Philosophical Analysis in the Twentieth Century, Volume 2: The Age of Meaning [Course Book ed.] 9781400825806

Table of contents :
CONTENTS
ACKNOWLEDGMENTS
INTRODUCTION: OVERVIEW AND HISTORICAL BACKGROUND
PART ONE: LUDWIG WITTGENSTEIN’S PHILOSOPHICAL INVESTIGATIONS
CHAPTER 1. Rejection of the Tractarian Conception of Language and Analysis
CHAPTER 2. Rule Following and the Private Language Argument
PART TWO: CLASSICS OF ORDINARY LANGUAGE PHILOSOPHY: TRUTH, GOODNESS, THE MIND, AND ANALYSIS
CHAPTER 3. Ryle’s Dilemmas
CHAPTER 4. Ryle’s Concept of Mind
CHAPTER 5. Strawson’s Performative Theory of Truth
CHAPTER 6. Hare’s Performative Theory of Goodness
PART THREE: MORE CLASSICS OF ORDINARY LANGUAGE PHILOSOPHY: THE RESPONSE TO RADICAL SKEPTICISM
CHAPTER 7. Malcolm’s Paradigm Case Argument
CHAPTER 8. Austin’s Sense and Sensibilia
PART FOUR: PAUL GRICE AND THE END OF ORDINARY LANGUAGE PHILOSOPHY
CHAPTER 9. Language Use and the Logic of Conversation
PART FIVE: THE PHILOSOPHICAL NATURALISM OF WILLARD VAN ORMAN QUINE
CHAPTER 10. The Indeterminacy of Translation
CHAPTER 11. Quine’s Radical Semantic Eliminativism
PART SIX: DONALD DAVIDSON ON TRUTH AND MEANING
CHAPTER 12. Theories of Truth as Theories of Meaning
CHAPTER 13. Truth, Interpretation, and the Alleged Unintelligibility of Alternative Conceptual Schemes
PART SEVEN: SAUL KRIPKE ON NAMING AND NECESSITY
CHAPTER 14. Names, Essence, and Possibility
CHAPTER 15. The Necessary Aposteriori
CHAPTER 16. The Contingent Apriori
CHAPTER 17. Natural Kind Terms and Theoretical Identification Statements
EPILOGUE: The Era of Specialization
Index

Citation preview

PHILOSOPHICAL ANALYSIS IN THE TWENTIETH CENTURY

This page intentionally left blank

PHILOSOPHICAL ANALYSIS in the

TWENTIETH CENTURY VOLUME 2 THE AGE OF MEANING

Scott Soames

PRINCETON UNIVERSITY PRESS PRINCETON AND OXFORD

Copyright © 2003 by Princeton University Press Published by Princeton University Press, 41 William Street, Princeton, New Jersey 08540 In the United Kingdom: Princeton University Press, 3 Market Place, Woodstock, Oxfordshire OX20 1SY All Rights Reserved ISBN: 0-691-11574-5 British Library Cataloging-in-Publication Data is available Library of Congress Cataloging-in-Publication Data Soames, Scott. Philosophical analysis in the twentieth century/Scott Soames. p. cm. Includes bibliographical references and index. Contents: v. 2. The age of meaning. ISBN: 0-691-11574-5 (v. 2: alk. paper) 1. Analysis (Philosophy) 2. Methodology—History—20th century. 3. Philosophy—History—20th century. I. Title. B808.5 S63 2003 2002042724 146'.4—dc21 This book has been composed in Galliard Printed on acid-free paper. ∞ www.pupress.princeton.edu Printed in the United States of America 10 9 8 7 6 5 4 3 2 1

this volume is dedicated to my son

BRIAN

This page intentionally left blank

CONTENTS

Acknowledgments xi Introduction to Volume 2

xiii

PART ONE: LUDWIG WITTGENSTEIN’S PHILOSOPHICAL INVESTIGATIONS 1 CHAPTER 1

Rejection of the Tractarian Conception of Language and Analysis 3 CHAPTER 2

Rule Following and the Private Language Argument 32 Suggested Further Reading

62

PART TWO: CLASSICS OF ORDINARY LANGUAGE PHILOSOPHY: TRUTH, GOODNESS, THE MIND, AND ANALYSIS 65 CHAPTER 3

Ryle’s Dilemmas

67

CHAPTER 4

Ryle’s Concept of Mind

92

CHAPTER 5

Strawson’s Performative Theory of Truth

115

CHAPTER 6

Hare’s Performative Theory of Goodness Suggested Further Reading

153

135

viii

CONTENTS

PART THREE: MORE CLASSICS OF ORDINARY LANGUAGE PHILOSOPHY: THE RESPONSE TO RADICAL SKEPTICISM 155 CHAPTER 7

Malcolm’s Paradigm Case Argument

157

CHAPTER 8

Austin’s Sense and Sensibilia 171 Suggested Further Reading

193

PART FOUR: PAUL GRICE AND THE END OF ORDINARY LANGUAGE PHILOSOPHY 195 CHAPTER 9

Language Use and the Logic of Conversation Suggested Further Reading

197

219

PART FIVE: THE PHILOSOPHICAL NATURALISM OF WILLARD VAN ORMAN QUINE 221 CHAPTER 10

The Indeterminacy of Translation

223

CHAPTER 11

Quine’s Radical Semantic Eliminativism Suggested Further Reading

259

287

PART SIX: DONALD DAVIDSON ON TRUTH AND MEANING 289 CHAPTER 12

Theories of Truth as Theories of Meaning

291

CHAPTER 13

Truth, Interpretation, and the Alleged Unintelligibility of Alternative Conceptual Schemes 312 Suggested Further Reading

331

CONTENTS

ix

PART SEVEN: SAUL KRIPKE ON NAMING AND NECESSITY 333 CHAPTER 14

Names, Essence, and Possibility

335

CHAPTER 15

The Necessary Aposteriori

372

CHAPTER 16

The Contingent Apriori

397

CHAPTER 17

Natural Kind Terms and Theoretical Identification Statements 423 Suggested Further Reading

457

EPILOGUE

The Era of Specialization Index 477

461

This page intentionally left blank

ACKNOWLEDGMENTS

L

IKE VOLUME 1, this volume grew out of a lecture course for upper-division undergraduates and beginning graduate students given several times at Princeton University—this one in 1998, 2000, and 2002. As such, it has benefited from those who attended the lectures and participated in discussions connected with them. In addition, I am very grateful to five people for reading and commenting on the manuscript of Volume 2—my Princeton colleagues, Professors Mark Greenberg and Gil Harman, Professor John Hawthorne of Rutgers University, my longtime friend and philosophical confidant Professor Ali Kazmi of the University of Calgary, and my Ph.D. student Jeff Speaks. All five read the manuscript carefully and provided me with detailed and extremely helpful criticisms. In addition, Ali and Jeff spent many hours with me discussing important philosophical issues connected with it. Without the contributions of these five philosophers, this work would have been much poorer. As with Volume 1, the people at Princeton University Press–particularly, Jodi Beder, Ian Malcolm, and Debbie Tegarden–did an outstanding job, and contributed much to the final shape of the work. Finally, I would again like to express my appreciation to Martha, who means so much to me, for all she has added both to my life and to the creation of these volumes.

This page intentionally left blank

INTRODUCTION OVERVIEW AND HISTORICAL BACKGROUND

Overview of the Period This volume continues the story of the leading developments in twentieth-century analytic philosophy begun in volume 1, which ended with the mid-century views of W. V. Quine. Taking up where that one left off, this volume covers the period starting roughly with Ludwig Wittgenstein’s Philosophical Investigations, published in 1953 but completed several years earlier, and ending about the time of Saul Kripke’s Naming and Necessity, originally given as three lectures at Princeton University in 1970. Topics covered will include the philosophy of the later Wittgenstein, the ordinary language school of Gilbert Ryle, John L. Austin, Peter Strawson, Richard M. Hare, and Norman Malcolm, the attack on the ordinary language school led by Paul Grice and the recognition of the need to distinguish meaning from use, Quine’s naturalism and skepticism about meaning, Donald Davidson’s systematic theory of truth and meaning, and Kripke’s reconceptualization of fundamental semantic and philosophical categories. The period studied in these two volumes has the distinction of being old enough to be not quite contemporary, while recent enough not to have achieved the venerable status of history. This makes for an interesting combination. On the one hand, we have achieved enough distance to be able to look back at the work done in the period and begin to form an overall picture of what was achieved and what was missed. On the other hand, since the philosophers studied in these volumes still cast long shadows over current debates, the critical overview we develop should be relevant to contemporary discussions in philosophy. This will, I think, become more apparent as we move through volume 2, and begin to encounter conceptual advances that not only ushered in a new philosophical future, but also transformed our view of our analytic past. The period discussed in volume 2 begins with the ascendancy of two leading ideas, both growing out of Wittgenstein’s Philosophical Investigations. The first was that philosophical problems are due solely to the misuse of language. Thus, the job of the philosopher is not to construct elaborate theories to solve philosophical problems, but to expose the linguistic confusions that fooled us into thinking there were genuine

xiv

INTRODUCTION TO VOLUME 2

problems to be solved in the first place. The second leading idea was that meaning itself—the key to progress in philosophy—was not to be studied from a theoretical or abstract scientific perspective. Rather than constructing general theories of meaning, philosophers were supposed to attend to subtle aspects of language use, and to show how misuse of certain words leads to philosophical perplexity and confusion. So what we have at the outset is a remarkable combination of views: all of philosophy depends on a proper understanding of meaning, but there is no systematic theory of meaning, or method of studying it, other than by informally assembling observations about aspects of the use of particular philosophically significant words in more or less ordinary situations. As one might have guessed, this combination of views proved to be unstable. There are too many factors in addition to meaning that influence when and how particular words are used in order for us to draw philosophically useful conclusions primarily from piecemeal observations about ordinary use. What is needed is some sort of systematic theory of what meaning is, and how it interacts with these other factors governing the use of language. This insight was something that gradually emerged during the 1950s and early 1960s as ordinary language philosophers wrestled with their dilemma. Two important milestones on the way were the development of the theory of speech acts by John L. Austin, and the work on conversational implicature by Paul Grice, both of which we will say something about in this volume. The end result was that at a certain point philosophers who were convinced that philosophical problems were simply linguistic problems came to recognize that they needed a systematic theory of meaning. However, it was unclear whether such a theory was possible, or, if it was, what it should look like. At the time, skepticism on the matter was fueled by Quine’s highly influential arguments in Word and Object and Ontological Relativity and Other Essays, which reject our ordinary notions of meaning and reference as scientifically hopeless, while proposing radically deflated substitutes. Meaning, as Quine conceived of it, was not the center of anything, certainly not philosophy. However, his was not the only voice. In the early 1960s an important development took place. Philosophers working in a different tradition, growing out of the development of formal logic, came up with a philosophical conception of meaning that many found irresistible. The conception was formulated by Donald Davidson, who conceived of a theory of meaning as a systematic theory of the truth conditions of the sentences of a language. To many this seemed like precisely the thing that was needed to fulfill

INTRODUCTION TO VOLUME 2

xv

the conception of philosophy as the analysis of meaning—no matter that the conception of meaning employed was a descendant of one that the later Wittgenstein and the ordinary language philosophers who followed him had earlier rejected as irrelevant. However, the story was not over. Shortly after the development of Davidson’s theory of meaning, Saul Kripke came along to explode the idea that the problems of philosophy were all problems of meaning or linguistic analysis. So we have a historical development with considerable irony. We start with the conviction that all problems of philosophy are really linguistic confusions to be resolved by a clear understanding of meaning. However, it is soon realized that in order to pursue this idea we need some theoretical understanding of meaning. This leads eventually to the widespread acceptance of a certain kind of logically and scientifically inspired theory of meaning, which—despite its many shortcomings—represents a significant advance. Then, at about this time, a powerful and persuasive position is developed that leads to the conclusion that however valuable it is to have an informative theory of meaning, it is a mistake to think that our most basic philosophical problems can be resolved by appealing to it. Such is the story that will be developed in volume 2.

Historical Background We begin the story of the period from Wittgenstein’s Philosophical Investigations to Kripke’s Naming and Necessity with a review of the historical background, covered in volume 1, that led up to it. Crucial to the development of the analytic approach was G. E. Moore’s conception of the proper starting point in philosophy. According to Moore, there is no defensible starting point in philosophy that is more privileged and beyond reasonable doubt than our most fundamental commonsense convictions—e.g., the conviction that we exist, that we are conscious beings, that we inhabit a world containing other conscious beings as well as material objects of various sorts; the conviction that not only do these things exist now, but many also existed in the past, some at times before we were born; and finally, our conviction that we have some genuine knowledge of the existence and character of many of these things. According to Moore, any attempt to ground these convictions on something more certain than they are is bound to fail; no such grounding is possible. Moreover, any claim that we can’t

xvi

INTRODUCTION TO VOLUME 2

know these commonsense propositions to be true is something that presupposes a philosophical conception of knowledge that stands more in need of justification than the commonsense propositions themselves. This point of view, forcefully expressed by Moore, recurs again and again in the analytic tradition throughout the twentieth century. Not all analytic philosophers have accepted every aspect of it. But it has remained a powerful and influential force. Although Moore thought that philosophy couldn’t contest our most basic commonsense convictions, he did believe that philosophy might be able to provide an analysis of their content—an analysis that would make clear how commonsense truths can genuinely be known. But how should such analysis proceed? Since Moore himself wasn’t sure, it was left to Bertrand Russell to provide what became the most widely accepted answer to this question. According to Russell, the task of philosophical analysis was primarily that of uncovering the hidden logical forms of sentences, which he took to be the forms of the thoughts that the sentences were used to express. Failure to identify the logical forms of sentences, and to distinguish them from the sentences’ grammatical forms, was, he believed, the source of many of the most serious errors in philosophy. Russell illustrated his point with the problem of negative existentials. A negative existential is a sentence that says that a certain thing, or a certain kind of thing, doesn’t exist—e.g., Carnivorous cows don’t exist, or The creature from the black lagoon doesn’t exist. Such sentences are grammatically of subject-predicate form. Standardly, we think that a sentence of this form is true if and only if the subject refers to something that has the property expressed by the predicate. For example, the sentence Pedro Martinez is a baseball player is true iff the referent of the subject expression—the man Pedro Martinez—has the property of being a baseball player. But if we say the same thing in the case of negative existentials, we get a paradox. Suppose some negative existential is true—e.g., the one about carnivorous cows. Then, it would seem that the subject expression—the phrase carnivorous cows—must refer to some things (carnivorous cows), and the predicate don’t exist must express a property (nonexistence) which those things have. But that, Russell thought, is impossible; if carnivorous cows are there to be referred to at all, then they must exist. Thus, it would seem that the sentence cannot be true; and, more generally, no negative existential can be true. But surely, that can’t be right. According to Russell, the solution to this problem lies in the fact that the grammatical form of negative existentials obscures their true logical

INTRODUCTION TO VOLUME 2

xvii

form. Roughly put, his view was that the logical form of the grammatically subject-predicate sentence Carnivorous cows don’t exist is given by the logical formula, For all x either x isn’t carnivorous or x isn’t a cow. The important point to notice is that this logical form does not contain a subject expression the job of which is to refer to something that is then said not to exist. Rather, Russell saw the logical form as making a claim about the property—expressed by either x is not carnivorous or x is not a cow—of not being carnivorous or not being a cow. The claim it makes is that this property is possessed by every object. The analysis of the negative existential The creature from the black lagoon doesn’t exist is similar. Roughly put, Russell takes this sentence to say that the property of being identical with an object o if and only if o is a creature from the black lagoon is a property that has no instances. He then generalizes this analysis by arguing that whenever a sentence contains a definite description—an expression of the form the so and so—its logical form will be of this complicated sort, and will not contain any single logical constituent corresponding to the grammatical unit the so and so. The end result is a conception of abstract logical form that is rather distant from surface grammatical form, and that has to be reached by a process of logical analysis. This theory of Russell’s—his theory of descriptions—became for many analytic philosophers in the first half of the century the paradigm of philosophical analysis. Russell extended the paradigm into the philosophy of mathematics, where he defended the view that all of mathematics could ultimately be reduced to pure logic. The motivation for this view was, in part, his desire to explain the certainty of mathematics and our knowledge of it. He initially thought that if mathematics could all be reduced to logic, then it would have the highest degree of certainty anything can have. The reduction was thought of as coming in two parts. First, higher mathematics was reduced to arithmetic, something that was taken to have been accomplished prior to Russell. Next, arithmetic was to be reduced to logic; this was the project to which Russell contributed. To accomplish the reduction, he formulated a set of what purported to be logical axioms, and he proposed a set of definitions of the central concepts of arithmetic—natural number, zero, and successor—in terms of what he took to be purely logical concepts. He then showed how, using his definitions, the axioms of arithmetic could be derived from his set of logical axioms. In effect, this involved viewing simple arithmetical sentences as being abbreviations of very complex logical formulas. Although this might at first seem counterintuitive, the philosophical advantages of the analysis were thought to overshadow any computational

xviii

INTRODUCTION TO VOLUME 2

complications. In the end, it is doubtful that anything did more to lay the foundations of the view that philosophy is logical and linguistic analyses than Russell’s theory of descriptions, and his reduction of formal theories of arithmetic to his system of logic. After Russell came Wittgenstein, who had briefly been a student of Russell’s. He took Russell’s conception of analysis, and his distinction between logical and grammatical form, and made them the foundations of an integrated philosophical system, which he presented in his early work, the Tractatus Logico-Philosophicus. In the Tractatus, he presents an aprioristic theory of meaning, intelligibility, and the limits of philosophy. According to the theory, all intelligible thoughts have contents and structures that are revealed by sentences in an ideal logical language, which constitutes the hidden core of our ordinary language. The theory classifies the sentences in this ideal language as falling into one or the other of two classes. The sentences in one class are contingent and capable of being known to be true, or to be false, only by doing observations and gathering evidence about the world. The sentences in the other class are either necessarily true or necessarily false. According to the theory, whenever a sentence of the ideal language is necessarily true (or necessarily false), it is capable of being shown to be such by logic alone. On this view, there can be no special philosophical sentences that give information about the world. If a sentence gives information about the world, then it is empirical and not decidable by philosophical reasoning. If a sentence is a necessary truth, then it is a tautology, and requires only rigorous logical proof. Hence, there are no meaningful philosophical sentences, no philosophical thoughts, and there is no subject matter for philosophy. According to the Tractatus, philosophical problems arise solely from the misuse and misunderstanding of language. Confronted with a philosophical problem, the proper response is to make clear precisely what the misuse or misunderstanding is. Ideally, this should be done by giving the proper philosophical analysis of the problematic sentences, preferably by showing how to express them in an ideal logical language. So says the Tractatus. Later, after Wittgenstein had rejected the Tractatus, he retained this linguistic conception of philosophy, while giving up the picture of analysis as revealing hidden logical structure. According to the Tractatus, the foundation of the hidden logical language consists of elementary or atomic sentences (propositions), which, if true, mirror the structure of the world. On this picture, the

INTRODUCTION TO VOLUME 2

xix

basis of all meaning is naming. The simplest linguistic expression is a name, the meaning of which is the object named. The simplest type of sentence, an atomic sentence, is a structured collection of names. The way the names are arranged in a sentence represents the way their bearers are portrayed as combining in the world—for example, the fact that the name a immediately precedes the symbol R, which in turn immediately precedes the symbol b, in the sentence a R b, may represent the referent of a as being located to the right of the referent of b. Atomic sentences are said to represent logically possible ways in which objects can be combined. Such combinations are called logically possible facts. A sentence (proposition) that stands for a logically possible fact is said to be true if and only if the fact really obtains, and so is actual rather than merely possible. According to the Tractatus, the truth values of all other sentences (propositions) are completely determined by the truth values of all the atomic sentences (propositions). In fact, all nonatomic sentences are claimed to be constructable from the atomic sentences by repeated applications of a simple operation that gathers together previously constructed sentences into a set, and jointly denies them. This is the picture of meaning that Wittgenstein constructs in the Tractatus, and that he later sets out to refute and replace in the Philosophical Investigations. Another position that preceded the Investigations was logical positivism. Although the positivists accepted the Tractatus conception of philosophy as linguistic analysis, they didn’t accept the Tractatus conception of meaning. In its place, they offered a replacement that linked meaning with verification. Again, sentences were separated into two classes—contingent and empirical vs. analytically true or analytically false, where an analytically true sentence was said to be true solely in virtue of meaning, independent of any possible state of the world (and similarly for analytically false sentences). An attempt was then made to provide a precise criterion for determining which non-analytic, empirical sentences were meaningful. These attempts fell into two categories. First, the positivists tried to define empirical meaningfulness in terms of strong verifiability or strong falsifiability. A sentence was said to be strongly verifiable if and only if it was logically entailed by some finite consistent set of observation statements; it was said to be strongly falsifiable if and only if its negation was logically entailed by some finite consistent set of observation statements. In effect, strongly verifiable sentences were supposed to be those the truth of which could, in principle, be completely established on the basis of sensory observations alone;

xx

INTRODUCTION TO VOLUME 2

strongly falsifiable sentences were those the falsity of which could, in principle, be established in this way. However, the idea that an empirical sentence is meaningful only if it is either strongly verifiable or strongly falsifiable quickly came to grief when it was realized that, by this criterion, great masses of ordinary commonsense claims about the world, as well as most of natural science, would wrongly be characterized as meaningless. This realization led the positivists to try a different approach. This time the idea was to define empirical meaningfulness in terms of weak verifiability, where a sentence was said to be weakly verifiable if and only if it, together with other claims, logically entailed observational predictions not entailed by the other claims alone. However, this too led swiftly to disaster, since, when the idea was made precise, it turned out that all sentences, even patently nonsensical ones, ended up being characterized as meaningful. From these failures an important lesson was drawn: typically, the verification of one sentence depends on taking other sentences for granted. So if meaning is verification, then, in general, individual sentences don’t have meanings, considered on their own. Rather it is systems of sentences—theories or entire conceptual schemes—that are the primary bearers of meaning. Sentences are meaningful only in so far as they contribute to such theories. As the first half of the twentieth century drew to a close, analytic philosophers fell into two main groups characterized by their response to this lesson. The first group was led by Willard Van Orman Quine, whose doctrine of holistic verificationism maintained that meaning is verification, and hence it is only reasonably extensive theories that have meanings on their own. In its simplest form, this view holds that the meaning of such a theory is the class of sensory observations that would support it, and hence that any two theories which would be supported by the same observations mean the same thing. This view was examined at the end of volume 1, where it was shown to lead to a number of paradoxical results, including some that paralleled problems which plagued earlier versions of verificationism. The second group of philosophers was influenced by the view that meaning is use, advanced by Wittgenstein in the Philosophical Investigations. In that work, Wittgenstein rejected views of meaning (including verificationism) that rest on what he took to be a false referential and descriptivist conception of language. In particular, he rejected any conception according to which the meaning of a word is what it stands for, and the meaning of a sentence is the potential fact or state of the world that it represents. By contrast, he thought that much of language isn’t descriptive or referential at all, and even that which is, isn’t entirely descriptive

INTRODUCTION TO VOLUME 2

xxi

or referential. He held that what gives an expression meaning is never simply that a correlation has been set up between it and something in the world; rather, an expression is meaningful only if there is an agreed-upon pattern of use of the expression to which language users conform that allows it to play a useful and intelligible role in their lives. For the Wittgenstein of the Investigations, sentences and other expressions don’t have to stand in any special justificatory relation to the world or to experience in order to be meaningful; any expression for which there are socially useful agreed-upon conditions of correct application qualifies as meaningful. What philosophers need to do, he insisted, is not to construct models of what they think meaning must be, but to look carefully at particular cases to see what the conventions governing the correct application of our words really are. It is to this conception of meaning that we will turn in chapter 1.

A Word about Notation In what follows I will use either single quotation or italics when I want to refer to particular words, expressions, or sentences—e.g., ‘good’ or good. Sometimes both will be used in a single example—e.g., ‘Knowledge is good’ is a true sentence of English iff knowledge is good. This italicized sentence refers to itself, a sentence the first constituent of which is the quote name of the English sentence that consists of the word ‘knowledge’ followed by the word ‘is’ followed by the word ‘good’. In addition to using italics for quotation, sometimes I will use them for emphasis, though normally I will use boldface for that purpose. I trust that in each case it will be clear from the context how these special notations are being used. In addition when formulating generalizations about words, expressions, or sentences, I will often use the notation of boldface italics, which is to be understood as equivalent to the technical device known as “corner quotes.” For example, when explaining how simple sentences of a language L are combined to form larger sentences, I may use an example like (1a), which has the meaning given in (1b). 1a. For any sentences A and B of the language L, A & B is a sentence of L. b. For any sentences A and B of the language L, the expression which consists of A followed by ‘&’ followed by B is a sentence of L.

xxii

INTRODUCTION TO VOLUME 2

Given (1), we know that if ‘knowledge is good’ and ‘ignorance is bad’ are sentences of L, then ‘knowledge is good & ignorance is bad’ and ‘ignorance is bad & knowledge is good’ are also sentences of L. Roughly speaking, a generalization of the sort illustrated by (2a) has the meaning given by (2b). 2a. For any (some) expression E, . . . E . . . is so and so. b. For any (some) expression E, the expression consisting of ‘. . .’, followed by E, followed by ‘. . .’, is so and so. One slightly tricky example of this is given in (3). 3a. For any name n in L, ‘n’ refers to n expresses a truth. b. For any name n in L, the expression consisting of the lefthand quote mark, followed by n, followed by the right-hand quote mark, followed by ‘refers to’, followed by n, expresses a truth. Particular instances of (3a) are given in (4). 4a. ‘Brian Soames’ refers to Brian Soames expresses a truth. b. ‘Greg Soames’ refers to Greg Soames expresses a truth. Finally, I frequently employ the expression iff as short for if and only if. Thus, (5a) is short for (5b). 5a. For all x, x is an action an agent ought to perform iff x is an action that produces a greater balance of good over bad consequences than any alternative action open to the agent. 5b. For all x, x is an action an agent ought to perform if and only if x is an action that produces a greater balance of good over bad consequences than any alternative action open to the agent.

PA R T O N E LUDWIG WITTGENSTEIN’S PHILOSOPHICAL INVESTIGATIONS

This page intentionally left blank

CHAPTER 1 REJECTION OF THE TRACTARIAN CONCEPTION OF LANGUAGE AND ANALYSIS chapter outline 1. Critique of the Tractarian Conception of Language The Augustinian picture vs. the conception of meaning as use Conceptual prerequisites of ostensive definitions Reference and analysis The meaning and reference of names Language games, family resemblances, and vagueness 2. Wittgenstein’s New Conception of Language and Linguistic Analysis Ordinary language is not to be understood on the model of logical calculi; sentences have neither hidden logical forms nor unique analyses Language use is not to be explained by speakers knowing and being guided by linguistic rules, but rather by unthinking, socially-conditioned agreement 3. Wittgenstein’s Deflationary Conception of Philosophy Roots of this conception in his identification of the philosophical with the necessary and apriori, of these with the analytic Doctrines of the Investigations as self-undermining because they lead to a conception of philosophy which they do not fit

Overview of The Philosophical Investigations There are three main topics in the Philosophical Investigations : (i) a critique of what Wittgenstein regards as the dominant referential conception of meaning, and a proposal to replace it with a conception in which to use language meaningfully is to master a certain kind of social practice; (ii) a critique of the previously dominant conception of philosophical analysis, and the substitution of a new conception of analysis to play the central role in philosophy; and (iii) the development of a new philosophical psychology in which what appear on the surface to be sentences that

4

WITTGENSTEIN ON LANGUAGE

report private sensations and other internal mental events or states are viewed as having meanings which license their assertion on the basis of public criteria having to do with behavior and external circumstances. The book’s center of gravity is the discussion of what it is to follow a (linguistic) rule, and the lessons drawn from it about (i), (ii), and (iii). However, Wittgenstein does not start with this. Instead, he begins with preliminary critiques of his earlier, Tractarian conceptions of language and analysis. He then uses the discussion of rule following to strengthen his critiques, to illuminate his new conceptions of meaning and analysis, and to illustrate their consequences by applying them to psychological sentences. We will follow him in this. In this chapter we will deal with (i) and (ii); in the next chapter we will be concerned with (iii).

The Critique of Tractarian Descriptivism The Augustinian Picture vs. the Conception of Meaning as Use

We begin with Wittgenstein’s critique of the central tenets of his earlier referentialism about language. The view under attack holds that the meaning of an expression is what it names or stands for, and the meaning of a sentence is a possible fact the actual existence of which would make the sentence true. Natural corollaries of the view stipulate that learning a language, and being a competent user of its expressions, is the result of recognizing correlations between words and the objects they stand for, and that being justified in accepting a contingent, empirical sentence S involves having reason to believe that the languageindependent possible state of affairs which constitutes the truth conditions of S actually obtains. Wittgenstein introduces this picture of language in section 1 of the Investigations with a quote from Augustine: When they (my elders) named some object, and accordingly moved towards something, I saw this and I grasped that the thing was called by the sound they uttered when they meant to point it out. Their intention was shown by their bodily movements, as it were the natural language of all peoples: the expression of the face, the play of the eyes, the movement of other parts of the body, and the tone of voice which expresses our state of mind in seeking, having, rejecting, or avoiding something. Thus, as I heard words repeatedly used in their proper places in various sentences, I gradually learnt to

THE TRACTARIAN CONCEPTION OF LANGUAGE

5

understand what objects they signified; and after I had trained my mouth to form these signs, I used them to express my own desires.1 Wittgenstein’s summary of the view is as follows: These words, it seems to me, give us a particular picture of the essence of human language. It is this: the individual words in language name objects—sentences are combinations of such names.— In this picture of language we find the roots of the following idea: Every word has a meaning. This meaning is correlated with the word. It is the object for which the word stands. Having presented this picture, Wittgenstein immediately challenges it with his example of the five red apples. He says: Now think of the following use of language: I send someone shopping. I give him a slip marked “five red apples”. He takes the slip to the shopkeeper, who opens the drawer marked “apples”; then he looks up the word “red” in a table and finds a color sample opposite it; then he says the series of cardinal numbers—I assume that he knows them by heart—up to the word “five” and for each number he takes an apple of the same color as the sample out of the drawer.—It is in this and similar ways that one operates with words.—“But how does he know where and how he is to look up the word ‘red’ and what he is to do with the word ‘five’?”—Well, I assume that he acts as I have described. Explanations come to an end somewhere.—But what is the meaning of the word “five”?— No such thing was in question here, only how the word “five” is used.2 In explaining what it is to understand the expression five red apples, Wittgenstein here recounts what one would do with it. Mastery of the numeral ‘five’ is not explained by finding some unique object for it to name; rather, mastery is a matter of engaging in certain sorts of routine that govern its application. The shopkeeper recites a series of sounds—one, two, three, four, five—and correlates them one-to-one with a series of acts—each act involving taking an apple out of the drawer. Mastery of the numerals is mastery of routines like this. This is Wittgenstein’s first example of the thesis that meaning is use. 1 Wittgenstein, Philosophical Investigations, translated by G.E.M. Anscombe, 3d edition (New York: Macmillan, 1958). 2 Philosophical Investigations, section 1.

6

WITTGENSTEIN ON LANGUAGE

In sections 2 and 6–21, he continues the theme of meaning as use with the example of the primitive language of a builder and his assistant. Here, the sentences—Pillar, Slab, Block, and Beam—are not used as descriptions; rather, they are used to give orders. Wittgenstein goes on to emphasize many different uses of sentences as a way of undermining the tendency to take describing, stating a fact, or asserting something as primary. The value of undermining this picture is that it allows him to concentrate on using language as participating in many kinds of social activity. In the elementary language game he describes, there is no referring to an object and predicating a property of it. There is only the coordination of words and actions. The moves in the language game are meaningful in so far as they contribute to the coordination of the actions of the builder and his assistant. Conceptual Prerequisites of Ostensive Definitions

One case study for the thesis that meaning is use is provided by names. At section 26 Wittgenstein starts talking about proper names, common nouns like cat, red, and round, and ostensive definitions. He has already pointed out that many parts of language are not names. Now he emphasizes that even when we introduce a name with an ostensive definition, the definition requires background assumptions in order to work. For example, suppose I point to one of the buttons on my shirt and say This is red, in an attempt to convey to you the meaning of the word red. In order for you to understand my intention, you need to know what I am pointing at—at myself, at my chest, at my shirt, at the button?—and you also need to know which aspect of the thing I am pointing at is the one I am characterizing—its size, its shape, its color, its price? This is where the needed background assumptions come in; it is only by relying on them that I can get my meaning across. It is certainly true that these background beliefs are necessary in order for an ostensive definition to work; however, Wittgenstein appears to go beyond this, and suggest something stronger—namely, that ostensive definition (in the sense of the Augustinian picture discussed in section 1) makes sense only if one has already mastered a significant part of language in advance. For example, if you are not sure how to interpret my ostensive definition of red, I might clarify things by saying The color of the buttons on my shirt is red. But that presupposes that you have already understood the words color, buttons, and shirt.

THE TRACTARIAN CONCEPTION OF LANGUAGE

7

Wittgenstein realizes that there are cases in which I could give my ostensive definition, This is red, and you would get the idea without my having to use any further words, because you would correctly guess that I was pointing to one of my buttons and talking about its color. In light of this, one wonders whether further language is always needed to understand an ostensive definition after all. Wittgenstein addresses this point in section 32. Someone coming into a strange country will sometimes learn the language of the inhabitants from ostensive definitions that they give him; and he will often have to guess the meaning of these definitions; and will guess sometimes right, sometimes wrong. And now, I think, we can say: Augustine describes the learning of human language as if the child came into a strange country and did not understand the language of the country; that is, as if it already had a language, only not this one. Or again: as if the child could already think, only not yet speak. And “think” would here mean something like “talk to itself ”. There is something quite striking about this passage. Wittgenstein seems to equate (i) a child’s ability to have thoughts about what the words he has been exposed to stand for with (ii) the child’s already having a language in which to express those thoughts. Since it is presumably absurd to suppose that the child already has such a language, Wittgenstein seems to be casting doubt on the idea that the child can think at all prior to having learned his first language. To this one is tempted to reply: “Of course the child can think before he has learned to speak. Have you ever been around young children? It is obvious that they have thoughts before they can speak. Moreover, they would be in pretty bad shape if they couldn’t have such thoughts. How could they learn anything—let alone language—if they couldn’t do at least some thinking first? ” However, Wittgenstein would not accept this reply. One of the themes of the Investigations is that terms like think and understand, which appear to refer to private mental events or processes, should really be understood as standing for complex behavioral and social dispositions, standardly including dispositions to use language. Of course, if one takes thinking to be something that essentially involves dispositions to use language in certain ways, then the idea of a child being able to think before he has language will seem to be a non-starter. Now, I am not sure that, in the end, Wittgenstein really wants to go so

8

WITTGENSTEIN ON LANGUAGE

far as to claim that all thinking requires language use or linguistic dispositions, and hence to deny that non-linguistic creatures can have any thoughts at all.3 However, he does seem to presuppose that they couldn’t have the kinds of thoughts needed to understand ostensive definitions. Why, precisely, we should take this to be so is something that, regrettably, he is not terribly clear about (at least at this stage of the Investigations). Still, the import of his view is clear enough: since the thoughts required to interpret an ostensive definition cannot be had prior to mastering a significant amount of language, ostensive definition cannot be the foundation of all language learning and language use. And if it can’t serve as the foundation of language learning and use, Wittgenstein seems to suggest, it is hard to see how the referentialist conception of language, with its emphasis on the importance of naming, can get off the ground. It is not that he is suggesting that we never succeed in naming anything ostensively, or in describing something once we have named it. Of course, we do. Rather, he is saying that naming and describing cannot constitute the essence of meaning, since in order to be able to name or describe anything, we must have a rich system of meaning already in place. To be sure, he hasn’t established this, or even really argued for it yet. He has just raised the issue, and started to paint an alternative picture. Reference and Analysis

In sections 37 and 38, he attacks a different part of the descriptivist, or referential, conception of meaning. He indicates that it is not clear what we have in mind when we talk about the naming or reference relation. What is the relation between name and thing named?—Well, what is it? Look at language-game (2) [the builder and his assistant] or at another one: there you can see the sort of thing this relation consists in. This relation may also consist, among many other things, in the fact that hearing the name calls before our mind the picture of what is named; and it also consists, among other things, in the

3 See Philosophical Investigations, section 330, for what seems to be a suggestion that actions and behavioral dispositions alone, without linguistic actions (and perhaps even linguistic dispositions), may sometimes count as sufficient for thought. See also (i) of Part II of the Investigations, where Wittgenstein recognizes that certain beliefs are attributable non-linguistic animals, but suggests that the mental states and contents available to such animals are extremely limited. Thanks to Jeff Speaks for a useful discussion of this point.

THE TRACTARIAN CONCEPTION OF LANGUAGE

9

name’s being written on the thing named or being pronounced when the thing is pointed at.4 But what, for example, is the word “this” the name of in language-game (8) [Here Wittgenstein refers to an expansion of the primitive language game of the builder and his assistant] or the word “that” in the ostensive definition “that is called . . .”?—If you do not want to produce confusion you will do best not to call these words names at all.—Yet, strange to say, the word “this” has been called the only genuine name; so that anything else we call a name was one only in an inexact, approximate sense. This queer conception springs from a tendency to sublime the logic of our language—as one might put it. The proper answer to it is: we call very different things “names”; the word “name” is used to characterize many different kinds of use of a word, related to one another in many different ways:—but the kind of use that “this” has is not among them. . . . This is connected with the conception of naming as, so to speak, an occult process. Naming appears as a queer connection of a word with an object.—And you really get such a queer connection when the philosopher tries to bring out the relation between name and thing by staring at an object in front of him and repeating a name or even the word “this” innumerable times. For philosophical problems arise when language goes on holiday. And here we may indeed fancy naming to be some remarkable act of mind, as it were a baptism of an object. And we can also say the word “this” to the object, as it were address the object as “this”—a queer use of this word, which doubtless only occurs in doing philosophy.5 The object of Wittgenstein’s ridicule here is (in part) our inclination to think that there is just one relation which all names bear to their referents. Wittgenstein thinks that this idea is obviously incorrect. We can say, if we like, that the numeral ‘5’ names or refers to the number five, but only if we do not take the naming or reference relation that we appeal to in this case to be the same as the relation between someone’s name and the person named, for example. In section 38, he indicates that he thinks that the philosophical problem of discovering the nature of the naming or reference relation falsely presupposes that there is just one such relation. He dismisses this idea with the general remark 4 5

Philosophical Investigations, section 37. Philosophical Investigations, section 38.

10

WITTGENSTEIN ON LANGUAGE

about philosophy, “For philosophical problems arise when language goes on holiday.” It is less clear what relevance his dismissal of the idea that there is a unique reference relation has for his overall attack on the referentialist conception of meaning. Perhaps the idea is this: the referentialist is inclined to think that the relation of reference is basic, that by taking it as our starting point we can give a general account of meaning and understanding, and by employing it in the philosophical analysis of language we can achieve a level of clarity and exactness not obtainable in any other way. By contrast, Wittgenstein thinks that talk about reference is something that stands in as much need of analysis and clarification as anything else. In certain cases we mean one thing when we talk about the reference of a word, and in other cases we mean something else. Since talk about reference is subject to the same sorts of ambiguity and unclarity as the ordinary talk it is supposed to be used to analyze, he seems to suggest, claims about reference can’t provide the basis of, or starting point for, all philosophical analyses of meaning and understanding. This point, though arresting, is telling only if supplemented with something further. Consider the view that to understand a word is to know what it refers to (which, we may suppose, includes, in the case of a predicate, what it applies to). It is one thing to be told that by reference we mean different things in different cases; it is quite another to be told that what we mean by reference in at least some of these cases is such that understanding a word cannot be explained as knowing what it refers to. Maybe the thought is that for any specific disambiguation of ‘refers’, the general claim that to understand any word is to know what it refers to in that specific sense must be false, since, at a minimum, for different words, different reference relations will be required. However, even if that were to turn out to be so, it would still leave open the possibility that understanding a word is always a matter of what it refers to, in the relevant sense of ‘refers’. To rule even this out, one might argue that in at least some cases—e.g., the terms ‘5’, ‘Uranus’, and ‘neutrino’—we don’t understand these terms because we know that ‘5’ refers to an object iff it is the number 5, that ‘Uranus’ refers to an object iff it is Uranus, and that ‘neutrino’ applies to an object iff it is a neutrino; rather, we know these elementary truths about reference because we understand and accept the sentences that express them, which in turn presupposes that we already understand the expressions ‘5’, ‘Uranus’, and ‘neutrino’ themselves. If such an argument could successfully be made, it would show that understanding an expression is at least sometimes conceptually prior

THE TRACTARIAN CONCEPTION OF LANGUAGE

11

to knowing what it refers to, in which case the latter cannot be the explanation of the former.6 Although Wittgenstein seems to have believed this, he did not explicitly argue in this way, and his argument at this stage of the Investigations is left inexplicit and incomplete. The Meaning and Reference of Names

In sections 39 and 40, he continues his assault on the referential conception of meaning with an argument that the meaning of a name can’t be its bearer. Here is section 39. But why does it occur to one to want to make precisely this word [the word “this”] into a name, when it evidently is not a name?— That is just the reason. For one is tempted to make an objection against what is ordinarily called a name. It can be put like this: a name ought really to signify a simple. And for this one might perhaps give the following reasons: The word “Excalibur”, say, is a proper name in the ordinary sense. The sword Excalibur consists of parts combined in a particular way. If they are combined differently Excalibur does not exist. But it is clear that the sentence “Excalibur has a sharp blade” makes sense whether Excalibur is still whole or is broken up. But if “Excalibur” is the name of an object, this object no longer exists when Excalibur is broken in pieces; and as no object would then correspond to the name it would have no meaning. But then the sentence “Excalibur has a sharp blade” would contain a word that had no meaning, and hence the sentence would be nonsense. But it does make sense; so there must always be something corresponding to the words of which it consists. So the word “Excalibur” must disappear when the sense is analyzed and its place be taken by words which name simples. It will be reasonable to call these words the real names. This is essentially one of Russell’s old arguments that ordinary names are not real, or logically proper, names, since ordinary names can have meanings even when they lack referents, whereas the meaning of a real (logically proper) name is nothing other than its referent. Unlike ordinary An argument of this sort is developed in the section “Semantic Competence and ‘The Augustinian Picture’ ” (pp. 587–91) of my “Semantics and Semantic Competence,” Philosophical Perspectives 3 (Philosophy of Mind and Action Theory), 1989, 575–96. It is discussed further in my “Facts, Truth Conditions, and the Skeptical Solution to the Rule-Following Paradox,” Philosophical Perspectives 12 (Language, Mind, and Ontology), 1998, 313–48.

6

12

WITTGENSTEIN ON LANGUAGE

names, Russell thought, at least some uses of the demonstrative ‘this’ conformed rather well to this criterion. If one says, pointing at nothing in particular, This is a fine red one, then it is plausible to suppose that since one hasn’t identified any object to which the predicate is supposed to be applied, one’s sentence lacks content, and one hasn’t meaningfully asserted anything. For reasons like this, Russell was inclined to say that the word ‘this’ functions as a genuine name, whereas ordinary proper names do not. Although Wittgenstein at one time accepted essentially this argument, by the time of the Investigations, he no longer did. Here is section 40: Let us first discuss this point of the argument: that a word has no meaning if nothing corresponds to it.—It is important to note that the word “meaning” is being used illicitly if it is used to signify the thing that ‘corresponds’ to the word. That is to confound the meaning of a name with the bearer of the name. When Mr. N. N. dies one says that the bearer of the name dies, not that the meaning dies. And it would be nonsensical to say that, for if the name ceased to have meaning it would make no sense to say “Mr. N. N. is dead”. In this passage, Wittgenstein turns the Russellian argument around and uses it to refute the contention that the referent of a name is its meaning. In addition, we get an anticipation of his rejection of “old-style” philosophical analysis (which assigned abstract logical forms to sentences). Russell and the early Wittgenstein would have said that the fact that a sentence containing an ordinary name would remain meaningful even if its referent ceased to exist, shows that what looks like a name really doesn’t function logically as a name; hence the whole sentence must be given a complex analysis. For Russell and the early Wittgenstein, genuine names must have referents the existence of which is somehow guaranteed. For Russell this meant that names could refer only to things that one could not be mistaken about, like private sense data, or the abstract objects with which one was directly acquainted. For the early Wittgenstein it meant that names must refer to indestructible metaphysical simples that could not fail to exist. Either way, sentences containing ordinary names had to be analyzed in terms of complicated logical forms, which, if they contained genuine, logically proper names at all, were construed as talking about a certain philosophically revealed subject matter. The Wittgenstein of the Investigations rejects the idea that sentences have analyses of this sort. In the sections following 40, he demolishes his old Tractarian view that there are indestructible, necessarily existing simples in the world

THE TRACTARIAN CONCEPTION OF LANGUAGE

13

that cannot be described but can only be named by the corresponding linguistic simples—i.e., by expressions satisfying Russell’s idealized conception of genuine names. The significance of this demolition is rather limited, however, since the views about metaphysical simples in the Tractatus were always seen to be implausible, and never had much of a following. Undermining them was not a great advance. A more interesting question is whether Wittgenstein’s argument in these sections really shows that the meaning of an ordinary name can’t be its referent. I am not convinced that it does. Certainly the sentence Socrates is dead continues to be meaningful even though the man Socrates no longer exists, and the name Socrates does not refer to any existing thing. However, is it so clear that Socrates fails to refer to anything at all? If it did fail to refer, then it would be hard to see how the sentence Socrates is dead could be true—since the subject expression wouldn’t refer to anything that had the property expressed by the predicate. So perhaps names can continue to refer to things that once existed but no longer do. If the things they refer to are their meanings, then names may remain meaningful, even when the things that are their meanings cease to exist. If this seems surprising, consider what the alternative might be. Suppose one thought, with Russell, that the meaning of a name like Socrates was given by a description associated with it by speakers. Suppose, for example, one took the name to be short for the description the philosopher who was convicted of corrupting the youth of Athens, and then applied Russell’s theory of descriptions. The result would be problematic in at least two respects. First, if 1. Socrates is dead meant the same as 2. There exists an x such that (i) x is identical with an individual iff that individual was a philosopher convicted of corrupting the youth of Athens, and (ii) x is dead, then (1) would entail 3. There exists an x such that x is identical with an individual iff that individual was a philosopher convicted of corrupting the youth of Athens, which, according to the analysis, means the same as 4. Socrates exists.

14

WITTGENSTEIN ON LANGUAGE

But it is absurd to suppose that Socrates is dead entails Socrates exists. Second, the descriptive counterpart (5) of (1) seems to be true, despite the fact that its most natural Russellian analysis, (2), is false. 5. The philosopher who was convicted of corrupting the youth of Athens is dead. Thus, the orthodox Russellian analysis is in trouble. Nor will it do to claim that so and so is dead has the logical form it is not the case that so and so is alive; it is not the case that the largest prime number is alive, but that doesn’t mean that that the largest prime number is dead. There is, however, a different, more complicated analysis of (5) that is more congenial to the Russellian.7 In giving the analysis, one begins by paraphrasing (5) as (7). 7. The philosopher who was convicted of corrupting the youth of Athens was once alive but no longer is. Next, one assigns (7) the logical form (8). 8. In the past it was the case that [there existed an x such that ( (i) x was identical with an individual iff that individual was a philosopher convicted of corrupting the youth of Athens, and (ii) x was alive but it is now the case that x is not alive)]. This sentence is true at the present time tnow iff for some earlier time tthen (i) there was (at tthen) an x such that x was identical with an individual iff that individual was a philosopher convicted of corrupting the youth of Athens, and (ii) x was alive (at tthen), but at tnow x is not alive. If this analysis is legitimate, then the Russellian can, in principle, account for (5). However, in order for the analysis to work, it must now be the case that the formula x is not alive is genuinely both meaningful and true relative to an assignment to the variable ‘x’ of an individual who once existed but no longer does. This is significant since, whatever may be true of other parts of language, the meanings of variables (relative to assignments) are simply the objects they are assigned as referents. Thus, if variables can be meaningful by virtue of referring to nolonger-existent objects, and if formulas containing them can be both 7 The analysis is derived from Russell’s suggestion, “To say that a person is dead is complicated. It is two statements rolled into one: ‘Socrates was alive’ and ‘Socrates is not alive’ ” (p. 78, of the Philosophy of Logical Atomism, La Salle, IL: Open Court; 1985; originally published in 1918). Thanks to Ali Kazmi for a discussion of this point.

THE TRACTARIAN CONCEPTION OF LANGUAGE

15

true and meaningful when the variables are taken to so refer, then the fact that Socrates no longer exists is no argument (a) that the name Socrates doesn’t refer to him, (b) that the referent of Socrates is not the meaning of Socrates, or (c) that if the meaning of Socrates is its referent, then sentences containing it cannot be meaningful. In light of this, the best thing to say about Wittgenstein’s example in section 40 seems to be that the name Mr. N. N. refers to someone who once existed, but no longer does. But then, the name does refer, even though its referent no longer exists. If this is right, then the name still has a referent, and Wittgenstein’s remarks do not show that its referent can’t be its meaning. The proponent of the view that Wittgenstein opposes may simply maintain that the name does mean something, even though the thing it means is something that no longer exists.8 Of course, the proponent of this view has not conclusively established his position either— for, as Wittgenstein might well insist, the view that the meaning of a name is its referent has other counterintuitive consequences. For example, if the meaning of the name Saul Kripke is the man Saul Kripke, then someone who has spoken to Saul Kripke should be able to correctly report, I have spoken with the meaning of the name ‘Saul Kripke’. But that sounds extremely odd—in a way that I have spoken to the bearer (or referent) of the name ‘Saul Kripke’ does not. Presumably, the two sides in this dispute would view the significance of this observation differently. Whereas Wittgenstein would see it as supporting his contention that the referent of a name is not its meaning, the defender of the view that they should be identified would take it to show only that we are not ordinarily accustomed to thinking in these terms (correct though this view may be).9

Language Games, Family Resemblances, and Vagueness

We will return to the issue of whether the meaning of a name is its referent when we discuss Wittgenstein’s remarks in section 79. Before doing that however, we take up his comparison of different uses of language to 8 See the appendix to chapter 3 of my Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity (New York: Oxford University Press, 2002) for a sketch of this view, which is derived from Nathan Salmon. For more details, see Salmon, “Existence,” in J. Tomberlin, ed., Philosophical Perspectives, vol. 1, Metaphysics (Atascadero, CA: Ridgeview, 1987), 49–108; and Salmon, “Nonexistence,” Noûs 32 (1998): 277–319. 9 See the section “Meaning, Semantic Content, and Speakers’ Intuitions,” of chapter 3 of Beyond Rigidity (pp. 67–72), for an explanation of how and why facts about meaning may sometimes diverge from the judgments of meaning given by ordinary competent speakers.

16

WITTGENSTEIN ON LANGUAGE

different games. He develops the comparison in the sections numbered in the middle sixties, where his main point is that just as there is no one thing unique to games that is common to them all, so there is no one thing unique to different uses of language that is common to them all. Here is the last paragraph of section 65. And this is true.—Instead of producing something common to all that we call language, I am saying that these phenomena have no one thing in common which makes us use the same word for all,—but that they are related to one another in many different ways. And it is because of this relationship, or these relationships, that we call them all “language”. In sections 66 and 67, Wittgenstein amplifies this idea by considering the concept of a game. Consider for example the proceedings that we call “games”. I mean board-games, card-games, ball-games, Olympic games, and so on. What is common to them all?—Don’t say: “There must be something common, or they would not be called ‘games’”—but look and see whether there is anything common to all.—For if you look at them you will not see something that is common to all, but similarities, relationships, and a whole series of them at that.10 I can think of no better expression to characterize these similarities than “family resemblances”; for the various resemblances between members of a family: build, features, color of eyes, gait, temperament, etc., etc., overlap and criss-cross in the same way.— And I shall say: ‘games’ form a family. And for instance the kinds of number form a family in the same way. Why do we call something a “number”? Well, perhaps because it has a—direct—relationship with several things that have hitherto been called number; and this can be said to give it an indirect relationship to other things we call the same name. And we extend our concept of number as in spinning a thread we twist fibre on fibre. And the strength of the thread does not reside in the fact that some one fibre runs through its whole length, but in the overlapping of many fibres. 10

Philosophical Investigations, section 66.

THE TRACTARIAN CONCEPTION OF LANGUAGE

17

But if someone wished to say: “There is something common to all these constructions—namely the disjunction of all their common properties”—I should reply: Now you are only playing with words. One might as well say: “Something runs through the whole thread—namely the continuous overlapping of those fibres”.11 This talk about games and families is brought up in the context of illustrating the contention that there is no essence of language or language use. There is no one thing that is common to all uses of language, one pattern that all must fit, but rather a system of overlapping similarities— just as in the case of games. To understand a particular use of language one must examine it on its own terms, not assume in advance that it must fit some single preconceived conception of meaning. At 69 and 70 we get the view that the concept of a game, like most other concepts, is vague. But that doesn’t make it defective or in need of any precisification by philosophical analysis. Similarly, the concept of a language is vague, but that doesn’t mean that there is anything wrong with the concept; nor does it mean that the way to understand ordinary language is to describe the relation it bears to some fully precise logical ideal. Sections 71 through 75 are concerned with how one masters an ordinary vague word or concept. It is not by grasping the thing common to all its instances; there is no such thing. Rather, to master a vague word is to be able to apply it in agreement with others. This idea of applying a word in agreement with others is central to Wittgenstein’s general approach to meaning and understanding, and is something he returns to again and again. The Use and Meaning of Names

In section 79, Wittgenstein applies his idea that often there is no one thing common to different uses of the same expression to proper names. Here is the first paragraph of that section. Consider this example. If one says “Moses did not exist”, this may mean various things. It may mean: the Israelites did not have a single leader when they withdrew from Egypt—or: their leader was not called Moses—or: there cannot have been anyone who accomplished all that the Bible relates of Moses—or: etc. etc.— We may say, following Russell: the name “Moses” can be defined 11

Ibid., section 67.

