will often be
unable to put words to the features or qualities making up a
mode of presentation: they will form an indescribable element
of his thinking. Now it might be argued that it is wrong to
ascribe to a subject thoughts which are individuated in ways
which the subject cannot himself put words to. But this is
unattractive. For we shall see that every theorist, not just the
universalist, is going to have to admit some universal thoughts,
regardless of whether the subject is good at articulating them.
Otherwise we deny too many obvious facts - such as the fact
that the subject in the empty world is actually thinking, when
he expresses himself on the apparent object of his experience.
This will become apparent later.

3. Singular Thought Theories

Many modern writers suppose that Russell's conclusion should
 REFERENCE 317

be avoided. It makes modes of presentation too much like inter-

mediaries, getting between us and the things we directly think
about. Once again there is a direct parallel with the philosophy
of perception, where nobody likes the idea that our experiences
act as intermediary, direct objects of perception, forming a veil
between us and the real world.
Consider again the universalist argument of the last section.
It required us to say that the subjects' thoughts are constant:
the same in wa, w,, and we, or the same across earth and
twin-earth. Why should we accept this? Certainly, if thoughts
are directly expressed using descriptions, as when I say 'the
richest man in Germany is hard-working', I will be thinking the
same regardless of who this fortunate person is. But if after an
informational episode I think 'that violin sounds lovely', why
can't we tie the identity of my thought to whichever violin it in
fact is, thence denying outright that in w, I think the same, or
even that in we I express any thought at all? Why not only accept
argument (l')-(3')?
Everyone is going to admit that the reports of thoughts must
be different according to whichever possible worlds we
describe. Just as on twin-earth people enjoy bathing in XYZ,
whereas we enjoy bathing in water, so in w. our musicians
thought that Stradivarius number l sounded lovely, whereas in
w, they thought that number 2 did. Since we here have names or
singular terms occurring inside the 'that' - clauses, which report
thoughts, we can only properly name items which are being
thought about, and this means one violin in one situation, and
another in the other. The question is whether this difference of
report marks a difference of thought reported, or whether it just
marks the difference of an external property of the thinking,
that in one case it is about one thing, and in the other case about
a different thing. In other words, the universalist will claim that
the thinking is essentially or intrinsically the same, in spite of
having a different object. But the opponent - whom I shall call
the singular thought theorist - claims that the thoughts are
themselves different, and that their identity is given by the
object referred to.
It is now necessary to define the opposition to the universalist
more closely. Does he say that there are no universal thoughts,
constant across the different possibilities? Or does he admit that
318 LANGUAGE AND THE \\'ORLD

there are, but also insist that there is a legitimate, self-standing

category of singular thoughts, individuated by reference? And
does he merely stand on the possibility of using argument
(l ')-(3'), perhaps admitting that the term 'thought' may
be ambiguous, and that we can classify thoughts differently,
according to whether our interest is in psychology or in facts
and truths? And in any case, are we talking about thoughts
which a subject has, or rather of the thoughts he manages to
express by his utterances? There are at least two different
versions of two different theories:
(SSTT) Strong singular thought theories. The subject
characteristically has/expresses by his utterance no thoughts
which are constant from world to world. He only has/expresses
genuinely singular thoughts.
(WSTT) Weak singular thought theories. The subject has/
expresses by his utterance universal thoughts. But in addition
he has/expresses by his utterance genuinely singular thoughts.
Clearly a particular problem for SSTT arises with the empty
world. If the identity of the thought is given by reference, what
of the case where there is apparently good, rational, thinking
going on, although through bad luck or bad management, there
is actually no object being referred to? Recently there has been a
renewed tendency for theorists to grasp the nettle. 5 They claim
that the kind of singular thought which can be had, about a
violin or a heavenly body, in w. or w, is simply "not available" in
we. When the experience was caused by nothing like a violin or
planet, then the subjects only purport to be expressing
thoughts, when they subsequently say things like 'we must plot
the orbit of that planet' or 'we must buy that violin however
much it costs'. The most forceful arguments in favour of this
surprising option are given by Gareth Evans. He offers two
direct considerations, and a number of indirect ones.
Evans tries to show that when there is no object nothing can
count as understanding utterances which contain singular
terms purporting to refer to an object of the relevant kind. Since
nothing counts as understanding them, no thought is voiced by
5
Gareth Evans, The Varieties of Reference, esp. chs. 5.1, 9.4, and 9.5, and the
appendices to ch. 6. Also John McDowell, 'On the Sense and Reference of a Proper
Name', Mind(1977).
 REFERENCE 319

them either (understanding just is grasping the thought

voiced). The first argument is in two parts. To begin with it is
argued that only someone who believes that a singular term refers
can understand sentences in which it is embedded. Then it is
argued that this belief, standing as it does as a necessary condi-
tion of understanding something, cannot itself be false. For
understanding is a success - it is telling which thought was
expressed, and that is apprehending a truth about the utterance
- and it can never be a condition on a success that it require a
belief in a falsehood: "truth is seamless; there can be no truth
which it requires acceptance of a falsehood to appreciate"
(p. 331). Myself, I suspect that this attractive principle
nevertheless fails when we come to appreciations of psychologi-
cal truths. For example, there is a truth about what it is really
like to enjoy neo-classical art, or the music of Wagner, but it
may require the acceptance of falsehood, or at least the indulg-
ence of various cognitive and emotional defects, to come to
understand what it is like to do so. Fortunately I do not want to
adjudicate this, for the first part of the argument fails. Consider
the astronomical case. Suppose one astronomer voices 'I will
compute the orbit of the planet, and plot its reappearance'.
Suppose he is heard by a colleague who has himself observed
the matter, and knows perfectly well why it is thought that there
is a new planet there, but is himself bewildered or agnostic. He
cannot bring himself to believe that they have really seen a new
planet, but has no other explanation of what is going on. Is it
really true that such a person cannot understand his more credul-
ous colleague's utterance? He himself might naturally fall into
using the singular term: 'Well, it's highly unlikely that there is a
planet there - still, you get the radio onto it and I'll do the
sums.' It is not belief which is necessary for understanding but-
well, understanding: knowing why the speaker is making that
remark, and why he is right to choose those words to do so.
(Notice that the agnostic astronomer does not pretend that there
is a planet there when he talks about plotting it: he merely
supposes that there is one, for the sake of planning what to do.)
Evans's second argument raises the problem for the condi-
tions for communication, when an empty singular term is used.
A problem which is left unresolved by the train of thought of the
last section, and by any simple talk of thought being identified
320 LANGUAGE AND THE WORLD

by modes of presentation, is to settle what is necessary and

sufficient for mutual understanding. Communication, or under-
standing the thoughts voiced by someone, would apparently
require knowing the mode of presentation under which they
think about or came across or recognize or remember whatever
thing they talk about. It would apparently not suffice for under-
standing that we just know which object they talk of, because for
the universalist it is not reference which identifies thought. Now
Evans believes that in the normal case, when there is successful
reference, it is quite sufficient for mutual understanding that
the hearer know which object was involved in the original
information episode, and hence which the speaker intends to
refer to. There is no requirement that the ways in which speaker
and hearer think of the object be the same, or even approxi-
mately the same. This will occupy us later. But it clearly opens
the way for a challenge. If shared reference is the requirement
permitting understanding in the normal case, what cor-
responds to it in the empty case?
So the challenge, to whose who wish to argue that information-
invoking referential communication can take place in the absence of
an object, is to state a communication-inducing relation between the
origin of the speaker's information and the origin of the hearer's
information which does not presuppose that the information origi-
nates in episodes involving the same object, but which, when the
information is from an object, holds in just those cases when it is from
the same object. (p. 337.)
But there is an easy answer to this challenge. Our two
astronomers are so related that had their information been
veridical, they would have been thinking of the same object. In
general, if the same illusion, or fiction, or cause of hallucination,
explains why you and I take ourselves to be talking of the same
thing, then we communicate. The condition is that there should
be an identity (or at least a sufficient overlap) in the explanation
of why we are each saying what we are. This is just the condition
which is met by ordinary communication when there is an
object. For then the same object figures in the informational
episodes which prompt our thinking, and itself provides the
same explanation of the thinking. (There will be a question of
borderline cases, when the way the object figures in the expla-
nation becomes violently different. On this, more in 9.5.) If the
 REFERENCE 321

condition is not met, then there is a lack of communication.

This explanatory criterion for communication is used when we
ask whether two people or two cultures think of the same
imaginary or mythical entities. The Greeks have a god Hermes,
and the Romans a god Mercury. Is it the same god? Their roles
are similar, in the Heavens of Greece and Rome; ifwe press the
question of identity, we are asking whether whatever explains
one cult also figures in explaining the other. Obviously direct
links ( the Romans having picked up the cult from the Greeks,
for example) suffice; or a common causal origin (in the doings of
some god in an ancestral cult, responsible for each of the later
developments), or perhaps simply response to some shared
psychological impulse to which talk of this god gives. 6
Is this criterion arbitrary, or is it dear why we operate it when
we think of the detail? Suppose our two astronomers, or two
violin seekers, are victims of the same illusion, causing each of
them to suppose that they are on the track of a definite violin or
planet. Then this criterion allows them to be communicating,
when they eack talk of it. But suppose that just at the time that
one receives the visual illusion, the other is victim of a faulty
radio receiver. These coincidental mistakes might lead them to
talk together ~f 'that planet ... '. On this account, they would
not be rightly understanding one another. For although they
suppose that their episodes are explained by a common source,
they are not. So it might have turned out that only one of them
was really talking of a planet, and the other was the victim of a
delusion. Had it turned out that way, there would certainly
have been no one single truth or falsehood which they would
each have expressed by the utterance 'that planet is ... '. Fo.r
one might have been right - in touch with the truth - and the
other not. So although they would have taken themselves to
express the same thought, they would be mistaken. Communi-
cation requires a correspondence between whatever it is which
explains my saying what I do, and whatever it is which the
hearer takes to explain those sayings. In the normal case the
object presented in the informational episodes provides the
6 G. E. Moore,' "Real" and "Imaginary" ', Lectures in Philosophy, ed. Casimir Lewy.

Evans quotes Moore's view with approval, and gives the same answer to problems of
communication about an avowedly fictional character or object. But, curiously, he does
not realize how this answers his own challenge (p. 338).
322 LANGUAGE AND THE WORLD

common explanation which cements our thinking together.

And, again, if one object explains one person's thinking, and
another explains another person's, then although they might
utter the same words and take themselves to be communicat-
ing, they will not be. For example, you may see a person and
describe him to me; I identify him with someone I too have
seen, and we go on to talk at cross-purposes, mistakenly suppos-
ing that our thoughts have the same relationship to reality; the
same truth-condition.
I distinguished above four different singular thought
theories. In the light of what we have sa;_J, the stronger versiops
(SSTT) become either unattr.tctive or unmotivated. For thhe
is no reason at all to deny that there is thinking going on in the
empty world, and that some of that thinking is just what is going
on in the other worlds. And there is no good reason, in my view,
for insisting that none of the universal thoughts are expressed
by the utterances the subjects make, and which enable them to
be understood: no good reason for saying that even if they have
universal thoughts, nevertheless they do not voice them, when
they go into the singular register. In these cases, understanding
their utterances is no different from understanding them. At
least, if this is denied it must be because of some semantic thesis
concerning expression: it may hinge upon our precise attitudes to
people who use empty terms. It is open to a society to regard
them as having failed to fulfil the conditions for assertion. But
this will be a boring issue, depending upon fine discriminations
in our attitudes to the deluded subjects. However, the weaker
versions (WSTT) may still justify a fight. So what is in play,
when we admit different truths across these worlds, if we yet
worry whether there are different thoughts?