18

WITTGENSTEIN ON LANGUAGE

by means of various descriptions. For example, as “the man who led the Israelites through the wilderness”, “the man who lived at that time and place and was then called ‘Moses’”, “the man who as a child was taken out of the Nile by Pharaoh’s daughter” and so on. And according as we assume one definition or another the proposition “Moses did not exist” acquires a different sense, and so does every other proposition about Moses.—And if we are told “N did not exist”, we do ask: “What do you mean? Do you want to say . . . . . . or . . . . . . etc.?” Let us construct a gloss for this passage. Take an arbitrary proper name N. In discussing this example, we will abstract away from the fact that different individuals can share the same name, and when we consider different uses of N we will always have in mind cases in which N is used to refer to the same individual. Even when this idealization is presupposed, it is clear that different speakers may associate different descriptive information with N, and a single speaker may associate different descriptions with it on different occasions of use. Because of this there is no one thing that speakers mean by N is F, or by N doesn’t exist, on different occasions. The information a hearer gathers from such an utterance depends on the descriptive information the hearer associates with the name. The information a speaker wishes to convey by an utterance depends on what descriptive information he thinks that he and his audience jointly associate with the name. All of this seems clear and at least roughly correct—to which I would add the following: anyone who is familiar with a name, and who is competent to use it, can be expected to associate some descriptive content with it. Nevertheless, there is no single description, or disjunction of descriptions, that anyone who is competent to use the name must associate with it in order to be competent. At least, there is no such description or disjunction of descriptions that picks anyone out uniquely. We may now consider what Wittgenstein has to say in the rest of the section. But when I make a statement about Moses,—am I always ready to substitute some one of these descriptions for “Moses”? I shall perhaps say: By “Moses” I understand the man who did what the Bible relates of Moses, or at any rate a good deal of it. But how much? Have I decided how much must be proved false for me to give up my proposition as false? Has the name “Moses” got a fixed unequivocal use for me in all possible cases?—Is it not the

THE TRACTARIAN CONCEPTION OF LANGUAGE

19

case that I have, so to speak, a whole series of props in readiness, and am ready to lean on one if another should be taken from under me and vice versa?—Consider another case. When I say “N is dead”, then something like the following may hold for the meaning of the name “N”: I believe that a human being has lived, whom I (i) have seen in such-and-such places, who (ii) looked like this (pictures), and (iii) has done such-and-such things, and (iv) bore the name “N” in social life.—Asked what I understand by “N”, I should enumerate all or some of these points, and different ones on different occasions. So my definition of “N” would perhaps be “the man of whom all this is true”.—But if some point now proves false?—Shall I be prepared to declare the proposition “N is dead” false—even if it is only something which strikes me as incidental that has turned out false? But where are the bounds of the incidental?—If I had given a definition of the name in such a case, I should now be ready to alter it. And this can be expressed like this: I use the name “N” without a fixed meaning. (But that detracts as little from its usefulness, as it detracts from that of a table that stands on four legs instead of three and so sometimes wobbles.) Should it be said that I am using a word whose meaning I don’t know, and so am talking nonsense?—Say what you choose, so long as it does not prevent you from seeing the facts. (And when you see them there is a good deal that you will not say.) (The fluctuation of scientific definitions: what to-day counts as an observed concomitant of a phenomenon will to-morrow be used to define it.) One of the things that Wittgenstein seems to be saying here is that on any particular occasion on which I make a statement using a particular name, what I mean and attempt to convey by my utterance is to a certain extent vague, because the descriptive information I associate with the name is to a substantial extent vague, and open-ended (as is the information I associate with ‘game’). This fits his overall picture of ordinary language use as perfectly meaningful, and in order as it stands, even though it may appear inexact when compared to the kinds of ideal calculi to which previous philosophers, himself included, sometimes tried to assimilate it. This observation about our use of names is unobjectionable, and even platitudinous, as are most of his other observations in the section. However, I doubt that they should be read as

20

WITTGENSTEIN ON LANGUAGE

constituting any sort of theory of the meaning of proper names, as they sometimes are. A better interpretation is to see them as laying down some obvious facts that any adequate theory must accommodate. As for the question of what such a theory would look like, I believe that Wittgenstein’s observations could be accommodated and extended in at least three different ways. Way 1: On this view, names have different meanings for different speakers at different times. The meaning of a name for a speaker at a time is something like an open-ended set of descriptions that the speaker associates with the name at the time (and, perhaps, takes to be common ground between himself and his audience). Roughly speaking, the meaning of n for a certain speaker is given by a vague description—the unique individual x who satisfies a sufficient number of the following descriptive claims: x is F, x is G, and so on. This is a theory that is often associated with Wittgenstein, though it is rather more specific than anything he said.12 In addition, there are different versions of the theory, depending on whether the vague description associated with a name is built from specific descriptions commonly associated with the name throughout the linguistic community, or whether it varies from speaker to speaker, being constructed in each case from particular descriptions that the speaker in question associates with the name. Either way, there are difficulties to be surmounted. On the more social version of the theory— which, given Wittgenstein’s general conception of language as a social institution, one might expect him to favor—one confronts the problem that for some names the associated descriptive information may vary so much from speaker to speaker that there is little commonly shared information—in which case the vague but allegedly meaninggiving description may fail to pick out any unique referent at all. On the individualized version of the theory, names are expressions the meanings of which change from speaker to speaker, and time to time; moreover, it may well be that the meaning of a name for one speaker is never exactly the same as its meaning for another speaker. This being the case, it is a problem for the theory to explain how I can report what you believe when you assent to a sentence containing a name n. In most cases I won’t know exactly which descriptions you associate 12 For a defense of a view of roughly this sort, see John Searle, “Proper Names,” Mind 67 (1958): 166–73.

THE TRACTARIAN CONCEPTION OF LANGUAGE

21

with n, and it seems likely that whatever those descriptions turn out to be won’t completely match the ones I associate with n. But surely when I say you believe that n is F, after hearing your utterance of n is F, what I am saying is that you believe the content of n is F as I use it in my belief report—and that content will depend on the descriptions I associate with the name.13 Way 2: This way of building on Wittgenstein’s remarks differs from the first in distinguishing what a speaker means by an expression at a time from what the expression means, considered as part of the common language—say, as part of English. On this view, what a speaker means (and asserts) when he says n is F on some particular occasion may include the claim that the individual who satisfies most of a certain set of descriptions (associated with n by the speaker) has the property expressed by F. However, this is not what the sentence means as part of our common language. Roughly speaking, what a sentence S means in English consists of information that any competent speaker who is familiar with it and understands it would both associate with S and use it to assert and convey in different circumstances. In other words, the meaning of S is information that must be grasped by any competent user of S, and that would be part of the information asserted and conveyed by any normal use of it. On this view, the meaning of a name in the language often won’t be any uniquely identifying description or set of such descriptions at all. Way 3: This way of extending Wittgenstein’s remarks is just like the second, with the addition of a positive claim about what the meaning of a name in the language is—to wit, its referent. Although Wittgenstein would not himself have welcomed this claim, I think it can be made compatible with his main observations in section 79. The justification for taking the meaning of a name in the public language to be its referent is that if we ask What information is common to the information asserted and conveyed by competent speakers by normal utterances of the sentence n is F in different contexts?, the answer seems to be that it is information that a certain individual—the one the name in fact refers to—has a certain property—the one expressed by F. To see this, imagine that some competent speaker, Ralph, assertively utters the sentence in some arbitrary context. No matter what descriptive information Extensive arguments against views of this sort will be considered in chapter 14, when we discuss Saul Kripke’s Naming and Necessity.

13

22

WITTGENSTEIN ON LANGUAGE

Ralph may associate with the name, normally I can correctly report his assertion by saying Ralph asserted that n is F. Further, if there is an individual that n refers to, it will follow that there is a certain individual such that Ralph asserted that he/she is F expresses a truth. This just says that Ralph asserted some singular proposition that attributes the property expressed by F to a certain individual, where the proposition in question doesn’t include any further description of that individual. Ralph’s assertive utterance may result in the assertion of other propositions as well, but the bare, singular proposition that predicates the property expressed by F of the referent of n will always be one of the propositions he asserts (assuming that the context is normal and that n has a referent). Arguably, it is the only proposition both asserted by utterances of the sentence in all normal contexts, and reasonably related to the meanings of the individual words and phrases of the sentence. For this reason, it qualifies as the meaning of the sentence in the language. In my opinion, this third view is more or less correct.14 Of course, in order to develop the view, something has to be said about the special case of negative existentials, like Moses did not exist, that Wittgenstein makes so much of. Suffice it to say that although these sentences do present serious problems for the view, there appear to be promising strategies for dealing with them.15 However, details aside, the point to be emphasized is that once the meaning of a sentence in the common language is distinguished from the information that the sentence is used to assert or convey on a particular occasion, the view just outlined is largely in harmony with Wittgenstein’s remarks in section 79. This point has been obscured by the fact that Wittgenstein himself often wrote as if he was oblivious to, or didn’t care about, the distinction between speaker meaning and sentence meaning. Indeed, it was many years after the Philosophical Investigations before philosophers and linguists, led by Paul Grice, developed the conceptual tools needed to make the distinction.16 Hence, in articulating the second and third views about the meanings of names, we have gone substantially beyond Wittgenstein’s text. See chapter 3 of my Beyond Rigidity. See the references given in n.8 above. 16 See Paul Grice, chapters 5, 6, and 14 (originally published as “Utterer’s Meaning and Intentions,” “Utterer’s Meaning, Sentence-Meaning, and Word-Meaning,” and “Meaning”, respectively) of his Studies in the Way of Words (Cambridge, MA, and London: Harvard University Press, 1989). 14 15

THE TRACTARIAN CONCEPTION OF LANGUAGE

23

With this in mind, let us sum up a few lessons about Wittgenstein’s discussion of names, and the general referentialist conception of language. First, his genuine insight that sentences containing names are often used to convey descriptive information that is open-ended, vague, and somewhat changeable, is not obviously incompatible with the view that the meaning of a proper name in the common language is simply its referent. Second, no matter what one concludes about that view, nothing in Wittgenstein’s remarks in section 79 takes us away from the general descriptive paradigm of language use—provided that we are willing to recognize that descriptions can be vague. Third, in understanding Wittgenstein’s critique of the descriptivist and referentialist paradigms, one must recognize that his arguments against specific theses about meanings of certain expressions—such as the thesis that the meaning of a name is its referent—are conceptually separable from his arguments against the view that understanding an expression is always to be explained as the result of knowing what it refers to, and understanding a sentence is always to be explained as the result of knowing the conditions under which it would be true. One could accept the latter arguments, even if one rejected the former. It is these arguments that are most important to Wittgenstein’s critique. His most central points are (a) that the real roots of linguistic meaning—the socially conditioned coordination of linguistic and nonlinguistic behavior—are conceptually prior to naming and describing, (b) that these roots must be in place prior to the introduction of either names or descriptive predicates, (c) that different instances of naming and referring do not always come to the same thing, and that because of this, talk about naming and referring requires as much analysis as other kinds of talk, and (d) that knowing that an expression refers to an object often presupposes the very linguistic competence that some descriptivists want to use it to explain. His fundamental positive thesis is that mastery of an expression is at base the mastery of a set of social practices by which speakers coordinate their activities. There may be some cases in which this mastery involves the independent recognition first of an object, then of a word, and finally of the fact that the community uses the word as a name of the object. But this is not always so—and not just because words can be meaningful without referring to anything. Even in cases in which a word does refer, there may be no such thing as identifying the object to which it refers, or of recognizing the naming relation, independently of understanding some word for the referent. In such cases, this prior understanding has to be

24

WITTGENSTEIN ON LANGUAGE

explained without presupposing any prior apprehension of the object, and intention to correlate it with an expression. This point concerning the foundations of meaning and understanding is independent of Wittgenstein’s rejection of the more specific semantic thesis that the meanings of ordinary proper names in English are their referents.

Wittgenstein’s New Conception of Language and Linguistic Analysis At this stage in the Investigations, around section 81, Wittgenstein turns from the rejection of his old Tractarian conception of language, and directs his attention to how changing his view of language affects his conception of philosophy. He continues to believe that philosophical problems are linguistic problems, and that philosophical analysis is a kind of linguistic analysis—but it is a new kind of linguistic analysis that no longer is seen as a species of logical analysis. In the sections between 81 and 133 he announces an interconnected set of what he takes to be obvious truths about meaning, the nature of philosophy, and the kinds of analyses it should try to provide. These would-be truisms are mostly negative in character, and are aimed at ruling out alternative views of meaning and analysis. In explicating these theses it is useful to consider them slightly out of order. We begin with what I will call thesis 1, which is discussed in sections 90 through 92. thesis 1 Natural language is not an incomplete realization of a linguistic ideal—a formalized calculus—nor do the sentences of ordinary language have meanings that are more revealingly expressed by sentences of such a calculus. There is no such thing as the logical form of a sentence (given by some sentence in the language of a calculus). In these sections we also find an indication of the kind of linguistic analysis that Wittgenstein endorses, and a warning about the illusion we fall into if we start thinking in terms of the (unique) analysis of a word or sentence. According to Wittgenstein, we do not give an analysis of a sentence because there is anything wrong with the sentence as it stands that demands clarification. Rather, we give an analysis when

THE TRACTARIAN CONCEPTION OF LANGUAGE

25

something about the sentence leads us into philosophical confusion. Conceivably, a sentence might receive different analyses if people become confused about it in different ways. Each analysis may be aimed at clearing up a particular confusion, even if no analysis clears up all confusions. This brings us to thesis 2. thesis 2 We don’t give the analysis of a word, phrase, or sentence by giving the linguistic rules that guide and govern our use of that expression. In using the words that we understand, we are not invariably guided by rules that we introspectively grasp, and that determine the correct application of our terms. Typically, we simply apply a word instinctively to a previously unconsidered case. What makes such an application correct is its agreement with applications made by the larger linguistic community. In some cases, we may be guided by rules, tables, or internal instructions. But even then, what we are guided by is something that could, in principle, be interpreted in different ways. In the end, such interpretations cannot themselves always be guided by further rules, tables, and instructions. Hence, our interpretations of many symbols must simply be instinctive and unthinking. This thesis is illustrated by the following passages drawn from sections 82, 83, 84, and 85. What do I call ‘the rule by which he proceeds’?—The hypothesis that satisfactorily describes his use of words, which we observe; or the rule which he looks up when he uses signs; or the one which he gives us in reply if we ask him what his rule is?—But what if observation does not enable us to see any clear rule, and the question brings none to light?—For he did indeed give me a definition when I asked him what he understood by “N”, but he was prepared to withdraw and alter it.—So how am I to determine the rule according to which he is playing? He does not know it himself.—Or, to ask a better question: What meaning is the expression “the rule by which he proceeds” supposed to have left to it here?17 17

Philosophical Investigations, section 82.

26

WITTGENSTEIN ON LANGUAGE

Doesn’t the analogy between language and games throw light here? We can easily imagine people amusing themselves in a field by playing with a ball so as to start various existing games, but playing many without finishing them and in between throwing the ball aimlessly into the air, chasing one another with the ball and bombarding one another for a joke and so on. And now someone says: The whole time they are playing a ball-game and following definite rules at every throw.18 I said that the application of a word is not everywhere bounded by rules. But what does a game look like that is everywhere bounded by rules? whose rules never let a doubt creep in, but stop up all the cracks where it might?—Can’t we imagine a rule determining the application of a rule, and a doubt which it removes— and so on?19 A rule stands there like a sign-post.—Does the sign-post leave no doubt open about the way I have to go? Does it show which direction I am to take when I have passed it; whether along the road or the footpath or cross-country? But where is it said which way I am to follow it; whether in the direction of its finger or (e.g.) in the opposite one?—And if there were, not a single sign-post, but a chain of adjacent ones or of chalk marks on the ground—is there only one way of interpreting them?20 In these passages, Wittgenstein is telling us that our use of words and other symbols is not always guided by rules; sometimes it is simply automatic. Moreover, even when we do have a rule in mind that we are attempting to follow, the rule itself is something that, in principle, might be subject to different interpretations. As a matter of simple empirical fact, our interpretations are often instinctive and unthinking, in the way that our interpretation of a directional sign with a finger pointing in one direction is. But even if, in some cases, our interpretation of one rule requires another rule to guide us, at some point the appeal to rules to interpret symbols has to give out, and we will find ourselves using words without thinking. These are cases in which we use words without following rules at all. This carries a lesson for the conception of philosophy as the analysis of language; just as the philosophical analysis Ibid., section 83. Ibid., section 84. 20 Ibid., section 85. 18 19

THE TRACTARIAN CONCEPTION OF LANGUAGE

27

of language is not a matter of giving the logical form of sentences, so it is not a matter of articulating the rules that we follow when we understand words. This leads Wittgenstein to a new and highly deflationary conception of philosophy.

Wittgenstein’s Deflationary Conception of Philosophy Wittgenstein’s conception of philosophy is summarized in thesis 3. thesis 3 The philosophical analysis of language does not aim at, and cannot issue in, theories of any kind. Philosophy (as Wittgenstein says in section 109) “is a battle against the bewitchment of our intelligence by means of language.” It is the untangling of linguistic confusions achieved by examining our words as they are ordinarily used, and contrasting that use with how the words are misused in philosophical theories and explanations. If philosophy were properly done, there would be no philosophical theories or explanations, since there is nothing for philosophy to explain. Its task is essentially therapeutic—to untangle particular confusions. This thesis is illustrated by sections 109, 116, 119, 124, 126, 127, and 128. It was true to say that our considerations could not be scientific ones. It was not of any possible interest to us to find out empirically ‘that, contrary to our preconceived ideas, it is possible to think such-and-such’—whatever that may mean. (The conception of thought as a gaseous medium.) And we may not advance any kind of theory. There must not be anything hypothetical in our considerations. We must do away with all explanation, and description alone must take its place. And this description gets its light, that is to say its purpose, from the philosophical problems. These are, of course, not empirical problems; they are solved, rather, by looking into the workings of our language, and that in such a way as to make us recognize those workings: in despite of an urge to misunderstand them. The problems are solved, not by giving new information, but by arranging what we have always

28

WITTGENSTEIN ON LANGUAGE

known. Philosophy is a battle against the bewitchment of our intelligence by means of language.21 When philosophers use a word—“knowledge”, “being”, “object”, “I”, “proposition”, “name”—and try to grasp the essence of the thing, one must always ask oneself: is the word ever actually used in this way in the language-game which is its original home? What we do is to bring words back from their metaphysical to their everyday use.22 The results of philosophy are the uncovering of one or another piece of plain nonsense and of bumps that the understanding has got by running its head up against the limits of language. These bumps make us see the value of the discovery.23 Philosophy may in no way interfere with the actual use of language; it can in the end only describe it. For it cannot give it any foundation either. It leaves everything as it is. It also leaves mathematics as it is, and no mathematical discovery can advance it. A “leading problem of mathematical logic” is for us a problem of mathematics like any other.24 Philosophy simply puts everything before us, and neither explains nor deduces anything.—Since everything lies open to view there is nothing to explain. For what is hidden, for example, is of no interest to us. One might also give the name “philosophy” to what is possible before all new discoveries and inventions.25 The work of the philosopher consists in assembling reminders for a particular purpose.26 If one tried to advance theses in philosophy, it would never be possible to debate them, because everyone would agree to them.27 We have in these passages a relentlessly deflationary view about philosophy, and what it can tell us about language, or anything else. However, Ibid., section 109. Ibid., section 116. 23 Ibid., section 119. 24 Ibid., section 124. 25 Ibid., section 126. 26 Ibid., section 127. 27 Ibid., section 128. 21 22

THE TRACTARIAN CONCEPTION OF LANGUAGE

29

there is something odd, even paradoxical, about Wittgenstein’s general theses. His broad negative theses are interesting and arresting; unlike most analytic philosophers, however, he gives little or no argument for them. One could say that in proceeding in this way he was being consistent. To argue for his theses would be to treat them as unobvious— as surprising and informative theoretical claims about language and philosophy. But that would be inconsistent with his claim that there are no genuine philosophical theories, and that all philosophical remarks are reminders of things we already know, yet sometimes disregard when we get confused. So there is a curious consistency in his not offering much in the way of argument for his philosophical views. His deflationary conception of philosophy is also consistent with, and even derivative from, his new ideas about meaning plus a set of unquestioned philosophical presuppositions he brings to the enterprise. The philosophical presuppositions include the then current and widespread assumptions (i) that philosophical theses are not empirical, and hence must be necessary and apriori, and (ii) that the necessary, the apriori, and the analytic are one and the same. Because he takes these assumptions for granted, he takes it for granted that if there are any philosophical truths, they must be analytic. To this he adds his new conception of meaning—with its rejection of abstract logical forms, its deflationary view of rule-following and algorithmic calculation, and its emphasis on social conditioning as generating agreement in our instinctive applications of words. Having jettisoned his old conception of meaning as something hidden—to be revealed by translating sentences into an idealized calculus of logical forms—and replaced it with a conception of meaning that sees it as arising from unquestioning but socially conditioned agreement, he has little room in his conceptual universe for surprising philosophical truths. Genuinely philosophical truths, if there should be any, can only be necessary and apriori, and so true in virtue of meaning. But how are the analytic truths of interest to a philosopher to be established, if they are not to be translated into the formulas of a logical calculus, and demonstrated by being given rigorous, but sometimes also innovative and insightful, logical proofs? For the Wittgenstein of the Investigations, the answer is that they don’t need to be established, since they are already accepted as beyond question by competent users of the language. To be sure, they may sometimes need to be brought into focus by assembling examples drawn from ordinary use that bring out the constitutive role they play in our language; but there is little room here for the discovery of genuinely

30

WITTGENSTEIN ON LANGUAGE

surprising philosophical truths. Thus, there are clear doctrinal reasons underlying Wittgenstein’s deflationary conception of philosophy. However, there is also an evident problem. His general theses about language and philosophy are not, by any means, obvious or already agreed upon; nor are they the sorts of things that one can just see to be true, once they are pointed out. On the contrary, they seem to require substantial explanation and argument, if they are to be accepted at all. Since Wittgenstein himself doesn’t, and for consistency’s sake can’t, always take on the task, the best we can do is to try to construct the needed arguments and explanations from what he gives us. Often, this requires going substantially beyond what is explicit in the text. It also makes the task of interpreting him treacherous. What, then, are we to make of his general remarks about philosophy and its relation to language? As I understand these remarks, we have no choice but to view them as, to some extent, self-undermining. As in the case of the Tractatus, Wittgenstein’s central philosophical theses in the Investigations lead to a view of philosophy at variance with the theses themselves. When this happens, something has to be wrong. It is certainly debatable what the chief source of the trouble is. However, it is hard not to see the double identification of the philosophical with (a subset of) the necessary and the apriori, and of these with the analytic, as central to the problem. Like so many other analytic philosophers in the twentieth century, Wittgenstein comes to grief over assumptions about these different modalities that he takes to be obvious. This theme, developed in volume 1, is one we will continue to trace as we discuss the philosophers covered in this volume—until we reach the point at which we have the conceptual tools to challenge the assumptions directly. As for the immediate historical impact of Wittgenstein’s strongly antitheoretical approach to language and philosophy in the Investigations, it was certainly felt by the ordinary language philosophers who followed and took much of their inspiration from him. Not surprisingly, we will find a similar disconnect in some of them—between the ambitious, and sometimes revisionary, philosophical theses they defended, and the officially modest and conservative methodology they claimed to practice. On the positive side, there certainly were successful cases in which careful examination of the ordinary use of philosophically significant words resulted in philosophical illumination. However, as we shall see, the ordinary language philosophers also suffered from the lack of a systematic theory of meaning and language use. However, before moving directly

THE TRACTARIAN CONCEPTION OF LANGUAGE

31

to that story, we need to examine what is arguably the core of the Investigations, namely, Wittgenstein’s discussion of rule following, and the contentious and revisionary doctrines about psychological language known as the private language argument that he derives from that discussion.

CHAPTER 2 RULE FOLLOWING AND THE PRIVATE LANGUAGE ARGUMENT chapter outline 1. Wittgenstein on Meaning, Understanding, and Following a Rule Understanding a word, or following a rule, is not an internal psychological state; standards of correct application involve agreement with a linguistic community 2. Evaluating the Argument A gap in the argument: the contents of some words, beliefs, and intentions may be inherited from the contents of perception Why agreement in meaning does not always involve simple point-by-point agreement in the application of words to individual objects; difficulties in appeals to more complicated conceptions of agreement in application 3. The Private Language Argument The basic idea: sensation words are associated with public criteria that determine correctness or incorrectness of their applications An ambiguity in the argument and two versions of the central thesis: (i) strong thesis: words like pain don’t refer to any private sensations; (ii) weak thesis: the meanings of such words do not consist simply in such reference

Collapse of the weak thesis into the strong 4. The Connection between Wittgenstein’s Views about Meaning and the Private Language Argument Why the views about meaning require the strong thesis, which is untenable 5. A Weaker, More Plausible, but Only Partially Wittgensteinian View of Sensation Language 6. Conclusion: The Legacies of the Investigations

THE PRIVATE LANGUAGE ARGUMENT

33

Wittgenstein on Meaning, Understanding, and Following a Rule Wittgenstein discusses understanding and following a rule in sections 143 through 155 and 179 through 202 of the Investigations. The main lesson of these sections is that understanding a word, or following a rule, is not an internal psychological state; rather, it is a dispositional state involving the use one makes of the word or rule. To understand a word is to be disposed to use or apply it in the correct way, over a wide range of cases, where by the correct way, we do not mean the way determined by a rule that the speaker knows and has internalized. The problem, as Wittgenstein sees it, with appealing to internalized rules to explain our understanding of words is that such rules are themselves made up of words or symbols which must be understood, if the rules are to be of any use. Obviously, this sort of explanation cannot go on forever. In the end, he thinks, we are left with a large class of words or symbols that we understand and are able to apply correctly, despite the fact that what guide us in these applications, and make them correct, are not further rules or symbols of any sort. When we reach rock bottom, we are not guided by internal rules at all; we simply apply words unthinkingly to new cases that are presented to us. What determines whether the new applications of such words are correct or not? If F is such a word, and I am inclined to apply it to certain things and not to others, what determines whether the things I am inclined to call F really are things to which the word correctly applies? One is tempted to say that I am right when I say of an object o it is F just in case the rule I mastered when I came to understand F, and continue to follow when I use it, says that F applies to a certain sort of object, and o is an object of that sort. Conformity with this rule, one may be inclined to think, is what makes some uses of F correct, while failure to conform to it makes other uses of F incorrect. However, Wittgenstein argues, this can’t be right, since typically there is no internalized rule that we follow when we use the word. What then does determine which applications of F to new cases are correct? The mere fact that I am inclined to say of something it is F can’t guarantee that the word correctly applies to the thing. If my use of F is to be meaningful at all, there must be some independent standard of evaluation that any particular application I make of it must live up to in order to be correct. Wittgenstein seems to think that the same argument

34

WITTGENSTEIN ON LANGUAGE

that shows that the standard of correctness cannot be determined by an internalized rule can be repeated to establish that it can’t be determined by any belief, intention, or other contentful mental state of mine. The problem, he thinks, is that in order to perform such a role, any such mental state must itself already have gotten its content from somewhere. A regress argument can then be used to conclude that the contents of all my words, and all my mental states, must, in the end, rest on something other than my mental states.1 Thus, Wittgenstein seems to suggest, the standard of correctness governing my use of a word cannot rest on anything internal to me, but must, instead, somehow come from something external. If this is right, then what makes my applications of a word correct is the fact that they conform sufficiently closely to an external standard. And what more natural place to look for this standard than in the linguistic community of which I am a part? Hence, Wittgenstein seems to think, to use a word correctly is to apply it in a manner that in some way conforms with how it is applied by others in one’s linguistic community. Of course, this presupposes substantial agreement in the community as a whole about how the word is to be applied. If there were no such agreement, then the word would have no meaning, and we could not speak of correct or incorrect applications of it at all. But where there is the necessary agreement, this agreement determines the standard of correctness for the word.2 1 A slight qualification should be noted here. As pointed out in chapter 1, although Wittgenstein seems to think that thought and language generally go hand in hand—so that what explains the contents of one explains the contents of the other—he seems to allow for a very restricted range of beliefs which are not tied to one’s use or mastery of language. Though the regress argument may not apply to these beliefs, their connection to behavior seems to be crucial. In addition, Wittgenstein clearly regards these beliefs as far too impoverished to provide the standards of correctness needed for our words. 2 As I read Wittgenstein, he does believe that some sort of conformity with community use provides standards of correctness for an individual’s use of many of our most basic terms. This interpretation appears to conflict with Saul Kripke’s interpretation in Wittgenstein on Rules and Private Language (Cambridge, MA: Harvard University Press, 1982). Kripke interprets Wittgenstein as presenting a “skeptical paradox” about meaning, and offering a “skeptical solution” to that paradox. According to Kripke, that solution involves giving up standards of correctness (that determine the truth or falsity of different applications of a term) in favor of conditions of justified assertability (that specify when one is justified in asserting that applications of the term are correct). (See in particular his discussion on pp. 110–12.) In my opinion, the view that Kripke attributes to Wittgenstein is, though fascinating, neither completely Wittgensteinian, nor ultimately coherent. See my “Facts, Truth Conditions, and the Skeptical Solution to the RuleFollowing Paradox,” in Philosophical Perspectives, vol. 12: Language, Mind and Ontology, edited by James Tomberlin (Oxford: Blackwell, 1998), 313–48; and my “Skepticism about Meaning: Indeterminacy, Normativity, and the Rule Following Paradox,” The Canadian Journal of Philosophy, supplementary vol. 23 (1997): Meaning and Reference, ed. Ali A. Kazmi, 1998, 211–49.

THE PRIVATE LANGUAGE ARGUMENT

35

Since, according to Wittgenstein, this is what it is for our most basic terms to be meaningful, all such terms must be associated with public criteria of application of some sort—criteria by which someone other than the speaker can, in principle, judge whether the speaker’s applications of the word are (generally) correct. This doesn’t mean that when we apply such a word to a new instance we invariably do so by checking to see that the public criteria are met. On the contrary, we often simply apply the word instinctively, without thinking. Nevertheless, Wittgenstein believes that in order for our use of the word to be meaningful, there must be public criteria that others could consult to determine whether some, at least, of our applications of it are correct. To sum up: one can’t mean anything by any word unless our most basic words (those not definable or derivable using other words) are such that sometimes it is possible for others to tell whether one’s use of them is correct. As we will soon see, this principle provides the basis for Wittgenstein’s famous and controversial private language argument.

Evaluating Wittgenstein’s Argument A Gap in the Argument: Perception, Intention, and Belief

Before we turn to the private language argument, however, we would do well to look more carefully at the argument just given, regarding meaning, understanding, and following a rule. That argument started with a demonstration that in the case of many words one is not guided by internalized rules or instructions (themselves thought of as containing words or symbols) when applying the word to new cases. This conclusion is clearly correct, and is established by the observation that denying it would invite an endless regress, which surely would be absurd. Next we are told that since, for many words, applications of them to new cases are not guided by internalized rules, the standards of correctness for evaluating such applications cannot be conformity to internalized rules that one is trying to follow. So far, so good. However, at this stage, we are invited to make a crucial inference. We are invited to infer that standards of correctness are not determined by any of one’s past or present beliefs or intentions, or indeed by anything that has been or is in one’s mind. This inference is questionable. We can bring this out by considering a simple example. Imagine that a man living alone on a desert island somehow gets the idea of setting

36

WITTGENSTEIN ON LANGUAGE

up a rudimentary language in order to bolster his memory and record the occurrence of significant events. Suppose he is able to perceive differences in colors, and on some occasion when looking at colored objects he forms the intention to use the symbol red to stand for one of the colors, the symbol blue to stand for another, and so on. In saying that he forms this intention, I mean that he sees one thing as being a certain color and another thing as being a different color; he attends to these colors and forms the intention to use the word red for one and blue for the other. Perhaps at the beginning, when he sees a new object for the first time he carries it back to the original objects for comparison, before judging the new object to be red or blue. However, as time goes on he grows confident in his inclinations to apply the words in accord with his original intentions, and his applications become instinctive. Later, the original objects are no longer available to him, and all he has to go on are his present inclinations to apply the words in some cases, but not in others. In this sort of case, no internal rule or set of instructions guides his applications. Nevertheless, what makes a particular application of one of the color words correct or incorrect, true or false, in any given case might still be whether or not the application conforms with the man’s original intention. If, all of a sudden, he gets intoxicated, and starts applying the word red to objects indistinguishable in color from objects to which in the past he applied the word blue, then his new applications are incorrect; when he says of the new objects they are red, what he says may be false, even if there are no practical means at his disposal that would allow him to uncover the error. Intuitively, this scenario seems to be possible.3 Here, the standard of correctness seems to be given by the agent’s past intention, which did The case of the solitary language user who introduces color words to stand not for logically private sensations, but for ordinary physical objects, is mentioned by Peter Strawson, in his “Review of Philosophical Investigations,” Mind 63 (1954): 70–99; reprinted in George Pitcher, ed., Wittgenstein (Garden City, NY: Anchor Books, Doubleday, 1966), 22–64, at pages 42 and 43. (All subsequent citations will refer to the Pitcher volume.) Strawson’s point is that any solitary user of a language has no check on correct application other than his own memory—no matter whether he uses words to stand for private sensations or physical objects. We wouldn’t want to conclude from this that no words can refer to physical objects; similarly, we shouldn’t conclude that the considerations adduced by Wittgenstein show that no words refer to private sensations. Nor does it seem that we should conclude that a language could never be spoken by a solitary individual. A. J. Ayer makes a further point along the same lines in his article, “Can There Be a Private Language?,” Proceedings of the Aristotelian Society, supplementary vol. 28, 1954, also reprinted in Pitcher. Ayer’s point is that unless we are justified in accepting at least some of our memories and some of our sense perceptions, the existence of other language users

3

THE PRIVATE LANGUAGE ARGUMENT

37

not itself involve the use of any language requiring interpretation. Moreover, the scenario is not ruled out by Wittgenstein’s regress argument that shows that our use of words is not always guided by internalized rules; the scenario grants that the agent is not guided by such rules. As I have indicated, Wittgenstein would, I think, ask how the agent’s initial intention to let the word red stand for the color got its content, and he would doubt that it could have done so without appeal to some antecedently understood language. In all likelihood he would raise similar questions about later intentions of the agent to use his words in accord with his earlier intentions. No doubt these are good questions, but simply to raise them is not to show that they are unanswerable. Wittgenstein has not, as far as I can see, shown that the agent couldn’t have the original intention to use the symbol red to stand for the property that his visual experience represents certain objects as having. Nothing in the Investigations rules out the possibility that perception provides the agent with representational content. We know that pre-linguistic agents can distinguish different colors. Hence, red things will typically look different to them than blue things do. The natural thought here is that the pre-linguistic agent’s visual experience represents red things in the environment as being red, blue things as being blue, and so on. But if the agent already has something—a part of his visual system—that represents things as being red, then it doesn’t seem to be a huge step to suppose he could introduce something else—a word—with that same content. Let us spell this out a bit. Suppose that, on some occasion, a pre-linguistic agent’s visual experience has the content that a certain object o is red. This being so, it would seem that the agent could form a belief that o is red which inherits its content from the content of the visual experience. In most cases, one would expect perceptual beliefs to arise naturally from perceptual experiences in this way. Once the content of the perceptual belief is in place, the agent still needs the ability to form an intention to use language to stand for that content. But why should this be problematic? Can’t individuals form intentions on their own, without the help of their neighbors? If they can, and if the content of the perceptual belief is already fixed, the linguistic intention to use an won’t help us check up on anything. In order for the linguistic behavior of others to have any weight with us, we have to be able to perceive and interpret that behavior. And if we are justified in doing that on our own, why isn’t a solitary language user justified in accepting his memory of how he intended to use a symbol as accurate? See in particular pages 256 and 257 in Pitcher.

38

WITTGENSTEIN ON LANGUAGE

expression to stand for that content should follow in train, and an instance of the meaningful use of language might be achieved, with its attendant standard of correctness.4 Of course, one could still go back to the beginning and ask how the agent’s perceptual experience got its representational content in the first place. Here it is plausible to suppose that the different causal relations in which the agent’s visual system stands to different classes of objects in his environment play important roles in determining the representational contents of his visual experiences. If this is right, then factors external to the agent are crucial in determining the contents of his perceptions, and, via his perceptions, the contents of his beliefs, intentions, and other mental states. In this sense, Wittgenstein may be right in thinking that what ultimately determines contents, and hence standards of correctness, are not things entirely internal to an agent. But even if the determinants of both mental and linguistic content must, at bottom, always include some external factors, he has given no good reason to suppose that the mental contents required by ostensive language learning of the sort of linguistic expressions and constructions found in our ordinary language must always involve facts about a wider linguistic community. This crucial gap in Wittgenstein’s argument prevents it from establishing his desired conclusion—that the standards of correctness required for the meaningful use of language can be found only in agreement between the agent’s use of an expression and that of a larger linguistic community.

Problems with Agreement: Doubts about Wittgenstein’s Conclusion

This is not the only problem with Wittgenstein’s argument. In addition to being inadequately supported, his conclusion is one there is reason to doubt. The desert island case—in which a language is invented and spoken by a single person—provides one such case. However, let us put that theoretical possibility aside, and concentrate instead on the normal case—that of an agent trying to learn the language spoken in some linguistic community. Let us focus on the point at which the agent is trying to learn the word F. Since F already has a meaning in the linguistic community, the agent has to learn to use the word in accord with that meaning; he has to come to understand it in a way that matches the way Note the significance of this case for our treatment in chapter 1 of Wittgenstein’s discussion of ostensive definition, and his criticism of the Augustinian picture of language acquisition.

4

THE PRIVATE LANGUAGE ARGUMENT

39

it is understood in the community as a whole. Since this is his goal, he recognizes that, in general, when he says of something, it is F, what he says will be subject to correction by other members of the community— their standards for applying F will be important determinants of the correctness of his utterances. In this sort of case, there is a trivial sense in which the standard of correctness governing his use of the term involves conformity with others in his linguistic community. The triviality is this: insofar as he is attempting to use the word with the meaning it has in the community, he uses it correctly only if his use conforms to that meaning. Of course. However, there are two points to keep in mind. First, the agent might mean something by the word even before his meaning matches what the community means by it. If his goal is to match the community meaning, he will not have achieved it. But he still will have meant something; so not all meaning is agreement with others. Second, even though his goal is to use the word with the meaning it has in the community, this does not establish the claim that what makes a particular application of the word to an object correct or incorrect is whether or not members of the community would apply the word to that object. Different people can mean the same thing by a word, even if they fail to agree on many applications of it; and in some cases a word may correctly apply to certain things even though most members of the linguistic community who understand the word do not recognize this, or even think otherwise. These points are commonplace, and have to be accommodated by any account of meaning. For concreteness, I will illustrate them with a pair of examples. Consider the ‘⫹’ sign.5 It stands for a certain function defined on all pairs of natural numbers. For any natural numbers n and m, there is one and only one number z which is the answer to the question, What is n ⫹ m? 6 There is, of course, lots of agreement on the use of the ‘⫹’ sign. Everyone agrees that 2 ⫹ 3 ⫽ 5, for example. Or at any rate, if someone doesn’t agree, we think that the person doesn’t know how to add, or perhaps doesn’t know what the words mean. However, there are also certain calculations about which there is no community-wide agreement. When the numbers get too large, we simply can’t do the math, and we may give no answers. Now suppose that two remarkable individuals come along. Both master the ‘⫹’ sign in the sense that they The example is based on Saul Kripke’s discussion in Wittgenstein on Rules and Private Language. 6 ‘n’ and ‘m’ are the numerals for n and m, respectively. 5

40

WITTGENSTEIN ON LANGUAGE

agree with the community at large on all standard calculations involving the ‘⫹’ sign that the community can perform. However, they differ from the community in being able to perform many calculations on numbers that are larger than the rest of us can handle. When given these numbers to add, the two individuals come up with definite answers, even though other members of the community just give up, without coming to an answer. Suppose, further, that our two individuals give different answers for some of these calculations. Since they give different answers to some of the new cases, (i) and (ii) cannot both be correct: (i) they both mean addition by ‘⫹’; (ii) they both give correct answers to all the new cases. Yet both take themselves to mean by ‘⫹’ just what the community means, and both agree equally well with the community on all addition problems to which the community would provide answers, if asked. Suppose that they do mean just what the community means by ‘⫹’—namely, addition. If the standard of correctness for particular answers were simply agreement with the rest of the community, then either the new answers of both agents would have to be correct—since they both agree with the community on all cases in which the community can provide answers, and disagree with the community on none—or the new answers of both would have to be incorrect—since both go beyond those that can be given by the community at large. But this result isn’t right. Surely it is possible that the new answers given by one of them are correct, whereas the new answers given by the other are incorrect, even though both mean what we do by ‘⫹’. If this is right, then the criterion for understanding a word as it is used in a linguistic community, and the standard for applying it correctly in any given case, cannot simply be agreement with the particular applications made of it by the rest of the community. Something other than simple agreement about particular applications must be at work. Next consider a different case. One day some people who live on an island come across a few little furry animals and decide to call them rabbits. They say, Let’s introduce the word ‘rabbit’ to apply to these animals, where by these animals they mean any animal of the same species as those they just encountered. (They have some idea of what a species is, and they can recognize instances of several different species, though they may have no definition of the notion.)7 Over time, people in the The idea that all of this is explicitly stipulated is used to simplify and clarify the example. The same situation could arise without such explicit stipulations.

7

THE PRIVATE LANGUAGE ARGUMENT

41

community learn the word rabbit, and apply it to the various rabbits they come across on the island. This goes on long after the original rabbits that prompted the introduction of the term have died. Then one day some islanders make an unprecedented trip to the mainland. Among the things they discover are animals that look a lot like the rabbits on their home island. They simply assume that these animals are of the same kind as those on the island, and they call them rabbits too. The travelers bring this news back to the island, and it is routinely accepted by most members of the population that there are rabbits on the mainland, as well. However, a few skeptics doubt this. Hence, although most members of the linguistic community apply the term rabbit both to animals on the island and to animals on the mainland, a few insist that the latter are not rabbits. If the standard of correctness for any application of the term were simply agreement with the applications actually made by the larger linguistic community, then it would follow that the skeptics in this case are, by definition, wrong when they refuse to apply the word to the animals on the mainland. But they are not wrong by definition. Suppose it is eventually discovered that the animals on the mainland have a different evolutionary history from the animals on the island, that they have different cellular structure and DNA, and that they are not members of the same species as the animals on the island. The proper conclusion to draw in such a case is that most members of the linguistic community were wrong when they said that the animals on the mainland were rabbits.8 But if that is so, then the standard of correctness for the application of a term is not simply agreement with the application that most members of the community actually make. Perhaps the standard should be agreement with the application that most members of the community would make if certain idealized conditions were satisfied—e.g., if they knew all the relevant facts (provided one could give some general, noncircular definition of what facts count as relevant that did not depend on a conceptually prior assumption about what the word really means).9 Or perhaps something 8 At least this would be so initially. Of course, over time the meaning of the word rabbit might change, if the islanders come to regard both the animals on the island and those on the mainland as paradigmatic examples of individuals to which the word applies. But this point about meaning change does not undermine the point about standards of correctness before the change. 9 The same appeal to idealized dispositions could be made in the case of our earlier example ‘⫹’. For a very good discussion (focusing on ‘⫹’) of the problems involved in giving noncircular specifications of idealized conditions to be satisfied if agreement with applications made by agents in those conditions is to be used as the standard of correctness for a word, see Kripke, Wittgenstein on Rules and Private Language, pp. 22–37, and p. 111.

42

WITTGENSTEIN ON LANGUAGE

different is at work. Either way, there are words the meanings of which are not determined by applications most members of the community actually agree on, or would agree on in ordinary, nonidealized situations. The case just considered was one in which the skeptics attached the same meaning to the term rabbit as most people in the community, even though they were right when they disagreed with most people on many particular applications. One can also construct analogous cases in which members of two groups are inclined to apply a word to exactly the same objects or individuals, yet the two groups really mean something different by it. For example, imagine that the islanders and the mainlanders each introduce the word rabbit with the stipulation that it is to apply to all and only members of whatever species it is that inhabits their area. So rabbit, as used by the islanders, applies to members of the species found on the island, and rabbit, as used by mainlanders, applies to members of the species found on the mainland. Later, when islanders go to the mainland, and mainlanders go to the island, both assume that the animals they see are what they call rabbits. So the two groups actually are disposed to predicate the same word of the same things. However, in fact, the species on the mainland and the species on the island are different. Because of this, the meanings the two groups attach to the word are different, and the particular applications they make of it sometimes differ in truth value. For these reasons, one certainly cannot accept the thesis that what it means for any word to correctly apply to an object is for most members of the community to agree that it does. However, it is not clear what relationship this clear, simple, but incorrect thesis bears to Wittgenstein’s far less clear thesis that at least for our most basic words—those that are understood without verbal definitions or appeals to rules using other words—the standards of correctness governing them involve agreement with uses of them by other members of the linguistic community. We cannot know for sure whether we should accept or reject his thesis until certain crucial matters are clarified. For one thing, we need to know which words we are talking about. A few pages back, I followed Kripke’s lead in using the symbol ‘⫹’ to illustrate and test Wittgenstein’s views about meaning. However, this may be the wrong word to test. Arguably, ‘⫹’ is one of the best examples of words which are associated with explicit rules that guide us and govern their use.10 Most of us The significance of this for Kripke’s interpretation of Wittgenstein is discussed on pp. 334–37 of my “Facts, Truth Conditions, and the Skeptical Solution to the Rule-Following Paradox.”

10

THE PRIVATE LANGUAGE ARGUMENT

43

learned rules in grade school for computing sums which still guide our application of ‘⫹’, and even without those relatively efficient rules, most of us know how, in principle, to define addition in terms of counting. To be sure, the rules that guide us are formulated using other words the meanings of which must be explained—words like zero, number, and successor (or the number following this number)—and these words may be primitive, and understood without definition. If so, they should qualify as words to which Wittgenstein’s thesis about meaning directly applies. This is significant, since, when words such as these are involved, it is harder to produce apparent counterexamples even to the simple thesis that what makes an individual’s application of a word to an object correct is that most members of the linguistic community would agree with it. When words and phrases like red, hard, sweet, up, under, on top of are involved, the idea that something resembling this simple agreement thesis is correct is more plausible than it is for words with more complicated or complex contents. Thus, by determining the proper scope of Wittgenstein’s thesis about meaning, one may render it more plausible and less subject to obvious counterexample.11 Nevertheless, this cannot be the whole story. The number of words that are understood without verbal definition or appeal to explicitly formulated rules that specify their use is large, as Wittgenstein would have been the first to emphasize. Thus, in many cases doubts of the sort we have raised about simple agreement as the standard of correctness cannot be avoided by appealing to the existence of rules and definitions that shift the burden onto other words. This brings us to the other main source of unclarity in Wittgenstein’s thesis about meaning. What exactly is meant by agreement with uses of the word by others in the linguistic community? I doubt that he literally meant that for every object o a relevant word w correctly applies to an object o iff most members of the linguistic community would actually agree, if asked, that w correctly applies to o. However, if he didn’t mean this, then it is not clear precisely what he did mean. It seems as if something must be involved more complicated than simple comparisons of who would actually predicate the word of which 11 It might be argued that even the term rabbit, used to construct the apparent counterexample just considered, should not be regarded as a basic word the standards of correctness for which are defined by substantial communal agreement in application to particular cases. Since in the example the word was introduced with a stipulation that it was to apply to all members of the species of certain individual animals, agreement on that stipulation may be all that is needed. Which, then, one wonders, are the basic words?

44

WITTGENSTEIN ON LANGUAGE

objects. Perhaps a more indirect kind of agreement is intended—like agreement under idealized circumstances, agreement in the pattern of reasoning that goes into making or justifying particular applications of the term, or agreement with well-placed and fully informed speakers. However, one must be careful that in searching for the needed form of agreement, one neither (i) illicitly appeals to the very meanings that agreement is supposed to define (e.g., in attempting to spell out the idealized conditions in which agreement is supposed to be constitutive of meaning), nor (ii) violates other Wittgensteinian commitments about the relationship between language, belief, and mental states generally. For example, one kind of agreement that may play an important role in grounding the meaning of a term in a language is agreement in the core beliefs and intentions that speakers of the language have concerning it. However, Wittgenstein cannot, in general, appeal to this sort of agreement, since, according to him, we often can make no sense of the contents of the relevant beliefs and intentions prior to establishing a framework of community agreement on the use of words in the first place. In light of problems like this, the prospects for a successful defense of his central thesis about the role of community agreement in grounding meaning and explaining linguistic understanding do not appear (to me) to be very promising. At this point, the best that might be said of the thesis is that it is too vague either to be persuasively defended, or to be conclusively refuted. But then, since no precise and viable interpretation has been put on offer, we are best advised to regard the thesis with skepticism.

The Private Language Argument The Basic Idea

We now turn to the private language argument found in sections 243 through 315. The basic idea behind the argument is this: Suppose I were to introduce some word to refer to a momentary sensation that was private to me—i.e., a particular sensation that others could not have or observe (though they could, perhaps, have similar sensations of the same type).12 Suppose, further, that in the future I wanted to use 12 Wittgenstein’s private language argument is standardly seen as directed against the view that sensation words apply to logically private inner experiences—things of a non-physical sort that would be conceptually impossible for anyone other than the agent to experience or observe in any way. However, I don’t believe that this restriction of the private to the logically private is

THE PRIVATE LANGUAGE ARGUMENT

45

the word to refer to other sensations of mine that were of the same type. In such a situation, neither I nor anyone else would be able to conclusively determine whether I was using the word correctly—i.e., to determine whether the sensations to which I now apply the word really are of the same type as the one (or ones) that prompted me to introduce it. Without such an ability, there could be no content to the claim that I was using the word correctly, and hence no distinction between correct and incorrect use. But whenever we mean anything by a term, there is always such a distinction. Thus, Wittgenstein concludes, no one can meaningfully use a term purely privately. He gives a brief sketch of the scenario in section 258. Let us imagine the following case. I want to keep a diary about the recurrence of a certain sensation. To this end I associate it with the sign “S” and write this sign in a calendar for every day on which I have the sensation.—I will remark first of all that a definition of the sign cannot be formulated.—But still I can give myself a kind of ostensive definition.—How? Can I point to the sensation? Not in the ordinary sense. But I speak, or write the sign down, and at the same time I concentrate my attention on the sensation—and so, as it were, point to it inwardly.—But what is this ceremony for? for that is all it seems to be! A definition surely serves to establish the meaning of a sign.—Well, that is done precisely by the concentrating of my attention; for in this way I impress upon myself the connection between the sign and the sensation.—But “I impress it on myself ” can only mean: this process brings it about that I remember the connection right in the future. But in the present case I have no criterion of correctness. One would like to say: whatever is going to seem right to me is right. And that only means that here we can’t talk about ‘right’. Here Wittgenstein seems to be saying that if all there is to the meaning of a certain sign is some type of sensation, instances of which it is supposed to apply to, and if those instances are entirely private to those that essential. In my opinion, much of what Wittgenstein says would remain intact, even if one took seriously the possibility that private sensations were particular physical events or states in one’s brain. Wittgenstein himself would not, I think, have taken this possibility seriously—for the same reasons our next philosopher, Gilbert Ryle, did not. Since these reasons are discussed and criticized at length in chapters 3 and 4, I will not go into them here. From now on, unless I explicitly indicate otherwise, when I speak of private momentary sensations, one can understand ‘private’ in either the more or the less restricted sense.

46

WITTGENSTEIN ON LANGUAGE

have them, then there can be no criterion for determining whether new applications of the sign are correct. But if there is no criterion, then there is no distinction between correct and incorrect application. And if that is the case, then there is no meaning. So we have an attempted reductio ad absurdum of the claim that the meaning of an expression simply consists in the kind of private sensation that it stands for.

An Ambiguity in the Argument

Things are not so simple, however. When we look closely at other sections of the Investigations, we find a certain ambiguity about the content of the thesis that Wittgenstein is attacking. In order to understand the private language argument, it is crucial to determine precisely what is supposed to be ruled out as impossible. One relevant passage is found in section 243. But could we also imagine a language in which a person could write down or give vocal expression to his inner experiences—his feelings, moods, and the rest—for his private use?—Well, can’t we do so in our ordinary language?—But that is not what I mean. The individual words of this language are to refer to what can only be known to the person speaking; to his immediate private sensations. So another person cannot understand the language. Note the last two sentences. The view that Wittgenstein is setting out to attack seems to be one which maintains that “individual words of this language are to refer to what can only be known to the person speaking; to his private sensations.” If he is going to reject this view, then his negative thesis must be stronger than the thesis that no word can be such that its meaning consists entirely in the kind of private sensation it stands for. Rather, his thesis must be that no word can stand for a private sensation at all. That may be the lesson of section 243. However, things become somewhat muddied in the sections that follows. Here is what Wittgenstein says in section 244: How do words refer to sensations?—There doesn’t seem to be any problem here; don’t we talk about sensations every day, and give them names? But how is this connection between the name and the thing named set up? This question is the same as: how does a human being learn the meaning of the names of sensations?—of the

THE PRIVATE LANGUAGE ARGUMENT

47

word “pain” for example. Here is one possibility: words are connected with the primitive, the natural, expressions of the sensation and used in their place. A child has hurt himself and he cries; and then adults talk to him and teach him exclamations and, later, sentences. They teach the child new pain-behavior. “So you are saying that the word ‘pain’ really means crying?”— On the contrary: the verbal expression of pain replaces crying, and does not describe it. According to Wittgenstein, uses of the word ‘pain’ are connected with certain natural behaviors like crying or groaning. He seems to suggest that when one learns to use the word, one learns to replace those behaviors with utterances of sentences containing the word ‘pain’, or other related words. Wittgenstein continues this point in section 256. Now, what about the language which describes my inner experiences and which only I myself can understand? How do I use words to stand for sensations?—As we ordinarily do? Then are my words for sensations tied up with my natural expressions of sensation? In that case my language is not a ‘private’ one. Someone else might understand it as well as I.—But suppose I didn’t have any natural expression for the sensation, but only had the sensation? And now I simply associate names with sensations and use these names in descriptions.13 Here, Wittgenstein builds on the suggestion that the word ‘pain’ is, by virtue of its meaning, correlated with certain natural behaviors. Because of this correlation, our ordinary language containing the word ‘pain’ isn’t a private language. This suggests that to be a genuinely private language, and hence the kind of thing attacked as impossible by the private language argument, the language must contain words the meanings of which are exhausted by the private sensations they stand for, rather than including some connection to publicly observable behavior. Taken together, the passages we have looked at contain an ambiguity regarding what a private language is supposed to be. In some passages, for example in sections 256 and 258, Wittgenstein seems to be saying that if all there is to the meaning of a certain sign is some entirely private type of sensation that it stands for, then there can be no criterion for 13

Philosophical Investigations, section 256.

48

WITTGENSTEIN ON LANGUAGE

determining whether new applications of the sign are correct or not. But if there is no criterion, then there is no distinction between correct and incorrect application; and if that is the case, then there is no meaning. So on this interpretation we have an attempted reductio ad absurdum of the claim that the meaning of an expression could consist simply, or entirely, in the kind of private sensation that it stands for. However, in other passages—for example, in section 243—Wittgenstein seems to be saying something stronger—namely, that there can be no language, or part of language, in which words stand for sensations that are entirely private to agents. And that is pretty close to the claim that there are no private sensations at all. What we have in these sections is an oscillation between a stronger and a weaker thesis as conclusion of the private language argument. strong thesis No meaningful word used by a speaker stands for any kind of private sensation, idea, or internal experience of the speaker— i.e., for any type instances of which the speaker experiences and which are private to one who has them. So, in particular, pain does not stand for a kind of sensation of this type. Thus, when a person says that he is in pain, or has a headache, the person is not reporting or describing any private internal happenings. Rather, the person is engaging in a verbal performance that replaces the natural behaviors of crying, screaming, rubbing one’s head, and so on. When a speaker says that someone else is in pain, the speaker is simply describing what has happened, or is happening, to the person, and that person’s disposition to behave in a certain ways. weak thesis No word used by a speaker is meaningful solely in virtue of standing for any kind of private sensation, idea, or internal experience. Like all words, any word applying to private sensations must also be associated with public criteria that allow others to determine whether the speaker is using the word correctly. So when a speaker s says that s is in pain, this can be taken as reporting a private experience so long as there are public criteria involving the behavior of s and what has happened, or is happening to s, which, in principle, determine whether s speaks correctly.

THE PRIVATE LANGUAGE ARGUMENT

49

On the face of it, Wittgenstein’s argument does not seem to establish the strong thesis—at least it doesn’t if the weak interpretation is available and viable. What his argument demands is that whenever there is meaning, it is possible in principle to determine whether a word has been used correctly. If a word can apply to private sensations while also being associated with public criteria that allow such determinate checking, then that would seem to be enough. Thus, it looks as if the weak thesis is the best that Wittgenstein’s argument could support. This, roughly, is the conclusion that Peter Strawson comes to in his review of the Philosophical Investigations.14 Although there is a good deal to be said for this interpretation, there is also a problem with it; it is not clear that the weak thesis is tenable. When pressed, one has a hard time keeping it either from collapsing into the strong thesis, or from being weakened to the point that it will no longer suit Wittgenstein’s philosophical purposes. The Collapse of the Weak Thesis into the Strong Thesis

To see this, let us begin by supposing that the weak thesis is correct, and that when I sincerely and correctly say “I have a headache,” it is true both that I refer to a private sensation of mine, and that there are public criteria available to others that could, in principle, often be used to determine whether or not I have applied the word correctly. Let us suppose that these public criteria involve behavioral evidence that could be used to assess the correctness or incorrectness of my report—behavior like being disposed to rub my head, take aspirin, complain that I don’t feel well, and so on. We further stipulate that, on this view, it is necessary for the truth of the claim that I have a headache that I experience a private sensation of a certain sort, but that this isn’t the only thing involved in determining the truth of my claim. There is, however, a problem in making sense of all this. Up to now we have tacitly assumed that in at least some substantial range of cases, the public criteria involving agreed-upon pain behavior are themselves supposed to be necessary and sufficient for determining the correctness of my claim. But how can this be? Isn’t it conceivable that in many cases the pain behavior could occur without being accompanied by any sensation of pain, or vice versa? One would certainly think so. The internal sensation is, on this view, one empirical state, the behavior typically 14

See Strawson, “Review of Philosophical Investigations,” pp. 41–49 in Pitcher.

50

WITTGENSTEIN ON LANGUAGE

manifested is another. It would seem that these are just two different states of affairs, with no inevitable, necessary connection between them. But if this is right, then the fact that the public criteria are satisfied in a range of cases doesn’t conclusively guarantee that any sensations of the right sort exist. Similarly, private headache sensations may exist, even if the public criteria aren’t satisfied. What then is the connection between the private sensation reported and the satisfaction of the public criteria? Perhaps satisfaction of the public criteria—my rubbing my head, taking aspirin, and complaining— is a good, but fallible, rule of thumb that others can use to justify their thought that my internal sensation exists. It seems reasonable to think so. Normally, we would all agree that behavior like this would be evidence that I have a headache, even if it would never constitute conclusive and irrefutable proof. However, if it would not constitute such proof, then it is not clear that satisfaction of public criteria can play the role that Wittgenstein’s views about meaning, and his use of them in the private language argument, seem to require. According to Wittgenstein, whenever there is meaning, the standards that determine correct use of an expression are based on publicly available facts that guide the use of the expression by the broader community.15 In the case of mentalistic and sensation language, these facts include behavioral facts, including facts about dispositions to behave in readily observable ways. Since the behavioral facts associated with such language are supposed to be the very determinants of correctness or incorrectness for a substantial range of sensation reports, it would seem that the public criteria associated with such reports cannot provide rules that are, in principle, either incomplete, or merely good, but fallible, rules of thumb. The problem is that it appears that this is all they can provide—if, as the weak thesis maintains, the correctness of such reports also requires reference to the private sensations of the agent. Hence, the weak thesis is untenable. Certainly, it cannot be the case (i) that necessarily, in some substantial range of cases, the claim that I have a headache is correct iff certain behavioral and other publicly available facts B about me hold, and (ii) that in every case, the occurrence of a private sensation of a certain type S in me is a necessary condition for my having a headache, unless (iii) in some substantial range of cases, the publicly available facts B necessitate 15 Although we have seen that this need not hold true for all words, we assume that for Wittgenstein it is supposed to hold true of sensation words like ‘pain’.

THE PRIVATE LANGUAGE ARGUMENT

51

the existence of a sensation of type S, rather than merely providing good, but fallible, evidence for it. Since, in fact, B doesn’t seem to necessitate the existence of any sensations of type S, (i) and (ii) cannot jointly be maintained. One could, of course, weaken (i) by replacing ‘iff’ with ‘only if’. But then satisfaction of Wittgenstein’s public criteria would never, in itself, determine the correctness of applications of a sensation term like headache. Since this result appears to be incompatible with his views about meaning, the weak thesis is not available to him. This explains, I think, his tendency to move back at crucial points to the strong thesis. According to the strong view, we never refer to private sensations at all. When I speak of my own sensations—for example, when I say that I am in pain—I am not describing anything; I am simply engaging in a verbal substitute for natural nonverbal behavior. When I speak of the sensations of others—for example, when I say that they are in pain—I am making a claim to the effect that they are presently disposed to produce just the sorts of nonverbal behavior for which saying the words I am in pain would constitute a natural replacement. Since this view of sensation language is broadly similar to the view developed in great detail by our next philosopher, Gilbert Ryle, in his widely influential book, The Concept of Mind, we will wait until then to present a critical, point-by-point evaluation of it. At this stage, the important thing for us to see is how Wittgenstein’s argument naturally leads him to this strong conclusion.

The Connection between Wittgenstein’s Views about Meaning and the Private Language Argument Why Wittgenstein’s Views about Meaning Require the Strong Thesis

The private language argument is supposed to be an application of Wittgenstein’s general views about meaning, according to which for an expression to be meaningful, something must determine to which objects it correctly applies, and to which it does not. If P is meaningful, then some cases in which we apply P to a particular object by saying this object is P are correct applications of the predicate and some are not. And what do we mean by correct? Presumably, an application of P to an object made by saying this object is P is correct iff what we say is true, and an application is incorrect iff what we say is not true. Thus, it is natural to interpret Wittgenstein as saying that if P is meaningful,

52

WITTGENSTEIN ON LANGUAGE

then something must determine which objects P is true of, and which P is not true of. What can that something be? Wittgenstein thinks that he has shown that it can’t in general be any internalized rule that the speaker has mastered in learning P. He also thinks that it can’t be any belief or intention internal to the speaker. It does not help to appeal to these, he thinks, because mental states rich enough to ground our applications of words must get their contents (and resulting correctness conditions) from somewhere, and he believes that the same insuperable problem arises in explaining what determines these as arises in explaining what determines the contents (and correctness conditions) of our words. So, he concludes, when P is one of the many predicates not learned by verbal definition, and not associated with an explicit rule that guides speakers and determines correct application, nothing internal to the speaker determines which objects his predications of P are true of.16 Instead, he concludes that what determines the contents (and correctness conditions) of P is some pattern of agreement in the linguistic community of which the speaker is a part. As we have seen, exactly what this agreement is supposed to consist in, and how extensive it is supposed to be, remains unclear. However, roughly speaking, the idea seems to be that for some substantial range of cases, for P to be true of an object o is for members of the community to be disposed to apply P to o in the right sorts of idealized conditions. And if one asks, What explains this agreement?, Wittgenstein will answer that we are simply trained by others to use words in certain publicly identifiable situations, and not in others. Earlier we saw that there is reason to be skeptical of all this. Be that as it may, when Wittgenstein comes to the private language argument, he sees his task as that of applying his general conception of meaning to mentalistic and sensation language.17 Here, it seems, he has a simple Here Wittgenstein neglects (among other things) the plausible suggestion that since perceptual experience has content in virtue of causal relations between parts of the perceptual system and things in the world, the contents of belief, intentions, and, ultimately, language might be inherited in substantial part from the contents of our perceptual experience. 17 It is, I think, a constraint on how to interpret what Wittgenstein means by community agreement, and its role in determining standards of correctness, that these standards yield the sorts of consequences he takes them to have for the private language argument. We have seen that his general discussion of meaning and rule following leaves it rather unclear what sort of community agreement is required in order for the meaningful use of terms to be possible. Since weak conceptions of agreement—those that do not require agreement on applications of the word to particular individuals, provided there is substantial uniformity involving, e.g., core beliefs and intentions concerning the word, including the reasons speakers give for their applications of it—would not yield the surprising conclusions about sensation language that 16

THE PRIVATE LANGUAGE ARGUMENT

53

thought. If there were nothing more to the meaning of a word like pain than the private experiences it stands for, then no publicly observable facts could possibly serve as the basis for community-wide agreement regarding particular applications of the word. But if that were so, then there could be no standard of correctness for the use of the word—since there would be nothing that determined which objects it was true of. Since such a standard of correctness is required in order for there to be meaning at all, there can be no word the meaning of which consists entirely of the private experiences for which it stands. Although this view of meaning may initially seem to motivate the weak thesis associated with the private language argument, we have seen that the weak thesis is unsustainable. If the public criteria associated with sensation words really do determine, for some substantial range of cases, which applications of the word are correct, then reference to internal sensations must be dropped altogether, in which case the weak thesis collapses into the strong thesis. The strong thesis arises by taking seriously the idea of a public criterion that determines at least some of the things a predicate like pain is true of. If, like Wittgenstein, one thinks that for some substantial range of cases what makes it the case that a certain predicate P is true of certain objects is that members of the community are disposed to agree (in the right sorts of circumstances) about its application to those objects, then one naturally will think that it is impossible for P not to apply to those objects, if members of the community would, in the proper circumstances, agree that it applies (and vice versa). But then, it is hard to see how any predicates could be true of private sensations at all. For surely, if there are such sensations, then whatever the public criteria that might be associated with a sensation predicate P, it is at least possible that the criteria might be satisfied in many cases, even though the private sensation did not exist. But to say this is just to say that it would be possible for the public criteria associated with P to be satisfied much of the time, without it being the case that P was true of anything. Since this contradicts the natural interpretation of Wittgenstein’s claim that public criteria associated with a word provide a standard of correctness that determines which uses of it are correct, his position is most naturally viewed as giving rise to the strong thesis associated with the private language argument. he is after, this is good reason to believe that his conceptions of community agreement, and the resulting standards of correctness, are intended to be quite demanding.

54

WITTGENSTEIN ON LANGUAGE

Why the Strong Thesis Is Untenable

In a way this is good, because it means that his argument has the farreaching philosophical consequences he intended it to have. However, it may also be bad, since the claim made by the strong thesis is subject to powerful objections. The first objection involves what seems to be a natural extension of the thesis. If Wittgenstein’s general view about language leads to the strong thesis about sensation language, then it would also seem to lead to similarly strong theses about other parts of language. Consider, for example, uses of language to talk about the past. Just as none of us can now observe or experience the private sensations of others, so none of us can now observe or experience past events. However, something must determine which claims about the past are true and which are not. According to Wittgenstein’s general conception of language, it would seem that the only thing we could appeal to is community-wide agreement about past events. But in order for there to be communitywide agreement now about past events, there must be criteria presently available to all on which to base such agreement. Surely, however, if our language about the past makes reference to past events, in many cases it will be conceptually possible for the presently available criteria associated with a sentence to be satisfied, even though the purported event referred to by the sentence never occurred. There will also be lots of cases in which it is conceptually possible that the past event occurred, even though the presently available criteria associated with the sentence are not satisfied. Since this seems to contradict the natural interpretation of what it means for public criteria associated with a sentence to provide a standard of correctness for the use of that sentence, Wittgenstein’s position could naturally be interpreted as giving rise to the strong thesis that none of our sentences ever describe past events. But that is absurd. Since the pattern of argument appears to parallel the pattern that gives rise to the strong thesis about sensation language, either some explanation of why that appearance is deceptive must be given, or both arguments should be rejected. Since Wittgenstein wishes to retain his views about private language, he has reason to look for a difference between the two cases. Perhaps there is some asymmetry between talk about sensation and talk about the past that could be exploited to retain the strong thesis about sensation language while rejecting it for temporal talk. If so, however, it is not immediately obvious what it involves, and Wittgenstein does not spell it out. As a result, his position remains potentially vulnerable.

THE PRIVATE LANGUAGE ARGUMENT

55

The second objection to the strong thesis focuses on sensation language itself. In understanding this objection, it is important to try to determine what the alternative to Wittgenstein’s strong thesis might be. At the time he wrote, the most salient competitor to his view was one in which sensations were taken to be non-physical things, logically private to each agent who experienced them, that possessed precisely the properties they appeared to have.18 Thus, it was natural to take his private language argument as attempting to refute the view that such things could be referents of our words. Now, it may well be that this view of sensations and sensation language is indefensible, and that Wittgenstein was right to oppose it. However, the problem with his argument is that it appears to rule out much more than this—including certain views that are considerably more plausible. For example, consider the following view: we begin with the sentences I have a headache, I have a pain in my lower back, and I feel a tingling sensation in my fingers. Wittgenstein’s objections aside, utterances of these sentences seem to refer to headaches, pains in one’s back, and tingling sensations in one’s fingers. How should we understand these apparent references? A natural way to do this is to take sensation reports to be reports of certain kinds of perception-like experiences. As with all perceptual experiences, these have contents that represent things as being certain ways. Although there are different alternatives regarding what these contents might be, one natural possibility is the following: When I truly say that I have a headache, I report a perception-like experience which represents my head as aching; when I truly say that I have a pain in my back, I report a perception-like experience which represents my back as hurting, or perhaps as being damaged; and when I truly say that I have a tingling sensation in my fingers, I report a perception-like experience which represents my fingers as tingling. In each of these cases, for an experience to represent o as Fing is for the experience to have the content that o F’s. The experiences that have these contents are the sensations; my headache is my perception-like experience that my head aches, the pain in my back is the perception-like experience that my back hurts, or is damaged; and the tingling sensation in my fingers is the perception-like experience that my fingers are tingling. These experiences are internal states of mine—roughly, states of my neurological system—that sensation They were taken to be entities of essentially the same type as G. E. Moore’s sense data, discussed in chapter 2 of volume 1.

18

56

WITTGENSTEIN ON LANGUAGE

words apply to and that are reported by my utterances of sensation sentences.19 Although I think that something like this view is plausible, I am not here asserting that it is ultimately the correct analysis of our sensation vocabulary.20 However, I am suggesting that the fact that it is inconsistent with the strong thesis of the private language argument should not be taken as refuting it. More strongly, I suggest that we do have such perception-like experiences, and that we could introduce sensation terminology more or less in accord with the above story (whether or not our already existing sensation vocabulary—‘headache’, ‘pain’, etc.— works exactly in this way). Since such terminology would refer to states and events that are, in a natural sense, internal to individual agents, the strong thesis of the private language argument should be regarded as incorrect.21 Recall that the strong thesis was derived from Wittgenstein’s general conception of meaning, which was itself subject to major objections. One of these concerned his failure to see how perceptual content might ground the contents of perceptual beliefs, and through them, the contents of at least some linguistic expressions—independent of any consideration of a larger linguistic community. We now see another problem 19 Quite apart from considerations involving rule following, or the private language argument, Wittgenstein would not, I think, have taken this alternative seriously. He would, I believe, have taken it as obvious that my headache can’t be identical with any state of my neurological system for the simple reason that if it were, then necessarily for my headache to exist would be for my neurological system to be in a certain state. But, he would have thought, if that were necessary, then it would also be apriori (and even analytic)—in which case, it would be knowable simply by reflection (and understanding what ‘headache’ means). Since it is not knowable in this way—since I can know that I have a headache without being in a position to conclude anything about the specific state of my neurological system—Wittgenstein would have taken this physicalistic construal of headaches and other sensations to be obviously incorrect, even absurd. As I will argue in the chapters on Ryle, this position and the reasoning behind it were characteristic features of the ordinary language school that grew out of a mistaken assimilation of necessity to aprioricity and analyticity. 20 There appears to be some equivocation in our ordinary talk about sensations like pain. For example, we talk of pain both as a kind of experience and as the thing experienced. Thought of in the former way, it is a kind of awareness with a representational content; thought of in the latter way, it is all or part of the content we apprehend. This, it seems to me, is a confusion, though a very common one. How, precisely that confusion is reflected in the meanings of our sensation words is a delicate question. 21 For illuminating discussions of pains and other sensations as perception-like experiences, see David Armstrong, Bodily Sensations (London: Routledge and Kegan Paul, 1962) and A Materialist Theory of the Mind (London: Routledge and Kegan Paul, 1968); George Pitcher, “Pain Perception,” The Philosophical Review 79 (1970): 368–93; Alex Byrne, “Intentionalism Defended,” The Philosophical Review 110 (2001): 199–240 (especially at 227–30); and Michael Thau, Consciousness and Cognition (New York: Oxford University Press, 2002), pp. 37–41.

THE PRIVATE LANGUAGE ARGUMENT

57

related to his neglect of perception. If sensations are, or are associated with, kinds of contentful perceptual experience, then the possibility of using language to refer to them should be roughly on a par with the possibility of using language to refer to perceptual experiences generally—like my experience of hearing a song, or seeing a movie. Surely, any thesis about meaning that rules out all such reference as illusory is too radical and insecurely supported to be accepted. As a result, I believe that Wittgenstein’s general conception of meaning should either be given up altogether, or modified and interpreted so that it doesn’t lead to the strong thesis of the private language argument. Although some such reformulation may prove possible, it is unclear exactly how it might go. Moreover, it seems likely that any such reinterpretation that made the account of meaning acceptable would render innocuous the conclusions about mentalistic and sensation language that could be drawn from it, thereby robbing Wittgenstein’s discussion of its dramatic philosophical import.

A Weaker, More Plausible, but Only Partially Wittgensteinian View of Sensation Language We can illustrate this point by thinking again about the weak thesis, this time asking not how it can be strengthened, but how it might be weakened so as it make it more plausible. Wittgenstein may well have a point when he says that the words of our public language—including our words for sensations—are tied up with publicly observable phenomena in some way. If they weren’t somehow tied up with such phenomena, it would be hard to explain how we succeed in teaching or learning them, so as to arrive at a common understanding. However, the private sensations linked by sensation words to publicly observable phenomena need not be regarded as necessary and inevitable concomitants of those phenomena. Rather, the publicly observable causes of the sensations and forms of behavior associated with them might be regarded as constituting important, but fallible, evidence for what pain-sentences attempt to describe. Thinking of things in this way would give us something like the following Even-Weaker Thesis, as opposed to the Weak and Strong theses discussed earlier. the even-weaker thesis No word in a public language is meaningful solely in virtue of standing for any kind of private sensation, idea, or internal

58

WITTGENSTEIN ON LANGUAGE

experience. Like all such words, any word referring to private sensations must also be associated with public criteria that provide others with a good, even if only fallible, basis to judge whether the speaker is using the word correctly. So when someone says he is in pain, this can be taken as reporting a private experience, so long as there are public criteria involving the behavior of the speaker (and the causes of the pain) that could, in principle, provide others with some reasonable, though not necessarily conclusive, grounds for assessing the correctness of his remark. There is, I think, a certain amount to be said for something like this thesis. When I speak of my headache, for example, I seem both to be referring to something private to me,22 and to be revealing something about my motivational structure. The headache, we are inclined to think, is an event or state located in my head, that only I can experience, which others can know only by observing its symptoms. This makes it plausible to suppose that part of what I do when I use the term “my headache” is to refer to a private sensation. However, in addition to referring to this sensation, I also seem to be describing it as a kind of pain, and hence as being the kind of thing which, all other things being equal, I wish to avoid. The idea of pain as something to be avoided is so strong that it is not out of the question to suppose that it is somehow built into the concept of pain itself. For example, imagine someone who described himself has “having a sharp, intense pain” of the sort people normally have when struck forcibly on the head with a heavy stick. Suppose further that the person went on to say that the pain he felt was, like most of his other pains, highly enjoyable, that it made him feel like laughing, that it was the kind of experience that he most desired purely for its own sake, and that the desire to feel these pains routinely motivated his actions. It would be hard to know how to respond to such a report. The story is certainly hard to accept—so hard that one might suspect the person of being insincere, delirious, or ignorant of what the word pain and related words really mean. Perhaps the story could be accepted as true, if some complex explanation were given of the person’s deviant neurological structure. Though not necessarily logically or conceptually private to me—if that is understood as ruling out the possibility that the headache I am now having is a particular physical event or state in my brain. In saying that the Even-Weaker Thesis is plausible, I intend to be understood in a way that allows for a physicalistic view of private sensations, ideas, and experiences.

22

THE PRIVATE LANGUAGE ARGUMENT

59

But one feels that this would have to be a highly unusual case. If this is right, it suggests that to describe a particular private sensation as a pain may be to identify it as being one instance of a type standard instances of which play certain well-known motivational roles in influencing other mental states of agents and, thereby, their behavior. It is this mediated connection to behavior that provides others with a reasonable, though fallible, basis for judging the sincerity or insincerity, as well as the correctness or incorrectness, of particular pain reports.23 This seems to be a rather plausible view. However, although it bears some similarities to Wittgenstein’s observations in the Investigations, it is not as revisionary, or philosophically far-reaching, as the view he seemed to be aiming at. Perhaps a case could be made that this conception of the claims made by sensation sentences—according to which publicly available phenomena provide fallible evidence for the existence of various mental states, including pains and other sensations—is strong enough to shift the burden of proof onto the philosophical skeptic, so that in ordinary circumstances we are justified in making claims about the sensations of others, unless the skeptic can provide a positive reason for thinking that those claims are false. If such a result could be established, it would be significant. However, the view need not arise from any overarching conception of meaning or language use in general, and it seems to be only indirectly related to the position developed by Wittgenstein.24

Conclusion: The Legacies of the Investigations This brings our discussion of the Investigations to a close. If the legacy of the work appears to be mixed, and its ultimate significance uncertain, that, I think, is as it should be. As will become evident in Parts 2 For a good statement and defense of this sort of view, see David Lewis’s classic article, “Mad Pain and Martian Pain,” reprinted in his Philosophical Papers, vol. 1 (New York and Oxford: Oxford University Press, 1983). 24 Since Wittgenstein did not state his position clearly, this is, of course, a matter of interpretation. Some would say that he should be taken as advocating something closer to the EvenWeaker Thesis than to my strong or weak theses. This difference in interpretation is related to broader differences, alluded to in n. 2, about whether Wittgenstein’s general conception of meaning should be seen as tying the truth or falsity of a speaker’s applications of a word to standards of community agreement, or whether such standards are relevant only to the question of whether one is justified in asserting that the speaker’s applications of the word are correct. Since my interpretation of Wittgenstein’s general conception of meaning has him taking a stand on truth conditions (rather than simply on assertability conditions), my interpretation of his private language argument is correspondingly strong. 23

60

WITTGENSTEIN ON LANGUAGE

and 3, the thought of the later Wittgenstein had an enormous impact on the philosophers of the ordinary language school who followed him—deeply influencing their views about language, analysis, and the nature of philosophy, as well as guiding their proposed solutions to central philosophical problems, including those posed by the nature of the mental, as well as various forms of skepticism. However, this influence also carried a historical price; when the ordinary language approach foundered on intractable problems, the crack-up could not but reflect on its origins. Looking back at the Investigations, there is no doubt that the work still retains much of its fascination. Nevertheless, it is hard, from today’s perspective, to ignore the flaws at its core—including (i) the identification of the philosophical with (a subset of ) the necessary and the apriori, (ii) the further identification of these with the analytic, (iii) the conception of meaning as essentially transparent, (iv) the closely related deflationary conception of philosophy and philosophical analysis, and (v) the unwaveringly informal and antitheoretical approach to the study of meaning. From today’s perspective, all of these appear highly questionable at best, and misguided at worst. In other, related areas, Wittgenstein’s legacies are more positive, if still somewhat mixed. For example, although descriptivist and referential theories of meaning are flourishing today despite his critique, many proponents of these theories have tried to incorporate certain of his undeniable insights.25 In addition, considerable interest continues to be shown in his discussion of rule following, which is seen as posing a serious challenge to attempts to explain how intentional facts—about the contents of both linguistic expressions and mental states—relate to the nonintentional facts on which, it seems, they must depend.26 This, of course, is not all. In my view, Wittgenstein’s most valuable insights and lasting contributions include (i) his emphasis on the social nature of language and linguistic meaning, (ii) his insistence that much of our thought is conceptually dependent on our mastery of language, and (iii) his implicit recognition, in consequence of (i) and (ii), that the contents of many of our thoughts are in part determined socially, See, for example, my “Semantics and Semantic Competence,” cited in chapter 1. See Saul Kripke, Wittgenstein on Rules and Private Language; Crispin Wright, “Kripke’s Account of the Argument against Private Language,” The Journal of Philosophy 81, 12 (1984): 759–78; Paul Boghossian, “The Rule-Following Considerations,” Mind 98 (1989): 507–49; George Wilson, “Kripke on Wittgenstein and Normativity,” Midwest Studies in Philosophy 19 (1994): 366–90; and Scott Soames, “Skepticism about Meaning: Indeterminacy, Normativity, and the Rule-Following Paradox.”

25 26

THE PRIVATE LANGUAGE ARGUMENT

61

rather than entirely individualistically. All of these may be seen as important precursors of much contemporary thinking. Finally, there are many areas of Wittgenstein’s later thought which we have not been able to touch on at all—such as his philosophy of logic and mathematics—that have continued to intrigue and inspire philosophers up to the present day.

SUGGESTED FURTHER READING FOR PART ONE

Main Primary Sources Discussed Wittgenstein, Ludwig. Philosophical Investigations. Translated by G.E.M. Anscombe, 3d edition. New York: Macmillan, 1958.

Additional Primary Sources Wittgenstein, Ludwig. Tractatus Logico-Philosophicus. Translated by C. K. Ogden. Mineola, NY: Dover, 1999; originally published in English in 1922 by Routledge. See also the translation by David Pears and Brian McGuinness (London: Routledge), 1974.

Additional Recommended Reading Armstrong, David. Bodily Sensations. London: Routledge and Kegan Paul, 1962. ———. A Materialist Theory of the Mind. London: Routledge and Kegan Paul, 1968. Ayer, A. J. “Can There Be a Private Language?” In George Pitcher, ed., Wittgenstein (Garden City, NY: Anchor Books, Doubleday, 1966), 251–66; originally published in Proceedings of the Aristotelian Society, supplementary vol. 28, 1954. Boghossian, Paul. “The Rule-Following Considerations.” Mind 98 (1989): 507–49. Byrne, Alex. “Intentionalism Defended.” Philosophical Review 110 (2001): 199–240. Fogelin, Robert. Wittgenstein, 2d ed. London and New York: Routledge, 1987 (first ed., 1976). Grice, Paul. Chapters 5, 6, and 14 (originally published as “Utterer’s Meaning and Intentions,” “Utterer’s Meaning, Sentence-Meaning, and Word-Meaning,” and “Meaning,” respectively) of his Studies in the Way of Words. Cambridge, MA and London: Harvard University Press, 1989.

SUGGESTED FURTHER READING

63

Kripke, Saul. Wittgenstein on Rules and Private Language. Cambridge, MA: Harvard University Press, 1982. Lewis, David. “Mad Pain and Martian Pain.” Reprinted in his Philosophical Papers, vol. 1 (New York and Oxford: Oxford University Press, 1983); originally published in 1980. Pitcher, George. “Pain Perception.” Philosophical Review 79 (1970): 368–93. Salmon, Nathan. “Existence.” In J. Tomberlin, ed., Philosophical Perspectives, vol. 1, Metaphysics (Atascadero, CA: Ridgeview, 1987), 49–108. ———. “Nonexistence.” Noûs 32 (1998): 277–319. Searle, John. “Proper Names.” Mind 67 (1958): 166–73. Strawson, Peter. “Review of Philosophical Investigations.” In George Pitcher, ed., Wittgenstein (Garden City, NY: Anchor Books, Doubleday, 1966), 22–64, originally published in Mind 63 (1954): 70–99. Soames, Scott. Chapter 3, Beyond Rigidity. New York: Oxford University Press, 2002. ———. “Facts, Truth Conditions, and the Skeptical Solution to the Rule-Following Paradox.” Philosophical Perspectives 12, Language, Mind, and Ontology (1998): 313–48. ———. “Semantics and Semantic Competence.” Philosophical Perspectives vol. 3, Philosophy of Mind and Action Theory, (1989): 575–96. ———. “Skepticism about Meaning: Indeterminacy, Normativity, and the Rule-Following Paradox.” In Ali Kazmi, ed., Meaning and Reference, Canadian Journal of Philosophy, supplementary vol. 23, 1998, 211–49. Thau, Michael. Consciousness and Cognition. New York: Oxford University Press, 2002. Wilson, George. “Kripke on Wittgenstein and Normativity.” Midwest Studies in Philosophy 19 (1994): 366–90. Wright, Crispin. “Kripke’s Account of the Argument against Private Language.” Journal of Philosophy 81, 12 (1984): 759–78.

This page intentionally left blank

PA R T T W O CLASSICS OF ORDINARY LANGUAGE PHILOSOPHY: TRUTH, GOODNESS, THE MIND, AND ANALYSIS

This page intentionally left blank

CHAPTER 3 RYLE’S DILEMMAS chapter outline 1. Philosophical Problems as Dilemmas 2. Case Studies in Ordinary Language Analysis Fatalism: the difference between Russellian logical analysis and ordinary language analysis The world of science and the world of common sense: supervenience and the problem for Ryle posed by the difference between necessary and apriori consequence Perception: Skepticism about perceptual knowledge; Ryle’s unsuccessful attempts to show that radical skepticism is conceptually incoherent The nature of perception: how limitations imposed by Ryle’s methodological views inhibited his discussions of the essential nature of perception, and of the question of whether seeing something is (partly) an internal state or process

3. Lessons

In this chapter and the next we will discuss Gilbert Ryle, who was one of the leading exponents of the “ordinary language” school of philosophy in postwar Britain. Ryle, a prominent philosopher at Oxford for more than thirty years, became well known in 1931 for his article “Systematically Misleading Expressions,” in which he argued that the task of philosophy is to expose and correct linguistic confusions that have spawned philosophical problems.1 His main work, The Concept of Mind, was published in 1949, and his book Dilemmas, which was also quite influential (having been reprinted seventeen times since its initial publication), originated as a series of lectures at Cambridge in 1953.2 Though his style of writing and thinking was very different from Wittgenstein’s, the philosophical views of the two men overlapped considerably. While Ryle, “Systematically Misleading Expressions,” Proceedings of the Aristotelian Society, 1931–32. 2 Ryle, The Concept of Mind (New York: Barnes and Noble, 1949); Ryle, Dilemmas (Cambridge: Cambridge University Press, 1953; most recent reprint 1998). 1

68

ORDINARY LANGUAGE PHILOSOPHY

influenced by Wittgenstein, he should not, I think, be considered a follower. Rather, he was, along with Wittgenstein, J. L. Austin, and Paul Grice, one of the prime movers in postwar philosophy in England. In this chapter, we will discuss Dilemmas, which provides a good introduction to his overall vision of philosophy.

Philosophical Problems and Their Resolution In chapter 1 Ryle introduces the idea of a dilemma that results from apparently obvious theories or commonsense platitudes coming into seeming conflict with one another. In these cases, a view that seems perfectly fine in dealing with a restricted set of questions comes to seem incompatible with another view that seems fine when confined to a different domain. When this happens, we find ourselves in the uncomfortable position of seeming to be unable to jointly maintain a pair of views, each of which appears correct when considered on its own. Ryle believes that in many cases the apparent conflict is an illusion that can be dispelled by philosophical analysis. Although in chapter 1 he doesn’t discuss any particular dilemma in detail, he does give examples of what he has in mind. Here is one: The neuro-physiologist who is studying the mechanism of perception, like the physiologist who is studying the mechanism of digestion or reproduction, bases his theories upon the most solid kind of evidence that his work in the labouratory can provide, namely, upon what he and his collaborators and assistants can see with the naked or instrumentally assisted eye, and upon what they can hear, say, from the Geiger counter. Yet the theory of perception at which he arrives seems constitutionally to entail that there is an unbridgeable crevasse between what people, including himself, see or hear and what is really there—a crevasse so wide that he has apparently and can have no labouratory evidence that there exists even any correlation between what we perceive and what is really there. If his theory is true, then everyone is systematically debarred from perceiving the physical and physiological properties of things; and yet his theories are based on the very best experimental and observational evidence about the physical and physiological properties of such things as ear-drums and nerve-fibres. While at work in the labouratory he makes the best possible use of his eyes and

RYLE’S DILEMMAS

69

ears; while writing up his results he has to deliver the severest possible censure on these sham witnesses. He is sure that what they tell us can never be anything like the truth just because what they told him in the labouratory was of the highest reliability. From one point of view, which is that of scientists and laymen alike while actually exploring the world, we find out what is there by perceiving. From the other point of view, that of the inquirer into the mechanism of perception, what we perceive never coincides with what is in the world.3 In this passage, Ryle tells us that the theory of the neurophysiologist seems to conflict with the platitude that we perceive things in the physical world around us, and thereby acquire accurate and reliable information about them. As he emphasizes, this conflict is paradoxical, since we would have no evidence for the theory if we couldn’t accept ordinary claims about what we perceive. One feels, in this case, that any apparent conflict must be illusory. Somehow, concepts and ideas which have a legitimate application in one domain must have been stretched and misused so as to foster the illusion of conflict with truisms about another domain. In chapter 7, on perception, Ryle tries to demonstrate this in detail. Another example of a dilemma from chapter 1 involves free will and moral responsibility. Consider, next, a very different sort of dilemma. Everyone knows that unless a child is properly brought up he will probably not behave properly when grown up; and if he is properly brought up he is quite likely to behave properly when grown up. Everyone knows, too, that though certain actions of lunatics, epileptics, kleptomaniacs and drowning men are regrettable, they are not reprehensible or, of course, commendable either, where similar actions of a normal adult in normal situations are both regrettable and reprehensible. Yet if a person’s bad conduct reflects his bad upbringing, it seems to follow that not he but his parents should be blamed—and then, of course, in their turn, his grandparents, his great-grandparents and in the end nobody at all. We feel quite sure both that a person can be made moral and that he cannot be made moral; and yet that both cannot be true. When considering the parents’ duties, we have no doubt that they are to blame if they do not mould their 3

Ryle, Dilemmas, p. 2.

70

ORDINARY LANGUAGE PHILOSOPHY

son’s conduct, feelings, and thoughts. When considering the son’s behaviour we have no doubt that he and not they should be blamed for some of the things that he does. Our answer to the one problem seems to rule out our answer to the other, and then at second remove to rule itself out too.4 Here Ryle discusses the connection between commonplace views about the causation of behavior and ascriptions of moral responsibility. Commonplace views about the positive effects of moral training are predicated on the idea that a person’s background can play an important causal role in shaping his character and influencing his actions. Also, certain obvious facts about instances in which an agent’s acts are caused or dictated by factors beyond his control lead us to think that in at least some such cases the agent is not responsible for his acts. But if we put these two obvious truths together, and generalize them in what may initially seem to be a plausible way, we get the view that all acts are ultimately caused by factors external to an agent, over which he had no control, and that, because of this, no one is ever morally responsible for anything. So, we arrive at a view that conflicts with the commonsense platitude that people are often morally responsible for what they do. Ryle has no tendency whatsoever to deny this platitude. Nor does he reject either the observation that moral training can causally shape later behavior, or the claim that sometimes people are not responsible for their behavior because they are unable to act otherwise. Consequently, he thinks that there must be something wrong with the ways in which these commonsense truisms are generalized and extended into a view that conflicts with ordinary ascriptions of moral responsibility. He doesn’t analyze this particular case in Dilemmas, so he doesn’t tell us precisely what fallacy has been committed. But he has no doubt that there must be such a fallacy. This attitude toward philosophical problems is very much in keeping both with G. E. Moore’s view that commonsense certainties cannot be overturned by philosophical doubt, and with Wittgenstein’s dictum that philosophy must leave everything as it is, that it must only describe and never explain. According to both Ryle and Wittgenstein, philosophical problems are dissolved not by establishing surprising and informative theories, but by assembling philosophical reminders of things we already know. Dilemmas is a collection of case studies illustrating this approach to philosophy. 4

Ibid., p. 4.

RYLE’S DILEMMAS

71

Fatalism The Problem

Chapter 2 is concerned with fatalism—the view that everything that happens does so inevitably. The fatalist reasons as follows: Suppose that something is going to happen tomorrow. Then it is true now that it is going to happen tomorrow. But if it is already true that it is going to happen tomorrow, then it can’t be that it won’t happen tomorrow. Since that is so, nothing can possibly stop it from happening; thus, it is inevitable. Since the same reasoning applies to every event, everything is inevitable. Let us focus on one particular case. Suppose it is true now that I will lecture on Ryle tomorrow. (You supply the date.) Then it was true last week, last year, or even 100 years ago, that I would lecture on Ryle tomorrow. Moreover, if it was true all those years ago, then necessarily I will lecture on Ryle tomorrow. But since it is necessary that I give the lecture, it is impossible for me not to give it; no matter what I or anyone else does in the intervening time, there will be a lecture on Ryle tomorrow. So, on that day an event of lecturing will occur which could not have been prevented. We all know this argument is unsound. Where does it go wrong? Ryle tries to show us, and in so doing to illustrate his conception of philosophical analysis. However, in my opinion, he misses the main problem with the argument. So I will proceed as follows: first I will diagnose what seems to me to be the main confusion in the fatalist’s argument; next I will talk about Ryle’s treatment of the case; finally, I will try to draw some general lessons about Ryle’s conception of philosophy from the strengths and weaknesses of his discussion of this particular problem. The Solution to the Problem

As I see it, the confusion can be dispelled quite easily. The argument starts out with the premise A. A. It was true 100 years ago that I would lecture on Ryle tomorrow. The next premise in the argument is expressed in English by the words if it was true 100 years ago that I would lecture on Ryle tomorrow, then, necessarily, I will lecture on Ryle then. This sentence is ambiguous. In

72

ORDINARY LANGUAGE PHILOSOPHY

general, an English sentence If P, then, necessarily, Q can express either the same thing as It is a necessary truth that if P, then Q , or the same thing as If P, then it is a necessary truth that Q. On the former interpretation, the next premise in the argument for fatalism is: B. It is a necessary truth that if it was true 100 years ago that I would lecture on Ryle tomorrow, then I will lecture on Ryle then. On this interpretation the premise is true, but the next step, C, doesn’t follow from A and B. C. It is a necessary truth that I will lecture on Ryle tomorrow. In order to derive C, we would have to combine B not with A but with the claim that A is a necessary truth, and we don’t have that. So on this interpretation of the ambiguous premise, the conclusion of the argument does not logically follow from the premises, and fatalism is not established. Of course, the ambiguous premise can be understood another way as well—namely as B⬘. B⬘. If it was true 100 years ago that I would lecture on Ryle tomorrow, then it is a necessary truth that I will lecture on Ryle then. The conclusion C does follow from A and B⬘. However, unlike B, B⬘ is not an obvious truth. In fact, there is no reason to accept it unless one already accepts fatalism in advance. But if that is the case, then, one cannot use it in the argument for fatalism. Thus, on this interpretation, the argument begs the question. To sum up, the problem with the argument for fatalism is that it relies on an ambiguous premise. On one way of disambiguating the premise, the argument is invalid, since the conclusion is not a logical consequence of the premises. On the other way of disambiguating the premise, the argument assumes what it is supposed to prove, and so begs the question. Before moving on to Ryle’s discussion, I want to extend the diagnosis I have given. There is another argument, very much like the argument for fatalism, that suffers from essentially the same defect. Fatalism is similar to another view, discussed at the end of chapter 5 of volume 1, that might be called rampant essentialism. Whereas the fatalist says that everything that happens could not have failed to happen, the rampant essentialist says that every property that an object has is an essential property of it—i.e., it is a property that the object could not have failed to have.

RYLE’S DILEMMAS

73

The argument for this conclusion is parallel to the argument for fatalism. We begin with the assumption that I have some arbitrary property—say, the property of wearing socks today.5 i. Ps We then appeal to an obvious necessary truth. ii. Necessarily, if Ps, then for any x, if ~Px, then x ⫽ s. (Necessarily, if I am wearing socks today, then anyone not wearing socks today isn’t me.) Next we appeal to what seems to be nothing more than a paraphrase of this obvious truth. iii. Thus if Ps, then necessarily, for any x, if ~Px, then x ⫽ s. (If I am wearing socks today, then, necessarily, anyone not wearing socks today isn’t me.) Finally, we infer (iv) from (i) and (iii). iv. It is a necessary truth that for any object whatsoever, if that object lacks the property of being P, then that object isn’t s. In other words, it is impossible for any object to lack this property while still being (identical with) s. So it is impossible for s to lack the property of being P while still being (identical with) s. Since it is impossible for s not to be identical with s, it is impossible for s not to have the property of being P— i.e., this property is one that s could not possibly have lacked. (The property of wearing socks today is a property I could not have lacked.) This argument fails for the same reason that the argument for fatalism fails. Premises (i) and (ii) are true. However, (iii) is ambiguous. On one reading its logical form is the same as the logical form of (ii)— namely, (Lii). Lii. It is a necessary truth that: if Ps, then for any x, if ~Px, then x ⫽ s. On this reading, the conclusion (iv) does not logically follow from (iii) plus (i), so the argument is invalid. On the other reading of (iii), its logical form is (Liii), which is not equivalent to (Lii).

5

In the argument that follows, I use ‘P’ as a schematic letter.

74

ORDINARY LANGUAGE PHILOSOPHY

Liii. If Ps, then it is a necessary truth that: for any x, if ~Px, then x ⫽ s. On this reading, (iii) states that if I have the property of being P, then the claim that anything that lacks this property is not me is a necessary truth. On this reading, (iv) follows from (i) and (iii), but there is a problem establishing (iii). It doesn’t logically follow from (ii). In fact, there is no reason to accept (iii)—on the reading (Liii)—unless we already accept rampant essentialism in advance. What it states is that if I have a certain property, then I have it essentially. Since this is what the argument is supposed to establish, we cannot assume it as a premise without begging the question. To sum up, the argument for rampant essentialism relies on an ambiguous step—(iii). On one way of disambiguating (iii), the conclusion (iv) does not follow from (i) and (iii). On the other way of disambiguating (iii), (iv) follows from (i) and (iii), but (iii) does not follow from (ii). Moreover, to simply introduce (iii) as an independent premise would be to assume what the argument is supposed to prove. So the argument is either invalid or question-begging, exactly as the argument for fatalism was. Here we have a pair of philosophical problems that can be dissolved by linguistic analysis. In that sense, these problems fit very well Ryle’s and Wittgenstein’s paradigm of philosophical problems as arising from confusions about language. However, these problems don’t fit very well the new paradigms of linguistic analysis favored by Wittgenstein and Ryle. Instead, they fit the old Russellian paradigm in which we look for the true logical forms of sentences. The basic linguistic point in both cases is that English sentences If P, then, necessarily, Q are ambiguous between two different logical forms: It is a necessary truth that if P, then Q and If P, then it is a necessary truth that Q.6 Once we see this, it is clear how the arguments for fatalism and for rampant essentialism fail. This does not, of course, prove that the fatalist or the rampant essentialist theses are false; but it does undermine our reason to believe them. Since the theses are diametrically opposed to what we ordinarily and pre-philosophically think, the effect of undermining the arguments is to allow us to retain our implicit rejection of fatalism and rampant essentialism with a clear conscience. There is a presumption here worth noting. If a particular claim is something that seems utterly obvious to us in our ordinary thinking, 6

‘P’ and ‘Q’ are here used as metalinguistic variables over sentences.

RYLE’S DILEMMAS

75

then if that claim is questioned philosophically, we are justified in continuing to accept it unless a compelling philosophical argument is given against it. One might put this dramatically by saying that not all propositions are philosophically equal. If p is a seemingly obvious bit of common sense, or a well-established part of ordinary science— of which we are already strongly convinced—then the burden of proof is on any philosophers who want to challenge p. If they can’t give a good reason to believe the negation of p, then we are justified in continuing to believe p. This general attitude was explicitly defended by G. E. Moore in the early part of the century, and it carried over to all ordinary language philosophers, including Wittgenstein and Ryle.

Ryle’s Discussion of Fatalism as an Illustration of His Philosophical Method

Our discussion of fatalism and rampant essentialism conforms both to this Moorean attitude and to the traditional, Russellian conception of logico-linguistic analysis in analytic philosophy very well. However, it does not fit the new paradigms of linguistic analysis favored by Wittgenstein and Ryle. This point can be appreciated more fully by taking a look at what Ryle says about the argument for fatalism. His statement of the argument is like mine in all relevant respects. At a certain moment yesterday evening I coughed and at a certain moment yesterday evening I went to bed. It was therefore true on Saturday that on Sunday I would cough at one moment and go to bed at the other. Indeed, it was true a thousand years ago that at certain moments on a certain Sunday a thousand years later I should cough and go to bed. But if it was true beforehand— forever beforehand—that I was to cough and go to bed at those two moments on Sunday, 25 January 1953, then it was impossible for me not to do so. There would be a contradiction in the joint assertion that it was true that I would do something at a certain time and that I did not do it. This argument is perfectly general. . . . So nothing that does occur could have been helped and nothing that has not actually been done could possibly have been done.7 7

Dilemmas, p. 15, my emphasis.

76

ORDINARY LANGUAGE PHILOSOPHY

Note the two emphasized sentences. The first is of the form If S, then it was impossible that ~R, which is equivalent to the corresponding sentence of the form If S, then necessarily R. Since these sentences are ambiguous, the crucial ambiguity pointed out in the previous section appears at this point in Ryle’s statement of the argument. The second emphasized sentence is used to establish the truth of the first. It is of the form It is a contradiction that S and ~R, which we may equate with It is impossible that S and ~R. This is equivalent to one of the two readings of the first emphasized sentence: namely, the reading that is equivalent to It is a necessary truth that if S, then R. On this reading, the first emphasized sentence is clearly true; however, the fatalist’s conclusion does not follow logically from it, and cannot be derived without assuming what was supposed to be proved. Although the conclusion can be derived from the second reading of the first emphasized sentence—in which it is equivalent to If S, then it is a necessary truth that R—on that reading the first emphasized sentence does not follow from the second, and cannot be established without begging the question. Hence, Ryle’s statement of the fatalist’s argument suffers from precisely the flaw that we have already diagnosed. Ryle does not point this out. Instead, on page 16 he asks what is wrong with the argument, and immediately launches into a discussion of how its first premise might be misunderstood. He says that we have a tendency of taking the premise that 100 years ago it was true that I would lecture on Ryle tomorrow as implying that 100 years ago it was known that I would lecture on Ryle tomorrow. In fact this is no part of the premise. So why, according to Ryle, do we have a tendency to misinterpret the premise in this way? He gives his answer in the following passage: Yet even when we try hard to bear this point in mind, it is very easy inadvertently to reinterpret this initial principle into the supposition that before the thing happened it was known by someone that it was booked to happen. For there is something intolerably vacuous in the idea of eternal but unsupported pre-existence of truths in the future tense. When we say ‘a thousand years ago it was true that I should now be saying what I am’, it is so difficult to give any body to this ‘it’ of which we say that it was then true, that we unwittingly fill it out with the familiar body of an expectation which someone once entertained, or of a piece of foreknowledge which someone once possessed. Yet to do this is to

RYLE’S DILEMMAS

77

convert a principle which was worrying because, in a way, totally truistic, into a supposition which is unworrying because quasi-historical, entirely without evidence and most likely just false.8 Although the gist of these remarks is mainly unobjectionable, there are two problems with them. First, at best Ryle has pointed out why some people are inclined to reject the first premise of the fatalist’s argument. But that doesn’t explain the attraction of the argument to those who feel it. Nor does it tell us what is really wrong with the argument—since in fact the first premise is fine. Thus, Ryle’s comments about the tendency to confuse truth with knowledge don’t help us locate the main difficulty. Second, Ryle says: For there is something intolerably vacuous in the idea of eternal but unsupported pre-existence of truths in the future tense. When we say ‘a thousand years ago it was true that I should now be saying what I am’, it is so difficult to give any body to this ‘it’ of which we say that it was then true, that we unwittingly fill it out with the familiar body of an expectation which someone once entertained, or of a piece of foreknowledge which someone once possessed. One of the things Ryle wants to know is, what is this it that is supposed to have been true? The natural answer is that it is a proposition about what was then the future.9 It, that proposition, was true at some past time. What is intolerably vacuous about this? When we say that in the past, before any life existed on earth, the oceans were formed, what we are saying is, roughly, something that is true iff the proposition that the oceans were formed was true at some time t in the past prior to the existence of life on earth. That seems pretty obvious. There is no need to bring in any knower existing before life on earth to make sense of this claim. The idea that a proposition that we now express can be evaluated at different times, or under different counterfactual circumstances, without presupposing a knower, is not “intolerably vacuous”; it is commonplace. Similarly commonplace is the idea that there are things that are true now that no one now knows. Presumably, for some particular n it is true that there are now exactly n coins in my two coin jars at home, even though no one knows it; and if that is true now without being known, then it will remain true forever that at precisely this time there are exactly n coins in those jars, even though it is unlikely that anyone will ever know it. Ibid., p. 17. In general, It was true that S seems to be a variant of that S was true, which in turn seems to have essentially the same status as the proposition that S was true.

8 9

78

ORDINARY LANGUAGE PHILOSOPHY

Since there is no general problem with truths that are unknown, this isn’t the difficulty with the fatalist’s argument. On page 18 Ryle continues his discussion of the argument with some remarks about truth and falsity, as well as about correctness and incorrectness. Regarding truth, he says that when we call something true, there is often a suggestion that believing it would be justified. Because of this suggestion, he thinks we are reluctant to call some guesses true, even if they turn out later to be correct. On this basis, he says, we should rewrite the fatalist’s first premise to say that in the past if anyone had guessed that tomorrow is a day that I would lecture on Ryle, then that guess would have been correct. He then casts doubt even on this premise, suggesting that guesses only come to have the property of being correct after the event they have predicted takes place; they aren’t correct guesses before the event that makes them correct happens. I doubt that this is so; however, it is not particularly important in any case. If it is relevant to the fatalist’s argument at all, it tends to cast doubt on the first premise, when formulated using the notion of a correct guess. However, as I have said, there is a perfectly good sense in which the fatalist’s first premise is true, and can be accepted as true without making any claim about what guesses would have been correct, or what propositions one would have been justified in believing 100 years ago. So even if Ryle is right that the first premise is false when reinterpreted along the lines he suggests, this is irrelevant to the status of the argument when the first premise is left as stated by the fatalist. In the middle of the essay (p. 21 and following) Ryle turns from interpreting the fatalist’s premises to interpreting his conclusion—namely, that since in the past it was true that I would lecture on Ryle in the future, necessarily I will lecture on Ryle in the future. He asks what this sense of necessity is. He thinks that we may be inclined to treat it as causal necessity. That is, he thinks that we may be tempted to interpret the fatalist’s conclusion as saying that the fact that in the past it was true that I would lecture on Ryle in the future causes my future lecture on Ryle, or makes it happen. He then tries to show that, interpreted in this way, the conclusion is unwarranted. This is shown by the fact that we could construct a parallel argument involving the future rather than the past. The argument involving the future starts with the premise that ten years from now (in the future) it will be true that tomorrow (you supply the date) I lectured on Ryle. The argument then moves to the premise that if this is so in the future, then it is impossible for me not to lecture on Ryle tomorrow. Ryle points out that no one would conclude that its

RYLE’S DILEMMAS

79

being true ten years from now is what causes me to give the lecture tomorrow.10 Since the argument is parallel to the fatalist’s argument, no one should conclude from the fatalist’s premises that its being true in the past that I would lecture on Ryle tomorrow causes me to do so. Ryle is right about this. He is also right in pointing out that to read the fatalist’s conclusion as making a causal claim is to confuse the causal relation, which holds between events, with the relation of logical or conceptual consequence, which holds between propositions. Dispelling these confusions is all to the good, and is an important step in dispelling the air of paradox engendered by the argument. Nevertheless, Ryle still hasn’t located what is arguably its central confusion—a confusion that it shares with the argument for rampant essentialism, and that it retains even when its conclusion is not read causally. This failure is not accidental. Ryle has rejected the old paradigm of analysis as involving the identification of precisely formulated logical forms. Instead of giving us a single crucial point in discussing the fatalist’s argument, he ends up giving us something that at times seems like a sort of fog of analysis—a collection of related points no one of which is absolutely fundamental. In fact, he seems to regard this way of approaching philosophical problems as a virtue. Now for some general morals which can be drawn from the existence of this dilemma and from attempts to resolve it. It arose out of two seemingly innocent and unquestionable propositions, propositions which are so well embedded in what I may vaguely call ‘common knowledge’ that we should hardly wish to give them the grand title of ‘theories’. These two propositions were, first, that some statements in the future tense are or come true, and, second, that we often can and sometimes should secure that certain things do happen and that certain other things do not happen. Neither of these innocent-seeming propositions is as yet a philosopher’s speculation, or even a scientist’s hypothesis or a theologian’s doctrine. They are just platitudes. . . .11 The two platitudes from which the trouble arose are not in direct conflict with one another. It is real or seeming deductions from the one which quarrel with the other, or else with real or seeming deductions from it. . . .12 Dilemmas, pp. 20–21. Ibid., p. 29. 12 Ibid., p. 30. 10 11

80

ORDINARY LANGUAGE PHILOSOPHY

I have indicated that the quandary, though relatively simple, does depend on a smallish number of concepts, namely, in the first instance, upon those of event, before and after, truth, necessity, cause, prevention, fault, and responsibility. Now there is not just one of these concepts which is the logical trouble-maker. The trouble arises out of the interplay between all of them. The litigation between the two initial platitudes involves a whole web of conflicting interests. There is not just a single recalcitrant knot in the middle of one of the concepts involved. All the strings between all of them are implicated in the one tangle. I mention this point because some people have got the idea from some of the professions though not, I think, the practices of philosophers, that doing philosophy consists or should consist of untying logical knots one at a time. . . .13 I have no special objection to or any special liking for the fashion of describing as ‘analysis’ the sort or sorts of conceptual examination which constitute philosophizing. But the idea is totally false that this examination is a sort of garage inspection of one conceptual vehicle at a time. On the contrary, to put it dogmatically, it is always a traffic inspector’s examination of a conceptual traffic-block, involving at least two streams of vehicles hailing from the theories, or points of view or platitudes which are at cross-purposes with one another.14 Ryle follows this last passage with a comparison of the conceptual examination of a concept with the description of the role of a wicket keeper in cricket. The crucial point is that one can’t describe the position of wicket keeper without describing how that position fits in with all the other positions. The analogy is supposed to be that one cannot analyze a concept without tracing its intricate connections with all the members of the family of concepts of which it is a part. Personally, I have no axe to grind about what it takes to analyze a concept. Very likely, there are different sorts of cases. It may well be that sometimes what we want from an analysis is the tracing of the sort of intricate web of conceptual relations in which Ryle delights. But there is little reason for thinking that this is always so—at least, if analysis is construed as whatever it is that philosophers do to solve their problems. What strikes me as worrisome is Ryle’s tendency to use 13 14

Ibid., p. 31. Ibid., p. 32.