4. Singular Thoughts and Method

I have urged that Evans's direct arguments in favour of the
singular thought theory, and against the universalist, are not
convincing. But what exactly is at stake ifwe ask whether there
is a sui generis singular thought, which changes its identity from
world to world, and vanishes altogether in the empty world, w.?
Let us be sure of common ground, before making quick deci-
sions. First of all, we have already acknowledged what we can
 REFERENCE 323

call the weak semantic thesis, that the thoughts we ascribe to

our subjects will be described differently, depending on the real
state of affairs surrounding them. This is because they will be
described by putting singular terms into the descriptions of the
thought; and hence although in one world subjects may be
thinking that Stradivarius number l is lovely to hear, if the
situation is different those same thinkers would be correctly
described as thinking that Stradivarius number 2 is lovely to
hear. The issue, remember, is not over the difference of descrip-
tion of thought. This is uncontroversial. The issue is whether
this difference of description is accidental, marking only an
external property of the thinking, or whether there is a thought
had by subjects in one world which is simply not to be had in the
others. Secondly, since the strong versions of the thesis are
implausible, any realistic defence of the singular thought theory
must allow the subjects to have some universal, quantification-
ally expressed, thoughts. For instance, once the illusion was
operating, the subjects who formed plans and projects because
of it certainly had some thoughts. They wanted to track down the
violin responsible for the sound, or to track whichever heavenly
body caused the observation. There is no problem about these:
they are universal, and capable of being entertained or
apprehended in any of the possible worlds. This is why any
weak thesis (WSTT) cannot use the fact that a subject may be
unable to put words to his general thoughts to oppose Russell,
for he himself must countenance an inexpressible element in the
subject's thinking.
The universalist analysis will be this. Someone is thinking
that a is cp, where a is any singular term, if a is rightly related to
his thoughts. But for that to be so, his thoughts need not be
intrinsically connected with a (so that under substitute or
empty possibilities, the thoughts themselves are different). It is
just that the same thinking may or may not have a different
external relation, to one object, or another, or to no object, and
will be describable accordingly. For the universalist, there are
two different styles of description of thoughts (sometimes called de
re, where the thinker is related to a thing, and de dicto, when he is
described in universal terms), but there is only one kind of
thought, which may be describable in different ways.
One natural idea here is that we know enough of what
324 LA:'1:GUAGE AND THE WORLD

thoughts are from our own consciousness of them, just to see

straight away that thoughts might be the same in each different
possible world. This line urges that we know what it would be
like, from the inside, to be in the position ofmy subjects in the
various possible cases. And when we think what it would be
like, we can see that the thoughts we would have under substi-
tute and empty possibilities would be just the same, regardless
of how they were caused. This after all is how the possibilities
are drawn up (so that everything shall app,ear the same to the
subjects). Hence, it is urged, the intrinsic nature of thought is
indeed universal. Modern singular thought theorists tend to
condemn this approach as "Cartesian": implying a spurious
authority in the subject, whose knowledge of the contents of his
own mind is made into something certain and incorrigible.
But this charge is in tum groundless. The doppelganger and
empty possibilities are drawn, as I have remarked, so that
everything is the same from the subject's point of view. This is a
legitimate thought-experiment. Hence there is a legitimate
category of things that are the same in these cases; notably
experience and awareness. Since this category is legitimate, it is
also legitimate to ask whether thoughts all belong to it. In fact
the matter is even worse for the charge ofCartesianism than this
implies. For the different possible worlds w. and w, are not only
the same from the subject's point of view, but also in respect of
the features (sounds, light waves) of the world which impinge
on the subject. So one might even hold that psychology, and the
theory of thoughts in a psychological sense, is concerned not
only with what happens "in the head", but also with certain
relations between the subject and his environment, yet still
oppose WSTT. There is a recent tendency in philosophy to see
the facts about a subject's thoughts as facts about his relation to
his environment. This line opposes the idea that possession of
thought might be just, for instance, a matter of the right inner
representations, and indeed we sympathized with suspicion of
this in chapter 2. Perhaps the question of what a subject is
thinking is at least partly a question of his place in an environ-
ment and his reactions to it. Even if this were so, it would not
establish a WSTT. For it remains quite open that the relevant
features of an environment are themselves universal. The
thinker in w, and even the thinker in we is impinged upon by the
 REFERENCE 325

same sounds, etc. as his counterparts: indeed the only case

which differs in this is that of the pure hallucinator. He is not
impinged upon by the same features. He only has the same
inner experience as if he was. An environmental theory of
thought can try to deny that this suffices to make it true that he
thinks the same: I do not say that this is plausible, but only that
even ifit is plausible, it still does not establish a WSTT. Anoth'er
way of putting the point is this. Some theorists suppose that if
we had a molecule-by-molecule replica of a subject's brain,
then the replica would be possessed of the same thoughts (the
psychological supervenes upon this part of physical reality).
Others might deny this, but say that the psychological only
supervenes upon facts about the brain and its relationship to the
environment. The present point is that even if this latter view
were correct, thoughts could still deserve to be individuated
universally, in terms of features of the environment, which
could in principle be duplicated by different things.
This will not convince the singular thought theorist. He can
reply that 'thought' is after all a theoretical term; if other
considerations of theory suggest that we ought to classify
thoughts as identical only when reference is constant, un-
tutored responses which suggest otherwise will scarcely count.
So we need to turn to theory, and primarily to the purpose of
ascribing thoughts to each other, to see whether that suggests
an answer. Since what is at issue is the existence ofa distinctive
kind of thought- the singular thought, whose identity is tied to
that of the object figuring in any actual informational episode
which prompted the thinking - we should ask what we know
about thoughts and the purpose of ascribing them to one
another. If all our purposes are served by theorizing in terms of
thoughts whose identity persists across the substitute and
empty possibilities, then the singular thought is an unnecessary
abstraction.
Here is an analogy which might suggest this. Suppose I
design a computer program for classifying candidates taking
some large examination. The program takes in names and
marks, and classifies according to some decision procedure.
Suppose one year I run it on Abbott to Zylkovich, and next year
on Acland to Zorowski. Did it function the same way or diffe-
rently each year? The first year it classified one lot of candi-
326 L.-\:\'GUAGE AND THE WORLD

dates, and the next year a different lot. So we could say that it
functioned differently. But this is not what we would say. We
would say that it did the same each year, provided that the
program it followed was unchanged. The decision procedure is
independent of whichever candidates are actually input. If I
call my program ClassStar and sell it, it always does the same.
Perhaps you have a rival, ClassMate. ClassStar and ClassMate
do different things, but only in a sense in which each of them
separately does the same thing, year after year. Now consider
earth and twin-earth. On twin-earth there are people who love
XYZ, who bathe a lot, fish and own yachts, and wake up
looking forward to a day on the XYZ, or giving themselves
kinaesthetic pleasures imagining the soft cool XYZ trickling
over their bodies. Such a twin-earther is just like a water-loving
earthman: in fact, if he were suddenly transported here .he
would not notice the difference, and would continue to enjoy the
wetter aspects of life just as before. But he and water-loving
earthmen function differently from other kinds of people. Those
are people who on twin-earth fear and mistrust XYZ, don't
bathe or fish or like yachting, and who on earth fear and
mistrust water similarly. They function differently. But this
sameness and difference of functioning cuts clean across the
question of whether they are actually faced with XYZ or H 2 0.
Now, the story continues, it is the similarity which cuts across
worlds - the one which groups together water- and XYZ-lovers,
and groups together water- and XYZ-haters, which determines
psychological classifications. People who are similar in that
sense are the ones who think the same, just as programs can be
working the same although in one year they are working on one
set of candidates and in the next year on another. Our
psychologies are determined by the way we react to what we are
aware of. Since the features whereby we are aware of things are
universal (water, or violins, or planets can look or sound or taste
or appear the same as other things of the same kind or even
different kinds), so are psychologies.
This argument certainly makes the case for the category of
universal, identity- or reference-independent thoughts. But
does it succeed in ruling out the other kind-genuinely singular
thoughts? A reply might be that we ought to consider another
aspect of the reason we ascribe thoughts to one another. This
 REFERENCE 327

time it is the explanation of behaviour. People do things

because of what they think and believe and desire. Hence the
objects of thought must be identified in a way which matches
the ways we identify doings. Now often what we want to explain
is why someone relates in some way to things: why did he kick
that table, tread on that snake, study this historical figure? If our
interest is in someone's doings in relation to a particular thing,
then he is best interpreted or explained by ascribing thoughts
which themselves concern that particular thing: he wanted that
table out of the way, he thought that snake was dangerous, he
wanted to emulate that historical figure. Once again, however
(and by now perhaps predictably) the reply invites a counter.
The universalist, remember, allows that it is quite legitimate to
ascribe thoughts in the de re style. This decomposes into saying
that some universal thinking is going on, and that some particu-
lar object relates in the right way to that thinking. It is when
these are both true that people act upon particular things, and it
is the joint truth which explains why they do so. In other words,
I explain why someone attacked that snake by seeing him as in
the grip of snake-hating thoughts which will, on this occasion,
be caused by, and, like the subject's attention, be focused on
that snake. Had another identical snake been the one to cross
his path, the thoughts would have been the same, but their
cause, and the subsequent action, would have been directed
against the substitute. Once again, constancy of universal
thought, in spite of difference of object.
By now it should be becoming doubtful whether the kinds of
argument in play are going to determine the issue between
singular thought theorists and universalists. Certainly there are
no easy victories: properly protected, each side can do justice to
the phenomena which the other side urges against it. Is there
any prospect for a new kind of argument, settling the matter? A
bold line would be to search for a "transcendental" considera-
tion, showing that thought of one kind is only possible because
thought of the other kind exists. So at least we would achieve a
ranking of importance. And here there is one gleam of hope for
the singular thought theorist. Remember how the notion of the
iden ti ty-independen t, universal or quantified way of presenting
information was introduced. It was introduced by imagining a
transformation on an original atomic sentence, containing a
328 L\~GUAGE AND THE WORLD

singular term and a predicate. Does this suggest that it is an

essentially secondary matter, only understood via an under-
standing of genuinely singular information? It is a tempting
line. To pursue it would, I believe, demand looking at issues
beyond this book: issues of the relation between discourse about
individual, identified things, and discourse which is only
"feature-placing" (like 'it is raining'). The issue over thought
and things which we have been discussing, and which seems in
some ways to peter out in indeterminate arguments, really gets
its steam from this old clash: the opposition between theories
taking the notion of a thing as somehow fundamental to our way
of conceptualizing the world, and those which believe we can
think best in terms of manifestation of properties, with the
world of enduring and reidentifiable things a secondary, con-
ceptual construction on top.

5. Modes of Presentation and Substitutivity Problems

We have been tackling the issue between singular thought
theorists, and universalists, strictly in terms of the conception of
thought which we need, and which might buttress either posi-
tion. And, in default of the highly ambitious argument sug-
gested at the end, we found the considerations indeterminate.
We might now go back to more detailed semantic phenomena to
see if they make a difference. That is, do the phenomena of
language, of the way we actually ascribe thoughts using singu-
lar terms, suggest one line rather than another?
This is the primary source of argument for the universalist
position. For singular terms inside 'that'-clauses often seem to
be doing more than just telling us ef what someone is thinking.
They seem to suggest not only the object of thought but the way
it is thought about. The classic discussion of this is Frege's 'On
Sense and Reference'. Frege considered cases where someone
comes across the same object by different ways, so that the
appearances do not make it obvious that it is the same object.
For example, a tourist might take the train up the highest
mountain in Wales one day. He steams up a rather bare and not
very steep ~illside, which he knows to be Snowdon, and doesn't
enjoy himself much. Next day he moves to better things, buys
the Ordnance Survey map, and takes the scramble up Yr
 REFERENCE 329

Wyddfa across Crib Goch. He goes back after a great day,

saying things like: 'Snowdon was lousy, .but Yr Wyddfa is
magnificent'; 'Today's mountain was a good deal better than
yesterday's'; and generally he compares Snowdon unfavour-
ably with the great Yr Wyddfa. Unfortunately they are the
same mountain. Then, it is urged, we have to be careful which
name of the mountain we put into reports of what he thinks, or
we can turn true reports into false ones:
He thinks that Snowdon is duller than Yr Wyddfa. (true)
He thinks that Snowdon is duller than Snowdon. (false)
There are now, in the enormous literature on substitutivity
problems, many variants of these cases. They affect all singular
terms equally, and can arise although there is no difference of
singular term (no two separate names or demonstratives). For
example, a conjuror might explain a trick, which depended
upon showing the same object in two different lights or under
different aspects, without the audience realizing that it was the
same: 'they didn't realize that this [he shows it one way] is
identical with this [he shows it the other way].' So it is not just a
problem for names, and hence is not to be approached by
considering anything peculiar to names, as opposed to demon-
stratives or other expressions. An important class of cases con-
cerns personal pronouns. Consider someone seeing himself in a
mirror, without realizing that it is himself he is seeing. Suppose
his trousers are on fire, and he sees that the person in the mirror
- whom he calls that person - has his trousers on fire:
He thinks that that person has his trousers on fire. (true)
He thinks that he has his trousers on fire. (false)
Or, in the first example, the subject might not have expressed
himself using names, but just demonstratives: this mountain is
a good deal nicer than that other one.
Why does it matter to semantic theory that we cannot substi-
tute co-referential singular terms inside these reports, and pre-
serve truth? After all, we can all see in these cases what is going
on. The subject is deceived by the two separate appearances of
the mountain, or by the fact that he doesn't recognize it to be
himself whose image is in the mirror. What's the problem? The
problem is to describe how singular terms function, so that we
330 L\'.\IGCAGE A'.\ID THE WORLD

can see why they don't substitute in these reports of thought. For
it is tempting to say that the two terms have exactly the same
semantic role: each occurrence of'Snowdon' and 'Yr Wyddfa'
refers to the same mountain: 'that person' and the pronoun 'he'
refer to the same person. If reference exhausts their contribu-
tion to the meaning of the sentences in which they occur, then
how can substitution of a different term with exactly the same
role change the truth-value of the report?
A theorist attracted to modes of presentation can now suggest
a semantic role for them, and one which ties in very nicely to the
universalist position on thought. He can say that modes of
presentation determine thought, by the considerations of the
last two sentences, and that what happens in these cases is that
there are different modes of presentation of the same thing. If
we shift the reference of the singular terms so that in these
reports they function to refer to modes of presentation, then the
problem is solved. Using the terminology of9.3, the subject is
aware of the mountain as Snowdon in one way (MP,), and as Yr
Wyddfa in another way (MP); the reports now say:
He thinks that D(MPs) is duller than D(MPJ. (true)
He thinks that D(MPJ is duller than D(MP;). (false)
Unfortunately this elegant solution cannot be quite right. For
consider that this is a valid argument:
Snowdon is a popular mountain;
X thinks that Snowdon is a popular mountain, so
X thinks something true.
Whereas this is not:
Snowdon is a popular mountain;
X thinks that D(MP,) is a popular mountain. so
X thinks something true.
This is not a valid argument, because X might be in the very
state reported in the second premise, although he is not think-
ing ofSnowdon. This would be so ifhe had been presented with
just the same ( universal) features which Snowdon in fact pre-
sented, but not because he was on Snowdon. He might have
been on some unpopular doppelganger mountain, in which case
he would believe only falsely that it was popular. In other
 REFERENCE 331