RYLE’S DILEMMAS

81

the web metaphor as a rationale for rejecting the old, Russellian conception of analysis, with its emphasis on precisely formulated logical forms, and replacing it with methodology which, in some cases, may degenerate into a recipe for generating a conceptual fog. It is all well and good to recognize that sometimes the concepts philosophers deal with will be vague, imprecise, and open-ended, with close conceptual connections to other concepts of the same sort. We do have to be able to deal with such cases—perhaps along the lines Ryle suggests. What is not good is a prior ideological commitment to blurred edges, indirectness, and an unwillingness to separate tangential from central issues. Sometimes Ryle and other ordinary language philosophers seem to go too far in this direction, substituting one confining orthodoxy about analysis for another. When this happens, central philosophical points get missed, as in the elementary example of fatalism.

The World of Science and the Everyday World In chapter 5, Ryle discusses the tension that sometimes can be felt between what one might call the world of science—in particular, the world of atomic and subatomic physics—on the one hand, and the world of common sense on the other. What is said about atoms, nuclei, electrons, and so on, is very different from what we are all pre-scientifically inclined to say and think about ordinary objects like tables and chairs. Much that is said in physics is very surprising from an ordinary, commonsense point of view. In addition, the physicist may tell you that everything that we talk about in ordinary life is in fact nothing more than a fantastically complex collection of atomic and subatomic particles, which are correctly described only in theoretical physics. Every chair, every table, and every property of every chair and table, is in some sense nothing more than the fundamental particles of matter that make it up, and the scientific properties those particles have. So far, so good. The problem is that all of this may lead one to think that there is a conflict between the world conceived of in the way it is presented by our best physical theory, and the world conceived of as we ordinarily do. And if there is a conflict, then both can’t be correct; to accept one is to be committed to recognizing that the other is not really correct. And that is very uncomfortable, because we are loath to give up either one. Ryle’s reaction to this is that the appearance of a conflict between physics and common sense is an illusion. He makes the elementary point

82

ORDINARY LANGUAGE PHILOSOPHY

that although every object falls within the scope of physics, physics doesn’t purport to tell us everything there is to know about those objects. So one could take physics to be entirely true and correct, and one could recognize that ordinary objects like tables and chairs are among the objects in its domain, while also holding that there are many other truths about those ordinary objects which could not be deduced, even in principle, from a correct physics. Physics ascribes one set of properties to things; common sense ascribes a different set of properties to the same things. Since the properties are different, we don’t have to choose between them. Both ascriptions can be correct. Although Ryle is clearly right about this, he doesn’t go far enough. It is natural to suppose that all the ordinary truths supervene on the physical truths. To say this is to say that it would be impossible for there to be a change in the ordinary truths, unless there were some corresponding change in the physical truths. This is just to say that given the physical truths, the ordinary truths couldn’t be any different than they are; in other words, the ordinary truths are necessary consequences of the physical truths. Ryle tells us, correctly, that the ordinary truths can’t be derived logically, or by any apriori reasoning, from the physical truths. Thus, the ordinary truths are not logical or apriori consequences of the physical truths. But now we seem to have arrived at the result that something can be a necessary consequence of something else, without being an apriori consequence of it. If this is right, then presumably a claim can be necessarily true without being knowable apriori. One might wonder how this can be so. Certainly Ryle has no account of it; in fact, he seems to be unaware of the possibility. The overall framework within which he works tacitly identifies necessary truths with conceptual truths, which in turn are true in virtue of meaning, and hence—one is told—knowable apriori. So it seems that there is a problem here that Ryle’s overall approach does not equip him to deal with. As we will see, this problem is not an isolated one.

Perception Skepticism about Perceptual Knowledge

Ryle’s discussion in chapter 7 is divided into two parts. In the first part, he is concerned with traditional arguments from the fallibility of sense perception to skepticism about our ability to acquire knowledge through sense perception. His first point is that the mere fact that a certain method

RYLE’S DILEMMAS

83

is fallible gives us no reason to doubt that it often provides us with knowledge. In fact, in some cases our very reason for thinking that a method has led us astray comes from other applications of the same method, which we presuppose to be correct. For example, consider a case in which recounting a group of objects establishes that our initial count was incorrect. In this sort of case, we discover by counting that sometimes we count incorrectly. Obviously, it would be absurd and self-defeating to conclude from this that we can never learn anything by counting. A similar lesson can be drawn regarding perceptual judgments. Sometimes the discovery that one of our perceptual judgments was wrong depends on the correctness of some of our other perceptual judgments. In this sort of case we cannot draw the general skeptical conclusion—that we can never know anything on the basis of perception—without undermining our claim to know the particular skeptical premise—namely, that in certain specific cases our perceptual judgments have been shown to be false. We can’t do this because our claim to have established the skeptical premise itself relies on the accuracy of some perceptual judgments. Consequently, the mere fact that some of our perceptual judgments are in error does not show that we are never justified in relying on perceptual judgments, or that we can never know such judgments to be true. This point is correct, as far as it goes. However, it doesn’t undermine all forms of skepticism about the senses, since there are forms of skepticism that don’t rely on the fact that sometimes we can establish by perception that certain identifiable perceptual judgments are mistaken. Imagine, for example, the following skeptical scenario. You live a normal life for twenty years. Naturally, you make many true perceptual judgments during that time; perhaps you never make a false one. Then, one night your brain is removed, put in a vat, and fed electrical impulses that simulate perception. You are entirely unaware of this, and think you are living a normal life. That is the scenario the skeptic asks you to consider. He contends that if you take it seriously, you will come to the conclusion that you have no evidence, right now, that it isn’t true—since any experience you might have, if you are living a normal life, is one that you would also have under the simulation. Once the skeptic has persuaded you of this, he will conclude that since none of your evidence is capable of ruling out the scenario, you have no choice but to agree that you do not now know that you are living a normal life, and that your perceptual judgments are true. This is an argument for global skepticism about one’s perceptual judgments that does not rely on the fact that certain perceptions have been shown to

84

ORDINARY LANGUAGE PHILOSOPHY

be mistaken by other, correct, perceptions. Thus, it is a form of skepticism that remains unaddressed by Ryle’s first point. Ryle’s second point is more ambitious. His first point was that certain specific forms of skeptical argument are self-defeating (because the knowledge required to establish their skeptical premises conflicts with their general skeptical conclusions). His second point is that skepticism itself is self-defeating. His idea seems to be something along the following lines: The claim that some F’s are erroneous, fraudulent, or incorrect logically or conceptually, presupposes that some F’s are not. Applying this idea to perceptual judgments, we conclude that the claim that some perceptual judgments are erroneous presupposes that some perceptual judgments are accurate. If that is right, then any skeptic who claims that all perceptual judgments are erroneous presupposes something incompatible with this—namely, that some are not. If Ryle is right, then any such skeptical position is incoherent. Suppose, though, that the skeptic is more cautious, and says, not that all our perceptual judgments are mistaken (i.e., false), but only that it might be the case, for all we know, that they are. Ryle would, I think, maintain that the skeptic’s position would still be incoherent. His thesis is that the claim that some F’s are erroneous or fraudulent conceptually presupposes that some F’s are not erroneous or fraudulent. This thesis doesn’t just say that if some F’s are erroneous, other F’s are not. Rather, it says that it is conceptually impossible for some F’s to be erroneous unless other F’s are not. Applied to perceptual judgments, this means that it is conceptually impossible for some of our perceptual judgments to be mistaken unless other perceptual judgments of ours are not; and this entails that it is conceptually impossible for all of our perceptual judgments to be mistaken (false). Thus, the skeptic who says that, for all we know, all our perceptual judgments may be mistaken is saying that, for all we know, a conceptual impossibility might be true. Since that cannot be, it follows that if Ryle’s principle is correct, and if it can correctly be applied to perceptual judgments, then skepticism about the truth of all such judgments can be decisively refuted. Ryle illustrates this with an analogy. A country which had no coinage would offer no scope to counterfeiters. There would be nothing for them to manufacture or pass counterfeits of. They could, if they wished, manufacture and give away decorated discs of brass or lead, which the public might be pleased to get. But these would not be false coins. There can

RYLE’S DILEMMAS

85

be false coins only where there are coins made of the proper materials by the proper authorities. In a country where there is a coinage, false coins can be manufactured and passed; and the counterfeiting might be so efficient that an ordinary citizen, unable to tell which were false and which were genuine coins, might become suspicious of the genuineness of any particular coin that he received. But however general his suspicions might be, there remains one proposition which he cannot entertain, the proposition, namely, that it is possible that all coins are counterfeits. For there must be an answer to the question ‘Counterfeits of what?’15 Ryle talks about the impossibility of counterfeit coins if there were no real coins. This is not really so obvious. A country might announce that it was going to make coins but not follow through. Counterfeiters might produce some coins, and people might accept them only because they believe they were made by the government, and would be redeemed by the government for a certain quantity of gold. That is, people might accept these phony coins only because they believe the coins to have a property they don’t have. One might describe this case by saying that in the situation imagined, the only coins are counterfeit coins—in apparent violation of Ryle’s principle.16 Could such a counterexample be created in the case of perceptual judgments? The brain-in-vat example may be like that. It seems that after the brains have been put in the vats, all their perceptual judgments about objects around them are incorrect. In a case like that, the agents make false judgments on the basis of sensory experience, because they think that their sensory experiences have a property that in fact they do not have—namely, the property of being caused in a certain way by external objects they are perceiving. (This parallels the example involving counterfeit coins.) If this scenario is conceptually possible, then it creates trouble for Ryle. Let us try to spell out exactly what kind of trouble it causes. We may take Ryle as suggesting the following principle: From the premise that sometimes we are perceptually deceived, it follows logically or conceptually that we are not always perceptually deceived. On the face of it, the brain-in-vat scenario seems to be a counterexample to the principle, since after the brains are placed in vats they are always deceived. If Ibid., pp. 94–95. An idea similar to the one behind Ryle’s argument is developed and given a better run for its money by Norman Malcolm. See the discussion of his “paradigm case argument” in chapter 7. 15 16

86

ORDINARY LANGUAGE PHILOSOPHY

that is right, then Ryle’s principle is false, and his argument that skepticism is always self-undermining fails for that reason. On the other hand, one might claim that the brain-in-vat scenario does not falsify the principle, when properly understood. Recall that it was part of the brain-in-vat story that prior to being put in vats the brains lived normal lives and made lots of correct perceptual judgments. So if we consider the whole life of each brain, it was not a life in which every perceptual judgment was erroneous. A defender of Ryle might claim that the principle should be understood only as ruling that out, while allowing for the possibility that after a certain point all the perceptual judgments of the brains are deceptive. There are various responses that one might make to this attempted defense of Ryle. But the simplest response, for our purposes, is that if we grant that Ryle’s principle is compatible with our brain-in-vat scenario, then the cost of saving the principle is to make it compatible with an extremely radical and discomforting form of skepticism. On this interpretation of it, the principle leaves important forms of skepticism intact, and so fails to provide any general refutation of skepticism. The Nature of Perception

At this point in the chapter, Ryle drops the topic of global skepticism about perceptual judgments and turns to a different issue about perception—namely, whether psycho-physiological research can tell us what perception is. So when we ask ‘How do we see trees?’ or ‘What happens in us when we see trees?’ we are predisposed to expect the same sorts of answers, namely reports of modifications in some of our internal states and processes. Further than that, we are predisposed to think that these reports will tell us not only what happens in us when we perceive but what perceiving is.17 Ryle goes on to say that seeing a tree is not an inner physiological or psychological process, because it is not a state or process that could be discovered by scientific means at all. Yet, however its details may be filled in, this sort of story leaves us uneasy. When asked whether I do or do not see a tree, I do not dream of postponing my reply until an anatomist or physiologist has 17

Ibid., p. 100, my emphasis.

RYLE’S DILEMMAS

87

probed my insides, any more than he, when asked whether he has seen the zigzag lines on his encephalogram, postpones replying until some other anatomist or physiologist has tested him by a second encephalogram. The question whether I have or have not seen a tree is not itself a question about the occurrence or non-occurrence of experimentally discoverable processes or states some way behind my eyelids, else no one could even make sense of the question whether he had seen a tree until he had been taught complicated lessons about what exists and occurs behind the eyelids.18 This passage is extremely interesting, but also fundamentally flawed. To illustrate this, we draw the following analogy to what is said in the passage: Whether or not the glass was full of water is not a question about the presence or absence of H2O in the glass, because if it were, no one could answer it without knowing chemistry. Ditto for lightning and electrical discharge, gold and the element with atomic number 79, heat and molecular motion, light and streams of photons. Clearly this pattern of reasoning, which is the same as Ryle’s, is faulty. The mistake may be put as follows: Let A and B be general terms or phrases for natural kinds, processes, or states. The claim expressed by A is B or A’s are B’s can be true, perhaps even necessary, even though the words A and B differ in meaning, so statements containing one are not translatable into statements about the other. In fact, the claim expressed by Such and such is B may not be derivable from the statement expressed by Such and such is A by any apriori reasoning at all. So questions using A may be conceptually unrelated to questions using B even though the statement A is B, or the statement A’s are B’s, is necessarily true. This recognition destroys Ryle’s argument that seeing a tree is not a physiological process. It is not obvious what, precisely, he was thinking. Here is one plausible reconstruction. He may have been thinking that if A is B is a statement that tells us what the essential nature of the referent of A is (in Ryle’s case, a statement telling us what the essential nature of perception is), then it must be a necessary truth; otherwise, the most it could express would be a contingent correlation. On his view, this would mean that the statement must also be an analytic, apriori truth, and hence knowable simply by conceptual analysis and reflection. But no statement that relates my seeing a tree to the complex workings of my brain can be known apriori, on the basis of conceptual analysis. Thus, no such statement can tell us what my seeing a tree, essentially, is. 18

Ibid., pp. 100–101.

88

ORDINARY LANGUAGE PHILOSOPHY

The problem with this line of reasoning lies in the implicit claim that if a statement is a necessary truth, then it is an analytic, apriori truth. At the time Ryle wrote, this methodological principle would have been accepted by virtually every analytic philosopher. The fact that most philosophers no longer believe the principle opens up options regarding the essential natures of things that were all but invisible to Ryle.19 Because these options are genuine, the argument he gives in the previous passage fails. There may, of course, be other reasons for thinking that the essential nature of seeing a tree is not given by any complex neurophysiological findings about me—e.g., our readiness to recognize that different animals with very different neurophysiologies see trees—but Ryle doesn’t cite these reasons. Moreover, even if the generic nature of seeing a tree is not to be specified neurophysiologically, it may still be that a certain neurophysiology is essential to me, and because of this, a certain neurophysiological process is essential to my seeing a tree. At this point Ryle gives a second argument that purports to show that seeing a tree is not a state or process. If it were such a state or process, then it would make sense to speak of the process as lasting a certain amount of time, or to speak of the state as enduring for a certain amount of time. But Ryle maintains that it makes no sense to speak of my seeing the tree during a certain interval. On pages 103 and 104 he draws analogies to scoring a goal and solving a problem. He observes that scoring a goal is an achievement which has no duration, even though the process leading to it does. Whatever one may think of this point, it seems clear that in extending it to the case of solving a problem, he is on thin ice. But I could not say ‘I am at present solving this anagram’. Either I have now got the solution or I have not yet got it. In short, a lot of biographical verbs like ‘find’, ‘see’, ‘detect’, and ‘solve’ share with a lot of other verbs of stopping and starting, which have no special biographical connotations, the negative property of not standing for processes taking place in or to things, or for states in which things remain.20 What should we think of this? It is true that if one has not solved a problem, but is in the process of doing the necessary calculation, then one might be reluctant to describe oneself as solving the problem. The 19 20

This point will be extensively discussed in Part 7 of this volume. Ryle, Dilemmas, p. 104.

RYLE’S DILEMMAS

89

reason for this is that nothing one is doing will count as solving the problem unless a correct solution is eventually arrived at. And if one has not yet got the solution, one may be in no position to assert that it will be forthcoming—hence the reluctance to describe oneself as solving the problem. However, this is compatible with the view that solving a problem is a process that may take time, but which must terminate in a certain achievement. After I have solved a problem, I can look back and say that I was solving the problem between 1:30 and 2:00 in the afternoon. So it looks to me as if Ryle was wrong on this point. Moreover, the error illustrates a potentially significant fallacy in his general pattern of reasoning: from the fact that normally it would be odd or inappropriate for me to say during a certain interval that I am doing so and so, it doesn’t follow that my doing so and so isn’t a process that takes place during that interval. The reason it doesn’t follow is that there may be some other explanation for the oddness or inappropriateness of the remark in most circumstances. As I have indicated, in the case of solving a problem, this seems to be so. The same may be true of perception. Imagine a conversation in which someone makes the following comment: “When I told you, repeatedly, over the phone that I had never seen or heard a yellow-bellied sapsucker, I didn’t realize that in fact I was seeing one, and hearing it sing, throughout our conversation.” That sounds as if it could be true. If so, then one can see and hear something during a period of time. On page 106, Ryle makes another point. To see a tree, there must be a tree there to be seen. Since the tree is not part of me, but its presence is necessary for me to see the tree, my seeing the tree cannot be a physiological process in me. This is correct; it can’t. But my seeing a tree could be a relational process or state involving both me and the tree; and the psychologist and physiologist might be able to tell us interesting things about it. Finally, Ryle tells us that he is anxious to rebut the view that what we perceive are things internal to us, rather than what we ordinarily take ourselves to perceive. He says: From some well-known facts of optics, acoustics and physiology it seemed to follow that what we see, hear or smell cannot be, as we ordinarily suppose, things and happenings outside us, but are on the contrary, things or happenings inside us.21 21

Ibid., p. 109.

90

ORDINARY LANGUAGE PHILOSOPHY

That is certainly something to be resisted, and Ryle is on target in wanting to resist it. But his way of resisting the conclusion seems dubious. He imagines his opponent starting from the position that seeing something is a process, and arguing that it is a process internal to us. From there, the person is imagined to conclude that what we see is internal to us. Ryle wants to resist this position by refusing to grant the assumption that seeing something is a state or process. A better strategy would be (i) to point out that it is a relational state or process and (ii) to observe that even though much of the process is internal, it does not follow that the thing we see is.

Lessons This concludes our discussion of Ryle’s Dilemmas. His aim in the book was to sketch a conception of paradigmatically philosophical problems, and to illustrate how the new style of ordinary language linguistic analysis could be used to solve them. As I have tried to indicate, his success in this was limited. On the positive side, his conception of dilemmas does seem to capture one important type of philosophical problem. In addition, his discussions of fatalism, the world of physics and the world of common sense, skepticism, and the nature of perception contain a number of useful conceptual clarifications which do throw some light on the problems he took up. In no case, however, did he solve, or really get to the heart of, the problem. Three major shortcomings stand out as worthy of notice. First, his rejection of old-style Russellian logical analysis was too sweeping, and sometimes, as in the case of fatalism, prevented him from grasping central philosophical issues. Second, his identification of the necessary and the apriori with the analytic, and of the philosophical with a subclass of the latter, blinded him to certain significant philosophical questions and alternatives, as we saw in his discussions of the world of science and the world of common sense, and of the nature of perception. Finally, he did not properly distinguish meaning from other factors influencing the use of expressions. As a result, he was sometimes prone to misdiagnose the oddness or inappropriateness of certain utterances as indicating a conceptual incoherence in the sentences uttered—as in his discussion of success terms like solving a problem or seeing a tree. These flaws were not idiosyncrasies of Ryle. In Part 1, we saw that Wittgenstein was not

RYLE’S DILEMMAS

91

free of them. In general, they were endemic to the ordinary language approach to philosophy that flourished during the period we are studying. While the first of Ryle’s flaws may be corrected by looking back to Russell, the latter two will require looking ahead to the work of Kripke and Grice.

CHAPTER 4 RYLE’S CONCEPT OF MIND chapter outline 1. “Descartes’s Myth” Ryle’s rejection of “the ghost in the machine” and the causation of behavior by private mental events of which the agent has direct knowledge; problems posed by apparent reliance on verificationism and the conflation of necessity and aprioricity 2. Knowing How, Knowing That, and Acting Intelligently The regress argument that to act intelligently is not to follow rules or to apply theoretical knowledge; the failure of the argument Ryle’s dispositionalist analysis of what it is to act skillfully or intelligently; the problems posed by Ryle’s use of mentalistic notions, including belief and desire, to characterize the relevant dispositions Problems with analyzing belief and desire in terms of purely behavioral dispositions 3. Explaining Action in Terms of Belief, Desire, and Emotion Two contending pictures; Ryle’s rejection of belief and desire explanations as positing internal mental causes, and his contention that they should be understood as dispositional; his analysis of “the man did so and so because he is vain” Purely behavioral dispositions vs. dispositions to think and feel in certain ways; vanity is a disposition of the latter kind. Why citing vanity in the explanation of an action involves attributing a mental cause of a certain sort 4. Non-Inferential First-Person Knowledge of Mental States The problem this poses for Ryle’s view, and his unsuccessful attempt to circumvent it by adopting a performative analysis of first-person avowals 5. Conclusions

Ryle’s task in his major work, The Concept of Mind, is to undermine a certain natural and commonly held picture of the mind, and to provide an

RYLE’S CONCEPT OF MIND

93

alternative analysis of our talk about the mental that takes it to be to talk about people’s behavior and the circumstances in which they find themselves. In this chapter, we will look at his project in some detail. However, there is something to be noted at the very outset. As we will see, what Ryle argues for amounts to a sweeping revision of our conception of the mental. This is noteworthy, since such revisionism clashes with his ordinary language ideology about the nature of philosophy. According to the ideology, philosophy is not supposed to give us new theories; instead, it is supposed to untangle linguistic confusions that have led past philosophers astray—presumably leaving us with a clearer, less muddled version of what we pre-theoretically thought. This is not a conception of philosophy that leads one to expect big surprises, or radical revisions of our ordinary world view. Nevertheless, that is what Ryle gives us. No doubt, we should keep an open mind about ways of resolving this tension. Surely, Ryle himself would have denied that his view about the mental was revisionary. However, philosophers’ professions of solidarity with the common man often need to be taken with a grain of salt. Even Berkeley, who maintained that the planet Earth is nothing more than a collection of the private sense impressions of different minds, claimed to be speaking and thinking with the vulgar—and opposing only an illogical and untenable doctrine of mind-independent material objects, oddly prevalent among philosophers. In the case of Ryle, the apparent tension between these two aspects of his thought— his commitment to ordinary language and a highly deflationary view of philosophy, on the one hand, and his apparently revisionary conception of the mental on the other—should get our attention, and lead us to scrutinize his position carefully.

“Descartes’s Myth” In chapter 1, Ryle presents and criticizes a view of the mind he calls “the dogma of the ghost in the machine.” The view he outlines includes the following theses: 1. Human beings have both minds and bodies, which are distinct substances, or kinds of things. Our bodies are physical, and exist in time and space; our minds are non-physical, and exist only in time. Since the two substances are different, it is conceivable that our minds could continue to exist after our bodies cease to exist.

94

ORDINARY LANGUAGE PHILOSOPHY

2. Being parts of the physical world, our bodies are subject to all normal physical laws. Since our minds are not parts of the physical world, they are not subject to these laws. Undoubtedly the body can influence the mind, and the mind can influence the body, but the nature of this interaction is mysterious. 3. Each person has direct awareness of the contents of his own mind, of his beliefs and desires, his moods and emotions, his memories, his perceptions, his pleasures and pains. By contrast, a person is not directly aware of the contents of other minds. At best, one person may make informed inferences about the beliefs, desires, moods, emotions, perceptions, pleasures, and pains of others, based on their behavior. So, when I describe someone else as believing, desiring, or remembering something, or as finding something amusing or repugnant, I am making a claim about the contents of that person’s mind that he can directly verify, but that I can only infer from his behavior and what he says. My own view of the matter is that (1) and (2) are implausible, though quite understandable and commonly held, views, but that (3), or something like it, is probably correct. Ryle disagrees. He tells us that this picture is not just wrong in some details, but completely wrong and fundamentally absurd. Such in outline is the official theory. I shall often speak of it, with deliberate abusiveness, as ‘the dogma of the Ghost in the Machine.’ I hope to prove that it is entirely false, and not in detail but in principle. It is one big mistake and a mistake of a special kind. It is, namely, a category-mistake. It represents the facts of mental life as if they belonged to one logical type or category (or range of types or categories) when they actually belong to another. The dogma is therefore a philosopher’s myth.1 In saying that the view he is attacking makes a category mistake, Ryle is claiming that it represents facts about our mental life as being of one conceptual type or category when in fact they are of a completely different type or category. He illustrates what he means by a category mistake as follows: A foreigner visiting Oxford or Cambridge for the first time is shown a number of colleges, libraries, playing fields, museums, scientific departments, and administrative offices. He then asks, 1

Ryle, The Concept of Mind, pp. 15–16.

RYLE’S CONCEPT OF MIND

95

‘But where is the University? I have seen where the members of the Colleges live, where the Registrar works, where the scientists experiment and the rest. But I have not yet seen the University in which reside and work the members of your University.’ It has then to be explained to him that the University is not another collateral institution, some ulterior counterpart to the colleges, laboratories, and offices which he has seen. The University is just the way in which all that he has already seen is organized. When they are seen and when their co-ordination is understood, the University has been seen.2 Someone who sees the different parts of the university, but then complains that he has not yet been shown the university, thinks of it as if it were some particular building alongside the psychology building, the residence halls, etc., or as some particular institution in addition to the departments and colleges that make it up. Ryle would say that he fails to realize that when we talk about the university we are really just talking about how all the different institutions, buildings, offices, people, etc., are coordinated. Similarly, Ryle thinks that someone who believes that the mind is something over and above the body fails to realize that the mind is not another thing, over and above the body. Instead, when we talk about a person’s mind we are really just talking about how his behavior and activities are coordinated. If Ryle is right about this, then many people have been suffering from a very serious confusion. Think of how confused the person is who, after touring the campus, asks where the university is. According to Ryle, people who think that the mind is distinct from the body are just as confused. How did they get so confused? In section 3 of chapter 1 he pays a remarkable compliment to the influence of philosophers. He says that the chief origin of the confusion was Descartes—though elsewhere he notes that other misguided philosophers and theologians have contributed to the confusion as well. One of the chief intellectual origins of what I have yet to prove to be the Cartesian category-mistake seems to be this. When Galileo showed that his methods of scientific discovery were competent to provide a mechanical theory which should cover every part of space, Descartes found in himself two conflicting motives. As a man of scientific genius he could not but endorse the claims of mechanics, 2

Ibid., p. 16.

96

ORDINARY LANGUAGE PHILOSOPHY

yet as a religious and moral man he could not accept, as Hobbes accepted, the discouraging rider to those claims, namely that human nature differs only in degree of complexity from clockwork. The mental could not be just a variety of the mechanical. He and other philosophers naturally availed themselves of the following escape-route. Since mental-conduct words are not to be construed as signifying the occurrence of mechanical processes, they must be construed as signifying the occurrence of non-mechanical processes.3 Consider a case in which I describe a person’s actions as intelligent, and explain that he did a particular thing because he desired that P and believed that by acting in a certain way he could bring it about that P. In describing the person in this way, I am not saying that his action is caused by such and such physical states and processes. But I do seem to be saying something about how it is caused. According to Ryle, Descartes and other philosophers concluded that I must be ascribing non-physical causes to the action. The difference between the human behaviors which we describe as intelligent and those which we describe as unintelligent must be a difference in their causation; so, while some movements of human tongues and limbs are the effects of mechanical causes, others must be the effects of non-mechanical causes, i.e. some issue from movements of particles of matter, others from workings of the mind.4 Ryle is rightly suspicious of this argument. Let me try to bring out why one should be suspicious. Suppose I describe Smith’s death by saying that he was shot by a gunman in a passing car. In describing Smith’s death in this way, I am not saying that it was caused by the Godfather, or his men. Still, I seem to be saying something about the cause of his death. We would not conclude from this that I must have been saying that his death was caused by someone other than the Godfather, or his men. Rather, we would conclude that I have identified the cause of Smith’s death, namely the gunman in the passing car, while leaving it open whether or not further investigation will reveal that the gunman was sent by the Godfather. Similarly, when I say that Brown performed a certain action because he desired that P and believed that by acting in 3 4

Ibid., pp. 18–19. Ibid., p. 19.

RYLE’S CONCEPT OF MIND

97

a certain way he would bring it about that P, I am saying that his action had a certain cause. I am not saying that the cause was a physical one, to be described by the laws of mechanics, or whatever other laws govern the physical world. But I am also not saying that the cause of Brown’s action wasn’t physical. What I say simply leaves it open whether or not future investigation will convince us that the cause of Brown’s action— his desiring that P and believing that by acting in a certain way he would bring it about that P—will turn out to be identical with some complex physical state of affairs. Thus, in my opinion, we can agree with Ryle that our ordinary explanations of behavior in terms of belief and desire do not commit us to the Cartesian picture, according to which our actions are the result of some special non-physical causation. However, neither do our ordinary explanations of behavior commit us to the rejection of Cartesianism. In addition, whether we accept or reject other aspects of Cartesianism, we can, if we wish, continue to view the beliefs and desires of others as states of their minds or brains which they know directly and non-inferentially, even though we can only fallibly infer their existence as parts of our best explanation of what they say and do. That is the lesson I think we should draw from Ryle’s account of the genesis of the Cartesian picture. However, it is not the lesson that he draws. He thinks that not only was the Cartesian picture not justified by the considerations he takes to have prompted it; in addition, it was absurd and virtually self-refuting from the very beginning. Theorists correctly assumed that any sane man could already recognize the difference between, say, rational and non-rational utterances or between purposive and automatic behaviour. Else there would be nothing requiring to be salved from mechanism. Yet the explanation given presupposed that one person could in principle never recognize the difference between the rational and irrational utterances issuing from other human bodies, since he could never get access to the postulated immaterial causes of some of their utterances. Save for the doubtful exception of himself, he could never tell the difference between a man and a Robot.5 It seems to me that this argument is far too ambitious, and for that reason ends up being extremely weak. The idea behind it seems to be some form of verificationism. If beliefs and desires were private mental states, 5

Ibid., pp. 20–21.

98

ORDINARY LANGUAGE PHILOSOPHY

then we could never observe the beliefs and desires of others.6 But if we couldn’t observe them, then we couldn’t know that they exist. In fact, if we couldn’t observe them, then we couldn’t even plausibly conjecture that they exist. Since we do have some reasonable beliefs, and even knowledge, about the beliefs and desires of others, beliefs and desires cannot be private mental states. This argument is no stronger than verificationism in general, which by 1949, when The Concept of Mind was published, had been abandoned by its main proponents, the logical positivists, for the simple reason that every precise formulation of it had been decisively refuted.7 But even putting such technical matters aside, it is easy to see why verificationism is problematic. We all take ourselves to be able to know many things about the past, and to be able plausibly to conjecture about other things in the past that we may not know—this, despite the fact that we cannot now observe anything about the past. So the fact that we can’t observe certain things doesn’t mean that we can’t know anything about them, or that we can’t justifiably distinguish plausible from implausible claims about them. A similar point holds for scientific theorizing. Many sciences, from geology to physics and astronomy, posit theoretical entities and events that we cannot observe. The fact that we cannot observe them doesn’t mean that the scientific theories don’t give us justifiable beliefs about them, and, in some cases, knowledge. If, on the basis of these considerations, we reject verificationism in general, then we have little reason to accept Ryle’s apparent use of verificationism when applied to claims about the mental life of others. Where does this leave us? There are, I think, a variety of reasons for being suspicious of certain aspects of the Cartesian view of the mind. However, Ryle’s quick verificationist argument doesn’t refute it. Ryle is correct in thinking that we should not jump to the conclusion that ordinary explanations of behavior using notions like belief and desire commit us to some non-physical kind of causation. However, he has not ruled out the possibility that beliefs and desires are internal states that are genuine causes of behavior, the ultimate natures of which are still unknown. For example, he has not ruled out the possibility that beliefs and desires may be neurophysiological, or more broadly physical, states. Could we always observe them if, as Ryle believed, they were behavioral dispositions? Although behavior is observable, dispositions to behave often are not. Since some inference is often needed to justify claims about behavioral dispositions, one wonders why inferences to other kinds of internal states should be regarded as so problematic. 7 See chapter 13 of volume 1. 6

RYLE’S CONCEPT OF MIND

99

Ryle himself would, I am sure, regard this suggestion as absurd, and he would point out that we can know perfectly well that some other person believes a certain thing without knowing anything about the inside of that person’s brain. But this response misses the point. The mistake inherent in it lies in thinking that because the connection between having a belief and being in a certain neurological or physical state is not a conceptual connection, discoverable by mere analysis and reflection, it therefore cannot be a necessary connection. I suspect that it is because Ryle makes this mistake that he is inclined to think that whatever we refer to when we talk about beliefs cannot be identified with neurological or physical states, but can at most be contingently correlated with them. Since he is understandably suspicious of non-physical causes, he thinks that beliefs and desires can’t be the internal causes of our actions, though they might turn out to be contingently correlated with such causes. But if beliefs and desires are not causes of action, what are they? The short answer is that Ryle takes them to be dispositions to behave in certain ways. To say that someone has a certain belief or desire is not to isolate a cause of his action; rather, it is to describe the person as being an individual who would act in a certain way if certain conditions were fulfilled. This is the picture that Ryle tries to develop in the rest of the book.

Knowing How, Knowing That, and Acting Intelligently The Argument That to Act Intelligently Is Not to Follow Rules or to Apply Theoretical Knowledge

We now turn to chapter 2, where Ryle sets out what he is up to on the first page. In this chapter I try to show that when we describe people as exercising qualities of mind, we are not referring to occult episodes of which their overt acts and utterances are effects; we are referring to those overt acts and utterances themselves.8 Ryle wants to show that we do not refer to private mental states and events when we use mental vocabulary such as intelligent, stupid, imaginative, ingenious, etc. Rather, when we use these and other mental terms, we ascribe to agents various complex behavioral dispositions. As we will see, his arguments for this thesis are rather limited. However, in 8

Ryle, The Concept of Mind, p. 25.

100

ORDINARY LANGUAGE PHILOSOPHY

the course of trying to provide compelling arguments, Ryle makes some interesting distinctions, and comes up with a number of illuminating observations. One of his main concerns is to distinguish knowing how to do something from knowing that such and such is the case. He is particularly interested in undermining something he calls the “intellectualist legend,” according to which whenever we know how to do something, our ability derives from the possession of theoretical knowledge that such and such is the case. When we speak of the intellect or, better, of the intellectual powers and performances of persons, we are referring primarily to that special class of operations which constitute theorizing. . . . The main object of this chapter is to show that there are many activities which directly display qualities of mind, yet are neither themselves intellectual operations nor yet effects of intellectual operations. Intelligent practice is not a step-child of theory. On the contrary theorizing is one practice amongst others and is itself intelligently or stupidly conducted.9 Ryle wants to show that many things that we know how to do, and that can be properly described as being done intelligently or stupidly, are not the result of an act of intellectual preplanning, or the application of prior knowledge that such and such is the case. If he can succeed in showing this, then he thinks he will have shown that some characterizations of a person as having done something intelligently can’t be understood as positing a certain kind of internal mental causation—intellectual preplanning or the application of prior knowledge—of the act, and so must be understood in some other way. He begins by trying to explain why philosophers have been inclined to think that when we say that someone did something intelligently, we are saying something about the private intellectual processes that caused the action. What is involved in our descriptions of people as knowing how to make and appreciate jokes, to talk grammatically, to play chess, to fish, or to argue? Part of what is meant is that, when they perform these operations, they tend to perform them well, i.e. correctly or efficiently or successfully. Their performances come up to certain standards, or satisfy certain criteria. . . . To be intelligent is not merely to satisfy certain criteria, but to apply them; to regulate one’s 9

Ibid., p. 26.

RYLE’S CONCEPT OF MIND

101

actions and not merely to be well-regulated. A person’s performance is described as careful or skilful, if in his operations he is ready to detect and correct lapses, to repeat and improve upon successes, to profit from the examples of others, and so forth. He applies criteria in performing critically, that is, in trying to get things right.10 According to Ryle, for a person to do something intelligently or skillfully is for the person to regulate his performance, with the effect that his performance comes to satisfy criteria of excellence. What is involved in a person’s regulating his actions? Ryle thinks that intellectualists try to assimilate this to applying theoretical knowledge. Champions of this legend are apt to try to reassimilate knowing how to knowing that by arguing that intelligent performance involves the observance of rules, or the application of criteria. It follows that the operation which is characterized as intelligent must be preceded by an intellectual acknowledgement of these rules or criteria; that is, the agent must first go through the internal process of avowing to himself certain propositions about what is to be done (‘maxims’, ‘imperatives’, or ‘regulative propositions’ as they are sometimes called); only then can he execute his performance in accordance with those dictates.11 Here, Ryle characterizes the view he opposes as maintaining that intelligent action (always) results from following an internally formulated plan. Ryle’s next move is to try to show that this general claim is false. He gives an argument designed to show that not all intelligent activity is the result of intellectual pre-planning (or the application of theoretical knowledge). The key point in the argument is the observation that preplanning (or the application of knowledge) is itself an activity that can be intelligent or not. This is supposed to create a regress. In order to avoid the regress, Ryle concludes that it is not the case that to call an action or activity intelligent is to say that it is preplanned. Here it is worth quoting the thesis that Ryle refutes: To put it quite generally, the absurd assumption made by the intellectualist legend is this, that a performance of any sort inherits all its title to intelligence from some anterior internal operation of planning what to do.12 Ibid., pp. 28–29. Ibid., p. 29. 12 Ibid., p. 31. 10 11

102

ORDINARY LANGUAGE PHILOSOPHY

We may paraphrase the thesis that Ryle refutes as follows: What makes a performance of any kind whatsoever intelligent is its being caused by a prior internal act (performance) of preplanning (or applying theoretical knowledge) which is itself intelligent. Since any prior act of preplanning (or applying theoretical knowledge) is itself taken to be a performance of some sort, this thesis leads directly to a regress. What this result shows is that either some performances are intelligent by virtue of intellectual preplanning (or the application of theoretical knowledge) that is not itself intelligent, or some performances are intelligent but not by virtue of intellectual preplanning (or the application of theoretical knowledge) at all. Although this is a rather weak conclusion, Ryle attempts to derive some stronger conclusions from it.13 First, he concludes that for some acts, the fact that they are intelligent doesn’t have anything to do with being caused by preplanning (or the application of theoretical knowledge). Next, he concludes that the notion of intelligence, as applied to an act, can’t be defined in terms of preplanning (or the application of theoretical knowledge). Finally, he suggests that whenever we say that any act was done intelligently, we are not saying anything about its being preplanned, or about any internal causes of the act. This goes well beyond what is established by the regress argument, which leaves open the possibility that all intelligent acts are intelligent in virtue of their causes, provided that it is acknowledged that such causes need not themselves always be intelligent in order to play this role.14 Ryle’s Positive Analysis of What It Is to Act Skillfully or Intelligently

For this reason, I don’t think that Ryle has refuted the view he opposes. Nevertheless, it is illuminating to try to spell out in more detail the positive alternative he advocates. What, according to Ryle, are we saying when we say that someone’s performance of an activity is intelligent or Ibid., pp. 31–32. Consider, for instance, one of Ryle’s examples of intelligent behavior—“knowing how to talk grammatically.” According to at least some interpretations of Chomsky, this involves the automatic, unconscious application of internal grammatical rules. On this view, the application of these rules may be an essential part both of the causal explanation of the agent’s speech behavior, and of the characterization of it as intelligent—i.e., as guided by complex internal computations in the service of a recognized end. This is not to say that the Chomsky-like view is correct; however, it is not refuted by Ryle’s regress argument.

13 14

RYLE’S CONCEPT OF MIND

103

skillful? What, for example, does it mean to reason or argue intelligently, or to be a skillful boxer, or marksman? Ryle first takes up the question of what it means to say that a certain performance was skillful, or was the exercise of a skill. To recognize that a performance is an exercise of a skill is indeed to appreciate it in the light of a factor which could not be separately recorded by a camera. But the reason why the skill exercised in a performance cannot be separately recorded by a camera is not that it is an occult or ghostly happening, but that it is not a happening at all. It is a disposition, or complex of dispositions, and a disposition is a factor of the wrong logical type to be seen or unseen, recorded or unrecorded.15 According to Ryle, when we say that someone performed an activity skillfully, we are saying (i) that he performed it, and (ii) that his performance was the manifestation of dispositions to behave in certain ways. What, then, does he mean by a disposition? He tells us: When we describe glass as brittle, or sugar as soluble, we are using dispositional concepts, the logical force of which is this. The brittleness of glass does not consist in the fact that it is at a given moment actually being shivered. It may be brittle without ever being shivered. To say that it is brittle is to say that if it ever is, or ever had been, struck or strained, it would fly, or have flown, into fragments. To say that sugar is soluble is to say that it would dissolve, or would have dissolved, if immersed in water. A statement ascribing a dispositional property to a thing has much, though not everything, in common with a statement subsuming the thing under a law. To possess a dispositional property is not to be in a particular state, or to undergo a particular change; it is to be bound or liable to be in a particular state, or to undergo a particular change, when a particular condition is realized.16 Let us, for the sake of argument, accept this characterization of a disposition. With it in mind, we go back to our question, What are we saying when we say that someone performed an activity skillfully? Ryle’s answer is that (i) we are saying that he performed the activity, and (ii) we are saying or implying that a variety of subjunctive conditionals if it had been 15 16

Ryle, The Concept of Mind, p. 33. Ibid., p. 43.

104

ORDINARY LANGUAGE PHILOSOPHY

the case that P, then it would have been the case that Q are true, where these conditionals talk about aspects of the performance of the activity. This naturally invites the question, Which particular subjunctive conditionals are we asserting when we say that someone’s performance of an activity was skillful? In the following passage, Ryle says something that indicates that we should not expect a definite and precise analysis. But the practice of considering such simple models of dispositions, though initially helpful, leads at a later stage to erroneous assumptions. There are many dispositions the actualizations of which can take a wide and perhaps unlimited variety of shapes; many disposition-concepts are determinable-concepts. When an object is described as hard, we do not mean only that it would resist deformation; we mean also that it would, for example, give out a sharp sound if struck, that it would cause us pain if we came into sharp contact with it, that resilient objects would bounce off of it, and so on indefinitely. If we wished to unpack all that is conveyed in describing an animal as gregarious, we should similarly have to produce an infinite series of different hypothetical propositions.17 The view seems to be that when we call a person’s performance of an activity skillful, we are (i) saying that he performed the activity and (ii) indicating in some vague and open-ended way our commitment to what may be an infinite number of subjunctive conditionals if it had been the case that P, then it would have been the case that Q—where these conditionals talk about aspects of the performance of the activity. Ryle doesn’t worry about this open-endedness, since he thinks that similar open-endedness characterizes virtually all language. We now come to his example of the skillful marksman. We observe, for example, a soldier scoring a bull’s eye. Was it luck or was it skill? If he has the skill, then he can get near the bull’s eye again, even if the wind strengthens, the range alters and the target moves. Or if his second shot is an outer, his third, fourth, and fifth shots will probably creep closer to the bull’s eye. He generally checks his breathing before pulling the trigger, as he did on this occasion; he is ready to advise his neighbor what allowances to make for refraction, wind, etc. Marksmanship is a complex of skills, and the question whether he hit the bull’s eye 17

Ibid., p. 44.

RYLE’S CONCEPT OF MIND

105

by luck or from good marksmanship is the question whether or not he has the skills and, if he has, whether he used them by making his shot with care, self-control, attention to the conditions and thought of his instructions.18 According to Ryle, part of what we are committed to when we call a man a skillful marksman is the claim that if he shoots under various conditions, then many of his shots will land near the target. Moreover, we are claiming that this will be so even if the wind strengthens, the range changes, or the target moves. Also, Ryle suggests, we are indicating that if the marksman is asked for advice, “he is ready to advise his neighbor what allowances to make for refraction, wind, etc.” Finally, Ryle says that the good marksman exercises “care, self-control, attention to the conditions, and thought of his instructions” (my emphasis). These last comments raise an important issue. We need to distinguish a strong from a weak positive thesis as possible conclusions of Ryle’s argument. The weak positive thesis : To call the performance of some activity intelligent, skillful, or the exercise of a skill, is to attribute certain dispositions to the agent regarding how he would perform in various circumstances. The strong positive thesis : To call the performance of some activity intelligent, skillful, or the exercise of a skill, is to attribute certain dispositions to the agent regarding how he would perform in various circumstances, where the dispositions are ultimately behavioral and do not themselves posit any internal states or events that cause or guide action. It is the stronger thesis that Ryle needs to establish. With this in mind we may return to his claim that part of being a skillful marksman is being ready to advise others, in relevant circumstances, what allowances to make for refraction, wind, etc.—as well as taking thought of one’s own instructions in making one’s shot. What is it to have such thoughts and to advise others on these matters? If it is to have certain beliefs, to express them, and to try to get others to share them, then the dispositions that Ryle has identified as being part of the claim that someone is a skillful marksman will be ultimately behavioral only if 18

Ibid., p. 45.

106

ORDINARY LANGUAGE PHILOSOPHY

having beliefs and desiring to get others to share them are purely behavioral dispositions. This is something that has not been established. The same problem arises in Ryle’s discussion of what it is to argue intelligently. [A]lthough there may occur a few stages in his argument which are so trite that he can go through them by rote, much of his argument is likely never to have been constructed before. He has to meet new objections, interpret new evidence and make connections between elements in the situation which had not previously been co-ordinated. In short he has to innovate, and where he innovates he is not operating from habit. He is not repeating hackneyed moves. That he is now thinking what he is doing is shown not only by this fact that he is operating without precedents, but also by the fact that he is ready to recast his expression of obscurely put points, on guard against ambiguities or else on the look out for chances to exploit them, taking care not to rely on easily refutable inferences, alert in meeting objections and resolute in steering the general course of his reasoning in the direction of his final goal.19 The description Ryle gives here is judicious, and more or less accurate. But it is filled with words and phrases that seem to refer to causally efficacious internal mental states—inferring, thinking, interpreting, responding to objections, being on the lookout for this, making sure not to rely on that, and so on. Unless all of these can be shown to be nothing more than behavioral dispositions, Ryle will not have succeeded in establishing that to argue intelligently is simply to manifest a variety of purely behavioral dispositions. Problems with Analyzing Belief and Desire in Terms of Purely Behavioral Dispositions

So what are the prospects of reducing all this talk simply to talk about what behavior would take place in various conditions? One of the most surprising things about The Concept of Mind is that Ryle doesn’t tackle this issue head on, or even say very much directly about it. For example, the most explicit discussion of belief in chapter 2 is the following. Epistemologists, among others, often fall into the trap of expecting dispositions to have uniform exercises. For instance, when 19

Ibid., p. 47.

RYLE’S CONCEPT OF MIND

107

they recognize that the verbs ‘know’ and ‘believe’ are ordinarily used dispositionally, they assume that there must therefore exist one-pattern intellectual processes in which these cognitive dispositions are actualized. Flouting the testimony of experience, they postulate that, for example, a man who believes that the earth is round must from time to time be going through some unique process of cognizing, ‘judging’, or internally re-asserting, with a feeling of confidence, ‘The earth is round’. In fact, of course, people do not harp on statements in this way, and even if they did do so and even if we knew that they did, we still should not be satisfied that they believed that the earth was round, unless we also found them inferring, imagining, saying and doing a great number of other things as well.20 What we need is an analysis which shows that to believe the earth is round is nothing more than to have certain purely behavioral dispositions—i.e., nothing more than for it to be the case that certain actions would be performed in certain specifiable conditions. But this is not what Ryle gives us. What we are given is simply the commonplace observation that if one believes that the earth is round, then one will be disposed to infer, to imagine, and to say things in certain cases, that one would not be disposed to infer, imagine, and say if one didn’t have the belief. Not only is this far too unspecific to make Ryle’s point; the dispositions referred to are not, on their face, purely behavioral. Rather, they involve notions like inferring, imagining, and asserting that carry the same prima facie implications of internal states and processes the belief does, and that Ryle’s analysis is supposed to dispense with. Finally, when in other passages Ryle talks about notions like inferring, he is prone to use the notions of belief and thought in explaining them. So how can he appeal to inferring when explaining belief? It is no use looking elsewhere in The Concept of Mind for a more fully worked-out version of his position. Nowhere does he give, or even attempt, the kind of thoroughly behavioral analyses of mental language required by his positive view. In my opinion, this is not entirely an idiosyncratic failure on his part; the prospects for coming up with thoroughgoing behavioral analyses of belief and other mental notions are essentially nil. Why? For one thing, many of our beliefs are only distantly and tenuously related to any actions we might perform. It is one thing to try to give a behavioral analysis of the belief that one’s pants are on fire. There 20

Ibid., p. 44.

108

ORDINARY LANGUAGE PHILOSOPHY

we have some idea of the actions almost anyone who had that belief would perform. It is quite another thing to try to specify the behavioral dispositions that go along with believing that arithmetic is incomplete, that the battle of Gettysburg nearly turned the tide of the Civil War in favor of the South, or that some clouds resemble balls of cotton. Faced with examples like these, the most natural move is to appeal to linguistic dispositions—e.g., the disposition to sincerely assent to, or assertively utter, some sentence which one understands that means that arithmetic is incomplete, that the battle of Gettysburg nearly turned the tide of the Civil War in favor of the South, or that some clouds resemble balls of cotton. But this carries with it difficulties of its own. For one thing, it requires a non-circular account of sincerity that doesn’t presuppose belief or other closely related notions. For another, it requires an analysis, for arbitrary sentences s and propositions p, of what it is for s to mean p— an analysis of what it is for the sentences used by speakers to mean what they do that does not itself make any circular appeal to the beliefs and intentions of speakers for which Ryle needs an explanation. To say that the task of providing such an analysis is daunting is to understate the case. Moreover, Ryle—with his ordinary language philosopher’s antitheoretical approach to questions about meaning—was in no position to undertake it. A second and equally important problem is that even when a belief is tied fairly closely to action, the most plausible disposition corresponding to the belief is often one that itself involves another mental concept of the same sort as belief. For example, consider the belief that there is a beer in the fridge. The most obvious corresponding disposition would be a disposition to go to the fridge and get the beer, if one wanted a beer. But in citing this disposition, all we have done is to move from a statement about one kind of internal state—belief—to a statement about what one would do if one were in another kind of internal state—namely, the state of desiring a beer. Unless we can give an independent dispositional analysis of what it is to desire a beer, we are no closer to Ryle’s ultimate goal. Suppose we tried to identify some behavioral state corresponding to desiring a beer. The most natural choice would be the disposition to perform an action that one believed would result in one’s getting hold of a beer. But here we have moved from a statement about desire to a disposition to perform some act about which one has a certain belief. Our dispositional treatment of belief presupposes desire, and our dispositional treatment of desire presupposes belief. If, somehow, we were given a purely behavioral account of belief to begin with, we

RYLE’S CONCEPT OF MIND

109

might be able to use it to give a purely behavioral account of desire (or vice versa). But we don’t have a purely behavioral account of belief (or desire) to begin with. We are caught in a circle. The attempt to fill out the behavioral content of either notion presupposes the other. For this reason, it is hard to see how purely behavioral accounts of either belief or desire could ever be successful. If this is right, then the failure of Ryle’s official behaviorist project goes well beyond the defects in his execution of it.21

Explaining Action in Terms of Belief, Desire, and Emotion Two Contending Pictures

Much the same can be said of Ryle’s treatment of emotions in chapter 4. There, he attacks a certain conception of the explanation of action. One common way of explaining what a person has done is to give his reason for doing it. When we explain an action in this way, we typically ascribe a motivating factor to the agent—e.g., a desire of a certain sort or an interest of a certain kind—and perhaps also some cognitive factor, like a belief that the action will bring about the desired result, or satisfy the interest in question. To put it generically, the agent does x because he desires something and believes that doing x will bring that thing about. It is very natural to think of these explanations of action as describing the agent as being in an internal state, or as undergoing an internal process, of which he is often conscious and to which he has direct access, but which others can only infer as likely causes of his behavior. According to this picture, actions are caused by private, internal cognitive and motivational states of which the agent is often non-inferentially aware, because they are part of his mental life, but which others cannot experience and so must posit on the basis of behavioral evidence. 21 Strictly speaking, there is another possibility that would have to be ruled out in order to reach this conclusion—the possibility that even though independent analyses of belief and desire cannot be given, the totality of an agent’s beliefs and desires supervenes on (are necessary consequences of) the agent’s behavioral dispositions. There are two problems with this idea: (i) the supervenience claim appears to be false—see chapter 1 of Robert Stalnaker, Inquiry (Cambridge, MA: MIT Press, 1984); for discussion and (ii) from Ryle’s point of view, including as it does the conflation of the modalities of necessity, aprioricity, and analyticity, the supervenience claim would have to be treated as asserting the grossly implausible thesis that all true propositions about the beliefs and desires of the agent are apriori consequences of truths about the agent’s behavioral dispositions.

110

ORDINARY LANGUAGE PHILOSOPHY

Ryle thinks that this picture is absurd, and needs to be replaced. As noted earlier, the main reason he thinks this seems to involve an uncritical commitment to a form of verificationism which is not made explicit.22 A sign that there is something wrong with his reasoning is that if it established what he takes it to establish, then analogous reasoning would seem to establish that we could never know about the past, because it is no longer observable, and that we could never know about the objects posited by a good physical theory, if they were not themselves observable.23 On the other hand, if one admits that we can sometimes know on the basis of theory of the existence of things we are in no position to observe, then Ryle is open to the challenge that we can know of the existence of private events in the mental life of others on the basis of having a set of informal psychological hypotheses about the explanation of their behavior. So Ryle’s negative case suffers from a serious weakness. What about his positive alternative? He gives an example in which we say that a man did a certain thing out of vanity, or because he was vain. Ryle thinks that to say this is not to give an explanation of the action as one that is caused by a certain kind of internal state. Rather, he maintains that to say that someone is vain is to attribute to him a disposition to behave in certain ways. It is to say that, if given a chance, the man will talk about himself, boast, and so on. To say that a man did x because he is vain is, according to Ryle, to say roughly (i) that he has a disposition to behave in certain ways in certain conditions, (ii) that in this case those conditions were satisfied, and (iii) that to do x is to behave in one of those ways. To give vanity as the reason for a person’s action is supposed to be analogous to saying that a glass broke because it was brittle. Purely Behavioral Dispositions vs. Dispositions to Think and Feel in Certain Ways

In evaluating this position, it is important to keep a certain critical distinction in mind. We must distinguish the claim that vanity is a disposition from the claim that it is a purely behavioral disposition. It seems likely that character traits such as vanity either are dispositions, or at least involve dispositions. But it is doubtful that they are ever purely behavioral dispositions. Certainly, to be vain is not just to be disposed to behave in certain ways. In addition, being vain involves being disposed See Ryle, The Concept of Mind, p. 90. As noted earlier, there might even be a problem about behavioral dispositions, since unlike actual behavior, dispositions to behave are not always observable.

22 23

RYLE’S CONCEPT OF MIND

111

to think and feel in certain ways—e.g., to be prone to daydream of one’s successes, to reject and disbelieve criticisms, to believe oneself to be unusually worthy of admiration, to desire that others come to share this belief, and to be disposed to try to bring this about. If, as seems reasonable, these cognitive and motivational dispositions are dispositions to be in private mental states, not directly observable by others, then attributions of vanity to someone do not fit Ryle’s behaviorist picture. Rather, they involve inferences about the unobservable that he wants to avoid. There is a subsidiary point that should also be mentioned. Even in cases in which we explain an action by attributing a dispositional character trait to the agent, it may turn out that the explanation we give allows us to pick out an internal cause of the action in question. Suppose I say that Jones performed a certain action because he is vain. Let us grant that in saying that he is vain, part of what I am trying to convey is that he is disposed, in certain sorts of circumstances, to do things out of a desire to get people to share the exceedingly high opinion he has of himself. Now it may be clear from the context of utterance that the circumstances for Jones’s forming such a desire, and acting on it, are satisfied. If so, then my ascription of vanity to Jones allows one to infer an internal motivational state—an occurrent desire that he had at the time—which was one of the causes of his action. Thus, the dispositional nature of the character trait does not preclude it from playing a role in locating genuine causal antecedents of action in the mind of the agent. Finally, in addition to explanations of action that cite character traits like vanity, there are explanations of action in which we explicitly identify the cause of an action as including an occurrent desire of the agent. For example, one may explain why I went to the fridge by saying that (i) I was hungry and (ii) I thought there was food in the fridge. Here it seems that we have given a causal explanation appealing to private internal states of which the agent is directly aware, but of which others only have indirect evidence. This is the sort of thing that Ryle has not been successful in explaining away.

Non-Inferential First-Person Knowledge of Mental States Ryle also discusses other cases in which we seem to use language to describe feelings, moods, or emotions, even when we are not trying to explain action. One particularly interesting problem that he discusses concerns cases in which one describes one’s own state of mind—cases

112

ORDINARY LANGUAGE PHILOSOPHY

in which we say things like “I am feeling depressed” or “I am bored.” These cases are problematic for the general view Ryle is advocating. According to him, when we say that someone is depressed or bored, we are ascribing certain behavior and behavioral dispositions to the person. How do we know that a person is behaving in a certain way, and that he has certain further behavioral dispositions? Normally, we know these things by observing his behavior, and making reasonable inferences based on our past experience of similar behavior in him and others. Notice what happens if we raise the same questions about ourselves. Suppose that the claim that I make when I say “I feel depressed” or “I am bored” is a claim about my present behavior, and also about further dispositions regarding how I would behave if certain conditions were fulfilled. Given this assumption, one would naturally think that I could come to know that I was feeling depressed or bored only by observing my behavior, and drawing reasonable conclusions from my observation based on past experience of similar behavior in myself and others. But that doesn’t seem right. It seems that I can know that I am feeling depressed or bored, without observing my behavior or drawing any conclusions from it. So these cases don’t seem to fit Ryle’s model. On the contrary, they seem to fit the model he is attacking. If depression and boredom essentially involve internal states of my mind of which I am directly aware, then there is no need for me to observe my behavior, and make inferences from it, to find out that I am feeling depressed or bored. Ryle confronts this problem in the following passage. If a person says ‘I feel bored’, or ‘I feel depressed’, we do not ask him for his evidence, or request him to make sure. We may accuse him of shamming to us or to himself, but we do not accuse him of having been careless in his observations or rash in his inferences, since we do not think that his avowal was a report of observations or conclusions. He has not been a good or bad detective; he has not been a detective at all. . . . If the avowal is to do its job, it must be said in a depressed tone of voice; it must be blurted out to a sympathizer, not reported to an investigator. Avowing ‘I feel depressed’ is doing one of the things, namely one of the conversational things, that depression is the mood to do. It is not a piece of scientific premise providing, but a piece of conversational moping.24 24

Ibid., p. 102. See also p. 103.

RYLE’S CONCEPT OF MIND

113

In this passage Ryle correctly notes that first-person reports of one’s moods generally are not based on observation and evidence. Usually, we don’t discover certain facts about ourselves by observation, and then conclude from these facts that we are feeling depressed or bored. However, it is not fully clear what Ryle thinks is going on in these cases. The key lines, I think, are Avowing ‘I feel depressed’ is doing one of the things, namely one of the conversational things, that depression is the mood to do. It is not a piece of scientific premise providing, but a piece of conversational moping. One way of understanding these remarks is as follows: To feel depressed is to be disposed to behave in certain ways. Among these ways of behaving are sighing, crying, avoiding certain sorts of activities, and uttering the sounds “I feel depressed” in the right tone of voice (if one is an English speaker). Hence, one who does utter “I feel depressed” in the right tone of voice, and in the right circumstances, qualifies for that very reason as feeling depressed. One doesn’t have to do any further detective work to discover that it is true. One doesn’t have to make observations and draw conclusions. The mere fact that one has said “I feel depressed” in the right tone of voice guarantees that one does feel depressed. The mere performance of the speech act guarantees the correctness of what is said. This account is more explicit than Ryle’s discussion, and I am not completely sure what he would think about it. But I think it is a reasonable approximation of at least part of his story. The problem I find with the view is simply that it is neither necessary nor sufficient for feeling depressed that one be disposed to say the words “I feel depressed” in the right tone of voice (even if one is an English speaker). At a minimum, the analysis must require the remark to be sincere. If this involves one’s really feeling depressed, then it seems to take us back to treating the remark as a description of an internal state of which we can be directly aware. If the analysis only requires that one think that one is depressed, then we are faced with our original problem—namely, explaining the basis for this judgment. Since Ryle’s point was to give such an explanation, he has not succeeded. This does not mean that being depressed isn’t a dispositional state. But if it is, the dispositions involved are not just dispositions to behave in a certain way. Rather, they are dispositions to feel and think in certain ways—in the sense that Ryle seems to find problematic.

114

ORDINARY LANGUAGE PHILOSOPHY

Conclusions That completes my discussion of Ryle’s philosophy. In closing, I would like to highlight certain general features of the work that have been emphasized here. First, we have seen, both in Dilemmas and in The Concept of Mind, an attempt to apply to particular philosophical issues the general contention that problems in philosophy are the result of linguistic confusion, and are to be resolved by paying attention to how we use language in ordinary, non-philosophical contexts. Second, we noted that there is a kind of philosophical conservatism inherent in this methodology, since it would seem that nothing too surprising or revisionary of our ordinary ways of thinking could come out of it. Nevertheless, we have also seen that this conservatism was not consistently followed by Ryle. On the one hand, when discussing fatalism and skeptical views about perceptual knowledge, he seems confident from the outset that no philosophical argument could ever establish conclusions that so fundamentally conflict with what we pre-theoretically think. Thus he sets out, with varying degrees of success, to diagnose the linguistic errors in the arguments for these radically revisionary conclusions. On the other hand, when discussing the nature of the mental, and our discourse about it, Ryle himself harbors a radically revisionary agenda that encompasses the highly contentious doctrines of verificationism and behaviorism. A third, and by now familiar, lesson involves the blurring of modal distinctions. For Ryle, all necessity is conceptual necessity, and all conceptual necessity is linguistic in origin. Thus, all necessity is thought to be discoverable simply by reflecting on what we really mean by words, which itself can be made manifest by attending to how we use words in everyday life. I have argued that this blurring of modal distinctions blinded Ryle to certain live philosophical options, and marred his discussions of perception, the mind, and the relation between the description of the world found in physics and the descriptions we all give in ordinary life. Finally, I argued that although Ryle may be right in thinking that not all philosophical analysis involves specifying logical forms of sentences, or necessary and sufficient conditions for the application of concepts, nevertheless, some philosophical analysis is of this sort, and should not be neglected.

CHAPTER 5 STRAWSON’S PERFORMATIVE THEORY OF TRUTH chapter outline 1. Strawson’s Performative Theory To say that a statement is true is not to describe it, to attribute a property to it, or to make any statement about it; instead it is to perform the speech act of endorsing, conceding, or confirming it 2. Counterarguments Argument 1: Saying that some statements are true without endorsing any Argument 2: Predicating truth of the statements made by utterances of sentences containing true Argument 3: The systematicity of meaning 3. A Lesson about Explicitly Performative Sentences 4. General Lessons about Meaning and Use Two shortcomings of the approach of the ordinary language philosopher: (i) overlooking the systematicity of meaning, and (ii) overlooking factors distinct from meaning that have important effects on use; the case of meaningful sentences with no ordinary assertive use as complete utterances; Strawson’s treatment of ‘S’ is true iff S as a case study in the errors inherent in (i) and (ii); a nonsemantic explanation of facts about language use that motivated the performative theory of truth

In this chapter, we will examine Peter Strawson’s theory of truth, presented in his 1949 article, “Truth.” This article is a good illustration of several aspects of the ordinary language approach to philosophy.1 For one thing, it takes on a large and central philosophical issue, which is typical of the leading works of the ordinary language school. We have seen Ryle tackle problems concerning the nature of the mind, and skepticism about perceptual knowledge; we now turn to Strawson’s attempt to give an analysis of truth. The ordinary language school has 1

Peter Strawson, “Truth,” Analysis 9, 6 (1949):83–97.

116

ORDINARY LANGUAGE PHILOSOPHY

sometimes been criticized for focusing on petty details, and not dealing with the central, perennial problems of philosophy. These attacks are, I believe, misguided, as is illustrated by the importance of the problems discussed by Wittgenstein, Ryle, Strawson, and others. Another notable feature of Strawson’s paper is the way in which it tries to implement the methodological slogan that the meaning of an expression is given by its use. Stated in the abstract, this slogan is too vague to be informative. It is only when the philosophers of the ordinary language school tried to put it into practice that it began to become clear which aspects of the use of an expression are central to its meaning, and which are not. Strawson’s paper on truth represents an early stage in this process. As we progress, I will point out certain pitfalls that he fell into because the conception of meaning as use that he was trying to employ was too undifferentiated and undeveloped at the time he wrote. In this respect, his paper is instructive as one step along the path that this school of philosophy would follow. Finally, I would like to emphasize that although I think that Strawson’s paper is deeply misguided in important respects, I also find it to be a work of considerable intelligence, sensitivity, and even charm—like much of the philosophy produced by the ordinary language philosophers. It is clear, provocative, and a joy to read. There is much to be learned, both from its genuine insights and from its evident failures.

Strawson’s Performative Theory Strawson’s theory starts from the observation that there is a very close relationship between saying that it is true that S, and simply saying that S. In fact, Strawson thinks that in the two cases one asserts the very same thing. For example, consider the sentences: It is true that the earth is round. The proposition that the earth is round is true. That’s true (said in response to an assertion that the earth is round).

Strawson holds that someone who assertively utters one of these sentences asserts the very same thing that is asserted by an utterance of The earth is round. Hence, the presence of the truth predicate in these sentences adds nothing to what they are used to assert. For example, someone who assertively utters the sentence The proposition that the earth is round is true does not refer to a proposition and predicate the

STRAWSON’S THEORY OF TRUTH

117

property truth of it. There is no property truth, and the phrase is true never functions to describe anything. Why, then, do we have the predicate, and how does it function? According to Strawson, to say that it is true that S is to assert that S, and only that S; however, it is also to do something other than make an assertion; it is to perform the speech act of confirming, endorsing, or conceding that S. It is in its contributions to these speech acts that he locates the central function of the truth predicate. Strawson is right in maintaining that there are subtle conversational differences between uses of It is true that S and That S is true, on the one hand, and S, on the other.2 Utterances of sentences containing true often carry suggestions that are not carried by utterances of corresponding sentences that don’t contain true. For example, imagine the following question being asked at a presidential press conference: Is it true that you are considering sending troops to the Middle East? This question carries the suggestion that it has been reported, or predicted, or conjectured that the President is considering sending troops to the Middle East. Because of this, the question is naturally understood as a request for the President to confirm or deny those reports, predictions, or conjectures. The question could have been put another way, without using the truth predicate. If it had been so expressed, it might not have lent itself so readily to an interpretation in which it is understood as a request for confirmation or denial of a previously made report. For example, the reporter might simply have asked, Are you considering sending troops to the Middle East? This way of putting the question asks for the same information as before, but it doesn’t carry the suggestion that the questioner is merely following up a previous remark, rather than, perhaps, himself placing the issue on the public agenda. Here is another example: The President says, We have proposed a comprehensive program to balance the budget while stimulating the economy. It is true that unemployment has already begun to drop, but not fast enough, so we still need the stimulus package. Or, Yes, unemployment has already begun to drop, but not fast enough. . . . In this context, it is true that and yes are rhetorical devices used to indicate that the speaker is aware that people have said, or thought, that unemployment is already dropping without the stimulus package, and to concede that point, while trying to deny the further implication that the stimulus 2 In this sentence I use ‘S’ as a metalinguistic variable over sentences, and boldface italics as corner quotes. In the previous paragraph, ‘S’ was used as a schematic letter.

118

ORDINARY LANGUAGE PHILOSOPHY

package is unnecessary. Again, true is used to suggest the existence of something to which a response is called for. Strawson is very sensitive to these subtle facts about our use of the truth predicate; they are the central facts that his theory is designed to accommodate. However, they are not the central facts governing our notion of truth, and they do not provide the chief rationale for a truth predicate to begin with. As we will show in due course, once this is seen, one can give another, quite different, explanation of the subtle conversational facts that Strawson wrongly takes to be central. The central reason the truth predicate is such a useful part of our language is provided by cases in which what is said to be true is not directly displayed or presented. Some of these are cases in which the truth predicate is used to state important logical or semantic generalizations that would be difficult to state in any other way—for example, the generalization that for any declarative sentences A and B, if both A and if A, then B are true, then B is also true, or the generalization that for every declarative sentence, either it or its negation is true. One may be able to eliminate it is true that from It is true that the earth is round without loss of informativeness, but one can’t eliminate true from these generalizations without losing what one was trying to say. Another example of a use of the truth predicate in which what is said to be true is not directly displayed or presented is (1). 1. What the policeman said is true. Strawson says that someone who utters this sentence can be regarded as implicitly making what he calls the “second-order, existential metastatement” (2). 2. The policeman made a statement. According to Strawson, (2) can be regarded as part [but not all] of the analysis of [(1)]. . . . To complete the analysis, then, of the entire sentence “What the policeman said is true”, we have to add, to the existential meta-assertion, a phrase which is not assertive but (if I may borrow Mr. Austin’s word) performatory. We might, e.g., offer, as a complete analysis of one case, the expression: “The policeman made a statement. I confirm it”; where, in uttering the words “I confirm it”, I am not describing something I do, but doing something.3 3

Strawson, “Truth,” pp. 92–93, my additions in brackets. Emphasis his.

STRAWSON’S THEORY OF TRUTH

119

On this view, the complete analysis of (1) is (1A). 1A. The policeman made a statement. I confirm it. At this point we need to clarify what Strawson means by calling an utterance performatory. A couple of standard examples introduced by the philosopher J. L. Austin can be used to illustrate this.4 Consider a marriage ceremony. At the end of the ceremony the presiding official says “I now pronounce you husband and wife.” In saying these words, the official is performing a certain act with social significance; he is marrying two people. According to Austin, and to Strawson following Austin, the words the official says do not describe the act that has been performed—rather the saying of these words, in the right circumstances, is the act. When the official says these words, he is not asserting a proposition, he is not making a claim or saying anything that can be true or false. He is simply performing a certain kind of speech act. Another example from Austin is this. Imagine a duly constituted authority—perhaps the Queen of England—standing before a new ship at a christening ceremony. At the proper moment she utters the sentence I hereby name this ship the Gilbert Ryle, and smashes a bottle of champagne across the bow. According to Austin, and to Strawson following him, to do this is to perform the speech act of naming; it is not to describe oneself as having performed the act; it is not to assert any proposition; and so it is not to say anything true or false. It is to engage in an entirely different kind of linguistic performance. These examples both involve formal occasions. However, Austin and Strawson believed that many of our ordinary linguistic performances can be understood as conforming to this model. In giving his analysis of truth, Strawson presupposes that examples like I confirm that so and so, I endorse that so and so, and I concede that so and so are performatory in Austin’s sense. On this view, to utter one of these sentences in the right circumstances is not to assert that one has confirmed, endorsed, or conceded something; rather, it is to perform the speech acts of confirming, endorsing, or conceding something. So when Strawson says that the analysis of (1)—What the policeman said is true—is The policeman made J. L. Austin, “Performative Utterances,” in Philosophical Papers (London, Oxford, and New York: Oxford University Press, 1961), and How to Do Things with Words (New York: Oxford University Press, 1965). The latter is reconstructed from lecture notes for the William James Lectures delivered at Harvard University in 1955; the former is the transcript of a talk aired on the BBC in 1956. However, Austin tells us that the main ideas in these works were formed in 1939, and his views on these matters were well known in England by the late ’40s.

4

120

ORDINARY LANGUAGE PHILOSOPHY

a statement. I confirm it, an utterance of the first part of the analysis— namely The policeman made a statement—is taken as making an assertion, whereas an utterance of the second part of the analysis—namely, I confirm it—is taken to be a performance that does not involve asserting anything. Although this view is admirably clear, I will argue that it fails in three fundamental respects. After the theory has been refuted, I will go back and extract general lessons about the shortcomings it illustrates in Strawson’s ordinary language methodology, his conception of meaning, and his view of philosophical analysis.

Counterarguments5 Argument 1: Saying That Some Statements Are True without Endorsing Any

The first counterargument against Strawson’s theory is directed against claim A, which slightly extends his remarks about (1). A. In uttering What the policeman said is true, one confirms, endorses, or concedes what the policeman said (provided that he said something); in uttering Several/some/many things the policeman said are true, one confirms, endorses, or concedes several/some/many things the policeman said (provided that he said a number of things). In showing this claim to be false, it is useful to note the contrast between the following pairs of sentences. 3a. John is caring for several/some/many kittens. b. There are several/some/many kittens that John is caring for. 4a. John is looking for several/some/many kittens. b. There are several/some/many kittens that John is looking for. Whereas (3a) entails (3b), (4a) does not entail (4b). Note, in this connection, that one can look for the fountain of youth, even though there is no fountain of youth; and one can look for some Aztec cities of gold, even though there are no Aztec cities of gold. These examples 5 These arguments first appeared in chapter 8 of my Understanding Truth (New York: Oxford University Press, 1999).

STRAWSON’S THEORY OF TRUTH

121

show that a claim of the form John is looking for so and so’s can be true even though the corresponding claim There are so and so’s is false, in which case the claim There are so and so’s that John is looking for is also false. There is also another way in which a claim of the form (4b) can be false, even when the corresponding claim (4a) is true. Suppose that John is looking for some kittens because he wants some pets, even though there are no particular kittens he has in mind. In this sort of case (4a) may be true. However, (4b) can’t be true unless there are particular kittens that John has in mind and is trying to find. These examples show that the logical facts involving the verb care for differ in certain respects from the logical facts involving the verb look for. Now notice that the verbs used in Strawson’s performative theory— confirm, endorse, concede—are like care for and unlike look for in this respect. One cannot confirm, endorse, or concede something unless there is something that one confirms, endorses, or concedes. Thus, (5a) entails (5b). 5a. John has confirmed several/some/many statements in the report. b. Several/some/many statements in the report are such that John has confirmed them. This fact can be used to show that Strawson’s view is incorrect. Imagine that John has some, but not complete, confidence in a certain policeman who has written a report. The policeman typically gets most of what he says right, but from time to time makes mistakes. Not knowing the contents of the report, John might nevertheless feel confident enough to claim that many statements in the report are true. Despite this, there is not even one statement in the report such that John has said that it is true. So there is not even one statement in the report that John has confirmed or endorsed. Therefore, it is not the case that many statements in the report are such that John has confirmed or endorsed them. But then, since (5b) is false, (5a) is also false. John has said that many statements in the report are true. But he has not confirmed or endorsed any of them. Thus (A) is false. Ironically, this result alone does not show that sentences of the form (i) differ in meaning from those of the form (ii). (i) I (hereby) endorse (confirm, etc.) some/several/many statements in the report. (ii) Some/several/many statements in the report are true.

122

ORDINARY LANGUAGE PHILOSOPHY

The reason it doesn’t show this is that utterances of sentences of the form (i) typically do not themselves constitute endorsements, confirmations, etc. To see this, consider an analogy. Imagine a duly constituted authority standing before several new ships at a christening ceremony. At the proper moment, she utters the sentence I hereby name at least one of these ships the J. L. Austin, and sits down. She does not go on to specify which ship she is supposedly naming, but simply lets her performance stand. In such a case she has not named anything; her performance is flawed, even absurd. In the absence of special stage setting (specifying which statements are being endorsed or confirmed), utterances of sentences like (i) share these defects. They are not endorsements or confirmations of anything; rather they are flawed, even absurd performances (unless they are construed as straightforward descriptions). Of course, this is not the case with utterances of (ii). This fact shows that sentences of type (ii) do not have the meanings of sentences of type (i). This point carries over to (6a), which cannot be analyzed as (6b).6 6a. Many/some statements that the policeman made are true. b. The policeman made a number of statements. I endorse (confirm) many/some of them. Argument 2: Predicating Truth of the Statements Made by Utterances of Sentences Containing ‘True’

The next argument shows that a primary function of the truth predicate in sentences containing it is descriptive. The idea is simple. We start with the observation that the things we evaluate for truth are statements or assertions. We don’t call questions, orders, or commands true. If someone says “Close the door!” or “Hello,” we don’t say “That’s true.” Utterances of sentences containing true are not concerned with all linguistic performances; rather, they are used to assess statements (i.e., that which is stated or asserted). Now notice that utterances of sentences containing the word true make claims that can It is important to distinguish this criticism of the performative analysis from the view that one can never endorse or confirm a statement unless one is in a position to make that statement oneself. If the policeman happened to make exactly three statements, and I say “I endorse every statement he made,” then it could be argued that I have in fact endorsed each one, even though I don’t know what exactly the policeman said. However, this point makes no difference in a case like (6b), the problem with which is that there is no possible way for anyone to single out the specific statements that are supposed to have been endorsed.

6

STRAWSON’S THEORY OF TRUTH

123

themselves be assessed by further uses of true. So these utterances must themselves make statements. Once this is seen, it is easy to show that the truth predicate itself contributes to the statements made by utterances of sentences containing it. For example, suppose that a person x testifying before a legislative committee says several things, including something false—i.e., in addition to asserting several truths, x asserts something false. The next day, y, testifying before the committee, makes several true claims, plus the claim (7). 7. Every assertion made by x to the committee is true. What is the status of y’s remark that every assertion made by x is true? Clearly it is false. Strangely, the performative theory does not allow one to acknowledge this. According to it, the only assertion made by y’s utterance of (7) was the assertion (8), which is true. 8. x made some assertions to the committee. Hence, according to the performative theory, all of y’s assertions were true; none was false. Hence, the proponent of the performative theory is committed to (9). 9. Every assertion y made to the committee is true. Since this is incompatible with the datum that y asserted something false, the performative theory fails. What these two counterarguments show is that the performative theory gives the wrong analysis of the use and meaning of sentences in which we predicate truth of sentences or propositions that we merely describe, but do not display. Since these sentences provide the main reason why we need a truth predicate in the first place, the performative theory misses the most important fact about the truth predicate. However, this is not its only error. Argument 3: The Systematicity of Meaning

In addition, the performative theory fails to properly distinguish meaning from use. For example, it is correct to observe that the sentence The proposition that S is true is often used, when uttered assertively, to endorse, confirm, or concede the proposition expressed by S. However, it is a mistake to think that this observation provides an analysis of the meaning of true, or of sentences containing it. To suppose otherwise is to ignore a crucial requirement on analyses of meaning—namely, that

124

ORDINARY LANGUAGE PHILOSOPHY

an adequate analysis of the meaning of an expression must specify the contribution made by the expression to the meanings of larger sentences or discourses in which the expression may be embedded. When these larger linguistic contexts are considered, it becomes obvious that the content (meaning) of The proposition that S is true has nothing essential to do with the speech acts that the performative theory associates with it, and that the sentence may be used with its literal meaning even though no such speech acts are performed, reported, or are in any other way in the offing. This point is illustrated by the following pairs of examples.7 10a. Either the proposition that there are three consecutive 7’s in the decimal expansion of ␲ is true or the proposition that there are not is true. b. Either I endorse/confirm/concede that there are three consecutive 7’s in the decimal expansion of ␲ or I endorse/ confirm/concede that there aren’t. 11a. I don’t know whether the proposition that budget deficits cause inflation is true. b. I don’t know whether I endorse/confirm/concede that budget deficits cause inflation. Obviously, the (b) sentences cannot be taken to be analyses of the (a) sentences. Moreover, the (b) sentences sound bizarre. They naturally read as if the clauses containing the performative verbs—endorse, confirm, concede—are being used performatively, even though the linguistic environment in which they occur is not appropriate for clauses used in this way. An example of a slightly different sort is provided by (12). 12a. If the proposition that there are three consecutive 7’s in the decimal expansion of ␲ is true, then there are three consecutive 7’s in the decimal expansion of ␲. 12b. If I endorse/confirm/concede that there are three consecutive 7’s in the decimal expansion of ␲, then there are three consecutive 7’s in the decimal expansion of ␲. 7 The point illustrated by these examples was made by Peter Geach, “Ascriptivism,” Philosophical Review 69 (1960), and John Searle, “Meaning and Speech Acts,” Philosophical Review 71, (1962).

STRAWSON’S THEORY OF TRUTH

125

Whereas (12a) is obviously correct, (12b) is more than a little doubtful, which suggests that the latter is not the analysis of the former. However, there is also something else of interest about this example. Although I endorse/confirm/concede that there are three consecutive 7’s in the decimal expansion of
is a sentence that can be used, on its own, to perform the act of endorsing, confirming, or conceding, when it occurs as the antecedent of (12b) it is not used performatively, but instead makes a straightforwardly descriptive contribution to the meaning of the entire sentence—which expresses the claim that if I perform a certain act of endorsing, etc., then a certain fact will hold. Should examples like these be taken to be counterexamples to the performative theory of truth or not? On the one hand, the proponent of the theory might argue that they should not be, since on his view truth claims are performative, whereas in these linguistic environments the explicitly performative sentences in terms of which such claims are normally analyzed do not carry their standard performative force. However, this is not an effective defense, since it avoids counterexample only by failing to give any analysis whatsoever of many important and highly significant occurrences of the truth predicate. On the other hand, the proponent of the performative theory of truth might acknowledge that when true occurs in linguistic environments like (10–12), it carries the same meaning it does in the simple sentences the theory was designed for. But then, the failure of the (b) sentences to provide analyses of the (a) sentences provides evidence that the performative theory of truth fails to give the correct analysis of any sentences of the sort The claim/statement/proposition that S is true. Further illustrations of this lesson are provided by (13) and (14). 13a. The ancients believed that the proposition that Hesperus is a star was true. b. The ancients believed that I endorsed/confirmed/conceded that Hesperus is a star. 14a. The proposition that the earth is round would have been true even if I hadn’t endorsed/confirmed/conceded it. b. I would have endorsed/confirmed/conceded that the earth is round even if I hadn’t endorsed/confirmed/conceded it. The (a) and (b) sentences of these pairs clearly differ in meaning, and even truth value. In addition, the members of the pairs are alike except for containing corresponding grammatical variants of (15a) and (15b).

126

ORDINARY LANGUAGE PHILOSOPHY

15a. The proposition that S is true b. I endorse/confirm/concede that S. Therefore, it is proper to conclude that (15a) differs in meaning from (15b). The same holds for other constructions involving the truth predicate. Thus, it would seem that the performative theory fails to give the correct analysis of the meaning of any sentence in which the word true appears. It should be noted that the argument based on examples (10–14) cannot be avoided by modifying the (b) sentences of those pairs. One might think, for example, that the analysis of (11a) should not be (11b), but rather should be I don’t know whether I should endorse/confirm/concede that budget deficits cause inflation. However, this suggestion runs up against the datum that sentences of the form (15a) don’t change meaning when they are embedded in one or another linguistic environment. Thus if I should endorse/confirm/concede that S is to be the analysis of (15a) in a case like (11), then it must serve as the analysis across the board. Since it cannot do so in all examples, it cannot be the correct analysis. One uniform analysis that might be suggested is the following: The proposition that S is true means The proposition that S is worthy of endorsement. But to adopt this view would be to give up the performative aspect of the theory altogether. To say that a proposition is worthy of endorsement is to assert that it has a certain property. Thus, on this view, true would function as a genuine predicate, and the property truth would be identified with the property of being worthy of endorsement, contrary to the performative theory. Moreover, the identification would be questionable. Although S is obviously equivalent to The proposition that S is true, it is not obviously equivalent to The proposition that S is worthy of endorsement. Sometimes propositions are true even though we have no evidence for them and so, presumably, we should not endorse them; there are even cases in which they are false, even though all available evidence points in their favor. For all these reasons, we have little option but to regard the problems facing the performative theory of truth as insurmountable.

A Lesson about Explicitly Performative Sentences At the time Strawson presented his performative theory of truth, he, along with other ordinary language philosophers, believed with Austin that explicitly performative sentences such as I promise to return the

STRAWSON’S THEORY OF TRUTH

127

book had a special kind of non-descriptive performative meaning. Unlike utterances of their descriptive counterparts, utterances of such sentences were not supposed to count as assertions, but rather were taken to be performances of the acts that they might otherwise have been thought to describe. Whether or not this idea is correct, it did, I think, naturally lead some to the further conclusion that such sentences do not express propositions that represent (describe) the world as being a certain way, and hence determine truth conditions for those sentences. However, there is a problem with this view that is brought out by the factors we have just been considering. Often, explicitly performative sentences can occur in larger linguistic environments, such as antecedents of conditionals, where they carry no special performative force.8 This is true of I promise to return the book. Although, in the right circumstances, uttering this sentence on its own may count as making a promise to return the book, to utter 16. If I promise to return the book, then you can be confident that it will be returned. is not itself to promise, but rather to make a straightforward assertion. This is relevant because, presumably, the sentence I promise to return the book has the same meaning when it occurs on its own as it does when it occurs as the antecedent of a conditional. Since in the latter environment it expresses a normal (descriptive) proposition that contributes to the proposition expressed by the conditional as a whole, it must also do so when it occurs on its own. In this sense, the explicitly performative sentence I promise to return the book has a straightforwardly descriptive meaning that represents the world as being a certain way, and so imposes truth conditions on it. What makes this sentence special, and performative, is that one can bring it about that these truth conditions are satisfied (i.e., one can bring it about that the proposition expressed by the sentence is true) simply by uttering the sentence in the right circumstances to the person to whom the promise is being made. This, it seems to me, is the broader lesson about the nature of explicitly performative sentences to be learned from examples like (12b), (13b), (14b), and (16). There is, however, one remaining question to be answered. What explains the fact that to assertively utter I promise to return 8 This point does not apply to intrinsically performative sentences like Hello. Such examples do have special performative meanings that specify the speech acts performed by uttering them, and they do not occur as antecedents of conditionals, or in other linguistic environments that require sentences expressing propositions.

128

ORDINARY LANGUAGE PHILOSOPHY

the book in the right circumstances counts as making a promise? Although there are different possible answers to this question, the one I find most plausible and intriguing is this: it is part of the concept of promising (the meaning of ‘promise’) that to assert to x that one promises x to do something is, in the right circumstances, to promise x to do it. If this is right, then someone who assertively utters I promise to return the book in the right circumstances does assert a proposition after all; but it is one that is made true by his very assertion of it.9 This suggests, not just that Strawson’s performative theory of truth is incorrect, but that the conception of performative uses of language that guided him may have been defective as well. In fact, even if we put this last point aside—about selfvalidating assertions of the propositions expressed by explicitly performative sentences—it seems clear that the ordinary language philosophers of the 1940s and 1950s exaggerated the extent to which explicitly performative sentences like I promise to return the book and I confirm that the President is planning to send troops to the Middle East differ from ordinary non-performative sentences. These philosophers were simply wrong to the extent that they supposed that, due to some sort of special performative meaning, these sentences do not express ordinary, descriptive propositions that provide the sentences with straightforward truth conditions.10

General Lessons about Meaning and Use The problems with Strawson’s performative theory of truth point to an important general lesson about the relationship between meaning and use that must be kept in mind by any philosopher who wants to examine the use of an expression as a guide to determining its meaning. The lesson is that the meaning of an expression is something it contributes to the meanings of all larger sentences of which it is a part; thus, if use is to be a guide to meaning, one must not limit one’s attention to simple (non-compound) sentences containing the expression.11 Linguistic meanFor more on this view, see E. J. Lemmon, “On Sentences Verifiable by Their Use,” Analysis 22 (1962):86–89, and David Lewis, section 7 of “Scorekeeping in a Language Game,” Journal of Philosophical Logic 8 (1979):339–59, reprinted in his Philosophical Papers, vol. 1. 10 The early views of Austin that so attracted the attention of ordinary language philosophers in the 1940s and ’50s matured in the ’60s and ’70s into a systematic study of speech acts and their relation to linguistic meaning. One of the most useful works of that period is John Searle’s Speech Acts (Cambridge: Cambridge University Press, 1969). 11 I here ignore certain exceptional cases, for example, sentences in which the expression under investigation occurs within quotes. 9

STRAWSON’S THEORY OF TRUTH

129

ing is systematic. Standardly, the meaning of a complex expression is determined by the meanings of its parts. If this were not so, we could not explain how language users are routinely able to understand new sentences that they have never previously encountered. In order to account for this fact, we need a theory of the meaning of an individual expression that makes clear how it is able to systematically contribute to the meanings of larger linguistic compounds that contain it. Strawson’s performative analysis of the truth predicate doesn’t do this. I haven’t, of course, said anything about what a theory of meaning should look like, if it is to capture the systematic nature of linguistic meaning. At the time that Strawson wrote his truth paper, this problem had not yet received the intense attention from philosophers that it would later receive. In Part 6, when we take up Donald Davidson’s views on truth and meaning, developed in the mid-1960s, we will discuss this topic in some detail. For the present, I want to contrast the approach to meaning as something systematic with Strawson’s approach. His approach—which he shared with the later Wittgenstein and other ordinary language philosophers—was something like this: When given a sentence, the ordinary language philosopher would try to determine in what circumstances the sentence would ordinarily be used, as a more or less complete utterance. If the philosopher could not find such circumstances, he would be inclined to dismiss the sentence as meaningless, or as making only a pseudo-statement. If he could find circumstances in which the sentence would be used as a complete utterance, then he would look to see what speakers in such circumstances would normally be using the sentence to accomplish, or get across. When the philosopher had brought this out, he would take himself to have elucidated the meaning of the sentence. One shortcoming of this approach is that it overlooks the idea that the meaning of a sentence is only one factor in determining whether it will be used in a given situation. Other factors include what speakers and hearers take to be obviously true, and hence not worth saying, or obviously false, and hence incorrect to say, as well as things that are obviously irrelevant to the conversation. Factors like these are crucial to bear in mind, if one is to avoid what is otherwise a virtually irresistible mistake inherent in the methodology of studying the meaning of a sentence by concentrating exclusively on cases in which the sentence would be used on its own as a more or less complete utterance. One who follows this methodology will be tempted to conclude, as Strawson does, that a sentence that has virtually no significant use as a stand-alone utterance is

130

ORDINARY LANGUAGE PHILOSOPHY

meaningless, absurd, or degenerate—and hence does not encode any significant information. That this is a mistake is illustrated by examples such as I am not speaking, I never use language, I don’t exist, and The English sentence ‘the earth is round’ is true iff the earth is round. These sentences would virtually never be assertively uttered on their own, as free-standing assertions, in normal circumstances involving ordinary speakers. Nevertheless, they are perfectly meaningful, and they encode significant, nontrivial information, as is indicated by their contributions to the meanings of the following larger sentences containing them. Bill believes that I am not speaking. Bill thinks that I never use language. That woman looks forward to the day when I don’t exist. Since Juan doesn’t know English, he doesn’t know that the English sentence ‘The earth is round’ is true iff the earth is round, even though he does know that the earth is round. That Strawson misses this point can be seen by looking at certain representative passages from the truth paper. The first passage occurs immediately after a quote from Carnap, about which he says the following: It will be noticed that the expressions ‘truth-condition’ and ‘meaning’ are used synonymously. And this suggests that even if there is no use of the phrase ‘is true’ in which that phrase is correctly applied to (used to talk about) sentences, there is, or might be, a use of the phrase ‘is true if and only if’ in which this phrase is correctly applied to (used to talk about) sentences; a use, namely, in which this phrase would be synonymous with the phrase ‘means that’; which certainly is used to talk about sentences. Suppose, for example, that we wish to give information about the meaning of the sentence ‘The monarch is deceased’. We can do this by making the following meta-statement. . . : (ia) ‘The monarch is deceased’ means in English that the king is dead. . . . . . . Now the suggestion is that we might, without unintelligibility, give the same information in exactly the same way, except that we should replace the phrase ‘mean[s] that’ with the phrase ‘is true if and only if ’, obtaining the contingent meta-statement: . . . (iia) ‘The monarch is deceased’ is true in English if and only if the king is dead.12 12

Strawson, “Truth,” pp. 85–86.

STRAWSON’S THEORY OF TRUTH

131

There are two noteworthy problems with this passage. First, since is true (in English) if and only if does have a use in which it is correctly applied to sentences (taking the name of a sentence as one of its arguments), it cannot be that the predicate is true (in English) has no use in which it is correctly applied to sentences. The former expression—is true (in English) if and only if—is a linguistic compound, the meaning of which is determined by its constituents—is true (in English) and if and only if. The expression if and only if is not applied to, or predicated of, sentences, as is shown by the fact that it doesn’t take names of sentences, including quote-names, as arguments. Rather, it connects unquoted sentences. Thus, the compound phrase is true (in English) if and only if can apply to sentences (thereby taking the name of a sentence as one of its arguments) only if the predicate is true (in English) can itself be applied to sentences. The fact that Strawson doesn’t see this illustrates his failure to recognize the systematicity of meaning. The second problem with the passage is that, contrary to what it suggests, statements of truth conditions cannot be equated with statements of meaning. The intelligibility of is true (in English) if and only if when applied to sentences is not due to its synonymy, or even its equivalence, with means (in English) that. It is not synonymous or (strongly) equivalent with means (in English) that. Rather, its meaning is compositionally determined from the meanings of is true (in English) and if and only if. A statement ‘S’ is true (in English) if and only if P is true just in case P and the claim ‘S’ is true (in English) have the same truth value. It is not required that S and P mean the same thing, or that the two be equivalent in meaning.13 The problems get worse as Strawson goes on. He continues his remarks as follows: [W]e shall be using the phrase ‘is true if and only if’, in a contingent statement, to talk about a sentence. Now consider a degenerate case of such meta-statements: the case exemplified in the sentences: (iii) “The monarch is deceased” means (in English) that the monarch is deceased. ‘S’ and ‘P’ are here used as metalinguistic variables over sentences. Boldface italics indicates corner quotes. In this paragraph I use the assumption that if and only if is equivalent to the material biconditional. As a claim about English, this is doubtful. However, strictly speaking, the assumption is not needed for my conclusion: on any reasonable semantics, P if and only if Q may be true even when P and Q are not synonymous (for example, it will be true when both P and Q are necessary truths). In addition, the use of if and only if in truth-conditional approaches to semantics of the roughly Tarskian kind that Strawson seems to be talking about, standardly is one in which they are treated as material biconditionals.

13

132

ORDINARY LANGUAGE PHILOSOPHY

(iv) “The monarch is deceased” is true (in English) if and only [if ] the monarch is deceased. It is difficult, and, perhaps, for the present purpose, not very important, to decide which status to assign to such sentences as these. . . . . . . If we demand what contingent matter they state . . . no answer is possible. One cannot state what a sentence means without the help of another sentence. For these reasons, I propose to continue to refer to statements (or pseudo-statements) like (iii) and (iv) not as necessary, nor as contingent, but simply as ‘degenerate cases’ of contingent metastatements.14 What is going on in this passage is something like the following: From the fact that one would virtually never use the sentence ‘S’ means (in English) that S or ‘S’ is true (in English) if and only if S to make a statement informing someone of the meaning, or truth conditions, of S, Strawson concludes that these sentences do not make either necessary or contingent statements. Rather, he says that they are degenerate cases, and make only pseudo-statements. This mistake comes from ignoring the fact that there is a trivial reason that assertive utterances of these sentences would normally not be informative. Anyone who understood the sentences would be expected to know that they are true; hence, there is normally no need to assertively utter them. Still, the sentences are perfectly meaningful by virtue of the meaningfulness of their parts. Moreover, their meanings make crucial contributions to the meanings of larger sentences in which they occur, as is shown by the following perfectly coherent examples. Juan knows that Bill Clinton was President of the United States, but since Juan doesn’t know English, he doesn’t know that the sentence ‘Bill Clinton was President of the United States’ means (in English) that Bill Clinton was President of the United States. In Old English, the sentence ‘London is dirty’ didn’t mean that London is dirty, but in modern English it does. It could have been the case that the sentence ‘Snow is white’ didn’t mean (in English) that snow is white. These observations point to one way in which use can differ from meaning. A sentence can have a perfectly coherent meaning and yet 14

Strawson, “Truth,” 86–87.

STRAWSON’S THEORY OF TRUTH

133

have no significant use as a stand-alone assertive utterance, because of obvious facts about what is involved in using sentences in conversational settings to convey information. In addition to this, there is another difference between meaning and use that is directly relevant to Strawson’s performative theory. In motivating his theory, Strawson noted certain conversational differences between utterances of sentences of the form It is true that S and The claim (or proposition) that S is true, on the one hand, and normal utterances of ‘S’, on the other. Normal assertive utterances of sentences of the former sort often carry the suggestion that it has previously been claimed, suggested, or conjectured that S, and that the speaker is commenting on that claim, suggestion, or conjecture by confirming it, endorsing it, or conceding it. This is not usually true of simple assertive utterances of ‘S’. What is the reason for the difference?15 Strawson seems to have assumed that the difference in use must indicate a significant difference in meaning. But there is no need to accept this. The sentences S, It is true that S, and The claim that S is true may have the same or, at any rate, very similar meanings, provided that there is some non-semantic explanation for the significant difference in their use. And there is. Even though S and The claim that S mean obviously equivalent things, the latter refers to the claim made by S and describes it in a certain way. In many conversational circumstances there would be no reason for the speaker to go to the length of referring to, and describing, this claim, if all he wanted to do was to convey the information carried by S itself. If that were all he was interested in doing, he would normally just utter S. The fact that he chooses to convey essentially the same information indirectly, by referring to the claim made by S, and describing it as true, suggests that the claim made by S is already salient in the conversation setting, and that the speaker feels called upon to say something about it.16 It is plausible to suppose that the principle that generates this suggestion is a rule of linguistic communication, not linguistic meaning. The rule is, roughly: Don’t choose a more complicated way of saying something without reason when a simpler way is available. If some such rule typically guides conversations, then a speaker’s decision to use The claim that S is true rather than the simpler sentence, S, will typically signify that he has some reason to refer to the claim made by S. In this paragraph I have used ‘S’ as a schematic letter. In this paragraph, I have used ‘S’ as a metalinguistic variable over sentences. As before, boldface italics indicate corner quotes.

15 16

134

ORDINARY LANGUAGE PHILOSOPHY

Often the reason for doing this will be that the claim made by S has already been made salient in the context, and that the speaker feels obliged to comment on it. When the speaker does so by calling it true, this will count as confirming, endorsing, or conceding that claim, not because true has some special non-descriptive, performative meaning, but because it is part of the meaning of confirm, endorse, and concede, that to describe something as true is one way of doing these things. Thus, we can explain the conversational facts that motivated Strawson’s theory without adopting his discredited analysis of meaning.17 As we will see in chapter 9, Paul Grice makes essentially this point at the end of essay 3 of “Logic and Conversation,” originally given as lectures at Harvard in 1967, and published in his Studies in the Way of Words (Cambridge, MA: Harvard University Press, 1989). See pp. 55–57. 17

CHAPTER 6 HARE’S PERFORMATIVE THEORY OF GOODNESS chapter outline 1. Overview The view that to call something good is not to describe it, but to perform the speech act of commending it 2. Commending and Calling Something Good Critique of Hare’s view that to call something good is to commend it, which is always to guide choices 3. “Secondary” Descriptive Meaning Critique of Hare’s account of the so-called secondary descriptive meaning of x is a good N 4. The Primary, Evaluative Meaning of x is a Good N Use of the Geach-Searle argument to refute the view that Hare’s sentence means the same as I commend x as an N 5. Hare’s Generic Mistake: Ignoring the Systematicity of Meaning 6. Hare’s Anti-Descriptivism The argument that good doesn’t express a property His mistake in thinking that because of this, it can’t be descriptive; the case of attributive adjectives, big, old, tall, etc. 7. The Hypothesis That ‘Good’ is a Descriptive Predicate Modifier x is a good N iff (i) x is an N and (ii) x satisfies the contextually relevant interests taken in N’s to a higher degree than most N’s do Advantages of this analysis

Overview In the previous chapter, we examined Peter Strawson’s performative theory of truth, as an illustration of the ordinary language philosopher’s conviction that traditional problems of philosophy can be dissolved by linguistic analysis. According to Strawson, the perennial problems of defining truth, of explaining what truth is, and of discovering the relationship between language and the world on which truth depends were

136

ORDINARY LANGUAGE PHILOSOPHY

misguided. He thought that these traditional problems arose from failing to understand how the word true functions linguistically. His hope was that once we understood that true does not express any property, and that its function lies not in making statements, but in performing entirely different speech acts, the old philosophical problems involving truth would dissolve. We have seen that his hope was not realized. However, the failure of his performative analysis is much more clearly visible now, in hindsight, than it was at the time. During the decade after he published his analysis of truth, the idea of giving performative analyses of other philosophical notions became popular. Several different notions were subjected to this treatment, but none seemed more ripe for such analyses than evaluative notions expressed by words like good, bad, right, and wrong. By far the most well-known and influential performative analysis of evaluative terms was given by the Oxford philosopher R. M. Hare in his book The Language of Morals, published in 1952.1 In this chapter, we will talk about his analysis of the word good, which was the topic to which he devoted most of his attention. In the book, Hare spends most of his time analyzing uses of good as it occurs in sentences of the form x is a good N—e.g., good carpenter, good watch, good knife, good strawberry, and good sunset.2 Although all of these uses are evaluative, many are non-moral. His strategy is to focus on these because he thinks that the logical and semantic behavior of good is basically the same in moral and non-moral evaluations. Since non-moral evaluations are in certain ways simpler and less controversial, he takes it to be best to start with them. The overall picture he presents goes something like this: Although the primary meaning of the word good is evaluative, in many cases phrases containing it also acquire a secondary descriptive meaning which is parasitic upon, and generated by, the word’s primary evaluative meaning. The evaluative meaning of good is given by the observation that to call x good is to perform the speech act of commending x. Another way of commending x is, of course, simply to say I commend x. Hare maintains that someone who does this performs the act of commending, without asserting that he is commending, and indeed without asserting anything. Similarly, someone who calls x good is understood as commending x, without asserting that he is commending x, or has commended x, and without R. M. Hare, The Language of Morals (Oxford: Oxford University Press, 1952). Here I use ‘N’ as a schematic letter, rather than a metalinguistic variable. I will follow this practice throughout the chapter, unless otherwise noted. 1 2

HARE’S THEORY OF GOODNESS

137

attributing to x any property that is part of the primary meaning of the word good. On its primary evaluative meaning, good does not express any property.

Commending and Calling Something Good What is it to commend something? Although Hare doesn’t say very much about this, he does say that “when we commend or condemn anything it is always in order, at least indirectly, to guide choices, our own or other people’s, now or in the future.”3 So if to call something good is always to commend it, to do so must always be to guide choices in some way. This emphasis on guiding choices fits many cases quite well. If we are trying to decide what movie to go to and someone tells us that Ed Norton’s new film is a good movie, then it would be natural to take that remark to be an attempt to guide our choice. However, not all cases are this direct. We often say that certain things are, or were, good so and so’s even though we don’t envision ourselves or others having the opportunity to make choices on the basis of that information. Personally, I would say that Ronald Reagan was a good president of the United States, even though I don’t expect anyone to have the opportunity to vote for him again—or even for anyone very much like him. Or, to use a nice example due to my former student Rebecca Entwistle, I would say that when the College of Cardinals selected Pope John Paul II, they chose a good pope. I am willing to say this to people despite the fact that I know that none of them is in the College of Cardinals, and they will never have any occasion to choose a pope, or even to influence such a choice. How does this square with Hare’s idea that to call something good is always to commend it, where to commend it is always to guide choices, directly or indirectly? Hare responds to this sort of worry by saying “the choice that is envisaged need not ever occur, nor even be expected ever to occur; it is enough for it to be envisaged as occurring, in order that we should be able to make a value-judgment about it.”4 Does this address the problem? Although we can easily imagine being faced with choices that we know we will never have, it is hard to make sense of the idea of seriously attempting to guide a choice that we know is only imaginary. In light of this, it is difficult to give a clear and plausible interpretation to Hare’s 3 4

Hare, The Language of Morals, p. 127. Ibid., p. 128.

138

ORDINARY LANGUAGE PHILOSOPHY

claim that to call something good is to commend, and to commend is always to guide choices. Certainly by calling something a good N one is not always trying to guide the choice of the person to whom one is speaking; nor is one always trying to guide a choice involving the particular object one has called good. Even if these requirements are relaxed, it is difficult to understand all instances of calling things good as instances of guiding choices. Perhaps the idea is that some claims about x being good are to be thought of as aimed at an audience wider than one’s interlocutor, and as being about not only x, but other things of the same sort as x. Are we, then, to hold that when I say that the choice of John Paul II as pope was a good one, I am issuing a sort of universal imperative: To whom it may concern, whenever you are choosing a pope, choose one like John Paul II!? That sounds pretty strange, but it does fit with other things that Hare says. In section 2 of chapter 8, he says that to call x a good N is to commend not only x, but all other N’s that are relevantly similar to x. He expresses this idea by saying that all value judgments are covertly universal: . . . As we shall see, all value judgements are covertly universal in character, which is the same as to say that they refer to, and express acceptance of, a standard which has an application to other similar instances . . . When we commend an object, our judgement is not solely about that particular object, but is inescapably about objects like it. Thus, if I say that a certain motor-car is a good one, I am not merely saying something about a particular motor-car. To say something about that particular car, merely, would not be to commend. To commend, as we have seen, is to guide choices. Now for guiding a particular choice we have a linguistic instrument which is not that of commendation, namely, the singular imperative. If I wish merely to tell someone to choose a particular car, with no thought of the kind of car to which it belongs, I can say ‘Take that one’. If instead of this I say ‘That is a good one’, I am saying something more. I am implying that if any motor-car were just like that one, it would be a good one too; whereas by saying ‘Take that one’. I do not imply that, if my hearer sees another car just like that one, he is to take it too. But further, the implication of the judgement ‘That is a good motor-car’ does not extend merely to motor-cars exactly like that one. If this were so,

HARE’S THEORY OF GOODNESS

139

the implication would be for practical purposes useless; for nothing is exactly like anything else. It extends to every motor-car that is like that one in the relevant particulars; and the relevant particulars are its virtues—those of its characteristics for which I was commending it, or which I was calling good about it. Whenever we commend, we have in mind something about the object commended which is the reason for our commendation.5 So how does one’s calling something a good car guide choices? Not necessarily by telling one’s audience to choose the particular car one called good, but by telling one’s audience to choose cars that are relevantly like that one (if they are in the business of choosing cars). And what of the case of popes? Is Hare maintaining that in saying that John Paul II is a good pope I am directing everyone who may be choosing a pope to select one with the virtues I find in John Paul II? If he isn’t maintaining that, it is not clear what he is saying. It seems to me that in the passage just cited, Hare mixes together three points, two of them true and one false. The first true thing is that when we call something a good N, we nearly always have some reason for making that evaluation. Often, there are some features or properties of the object or individual in question that lead us to call it a good N, and hence to commend it. But if that is so, then in calling something a good N, we are implicitly using a standard which may be applied to other cases as well. If my entire reason for calling x a good N is that it has properties P and Q, then presumably I should be ready to call a different thing y of the same sort a good N if it has those same properties. Of course, this is not a special feature of the word good. For virtually any word or phrase w, if I say of something x, It is w, then normally I have some reason for doing so. Thus, in saying of x, It is w, I am often implicitly committed to a standard that may be applied to other cases. If my entire reason for predicating w of x is that x has properties P and Q, then presumably I should be ready to predicate w of a different thing y of the same sort, if y has those same properties— exactly as with good N. Thus, Hare’s correct observation about good is a perfectly general one that could be made about virtually any word or phrase that functions as a predicate. There is a second correct point here that could also be made. Every ordinary descriptive predicate is associated with standards of correctness that could in principle be applied to any object whatsoever. These 5

Ibid., pp. 129–30.