words, this solution (Frege's own) does not preserve the fact
that when a singular term occurs inside a report of thought, the
subject must be related in some favourable way to the item the
singular term normally refers to. 7
Obviously we want to preserve the instinct that it is the
difference of modes of presentation which causes the apparent
lack of substitutivity. But the argument just presented seems to
block the direct way of doing this, which is to shift the reference
of singular terms, in reports of thought, to those very modes of
presentation. No logic of the issue can afford to shift the refer-
ences of terms from occasion to occasion of use, or quite ordi-
nary arguments become invalid.
To avoid this problem a theorist using modes of presentation
must bring in the right relation (RR) which a subject must have
to a particular thing in order to count as thinking about it. The
argument can then become:
Snowdon is a popular mountain;
X thinks that D(MP,) is a popular mountain;
Snowdon is RR to this thinking of X; so
X thinks something true.
This can be valid. Xis now thinking of Snowdon, that it is
popular, and anybody thinking that is thinking truly. But pro-
tecting the argument this way, we open the way to this:
X thinks that D(MPJ is not a dull mountain;
Snowdon is RR to this thinking of X;
X thinks that D(MP,) is a dull mountain;
Snowdon is RR to this thinking of X; so
X thinks of Snowdon that it is dull, and that it is not dull
The same form of argument will show that X thinks of Snowdon
that it is duller than Snowdon. But this can be tolerated. The
deluded climber does think, of Snowdon, that it is a duller
mountain than Snowdon. But he wouldn't recognize this as a
correct way of putting anything that he thinks. The idea will be
that there will be true reports of people, in the de re style, which
seem to impute actual contradiction to them. But that is just
because someone might believe of the one mountain that it is
7
This argument is one which Russell used against Frege. I give more references to
literature on this difficult point in the notes.
332 L\'.\:GlJAGE AND THE WORLD

dull and that it is not; ifhe gets into this state after the episodes
we described, he makes no logical mistake. His "notional"
world (described in the universal, D(MP) style) is perfectly
consistent. Once again an analogy may help. Imagine a compu-
ter programmed to sort books out. It would seem bad - a
mistake in the program - if it said that the one book was to go
into Philosophy and, inconsistently, into Civil Engineering. But
it wouldn't be bad if the book had been entered differently on
separate occasions - say, once as a book by the philosopher
John Mackie (whose name the machine recognizes), and then
separately as The Cement of the Universe.
Does the climber then believe that Snowdon is duller than
Snowdon? He believes as much efSnowdon. Ifwe dislike put-
ting the contradiction inside the 'that' -clause we will need to
modify the semantics in this way. We would say that although
having any of a variety of universal thoughts, and being RR to
Snowdon, is enough to count as thinking of Snowdon, only the
right kinds of universal thought- ones involving some favoured
Snowdony mode of presentation (MP,) - permit you to be
reported as thinking that Snowdon ... Reports putting singular
terms into the content of what is thought, doubted, etc. will
have two responsibilities. The thinker will have to be thinking of
the right object. And he will have to think ofit in some favoured
way.
My own view is that singular terms do not normally bear this
dual responsibility (some do - see 9. 7). Normally, we truly
report someone as thinking that a was cp, provided his thinking
bears RR to a. There is no further requirement that he think ofit
as a, or in a way especially tied to our own use of the name. So,
for example, I can report someone as thinking that John was a
spy, even ifhe would not recognize ourJohn. Probably John was
dressed queerly, behaving queerly, seeming quite unlike him-
self, when he was encountered. The subject need not have come
across him in any favoured way, or under any particular mode
of presentation. However, when different modes of presentation
are in question we take more care. When it matters, for instance
to the rationality of the subject, that the same object was
encountered in two different ways, so.that he did not realize that
it was the same, we are likely to want to signal this. And
choosing separate terms to relate the situation is a good way of
 REFERENCE 333

doing it: it alerts the audience to the possibility that the subject
failed to realize that it was the same thing twice over. On this
account someone may truly be said to think that Snowdon is
duller than Snowdon. But the true report is not the most
helpful: reporting him as believing that Snowdon is duller than
Yr Wyddfa enables the audience, especially when it knows that
these are just two different names for the same mountain, to
understand what happened.
This account will be highly congenial to the universalist.
Underneath the reports made by relating people to objects
(putting singular terms into reports of thought) there is another
description of their thoughts, made via modes of presentation or
universal features which do not tie them to particular objects.
This description may not in practice be easily expressed, for we
have alreapy noticed that we often cannot put words to the
features of things which impress us. But it is this theoretical
description which matters when we are considering how logical
people are.
Clearly any approach along these lines requires an account of
the right relation, RR. But before sketching that, we should
notice some other semantic problems to which it is relevant.

6. The Contingent A Priori and Negative Existentials

Suppose I know that someone has made the first ascent of the
climb in Wales called Tensor. I know that (3:x)x climbed
Tensor first. But I don't know who did it. Sitting in my
armchair I can make identity-independent judgements, like the
existentially quantified one just given. Suppose, however, that I
attempt a different style of expression. I might say, 'Take
whoever climbed Tensor first. Call him Cedric. Then Cedric
climbed Tensor.' (Evans calls names introduced Hke this
'descriptive names.') Does my .utterance 'Cedric clltnbed
Tensor' express knowledge? Do I know Cedric climbed Tensor?
If so, I seem to have a peculiar, armchair way of obhiining
mountaineering history. For according to my usage, 'Cedric' is
going to refer to some definite person: Joe Brown, say. Joe
Brown actually climbed Tensor first. I know that Cedric
climbed Tensor first; that is, I know of Cedric that he did this.
Cedric is Joe Brown, so I know ofJoe Brown' that he did this.
334 LA'.'IGUAGE AND THE WORLD

Compare this case. The sentence 'I am here' has the peculiar
property that whenever and wherever I utter it, it is bound to be
true. Even if I am lost and do not know where I am, I can
bravely say 'I am here', and know that I am expressing a truth.
But how can this be so, if, as seems evident, the proposition or
information making the content of this sentence is only true
contingently- true, although it might have been false? To know
where I am demands some kind of knowledge, in fact, precisely
that which l don't have when I am lost. So we have the queer
combination that I can say the sentence in perfect confidence,
knowing that I have expressed a truth, although I am totally
ignorant of which truth it expresses. But the puzzle is only
superficial. By coming out with the utterance 'I am here' when I
am lost I do not express knowledge of where I am, precisely
because I do not know the reference of 'here', on the occasion. In
the terminology of 9.1, I know the character of the sentence,
since I am a perfectly competent English speaker and can use
the word 'here' properly, but on this occasion do not know to
which place it refers. Now it would be quite wrong of someone
to say at some later time, indicating where I in fact was:
'Blackburn knew that he was here.' That is what I didn't know,
being lost. When I came up with my sentence, I knew that it
expressed a truth, but, being ignorant of where I was, I did not
know which truth it expressed. This removes the appearance of
a kind of sneaky way of having knowledge, without going
through the normal empirical requirements oflocating myself.
The same analysis suggests itself when we introduce 'Cedric'.
Let us allow that the introduction of a name in this way is
legitimate, so that the right thing to say is that 'Cedric' does
indeed become a name ofJoe Brown. And when I said 'Cedric
climbed Tensor first' I knew I had expressed a truth. But I
didn't know of Joe Brown (i.e. of Cedric) that he did this. I
didn't know that of anybody at all: there was nobody whom I
knew to have climbed Tensor first. Hence, I didn't know that
Cedric climbed Tensor first. Again we can best see this by·
thinking of other people's report of me. Suppose I go through
my little christening and subsequent utterance when Joe Brown
is keeping quiet about having climbed Tensor, to tease the
climbing fraternity. But it is known that Tensor has been
climbed. People are wondering who did it. It later emerges that
 REFERENCE 335

Joe Brown did it, and my name 'Cedric' sticks to him. Even so,
nobody can properly say 'Blackburn knew, that first evening,
that Cedric climbed Tensor first'. Of course I didn't: for all the
story shows, I was as ignorant of that as anybody else.
The alternative view would introduce a category of contin-
gent, yet a priori truth: truths which could have been otherwise,
yet which can be known "from the armchair". The idea that
there should be such a category was first broached by Saul
Kripke, in his famous lecture 'Naming and Necessity'. His
examples were substantially similar: cases where a singular
term is quite properly introduced on the basis of a description,
and where it is then announced, of its reference (by using the
term), that it satisfies the description. The trick is that we can
do this from the armchair, yet it may be contingent that the
particular thing we refer to has the properties used in the
description. However, the above analysis suggests that the trick
fails. For although the subject can utter a truth, and can know
that his sentence is true, and can know what it means (the
subject knows what 'here' means, and has himself defined the
word 'Cedric'), yet it does not follow that he can be presented as
knowing that a is <p, where a is the object referred to. The
knowledge the subject does have, at the end of his christening
ceremony, is just that which he had before, and it remains
entirely identity-independent.
A closely related family of puzzles concern negative existen-
tials. Consider the difference between:
Take whichever man climbed Tensor first: he exists, and
There is a man who climbed Tensor first.
The second is the kind of sentence to which a quantifier analysis
is appropriate. It tells us that (3x)(x climbed Tensor & not
(3y)(y climbed Tensor before x) ). This would be true regard-
less of who got to the top first. The condition it puts upon a
world is that something beat all other things up the climb. It
could have been true even had Joe Brown never existed, pro-
vided someone else got up instead. But the same is not true of
the first sentence. It is only going to be true provided the very
man who did the climb - Joe Brown - exists. It is not true of
possible worlds in which Joe Brown does not exist. (In case that
terminology is unfamiliar, think of it like this. You can tell a
336 LA:-,JGUAGE AND THE WORLD

story representing the world as though Joe Brown didn't exist,

but someone else got up Tensor first. The first sentence I have
put down is false of the world, as it is represented in your story,
but the second is true.) The first sentence is an example ofan
identity-dependent existence statement. In other words, part
of the sentence functions as a singular term picking out an
object, and the rest tells us that the object exists. We can do
the same with names and demonstratives: Margaret Thatcher
exists; this typewriter exists. Some such things we believe, and
some we don't. For example, there are people who believe that I
exist, but for all I know there are people who believe that I do
not. They may be convinced that I am a fiction or unreal, and
that my writings and other symptoms of my existence are
unreliable. If there is a God, one day he may point out to me
that I thought that He didn't exist, and I will have to admit it.
But now what is it that I am supposed to believe of a thing,
when I believe of that thing that it doesn't exist?
It would be easy if we could treat my belief as identity-
independent: the belief that some combination of character-
istics has no instance, expressed as its not being the case that
(3x) ... x. But the argument of the last paragraph shows that
this is too simple. It misrepresents the proposition, in the same
way that 'D(MP,) is <p' misrepresents the proposition that
Snowdon is <p (p. 330). However we can profit from the uni-
versalist theory of that. On that view a person counts as think-
ing, of some particular, that it is <p if some of his thinking has the
right relation, RR, to that thing. So a person can count as
believing of a particular thing that it does not exist,just in virtue
of believing some identity-independent, universal proposition,
provided that this proposition has the right relation to the
particular object. Thus in virtue of my belief that the universe
has no particularly good creator with human welfare at heart I
will count, if I meet anything much like the Christian God, as
having believed that he does not exist. In virtue ofbelieving that
the name 'Simon Blackburn' is the alias for a committee of old
bores, that my wife has no husband, and other such things, you
would count as believing that I do not exist. Notice that only
things that do exist can stand in the right relation to your
thinking; one can only judge of things that do exist that they do
not.
 REFERE:\TCE 337

There used to be considerable controversy over whether

'existence is a predicate'. Russell in particular responded to the
insights of quantification theory by denying that existence
could be claimed or denied of an individual. 8 The only mean-
ingful judgement, he thought, must be that a certain condition
was met or was not met; existence became a second-level prop-
erty - a property of properties. In this respect it is like number.
When we say that there are seven oranges on the table, we are
not describing the individual oranges (as if we said that there
are mouldy oranges on the table); we are saying that the set of
oranges on the table, or the property of being an orange on the
table, has seven members or instances. And, in Frege's memor-
able phrase, 'affirmation of existence is in fact nothing but
denial of the number nought'. 9 What then of existential state-
ments and their denials which involve singular terms? We can
appreciate a sense in which Russell was wrong, and a sense in
which he was right. He was wrong in that, as we have seen,
there are sentences saying of particular things that they exist or
that they do not. But Russell may, again, have been right in his
instinct: the fundamental kind of judgement can be seen as
quantificational, and not dependent upon the existence of any
actual thing for its identity. Such judgements count as affirming
or denying existence of a thing if there is an object related to
them in an appropriate way.
The light it casts upon these phenomena gives considerable
credit to the universalist position. It is, in effect, by playing the
universal judgement against the singular form of expression
that we go some way to solving problems of substitutivity,
negative existentials, and the contingent apriori. But these areas
are highly controversial, and in the notes I give references to
other points of view.