140

ORDINARY LANGUAGE PHILOSOPHY

standards are part of the descriptive meaning or content of the predicate. Thus, where P is such a predicate, we could regard it as part of the content of the claim x is P, that any object that satisfies these standards is such that to predicate P of it is to say something true. Of course, this point is not much help in analyzing the content of the claim x is a good N, unless good N is itself a descriptive predicate, and Hare denies that it is. Thus, Hare needs an analog to this point about descriptive content that applies to good N. This is where he appears to go wrong. He seems to suggest that it is part of the content (meaning) of my remark that x is a good N, that all N’s which share the virtues that I find in x are also good N’s, and that in commending x, I am implicitly commending them too. Why should we believe this? Suppose that my reason for calling Brian a good student is that I believe he got an A on the exam. Then, on Hare’s proposal, when I say, on this occasion, “Brian is a good student,” what I am saying is, in effect, that Brian, along with all the other students who got A’s on the exam, are good students. But that isn’t right. In general, there is a distinction between what I say and the reasons I say it. Two people can say the same thing, even if their reasons for saying it are different. Similarly, two people can both utter “Brian is a good student” and be saying the same thing, even if they do so for different reasons. In such a case, both might be commending Brian, and both might be willing to commend other students as well if asked, but their linguistic performance is a commendation of Brian only; which others they might be willing to commend is a matter not of what they say, but why they say it. Hare’s failure to take this into account in the passage cited earlier is not in itself of great moment. It would be innocent except for its role in defending his doctrines that to use the word good is to commend, and to commend is always to guide choices. As we have seen, those doctrines require him to twist simple examples like John Paul II is a good pope into something resembling universal imperatives. The counterintuitiveness of this result is masked by Hare’s general doctrine that because evaluative judgments are backed by reasons, they are all covertly universal. Once we see that this general doctrine is flawed, problematic counterexamples like the good pope are reinstated. What they show is that the connection between calling something a good N and guiding choices is considerably more indirect than Hare maintains. To recapitulate, Hare

HARE’S THEORY OF GOODNESS

141

claims that the primary meaning of the word good is performative, and is given by the observation that to call something a good N is to do essentially the same thing as to say I commend this N. If asked what it is to commend something, Hare will tell you that it is to attempt to guide possible choices. I have raised a problem for this view by arguing that it is doubtful that to call x a good N is always to guide choices. One might argue that to call x good is indirectly relevant to guiding choices in that it attributes a property to x that would constitute a reason for choosing x, if the opportunity ever arose—but that invokes the very descriptivism about the meaning of good that Hare wishes to avoid.

“Secondary” Descriptive Meaning Or does it? It is, after all, part of Hare’s view that x is a good N often carries a secondary descriptive meaning. Hare’s idea is roughly as follows: When I say that x is a good N, I am commending it as an N. I don’t commend N’s for no reason; rather, I look for certain properties. For example, I commend strawberries that are sweet, red, and ripe, I commend students who are intelligent, hardworking, curious, intellectually honest, and creative, and I commend knives that cut well and are easy to use. For different classes of things—strawberries, students, and knives—I look for different properties in determining the good ones of that kind. These different descriptive properties or standards are what Hare calls the different secondary, descriptive meanings of the phrases good strawberry, good student, and good knife. So, for Hare, the primary meaning of the sentence ‘x is a good student’ is its performative meaning, which is roughly that of ‘I commend x as a student’. However, since one who knows the accepted standards for commending students can gain descriptive information from such an utterance, the descriptive meaning of my utterance is that x is a student who is intelligent, curious, hardworking, etc. Hare brings this out in the following passage. The first similarity between ‘M is a red motor-car’ and ‘M is a good motor-car’ is that both can be, and often are, used for conveying information of a purely factual or descriptive character. If I say to someone ‘M is a good motor-car’, and he himself has not seen, and knows nothing of M, but does on the other hand know what sorts of motor-car we are accustomed to call ‘good’ (knows

142

ORDINARY LANGUAGE PHILOSOPHY

what is the expected standard of goodness in motor-cars), he undoubtedly receives information from my remark about what kind of motor-car it is. He will complain that I have misled him, if he subsequently discovers that M will not go over 30 m.p.h., or uses as much oil as petrol, or is covered with rust, or has large holes in the roof. His reason for complaining will be the same as it would have been if I had said that the car was red and he subsequently discovered that it was black. . . . This is a sense of ‘mean’ about which, as we have seen, we must be on our guard. . . . And so, both with ‘good’ and with ‘red’, there is this process, which in the case of ‘red’ we may call ‘explaining the meaning’, but in the case of ‘good’ may only call it so loosely and in a secondary sense; to be clear we must call it something like ‘explaining or conveying or setting forth the standard of goodness in motor-cars’.6 This passage shows that Hare can accommodate the fact that one can often gain descriptive information from claims to the effect that x is a good N. However, it is doubtful that he should conclude that this descriptive information is part of the meaning of the sentence x is a good N. For any case that he can make that it is part of the meaning of this sentence, there is a corresponding case to be made for the claim that similar descriptive information is part of the meaning of the sentence I commend x as an N. But the descriptive criteria I generally use for commending things to which N applies do not seem to be part of the meaning of that sentence. In general, not all information that one can gather from an utterance is part of the meaning of the sentence uttered. Moreover, the descriptive information that Hare points to regarding uses of the word good simply results from collateral information one has about the usual reasons for calling something good. Suppose that a certain manager will say “Happy New Year!” only to her superiors. Because of this, when you hear her saying this to someone whom you don’t know, you may gain the information that the person is a higherup in the organization. But that doesn’t show that the words the manager utters include as part of their meaning the claim that the person addressed is the speaker’s superior. Hare confronts this objection in section 3 of chapter 7. However, his response is equivocal. 6

Ibid., pp. 112–14.

HARE’S THEORY OF GOODNESS

143

Such objectors might hold that the meaning of ‘good’ is adequately characterized by saying that it is used for commending, and that any information we get from its use is not a question of meaning at all. When I say ‘M is a good motor-car’, my meaning, on this view, is to commend M; if a hearer gets from my remark, together with his knowledge of the standard habitually used by me in assessing the merits of motor-cars, information about what description of motor-car it is, this is not part of my meaning; all my hearer has done is to make an inductive inference from ‘Hare has usually in the past commended motor-cars of a certain description’ and ‘Hare has commended M’ to ‘M is of the same description’. I suspect that this objection is largely a verbal one, and I have no wish to take sides against it. On the one hand, we must insist that to know the criteria for applying the word ‘good’ to motor-cars is not to know—at any rate in the full or primary sense—the meaning of the expression ‘good motor-car’; to this extent the objection must be agreed with. On the other hand, the relation of the expression ‘good motor-car’ to the criteria for its application is very like the relation of a descriptive expression to its defining characteristics, and this likeness finds an echo in our language when we ask, ‘What do you mean, good?’, and get the answer ‘I mean it’ll do 80 and never breaks down’. In view of this undoubted fact of usage, I deem it best to adopt the term ‘descriptive meaning’.7 In this passage, Hare is impressed by the fact that when asked what one means by calling something a good car, one can respond by saying I mean that it goes fast and never breaks down. But answers to questions of this sort—What do you mean by saying that x is F?—are often requests, not for the linguistic meaning of one’s sentences, but for the reasons one has for predicating F of some particular thing. For example, one might ask: What do you mean such and such theory is false?— and get the answer, I mean that it wrongly predicts that names are not rigid designators, but the claim that the theory wrongly predicts something about names is not part of the meaning of the sentence such and such theory is false. The same point applies to Hare’s example involving the word good.8 Ibid., pp. 117–18. In this paragraph and the last, ‘N’ and ‘F’ are used as metalinguistic variables over predicates. 7 8

144

ORDINARY LANGUAGE PHILOSOPHY

The Primary, Evaluative Meaning of x is a good N Having defended his notion of secondary, descriptive meaning, Hare goes on to tell us what he means by calling the evaluative meaning the primary meaning of good: It is time now to justify my calling the descriptive meaning of ‘good’ secondary to the evaluative meaning. My reasons for doing so are two. First, the evaluative meaning is constant for every class of object for which the word is used. When we call a motor-car or a chronometer or a cricket-bat or a picture good, we are commending all of them. But because we are commending all of them for different reasons, the descriptive meaning is different in all cases. . . . The second reason for calling the evaluative meaning primary is, that we can use the evaluative force of the word in order to change the descriptive meaning for any class of objects.9 I will return to the question of descriptive meaning later. For now, I will simply take Hare as analyzing ‘this is a good N’ as having a meaning which may, roughly, be paraphrased as ‘I commend this N’.10 On this view, since both sentences are performatives, neither expresses a descriptive proposition, despite the fact that often one can obtain collateral descriptive information about the item referred to, if one knows the evaluative standards of the speaker. The sentences themselves—This is a good N and I commend this N—are not used, on their primary meaning, to assert anything, and so they are not subject to evaluation as being true or false. Instead, they are used to perform the speech act of commending. We are now ready to apply the Geach-Searle objection to Hare’s position. Searle gives the following examples:11 Ibid., pp. 118–19. In his article, “Meaning and Speech Acts” (Philosophical Review 79 [1970]), published, in response to criticism, eighteen years after the publication of The Language of Morals, Hare implicitly denies that he ever held that ‘This is a good N’ was synonymous with ‘I commend this as an N’. However, he admits that according to him, to know the (primary) meaning of ‘good’ is to know that to call something a good N is to commend it as an N. His reason for making this difficult-to-discern distinction has to do with his response to the Geach-Searle objection presented below. Whether or not the two sentences are strictly synonymous on his view, he certainly treats them as rough paraphrases, both of which are performative, nondescriptive, and non-truth-evaluable. 11 Peter Geach, “Ascriptivism,” Philosophical Review, 69 (1960), and John Searle, “Meaning and Speech Acts,” Philosophical Review, 71 (1962). The examples are from Searle, p. 425. 9

10

HARE’S THEORY OF GOODNESS

145

If this is a good electric blanket, then perhaps we ought to buy it for Aunt Nellie. I wonder if this is a good electric blanket. I don’t know whether this is a good electric blanket. Let us hope this is a good electric blanket. Each of these examples contains the clause this is a good electric blanket. In addition, the words in the sentence occur in these examples with their normal literal meanings. Thus, if the meaning of this is a good N is roughly the same as I commend this N, then Searle’s examples should be roughly equivalent to the following purported paraphrases: If I commend this electric blanket, then perhaps we ought to buy it for Aunt Nellie. I wonder if I commend this electric blanket. I don’t know whether I commend this electric blanket. Let us hope that I commend this electric blanket. Of course, these purported paraphrases are not anywhere near equivalent to the original sentences containing the clause this is a good electric blanket. Because of this, Searle concludes, correctly in my view, that Hare’s thesis about the meaning of sentences like this is incorrect. This argument can’t be avoided by fiddling with the details of the analysis. Suppose, for example, someone were to modify Hare’s thesis so that it said that this is a good N means not that I commend this N, but that I should commend this N. This modification would make some of the purported paraphrases come out more naturally. For example, I wonder if this is a good electric blanket would come out I wonder if I should commend this electric blanket. That sounds a bit better. Nevertheless, the proposal won’t work. First, the modified proposal does not produce reasonable paraphrases in all cases. For example, The Greeks believed that democracy was a good form of government is not equivalent to The Greeks believed that I should commend democracy. Second, and more importantly, the modified analysis just substitutes one evaluative word, should, for another evaluative word, good. Hare’s task is to give performative analyses of all these words; thus, he doesn’t want to take any of them as primitive. So the suggested modification doesn’t help.

146

ORDINARY LANGUAGE PHILOSOPHY

A different possible modification would be to claim that this is a good N means the same as this N warrants commendation. But again, this claim would defeat Hare’s purpose. Presumably, to say that something warrants commendation is to describe it as having a certain property. But it is central to Hare’s analysis that calling something a good N is not, on its primary interpretation, to describe it at all, apart from saying that it is an N. (If to say that something warrants commendation is not to describe it as having a certain property, then presumably warranting commendation is evaluative, and to analyze good in terms of it is subject to the criticism just made of analyzing one evaluative notion in terms of another.) Thus, the Geach-Searle argument refutes Hare’s analysis as stated. There seems to be no obvious restatement of the analysis that both preserves Hare’s central idea and avoids refutation.12

Hare’s Generic Mistake Hare’s mistake regarding good was like Strawson’s mistake regarding true. Both philosophers set out to answer the question “What do these words mean?” Since they accepted the slogan “Meaning is use,” they took their question about meaning to be the question, “How are these words used?” However, in trying to answer this question, they focused almost exclusively on how the words are used in simple sentences of the form x is true/good. They asked “How are these simple sentences used?” When they noticed that someone who utters such sentences typically performs certain speech acts, they mistakenly jumped to the conclusion that the meaning of these sentences lies simply in the fact that they are used to perform those speech acts, and hence that the meanings of the words they were interested in is given by this fact. What they missed was the systematic nature of both meaning and use. Linguistic expressions are used in all sorts of linguistic environments, 12 As mentioned earlier, Hare responds to the Geach-Searle objection in his 1970 article, “Meaning and Speech Acts.” The main burden of his response is to show how when S is a simple declarative sentence the meaning of which is (allegedly) given by specifying the speech act (other than assertion) it is standardly used to perform, S may occur in more complex linguistic environments—questions, negations, propositional attitude ascriptions, conditionals, etc.—even though one who utters one of the more complicated sentences doesn’t perform the speech act in question. Hare tries, in a highly programmatic way, to sketch how the meanings of uses of S in complex sentences of various types might, in principle, be explained in terms of their allegedly primary performative uses. In my view, the most glaring defect in his response is that he gives no plausible analysis of the propositional contents contributed by simple clauses containing ‘good’ to complex sentences containing them.

HARE’S THEORY OF GOODNESS

147

simple and complex. An account of the literal meaning of an expression must explain its contribution to all the linguistic contexts in which it occurs. The fact that simple sentences containing true or good are often used to perform certain speech acts does not provide us with analyses of their meanings, because it does not do this.

Hare’s Anti-Descriptivist Motivation For this reason, Hare’s performative analysis of the word good fails. Having said this, I want to look more closely at what motivated him to look for a performative analysis in the first place. Perhaps a way can be found to accommodate that motivation without subscribing to his positive theory. There is no doubt that a large part of Hare’s motivation for offering the performative theory came from his conviction that no purely descriptive analysis of the word good was possible. He takes up this issue in chapters 5 and 6 of The Language of Morals, where he tries to show that the word good, as it occurs in the phrase good N, does not stand for any property, simple or complex. Once he has shown this, he takes himself to have shown that good can’t be descriptive, and so must have some other kind of meaning. The crucial assumption he makes in chapters 5 and 6 is that if the primary meaning of good is descriptive, then it must stand for a property. Moreover, if it stands for a property, then a sentence x is a good N must be equivalent to a corresponding claim x is an N and x is C, where C specifies the properties that good stands for.13 This assumption is illustrated in the following passage. Let us suppose for the sake of argument that there are some ‘defining characteristics’ of a good picture. It does not matter of what sort they are; they can be a single characteristic, or a conjunction of characteristics, or a disjunction of alternative characteristics. Let us call the group of these characteristics C. ‘P is a good picture’ will then mean the same as ‘P is a picture and P is C’. . . . Let us generalize. If ‘P is a good picture’ is held to mean the same as ‘P is a picture and P is C’, then it will become impossible to commend pictures for being C; it will be possible only to say that they are C. . . . 13

Here again, ‘N’ is a metalinguistic variable, as is ‘C’.

148

ORDINARY LANGUAGE PHILOSOPHY

Let us illustrate this with another example. . . . Let us consider the sentence ‘S is a good strawberry’. We might naturally suppose that this means nothing more than ‘S is a strawberry and S is sweet, juicy, firm, red, and large’. But it then becomes impossible for us to say certain things which in our ordinary talk we do say. We sometimes want to say that a strawberry is a good strawberry because it is sweet, &c. This—as we can at once see if we think of ourselves saying it—does not mean the same as saying that a strawberry is a sweet, &c., strawberry because it is sweet, &c. But according to the proposed definition this is what it would mean.14 This passage is, of course, reminiscent of G. E. Moore’s open question argument—which is quite striking when one remembers that The Language of Morals was published forty-nine years after Principia Ethica. Much of Hare’s chapter 5 is given over to his own version of Moore’s argument—from which he draws the conclusion that good can’t be defined, and so can’t stand for any complex property. In chapter 6 he gives a more general argument designed to show that the word can’t stand for any property at all. That more general argument is essentially the following. Consider the different good-making properties for strawberries, knives, clocks, sunsets, etc. Roughly, something is a good strawberry iff it is sweet, red, and ripe. Something is a good knife iff it cuts well and is easy to use; something is a good clock iff it is durable and keeps time accurately; something is a good sunset iff it is beautiful and enjoyable to watch. When we consider all these cases, we see that the good-making characteristics for one class of things may be completely different from the good-making characteristics for another class of things. In particular, there is no property that is a good-making characteristic for objects of all arbitrary classes. Nevertheless, when we say x is a good strawberry, y is a good knife, z is a good clock, and s is a good sunset, the meaning of the word good does not change from case to case. Since there is no property that it picks out in all cases, the meaning of good is not any property at all. But then, since good doesn’t stand for any property at all, its meaning must not be descriptive. Rather, it must have some other kind of meaning, which is its primary meaning. This is given by its allegedly choice-guiding function to commend. Let us trace the way Hare make this argument in chapter 6. He begins the chapter by noting that we use the word good to evaluate objects of all sorts of different classes. We say x is a good N for all sorts of 14

Hare, The Language of Morals, pp. 84–85.

HARE’S THEORY OF GOODNESS

149

predicates N.15 Nevertheless, there is a common meaning of the word good that runs through all these cases. He characterizes the view that good is descriptive as claiming that this common meaning must be a property that all good things of any type share, just as the meaning of red is supposed to be a property that all red things share.16 He then runs through the argument I have just sketched, and concludes that the meaning of the word good isn’t any property at all. To teach what makes a member of any class a good member of the class is indeed a new lesson for each class of objects; but nevertheless the word ‘good’ has a constant meaning which, once learnt, can be understood no matter what class of objects is being discussed. We have, as I have already said, to make a distinction between the meaning of the word ‘good’ and the criteria for its application. . . . . . . To sum up: there is no common property which is recognizable in all cases where a member of a class—no matter what class— is said to be ‘instrumentally good’.17 What shall we say about this argument? The first thing to say is that Hare is right to conclude that good does not stand for a property. The problem comes when he moves from this correct claim to the conclusion that the meaning of good is not descriptive, but rather must be given by its role in guiding choices and commending things. This does not follow, because there is a descriptive alternative that he has failed to consider. Consider, for example, the word big. Like good, we use big in connection with objects of all sorts of different classes. We say x is a big N for all sorts of predicates N—big flea, big dog, big animal, big house, big bicycle, big city, big crowd, etc.18 There is, of course, a common meaning of the word big that runs through all these cases. However, if we look at the properties of big fleas, big dogs, big houses, big crowds, and so on, we will not find any single property that is a big-making property for each class of objects. It would clearly be wrong to conclude from this that big is not a descriptive word. So how does it function? Big is one of those adjectives that is not independently predicative. The statement x is a big N is not equivalent to the conjunction x is big ‘N’ is used as a metalinguistic variable. Hare, The Language of Morals, p. 97. This may be an oversimplification even with red. When we say that someone has red hair (or red skin after being out in the sun), we are standardly indicating a color that we would not call red in cherries, blood, or roses. 17 Ibid., pp. 102–3. 18 In this paragraph and the next, ‘N’ is again used as a metalinguistic variable. 15 16

150

ORDINARY LANGUAGE PHILOSOPHY

and x is an N. If it were, then it would follow from the claim that x is a big flea plus the claim that x is an animal that x is a big animal, which of course it does not. So how does big function? It is a predicate modifier. It is a modifier which, when attached to a predicate N, forms a complex predicate big N, which applies to an object iff (i) N applies to the object, and (ii) that object is larger, along some relevant dimension of size, than most of the objects to which N applies. Other descriptive predicate modifiers of this type are heavy, tall, young, and old.19 For example, x is a heavy N iff x is an N, and x weighs more than most N’s, x is a tall redwood tree iff x is a redwood tree and x’s height exceeds the height of most redwood trees, and so on.20

The Hypothesis that Good is a Descriptive Predicate Modifier In light of this, we can specify a descriptive alternative regarding the meaning of good, as it is used in expressions of the form good N, that Hare failed to consider. Perhaps good is a descriptive predicate modifier.21 On this view, x is a good N means something like (i) x is an N and (ii) x satisfies the contextually relevant interests taken in N’s to a higher degree than most N’s do. If this view is correct, then good is a descriptive word, even though it doesn’t stand for any property which the word is used to predicate of an object.22 There are several further points worth noticing here. First, if this view is correct, then it explains the general connection that Hare is so interested in between calling something a good N and guiding choices, without leading to his questionable doctrine that each and every case of calling something a good N is by its very nature directed at guiding some choice. On the view just sketched, if I say that John Paul II is a good pope, I am saying that he satisfies the contextually relevant interests taken in popes to a high degree. If, for example, those interests are the normal interests of members of the Catholic Church, then they will include being devout, wise, understanding, inspirational, and so on. And perhaps red, as noted three footnotes back. In this sentence I switch back to using ‘N’ as a schematic letter. 21 This point is forcefully made, along with the analogy between good and big by Peter Geach in “Good and Evil,” Analysis 17 (1956); reprinted in Philippa Foot, ed., Theories of Ethics (Oxford: Oxford University Press, 1967). 22 Here again, ‘N’ is a schematic letter. 19 20

HARE’S THEORY OF GOODNESS

151

Thus, in a conversation in which this is taken for granted, you may presume that I take John Paul II to possess these qualities in abundance. On this account, what I say when I say that he is a good pope is something that would be relevant to anyone who was ever in a position to make choices involving popes, even though my utterance to you on this occasion is not aimed at influencing any real or imagined choice of yours. This preserves the general connection between calling something good and guiding choices, without wrongly claiming that every time one calls something good one does so with the intention of guiding a choice. Second, the view provides a natural explanation of what Hare calls the secondary, descriptive meaning of phrases of the form good N. For example, on the predicate-modifier analysis, the sentence x is a good book means that x is a book that satisfies to a high degree the contextually relevant interests taken in books. If these interests are those of being informed and entertained, then the claim that x is a good book will carry the information that x is informative and entertaining. On the predicate-modifier account, this secondary, descriptive information arises from the interaction of the descriptive meaning of the sentence together with information about the contextually relevant interest taken in books. Third, the Geach-Searle argument, which refuted the performative analysis, does not pose any threat to the predicate-modifier analysis. Fourth, the analysis also withstands the open question argument. Since the conversationally relevant interests invoked by the analysis are those that are conversationally agreed upon as appropriate for evaluating N’s, the question Granted x satisfies the interests we agree to be appropriate for evaluating N’s to a higher degree than most N’s do, is x a good N? is obtuse; it is a trivially self-answering question. Finally, the predicate modifier analysis leads one to certain expectations about linguistic facts that seem to be born out. For example, it leads one to expect that when good is combined with a predicate that stands for a class of things that no one has much of any interest in—such as particles of dust—the resulting claim—e.g., that x is a good particle of dust—will seem strange and hard to make sense of. In addition, the analysis leads one to expect that when N stands for a class of things that people do have interests in, but interests of many different and varied kinds, it may be difficult to pin down what the compound predicate good N applies to. A case in point here is the example x is a good man. One reason it may be hard to get a definite handle on this is because it is hard to pin down exactly which interests are the relevant ones in a given

152

ORDINARY LANGUAGE PHILOSOPHY

context. In my opinion, all of this gives at least some credence to a descriptive predicate-modifier analysis of some uses of the word good. Whether the analysis, or some refinement of it, proves ultimately to be correct, and whether it applies to all uses of good, or only to a proper subset of them, are questions we will leave open.23 23

In this paragraph, ‘N’ is used as a metalinguistic variable.

SUGGESTED FURTHER READING FOR PART TWO

Main Primary Sources Discussed Geach, Peter. “Ascriptivism.” Philosophical Review, 69, (1960): 221–25. ———. “Good and Evil.” In Philippa Foot, ed., Theories of Ethics, (Oxford: Oxford University Press, 1967), originally published in Analysis 17 (1956): 33–42. Hare, R. M. The Language of Morals. Oxford: Oxford University Press, 1952. Ryle, Gilbert. The Concept of Mind. New York: Barnes and Noble, 1949. ———. Dilemmas. Cambridge: Cambridge University Press, 1953. Searle, John. “Meaning and Speech Acts.” Philosophical Review 71 (1962): 423–32. Strawson, Peter. “Truth.” Analysis 9, 6 (1949): 83–97.

Additional Primary Sources Austin, J. L. How to Do Things with Words. New York: Oxford University Press, 1965. ———. “Performative Utterances.” In Philosophical Papers (London, Oxford, and New York: Oxford University Press, 1961). Hare, R. M. Freedom and Reason. New York: Oxford University Press, 1965. ———Meaning and Speech Acts.” Philosophical Review 79 (1970): 3–24. Ryle, Gilbert. “Systematically Misleading Expressions.” Proceedings of the Aristotelian Society, 1931–32. Strawson, Peter. “Truth.” Proceedings of the Aristotelian Society, supplementary vol. 24, 1950.

Additional Recommended Reading Grice, Paul. “Logic and Conversation.” Originally given as lectures at Harvard in 1967; published in his Studies in the Way of Words (Cambridge, MA: Harvard University Press, 1989).

154

SUGGESTED FURTHER READING

Lemmon, E. J. “On Sentences Verifiable by Their Use.” Analysis 22 (1962): 86–89. Lewis, David. “Scorekeeping in a Language Game.” Reprinted in his Philosophical Papers, vol. 1 (New York and Oxford: Oxford University Press, 1983); originally published in Journal of Philosophical Logic 8 (1979): 339–59. Searle, John. Speech Acts. Cambridge: Cambridge University Press, 1969. Stalnaker, Robert. Inquiry. Cambridge, MA: MIT Press, 1984. Stanley, Jason, and Timothy Williamson. “Knowing How.” Journal of Philosophy 98 (2001): 411–44.

PA R T T H R E E MORE CLASSICS OF ORDINARY LANGUAGE PHILOSOPHY: THE RESPONSE TO RADICAL SKEPTICISM

This page intentionally left blank

CHAPTER 7 MALCOLM’S PARADIGM CASE ARGUMENT chapter outline 1. Moore’s Challenge to Skepticism The view that skepticism is based on premises of its own that cannot be justified 2. Malcolm’s Attempt to Extend Moore’s Challenge The view that skepticism is self-defeating; if its conclusions were correct, then the skeptic’s own words would be meaningless The need to reconstruct Malcolm’s argument to avoid his conflation of necessary falsehood with self-contradiction The crucial role of words the meanings of which are entirely constituted by their correct application in particular cases Malcolm’s failure to show that know is such a word; consequences for his general argument against skepticism 3. Limited Anti-Skeptical Lessons of the Paradigm Case Argument The difference between diagnosing (semantic) incoherence in a skeptical view and refuting the skeptic on grounds that even he must accept The potential significance of versions of the paradigm case argument for diagnosing incoherence and error in certain limited forms of skepticism

Moore’s Challenge to Skepticism In Part 2, we looked at attempts by ordinary language philosophers to dissolve philosophical problems about truth, goodness, and the mind by applying the techniques of ordinary language analysis to sentences containing true, good, and various terms purporting to denote mental states and sensations. In Part 3, we will examine attempts to deal with radical philosophical skepticism within the ordinary language framework. Our first topic is a famous and very interesting argument known as the paradigm case argument, first presented by Norman Malcolm in his article on G. E. Moore, “Moore and Ordinary Language,” originally published in 1942.1 Malcolm was an American who had studied in Norman Malcolm, “Moore and Ordinary Language,” in The Philosophy of G. E. Moore, vol. 1, ed. P. A. Schilpp (La Salle, IL: Open Court, 1942), 343–68.

1

158

THE RESPONSE TO RADICAL SKEPTICISM

Britain with Wittgenstein. In his article, he discusses the way in which G. E. Moore dealt with radical versions of skepticism, and tries to extract a lesson from Moore about the role of ordinary language in answering the skeptic—a lesson that Moore himself never drew. We begin by reviewing Moore’s response to the skeptic in “A Proof of an External World.”2 There, Moore begins by noting that many philosophers, including Kant, have regarded it as “a scandal to philosophy” that no philosopher has succeeded in refuting the skeptic by proving the existence of objects external to the mind. Moore concludes from this that many philosophers must think that it is the job of philosophy to refute the skeptic by proving that there is an external world. Presumably, to do this one would have to provide a proof of the existence of objects external to the mind that even the skeptic would have to accept. With this in mind, Moore tries to state precisely what it is that philosophers want proven. This leads him to an analysis of what they mean by an object external to our minds. The key point, Moore believes, is that for an object to be external to our minds, it must be the sort of thing that could exist, even if no one were having experiences of any sort, and hence even if no one were experiencing or perceiving it. Having clarified that this is what many philosophers want to prove the existence of, Moore goes on to offer his proof. The two premises he offers are expressed as follows: “This is one hand,” and “This is another hand.” He then concludes that there are at least two hands. Since hands are the sorts of things which could exist even if no one were experiencing anything, he concludes that there are at least two things external to the mind. In giving this proof, Moore realizes that those philosophers who originally demanded a proof will deny that he has succeeded in giving one. He believes they are wrong. Insisting that his proof really is a proof, he sets out three criteria for being a genuine proof, and claims that his proof satisfies them. First, in any genuine proof, the premises must conceptually entail the conclusion. But Moore’s premises—expressed by “this is one hand” and “this is another”—do entail that there are at least two hands. Moreover, we understand that hands are the sorts of things which could exist even if no one was having any experiences. Since this is what it means for something to be external to the mind, the claim that there are two hands entails that there are at least two objects external to the mind. So Moore’s proof satisfies the first criterion for being a proof. 2 G. E. Moore, “Proof of an External World,” Proceedings of the British Academy, vol. 25, 1939; reprinted in his Philosophical Papers. See chapter 2 of volume 1 for further discussion.

MALCOLM’S PARADIGM CASE ARGUMENT

159

Second, in any genuine proof, the premises are different from the conclusion. His proof satisfies this criterion as well. Since the conclusion could be true even if there were no hands, and hence even if the premises were not true, the conclusion cannot be identical with the premises. Moore’s third criterion is that the premises of a genuine proof must be known to be true. Well, if he were standing in front of you giving his proof, he would claim that he really did know that the things he was holding up were hands. This is the crucial point. Many of those who think that we need a proof of the external world in the first place would be inclined reject the claim that Moore knows his premises to be true. How would Moore respond? First, he would point out that what he claims to know is a sort of paradigmatic example of what we all confidently take ourselves to know. Consequently, he would maintain, we have a strong prima facie reason to accept it. Though he doesn’t explain what the grounds of this justification are, he would insist that if one rejects fundamental commonsense convictions of this sort, one must have a very strong justification for doing so. What, then, he would ask, is the skeptic’s justification for rejecting the claim that he knows that he has hands? At this point, Moore has turned the tables on the skeptic. The skeptic starts out asking us to justify our claims. But in the face of Moore’s argument it becomes clear that the skeptic has something to justify as well. Presumably, the skeptic’s reason for rejecting Moore’s claim that he knows the premises of his proof to be true is that the skeptic has some conception of what knowledge is that is incompatible with Moore’s claim to know that he has hands. For example, the skeptic may think that one can know an empirical proposition p about the world only if one’s evidence for p conceptually guarantees its truth—e.g., only if it would be conceptually impossible for one to have the same qualitative experiences one actually has if p were not true. Since it seems to be conceptually possible that someone could have experiences qualitatively indistinguishable from Moore’s, even if he were a brain in a vat and had no hands, Moore’s experience doesn’t provide him with evidence that passes the skeptic’s test for knowledge. Hence the skeptic would deny that Moore knows that he has hands. Moore’s response to this is that the skeptic’s principle about what knowledge consists in amounts to a speculative philosophical theory that is not obviously true, and is itself very far from being beyond doubt. Consequently, it stands in need of justification. How do we justify broad philosophical theories of this sort—theories about what knowledge is, about what it is for an act to be right, about the conditions that have to

160

THE RESPONSE TO RADICAL SKEPTICISM

be satisfied if a sentence is to be meaningful, and so on? Well, Moore would say, one thing we do to test any such theory is to see what consequences it has for particular examples that we already feel confident really are instances of knowledge, right action, or meaningful sentences. Any general theory that conflicts with too many of our most confident intuitive judgments about particular cases is one that we are justified in regarding as implausible for that very reason. This general method for testing philosophical theories is a common one that has been used by analytic philosophers (following Moore’s lead) again and again. For example, when the logical positivists formulated their different versions of the verifiability criterion of meaning, they constantly found that the general principles they formulated either characterized many sentences that seemed obviously to be meaningful as meaningless, or many sentences that seemed meaningless as meaningful. In light of this, they were willing, in the end, to reject their initially plausible-sounding principles, thereby illustrating the Moorean point that general philosophical principles can often be tested by the degree to which they conform with our pre-theoretic judgments about particular cases. Moore’s own use of the method focused on the skeptic’s theory of what counts as knowledge. When we apply the method to this case, the very consequence of which the skeptic is so proud— namely, that nearly all examples of what we would ordinarily take ourselves to know about the world do not qualify as knowledge for the skeptic—has to be understood as strong and persuasive evidence against the skeptic’s theory of what knowledge is. That is the core of Moore’s response to skepticism about the external world, as well as to other forms of radical philosophical skepticism; and a very effective response it is. However, the response does leave some important questions open. For one thing, the response doesn’t tell us why we are prima facie justified in accepting paradigmatic examples of knowledge drawn from everyday life. For another thing, Moore’s response does not conclusively demonstrate that no form of skepticism could ever prevail. All it says is that the skeptic’s position typically relies on some general philosophical thesis that must somehow be given a justification sufficiently strong as to outweigh its otherwise counterintuitive consequences. Nothing in the Moorean methodology demonstrates that no such justification could ever be found. In the particular case of skepticism about the external world, he expresses confidence that no sufficiently strong justification for any skeptical conception of knowledge could ever be forthcoming—and that seems right. But what

MALCOLM’S PARADIGM CASE ARGUMENT

161

is the source of this confidence? What makes Moore, and many of us who are inclined to follow him, so sure that no conception of knowledge that leads to skepticism about the external world could ever be justified? Is there some error inherent in skepticism that guarantees this to be so?

Malcolm’s Attempt to Extend Moore’s Challenge This, in essence, is the question that Norman Malcolm sets out to answer in his article on Moore. His answer is that skepticism is wrong because it violates ordinary language. Malcolm spells this out in an argument that can be reconstructed as follows: P1. If the skeptic about knowledge of the external world were right, then not only would no one know any proposition about the external world, no one could know any such proposition no matter how much evidence one had for it. Therefore, according to the skeptic, if S expresses a proposition about the external world, and N is a name of some person, then the sentence N knows that S could not be used to correctly describe any possible situation. (Malcolm calls sentences that could not be used to correctly describe any possible situation, “self-contradictory.”) P2. Nevertheless, these sentences—N knows that S—are meaningful sentences of ordinary language that have ordinary uses. P3. No sentence with an ordinary use is self-contradictory. That is, any sentence with an ordinary use is one that could be used by speakers to correctly describe some possible situation. C. Therefore, skepticism about the external world is incorrect. The key step in this argument is P3. Here is a passage in which Malcolm talks about it. The assumption underlying all of these theories is that an ordinary expression can be self-contradictory. This assumption seems to me to be false. By an “ordinary expression” I mean an expression which has an ordinary use, i.e., which is ordinarily used to describe a certain sort of situation. By this I do not mean that the expression need

162

THE RESPONSE TO RADICAL SKEPTICISM

be one which is frequently used. It need only be an expression which would be used to describe situations of a certain sort, if situations of that sort were to exist, or were believed to exist.3 Note, Malcolm begins by describing an expression with an ordinary use as one which is used to describe a certain sort of situation. But he immediately backs off, and indicates that in order for an expression to have an ordinary use, it is not in fact necessary that it ever be used. He then tells us that an expression with an ordinary use is one “which would be used to describe situations of a certain sort, if situations of that sort were to exist, or were believed to exist.” But now we have a problem. P3 characterizes a sentence with an ordinary use as one that could be used to correctly describe some possible situation.4 But this isn’t exactly what Malcolm says. In the passage, he characterizes an expression with an ordinary use as one that would be used to describe certain situations if they existed, or if they were believed to exist. In light of this it might seem that what Malcolm has in mind for his third premise is P3* rather than P3. P3*. A sentence with an ordinary use is one that would be used to describe a certain sort of situation if it were to exist, or if it were believed to exist. However, if we substitute P3* for P3, then the argument is no longer logically valid. The problem is that nothing in the new argument rules out the possibility that some sentences might be used by speakers to describe a certain sort of situation that they believed to exist, but which in fact couldn’t possibly exist. As a result, with this characterization of what it is for an expression to have an ordinary use, it is conceivable that a sentence could have an ordinary use while still being what Malcolm calls “self-contradictory.” Consequently, this version of the argument fails.5 Malcolm, “Moore and Ordinary Language,” p. 358. Here, and throughout, when discussing whether a sentence S is such that, if a certain counterfactual situation were to exist, then S would be true if S were used in that situation to describe it, we will always take it for granted that in that situation S would mean precisely what it actually means—i.e., we don’t change the meaning of S when we consider its use in a counterfactual situation. 5 Malcolm’s use of the term ‘self-contradictory’ suggests that he is guilty of the all too familiar conflation of logically self-contradictory with apriori false, necessarily false, and impossible. Since speakers can generally tell when a statement is self-contradictory, this conflation may have tempted him to conclude that they can generally tell when a statement describes a situation that couldn’t possibly exist. One who succumbs to this temptation will tend to disregard the 3 4

MALCOLM’S PARADIGM CASE ARGUMENT

163

To get a version of the argument in which the premises entail the conclusion, we need to stick with the original version of P3. But now we need to find some convincing reason to think that P3 is true. We need some reason to believe that the fact that a sentence would be used in certain situations somehow guarantees that it would be used correctly to describe those situations. Malcolm does seem to think that there must be some guarantee of this sort. The reason that no ordinary expression is self-contradictory, is that a self-contradictory expression is an expression which would never be used to describe any sort of situation. It does not have a descriptive usage. An ordinary expression is an expression which would be used to describe a certain sort of situation; and since it would be used to describe a certain sort of situation, it does describe that sort of situation. A self-contradictory expression, on the contrary, describes nothing.6 Here Malcolm makes two extraordinary claims: (i) no self-contradictory sentence—that is, no sentence which could never be true, and so which could not be used to correctly describe any possible situation— would be used to describe any situation; and (ii) a sentence with an ordinary use is one that would be used by speakers to describe certain possible situations, and, because of this, it would correctly describe those situations. When taken as unrestricted generalizations, these claims are clearly false. As we will see in Part 7, many necessary falsehoods—such as the propositions expressed by Hesperus isn’t Phosphorus and Water isn’t H2O—are conceptually coherent and perfectly capable of having ordinary uses when employed by speakers who fully understand the sentences that express them, while wrongly believing them to be true. The reason Malcolm doesn’t see this is that he wrongly conflates self-contradiction with logical inconsistency, analytic falsehood, apriori falsehood, necessary falsehood, and genuine metaphysical impossibility. If his argument is to be salvaged, it must be reconstructed so as to avoid these confused identifications. What he needs is some guarantee that for certain kinds of sentence S, if S is meaningful, then S expresses a proposition that could be true, and so is not necessarily false. possibility that speakers may wrongly think that something is genuinely possible when it is not. For such a philosopher, the significant difference between P3 and P3* is easy to miss. 6 Malcolm, “Moore and Ordinary Language,” p. 359.

164

THE RESPONSE TO RADICAL SKEPTICISM

The guarantee he is looking for comes from a view he holds about how certain words get their meanings. According to Malcolm, the meanings of some words are given by paradigmatic examples to which the words correctly apply. For these words, (i) learning their meanings simply is learning to apply them to the individuals to which they clearly and correctly apply, and (ii) if there were no individuals to which they correctly applied, then these words would have no meanings. Putative examples of words of this sort are red, to the left of, earlier, later, etc. Malcolm thinks that for any word w that fits this picture, the sentence This is w will have uses in which it is true; it will therefore not be what he calls “self-contradictory”—more properly, it will not be necessarily false. If he is right about this, then radical skepticism about the truth of There are w’s will be inconsistent with the claim that w is a meaningful expression of this kind. Which expressions does Malcolm take to fall into this category? He gives us a clue in the following passage. The difference is that, whereas you can teach a person the meaning of the word ‘ghost’ without showing him an instance of the true application of that word, you cannot teach a person the meaning of these other expressions without showing him instances of the true application of those expressions. People could not have learned the meaning of the expressions ‘to the left of’, or ‘above’, unless they had actually been shown instances of one thing being to the left of another, and one thing being above another. . . . In the case of all expressions the meanings of which must be shown and cannot be explained, as can the meaning of ‘ghost’, it follows, from the fact that they are ordinary expressions in the language, that there have been many situations of the kind which they describe; otherwise so many people could not have learned the correct use of those expressions.7 In light of this passage, it is clear that Malcolm’s argument is intended to apply to expressions that, in his words, have meanings which “must be shown and cannot be explained.” How should our original argument be reformulated so as to reflect this? Perhaps the idea is that the word know has a meaning that must be shown rather than explained. Perhaps it has a meaning which can be learned only by recognizing 7

Ibid., pp. 360–61.

MALCOLM’S PARADIGM CASE ARGUMENT

165

some cases in which it correctly applies to an agent and a proposition, and perhaps if there where no such pairs of agent and proposition, then it would have no meaning at all. In fact, we will have to say something more specific than this—namely, (i) that any compound predicate knows that S in which S expresses an empirical proposition about the external world has a meaning which can’t be learned unless one knows that some compound predicates of this sort correctly apply to an agent and an empirical proposition about the external world, and (ii) that if no such predicates of this type correctly applied to pairs of agents and empirical propositions, then no such predicate would be meaningful. Using this idea, we can construct a new argument as follows: P1. If a compound predicate knows that S in which S expresses an empirical proposition about the external world has a meaning in ordinary language, then that meaning couldn’t be learned without learning that some compound predicates of this form correctly apply to an agent and an empirical proposition about the external world. P2. Some compound predicates of the above sorts have meanings in ordinary language that can be learned. C. Some sentences of the form N knows that S are true, in which N names an agent and S expresses an empirical proposition about the external world. Here we have a linguistically based refutation of skepticism about the external world. How successful is it? Although the argument is interesting, the linguistic claim expressed by P1 is one that Malcolm has done nothing to establish. Notice that it is one thing to say that in order to learn the meaning of the word know one must recognize some genuine instances of knowledge, and quite another to say that one must recognize genuine instances of knowledge of propositions about the external world. The skeptic might respond to the argument by rejecting P1, and maintaining that he recognizes instances of apriori knowledge, and of knowledge of propositions about himself and his own experiences, but no instances of knowledge of the external world. He might also claim that in order to understand the word know, one must recognize certain conceptual connections between knowledge and other things. For example, he might maintain that it is part of the meaning of the word know that (i) and (ii) are true: (i) if anyone

166

THE RESPONSE TO RADICAL SKEPTICISM

knows that S, then it must be true that S; (ii) if a person knows that S, then that person must have good grounds for believing that S (i.e., grounds that guarantee that it is true that S, or that make it overwhelmingly likely that S). The idea that there are conceptual truths about knowledge of this general sort is very persuasive. The skeptic takes advantage of this by using these apparent truths to argue that, contrary to what we ordinarily think, the conceptually necessary conditions for knowledge are not satisfied by alleged knowledge of the external world. One can agree with Moore that the skeptic will never succeed in justifying this claim; but Malcolm is trying to provide a more powerful response to the skeptic than Moore did. In effect, he is trying to show that the skeptic’s very willingness to use the word know, and to recognize it as meaningful, can be used to produce a refutation of skepticism that even the skeptic would have to accept. It seems evident that he has not done enough to establish this.

Limited Anti-Skeptical Lessons of the Paradigm Case Argument Diagnosing Semantic Incoherence in Skepticism vs. Refuting the Skeptic

Despite the failure of the paradigm case argument to establish Malcolm’s desired conclusion, there are interesting aspects of the general idea behind it that deserve to be explored further. His strategy is to extend and deepen Moore’s challenge to the skeptic. Moore challenged the skeptic to justify his conception of knowledge. Malcolm challenges the skeptic to explain how, if the skeptical conclusion is correct, the skeptic’s own words could have the meanings he presupposes them to have. Forget about the fact that Malcolm didn’t do enough to prove that the skeptic could never meet this demand. Just concentrate on the demand itself. The skeptic needs to be able to say something about how his words and thoughts acquire the contents he takes them to have. This is an issue that skeptics often have ignored. Typically, the skeptic takes it for granted, without question, that his words and thoughts have certain contents, and he never bothers to explain or question how this could be so. One sort of answer to the question of how a speaker’s words and thoughts acquire their contents lies in the relations those words and

MALCOLM’S PARADIGM CASE ARGUMENT

167

thoughts bear to objects in the speaker’s environment, and to the practices of the larger linguistic community of which the speaker is a part. Of course, it is hard to see how the radical skeptic about the external world could appeal to any such account. But if such an appeal is disallowed, then the skeptic needs to come up with some other account, and it is not easy to do so. How might such an account go? Here is one possibility: Let w be a word the meaning of which can only be shown, not explained, as Malcolm puts it. Perhaps the skeptic would grant that w is not learned by descriptive explanation of the properties with which it is associated. Still, the skeptic might say, there have been experiential differences (purely internal to the agent) between the cases in which the agent has applied, or attempted to apply, w, and those in which the agent has refused to apply w. In light of this, the skeptic might claim that these experiential differences, rather than any objects to which it has correctly been applied, are what provide w with its meaning; according to the skeptic, there is no reason to think that w has ever been, or ever could be, applied correctly to any object. Nothing in Malcolm’s discussion refutes this purely internalist account of the meanings of words.8 Of course, given that we are not skeptics, we are not likely to find any such skeptical semantic story plausible. From our ordinary, nonskeptical point of view, it is very plausible to suppose that many expressions, and many mental states, acquire their contents in substantial part from standing in certain relations to things in the world. If this is right, then when the skeptic questions the existence of such a world, or of our knowledge of it, he is, in fact, questioning the meaningfulness of the very terms he is using, or at least, of his, and our, knowledge of that meaning. Because of this, we nonskeptics may naturally conclude that the skeptic’s position is paradoxical or incoherent. And we may be right to draw this conclusion; his position may be paradoxical or incoherent, in just the way we say it is. The skeptic may, in actual fact, use words the meanings of which depend on the truth of principles that he either denies, or denies that we can ever know to be For types of skepticism less radical than global skepticism about the external world, the skeptic’s semantic story may be less far-fetched. Here is a possibility suggested by John Hawthorne about a skeptical semantic story that might accompany the skeptical view that nothing is, strictly speaking, flat: (a) the semantic content of ‘flat’ is an idealized condition that ordinary things approach but never satisfy, and (b) though an utterance of x is flat, said of an object o, is always strictly false, it may be conversationally acceptable when o comes what is, for purposes of the conversation, close enough to satisfying the condition for flatness. Though nonskeptics about flatness would certainly contest this account, it illustrates how proponents of some limited versions of skepticism might respond to a Malcolm-like semantic challenge.

8

168

THE RESPONSE TO RADICAL SKEPTICISM

true. However, this doesn’t mean that we have provided the skeptic with rational grounds which even he must accept for abandoning his skepticism. As we have seen, the skeptic may have his own skeptical semantic story, and there may be nothing we can say that will convince him that he is wrong. The lesson here is that diagnosing (semantic) incoherence in some form of radical skepticism is one thing; refuting the skeptic on his own grounds is another. As I see it, the interest in Malcolm’s paradigm case argument lies more in what it contributes to the former task than in what it contributes to the latter. The Limited Range of Cases in which a Diagnosis of Semantic Incoherence May Be Correct

Even when this is understood, Malcolm’s own discussion is not wholly successful. As we have seen, he did not succeed in showing that radical skepticism about the external world is semantically incoherent, even from a decidedly nonskeptical semantic perspective. However, his paradigm case argument does suggest a general line of attack that may be worth pursuing in certain cases. In this respect, I think he touched on something potentially important, even though he was not able to develop it sufficiently. This may be seen by considering forms of skepticism that are much more limited than the radical skepticism about the external world that we have been concerned with until now. Suppose, to take one highly restricted example, someone accepted Malcolm’s contention that (i) the word red has a meaning that can’t be learned without recognizing some examples to which the word correctly applies, and (ii) if there were no such examples to which it correctly applied, then it would be meaningless.9 Could such a person consistently use the word red to deny that there are any red objects? Could such a person coherently say There are no red objects, while at the same time presupposing that his words are meaningful? Yes, he could, but only if he granted that when he learned the meaning of the word red, there were objects to which it correctly applied, or at least that there were once red objects. Thus, even if he were to assertively utter, There are 9 It is worth noting that (ii) is not an automatic consequence of (i). See chapter 10 of my Beyond Rigidity for a discussion of the potential difference between the two as it applies to natural kind terms like water, gold, tiger, etc. Note also that in the case of color words like red, in order to establish (i) a philosopher would have to deal with the objection that one can learn their meanings by being shown white objects that are illuminated in nonstandard ways so as to appear to have one color or another. Thanks to Gil Harman for bringing this to my attention.

MALCOLM’S PARADIGM CASE ARGUMENT

169

no red objects, he would have to be willing to accept There have been red objects in the past. This illustrates the point that if it is correct that the meaningfulness of the word red depends on its correctly applying to some objects, then any form of skepticism which denies that there are, or ever have been, any red objects is semantically incoherent; if such a denial were correct, then the words used to express it would be meaningless. One might say something even stronger. Suppose a person both accepted a Malcolm-style linguistic account of red, and was willing to grant that he knew that his use of the word was meaningful. It is hard to see how such a person could then coherently deny that he knew that at one time his word correctly applied to certain things. Rather, such a person should be willing to say, Since I know that my word ‘red’ is meaningful, and since I know that this could not be so unless there had been red things to which it correctly applied, I know that there have been red objects. Suppose this is correct. We may then extend the diagnosis of semantic incoherence to versions of skepticism which deny that anyone now knows, or has ever known, that there have been red objects; if such denials were correct, then no one could know the words used to express them were meaningful. These conclusions are, of course, both conditional and programmatic. It would be an interesting project to try to formulate Malcolmstyle linguistic principles more precisely, to assess their plausibility, and to try to determine the scope of their anti-skeptical consequences, if any. There does seem to be something right about the idea that certain words get their meanings neither by being defined in terms of other words nor by satisfying some set of descriptive conditions that speakers associate with them, but simply by being applied to particular sorts of examples. If this is so, there may turn out to be some (limited) forms of skepticism for which this observation constitutes both a significant challenge and a revealing diagnosis of error. One example of this general sort that was actually discussed during the ordinary language period we are studying involves a skeptical claim arising from the observation that what modern physics tells us about the ordinary objects around us is sometimes at variance with what we previously believed. For example, we have learned that familiar material objects, which appear solid, are made up of many tiny atoms at some distance from one another. This has led some to the modest skeptical conclusion that although we normally think that bricks, concrete blocks, and slabs of steel are solid, in fact they really are not solid, since

170

THE RESPONSE TO RADICAL SKEPTICISM

most of the space they occupy is empty, owing to gaps between the atoms that make them up. The proper Malcolm-style response to such skeptics is that the word solid is defined by our intention that it is to apply to just such objects as bricks, steel, etc., and not by any abstract principle about how much empty space might exist at microscopic and submicroscopic levels. The putative linguistic fact here is that it is constitutive of the meaning of the word solid that it apply to such objects; it is not constitutive of its meaning that the things to which it applies must satisfy abstract descriptive principles that could be falsified by modern physics. Examples like this make it plausible to suppose that linguistic observations of the sort that Malcolm was trying to articulate may sometimes play a significant role in identifying the errors in certain skeptical views, even though they don’t constitute the sort of trump card against skepticism in general for which he seemed to be hoping.

CHAPTER 8 AUSTIN’S SENSE AND SENSIBILIA chapter outline 1. Austin’s Goal and His Strategies for Achieving It Ayer’s contention that appearance statements constitute all the evidence we have for believing in the external world Austin’s strategies for denying this, and his goal of showing skepticism to be incoherent by undermining this contention 2. The Argument That Knowledge of the World Does Not Always Require Evidence Cases in which we know a material object statement p to be true, even though it would be an abuse of language to say that we have evidence for p; why Austin is wrong to conclude that in these cases we don’t have evidence for p 3. Appearance Statements: Austin’s Criticism of the Sense Data Theory Main tenets of the sense data theory Austin’s criticism of Ayer’s use of the argument from illusion The strongest argument for sense data; the use of ordinary perceptual sentences to report the contents of radically nonveridical perceptual experiences Critique of the argument; how to understand these perceptual reports 4. The Argument That Appearance Statements Conceptually Depend on the Truth of Non-Appearance Statements The prospects for showing that appearance statements presuppose the truth of some material object statements The difference between refuting certain forms of skepticism and showing them to be semantically incoherent; why the prospects for doing the latter are better than those for doing the former An Austinian view of perceptual justification as a modest response to skepticism

Austin’s Goal and His Strategies for Achieving It In the last chapter we discussed attempts to show different kinds of skepticism to be incoherent by arguing that the crucial words used to state the skeptic’s position are ordinary words of a certain special type.