7. The Right Relation

What is the right relation, RR? We know of things in a
multitude of ways: by perception, by memory, by inference, by
testimony of others. Is there likely to be a clear-cut decision
procedure, telling us that if someone's thought bears just such-
• Russell, 'The Philosophy of Logical Atomism', pp. 233 ff.
• Frege, The Foundations ef Arithmetic, p. 65.
338 L\'\JGUAGE AND THE WORLD

and-such relations to a thing, then he count~ as thinking of it,

whereas if they do not, then he doesn't? Here is Tyler Burge:
The paradigm of this relation is perception. But projections from the
paradigm include memory, many introspective beliefs, certain histor-
ical beliefs, beliefs about the future, perhaps beliefs in pure
mathematics, and so on. There is no adequate general explication of
the appropriate non-conceptual relations which covers even the most
widely accepted projections from the perceptual paradigm. 10
In other words, we would have no clear rationale for drawing
the dividing line in just one place.
For the universalist, a first thought might be that ifa thing a
actually bears the properties involved in X's thinking, then Xis
thinking of a. The right relation would simply be fitting the
descriptions involved in the universal thought. But this seems
wrong. Suppose I believe a piece of gossip to the effect that
Princess Y is consorting with the world spaghetti-eating cham-
pion (perhaps my informant believes that Luigi Pastrami, who
is consorting with Princess Y, is the world spaghetti-eating
champion). Suppose that in fact Sister Maria Mozarella of the
Blessed Convent is world spaghetti-eating champion. Then she
would fit the description, but I do not on that account believe
that Princess Y is consorting with her. That a thing fits my
description is not enough to make it true that I am thinking ofit,
or referring to it. Nor is fit necessary - a thing may be quite
unlike the way I am thinking of it, yet it might still be it which I
am thinking of.
Certainly there are distinctive and important failures in our
thinking. Notably, I might conceive of something as the wror.g
sort of thing in some fundamental way. Someone, for example,
who fails to realize that I am a person, who believes something
which he expresses by saying 'Simon Blackburn is a commit-
tee', may only doubtfully be said to believe that I am a commit-
tee; he is doubtfully thinking of me. Someone whose relation to
any original informational episode involving a proposed object
of thought is thin and garbled eventually loses any right to be
counted as thinking of that thing. What about cases where the
object played no role in the genesis of the thinking: where it
didn't cause or explain why the subject is thinking as he is?
10
'Belief De Re',Joumal of Philosophy (1977).
 REFERE:\"CE 339

Surely then we are too far away from any informational episode
to allow that thinking of an object is taking place. But if
untutored reports are allowed, even these can be doubtful. If a
prophet predicts in sufficiently detailed terms we may say that
he predicted the coming of this person and not just predicted
that there would be someone who ... (in identity-independent
terms). Or suppose the wife learns that the husband often takes
out and adores a handkerchief with traces of the mistress upon
it. So she buys a new one, identical except for the marks and
substitutes it. She says, as she gloatingly learns of his strangled
gasp of surprise when· he took it out: 'He expected it to have
lipstick marks on it!' She attributes an expectation to the
husband relating him de re to the substitute handkerchief. But
this had no causal influence on the husband's expectation at all.
Another idea is that you may count as thinking of some
particular only if you would recognize it as the object of your
thought under some favourable conditions or other. In some
sense, you should be able to know what it is that you are
thinking about. I believe that this flatters our relations with the
past. For there is no practical sense in which I would recognize,
. say, the rabbit I had as a pet when I was a child. If it were
resurrected I couldn't tell it from a million other rabbits. The
fact that I am thinking of it is not manifested as a skill in
identifying it. All that is true is that I do some thinking which is
explained by my acquaintance with that rabbit. The thinking
may need to satisfy some conditions in itself. For instance, I
may need to have mastered various concepts - that ofa rabbit,
or of a scheme of temporally enduring objects amongst which
we move (and which we frequently lose track of, distort in our
memories, and so on). I can talk intelligently of rabbits and of
the general nature of the world I inhabit. After that, my
multifarious relations with particular items in that world can
act and blend in any variety of ways to create my thoughts: why
expect a principled cut-off point, before which I am thinking of
some of them in particular, and after which I am not?
To think of things as they are - persons, ordinary physical
objects, places, times - demands abilities. It demands what is
commonly gestured at with the phrase 'mastery of a conceptual
scheme'. At the very least this means being able to locate
oneself, as_ an enduring object with one perspective on the world
340 L..\'.\;GlJAGE AND THE WORLD

of similarly enduring and spatially arrayed objects. These are

general abilities. But it does not follow at all that thinking of
some particular thing or place or time demands any special
abilities with respect to it, such as the ability to locate it or to
keep track ofit. On the face ofit we can think of many things we
have lost track of; that is what makes a great deal ofour thinking
so poignant.

8. Perspectival Thinking
In 9.5 I expressed the view that there is no convention
or component of meaning to most singular terms which puts
restrictions on the way a subject can think of their referen<::e,
when we report the subject by putting the term into a 'that'-
clause. 'X thinks that a is qJ' allows for X to think of a under an
indefinite variety of modes of presentation. Apparent failures of
substitutivity- 'he thinks that Snowdon is dull, but he doesn't
think that Yr Wyddfa is dull' - are handled pragmatically. That
is, it is part of being helpful and avoiding misunderstandings
that when a subject has failed to realize the identity of things he
is dealing with, we take care, in reporting him, to give the
audience clues that this is so. And changing the singular term is
one way of doing this.
Putting things this way means identifying the content of these
'that' -clauses by reference. It is the reference of the singular
terms in them which determines the particular truth or falsity to
which the reported subject is related. The truth expressed by
'Snowdon is dull' is a truth about Snowdon and this is the truth
to which X is related when we report that X thinks that
Snowdon is dull. But just as the same mountain may present
very different appearances, and as different mountains might
present the same appearance, so these singular truths may be
capable of different presentations. This leads to twin-earth
cases and to cases of people maintaining conflicting attitudes to
the very same singular thought. To unravel the puzzles we have
been led to distinguish the universal element in thinking: the
identity-independent or notional element, which depends upon
the way in which the object of thought is known by the thinker.
Singular thoughts, identified through reference, do not obey
a principle due to Frege. This is that if someone understands a
 REFERENCE 341

sentence S, and understands a sentence S', then if S expresses

the same thought as S' the subject cannot fail to realize that it
does. According to this principle, thoughts should be, as it were,
transparent to the thinker. There should be no possibility of his
thinking that S, but not thinking that S', as there would be if
thoughts, like mountains, could present different facets. But the
singular thought can present as many different facets as the
object to which reference is made. What we are really up
against is the need to pull thoughts, identified as the truths and
falsehoods about the world, away from thoughts identified as
the objects of propositional attitudes. In the former capacity
thoughts are well individuated by reference. Notice in this
connection that X, thinking that water is <p on earth, may be
thinking truly, whilst his doppelganger, thinking that XYZ is <p,
· may be thinking falsely, in spite of the identity of everything as it
appears to them. Perhaps <p = 'is made of H 2 0'. When
thoughts are tied to truth and falsity, they must be individuated
by reference. But when we tum to thoughts as things which
characterize thinkers, it is not the singular thought we need, but
the idea of a mode of presentation which it, like its object, can
· bear, and which is not tied to reference at all. If we avoid the
category of singular thought, things are slightly easier. We can
accept the same semantics for 'that'-clauses with singular
terms. But we will not puzzle over the phenomenon of the one
subject's different attitude to 'the same singular thought'. For
that will be a bad way of putting the simple truth that someone
can be said to believe that a is <p and that a is not <p when two
consistent universal thoughts which he has are both RR to the
thing a.
But is this adequate to all cases? Consider the example of the
person who believes that that man's trousers are on fire (he
points to the image in the mirror) whilst not believing that his
own trousers are on fire. If singular terms substitute without
fail, then he really believes that his trousers are on fire. This
seems wrong; he doesn't behave like someone who believes that
his trousers are on fire. He makes no attempt to put them out,
for instance. He would be surprised on being told that his
trousers are on fire. (Or at least he would ifhe were told this in a
way which made him realize that it was himself. He wouldn't be
surprised if he told it in a way which left him thinking of the
342 LA:\'GUAGE AND THE WORLD

victim as the man in the mirror.) Isn't it right here to admit a

category of thought identified by something further than refer-
ence, so that the person already knowing of himself that his
trousers are on fire, nevertheless learns something further when
he learns that his trousers are on fire? The connection of
thought and behaviour matters here. Consider:

X thinks that that person has his trousers on fire.

X thinks that he has his trousers on fire.
X thinks that a bomb will soon go off in Pembroke College.
X thinks that a bomb will soon go off here.

X thinks that the world ends at midnight, 1 January 1983.

X thinks that the world ends now.

It seems plausible to say that the second member of each pair

tells us something different about X, even if Xis so situated that
the terms in the first report refer to him, to the place that he is,
and to the time of his thinking (they are his trousers, he is in
Pembroke College, and it is midnight, 1 January 1983). They
tell us something different because they do introduce the way X
thinks of the person, place or time. He thinks of it as himself, or
as here or now. In other words, with these particular indexicals
we have an exception to the general point I suggested, that ways
of thinking of things are not conventionally or as a matter of
meaning tied to particular singular terms. The point is vivid if
we think of communication. IfI try to tell X that his trousers are
on fire, that that bomb will go off here, or that the world ends
now, I just fail to communicate unless he realizes that this
means him, here and now. It is not enough if he realizes that
I am talking of that person, Pembroke College, and midnight,
1 January 1983. It is as though I am trying to communicate an
essentially perspectival truth; one which needs appreciating in
the light of its relation to him, there and then. And this essen-
tially perspectival nature is not revealed if we locate the truth
simply through the reference of singular terms used in reporting
it. For substituting co-referential terms loses the essential rela-
tion to the here and now. These are the important cases for
pressing the distinction between modes of presentation and
reference. For they are precisely cases where the aspect under
 REFERENCE 343

which things present themselves (as me, as here, as now) domi-

nates the thinking that the subject does.
The reader may readily appreciate the difficulty that this
category of truths is likely to cause. For what gives us the
identity of these perspectival thoughts? If it is a different truth
which I come to appreciate when I learn that my trousers are on
fire, although I already knew of myself (but not as myself) that
my trousers were on fire, then is this different truth the one
which you also appreciate when you learn that my trousers are
on fire? Is the thought that today is fine, entertained now, the
same as the thought that yesterday was fine, entertained tomor-
row, or has the perspective changed by then in such a way that
the thought is different? Are perspectival thoughts strictly tied
to one perspective on things, so that one which is apprehended
· as about me, here, now is then lost for ever, and only approxi-
mated to (referentially) by anyone who later thinks about him,
there, then? Notice that there is a strong intuitive sense in which
there is just one fact - today's fineness - which indifferently
makes any of these apparently different thoughts true. There is
no theory yet, in the philosophy of these things, which gives us a
satisfactory picture of the "cognitive dynamics" of these cases:
in other words, which tells us how to classify sameness and
difference of thought across times, or across the utterly different
mode of presentations under which I can think of myself, and
under which you can think of me. It is this problem of indi-
viduating such thoughts which makes it so tempting to retreat
into a purely referential classification, in which we think the
same provided we attribute the same property to the same
object, or time or place, no matter how we think of that object or
time or place.
Adjusting the relations between these essentially perspecti-
val thoughts, and thoughts conceived of in an objective,
context-independent way (timeless truths and falsities), is one of
the hardest problems in metaphysics. It is all very well to say
that a particular angle on things is required when the subject
thinks that he ... , or that now it is ... , or that here it is ... ; but
what is it to apprehend a person as oneself, or a time as the
present, or a place as tqe one which one occupies? It is even
controversial whether one of these egocentric categories of
thought is fundamental and can be used to explain the others,
344 LA'.\!GUAGE AND THE WORLD

or whether there is no ranking. For example, one might suppose

that if we understood self-awareness and self-reference then we
could define now as the time of my utterance or experiences; here
is of course the place where I am. But now is only the time ofmy
current experiences or utterances; here is where I am now. And
then reference to myself is plausibly argued to be possible only
because I can conceive of myself as a being with a spatial
position and a history in time. 11

Further exploration of this takes us deeply into metaphysics,

and into our general understanding of the way in which we
know about the world we inhabit. It is one more of the great
problems into which the philosophy of language blends. It
would be the next step in a journey of which I have tried to
provide the first step, which was to understand why the journey
is worth taking, and to appreciate some of the directions it
ought to take. In this part of philosophy there is no distinction
between general problems of metaphysics and specific prob-
lems in the philosophy of language. The language we want to
understand is that of people who have the puzzling perspective.
This should not leave anyone pessimistic about the importance
of the philosophy oflanguage. The linguistic mode in which the
problems are now couched is no disadvantage, for the lights
provided by such thinkers as Frege and Russell are better lights
in which to improve our understandings than any that existed
before them. To remember the morals derived from the explo-
ration of truth, there is no limit to the improvements we can
struggle towards, even if we do not know whether the struggle is
towards a final theory, fated to be agreed on by all enquiry. So
we should not. find it too pessimistic to conclude about reference
and thought what Conrad concludes about the description of
human beings in general:
And besides, the last word is not said - probably shall never be said.
Are not our lives too short for that full utterance which through all our
stammerings is our only and abiding intention? I have given up
expecting those last words, whose ring, if they could only be pro-
11
Evans, op. cit., ch. 7 is good on this. The relevant literature includes David
Kaplan's work on demonstratives (unfortunately still only available in manuscript)
and the seminal papers by John Perry, 'Frege on Demonstratives', Philosophical Review
( 1977), and 'The Problem of the Essential lndexical', Nous (1979).
 REFERENCE 345

nounced, would shake both heaven and earth. There is never time to
say our last word - the last word of our love, of our desire, faith,
remorse, submission, revolt. The heaven and earth must not be
shaken I suppose - at least not by us who know so many truths about
either. (Lordjim.)