172

THE RESPONSE TO RADICAL SKEPTICISM

Having discussed this strategy, we turn now to a different attempt to undermine skepticism about the external world by linguistic means. The attempt is set out in John Austin’s book, Sense and Sensibilia,1 which was put together after his death in 1960 from notes he used for a series of lectures given, in somewhat different versions, several times between 1947 and 1959. The book is concerned with a variety of related questions in the philosophy of perception: Do we ever perceive mind-dependent objects that philosophers have called “sense data”? If we do ever perceive sense data, are sense data the only things that we ever directly perceive? Is perceiving a table, a chair, or an automobile a matter of directly perceiving sense data that are somehow related to a table, a chair, or an automobile? Do sense data statements, or statements about how things appear to us, constitute evidence for our ordinary beliefs about material objects—such as my belief that there are three chairs in my office? Is it conceivable that all, or nearly all, of our ordinary beliefs about material objects could be false, even though all our beliefs about how things appear to us are true? If this is conceivable, how do we answer the skeptic who maintains that without ruling out this possibility, we cannot know that our ordinary perceptual judgments about material objects are true? One of the peculiarities of Austin’s book is that he doesn’t always approach these questions directly. In fact, he is extremely critical of the terms in which the questions are expressed, and of the whole framework of ideas in which they arise. It is not too much to say that he thinks that the basic terms used in framing the debate are fundamentally misguided and confused. Thus, his method is to trace how the main ideas that he opposes have been developed in the work of other philosophers, and to try to expose the mistakes that lie at the foundations of their work. Although Austin selects several targets, the main object of his criticism is A. J. Ayer’s book, The Foundations of Empirical Knowledge, published in 1940.2 In that book Ayer outlines a view that may be summarized as containing the following major points. 1. There is a class of appearance propositions that state how things seem or appear to one at a given time on the basis of one’s sense experience at the time. These appearance propositions do not conceptually entail the truth of claims asserting the existence of J. L. Austin, Sense and Sensibilia, ed. G. J. Warnock (New York: Oxford University Press, 1964). 2 A. J. Ayer, The Foundations of Empirical Knowledge (New York : Macmillan, 1940). 1

AUSTIN’S SENSE AND SENSIBILIA

173

anything except the agent and the contents of his ideas, experiences, or mental states. In particular, they do not conceptually entail the truth of statements about material objects or other people distinct from the agent. Nevertheless, these propositions constitute all the evidence we have for statements about material objects and other people. 2. These appearance statements report the sense data that an agent is perceiving at a given time. Sense data have the following characteristics: (a) they are things that we directly see, hear, smell, etc.; (b) we perceive other things by virtue of perceiving them; (c) the sense data that one perceives are private to oneself; different people do not perceive the same sense data; (d) visual sense data have precisely the colors, shapes, and sizes they appear to have; (e) sense data are not physical objects of any sort; they are elements of one’s consciousness. 3. Material object statements are equivalent to infinite sets of categorical and counterfactual statements about sense data. When you assert a material object statement, what you are doing is making an infinite, open-ended set of predictions about sense experiences that would occur if certain conditions were fulfilled. We may express this by saying that material objects are logical constructions out of sense data. For our purposes, the third of these claims—about logical constructions—may be set aside. The doctrine that material objects are logical constructions out of sense data is something discussed at length, and criticized, in volume 1.3 It doesn’t play much of a role in Sense and Sensibilia; Austin does talk about it a little in chapters 6 and 10, but it is not his main concern, and it can be separated from points 1 and 2 without losing anything essential. Of Ayer’s three main points, the first is the most important. If there is a class of appearance statements that constitute all the evidence we have for statements about the external world, and if these appearance statements make claims about us and the contents of our consciousness, but do not conceptually entail statements about external objects, then skepticism about these objects becomes a real, coherent alternative. Austin’s ultimate goal is to undermine the coherence of skepticism. His aim is not just to show that skepticism is unjustified, or implausible, or that it is a 3

See chapter 7.

174

THE RESPONSE TO RADICAL SKEPTICISM

position that no one has reason to accept. Rather, his goal is to prevent skepticism from getting off the ground by denying skeptics their starting point. He believes that this can be done by refuting Ayer’s claims about the way in which appearance statements provide evidence for statements about material objects. He has two main strategies for doing this. One is to try to show that there are no appearance statements that make claims only about us and the contents of our conscious states without carrying any implications about other objects in the world. The other strategy is to try to show that statements like the one expressed by this is a table, as said by me right now looking at the table, are things that we know without having evidence for them at all. They are claims for which the philosophical demand for evidence makes no sense.

The Argument That Knowledge of the World Does Not Always Require Evidence We will start by looking at the second strategy. According to Austin, we speak of having evidence for some claim only when something special about the circumstances provides a reason to doubt that the claim is true. When there is no reason to doubt, when it is completely obvious that the claim is true, no one would raise a question about evidence. First, it is not the case, as this doctrine implies, that whenever a ‘material-object’ statement is made, the speaker must have or could produce evidence for it. This may sound plausible enough; but it involves a gross misuse of the notion of ‘evidence’. The situation in which I would properly be said to have evidence for the statement that some animal is a pig is that, for example, in which the beast itself is not actually on view, but I can see plenty of piglike marks on the ground outside its retreat. If I find a few buckets of pig-food, that’s a bit more evidence, and the noises and smell may provide better evidence still. But if the animal then emerges and stands there plainly in view, there is no longer any question of collecting evidence; its coming into view doesn’t provide me with more evidence that it’s a pig, I can now just see that it is, the question is settled. And of course I might, in different circumstances, have just seen this in the first place, and not had to bother with collecting evidence at all. Again, if I actually see one man shoot another, I may

AUSTIN’S SENSE AND SENSIBILIA

175

give evidence, as an eye-witness, to those less favorably placed; but I don’t have evidence for my own statement that the shooting took place, I actually saw it. Once again, then, we find that you have to take into account, not just the words used, but the situation in which they are used; one who says ‘It’s a pig’ will sometimes have evidence for saying so, sometimes not; one can’t say that the sentence ‘It’s a pig’, as such, is of a kind for which evidence is essentially required.”4 Part of what Austin is saying here may be summed up using the following linguistic doctrine. E1. If p is an empirical proposition about some external object— like the proposition that a certain animal is a pig—then it is proper to assert that one has evidence for p only if (a) one is not optimally situated to determine whether p is true and so one does not take oneself to know for a fact that p is true, and (b) the evidence one cites makes it more likely that p is true than it would otherwise be. This, or something close to it, seems to be an important part of what Austin is saying; but it isn’t all he is saying. In addition to this principle, he seems to accept a related but stronger principle. We may express that principle as follows: E2. If a person x is in a good position to judge whether p is true, and if, being in that position, x sees that p is true without having to perform any special investigation, or make any inferences to the truth of p from other propositions, then not only would it be improper for x to assert that x has evidence for p, it would be improper, and indeed false, for anyone to assert that x has evidence for p. In such a case x doesn’t have evidence for p, though, of course, x knows p. We needn’t worry about the details of these two principles, which might well need a certain amount of fine-tuning, if applied with complete generality. The point is that they fit what Austin has to say about the particular example he chooses. Notice the relationship between the two. The first principle says, roughly, that it is improper for one to assert that one has evidence for p in certain circumstances in which p is obvious. The second principle says that in these circumstances one 4

Sense and Sensibilia, pp. 115–16.

176

THE RESPONSE TO RADICAL SKEPTICISM

doesn’t have evidence for p. The idea connecting the two is that the reason it is improper for one to assert that one has evidence for p in the relevant circumstances is that such an assertion would be false. Although this idea is certainly understandable, it is also questionable. There are other reasons why an assertion might be improper, or out of place, that don’t involve its being false. Ayer makes this point in his reply to Austin, entitled “Has Austin Refuted the Sense Datum Theory?”, published in 1967. Now it is perfectly true that when it is a question of an object which we have no trouble in identifying, like the pig in Austin’s example, we do not normally go through the process of saying to ourselves: ‘It looks so and so, it feels so and so, it has such and such a smell, therefore probably, almost certainly, it is a so and so’. We just take it straight off to be a pig, or whatever. It is also true that in a situation of this kind it would sound odd to ask someone what evidence he had that he saw a pig. We should say this only if we had some reason to mistrust his identification. . . . The point here is that seeing signs of x is to some extent contrasted with seeing x. In any normal case in which I am seeing or handling a physical object, it is an abuse of language to say that I thereby obtain evidence of its existence, especially if the implication is that the evidence is not conclusive.5 Here, Ayer endorses the first of our principles—the linguistic one about when it is proper to assert that one has evidence for something. However, he goes on to reject the claim that in these cases one doesn’t have evidence, and so he finds reasons for the linguistic impropriety that do not imply the falsity of the claim that the agent has evidence. All this may be accepted, as a comment on ordinary usage. As a general rule, when one speaks of having evidence for a proposition p, one expects it to be understood that one is not entirely convinced of the truth of p, that one does not answer for its being more than probable. If I think that I know that p, I am underplaying my hand, and so misleading my audience, if I say no more than that I have good evidence for p. It would, however, be rash to lay any weight on this in the present context, since my knowing A. J. Ayer, “Has Austin Refuted the Sense-Datum Theory?” Synthese 17 (1967); reprinted in Metaphysics and Common Sense (San Francisco: Freeman, Cooper, & Co., 1970). The quoted passage occurs on pp. 129–30 of the reprinted version. Further citations will be to that version as well.

5

AUSTIN’S SENSE AND SENSIBILIA

177

that p is certainly not inconsistent with my having good evidence for it. On the contrary, in very many instances it would not be proper for me to claim to know that p unless I did have such evidence. It is just that I am taken to commit myself more strongly to the truth of p by straightforwardly asserting it than by asserting that there is good evidence for it, and our habit is not to make the weaker claim when we are in a position to make the stronger one. But it does not follow from this that when I know that p I have not got evidence for it, any more than, from the fact that when I think I know something, it is misleading for me to say only that I believe it, it follows that if I do know something I do not believe it. Consequently Austin’s example fails to prove his point.6 Consider Ayer’s contrast—asserting p vs. asserting that there is considerable evidence for p. One who asserts p commits oneself to more than does one who asserts only that there is considerable evidence for p. One who asserts p commits himself both to p, and also to having considerable evidence for p. This latter commitment arises from the fact that we take it to be a rule of good conversational citizenship that one not assert things for which one does not have good evidence. But one who says only that there is considerable evidence for p does not commit oneself to p. So asserting p involves a greater conversational commitment than merely asserting that one has considerable evidence for p. Suppose now that one is in a conversational situation in which the truth or falsity of p is at issue. Suppose further that in this situation one says only that there is considerable evidence for p. In this situation, one’s audience would be justified in concluding that the reason one said this, rather than simply asserting p, is that one wanted to avoid the stronger commitment that would come from asserting p. In most conversational situations, the most natural reason for wanting to avoid this commitment would be that one has some doubt about p, and so one doesn’t take oneself to know it for a fact. If this reasoning is correct, then when one asserts that one has evidence for p, one will normally be taken as implying or suggesting that one doesn’t know for a fact that p is true. Consequently, when one does know for a fact that p is true, one will not make the weaker assertion about evidence. The Ibid., p. 130. As we will see in the next chapter, the underlying principle on which this argument is based was articulated by Paul Grice in “The Causal Theory of Perception,” 1961, and “Logic and Conversation,” given in 1967 as the William James lectures at Harvard, both reprinted in Studies in the Way of Words (Cambridge, MA and London: Harvard University Press, 1989).

6

178

THE RESPONSE TO RADICAL SKEPTICISM

reason that one won’t make this weaker assertion is not that it would be false to do so; it would not be false. Rather, to make the weaker assertion by itself would be misleading, whereas to make the weaker assertion in conjunction with the stronger assertion of p would be redundant. If this is right, then principle E1, extracted from Austin’s discussion, is true, even though neither E2 nor the larger philosophical conclusion that Austin tries to derive from it is justified. Austin wants to derive the conclusion that many run-of-the-mill material object statements that we make are statements we know to be true even though we don’t have evidence for them. The reason he wants to draw this conclusion is that it would deprive the skeptic of the ability to get his view off the ground. The skeptic thinks it is obvious that we know material object statements to be true only if we have evidence for them; he then wants to add that no evidence we are able to gather will ever be enough. One doesn’t have to agree with the skeptic about this in order to conclude that Austin’s argument fails to deprive the skeptic of his starting point; Austin’s observations about when we ask for evidence and when we don’t do not establish that we sometimes know material object statements without evidence.

Appearance Statements: Austin’s Criticism of the Sense Data Theory Let us now turn to the other strategy that Austin employs to try to deprive the skeptic of his starting point. This strategy concerns the nature and existence of appearance statements—propositions about how things appear to particular observers at particular times. Austin would like to be able to show that there is no class of propositions of this sort, which talk only about individuals and the contents of their minds, but which nevertheless constitute their evidence for claims about material objects. Much of his discussion of this issue takes the form of criticisms of arguments for sense data. The standard treatment of visual sense data can be summarized roughly as follows: (i) The sense data theorist accepts the principle that whenever it seems to us as if we are seeing something of a certain color, shape, or size, we are directly seeing something that is of that color, shape,

AUSTIN’S SENSE AND SENSIBILIA

179

or size. (Think of staring at a bright light, turning away to face a blank wall, and “seeing a gold circle” float before one’s eyes. The claim is that even in a case like this, one is seeing something gold in color.) (ii) The theorist then reasons that since ordinary material objects like sticks, trees, and furniture sometimes appear to be of a certain color, shape, or size while really not being of that color, shape, or size, our seeing a material object that looks so and so but really is not must involve directly seeing something other than that material object. This other thing that we directly see is a visual sense datum, distinct from the material object we perceive. (iii) The theorist draws a similar conclusion about visual hallucinations. When we hallucinate pink rats, we directly see something pink, even though there are no rats that we see. In these cases, as well, we directly see sense data. (iv) Since in all such cases the sense data one directly sees can’t plausibly be identified with any material objects, the sense data theorist concludes that they are not material objects, but rather something private to one’s mind. In this way, the theorist arrives at the conclusion that we directly see sense data in all cases of “non-veridical visual perception,” i.e., in cases in which it visually appears to us that things in our environment are a certain way, when in fact they are not that way. (v) Next, the theorist notices that cases in which things in our environment are the ways they are perceived to be may be indistinguishable from cases in which things are not as they appear. The theorist takes it to be a fact that for every case of an accurate visual perception p, in which an external object which is F is seen by an agent as being F, there is a possible perceptual experience p*, qualitatively identical with p, even though in experiencing p* the agent isn’t seeing anything in the world that is F, because the agent’s brain is being artificially stimulated to create this impression. (vi) Since the theorist has already concluded that in the “nonveridical” case the agent’s visual experience p* is enough to guarantee that he is directly seeing a sense datum, the theorist now concludes that the agent’s visual experience p in the “veridical” case guarantees the same thing. By this means, the theorist arrives at the conclusion that the things we directly perceive are sense data, both in cases of so-called “illusion” and in

180

THE RESPONSE TO RADICAL SKEPTICISM

so-called “veridical cases.” The two cases are alike in that the agent directly perceives things of precisely the same type in both; the cases differ in that in the veridical case the agent’s directly perceiving the sense data counts as seeing (in a related sense) the objects in the environment that he is looking at, which are responsible for his perceptual experience. Austin rejects this picture, bringing objections against every aspect of the positive case for sense data. We can illustrate some of these objections by reviewing his criticism of the so-called “argument from illusion” in chapter 3. We begin with his summary of the cases. In Ayer’s statement it runs as follows. It is ‘based on the fact that material things may present different appearances to different observers, or to the same observer in different conditions, and that the character of these appearances is to some extent causally determined by the state of the conditions and the observer’. . . . First, refraction—the stick which normally ‘appears straight’ but ‘looks bent’ when seen in water. He makes the ‘assumptions’ (a) that the stick does not really change its shape when it is placed in water, and (b) that it cannot be both crooked and straight. He then concludes (‘it follows’) that ‘at least one of the visual appearances of the stick is delusive’. Nevertheless, even when ‘what we see is not the real quality of a material thing, it is supposed that we are still seeing something’—and this something is to be called a ‘sensedatum’. A sense-datum is to be ‘the object of which we are directly aware, in perception, if it is not part of any material thing’. . . . Next, mirages. A man who sees a mirage, he says, is ‘not perceiving any material thing; for the oasis which he thinks he is perceiving does not exist’. But ‘his experience is not an experience of nothing’; thus ‘it is said that he is experiencing sense-data, which are similar in character to what he would be experiencing if he were seeing a real oasis, but are delusive in the sense that the material thing which they appear to present is not really there’. Lastly, reflections. When I look at myself in a mirror ‘my body appears to be some distance behind the glass’; but it cannot actually be in two places at once; thus, my perceptions in this case ‘cannot all be veridical’. But I do see something; and if ‘there really is no such thing as my body in the place where it appears to be, what is it that I am seeing?’ Answer—a sense-datum.7 7

Sense and Sensibilia, pp. 20–22.

AUSTIN’S SENSE AND SENSIBILIA

181

Austin makes three main points about these examples. (i) They are not all cases of illusions. For example, no one is fooled by the stick in the water because no one is foolish enough to think that a stick must always look the same whatever circumstances it is in. As for Ayer’s other examples, the mirage might be an illusion, but the image in the mirror is not. (ii) The image in the mirror is not some private sense datum. Rather it is something that can be photographed and seen by other people. (iii) In discussing the examples, Ayer runs together illusions and delusions. For example, though the cases are all supposed to be examples of illusions, Ayer characterizes the visual experiences in these cases as delusive. What is the difference between illusions and delusions, and how is this difference supposed to matter in “the argument from illusion”? What, then, would be some genuine examples of illusion? (The fact is that hardly any of the cases cited by Ayer is, at any rate without stretching things, a case of illusion at all.) Well, first, there are some quite clear cases of optical illusion—for instance the case we mentioned earlier in which, of two lines of equal length, one is made to look longer than the other. Then again there are illusions produced by professional ‘illusionists’, conjurors—for instance the Headless Woman on stage, who is made to look headless, or the ventriloquist’s dummy, which is made to appear to be talking. Rather different—not (usually) produced on purpose—is the case where wheels rotating rapidly enough in one direction may look as if they were rotating quite slowly in the opposite direction. Delusions, on the other hand, are something altogether different from this. Typical cases would be delusions of persecution, delusions of grandeur. These are primarily a matter of grossly disordered beliefs (and so, probably, behavior) and may well have nothing in particular to do with perception. But I think we might also say that the patient who sees pink rats has (suffers from) delusions—particularly, no doubt, if, as would probably be the case, he is not clearly aware that his pink rats aren’t real rats. The most important differences here are that the term ‘an illusion’ (in a perceptual context) does not suggest that something totally unreal is conjured up—on the contrary, there just is the arrangement of lines and arrows on the page, the woman on stage with her head in a black bag, the rotating wheels; whereas the term ‘delusion’ does suggest something totally unreal, not really there at all.8 8

Ibid., pp. 22–23.

182

THE RESPONSE TO RADICAL SKEPTICISM

In this passage, Austin nicely brings out some differences between illusion and delusion in a subtle and illuminating way. Still, one wonders, what difference does the distinction make to the argument? He tries to tell us in the following passage. The way in which the ‘argument from illusion’ positively trades on not distinguishing illusions from delusions is, I think, this. So long as it is being suggested that the cases paraded for our attention are cases of illusion, there is the implication (from the ordinary use of the word) that there really is something there that we perceive. But then, when these cases begin to be quietly called delusive, there comes in the very different suggestion of something being conjured up, something unreal or at any rate ‘immaterial’. These two implications taken together may then subtly insinuate that in the cases cited there really is something that we are perceiving, but that this is an immaterial something; and this insinuation, even if not conclusive by itself, is certainly well calculated to edge us a little closer towards just the position where the sense-datum theorist wants to have us.9 In my opinion, Austin’s criticisms of certain details of Ayer’s discussion of some particular cases are legitimate, and more or less accurate, as is his complaint that ignoring the difference between illusion and delusion may have the rhetorical effect of subtly stacking the deck in favor of the sense data theorist (though I see no reason to think that Ayer “calculated” any of this). But as Austin himself seems to recognize, these criticisms are hardly conclusive in themselves. Surely the central core of any attack on Ayer’s position regarding sense data ought to involve an attack on the principle that if something looks to one as if it is so and so, then one must be directly seeing something that is so and so. After all, if one grants this problematic principle, then it is hard to deny that we see sense data in cases of illusion, delusion, and hallucination; and if one grants that the nature of our perceptual experiences in those “non-veridical” cases is sufficient to bring visual sense data into being, then one will have a hard time denying the presence of the same sort of sense data in qualitatively identical instances of ordinary “veridical perception.” To be sure, a good deal of Austin’s critical discussion can be viewed as an attack on Ayer’s problematic principle, and when it is, a good 9

Ibid., p. 25.

AUSTIN’S SENSE AND SENSIBILIA

183

deal of what he says is persuasive. Nevertheless, before we accept Austin’s conclusions, we need to focus on the most important and difficult cases used to motivate the principle. First, consider hallucinations. There seems to be a well-entrenched habit of describing hallucinations using ordinary perceptual verbs. Drunks are said to see pink rats running across the floor. Macbeth says that he sees a dagger before him. In these cases, subjects seem to be having visual experiences. It seems to the drunkard as if he were seeing pink rats; nevertheless no material object is looking to him like a pink rat. So if he is seeing anything at all that looks like a pink rat, it is not unreasonable to think that it is a sense datum private to him. The fact that we tend to be willing to accept the description of the drunkard as seeing something thus provides some measure of support for the sense data theory. Second, consider the phenomenon of double vision. When I put my finger very close to my face, I see double. If you ask what I mean by this, I am inclined to say I see two images of my finger. But then you may ask, What are these images that you see? The two images are not identical with one another, or else they wouldn’t be two. But then it would seem that at least one of the images is not identical with my finger, since no one thing (my finger) can be identical with two different things. Moreover this image, which is not my finger, is not identical with any other material object. So a prima facie case can be made that it is a sense datum. And if this image is a sense datum, why not the other one as well? Third, consider afterimages. When I stare at a bright fluorescent light and then look at a blank wall, I have a kind of visual sensation that, in my unguarded non-philosophical moments, I am inclined to describe as seeing a faint gold-shaped cylinder float in front of my eyes. Again, if there is anything that I see in a situation like this, it would seem to be a private sense datum rather than a material object. Finally, there is the brain in a vat, which is being stimulated to produce experiences qualitatively identical with those one has in normal life. How would we describe the experiences of this brain? It seems natural to say that it has visual experiences of colors, shapes, and sizes even though it never sees any object in its environment whatsoever. But if it has these experiences of color, shape, and size, it is tempting to say that it sees (in a certain direct sense of ‘see’) colored images of various shapes and sizes. And if that is so, then it would seem that these images must be private sense data rather than material objects. In my view, these examples provide the strongest reason to believe that, sometimes at least, we do (directly) see sense data; and if having

184

THE RESPONSE TO RADICAL SKEPTICISM

such experiences in these “non-veridical cases” is enough to count as (directly) seeing sense data, it is hard to deny that having qualitatively identical experiences with the same perceptual content in “veridical cases” also counts as (directly) seeing sense data. So there is a case to be made for the sense-data theory. But, is it, in the end, convincing? Are the examples of hallucination, double vision, afterimages, and brains in vats compelling ? I think not. Take the case of hallucinations. Although we are inclined to say, The drunkard sees pink rats, we know that there are no pink rats that he sees. We know that it only seems to him as if he were seeing pink rats. That, or something like it, is what we are trying to convey when we utter the sentence He is so drunk that he is seeing pink rats. The sentence itself is literally false, though we use it in a loose and extended way to convey something true—namely that his visual experience is like that of one who is seeing pink rats. The case is analogous to one involving a mental patient suffering from the delusion that his body is covered with bugs. Seeing him repeatedly going through the motions of plucking things from different parts of his body, we explain to a visiting doctor, He is removing the bugs that attack him when he sleeps. Since there are no bugs that attack him when he sleeps, the sentence we utter is literally false, though we use it successfully to explain the patient’s action, as it seems to him. In a case like this, there is little temptation to suppose that there is a special sense of the verb removes according to which the sentence we utter is a literal truth that reports the removal of attacking bugs that are private to the patient. For the same reason, our ability to convey something true by using the sentence He is seeing pink rats when talking about the drunkard shouldn’t lead us to the conclusion that there is a special direct sense of see such that, in this sense, the drunkard really does see certain private ideas of his own.10 Similar remarks apply to the other cases. For example, when I remark that I am seeing two images of my finger, what I am really trying to convey is that my finger appears to me to be in two different locations simultaneously. Of course I know that it is not in two different places. Since I don’t want to give others the impression that my finger 10 The conclusion that in this sort of case the drunkard doesn’t see anything, and in particular doesn’t see anything that either is, or looks, pink, should not be confused with the claim that the drunkard’s visual experience does not represent the property of being pink as being instantiated in the immediate environment. On the contrary, although the drunkard’s visual experience does represent the environment as containing something pink, it may not be a visual experience in which anything that is seen is represented as being pink.

AUSTIN’S SENSE AND SENSIBILIA

185

might really be in two different places, or that I think it might be, I say that I see images of my finger. Linguistically, what is going on is that I start with facts about how my finger appears and then express these facts in terms of talk about appearances—as if appearances were extra objects that I see. For ordinary conversational purposes this is innocent enough, because it allows us to convey what we want to convey— something about how my finger appears. However, when we think philosophically about what is involved in treating appearances as special entities that we see, it seems reasonable to reject the idea that there really are such entities. So I would say that the claim that I see two images of my finger is literally false, even though I use it to convey something true. I am inclined to treat the other examples along the same lines. This involves treating certain things that speakers would ordinarily say in some not-too-common circumstances as literally false, although conversationally useful in conveying other things that are true. One could characterize the view I am urging as a revision of ordinary language, in the sense that in philosophical discussion of these cases, where we are concerned with the literal truth, we would not describe the perceptual situations in exactly the same way that speakers do in ordinary conversation. On the other hand, this limited philosophical revisionism need not involve any suggestion that ordinary conversational practice should be changed or given up. The practice of sometimes using sentences that are literally false to convey something true is quite widespread in language, no doubt because it fulfills important communicative purposes. So ordinary conversational talk about perception may safely remain as is. In this last interlude I have advanced my own view, rather than sticking closely to Austin’s text. Nevertheless, I think that the view we have arrived at is reasonably close to his. I will illustrate this with a passage from chapter 9, where he discusses Ayer’s claims about the meaning, or meanings, of the perceptual verb to see. Both Austin and Ayer agree that it has a sense in which to say that someone sees so and so conceptually entails that so and so exists, though it does not entail that so and so has the properties it appears to have. This is the ordinary sense of see, in which we talk about someone seeing my house, and in which the statement that I have a house follows from the statement that someone sees it. Ayer thinks that in addition to this sense, there is at least one other ordinary sense of the verb to see in which there is such a thing as so and so does not follow from x sees so and so. Austin disagrees

186

THE RESPONSE TO RADICAL SKEPTICISM

with this; instead, he seems to think that the verb has only the one meaning (when it is used to report a kind of perception). Here is the passage from Austin that I want to compare with the view of reports of hallucination, double vision, and the like, that I have sketched. The third example, of double vision, is less easily dealt with. Here Ayer says: ‘If I say that I am perceiving two pieces of paper, I need not be implying that there really are two pieces of paper there’. Now this, I think, is true only with some qualification. It is, I suppose, true that, if I know that I am suffering from double vision, I may say ‘I am perceiving two pieces of paper’ and, in saying this, not mean that there really are two pieces of paper there; but for all that, I think, my utterance does imply that there are, in the sense that anyone not apprised of the special circumstances of the case would naturally and properly, in view of my utterance, suppose that I thought there were two pieces of paper. . . . But, it may be said, have we not conceded enough to justify the main point that Ayer is making here? For whatever the case may be with ‘really are perceived’, we have agreed that I may properly say, ‘I am perceiving two pieces of paper’, in the full knowledge that there are not really two pieces before me. And since it is undeniable that these words may also be used as to imply that there really are two pieces of paper, do we not have to agree that there are two different senses of ‘perceive’? Well, no, we don’t. . . . Since, in this exceptional situation, though there is only one piece of paper I seem to see two, I may want to say, ‘I am perceiving two pieces of paper’ faute de mieux [for want of a better word], knowing quite well that the situation isn’t really that in which these words are perfectly appropriate. But the fact that an exceptional situation may thus induce me to use words primarily appropriate for a different, normal situation is nothing like enough to establish that there are, in general, two different, normal (‘correct and familiar’) senses of the words I use, or of any one of them.11 In this passage Austin implicitly distinguishes what the sentence I utter means from what I mean by uttering the sentence on a certain occasion. What my sentence means is something determined by the 11

Sense and Sensibilia, pp. 89–91.

AUSTIN’S SENSE AND SENSIBILIA

187

conventional, or literal, meanings of the words I use. What I mean may, in certain cases, be something different and more specific to the situation than that. Moreover, Austin suggests that when I use the sentence to describe a case of double vision, I know that the circumstances are unusual, and that the literal meaning of the sentence does not exactly fit them. In a case like this, I use the sentence because its meaning is close enough to convey what I have in mind. Just so. Here we see the subtlety and perceptiveness of Austin at his best.

The Argument That Appearance Statements Conceptually Depend on the Truth of Non-Appearance Statements Suppose, in light of all this, that we come to agree with Austin that there is no good reason to think that we ever perceive sense data, let alone that sense data are what we always (directly) perceive. This involves rejecting point 2 above, from Ayer’s The Foundations of Empirical Knowledge—namely, the claim that sentences about how things appear to an agent report sense data that the agent is perceiving. However, this still leaves Ayer’s point 1, which is really the crucial issue, in any case. 1. There is a class of appearance propositions that state how things seem or appear to one at a given time on the basis of one’s sense experience at the time. These appearance propositions do not conceptually entail the truth of claims asserting the existence of anything except the agent and the contents of his ideas, experiences, or mental states. In particular, they do not conceptually entail the truth of statements about material objects or other people distinct from the agent. Nevertheless, these propositions constitute all the evidence we have for statements about material objects and other people. All we have concluded so far is that appearance statements do not report sense data the agent is perceiving; but that doesn’t mean that there is no class of appearance statements that bear the evidentiary relation to material objects that is claimed by Ayer’s point 1. As long as point 1 remains, skepticism about the external world remains a challenging, coherently statable view, and Austin’s goal of producing a linguistic refutation of it, or of robbing it of its starting point, remains unrealized.

188

THE RESPONSE TO RADICAL SKEPTICISM

For example, consider the brain in the vat. Couldn’t we describe the perceptual experience of the brain using appearance statements like: At time t it seems to the brain as if it is seeing a blackboard with white marks on it; it also seems to the brain as if it is hearing a voice and seeing a person in a tan-colored shirt sitting in front of it, and so on? None of these statements commits us to the claim that the brain is really seeing or hearing anything, only that it is having experiences like ones it would have if it were seeing and hearing various things. Now consider all true appearance claims of this type about the brain. Two things seem prima facie possible. First, it seems possible that the set of true claims describing how things appear to the brain could exactly parallel the set of true claims describing how things appear to you. That is, whenever it seems to the brain as if it is seeing or hearing so and so, it seems to you as if you are seeing or hearing so and so, and vice versa. So things appear to the brain exactly as they appear to you. The second thing that seems possible is that despite this fact, the brain never really perceives any other person or material object. If all of that really is possible, then the skeptic can formulate his challenge. First, he will maintain that statements about how things appear to us provide our evidence for the statements about other people and material objects that we take ourselves to know. Second, he will cite the possibility of the brain in the vat as showing that the totality of this evidence is logically or conceptually compatible with the falsity of the statements we take ourselves to know on the basis of that evidence. Third, he will conclude from this that our evidence does not guarantee that the statements about other people and material objects that we take ourselves to know are really true. Finally, he will maintain that since our evidence does not guarantee that these statements are true, we do not really know that they are true. That is the skeptical challenge. Of course, the fact that we can formulate the challenge does not show that it is correct. For example, one might reject the skeptic’s implicit assumption that in order for one to know p, one’s evidence for p must logically or conceptually entail p. However, this is not Austin’s strategy. Instead, he seems to want to show that skepticism is incoherent, self-undermining, or impossible to get off the ground; and so far we don’t have anything remotely resembling this result, even if we agree with his rejection of sense data. With this in mind, I will briefly consider a way in which Austin tries to extend his attack on sense data to an attack on appearance statements that don’t mention sense data. The idea is to attack the view

AUSTIN’S SENSE AND SENSIBILIA

189

that appearance statements are epistemologically prior to material object statements. Both Ayer and the skeptic seem to assume that appearance statements provide evidence for material object statements without themselves presupposing the truth of any material object statements. Suppose, however, that it could be shown that appearance statements themselves somehow presuppose that some material objects statements are true. It might then turn out that the skeptic couldn’t get his challenge off the ground. The following passage from chapter 11, in which Austin criticizes G. J. Warnock’s book on Berkeley,12 seems to indicate that Austin was attracted to this line of reasoning. Furthermore, Warnock’s picture of the situation gets it upsidedown as well as distorted. His statements of ‘immediate perception’, so far from being that from which we advance to more ordinary statements, are actually arrived at, and are so arrived at in his own account, by retreating from more ordinary statements, by progressive hedging. (There’s a tiger—there seems to be a tiger—it seems to me that there’s a tiger—it seems to me now that there’s a tiger—it seems to me now as if there were a tiger.) It seems extraordinarily perverse to represent as that on which ordinary statements are based a form of words which, starting from and moreover incorporating an ordinary statement, qualifies and hedges it in various ways. You’ve got to get something on your plate before you can begin messing it around. It is not, as Warnock’s language suggests, that we can stop hedging if there is a good case for coming right out with it; the fact is that we don’t begin to hedge unless there is some special reason for doing so, something a bit strange and off-color about the particular situation.13 Although this passage is suggestive, the idea is not worked out in any detail. It is possible that what Austin had in mind here is a version of Ryle’s claim that deception makes sense only against a general background of discoverable non-deception; the possibility of error conceptually presupposes the reality of discoverable truth; the possibility of a counterfeit presupposes the existence of the genuine. If that is what Austin was getting at, then his claim is as problematic as we found Ryle’s to be. 12 G. J. Warnock, Berkeley (New York: Pelican Books, 1953; reprinted with a new preface by Peregrine Books, 1969). 13 Ibid., pp. 141–42.

190

THE RESPONSE TO RADICAL SKEPTICISM

Perhaps, however, his idea could be developed in a different way. Here is a thought. Recall Malcolm’s paradigm case argument. Perhaps an assertive utterance of a sentence It appears to me now that I am perceiving a so and so presupposes the meaningfulness of the expression so and so; and perhaps, at least in the case of certain expressions so and so, their being meaningful presupposes that at some time or other there have been so and so’s. If this were right, it wouldn’t rule out all skepticism. For example, it wouldn’t rule out skepticism about whether we can know that there are now so and so’s, but it would rule out some forms of skepticism. If the skeptic were cooperative, we might do even better. If he were to admit (i) that our appearance statements are meaningful, (ii) that in order for them to be meaningful, there must have been so and so’s, and (iii) that both (i) and (ii) are knowable, then he may have to admit that we can know that some so and so’s must have existed. One may judge for oneself the chances of finding such a cooperative skeptic—one who not only accepts a nonskeptical story about the meaning of the relevant expression so and so, but also accepts that this story is knowable. However, all is not lost, even if our skeptics are not so accommodating. If we nonskeptics can justify, by our normal nonskeptical standards, the relevant nonskeptical account of the meaning of so and so, then we may be able to convict recalcitrant skeptics of incoherence, despite their protestations. How good a result would this be? Here, I think we have to give a mixed evaluation. On the positive side, it would strengthen our response to the skeptic, and deepen our diagnosis of what is wrong with some radical forms of skepticism. Moore made the case that radical skepticism typically presupposes a philosophical conception of knowledge the justification of which is, to put it mildly, highly problematic. The success of the Malcolm-Austin line of attack would further convict at least some skeptics of advocating theses the truth of which would deprive their own words of the meanings they presuppose them to have. Such a result would help explain the lurking sense of unreasonableness and even incoherence that these forms of skepticism arouse in most of us nonskeptics. On the other hand, it would not, I think, be correct to describe the Malcolm-Austin attack, even if successful, as a refutation of skepticism. At least, this would not be correct if such a refutation were required to be a dialectically persuasive argument from premises that even the skeptic must accept to the conclusion that radical skepticism is incorrect. The problem is that even if the premises of such a Malcolm Austin–style argument against some forms of skepticism are true, and even if they

AUSTIN’S SENSE AND SENSIBILIA

191

are knowable by us, there is little reason to think that there is a nonquestion-begging and dialectically effective way of convincing the skeptic of this. Premises—even true ones—that involve claims about meanings are no more immune to skeptical doubt than premises about the existence of hands or the nature of knowledge. In this sense, even the limited success imagined for the Malcolm-Austin–style attack against certain forms of skepticism might not put those forms of skepticism to rest. Finally, there is another view about our knowledge of material object statements that Austin’s brief comments about the epistemological priority of ordinary material object statements over appearance statements might be taken to suggest. When those comments are combined with an idea that may have been behind his earlier expressed contention that in certain cases—such as seeing a pig directly in front of one—one knows material object statements to be true without having evidence for them, it is possible to construct a view of our justification for material object statements which is somewhat different from that expressed by Ayer’s point 1, and which may well be worth further investigation. The view is this: Our perceptual experiences have contents that represent the world to us as being various ways. For example, my visual experience right now represents a certain tree t, which is located outside my window, as being of a certain size, shape, and color. On the basis of this experience I form the perceptual belief that t has that size, shape, and color. Here, the content of my belief—that which I believe—is the same as the content of my visual experience. In normal, everyday situations, when one has no reason to doubt the accuracy of one’s perceptions, perceptual beliefs like this arise naturally and unreflectively from perceptual experiences that share their contents. The fact that these beliefs share the contents of the perceptions they arise from is what justifies them. Thus, the justification for these beliefs is provided by the perceptual experiences themselves, rather than by any more cautious set of beliefs about how things appear to us. In the present case, I didn’t infer that t has the properties P, Q, and R from more primitive beliefs to the effect that t (or something) appears to have P, Q, and R. Normally, I wouldn’t even entertain the appearance proposition, unless I had been given some positive reason for thinking that my perception shouldn’t be trusted. Thus, appearance propositions are not that from which we advance to ordinary material object statements— either to infer them in the first place, or to justify them after the fact. Rather, they are that to which we retreat in special cases when we have reason to think that something may be amiss.

192

THE RESPONSE TO RADICAL SKEPTICISM

Although much of this story goes beyond what is explicit in Austin himself, the overall view is one that provides a sympathetic treatment of some of his central concerns. How much adopting this view would alter his response to radical skepticism is, I think, a matter that is open to interpretation. Since there is nothing in the view of perception just sketched that would convict the skeptic of semantic incoherence, this aspect of the view might be disappointing to Austin. In addition, I see no way of using the view to construct a response to skepticism that even the skeptic would have to agree was dialectically effective. So, if Austin was hoping for such a response, he would not find it here. Nevertheless, the prospects for using the view to construct a response that gives a good account of what we nonskeptics rightly find wrong with skepticism might well be appealing to him. The promising thought arising from the view is that our most basic perceptual beliefs about material objects are prima facie justified by the very perceptions they arise from. Unless this justification is defeated by positive reasons for taking the beliefs to be false, these beliefs should qualify as knowledge (provided they are true). Of course, in certain cases—e.g., when one is in the House of Illusions—there are such reasons, and perceptual experiences that would normally give rise to knowledge do not. However, special circumstances aside, the radical skeptic is now properly viewed as facing an enormously daunting task. In order to establish his conclusion that none of our perceptual beliefs about material objects are known or justified, he must provide a positive reason for taking them all to be false. For this it is not enough to point out the mere possibility that appearances could remain as they are, even if no material objects were being perceived.14 14 For a good discussion of the use of the kind of account of perception sketched above to provide a response to skepticism, see Jim Pryor, “The Skeptic and the Dogmatist,” Noûs 34 (2000): 517–49.

SUGGESTED FURTHER READING FOR PART THREE

Main Primary Sources Discussed Austin, J. L. Sense and Sensibilia, ed. G. J. Warnock. New York: Oxford University Press, 1964. Ayer, A. J. The Foundations of Empirical Knowledge. New York: Macmillan, 1940. ———. “Has Austin Refuted the Sense-Datum Theory?” In Metaphysics and Common Sense (San Francisco: Freeman, Cooper, & Co., 1970); originally published in Synthese 17 (1967). Malcolm, Norman. “Moore and Ordinary Language.” In The Philosophy of G. E. Moore, vol. 1, edited by P. A. Schilpp (La Salle, IL: Open Court, 1942), 343–68.

Additional Primary Sources Grice, Paul. “The Causal Theory of Perception,” 1961, and “Logic and Conversation,” given in 1967 as the William James lectures at Harvard University, both reprinted in Studies in the Way of Words (Cambridge; MA and London: Harvard University Press, 1989). Moore, G. E. “Proof of an External World.” Proceedings of the British Academy, vol. 25, 1939; reprinted in his Philosophical Papers.

Additional Recommended Reading Pryor, Jim. “The Skeptic and the Dogmatist.” Noûs 34 (2000):517–49. Thau, Michael. “What Is Disjunctivism?” In Proceedings of the ThirtyFifth Oberlin Colloquium in Philosophy, ed. Todd Ganson and Michael Martin, to appear as a special issue of Philosophical Studies. Warnock, G. J. Berkeley. New York: Pelican Books, 1953; reprinted with a new preface by Peregrine Books, 1969.

This page intentionally left blank

PA R T F O U R PAUL GRICE AND THE END OF ORDINARY LANGUAGE PHILOSOPHY

This page intentionally left blank

CHAPTER 9 LANGUAGE USE AND THE LOGIC OF CONVERSATION chapter outline 1. Grice’s Critique of Ordinary Language Philosophy The paradigmatic method of the ordinary language philosophers; the need to distinguish meaning from other factors governing language use 2. Grice’s Central Idea: the Logic of Conversation Conversational maxims and conversational implicatures; the distinction between conversational and conventional implicature; particularized and generalized conversational implicatures; cancellation and detachment; meaning, assertion, and implicature 3. Puzzles and Unresolved Problems Grice’s paradox: since conversational implicature presupposes that speakers know what is meant and asserted, there should be little or no serious controversy about how to distinguish both from what is conversationally implicated; but there is A controversial example, and an argument that conversational implicatures are sometimes part of what is asserted 4. An Example of the Importance of Gricean Ideas for Ordinary Language Analysis Grice’s diagnosis of the error behind Strawson’s performative theory of truth 5. The End of the Ordinary Language School and the Rise of Alternative Conceptions of Philosophy Rethinking the study of meaning, and its role in philosophy; anticipation of things to come

Grice’s Critique of Ordinary Language Philosophy In this chapter we will discuss the first three lectures of Paul Grice’s important work, “Logic and Conversation,” originally delivered at Harvard

198

THE END OF ORDINARY LANGUAGE PHILOSOPHY

University in 1967, as the William James Lectures.1 Grice was a philosopher at Oxford from 1938 to 1967, when he moved to Berkeley, where he spent the next 20 years. He was a leading figure of the post–World War II philosophical movement at Oxford that emphasized the importance of ordinary language, and he co-authored an important and influential article with Peter Strawson on the analytic/synthetic distinction.2 Despite his ordinary language credentials, he was never doctrinaire or uncritical about the implicit philosophical method that characterized much of the philosophy of his era. By the late 1950s and early ’60s he had come to have systematic doubts about the way in which the slogan, “Meaning is Use,” was being implemented by the philosophers he knew.3 The first section of his “Logic and Conversation” lectures begins with an abstract description of a certain linguistic procedure employed by many of his philosophical contemporaries. It goes like this: One is interested in a certain expression or set of expressions—e.g., right/wrong, good/bad, voluntary/involuntary, true/false, knowledge/belief, or any number of others. One then finds some range of simple sentences in which these words appear. Next, one examines the use of these sentences, and one finds that whenever the sentences are used in ordinary, acceptable ways, a certain condition C is fulfilled. Examples of such conditions from the philosophers discussed in previous chapters are: (i) one says that something is good only if one is commending it; (ii) one says that it is true that so and so, only if one is endorsing some previous remark that so and so, or conceding the point of some remark that might be made; and (iii) one says that one has evidence for p only if one doesn’t already know for a fact that p is true. Having found a condition C that must be satisfied if the sentence is to be used in the normal way, one then attributes this condition to the meaning of the expression one is investigating. The satisfaction of C is then said to be part of the meanings of, or propositions expressed by, sentences containing the word. Simple sentences containing the word are said to entail, or presuppose, that C is satisfied. Grice’s reason for outlining this linguistic procedure is to indicate his doubts about it. Although he grants that something like it sometimes 1 Paul Grice, “Logic and Conversation,” in Studies in the Way of Words (Cambridge, MA: Harvard University Press, 1989). 2 Paul Grice and Peter Strawson, “In Defense of a Dogma,” Philosophical Review 65 (1956). See chapter 16 of volume 1 for discussion. 3 See Paul Grice, “The Causal Theory of Perception,” Proceedings of the Aristotelian Society 35 (1961), reprinted in Studies in the Way of Words.

THE LOGIC OF CONVERSATION

199

produces correct results, he also notes that it has been the source of important philosophical errors. Thus, he thinks, we need to find some criterion to distinguish legitimate from illegitimate applications of the procedure. To do this we need to develop a theory of language that neither identifies meaning with use, nor completely divorces the two. We need to recognize that although meaning is one of the most important factors governing the use of an expression, it is not the only factor. The main purpose of Grice’s lectures is to begin the task of constructing a theory of language that tells us what some of these other factors are, and how they interact with meaning. There are two points here to be emphasized—the critique of the reigning philosophical methodology, and the recognition of the need for theory. First, Grice was moved by the idea that something systematic seemed to be going wrong with many of the analyses offered by ordinary language philosophers. It seemed to him that the source of a number of the errors present in these analyses could be traced to the methodology employed, which directly reflected philosophical views about the relationship between language and philosophy. The general picture endorsed by these philosophers was that philosophical problems arise from misunderstanding and misusing ordinary expressions. This picture was grounded in the ideas (i) that philosophy is concerned with questions of conceptual necessity, rather than with empirical or contingent matters, and (ii) that all necessity is somehow linguistic in origin. Thus, to make progress on the conceptual puzzles of philosophy, it was assumed to be necessary to investigate language. In addition, there was a fairly definite idea about how this was to be done. The claim that meaning is use was seen as a rejection of the idea that the meanings of sentences are Platonic entities that can be grasped, or recognized in the mind’s eye. Meaning was not to be thought of either as something hidden, which could be revealed by elaborate theorizing, or as something transparent to introspection. Instead, linguistic meaning was thought to lie fully in view, in the patterns of use we make of expressions and how they relate to one another. This was the reigning ordinary language picture of philosophy and its relation to language. Grice by no means rejected all of it. He accepted the linguistic explanation of necessity, and the centrality of the study of language for philosophy. His concern was that the investigation of meaning and use should be made theoretically more sophisticated and systematic. He understood that the meaning of an expression could not directly be read off the situations in which simple sentences containing it are

200

THE END OF ORDINARY LANGUAGE PHILOSOPHY

used. He believed that by following a systematic approach, we would see language use as the product of several interacting systems—including a system of conventionally encoded meanings, plus communication systems governing the efficient exchange of information. This is the second point to emphasize about Grice’s project—the need to take a systematic and theoretical approach to language. The recognition of this point was an important departure from Wittgenstein’s idea that there is no room in philosophy for theories or explanations, and that all the philosopher has to do is describe, or bring to the forefront, the meanings of expressions which are already there, and fully revealed by how they are used. That approach had been tried for close to two decades, and its limitations had become evident. Grice’s lectures were a clear signal that many analytical philosophers—in England as well as the United States—were moving beyond the antitheoretical paradigm that so many ordinary language philosophers had previously followed.

Grice’s Central Idea: The Logic of Conversation Grice’s key theoretical insight is that there are certain natural principles which guide the efficient and rational exchange of information by cooperative language users, and that speakers relying on these principles may use sentences to convey information that goes well beyond the information given by the meanings of the sentences uttered, or the propositions they semantically express. He introduces the idea of a principle of rational and cooperative exchange of information in the second essay. Our talk exchanges do not normally consist of a succession of disconnected remarks, and would not be rational if they did. They are characteristically, to some degree at least, cooperative efforts; and each participant recognizes in them, to some extent, a common purpose or set of purposes, or at least a mutually accepted direction. This purpose or direction may be fixed from the start (e.g., by an initial proposal of a question for discussion), or it may evolve during the exchange; it may be fairly definite, or it may be so indefinite as to leave very considerable latitude to the participants (as in a casual conversation). But at each stage, some possible

THE LOGIC OF CONVERSATION

201

conversational moves would be excluded as conversationally unsuitable. We might then formulate a rough general principle which participants will be expected (ceteris paribus) to observe, namely: Make your conversational contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which you are engaged. One might label this the Cooperative Principle.4 He breaks down the general idea expressed by the Cooperative Principle into four sets of conversational maxims. maxims of quantity 1. Make your conversational contribution as informative as is required (by current conversational purposes). In other words, don’t say too little. 2. Don’t make your conversational contribution more informative than is required. Don’t say too much. maxims of quality 1. Don’t say what you believe to be false. 2. Don’t say that for which you lack adequate evidence. maxim of relevance Make your conversational contribution relevant to the purpose of the conversation—i.e., be relevant. maxims of manner 1. Avoid obscurity of expression. 2. Avoid ambiguity. 3. Be brief. 4. Be orderly. Grice uses these maxims to define the notion of a conversational implicature. The basic picture is this. A person utters a sentence S in a certain context. Because of the meaning of S, the utterance of it semantically expresses a certain proposition. This proposition is normally what the speaker says or asserts in the context. Standardly, it is part 4

Ibid., p. 26.

202

THE END OF ORDINARY LANGUAGE PHILOSOPHY

of the information the speaker conveys in the context. However, in many cases it does not exhaust the information the speaker conveys. In addition to what he says, the speaker also implicates various things. Grice’s definition of conversational implicature characterizes one particularly important kind of implicature. The definition may be paraphrased roughly as follows: conversational implicature A speaker conversationally implicates q by saying p iff (i) the speaker is presumed to be observing the conversational maxims, (ii) the supposition that the speaker believes q is required in order to make his saying p consistent with the presumption that he is obeying the maxims, and (iii) the speaker thinks that the hearers can recognize this requirement, and also that they can recognize that the speaker thinks that they can recognize it. Grice comments on this definition on page 31. The presence of a conversational implicature must be capable of being worked out; for even if it can in fact be intuitively grasped, unless the intuition is replaceable by argument, the implicature (if present at all) will not count as a conversational implicature; it will be a conventional implicature. To work out that a particular conversational implicature is present, the hearer will rely on the following data: (1) the conventional meanings of the words used, together with the identity of any references that may be involved; (2) the Cooperative Principle and its maxims; (3) the context, linguistic or otherwise, of the utterance; (4) other items of background knowledge; and (5) the fact (or supposed fact) that all relevant items falling under the previous headings are available to both participants and both participants know or assume this to be the case. A general pattern for the working out of a conversational implicature might be given as follows: “He has said that p; there is no reason to suppose that he is not observing the maxims, or at least the Cooperative Principle; he could not be doing this unless he thought that q; he knows (and knows that I know that he knows) that I can see that the supposition that he thinks that q is required; he has done nothing to stop me thinking that q; he intends me to think, or is at least willing to allow me to think, that q; and so he has implicated that q.”5 5

Ibid., p. 31.

THE LOGIC OF CONVERSATION

203

At this point, a word must be said about the difference between conversational implicature, which we have been examining, and conventional implicature, which Grice mentions in passing in the previous passage. Here are some examples of conventional implicature: An utterance of She is poor but honest is said to conventionally implicate—in virtue of the conventional meaning of but—that there is some contrast between poverty and honesty; an utterance of He is an Englishman, and, therefore, brave is said to conventionally implicate—in virtue of the conventional meaning of therefore—that being brave is an expected consequence of being an Englishman; an utterance of He isn’t here yet is said to conventionally implicate—in virtue of the conventional meaning of yet—that he is expected (at some point) to be here; and an utterance of It wasn’t Sam who solved the problem is said by some to conventionally implicate—in virtue of the conventional meaning attached to the construction It was (wasn’t) NP who VPed—that someone solved the problem. (Contrast this last with Sam didn’t solve the problem, which does not carry the suggestion that someone solved the problem.) In each case, the conventional implicature is part of the information conveyed by an utterance of the sentence. In each case, this implicature is generated by some aspect of the meaning of the sentence—either by the meaning of one of the words it contains, or by the meaning associated with its particular syntactic structure. However, despite the fact that the implicature arises from an aspect of the meaning of the sentence, the proposition implicated is not, according to Grice, part of what is asserted by an utterance of the sentence, nor, typically, is it entailed by what is asserted. Moreover, when a sentence like Martha hasn’t arrived yet is embedded under a verb like say or assert and used to report the assertions of others—as in The butler at the door said that Martha hasn’t arrived yet—the conventional implicature, that Martha is expected, is often understood as attaching to the speaker’s complete utterance reporting the assertion, rather than understood as part of the content of that which the butler is reported to have asserted.6 The Often but not always. For example, an utterance of Margaret believes that he is an Englishman, and therefore brave can easily be understood as attributing to Margaret, rather than the speaker, the implication that being brave is a consequence of being English. There is considerable contextual variation about how the conventional implicature of a constituent clause is incorporated into the information conveyed by utterances of larger sentences containing it. This feature of implicatures is rather special, and seems to mark them off from aspects of the proposition semantically expressed by a sentence (in a context).

6

204

THE END OF ORDINARY LANGUAGE PHILOSOPHY

butler himself might simply have responded—No she hasn’t, sir—when asked whether Martha has appeared at the party, without himself asserting or even suggesting that Martha is expected. The fact that this counts as the butler’s having asserted that Martha hasn’t arrived yet supports the idea that the implicature that she is expected is not part of the proposition semantically expressed by the sentence Martha hasn’t arrived yet, used in reporting the assertion, but rather is an extra suggestion that is attached in various ways to utterances of that sentence, or to utterances of larger sentences containing it. Note also that, even if the speaker turns out to be mistaken in thinking that Martha is expected, his report of what the butler said may still be taken to be true, despite the misleading suggestion of her expected arrival to which it gives rise. All of this supports the contention that conventional implicature is the result of a rather special sort of meaning that some expressions have, and that the conventional implicatures of sentences are standardly not parts of the propositions they semantically express, or of the propositions asserted by utterances of them. According to Grice, conversational implicatures are like conventional implicatures in some ways, but not others. Like conventional implicatures, they are typically not part of, or entailed by, what is said. Unlike conventional implicatures, they do not arise from any special linguistic conventions, or meanings, associated with the sentences used. Rather, they are generated by the interaction of features of particular contexts of utterance with the meaning of the sentence uttered, plus the general conversational maxims that regulate the rational and cooperative exchange of information in conversations. Among the examples of conversational implicatures given by Grice are the following: (i) A is standing by an obviously immobilized car and is approached by B; the following exchange takes place: A: I am out of petrol. B: There is a garage around the corner. (Gloss: B would be infringing the maxim “Be relevant” unless he thinks, or thinks it possible, that the garage is open, and has petrol to sell; so he implicates that the garage is, or at least may be, open, etc.)7

7

Grice, “The Causal Theory of Perception,” p. 32.