Notes to Chapter 9
9.1 'Russell discovered the crucial ... '
The best sources of Russell's views are the papers:
'Knowledge by Acquaintance and Knowledge by Description', in
Mysticism and Logic.
'The Philosophy of Logical Atomism', in Logic and Knowledge.
Chs. XV and XVI of Introduction to Mathematical Philosophy.
'On Denoting', also in Logic and Knowledge, is relatively hard.
' ... Tarski's achievement ... ' Can be appreciated informally like this:
Suppose we are given a metalanguage which enables us to name things,
including the terms of an object language which we are to describe;
which includes a stock of predicates, negation and conjunction
operators, and constructions enabling us to say 'there are ... ' and
'everything is .. .';
Suppose we are to describe the ways in which the meanings of sentences in
the object language, L, are built up;
Suppose we can say what the names ofL are, and what it is that they refer
to. What the primitive predicates of L are, and what they mean.
Which expressions ofL correspond to negation, conjunction, and the
existential and universal quantifier;
Then the problem is to show what any arbitrary sentence ofL means, by
first giving a structural description ofit-i.e. describing how it is made
up from the stock of names, predicates, operators, and quantifiers -
and having a set of rules determining what those things in those places
do by way of contributing to meaning, or truth-conditions.
We can proceed by listing the translations we can. I shall use the
convention that we underline object language expressions, and sup-
pose that their translation is a non-underlined term of the same form .
.fL In L refers to a
P applies to things just in case they are P
Pa is true iff Pa.
346 LANGUAGE AND THE WORLD

But in fact we don't go for truth directly. We go for the notion ofan
assignment satisfying a predicate ( the implications of this I mention in
a moment, after seeing how it works). This is to exploit the idea of
treating a variable as an 'instantaneous name' of an object, as we
range over a domain in order to evaluate general truths, as explained
in the text. We shall call the assignment of a thing a to a variable.:!,
Cr. - a) The denotation or reference of a variable, under an assign-
ments, we can call s[x ]. So for instance, (x - a)[J:] is just the thing a.
An assignment (K -a) satisfies an open sentence or predicate Fx iff
F(JS. - a)[x]: i.e. iff Fa.
A sentence with no variables in it can be evaluated or interpreted
directly without making any assignment. We can say that the null
assignment, A, satisfies Pa, iff Pa. But for (3:x)Fx to be true, there
must be an extension of the null assignment (A + (J:-a) which
satisfies Fx. In plain terms, there must be an object a, such that when
'F this?' lights on it, the answer is yes. The trick now is to show how
the structure of a sentence is mimicked by assignments which pro-
gressively extend A.
Under any assignment, including A, the reference of a name stays
the same. s [a)is always a. Now we can treat both names and variables
as terms, and collect together the informal explanations like this:
( 1) s satisfies P( t1 ••• tn) iff P (s[t 1 ] ••• s[tn])
(2) s satisfies I cp iff nots satisfies cp
(3) s satisfies p & 1/J iff s satisfies cp ands satisfies '!P_
(4) s satisfies (3:x)cp iff(3:a)s + (:!:. - a) satisfies !E
(5) ssatisfies(Vx)cp iff(Va)s + (K-a)satisfiesp_.
A sentence p_ is true iff A satisfies it.
cp here represents any open or closed sentence, i.e. any predicate or
full sentence of the object language. (3:a) ... just says that there is a
thing which ...
These explanations give us a sequence of moves with which to con-
struct the meaning ofan object language sentence. For example:
Fa is true iffA satisfies it, i.e. iffFa. (By (1).)
(3x)Fx is true iff A satisfies it, i.e. iff (3:a)A + (x - a)
satisfies Fx. This will be so iff(3:a)F(A +(x --a)fa:]), i.e. iff
(3:a)Fa.(By (4) and (1).) -
Notice that as with the description of the recursive structure of the
numeral notation ANN ( 1.3) it is not the final result that is news. It is
that going through the process of finding the translation into the
metalanguage reveals the role of that quantifier or operator or name in
that place, and in that sense reveals the structure of the language.
 REFERE'.\iCE 347

(Readers who are familiar with other expositions of this will notice
that this way of putting it does not use the full panoply of satisfaction
by sequences. This is because the notion of progressively extending
the null assignment gets the same effect more naturally. It mimics the
natural understanding of the quantifiers.)
What is the philosophical impact of defining truth in terms of
satisfaction by an assignment- indeed satisfaction by the null assign-
ment? Not much. It would be quite wrong to conclude that because
we here treat both sentences and predicates as similar (both rep-
resented by cp) somehow the notion of the satisfaction ofa predicate
has a conceptual priority. On the contrary, everything hinges on an
antecedent understanding of what it is for the atomic sentence, with
just a predicate from the basic vocabulary and a name or set of names,
Pa 1 ••• a11 to be true. A satisfies such a sentence iff P(a 1 ••• a11 ), which
means that to determine whether A satisfies such a sentence you have
to make ajudgement;judgements are the bearers of truth and falsity.
There is no way of squeezing past the priority of truth, unless, indeed,
the other arguments for the redundancy theory are felt compelling
(7.3).
The basic notation in this exposition is that of Neil Tennant,
Natural Logic. There are reflections on the rivalry between truth and
satisfaction in G. Evans, 'Pronouns, Quantifiers and Relative
Clauses', Canadian journal of Philosophy ( 1977). I agree with Dummett
(Frege, pp. 519 ff.) against Evans, that the priority of truth imposes no
need to avoid this kind of truth theory.
9.1 The issues arising out of Donnellan's case were discussed in a
number of papers:
A. F. MacKay, 'Donnellan and Humpty Dumpty on Referring',
Philosophical Review ( 1968).
S. Kripke, 'Speaker's Reference and Semantic Reference', in
French et al. ( 1977).
G. Evans, ch. 9.2 ofhis (1982).
Part of the debate concerns the same matters as 4.5: the potential
conflict between what the speaker intends to convey, and the strict
and literal meanings of the words he uses to convey it.
9.2 'The tension which dominates ... '
A good collection for studying different views on this is A. Woodfield
(ed.), Thought and Object. Other seminal papers include:
J. Fodor, 'M~thodological Solipsism Considered as a Research
Strategy in Cognitive Psychology', The Behavioural and Brain
Sciences ( 1980) .
348 LANGUAGE AND THE WORLD

S. Schiffer, 'The Basis of Reference', Erkenntnis ( 1978).

T. Burge, 'Belief De Re' ,Journal of Philosophy (1977).
- 'Individualism and the Mental', in French et al. ( 1979).

9.4 This section is highly programmatic. One position which I find

attractive, but which is not developed in the literature, would find a
compromise between singular thought and universalists: The basis
for a compromise is this. It is noteworthy that the best cases for the
universalist concern items about which we know relatively little -
historical personages, possible figures of myth, and the like. It is
plausible to say that this is because we can keep so much of our
thinking (fixed de re) constant, whilst as it were 'picking off' any single
such individual. It may even be true that when we come nearer home
we can imagine any single focus of thought to be absent or sub-
stituted, whilst the thinker is the same. But the essentially public and
behavioural aspect of thinking could be protected not by denying
sameness of thought in these single cases, but by claiming that they
are necessarily exceptional (a similar move to that in the philosophy
of perception which allows that any one experience could be illusory,
but not that they could all be so). Such a compromise will not do
justice to all the intuitions behind the universalist position, but it may
be nearer the truth than either extreme.

9.5 I believe that it is plausible to see Russell as trying to express

this argument, in a famous passage in 'On Denoting'. See:
S. Blackburn and A. Code, 'The Power of Russell's Criticism of
Frege', Ana(ysis ( 1978).
In effect, the point is that you cannot at all switch the semantic role-
the reference or semantic value-of names between ordinary contexts
and places when they occur inside reports of thought. A writer who
emphasizes this requirement is D. Davidson ('On Saying That').
There are quite alarming consequences of the fact that every kind of
term raises substitutivity problems. For at least in the cases of ordi-
nary reference we can see what is going on. We know how it is that a
speaker can come across an object under two different aspects. But
predicates are supposed to express properties or features of things. If
they fail to substitute, then how do we conceive ofa subject "coming
across" a feature under two different aspects? And they appear to fail
in just the same way:
John realized that the figure was a circle. True
John realized that the figure was the shape defined by the
locus ofa point equidistant from another point. False
 REFERENCE 349

A circle just is that shape: so how canJohn realize the one truth and
not realize the other: how is there room for two separate cognitive
contents? The problem of the failure of even bona fide synonyms to
substitute for one another in propositional attitude contexts is the
'paradox of analysis', discovered and explored by G. E. Moore.
Myself! believe that some at least of the mystery is taken out ofit ifwe
realize that a person may exhibit understanding of a concept (and
hence of a predicate expressing it) in his actions; hence he may act in
ways sufficient to certify that he realizes that Xis ({), although he is not
a ware of a true equivalence between being cp and being (} 1 ••• (} n· This
can happen not because the person is ignorant of the meanings of the
constituent terms, nor because his understanding of any of the con-
cepts individually is deficient, but because he doesn't put together
what he already knows. The importance of mentioning actions here is
to remove a misleading model, according to which understanding a
concept is having it as a literal, diaphanous presence in the mind (like
an image- see 2.2). Ifit were like this, it would be hard to see how one
could understand two synonymous terms without immediately seeing
that they were synonymous. (Dummett endorses this principle in
Frege , p. 95, and I express reservations about it in 'The Identity of
Propositions;.) But as with ordinary referring expressions, there is
still the semantic problem of saying how predicates contribute to the
sense of utterances so that substitutivity fails, even if we feel we
understand what is going on. The tradition Dummett is expressing
thinks of 'whatJ ohn realized' as an object- a Thought or proposition
- with its own identity. The report of John relates him to just this
thought, and if there is only one thought to be had, how is it that
substitutivity fails? Again, I favour a pragmatically inclined answer:
John's actions may show that he grasps the concept ofa circle (he can
recognize them, draw consequences from the fact that a circle is
present, and so on), but also indicate that he does not connect this
concept with that of the locus of a point. When this is so, it would be
misleading to represent him as though he had made the connection,
which is what the second report does.

' ... singular terms do not normally bear this dual responsibility ... '
The essential argument here is this: if ordinary names and demon-
stratives did bear such a responsibility, we might talk of the same
thing by using one, and know ourselves to be doing so, but not
communicate successfully (your term might bear a different sense and
hence express a different thought from the one I take it to bear). But
this seems not to happen. All that happens is that when someone does
come across an object in a different way, it can sometimes be
350 LANGUAGE AND THE \\'ORLD

important to realize that this is so. It does not follow from this that we
do not understand what he says until we realize it.

9.6 For singular terms and existential statements, see:

B. Russell, 'The Philosophy of Logical Atomism'.
G. E. Moore, 'Is Existence a Predicate?', Proceedings cif the Aristotelian
Society ( 1936).
J. Mackie, 'The Riddle of Existence', Aristotelian Society Supplemen-
tary Volume ( 1976).
G. Evans, The Varieties of Reference, ch. I 0.
For the contingent a priori and surrounding matters, see:
S. Kripke, Naming and Necessity.
K. Donnellan, 'The Contingent A Priori and Rigid Designation', in
French et al. (1977).
S. Schiffer, 'Naming and Knowing', in the same volume.
G. Evans, 'Reference and Contingency', The Monist ( 1979).

9.8 The literature on perspectival thoughts includes, as well as the

papers by Perry footnoted in the text, the following (difficult)
con tri bu tions:
H. Castaneda,' "He": a Study in the Logic of Self-Consciousness',
Ratio ( 1966).
D. Lewis, 'Attitudes De Dicto and De Se', Philosophical Review ( 1979).
E. Sosa, 'Propositions and Indexical Attitudes', in Parret and
Bouveresse ( 1982).
M. Dummett, 'A Defence ofMcTaggart's Proofofthe Unreality of
Time', in Truth and Other Enigmas.
 Glossary

Analytic. The original definition (Kant) was that a proposition is

analytic if the predicate is contained in the subject. The more
general modern notion is that a proposition is analytic ifit is 'true in
virtue of the meanings of its tdrms'. The purpose of defining such a
category is to find a harmlds category of necessary truths; ones
which are trivial enough not to raise worries of epistemology.
Atomic. An atomic sentence is one which contains no logical
operators.
A posteriori. Not a priori.
A priori. A proposition can be known a priori if it can be known by
merely thinking, by conceptual means alone.
Bivalence. The principle of bivalence states that every sentence is
either true or false.
Cartesian. Doctrine associated with Descartes. A Cartesian view in
the philosophy of mind sharply separates the mental from the
physical. It may include the view that we have immediate private
access to the contents of our own minds.
Conditional. A sentence of the form 'if ... then ... ' expresses a
conditional. The first gap is filled by the sentence expressing the
antecedent, the second by that expressing the consequent.
Consistent. A set of propositions is consistent if no contradiction can
be derived from it.
Content. See proposition.
Contingent. A proposition which is neither necessarily true, nor
necessarily false.
Conventionalism. The attempt to explain some feature of our intel-
lectual practices as the outcome of convention.
Convention. see 4.3.
Counterfactual. A conditional couched in the subjunctive: 'If ... had
been/were to be the case, then ... would have been/would be the
case.' The supposition made in the antecedent is (usually) taken to
be contrary-to-fact.
Defeasible. A claim whose truth is open to defeat by any of an
indefinite range of countervailing circumstances.
De dicto and de re. A belief or other propositional attitude is construed
de dicto when the content is supposed to be a general one, specified
by a sentence using quantifiers and general terms. It is construed de
352 GLOSSARY

re when the subject is related to some specific thing. Thus 'John

wants a sloop' may mean that John wants, in Quine's phrase, relief
from slooplessness (de dicto) or it may mean that there is some
particular sloop which John wants (de re).
Domain. The set of things in question in a given discourse; the
universe of discourse or more formally the range of variables of a
theory.
Empirical. An empirical truth is one which can be certified true by
experience (see 5.4 and 7.4). An empiricist believes that all truths
which we can know have this property.
Epistemology. The theory of knowledge.
Extension and intension. See 8.5.
Force. The force of an utterance is the speech act, in the sense of
asserting, commanding, questioning, etc. which it is used to
perform. The theory of force needs to find a way of classifying such
acts, and saying what it is, in the way of speaker's intentions, etc.,
which determines the particular force of an utterance on an
occasion.
Functor. A term expressing a function. '( )2' is a functor expressing
the mathematical function of the square of a number; if a term
referring to a number is introduced as an argument, the overall
expression denotes the square of the argument: this is the value of
the function for that argument.
Holism. In the theory of meaning, the view that the meaning of an
individual sentence is a function of its place in a larger whole, such
as a theory or a language.
Idealism. The doctrine that reality is in some way dependent upon
the mind.
Implicatures. Conclusions which may be drawn from thefact that an
utterance was made (perhaps in a certain way). These are distin-
guished from strict logical consequences following from what was
said. Implicatures may arise from conversational factors, or from
conventional factors. See 4.5. and 9.2.
Indexical. A term used to refer, but whose reference is determined by
some feature or features of the context of utterance. See 9.8.
Language. In the abstract, a language is characterized by a syntax,
stating which sequences of signs make up well-formed sentences,
and a semantics stating what meaning the sentences have, in virtue
of the signs they contain and the order they are in.
Metaphysics. The theory of the kind of things and facts that there are.
Modal. The mode in which an assertion is made; as necessarily true or
as possibly true. The extent of the analogy between these modalities
and others (probably true? true now? true in the past?) is con-
troversial.
 GLOSSARY 353

Name. A term whose semantic function is to introduce an object,

about which something is said.
Necessary. A proposition is necessary, or necessarily true, ifit could
not have been false; ifit is true in all possible worlds.
Positivism. The view that scientific or empirical (q.v.) truth is all
that can be understood. The logical positivists believed that by a
process of logical analysis, any meaningful proposition could be
reduced to an empirical one, and any others discarded. This thesis
is the verification principle.
Pragmatics. The pragmatics of an utterance are those features which
concern its effects, the likely consequences of its being made.
Pragmatics is the study of force and of the conversational and
conventional implicatures.
Predicate. Intuitively, the part ofa sentence which can be used to say
of a thing, or set of things, that something is true of them. In this
work a predicate is a linguistic category.
Propositional attitude. A subject's propositional attitudes are
described when he is said to believe, doubt, desire, expect, hope,
regret, etc. that ... ; the gap is filled by a sentence identifying the
content of his beliefs, etc.
Proposition. The content, or piece of information, or individual truth
or falsehood expressed by a sentence.
Quantifier. An expression which can turn a predicate either into
another predicate or into a whole sentence, whose satisfaction
condition or truth-condition is then a matter of the presence or
absence or quantity of things in the domain satisfying the original
predicate. See notes to 9.1.
Realism. See 5.1.
Recursive account/definition. One which derives some value for a
complex expression from repeated applications of a function on
the values of some set of base expressions.
Semantics. See language (see also 1.3, and 8).
Sentence. A unit of languge capable of expressing a single truth or
falsehood.
Speech act. See force.
Syntax. See language (see also 1.3).
Synthetic. Not analytic.
Tautology. A tautology is strictly a truth-function (q.v.) which
reduces to the value T or true whatever the values of its input
arguments. The term is sometimes used loosely for any analytic
statement.
Truth-function. A truth-function is one which, given a sentence or
sentences with a definite truth-value as argument, yields another
354 GLOSSARY

sentence, whose truth-value is a determinate function of the truth-

values of the arguments.
Truth-value. The truth or falsity of a proposition.
Universal. The universal element in thinking is our ability to
take different things to instance the same property or feature. See
chapter 3.
Verification Principle. See Positivism.
Word. The presence of a word is a feature of a sentence which plays a
role in determining its meaning.
 Bibliography
Adams, E. TheLogicefConditionals, Reidel (1975).
Aristotle, De /nterpretatione, ed. J. Ackrill, Oxford University Press
(1963).
- Metaphysics, transl. W. D. Ross, Oxford University Press (1908).
Ayer, A. J. Language Truth and Logic, Gollancz ( 1936).
- The Foundations of Empirical Knowledge, Macmillan ( 1940).
Baldwin, T. 'Prior and Davidson on Indirect Speech', Philosophical
Studies, xxv ( 1982).
Barker, S. and Achinstein, P. 'On the New Riddle of Induction',
Philosophical Review, lxix ( 1960).
Bennett, J. F. Rationality, Routledge & Kegan Paul ( 1964).
- Locke, Berkeley, Hume: Central Themes, Oxford University Press
(1971).
- 'The Meaning Nominalist Strategy', Foundations ef Language, x
(1973).
- Linguistic Behaviour, Cambridge University Press ( 1976).
Berkeley, Bishop. The Principles ef Human Knowledge, Dublin 1710.
Blackburn, S. Reason and Prediction, Cambridge University Press
(1973).
- (ed.) Meaning, Reference and Necessiry, Cambridge University Press
(1975).
- 'The Identity of Propositions', in (1975).
- 'Thought and Things', Aristotelian Society, Supplementary Volume, !iii
(1979).
- 'Truth, Realism and the Regulation of Theory', in French et al.
(1980).
- 'Opinions and Chances', in Mellor ( 1980).
- 'Rule-Following and Moral Realism', in Holtzman and Leich
( 1981).
- 'The Individual Strikes Back', Synthese ( 1984).
Blackburn, S. and Code, A. 'The Power of Russell's Criticism of
Frege: "On Denoting", pp. 48-50', Ana(ysis ( 1978).
Block, I. (ed.) Perspectives on the Philosophy of Wittgenstein, Blackwell
( 1981).
Boorse, C. 'The Origins of the Indeterminacy Thesis', Journal of
Philosophy, lxxii (1975).
Bosanquet, B. Logic, Oxford University Press ( 1911).
356 BIBLIOGRAPHY

Bradley, F. H. Essays on Truth and Reality, Oxford University Press

(1914).
Burge, T. 'Truth and Singular Terms', Nous, viii ( 1974).
- 'On Knowledge and Convention', Philosophical Review, lxxxiv
(1975).
- 'Belief De Re' ,Journal of Philosophy, lxxiv ( 1977).
- 'Individualism and the Mental', in French eta!. (1979).
Butler, R. (ed.) Analytical Philosophy, 1st Series, Blackwell ( 1962).
Carnap, R. Introduction to Semantics, Harvard ( 1948).
- 'The Methodological Status of Theoretical Concepts', Minnesota
Studies in the Philosophy of Science, vol. i, University of Minnesota
Press ( 1956).
Castaneda, H. ' "He": A study in the Logic of Self-Consciousness',
Ratio, vii ( 1966).
- 'Indicators and Quasi Indicators', American Philosophical Quarterly,
iv (1967).
Carruthers, P. 'Frege's Regress', Proceedings of the Aristotelian Society,
lxxxii (1981).
Chappell, V. C. (ed.) The Philosophy of Mind, Prentice Hall (1962).
Chisholm, R. 'Knowledge and Belief: 'De Dicto' and 'De Re'',
Philosophical Studies, xxix ( 1976).
Chomsky, N. 'Quine's Empirical Assumptions' in Davidson and
Hintikka (1969).
- 'Knowledge and Language', in Gunderson (1975).
- Rules and Representations, Blackwell ( 1980).
- 'Principles and Parameters in Syntactic Theory', in Hornstein and
Lightfoot ( 1981 ).
Coffa, A. 'Carnap's Sprachanschauung Circa 1932', Philosophy of
Science Association, ii ( 1976).
Craig, E. J. 'The Problem of Necessary Truth' in Blackburn ( 1975).
- 'Meaning, Use and Privacy', Mindxci (1982).
Dancy,J. 'On Moral Properties', Mindxc (1981).
Davidson, D. 'Truth and Meaning', Synthese, vii (1967).
- 'True to the Facts', Journal of Philosophy, lxvi ( 1969).
- 'On Saying That', in Davidson and Hintikka (1969).
- 'On The Very Idea of a Conceptual Scheme', Proceedings of the
American Philosophical Association, xi ( 1973).
- 'Radical Interpretation', Dialectica, xxvii (1974).
- 'Belief and the Basis of Meaning', Synthese, xxvii ( 1974).
- 'Mental Events', in (1980).
- 'Reality without Reference', in Platts ( 1980).
- 'What Metaphors Mean', in Platts (1980).
- Essays on Actions and Events, Oxford University Press (1980).
 BIBLIOGRAPHY 357

- Inquiries into Truth and Interpretation, Oxford University Press

( 1983).
Davidson, D. and Harman, G. (eds.) Semantics of Natural Languages,
Reidel ( 1972).
Davidson, D. and Hintikka,J. (eds.) Words and Objections: Essays on the
Work ef W. V. Quine, Reidel ( 1969).
Davies, M. Meaning, Quantification, and Necessity, Routledge & Kegan
Paul (1981).
Dennett, D. Brainstorms, Harvester (1978).
- 'A Cure for the Common Code', in his ( 1978).
Dennett, D. and Hofstadter, D. The Mind's 'I', Penguin (1982).
Derrida, J. 'Signature, Event, Context', Glyph, i ( 1977).
Donnellan, K. 'Reference and Definite Descriptions', Philosophical
Review, lxxv ( 1966).
Dummett, M. Frege: Philosophy of Language, Duckworth (1973).
- Truth and Other Enigmas, Duckworth (1978).
- 'Truth', Proceedings of the Aristotelian Society, lix ( 1958).
- 'What is a Theory of Meaning?', in Guttenplan (1975).
- 'A Defence of McTaggart's Proof of the Unreality of Time', in
(1978).
- 'The Philosophical Basis ofintuitionistic Logic', in ( 1978).
Dworkin, R. 'No Right Answer', in Hacker and Raz ( 1977). ···
Edgington, D. 'Meaning, Bivalence and Realism', Proceedings of the
Aristotelian Society, lxxxi ( 1980).
Evans, G. 'Identity and Predication', Journal ef Philosophy, lxxii
(1975).
- 'Pronouns, Quantifiers and Relative Clauses', Canadian]ournal of
Philosophy, vii ( 1977). Also in Platts ( 1980).
- 'Reference and Contingency', Monist, Ixii (1979).
- 'Semantic Theory and Tacit Knowledge', in Holtzman and Leich
( 1981).
- The Varieties ef Reference, Oxford University Press ( 1982).
Evans, G. and McDowell, J. Truth and Meaning, Oxford University
Press (1976).
Feigl, H. and Sellars, W. (eds.) Readings in Philosophical Analysis,
Appleton Century Crofts ( 1949).
Field, H. 'Tarski's Theory ofTruth' ,Journal of Philosophy, lix ( 1972).
- 'Mental Representations', Erkenntnis, xiii (1978).
- 'Realism and Relativism' ,journal ef Philosophy, lxxxiv ( 1982).
Fodor,J. The Language efThought, Harvester (1976).
- 'Methodological Solipsism Considered as a Research Strategy in
Cognitive Psychology', The Behavioural and Brain Sciences, iii ( 1980).
Fraassen, B. van. The Scientific Image, Oxford University Press (1980).
358 BIBLIOGRAPHY

Frege, G. The Foundations of Arithmetic, transl. J. L. Austin, Blackwell

(1959).
- Philosophical Writings: Translations, P. Geach and M. Black,
Blackwell ( 1960).
- Posthumous Writings, transl. P. Long and R. White, Blackwell
(1979).
- 'On Sense and Reference', in (1960).
- 'The Thought: A Logical Inquiry', transl. A. M. and M. Quinton,
in P. F. Strawson, ed. Philosophical Logic, Oxford University Press
(1967).
French, P., Uehling, T. and Wettstein, H. (eds.)
Midwest Studies in Philosophy, vol. 2, Studies in Semantics ( 1977).
Midwest Studies in Philosophy, vol. 4, Studies in Metaphysics ( 1979).
Midwest Studies in Philosophy, vol. 5, Studies in Epistemology ( 1980).
University of Minnesota Press.
Fricker, E. 'Semantic Structure and Speaker's Understanding',
Proceedings ef the Aristotelian Society, vol. lxxxii ( 1982).
Friedman, M. 'Physicalism and the Indeterminacy of Translation',
Nous, ix (1975).
Geach, P. Reference and Generality, Cornell University Press ( 1962).
- 'Assertion', Philosophical Review, lxxiv (1965).
Goodman, N. Fact, Fiction and Forecast, University of London Press
( 1955).
- 'Symposium on Innate Ideas' in Searle (1971).
Graves, C. et al. 'Tacit Knowledge',Journal of Philosophy, lxx ( 1973).
Grice, H. P. 'Meaning', Philosophical Review, lxvi ( 195 7).
- 'Utterer's Meaning and Intentions', Philosophical Review, lxxviii
(1969).
Grover, D. et al. 'A Prosentential Theory of Truth', Philosophical
Studies, xxvii ( 1974).
Guenthner, F. and Guenthner-Reutter, M. (eds.) Meaning and Trans-
lation, Duckworth ( 1978).
Gunderson, K. (ed.) Language, Mind and Knowledge, Minnesota
Studies in the PhilosophyofScience, vol. vii (1975).
Gustafson, D. Essays in Philosophical Psychology, Macmillan (1967).
Guttenplan, S. (ed.) Mind and Language, Oxford University Press
(1975).
Haack, S. Philosophy of Logics, Cambridge University Press (1978).
Hacker, P. and Raz, ]. (eds.) Law Morality and Society, Oxford
University Press ( 1977).
Hanfling, 0. (ed.) Essential Readings in Logical Positivism, Blackwell
(1981).
Hare, R. Moral Thinking, Oxford University Press ( 1981).
 BIBLIOGRAPHY 359

Harman, G. Thought, Princeton University Press (1973).

Harrison, B. An Introduction to the Philosophy of Language, Macmillan
(1979).
Heil, J. 'Does Cognitive Psychology Rest on a Mistake?', Mind, xc
(1980).
Hempel, C. 'On The Logical Positivists' Theory ofTruth', Anarysis, ii
(1935).
Hobbes, Leviathan, ed. M. Oakeshott, Blackwell (n.d.).
Hodges, W. Logic, Penguin (1977).
Holtzman, S. and Leich, C. Wittgenstein: To Follow a Rule, Routledge &
Kegan Paul (1981). .
Hornstein, N. and Lightfoot, D. Explanation in Linguistics, Longman
( 1981).
Hume, D. A Treatise of Human Nature, ed. Selby-Bigge, Oxford
University Press (1888).
- Enquiry Concerning Human Understanding, ed. Selby-Bigge, Oxford
University Press (1893).
Ishiguro, H. 'Imagination', in Williams and Montefiore ( 1966).
Joachim, H. H. The Nature of Truth, Oxford University Press (1906).
Jones, 0. R. (ed.) The Private Language Argument, Macmillan (1971).
Jones, R. F. The Seventeenth Century, Stanford University Press ( 1951).
Kant, I. Critique of Pure Reason, transl. N. Kemp Smith, Macmillan
( 1929).
Kenny, A. 'The Verification Principle and the Private Language
Argument', in 0. R.Jones (1971).
Kim,]. 'Supervenience and Nomological Incommensurables', American
Philosophical Quarterry xv ( 1978).
Kirk, R. 'Underdetermination of Theory and Indeterminacy of
Translation' Anarysis, xxxiii (1973).
Kneale, W. C. 'Are Necessary Truths True by Convention?', Proceedings
of the Aristotelian Society, Supplementary Volume, xxi ( 194 7).
Kripke, S. 'Is There a Problem about Substitutional Quantification?',
in Evans and McDowell ( 1976).
- 'Speaker's Reference and Semantic Reference', in French et al.
(1977).
- Naming and Necessity, Blackwell ( 1980).
- Wittgenstein on Rules and Private Language, Blackwell ( 1982).
Lakatos, I. 'Falsification and the Methodology of Scientific Research
Programmes', Philosophical Papers, vol. i, ed. J. Worrall and
G. Currie, Cambridge University Press (1978).
Lewis, D. Convention, Harvard ( 1969).
- Counteifactuals, Blackwell ( 1973).
- 'Radical Interpretations', Synthese, xxvii (1974).
360 BIBLIOGRAPHY

- 'Attitudes De Dicto and De Se', Philosophical Review, lxxxviii ( 1979).

Lewy, C. Meaning and Modality, Cambridge University Press ( 1976).
Locke,J. An Essay Concerning Human Understanding, ed. P.H. Nidditch,
Oxford University Press (1975).
Macintyre, A. After Virtue, Duckworth (1981).
Mackay, A. F. 'Donnellan and Humpty Dumpty on Reference',
Philosophical Review, lxxvii ( 1968).
Mackie,]. 'Simple Truth', Philosophical Quarterly, xx (1970).
- 'The Riddle of Existence', Proceedings ef the Aristotelian Society,
Supplementary Volume, i ( 1976).
- Ethics: Inventing Right and Wrong, Penguin ( 1977).
- The Miracle of Theism, Oxford University Press ( 1982).
Margalit, A. Meaning and Use, Reidel ( 1978).
Maxwell, G. 'The Ontological Status of Theoretical Entities',
Minnesota Studies in the Philosophy ef Science, vol. iii, University
of Minnesota Press ( 1962)
McDowell,]. 'On The Sense and Reference ofa Proper Name', Mind,
lxxxvi (1977).
- 'Meaning, Communication and Knowledge', in van Straaten
( 1980).
McGinn, C. 'Charity, Interpretation and Beiief',joumal efPhilosoplry,
lxxiv (1977).
- 'Truth and Use', in Platts (1980).
- 'The Structure of Content', in Woodfield (1982).
Mellor, H. Prospects for Pragmatism, Cambridge University Press
( 1980).
Moore, G. E. 'Is Existence a Predicate?', Philosophical Papers, Allen &
Unwin (1956).
- ' "Real" and 'Imaginary"', Lectures in Philosophy, ed. Casimir
Lewy, Allen & Unwin (1966).
Nagel, T. The Possibility of Altruism, Princeton (1970).
- Mortal Questions, Cambridge University Press (1979).
Neurath, 0. 'Protocol Sentences', in Hanfling (1981 ).
Ortony, A. (ed.) Metaphor and Thought, Cambridge University Press
(1979).
Parret, H. and Bouveresse, J. (eds.) Meaning and Understanding, de
Gruyter (1981).
Peacocke, C. Holistic Explanation, Oxford University Press (1979).
- 'Causal Modalities and Realism', in Platts (1980).
- 'Rule Following: The Nature of Wittgenstein's Arguments', in
Holtzman and Leich (1981).
Perry, J. 'Frege on Demonstratives', Philosophical Review, lxxxvi
(1977).
 BIBLIOGRAPHY 361

- 'The Problem of the Es.sential Indexical', Nous, xii ( 1979).

Pitcher, G. Truth, Prentice Hall (1964).
Plantinga, A. The Nature of Necessity, Oxford University Press ( 1974).
Plato. Protagoras and Meno, transl. W. K. C. Guthrie, Penguin Classics
(1956).
Platts, M. The Ways of Meaning, Routledge & Kegan Paul ( 1979).
- (ed.) Reference, Truth and Reality, Routledge & Kegan Paul (1980).
Prawitz, D. 'Meaning and Proofs', Theoria, xxiii ( 1977).
Putnam, H. 'Symposium on Innate Ideas', in Searle ( 1971).
- 'The Meaning of Meaning', Mind, Language and Reality, Cambridge
University Press (1975).
- 'Meaning and Reference', in Schwartz (1977).
- Reason, Truth and History, Cambridge University Press ( 1982).
Quine, W. V. 0. From a Logical Point of View, Harper (1953).
- Word and Object, M.I.T. Press (1960).
- Ontological Relativity and Other Essays, Columbia ( 1969).
- Philosophy of Logic, Prentice Hall (1970).
- The Roots of Reference, Open Court (1974).
- 'Two Dogmas of Empiricism', in (1953).
- 'Notes on the Theory of Reference', in ( 1953).
- 'Epistemology Naturalized', in (1969).
- 'Reply to Chomsky', in Davidson and Hintikka ( 1969).
- 'On the Reasons for the Indeterminacy ofTranslation',journal of
Philosophy, lxvii ( 1970).
- 'Methodological Reflections on Current Linguistic Theory', in
Davidson and Harman (1972).
Ramsey, F. The Foundations of Mathematics, Routledge & Kegan Paul
(1931).
- 'General Propositions and Causality', in Foundations: Essays in
Philosophy, Logic, Mathematics and Economics, ed. D. H. Mellor,
Routledge & Kegan Paul (1978).
Rorty, R. Philosophy and the Mirror ofNature, Blackwell (1979).
- The Consequences of Pragmatism, Harvester ( 1982).
Russell, B. Introduction to Mathematical Philosophy, Allen & Unwin
(1919).
- An Inquiry into Meaning and Truth, Allen & U nwin ( 1940).
- Logic and Knowledge, ed. R. Marsh, Allen & U nwin ( 1956).
- Mysticism and Logic, Allen & U nwin ( 1963).
- 'On Denoting' in (1956).
- 'The Philosophy of Logical Atomism', in (1956).
- 'Knowledge by Acquaintance and Knowledge by Description' in
( 1963).
Ryle, G. 'Mowgli in Babel', On Thinking, Blackwell (1979).
362 BIBLIOGRAPHY

- 'Imagination', in Gustafson (1964).

Scheffler, I. Science and Subjectivity, Bobbs-Merrill ( 1967).
- The Anatomy of Inquiry, Harvard ( 1963).
Schiffer, S. Meaning, Oxford University Press (1972).
- 'The Basis of Reference', Erkenntnis, xiii ( 1978).
Schwartz, S. (ed.) Naming, Necessity, and Natural Kinds, Cornell
University Press (1977).
Scruton, R. Art and Imagination, Methuen ( 1974).
Searle,J. Speech Acts, Cambridge University Press ( 1969).
- (ed.) Philosophy of Language, Oxford University Press ( 1971).
- 'Meaning and Speech Acts', Philosophical Review, lxxi (1962).
- 'Reiterating the Difference', Glyph, i (1977).
- 'Minds, Brains, and Programs', in Dennett and Hofstadter (1981).
- 'The Myths of the Computer', New York Review of Books, xxix, no. 7
( 1982).
Shoemaker, S. 'On Projecting the U nprojectible', Philosophical Review,
lxxiv (1975).
Sluga, H. Frege (Arguments of the Philosophers), Routledge & Kegan
Paul (1980).
Sosa, E. 'Propositions and Indexical Attitudes', in Parret and
Bouveresse ( 1981).
Sprat, T. History of the Royal Society, London ( 1667).
Stitch, S. P. 'What Every Speaker Knows', Philosophical Review, lxxx
(1971).
- 'Grammar, Psychology and Indeterminacy',Journal of Philosophy,
lxxix (1972).
Straaten, Z. van. (ed.) Philosophical Subjects, Oxford University Press
(1980).
Strawson, P. F. 'On Referring', Mind (1950), and in (1971).
- Individuals. Methuen (1959).
- Logico-Linguistic Papers, Methuen (1971).
- 'Intention and Convention in Speech Acts', in ( 1971).
- 'Meaning and Truth', in (1971).
- 'Scruton and Wright on Anti-Realism', Proceedings of the Aristotelian
Society, lxxvii (1976).
Stroud, B. 'Wittgenstein and Logical Necessity', Philosophical Review,
lxxiv (1965).
Swift, J. Gulliver's Travels, ed. L. Landa, Oxford University Press
(1976).
Tarski, A.' Logic, Semantics, Metamathematics, Oxford University Press
(1956).
- 'The Concept of Truth in a Formalized Language', in ( 1956).
- 'The Establishment of Scientific Semantics', in ( 1956).
 BIBLIOGRAPHY 363

- 'The Semantic Conception ofTruth', in Feig! and Sellars ( 1949).

Taylor, B. 'On the Need for a Meaning Theory in a Theory of
Meaning', Mind, xci (1982).
Teller, P. 'Goodman's Theory of Projection', British Journal for the
Philosophy of Science, xx ( 1969).
Tennant, N. Natural Logic, Edinburgh University Press ( 1978).
Thomson, J. J. 'Grue' Journal of Philosophy, !xiii ( 1966).
- 'The Verification Principle and the Private Language Argument',
in 0. R.Jones (1971).
'Williams, B. and Montefiore, A. (eds.) British Analytical Philosopky,
Routledge & Kegan Paul ( 1966).
\Villiams, B. Moral Luck, Cambridge University Press ( 1981).
- '\'\'ittgenstein and Idealism' in ( 198 l).
Williams, C. J. F. What is Truth? Cambridge University Press ( 1976).
Wittgenstein, L. Philosophical Investigations, Blackwell (1953).
- Remarks on the Foundations of Mathematics, Blackwell ( 1956).
- The Blue and Brown Books, Blackwell ( 1964).
- On Certainty, Blackwell (1969).
- Remarks on Philosopmcal Psychology, Blackwell ( 1980).
Woodfield, A. (ed.) Thought and Object, Oxford University Press
( 1982).
\Vright, C. Wittgenstein on the Foundations of Mathematics, Duckworth
(1980).
Zifl: P. 'On H.P. Grice's Account of Meaning', Analysis, xxviii ( 1967).
 Index

abstract ideas 46, 47 conditionals 192-5, 294, 295

.\chinstein, P. I 08 Conrad,J. 344
Adams, E. 294 consistency 198
analysis, paradox of 155, 292, 348-9 contingent a priori 333-5
;\ristotle 28, 40, 51, 224 convention 22, 78, 79, 118-24, 131,
attitudes 101, 102, !47, !49, !67, 169, 251-3
189-95, 256 Convention T 271, 272
Ayer, .\.j. 108, 167, 171, 179 correspondence, and truth 224, 225,
226-9, 233, 243 ff
and epistemology 248-55, 275
Baldwin, T. 291 n
coun terfactual conditionals 209,
Barker, S. I 08
213-17
beha,·iourism !52
Craig, E. 66 n, 93 n, 248 n
Bennett,J. 67, 112, 129
Crusoe, Robinson 84-92
bent rules 74, 80, 96, 97, 128, I 78, 25 l
Berkeley, Bishop 8, 42, 46, 47, 105,
Dancy,.]. 221
157
Davidson, D. 60, 62, 64, 134, 174,
bi\'alence 203-10
175, 184, 185,274,275,276,282,
Blackburn,S.68, 108,221,348
291,298,348
Boorse, C. 68
Davies, .\1. 33, 142
Bosanquet, B. 238
deferential conventions 90
Bradley, F. H. 235, 242, 254
definite descriptions, theory of 306,
Burge, T. 121, 338, 348
307-10
demonstratives, see indexicals
Carnap, R. 8, I 79, 245, 259 Dennett, D. 67, !36
Carruthers, P. 258 Derrida, .J. 141
Cartesianism 94, 95, !03, !09, !86, Descartes, 4
324 displays 123-5, 133
Castaneda, H. 350 Donnellan, K. 303, 308, 347, 350
causal connections 209, 2 IO Dummett, '.\1. 40, 64, 65, 68, 84, 91,
Chappell, \'. C:. I 09 203,222,226,228,348,350
charity, principle of278 ff Dworkin, R. 206
( :J10msky, ~- 27, 3 l, 33, 37, 38, 53, 68
(:ode . .\. 348 Edgington, D. 223
Col fa, .J. A 259 emoti,·e language 148-9, 167 ff
coherence 197, 235-7, 240-3, 246 entrenchment 77
and epistemology 248 ff eq11i,·alence thesis 226-32
commands !9! Ernns, G. 33 n, 35 n, 36 n, 30!, 302,
competence 25, 27, 32, 63 311, 318, 319, 320, 322, 344, 347,
composition 127, 128-40 350
366 !:'-:DEX

existential propositions 305- 7, information, identity dependent or

335-7 independent303,304,311,
expression 167-9 314-7, 327
extensionality 281, 286-93, 295 informational episodes 3 I I, 3 I 2, 313,
317,318,320,321
Feig!, H. 262 n innateness of grammar 28-30
fiction 204 intensional contexts 286, 288, 290
Field, H. 67, 246, 265, 301 intentions 72, I I 1-14
Fodor,J. 51, 55, 56, 58, 347 interpretation, radical I 9, 57, 58, 60,
Fogelin, R. 108 61-6, 134, 275, 277-81, 284,
force of utterances 122-6 289, 293
Frege, G. 189, 226, 228, 229, 232, Ishiguro, H. 67
258,286,302,313,328,331,340,
344 James, W. 56
Fricker, E. 33 n Joachim, H. 235, 236, 259
Friedman, M. 300 Jones, 0. R. 97 n
functionalism 53, 54, 55 judgement 95, 231, 234

Geach, P. 189, 196 n, 258 Kant, I. 4, 6, 83 n, I 96, 2 I 2, 222, 233,

Goodman, N. 28, 69, 74, 75-7, 107, 234,236
128, 198, 251, 255 Kaplan, D. 344
gorillas I 73 Kenny, A. 97 n
Grice, H. 110-14, 116, 117, 118, 133 Kim,J. 221
Grover, D. 258 Kirk, R. 68
knowledge, theory of 41, 234, 248-56
Haack, S. 300 Kripke, S. 81 n, 84, 85 n, 87 n, 99,
Hacker, P. 206 n 101 n, 107, 108, 223, 301, 302,
Harman, G. 67 347,350
Harrison, B. 129
Hart, H. L. 207 n Lakatos, I. 256
Hempel, C. 259 language of thought 51, 56
Hegel 201 n language, private 92- I 08, I 22
Heil,J. 67 legal verdicts 132, 205-9
Hobbes, T.41,51, 172, 175, 179 Leibniz 8
holism 63, 64, 158-63, 236 Leich, C. 221
Holtzman, S. 221 Lewis, D. 59, I 18, I 19-23, 213, 215,
humanity, principle of278-80 300,350
Hume, D. 6, 51, 105, 199, 200, 208, linguistic nominalism I 13-5
210,211,212,234,259 literary criticism I 99, 200, 204
Locke,J.4,40,41,51
idealism 217, 238, 243, 247 logicism I 52
ideas 40, 41 Lowenheim-Skolem theorem 278
idiolect 90
imagination and images 45-50, 67, machine language 52-4
234 Macintyre, A. 197
implicatures 308, 309 Mackay, A. 347
indexical expressions 297, 334, Mackie,J. 108, I 71, 180, 216 n, 221,
340-4 258,332,350
 I'.\'DEX 367

'.\Ialcolm, N. 109 projectivism 171, 180, 182, 187,

mathematical truth 23, 208 193-5,210,212,216,218
'.\fauthner, F. 8 propositional attitudes 290-1,
l\icDowell,J. 108, 142, 318 329-31, 340
:\IcGinn, C. 68, 300 Putnam, H. 28, 131, 223, 250, 260,
Mellor, H. 221 300-1, 310, 312
metaphor 171-9
metaphysics 3, 5, 39, 54, 343, 344 quantifiers 305, 306, 307, 346-7
:\fill, J. S. 5 quasi-realism 171, 180, 189, 195, 209,
modes of presentation 313-5, 328- 210, 211,216,218,219,224
33 quietism 146
.Moore, G. E. 321 n, 350 Queen, divorce of 309-10
moral truth 197-202, 217, 218, 241 Quine,\\'. V. 30, 31, 40, 57, 59, 63,
64,65,68, 160, 161,217,236,
Nagel, T. 115, 222 253-4,269,280,281,284
names, ambiguous 296-8
see also singular terms Ramsey, F. P. 229, 258
naturalism, ethical 152, 168 Raz,J. 206 n
necessary truth 23 n, 216- 7 realism 146, 147, 169, 182, 185, 196,
Neurath, 0. 243, 246, 259 203, 209,214,223,249
Neitzsche, F. 8, 172, 177, 181 realisation, variable 55
nominalism 79, 80 reconstruction, rational 58, 156, 15 7
numeral systems 12, 13, 14, 15, 20, reductions 146, 151-8, 161, 162, 179,
21,3~ 128,267-8,271 210
redundancy, theory of truth 229, 230,
obligations 197, 201 270
observation 160, 161, 253-5 reference 302-45
Ortony, A. 180 relativism 77-9, 86, 197, 199, 201,
222
Peacocke,C.97, 109, 142 representation 39, 42, 44, 49, 53, 54,
Peirce,C.S.160,241,250 56
Pentagon, definitions of truth of 225, Richards, I. A. I 74
227 Rorty, R. 223, 254, 259
Perry,J. 344 Royal Society 8, 50 n
personal identity 162-6 rules of application 69 ff
phenomenalism 152 rules of grammar 26-37
Pitcher, G. 257 Russell, B. 8, 69, 107, 157, 238, 239,
Plantinga, A. 223 240, 259, 304, 307, 309, 310, 315,
Plato 56 330 n, 344, 345, 350
Platts, M. 127, 276 n, 282 n, 300 Ryle, G. 28, 67
Polo, M. 86
Popper, K. 256 Scheffler, I. 179, 259
positivism,logical6,37, 152, 159,238 schemes,conceptual60,61,339
possible worlds 213-6 Schiffer, S. 141, 348, 350
pragmatism 207, 241, 242, 259 Scruton, R. 67
Prawitz, D. 223 Searle,]. 136, 137, 141, 170 n
private language Sellars, \V. 262 n
see language, private semantic, theoty of truth 271, 272
368 !'.\,DEX

semantics, compositional 9, 11, 18, translation 5 7, 58, 63

29,263,282-3,293-6 see also interpretation
pure and descriptive 264, 268 transparency (of truth) 227-9, 231
Semantic externalism 93, 104, 105 truth, theories of 9, 14 7, 220, 228,
sense 292, 313 ff, 328 229 fl; 261 ff
Shoemaker, S. 108 truth-conditions I 02, 146, 195, 219
singular terms 304, 313, 319, 328, 340 twin-earth arguments 312 fl; 326,
singular thoughts 316-25, 340-4 340
Sosa, E. 350
speech, direct and indirect 20, 23 unasserted Contexts 190-6, 218
Sluga, H. 8 n universals, problem of69 ff
speech acts 168, 169, 170 universal view of thoughts 313-18,
Sprat, T. 50 n 322,323,340
Stonehenge 161
Strawson, P. F. 36 n, 68, 114, 115, Van Fraassen, B. 146, 179, 223
133 verification principle 61, 97-100
subjectivity, and truth 197, 198
supervenience 182-90, 220 Wagner, R. 319
Swift,]. 50, 65 Williams, B. 223
Sydney Opera House 161 Williams, C.J. F. 258
syntax 18, 29 \\'inch, P. 109
Wittgenstein, L. 7, 8, 37, 48, 49, 69,
72, 74-88, 92, 95-103, 106, 107,
Tarski, A. 257, 261, 262, 265, 266, 159, 186, 219 n, 226, 229, 234,
269,270,272,275,297,299,300, 236,253
306,345 wooden uses of terms 71, 96 ff
Teller, P. 108 words 21, 22, 24, 43, 129, 274-6
Tennant, N. 344 Wright, C. 68, 98 n, 108, 223
Thomson, J. I 08
Tipton, I. 67 Ziff, P. 141
In this book Simon Blackburn provides a route into the central problems of modern
phi losophy of language. The text is designed not to give the student a superficial
acquaintance with well-known writers and their results , but to foster a genuine
appreciation of the problems which have dominated the area, and of the place these
problems have in a wider philosophical context. Individual chapters on rule-following,
meaning and convention, realism . theories of truth, semantics , and reference , enable the
reader to appreciate the rea l import of recent investigations , and to understand the
perennial concern of philosophers with the language we use to describe and change our
world . ·

'This valuable book should certainly be recommended to anyone studying the philosophy
of language : it is the best introduction to the subject I know.'
R. M. Sainsbury in The British Journal for the Philosophy of Science
Simon Blackbu.rn was Lecturer in Philosophy in the University of Oxford and Fellow of
Pembroke Col lege, Oxford from 1969- 1990, and is now Distinguished Professor at the
University of North Carolina, Chapel Hill. He is author of Reason and Prediction
(Cambridge University Press , 1973), and editor of Meaning, Reference and Necessity
(Cambridge University Press , 1975).
ALSO PUBLISHED BY OXFORD UNI VERS ITY PRESS
The Varieties of Reference Essays on Actions and Events
Gareth Evans Donald Davidson
Edited by John McDowell Paperback
Paperback
Inquiries into Truth and Interpretation
The Philosophy of Language Donald Davidson
Edited by J. R. Searle P,werback
(Oxford Readings in Philosophy)
Paperback ISBN 0 - 19 - 824651-X
The Game of the Name
Gregory McCulloch
Also available 'in paperback

OXFORD UNIVERSITY PRESS

9 780198 246510