THE LOGIC OF CONVERSATION

205

(ii) A is planning with B an itinerary for a holiday in France. Both know that A wants to see his friend C, if to do so would not involve too great a prolongation of his journey: A: Where does C live? B: Somewhere in the South of France. (Gloss: There is no reason to suppose that B is opting out; his answer is, as he well knows, less informative than is required to meet A’s needs. This infringement of the first maxim of Quantity can be explained only by the supposition that B is aware that to be more informative would be to say something that infringed the second maxim of Quality, “Don’t say what you lack adequate evidence for,” so B implicates that he does not know in which town C lives.)8 [In this example, it is implicated that B doesn’t know where in the South of France C lives, since if B did know this, he would have to provide that information in accord with the first maxim of quantity—don’t say too little.] (iii) A is writing a testimonial about a pupil who is a candidate for a philosophy job, and his letter reads as follows: “Dear Sir, Mr. X’s command of English is excellent, and his attendance at tutorials has been regular. Yours, etc.” (Gloss: A cannot be opting out, since if he wished to be uncooperative, why write at all? He cannot be unable, through ignorance, to say more, since the man is his pupil; moreover, he knows that more information than this is wanted. He must, therefore, be wishing to impart information that he is reluctant to write down. This supposition is tenable only if he thinks Mr. X is no good at philosophy. This, then, is what he is implicating.)9 [Here, the writer of the recommendation is evidently not providing the required information. He is saying too little. He knows that we will see this, and he knows that we will conclude that he doesn’t have anything good to say about the student because he thinks the student is no good. So this is what is implicated.]

8 9

Ibid., pp. 32–33. Ibid., p. 33.

206

THE END OF ORDINARY LANGUAGE PHILOSOPHY

(iv) Irony. X, with whom A has been on close terms until now, has betrayed a secret of A’s to a business rival. A and his audience both know this. A says X is a fine friend.10 [In this case, the speaker’s remark X is a fine friend is obviously false. The speaker must be trying to convey something else. A natural assumption is that he is intending to convey the opposite. Note, the speaker has not asserted the proposition that X is a fine friend, and implicated something else. He hasn’t asserted that proposition at all. Note also that nothing in Grice’s theory explicitly predicts which related proposition he will be trying to convey. That must come from something else.] (v) Metaphor. Grice gives an example involving the metaphor, You are the cream in my coffee.11 [Here, the sentence is literally false, since you refers to a person. This is so obvious that the speaker can’t be trying to convey it. Nor is the speaker trying to convey its negation, the literal truth of which is so obvious that it doesn’t need to be conveyed. So it is natural to suppose that the speaker wants to convey the idea that you are as crucial to me as cream is to a good cup of coffee, or something like that.] All of these examples—except perhaps the last case involving the particular metaphor that we chose—are examples of what Grice calls particularized conversational implicatures. They are cases in which assertively uttering some sentence S carries a particular implicature because of special features of the context of utterance. In such cases, the same sentence might be uttered assertively in other normal contexts without carrying the implicature it carried in the first context. One can say He is a fine friend without implicating the opposite; one can say There is a garage around the corner in a conversation in which no one presently needs gas, without implicating that the garage is now open and has gas to sell, and so on. Grice contrasts particularized conversational implicatures with generalized conversational implicatures. A generalized conversational implicature is one that is carried by the sentence uttered in any normal context in which the speaker is presumed to be obeying the conversational 10 11

Ibid., p. 34. Ibid., p. 34.

THE LOGIC OF CONVERSATION

207

maxims, or the Cooperative Principle. A good example of this, and one which is theoretically interesting in its own right, involves disjunctions— sentences of the form A or B. In formal logic, disjunction is standardly treated as a strict truth-functional operation. In particular, the logician’s A or B is true iff either one or both of the disjuncts are true; it is false iff both disjuncts are false. However, philosophers have sometimes questioned whether or has the same meaning in ordinary language as it does in logic. One reason they have questioned this comes from the observation that nearly always when we assert the disjunction of A and B in ordinary language, we do so not because we already know that A is true, or because we already know that B is true. Rather, we assert the disjunction because we have some reason for thinking that it is highly unlikely, perhaps even impossible, that both A and B will fail to be true. In these cases, although we don’t know whether A is true or false, and we also don’t know whether B is true or false, we do have grounds for being confident that B will be true, should A turn out to be false, and we also have grounds for thinking that A will be true, should B turn out to be false. We may express this by saying that when we assert the proposition expressed by a disjunction in ordinary conversation, we generally have non-truth-functional grounds for doing so—grounds for our assertion which are not grounds for asserting the proposition expressed by either disjunct. This is such a pervasive feature of our ordinary use of disjunctions that it might seem to constitute a special feature of the meaning of or in ordinary language which is not part of the meaning it has when used in formal logic. Grice disputes this. An important part of the argument is his explanation of how the alleged special feature of the ordinary meaning of or is, in reality, not a feature of its meaning at all, but rather is a generalized conversational implicature of disjunctions. Suppose that someone assertively utters A or B in a conversation in which he has done nothing to suggest that he is not abiding by the Cooperative Principle and the various conversational maxims. Suppose further that we assume that the or in question is the logician’s or, and that the disjunction has the standard truth conditions given in formal logic. Under these assumptions, the statement the speaker made—the one expressed by A or B—is weaker, and less informative, than the statement expressed by A, as is shown by the fact that the statement expressed by the entire disjunction is entailed by the statement expressed by the disjunct, A, but not vice versa. For the same reason, the statement expressed by A

208

THE END OF ORDINARY LANGUAGE PHILOSOPHY

or B is weaker, and less informative, than the statement expressed by B. The first conversational maxim of quantity directs the speaker to make the strongest, most informative statement he can, provided that the statement would be relevant to the conversation, and provided also that in making the statement he would not be violating other maxims, including the maxim of quality that tells him not to assert that for which he lacks adequate evidence. Since the speaker has chosen to assert the weaker statement—expressed by the disjunction—rather than to assert the statement expressed by either disjunct on its own, and since, presumably, if the disjunction is relevant to the conversation, then each disjunct would also be relevant, the presumption that the speaker is obeying the conversational maxims requires us to conclude that he lacks adequate evidence to assert the statement expressed by either disjunct on its own. But since he must have adequate evidence to assert the statement expressed by the disjunction, he must have non-truth-functional grounds for asserting it—i.e., grounds for asserting it that are not grounds for asserting the statement expressed by either disjunct. In short, someone who asserts the proposition expressed by A or B typically conversationally implicates that he doesn’t know, or have adequate grounds to assert, the proposition expressed by A; he conversationally implicates that he doesn’t know, or have adequate grounds to assert, the proposition expressed by B; and he conversationally implicates that he does have adequate grounds for thinking that both can’t be false. This is precisely the feature of our ordinary use of or that led some philosophers to question whether the ordinary or and the logical or mean the same thing. What Grice has shown is that there is no reason to doubt that they do. Even if we start with a notion of disjunction which is strictly the logical one, we will end up concluding that assertive utterances of these disjunctions conversationally implicate precisely what is implicated by utterances of ordinary disjunctions. Thus, we have no reason to posit a special meaning of or in ordinary language, distinct from its meaning in logic. The feature of the use of disjunctions that Grice and other philosophers have observed can be accounted for by conversational maxims which are needed independently to explain the dynamics of conversation and the exchange of information. We have just argued that assertive utterances of disjunctions normally conversationally implicate that the speaker doesn’t know either disjunct to be true. However, this is not always so. The speaker can, if he likes, cancel the normal implicature. Someone staging a treasure

THE LOGIC OF CONVERSATION

209

hunt may say, Either the prize is in the attic or the prize is in the garage, but I am not saying which, since that would spoil the game. In making this remark, the speaker makes it clear that he is opting out of the normal conversational maxims. In particular, he is not trying to be as informative as possible, since doing so would defeat his special aims. A speaker may indicate this either explicitly, as in the case just cited, or implicitly, as in the case of a hostile witness who, though perhaps answering truthfully, makes it evident that he is trying to give as little information as possible. When it is clear that a speaker has opted out of the Cooperative Principle, and the usual conversational maxims, utterances of sentences that would normally carry conversational implicatures do not carry them in these special circumstances. In these cases, the conversational implicatures are said to be canceled. By contrast, conventional implicatures—which are parts of the meanings of the sentences that carry them—are said not to be cancelable. Another feature of many conversational implicatures, in addition to cancelability, is what Grice calls nondetachability. An implicature which arises from an assertive utterance of a sentence is nondetachable just in case any other way of saying—i.e., asserting—the same thing, for example by uttering some other sentence, would also carry the implicature. If one looks over Grice’s examples, one finds that most of them involve applications of the maxims of quality, quantity, and relevance to what one says or asserts, as opposed to applications of the maxims of manner to the way one chooses to express what one asserts. Conversational implicatures arising in the first way should be nondetachable. Implicatures arising from the maxims of manner may not be.12 For example, compare the following two cases: (i) the speaker says John took off his pants and got into bed; (ii) the speaker says John got into bed and took off his pants. Some might argue that the speaker asserts the same thing in both cases, but conversationally implicates different things. In the first case, the speaker implicates that John took off his pants before he got into bed; in the second case, the speaker implicates that John got into bed and then took off his pants. These implicatures might be traced to the conversational maxim of manner that directs speakers to be orderly. Presumably, this maxim directs speakers to provide a comprehensible order when narrating a sequence of events. This may mean that, in the absence of some indication to the contrary, the order in which one mentions the events should match the order in which they occurred. 12

Grice makes this point on p. 43 of Studies in the Way of Words.

210

THE END OF ORDINARY LANGUAGE PHILOSOPHY

So conversational implicatures are cancelable, and many but not all of them are nondetachable. Another feature of Gricean conversational implicatures is that, standardly, they are not supposed to be parts of the truth conditions of what is said. If a speaker who assertively utters sentence S conversationally implicates Q, and Q turns out to be false, it does not follow that S is false, or that the speaker has asserted anything false, though it does follow that the speaker has said something that is liable to mislead his audience. A conversational implicature is taken to be something over and above the meaning of the sentence uttered, and the truth conditions of that which is said by uttering it. Often, the conversational implicature is generated in part by the meaning of the sentence uttered, and in order to calculate the implicature, it is typically necessary to understand both that meaning and the resulting assertion; but the implicature, as understood by Grice, is usually taken to be something additional to these.

Puzzles and Unresolved Problems This dependence of conversational implicature on meaning (and the content of what is said) gives rise to a certain puzzle that Grice discusses in the third essay. The puzzle arises from the fact that, in many cases, in order to use sentences to generate conversational implicatures, speakers must have a good grasp of what those sentences mean, and of what is said or asserted by utterances of them. For Grice, nothing counts as a conversational implicature unless speakers and hearers are able to work through a pattern of reasoning typically starting with what the sentence uttered means and what the speaker has said by uttering it, and ending with what he must believe if he is obeying the conversational maxims. In light of this, it would seem that it ought to be clear, and not subject to substantial theoretical doubt, how much of the information conveyed by a given utterance is part of the meaning of the sentence uttered, and of what the speaker said, as opposed to how much of the information conveyed has been conversationally implicated. In fact, however, this often does not seem very clear at all. Theorists working on conversational implicature have generated considerable debate regarding how much of the information conveyed by different utterances is due to meaning, and how much is due to conversational implicature. Here is an example of the controversy. Consider the sentence

THE LOGIC OF CONVERSATION

211

I have two children. Assertive utterances of this sentence would normally carry the information that I have exactly two children. But how much of this is due to meaning, and how much is due to implicature? Does the sentence literally mean that I have exactly two children, or does it literally mean that I have at least two children, and conversationally implicate, via the maxims of quantity and relevance, that I have at most two children as well? The answer to this question is not immediately obvious, and so it is not surprising that theorists have argued about it. Here, in my opinion, is a step in the right direction. Consider the analogous sentence I have two beers in the fridge. There is good reason to believe (i) that this sentence literally means that I have at least two beers in the fridge, and (ii) that the implicature that I have at most two beers in the fridge arises only in cases in which the question of whether I have exactly two beers is conversationally relevant. Although that question may sometimes be relevant, it is not always so. For example, someone interested in having a quick beer with me before leaving might ask Do you have two beers in the fridge? In this case, it may be clear that the person wants to know whether I have at least two. Hence, when I answer, Yes, I have two beers in the fridge, and he goes to the fridge and sees I have six, he won’t accuse me of lying, or even of misleading him. Since he could reasonably accuse me of this if my sentence literally meant, and I had actually asserted, that I have exactly two beers in the fridge, there is reason to believe that the sentence doesn’t mean this, and that, in the case imagined, I didn’t assert it. In a different conversational setting, the information conveyed may be richer. Had my guest asked, How many beers do you have in the fridge?, I would naturally have assumed that he wanted to know the precise number of beers in the fridge. Had I answered, I have two beers in the fridge, I would have expected him to understand that I was telling him that I have exactly two. Assuming that this conclusion could, in principle, be reached by applying Gricean principles, one may treat the extra information conveyed in this case as a conversational implicature. Presumably, a similar story could be told about the linguistically parallel sentence I have two children. What it literally means is that I have at least two children, even though utterances of it often carry the additional information that I have at most, and hence exactly, two as well. Is this extra information part of what I assert, when I assertively utter the sentence, or is it merely something I conversationally implicate? It would be convenient for the Gricean model if it were merely

212

THE END OF ORDINARY LANGUAGE PHILOSOPHY

implicated. I must confess, however, that I can’t believe this. If I ask someone how many children he has, and he answers I have two children, it seems to me that he has asserted that he has exactly two; if it turns out that he has more, he would rightly be accused of lying, or at least of saying something false. Similarly, if my guest asks me How many beers do you have in the fridge? and I say I have two beers in the fridge, it seems to me that I have asserted that there are exactly two. These, it seems to me, are cases in which what one asserts goes beyond the meaning of the sentence uttered and the proposition it semantically expresses in the context. Presumably, Gricean maxims play some role in the assertion of this extra information. However, precisely what role they play is unclear, due to the manner in which Grice developed his model. He tended to contrast what is conversationally implicated with what is said, where what is said is usually taken to be asserted, and what is implicated is not. For this reason, our example does not quite fit the picture he articulated.13 This, then, is one example of how disputes can arise about the relationship between what a sentence means, what is asserted by an utterance of it, and what is conversationally implicated by the utterance. If I am right about the proper resolution of this particular dispute, then it shows that there is something amiss, or incomplete, in the Gricean picture. However, even if we suspend judgment about the proper resolution of the dispute over this example, there is something puzzling about the mere fact that a dispute exists. If Grice is right about conversational implicatures, the generation of such implicatures would seem to require speakers to have a firm grasp of precisely what their sentences mean, and what they say or assert by uttering them. The basic Gricean picture underlying most conversational implicatures is this: a speaker utters a sentence that means (or expresses the proposition) p; both speaker and hearers recognize this and see that the speaker has asserted p; they also see, or at least are capable of seeing, that if he is obeying the conversational maxims, then he would not be asserting p unless he also believed q, and wanted them to believe q as well. But if 13 My reason for stating this somewhat vaguely, “He tended to contrast . . . ,” was that Grice himself was cautious (and somewhat unclear) about the precise content of his locutions what is said and what he said. As used in his theory of implicature, say was meant to be a technical term (see p. 121 of Studies in the Way of Words) that nevertheless was supposed to bear some reasonably close relationship to what we ordinarily mean by say and assert. Since Grice never fully clarified this relationship, the content of his model and the precise nature of possible counterexamples to it remain a bit fuzzy. Nevertheless, the example developed above constitutes a challenge to one common and natural way of understanding his theory.

THE LOGIC OF CONVERSATION

213

ordinary competent speakers see, or are capable of seeing, all of this, then they see, or are capable of seeing, precisely what information is part of the meaning of (or proposition expressed by) the sentence uttered, precisely what has been said or asserted, and precisely what further information—which is not part of the meaning of (or the proposition expressed by) the sentence, and which also is not part of what has been asserted—has merely been conversationally implicated. And if all this is so, how could serious theoretical controversy arise? Grice expresses worry about this puzzle on page 49. We must of course give due (but not undue) weight to intuitions about the existence or nonexistence of putative senses of a word (how could we do without them?). Indeed if the scheme which I have been proposing is even proceeding in the right direction, at least some reliance must be placed on such intuitions. For in order that a nonconventional implicature should be present in a given case, my account requires that a speaker shall be able to utilize the conventional meaning of a sentence. If nonconventional implicature is built on what is said, if what is said is closely related to the conventional force of the words used, and if the presence of the implicature depends on the intentions of the speaker, or at least on his assumptions, with regard to the possibility of the implicature being worked out, then it would appear that the speaker must (in some sense or other of the word know) know what is the conventional force of the words which he is using. This indeed seems to lead to a sort of paradox: If we, as speakers, have the requisite knowledge of the conventional meanings of sentences we employ to implicate, when uttering them, something the implication of which depends on the conventional meaning in question, how can we, as theorists, have difficulty with respect to just those cases in deciding where conventional meaning ends and implicature begins? If it is true, for example, that one who says that A or B implicates the existence of nontruth-functional grounds for A or B, how can there be any doubt about whether the word “or” has a strong or weak sense?14 The general question may be put this way: How transparent does meaning have to be if Grice’s account of conversational implicature is to be correct? In addition, we may ask, is there room for some conception of 14

Grice, Studies in the Way of Words, p. 49.

214

THE END OF ORDINARY LANGUAGE PHILOSOPHY

implicature—perhaps different from Grice’s—in which there are theoretical reasons for distinguishing meaning from implicature, even though it is not assumed that speakers themselves, just by virtue of understanding their language, must possess the means for dividing the information conveyed by utterances into that due to meaning and that due to other things? I myself think there must be some important notion of implicature that fits this broader picture, since I don’t see how simply being a competent speaker could be enough to guarantee that one can always distinguish information conveyed due to implicature from information conveyed due to meaning. However, it must be admitted that our understanding of this issue, even today, is fragmentary and far from complete. Moreover, whatever progress we have been able to make owes an enormous debt to the original work on logic and conversation done by Grice.15

An Example of the Importance of Gricean Ideas for Ordinary Language Analysis I will close the discussion of Grice with an illustration of how he applied his ideas about conversational implicature to a particular ordinary language analysis of an important philosophical concept that we looked at in an earlier chapter. Grice finishes up the third of his essays in “Logic and Conversation” with a comment on Strawson’s performative theory of truth. The comment illustrates how Grice’s ideas about implicature allowed him to accommodate certain linguistic observations which, in the absence of any systematic theory of language use, led other ordinary language philosophers astray. What I wish to do is to show that, on the assumption that a certain sort of theory of this kind [in which truth is a property of statements or utterances that correspond to facts] is correct, then . . . it is possible to accommodate the linguistic phenomena which led Strawson to formulate the original version of the speech-act theory. . . .

For some of my own further thoughts on this issue, see my Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity (New York: Oxford University Press, 2002), chapter 3; and my “Naming and Asserting,” in Zoltan Szabo, ed., Semantics vs. Pragmatics (Oxford University Press, forthcoming).

15

THE LOGIC OF CONVERSATION

215

If some such account of It is true that . . . is correct (or indeed any account which represents saying It is true that p as equivalent to saying something about utterances) then it is possible to deal with the linguistic facts noted by Strawson. To say Smith is happy is not to make a (concealed) reference to utterances of a certain sort, whereas to say It is true that Smith is happy is to do just that, though of course if Smith is happy, it is true that Smith is happy. If I choose the form which does make a concealed reference to utterances, and which is also the more complex form, in preference to the simpler form, it is natural to suppose that I do so because an utterance to the effect that Smith is happy has been made by myself or someone else, or might be so made. Such speech acts as endorsing, agreeing, confirming, and conceding, which Strawson (presumably) supposed to be conventionally signaled by the use of the word true, are just those which, in saying in response to some remark “That’s true,” one would be performing (without any special signal). And supposing no one actually to have said that Smith is happy, if I say “It is true that Smith is happy” (e.g., concessively) I shall implicate that someone might say so; and I do not select this form of words as, for example, an inquiry whether Smith is happy when I do not wish this implicature to be present.16 Grice here gives essentially the same explanation that we gave in chapter 5 of the feature of the use of sentences containing the word true that motivated Strawson’s performative theory of truth. What Grice is saying is that although S and It is true that S are used to make assertions that are equivalent in content, uttering It is true that S carries an additional conversational implicature that we are responding to a statement that someone has either actually made, or might potentially make. It carries this implicature because the maxims of manner require us to get our point across in the simplest, most efficient manner. If all we are interested in doing is conveying that Smith is happy, then, although this could be conveyed either by uttering Smith is happy or by uttering It is true that Smith is happy, we ought to utter the former, because it is simpler and more direct. (Be brief!) Hence, on those occasions in which the speaker chooses the latter sentence, the presumption that he is obeying the maxims requires one to look for 16

Grice, “The Causal Theory of Perception,” pp. 56–57, my addition in brackets.

216

THE END OF ORDINARY LANGUAGE PHILOSOPHY

some additional reason for his remark. Since the more complex form It is true that Smith is happy predicates truth of a statement, it is natural to suppose that the speaker intends his remark as a response to that statement—an endorsement of the statement, if it has previously been made, or a concession of the statement, if one imagines it being made. Because we can explain this fact by appeal to the general conversational maxims, we don’t need a special performative theory of the meaning of sentences containing true. In this way, having a clear conception of the distinction between meaning and implicature helps guard against philosophical error.

The End of the Ordinary Language School and the Rise of Alternative Conceptions of Philosophy This concludes our discussion of the ordinary language school of philosophy that flourished in Great Britain in the two decades after World War II. The central doctrines guiding this approach were (i) that philosophical problems arise from the misuse of language and are to be solved by getting clear about the meanings of words, (ii) that philosophical analysis consists less in uncovering hidden logical forms and formulating precise necessary and sufficient conditions for the application of a word or concept, than in opportunistically assembling reminders about how philosophically significant words are used in ordinary settings, (iii) that meaning is use, (iv) that the philosophical study of meaning is to proceed by informal, case-by-case investigations, and (v) that systematic theories of meaning are not required and are not to be sought. By the time Grice gave his lectures, “Logic and Conversation,” these ideas had run their course. The antitheoretical approach of the ordinary language philosophers simply could not sustain the emphasis on discovering truth about meaning that their conception of philosophical problems demanded. Grice’s recognition of this came from within the ordinary language school itself, and pointed in one direction beyond it. Across the Atlantic in America, other fundamentally different views were brewing—about meaning, about the importance of theory, and about the nature of philosophical problems. In the chapters that follow, we will look at three of these views. First, we will examine Willard Van Orman Quine’s attack on the notions of meaning and reference—encapsulated in his doctrines of the indeterminacy of translation and the inscrutability of reference, which

THE LOGIC OF CONVERSATION

217

are presented in his book Word and Object, published in 1960, and in his essay “Ontological Relativity,” delivered at Columbia University as the Dewey Lectures in 1968, and published in 1969.17 Quine’s attack, which initially grew out of his rejection of analyticity and reformulation of the verificationism of the logical positivists, sought to eliminate our ordinary notions of meaning and reference, and replace them with more scientifically acceptable substitutes. He viewed meaning and reference, as ordinarily understood, as pre-scientific concepts that have no substantial role to play either in science or philosophy. Accordingly, he proposed that philosophical problems are not linguistic in nature, but rather represent abstract, highly theoretical issues that arise in the construction of a unified empirical theory of the world. After discussing Quine, we will turn to Donald Davidson’s approach to truth and meaning. While retaining some of Quine’s skepticism about meaning as ordinarily understood, Davidson held that a theoretically fruitful study of how language is used to represent the world is both possible and important for philosophy. On his view, the key to such a theory lies in applying to natural languages like English the methods introduced by the great logician, Alfred Tarski, to define truth for formalized languages.18 For Davidson, the central task for a theory of the interpretation of a natural language is the development of a systematic theory that derives statements of the truth conditions of the infinitely many sentences of the language from axioms about the interpretations of its finitely many primitive expressions. In articulating this view, Davidson reconnected the thriving formal tradition in logic, philosophical logic, and the philosophy of language—temporarily shunted aside by ordinary language philosophers—with the larger philosophical mainstream. Although Davidson’s views were influential throughout the world of analytic philosophy, they were particularly well received in Great Britain. The reason for this is evident when one remembers both that British philosophers had long been convinced that the key to progress in philosophy was a proper understanding of meaning, and that the ordinary language school had foundered precisely for want of a theoretically fruitful way of pursuing this end. Thus, it is not surprising that when Donald Davidson came along, 17 Quine, Word and Object (Cambridge, MA: MIT Press, 1960), and Ontological Relativity and Other Essays (New York and London: Columbia University Press, 1969). 18 Alfred Tarski, “The Concept of Truth in Formalized Languages” (1935), translated and reprinted in Tarski, Logic, Semantics, Meta-Mathematics, 2d edition (Indianapolis: Hackett, 1983).

218

THE END OF ORDINARY LANGUAGE PHILOSOPHY

many younger British philosophers felt that they had found precisely what their elders had so sorely needed. Finally, we will close our discussion of the period by examining Saul Kripke’s book Naming and Necessity, originally given as three lectures at Princeton University in January of 1970.19 In those lectures Kripke brushed aside Quinean criticism of the traditional metaphysical notions of necessity and possibility, exploded the long-held doctrine of the linguistic nature of necessity, clearly distinguished the metaphysically necessary from the epistemologically apriori, reintroduced essence and essential properties as legitimate subjects of philosophical inquiry, and broke the dominance of the linguistic model as the source for all philosophical theorizing. Though no enemy of meaning, conceptual analysis, or the analytic, Kripke opened the doors of twentieth-century analytic philosophy to a broader set of philosophical concerns. 19

Saul Kripke, Naming and Necessity (Cambridge, MA: Harvard, University Press, 1980).

SUGGESTED FURTHER READING FOR PART FOUR

Main Primary Sources Discussed Grice, Paul. “The Causal Theory of Perception.” Proceedings of the Aristotelian Society, 35, 1961, reprinted in Studies in the Way of Words. ———. “Logic and Conversation.” Given in 1967 as the William James lectures at Harvard University; reprinted in Studies in the Way of Words (Cambridge, MA: Harvard University Press, 1989).

Additional Primary Sources Grice, Paul, and Peter Strawson. “In Defense of a Dogma.” Philosophical Review 65 (1956). Strawson, Peter. “Truth.” Analysis 9, 6 (1949):83–97.

Additional Recommended Reading Soames, Scott. Chapter 3 of Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity (New York: Oxford University Press). ———. “Naming and Asserting.” In Zoltan Szabo, ed., Semantics vs. Pragmatics (Oxford University Press, forthcoming). Thau, Michael. Chapter 4 of Consciousness and Cognition. New York: Oxford University Press, 2002.

This page intentionally left blank

PA R T F I V E THE PHILOSOPHICAL NATURALISM OF WILLARD VAN ORMAN QUINE

This page intentionally left blank

CHAPTER 10 THE INDETERMINACY OF TRANSLATION chapter outline 1. The Road to Word and Object Quine’s conception of philosophy; his rejection of the analytic/synthetic distinction; criticism by Grice, Strawson, and Carnap; Word and Object as Quine’s response 2. Quine’s Central Theses and His Strategy for Establishing Them The underdetermination of translation by data and the indeterminacy of translation 3. The Argument for the Underdetermination Thesis Translation theories, stimulus meanings as data, and the empirical predictions made by translation theories Radical translation: Quine’s example of rabbit and gavagai “Translations” of different speakers of the same language Why it is plausible that “incompatible” theories of translation may be compatible with all data about stimulus meanings, and other observational facts about language use Why the underdetermination thesis should not be surprising 4. Quine’s Arguments for the Indeterminacy of Translation The dubious argument from behaviorism The challenging argument from physicalism plus the alleged underdetermination of translation of physical truths 5. Evaluating Quine’s Case for Indeterminacy The danger of equivocation about the determination relation Different interpretations of the determination relation; on some physicalism is plausible, on some the underdetermination of translation by physics is plausible, but on no interpretation are both plausible; how the mistaken identification of necessity with aprioricity contributed to the illusion that Quine’s argument was strong; failure of the argument

The Road to Word and Object The subject of this chapter is Quine’s Word and Object, which was written in the late 1950s and published in 1960, and which is widely

224

QUINE’S PHILOSOPHICAL NATURALISM

regarded as his most systematic and important work.1 The man who reveals himself in these pages is a very different kind of philosopher from the ordinary language philosophers we have been discussing. He rejects the doctrine that philosophical problems arise from confusion about the meanings of words or sentences, and rejects the conception of philosophy that sees its central task as providing analyses of meaning that will solve or dissolve philosophical problems. He rejects these views because he rejects their presuppositions—the presupposition that words and sentences have meanings in the sense that previous philosophers have assumed that they do, and the presupposition that we can separate out facts about meanings or linguistic conventions from the totality of all empirical facts. For Quine, philosophy is continuous with science. It has no special subject matter of its own, and it is not concerned with the meanings of words in any special sense. Philosophical problems are simply problems of a more abstract and foundational sort than the ordinary problems of day-to-day science. In chapter 16 of volume 1, I discussed Quine’s famous attack on the analytic/synthetic distinction, given in his paper, “Two Dogmas of Empiricism.”2 There, I argued that although his attack advanced the discussion of this distinction in significant ways, he did not establish his central conclusions. I will not here go over the details of his argument or my criticism of it. However, I will say a few general words to set the stage for what he does in Word and Object, which we will examine carefully. In “Two Dogmas,” Quine isolates what he takes to be a family of notions, including analyticity (or truth in virtue of meaning), necessity, and synonymy (sameness of meaning). He argues that the notions in this family can be defined in terms of one another, but cannot be defined or explained using terms outside the family. This is significant, since he thinks that all of these notions are unclear, suspect, and in need of explanation if they are to be legitimately employed. Since he believes that they cannot be given the required explanation, he concludes that there is no such thing as analyticity, no such thing as necessity, and no such thing as synonymy. His conclusions, though widely influential, provoked vigorous rebuttals in the mid-’50s by several important 1 2

W. V. Quine, Word and Object (Cambridge, MA: MIT Press, 1960). W. V. Quine, “Two Dogmas of Empiricism,” The Philosophical Review 60 (1951).

THE INDETERMINACY OF TRANSLATION

225

philosophers. For example, Paul Grice and Peter Strawson jointly wrote an article in which they criticized Quine for drawing what seemed to them to be absurdly radical conclusions on the basis of an unacceptably restrictive conception of what constitutes an adequate explanation of a term or concept.3 As to the radical nature of his conclusions, they pointed out that if it makes sense to talk about a word or a sentence meaning something, then it must make sense to talk about two words, or two sentences, meaning the same thing, and hence being synonymous. Consequently, Quine’s conclusion that it doesn’t make sense to talk about words, or sentences, being synonymous carries with it the obvious corollary that it makes no sense to speak of a word or a sentence meaning anything at all. If no expressions can be synonymous, then no expressions can have meanings, and hence no expressions can be meaningful. This conclusion, quite understandably, seemed to Grice and Strawson to be far too paradoxical and extreme to be acceptable. Another philosopher who rejected Quine’s view in “Two Dogmas of Empiricism” was his friend and teacher, Rudolf Carnap, to whom Word and Object is dedicated. In the mid-’50s Carnap wrote what was, essentially, a reply to Quine’s rejection of the notion of meaning. It is reprinted in Carnap’s book Meaning and Necessity, as appendix D, “Meaning and Synonymy in Natural Languages.”4 There Carnap tries to show how the notion of the meaning of a term, over and above its reference, can play an important role in empirical theorizing about language users. He argues that although there are empirical uncertainties in establishing what either the meaning or the reference of a term is, there are reasonable empirical methods for bringing evidence to bear on both questions; as a result, hypotheses about meaning and reference are in the same boat. If Carnap is right, then Quine was wrong to dismiss notions like meaning and synonymy while apparently retaining such notions as reference. Contrary to Quine, Carnap maintains that the two are scientifically on a par, both scientifically required, and both scientifically respectable. Word and Object can be seen as Quine’s response to these criticisms. In this book he tries to articulate a scientifically acceptable account of 3 H. P. Grice and P. F. Strawson, “In Defense of a Dogma,” Philosophical Review 65 (1956). See chapter 16 of volume 1 for discussion. 4 Rudolf Carnap, Meaning and Necessity (Chicago: University of Chicago Press, 1956).

226

QUINE’S PHILOSOPHICAL NATURALISM

language use in which a language is viewed as a set of verbal responses to verbal and non-verbal stimuli. The ordinary notions of meaning and reference are rejected as illegitimate pre-scientific concepts that have no place in a scientifically respectable description of the world. As Quine sees it, claims about meaning and reference cannot be given the empirical support needed to justify accepting them. Moreover, he thinks that even if we were somehow given the complete totality of facts provided by an ideal physics, we still would not have any way of making a correct and determinate assignment of meanings and referents to expressions. The totality of physical facts simply leaves it open which of a large number of different meanings particular words might have, and which of a large number of different objects they might refer to. The problem, in Quine’s view, is not that we are in danger of forever remaining ignorant of the facts about meaning and reference. The problem is that there are no genuine facts to be ignorant of. If the totality of physical truths leaves it undetermined whether a certain word means such and such, or refers to so and so, then the claim that it does mean such and such, or that it does refer to so and so, is not a genuine truth. Since there are no significant truths about meaning and reference, the ordinary notions of meaning and reference have no place in scientific descriptions of the world, and so should be replaced with scientifically respectable substitutes. That is the overall picture painted by Quine.

Quine’s Central Theses and His Strategy for Establishing Them5 Quine attempts to establish his view by starting with an empirical theory in which the notion of meaning plays a central role, and showing (i) that the class of all possible data for such a theory radically underdetermines the claims about meaning that it makes, and (ii) that this indeterminacy could not be resolved even if we had access to all physical facts. The empirical theories that he chooses to work with are theories of translation—that is, theories which attempt to translate the words and 5 An early and abbreviated version of the material in the remainder of this chapter appears in section I of my article, “The Indeterminacy of Translation and the Inscrutability of Reference,” Canadian Journal of Philosophy 29 (1999): 321–370.

THE INDETERMINACY OF TRANSLATION

227

sentences of one language into the words and sentences of a different language. Like all scientific theories, translation theories are tested against observational data—in this case, observations about the linguistic behavior of the speakers of the two languages. Quine’s first task in chapter 2 of Word and Object is to specify precisely what this data amounts to, and how it bears on theories of translation. Next, he argues that the class of all possible data of this sort vastly underdetermines the choice of a theory of translation. That is, he tries to establish a thesis that we might call the underdetermination of translation by data. the underdetermination of translation by data Let L1 and L2 be arbitrary languages, and let D be the set of all observational truths (known and unknown) relevant to translation from one to the other. For any theory of translation T for L1 and L2, compatible with D, there is a theory T⬘, incompatible with T, that is compatible with, and equally well supported by, D. It should be noted that, according to Quine, the incompatibility of different theories, each equally supported by all possible data, is not confined to matters of poetic connotation or subtle matters of style. Rather, the difference between alternative translations is alleged to be vast, going to the heart of the languages and the conception of the world shared by their speakers. Even worse, Quine maintains, the incompatibility is not resolvable even by the addition of all physical facts, which he takes to be the totality of all truths of nature.6 The doctrine that this is so is his thesis of the indeterminacy of translation. the indeterminacy of translation Translation is not determined by the set N of all truths of nature, known and unknown. For any pair of languages and theory of translation T for those languages, there are alternative theories of translation, incompatible with T, that accord equally well with N. All such theories are equally true to the

See, Quine, “Reply to Chomsky,” in D. Davidson and J. Hintikka, eds., Words and Objections (Dordrecht: Reidel, 1969), 303; also Quine, “On the Reasons for Indeterminacy of Translation,” Journal of Philosophy 67 (1970): 178–83.

6

228

QUINE’S PHILOSOPHICAL NATURALISM

facts; there is no objective matter of fact on which they disagree, and no objective sense in which one is true and the other is not. In this chapter, our task will be (i) to reconstruct the argument Quine gives for this thesis, and (ii) to evaluate whether the argument is persuasive in establishing its conclusion. In the next chapter we will continue the discussion by (iii) tracing how Quine extends the argument to encompass indeterminacy of reference as well as meaning, (iv) sketching his rationale for the rejection of both of these notions and their replacement by allegedly more scientifically respectable substitutes, (v) drawing out further radical consequences of his position, and (vi) arguing that these consequences can be turned against Quine himself in a way that shows his position to be self-undermining.

The Argument for the Underdetermination Thesis The place to begin is with the nature of theories, and the data against which they are to be tested. First, consider theories of translation. A theory of translation for two languages correlates individual words of each language with words or phrases of the other language; this correlation is then used to correlate the sentences of the two languages. Any system of establishing such correlations may be counted as a translation manual or theory. We may take such a theory as yielding (infinitely many) theorems of the form: Word or phrase w1 in L1 means the same as word or phrase w2 in L2. Sentence S1 in L1 means the same as sentence S2 in L2. What empirical predictions do theories of this type make? What data are they tested against? Obviously, they make claims about what various expressions mean. However, there is a difficulty in discovering what the expressions in someone else’s language mean. We can’t ask the foreigner to tell us what his words mean, since before we have a translation manual we can’t understand him, and he can’t understand us. All we can do, Quine thinks, is observe the situations in which the

THE INDETERMINACY OF TRANSLATION

229

foreigner uses certain expressions and compare those with the situations in which we would use various expressions of our own.7 For example, suppose that L1 is English and L2 is some native language never before encountered by English speakers. (Quine refers to the translation of such languages as “radical translation.”) Suppose also that we can somehow discover the native’s words for yes and no, as well as discern the difference between the ways questions and assertions are expressed in the native’s language. (I will return to this point later.) It then seems possible that we might discover that the native will answer yes in response to an utterance of the one-word interrogative sentence Gavagai? in essentially the same situations that we would answer yes to an utterance of the one-word interrogative sentence Rabbit? On this basis, it would seem reasonable to translate the two sentences onto each other, as well as translating the one-word exclamation Gavagai! as Rabbit! or Lo, a rabbit! We might then do the same thing in a range of other cases. For example we might find that we could translate the native sentence Gleep! as Red! or Lo, something red! It should be noted that although this procedure will work in getting our system of translation started, there are real limits to it. Two things make it work. (i) Gavagai? and Rabbit? are sentences that speakers will assent to, or dissent from, in part on the basis of what they are observing at the moment. In this way they are unlike the sentences The earth is round, Columbus discovered America, and Molecules are made up of atoms. These latter are sentences that speakers will assent to, or dissent from, independent of what they are observing at the time. Thus, the technique that worked for translating Gavagai and Rabbit will not work for translating all sentences. (ii) Assent to, or dissent from, Gavagai? and Rabbit? depends on observation plus a minimum of supplementary background beliefs. For example, compare the one-word sentences Rabbit? and Red?, on the one hand, with the sentences Bachelor? and Philosopher?, on the other. Intuitively, assent to, or dissent from, the last two sentences depends on observation, understanding the words, and considerable background Here, Quine elaborates on Carnap’s suggestions in sections 2 and 3 of “Meaning and Synonymy in Natural Languages.” His aim is to show that what amount essentially to Carnap’s own empirical procedures radically underdetermine both extension (reference) and intension (meaning).

7

230

QUINE’S PHILOSOPHICAL NATURALISM

information about the individual one is observing. Because of this, varying the background information available to the speaker may result in great variation in assent and dissent. Thus, it wouldn’t be reasonable to require the native’s assent to Leepus?, when queried about an arbitrary individual, to always match my assent to Bachelor?, when queried about the same individual, in order to translate the one as the other. We can bring these points together with the help of the following definitions. stimulus meaning The stimulus meaning of a sentence S (for a speaker at a time) is a pair of classes—the class of situations which would prompt the speaker to assent to S if queried (the affirmative stimulus meaning of S), and the class of situations which would prompt the speaker to dissent from S if queried (the negative stimulus meaning of S). (Quine does not take the affirmative and negative stimulus meanings of S to exhaust all situations in which the speaker is queried about S; they don’t include situations in which the speaker withholds or suspends judgment.)8 occasion sentences S is an occasion sentence (for a speaker) iff the speaker’s assent to, or dissent from, S depends (in part) on what the speaker is observing.9 observation sentences S is an observation sentence in a language L iff (i) S is an occasion sentence for speakers of L, and (ii) the stimulus meaning of S varies very little from one speaker of L to another. (The thought behind (ii) is that sameness of stimulus meaning of S across a population is a reasonable approximation of the maximal degree to which assent to S, or dissent from S, is dependent on observation alone, free of substantial background assumptions that may be idiosyncratic to particular speakers.)10 See Word and Object, pp. 32–33. See ibid., pp. 35–40. 10 See ibid., pp. 41–46. 8 9

THE INDETERMINACY OF TRANSLATION

231

Using these notions, we can give Quine’s conception of the class of empirical data against which theories of translation are to be tested. According to him, the empirical data relevant to theories of translation are statements about the stimulus meanings of sentences.11 The empirical predictions made by translation theories are summarized by the following principles: empirical predictions of translation theories (i) Translation of observation sentences must preserve stimulus meaning. If a translation theory states that an observation sentence S1 in L1 means the same as a sentence S2 in L2, then the theory predicts that S1 and S2 have the same stimulus meanings in their respective linguistic communities.12 (ii) Translation must preserve the stimulus synonymy of pairs of occasion sentences. If S1 and S2 are occasion sentences of L1, and if a translation theory states both that S1 means the same in L1 as P1 in L2 and that S2 means the same in L1 as P2 in L2, then the theory predicts that S1 and S2 have the same stimulus meanings for speakers of L1 iff P1 and P2 have the same stimulus meanings for speakers of L2.13 (iii) Translations of truth functional operators—and, or, not, etc.—have recognizable effects on stimulus meaning. For example, if a theory translates an expression e of a language L as meaning the same as not in English, then the theory predicts that adding e to sentences of L reverses stimulus meaning; and if it translates e as meaning the same as or in English, then a sentence of L that results from using e to combine a pair of sentences A and B is assented to whenever either A is assented to or B is assented to, and it is dissented from whenever 11 Statements about stimulus meaning are established by induction and generalization based on regularities in the patterns of observed assents and dissents. Although strictly speaking, the contents of these statements go beyond the observations used to justify them, I will continue to refer to facts about stimulus meaning as observational data for theories of translation (using observational in a somewhat relaxed sense). 12 Word and Object, page 44. 13 Ibid., pp. 46–51.

232

QUINE’S PHILOSOPHICAL NATURALISM

both are dissented from. Similar claims hold for other truth-functional operators.14 The rationale behind the second principle can be illustrated as follows. Consider the two English sentences Bachelor! and Unmarried man!. We have already noted that these will not be observation sentences, since their stimulus meanings will vary from speaker to speaker, depending on the background information available to the speaker about the individual being observed. However, they are still occasion sentences—i.e., they are sentences for which assent or dissent depends in part on observation (and typically varies with who is observed). Moreover, even though the stimulus meanings of these sentences vary from speaker to speaker, they will vary together—as a pair. Thus, we can expect that for every individual speaker, Bachelor! and Unmarried man! will have the same stimulus meanings.15 This places an empirical constraint on translation. If Leepus! in the native language is translated as Bachelor! in English, and Lap nox! is translated as Unmarried man!, then the two native sentences must be stimulus-synonymous in the native language, since the two sentences they are translated into are stimulus synonymous occasion sentences of English. That is what the second principle tells us. The third principle is straightforward as far as it goes. However, there are some potential problems with it that I have sidestepped by stating the rules for translating the different truth-functional operators in a weak form, as conditionals, rather than in a strong form, as biconditionals. In Word and Object Quine gave the strong forms, but later retreated after recognizing that they were untenable. This is acknowledged in The Roots of Reference, where the problem is attributed to logically compound sentences containing component sentences on which speakers withhold judgment (and hence neither assent to nor dissent from).16 Consider, for example, disjunction. Although it is true that English speakers will assent to a disjunction A or B whenever they Ibid., pp. 57–61. This is, of course, an oversimplification. As Quine observes in a footnote on page 46, various complications having to do with ages, divorce, bachelors of arts, and other matters lead to some relatively small differences in the individuals that many ordinary speakers are willing to call ‘bachelors’ as opposed to ‘unmarried men’. In the interest of simplicity, Quine proposes to overlook these complications, as we will do here. In general, when in principles like (i–iii), we talk of sameness of stimulus meaning, what we really have in mind are stimulus meanings that are more or less the same. 16 W. V. Quine, The Roots of Reference (La Salle, IL: Open Court, 1974). 14 15

THE INDETERMINACY OF TRANSLATION

233

assent to A or to B, and dissent from A or B whenever they dissent from both, there is no general rule determining the verdict they will give to A or B in cases in which they suspend judgment on both A and B. In some cases they will suspend judgment on the disjunction as well, while in others they will assent to it. Which cases fall into which category depends on the particular disjuncts and the particular speakers being queried. In The Roots of Reference, Quine indicates that this observation about the stimulus meanings of disjunctions would hold true both for communities that assign intuitionist interpretations to the usual logical operators, and for communities that assign the classical truth-functional interpretations.17 From this and similar observations, he concludes that the difference in meanings assigned to the logical operators by intuitionists and classical logicians is not determinable solely on the basis of stimulus meaning.18 Despite this, he goes on to argue that the similarity in the two interpretations can be captured by appeal to stimulus meaning, and that the set of those logical truths that are accepted by both intuitionists and classical logicians is something that can, in effect, be read off community-wide dispositions to assent and dissent. Unfortunately, even this claim turns out to be highly problematic.19 In general, the task of deriving interpretations of the logical operators from community-wide stimulus meanings is fraught with difficulty, and the overall goal of giving a purely behavioral analysis of the basis of logical truths appears well-nigh unachievable. Since the details of this would take us too far afield, we will avoid the problem by sticking to the weak version of (iii). We now have three sets of empirical data about which we can take translation theories as making predictions. According to Quine, this virtually exhausts all the empirical data there are for judging translation theories. The only addition he seriously considers involves constraints based on the notions of stimulus analyticity and stimulus contradiction.20 Intuitionists differ from classical logicians in, among other things, not accepting all instances of A or ~A. 18 The Roots of Reference, p. 78. 19 For an excellent discussion of the many problems with Quine’s contention, see Alan Berger, “Quine on ‘Alternative Logics’ and Verdict Tables,” Journal of Philosophy 77 (1980): 259–77. One problem particularly worth noting is the difficulty in coming up with an interpretation of one of the key notions involved in the definition of stimulus meaning—namely, what it is to dissent from a sentence—that is neutral between classical and intuitionist understandings of negation. Berger discusses this on pp. 276–77. 20 See Word and Object, pp. 55 and 66. 17

234

QUINE’S PHILOSOPHICAL NATURALISM

stimulus analyticity and stimulus contradiction S is stimulus analytic in L iff virtually all speakers of L assent to S in all stimulus situations. S is stimulus contradictory iff virtually all speakers of L dissent from S in all stimulus situations. Quine considers adding principle (iv) as a possible empirical constraint on translation.21 (iv) Translation must preserve stimulus analytic and stimulus contradictory sentences. If S1 is a stimulus analytic (contradictory) sentence of L1, and if a theory of translation states that S1 means the same in L1 as S2 in L2, then the theory predicts that S1 is stimulus analytic (contradictory) in L1 iff S2 is stimulus analytic (contradictory) in L2. Nevertheless, Quine realizes that this potential requirement is problematic, since it might easily fail to distinguish universally believed native falsehoods from English truths (and vice versa).22 For example, the criterion would allow us to translate There have been black dogs onto a native sentence that might more naturally be read The sun god is mighty. Worse, it might prevent translating The sun god is mighty onto a sentence in the native language which one would intuitively think does say that the sun god is mighty, simply because English speakers don’t have the same beliefs about the universe as the natives do. For reasons like this, the status of (iv) is a bit up in the air. In addition, there might be other adjustments we would want to make to the principles. In general, these adequacy conditions on translation should be taken with a grain of salt. Stimulus meanings, which are based on assent and dissent, are rather coarse-grained in the distinctions they allow for. As we have seen from our study of Grice, the information conveyed by an utterance is due to a variety of factors—including the (truth-conditional) meaning of the sentence uttered, its conventional implicatures, and its conversational implicatures—either particularized or generalized. All of these may be factors in prompting assent or dissent.23 This raises the possibility that two sentences might differ in meaning, even though utterances of them would generally convey the Ibid., p. 68. Ibid., p. 69. 23 Thanks to John Hawthorne for impressing on me the significance of this point for Quine’s adequacy conditions on translations. 21 22

THE INDETERMINACY OF TRANSLATION

235

same information (and hence prompt the same assents or dissents) owing to the fact that utterances of one of the sentences generally would implicate (rather than entail) a proposition that was part of the meaning of the other.24 In such cases, the difference in meaning between the two sentences would be all but invisible to Quine’s radical translator, and Quine’s constraints on the empirical adequacy of translation would allow the sentences to be assimilated to one another. Similarly, if there are sentences that have the same literal (truth-conditional) meanings, but differ regarding their implicatures, they might differ substantially in stimulus meaning—in which case Quine’s constraints on empirical adequacy would rule out translating the sentences as meaning the same thing. In light of these possibilities, one might try to refine the account of stimulus meaning, or the adequacy conditions on translation stated in terms of them, so as to be sensitive to the subtle differences that mark the way we treat conveyed information in different Gricean categories. (Think of the Gricean tests involving cancelation and detachability, for example.) However, it is far from clear how much refinement of this sort could be achieved within Quine’s starkly behavioristic perspective on radical translation. At some point, one must recognize the severe limitations he has imposed on himself, as well as the tentative and approximate character of his adequacy conditions on translation. The main point to focus on is not the exact formulation of these conditions, but rather the paucity of behavioral evidence about stimulus meaning, when compared to the translation theories that arise from them. For Quine, this is all the data we have, or could have, that bears on the truth or falsity of theories of translation. His first major claim is that the totality of all this evidence fails to decide among radically different translation theories. Just how different these theories can be is brought out by his famous rabbit/gavagai example. We have seen that we can translate the one-word native sentence Gavagai! as the one-word English sentence Rabbit!. This might encourage one to think that we can translate 24 Grice specialized in finding cases like this and using them as counterexamples to the ordinary language philosophers. One such case, which we discussed in chapter 9, involved the ordinary truth-functional or that we are familiar with from classical logic, and the imagined ordinary language philosopher’s or, which builds into its meaning the condition that someone who utters a disjunction containing it has non-truth-functional grounds for his assertion. Since disjunctions with these two operators would exhibit essentially the same patterns of assent and dissent, their stimulus meanings would be essentially the same, even though their meanings are clearly different.

236

QUINE’S PHILOSOPHICAL NATURALISM

the English word rabbit, which occurs in many English sentences, as the native word gavagai, which, we may suppose, occurs in many sentences of the native language. However, this thought is wrong. According to Quine, to do this would be to make a claim that goes well beyond all available evidence. For, consider ‘gavagai’. Who knows but what the objects to which this term applies are not rabbits after all, but mere stages, or brief temporal segments, of rabbits? In either event the stimulus situations that prompt assent to ‘Gavagai’ would be the same as for ‘Rabbit’. Or perhaps the objects to which ‘gavagai’ applies are all and sundry undetached parts of rabbits; again the stimulus meanings would register no difference. When from the sameness of stimulus meanings of ‘Gavagai’ and ‘Rabbit’ the linguist leaps to the conclusion that a gavagai is a whole enduring rabbit, he is just taking for granted that the native is enough like us to have a brief general term for rabbits and no brief general term for rabbit stages or parts. A further alternative likewise compatible with the same old stimulus meaning is to take ‘gavagai’ as a singular term naming the fusion, in Goodman’s sense, of all rabbits: that single though discontinuous portion of the spatio-temporal world that consists of rabbits. . . . And a still further alternative in the case of ‘gavagai’ is to take it as a singular term naming a recurring universal, rabbithood. The distinction between concrete and abstract object, as well as that between general and singular term, is independent of stimulus meaning.25 Does it seem that the imagined indecision between rabbits, stages of rabbits, integral parts of rabbits, the rabbit fusion, and rabbithood must be due merely to some special fault in our formulation of stimulus meaning, and that it should be resoluble by a little supplementary pointing and questioning? Consider, then, how. Point to a rabbit and you have pointed to a stage of a rabbit, to an integral part of a rabbit, to the rabbit fusion, and to where rabbithood is manifested. Point to an integral part of a rabbit and you have pointed again to the remaining four sorts of things; and so on around. Nothing not distinguished in stimulus meaning itself is to be distinguished by pointing, unless the pointing is 25

Word and Object, pp. 51–52.

THE INDETERMINACY OF TRANSLATION

237

accompanied by questions of identity and diversity: ‘Is this the same gavagai as that?’, ‘Do we have here one gavagai or two?’.26 Here, Quine enumerates several quite different hypotheses about what the native’s word gavagai means and refers to that seem to be compatible with all the behavioral evidence we might gather about his use of the term—save, perhaps, for information we might gather by asking the native questions involving the notions of identity or divergence. Having gotten this far, Quine next takes up the questions: How are we to ask the native questions about identity? and How are we to translate, on the basis of stimulus meaning, any native expression as meaning what we mean by strict, numerical identity? Here is his response: It will perhaps be countered that there is no essential difficulty in spotting judgments of identity on the part of the jungle native, or even of a speechless animal. This is true enough for qualitative identity, better called resemblance. In an organism’s susceptibility to the conditioning of responses we have plentiful criteria for his standards of resemblance of stimulations. But what is relevant to the preceding reflections is numerical identity. Two pointings may be pointings to a numerically identical rabbit, to numerically distinct rabbit parts, and to numerically distinct rabbit stages; the inscrutability lies not in resemblance, but in the anatomy of sentences. We could equate a native expression with any of the disparate English terms ‘rabbit’, ‘rabbit stage’, ‘undetached rabbit part’, etc., and still, by compensatorily juggling the translation of numerical identity and associated particles, preserve conformity to stimulus meanings of occasion sentences.27 Quine’s contention is that the question of whether the native’s word gavagai means the same as, and hence refers to the same things as, our word rabbit—as opposed to meaning the same as, and referring to the same things as, our expressions undetached spatial part of a rabbit, temporal stage of a rabbit, or rabbit-fusion—is dependent on the question of what, if anything, in the native language means what we mean by is the very same thing as or is identical with. The problem, he believes, is that behavioral evidence about stimulus meaning doesn’t decide this question. Imagine that squiggle is a word in the native language that is a candidate for expressing the notion of identity. Pointing at different spatial 26 27

Ibid., pp. 52–53. Ibid., pp. 53–54.

238

QUINE’S PHILOSOPHICAL NATURALISM

or temporal parts of the same rabbit, we might ask Gavagai (point) squiggle gavagai (point)? with the thought that if the native assents, we will know that his word gavagai doesn’t mean what we mean by undetached spatial part of a rabbit or temporal stage of a rabbit, since the spatial or temporal parts we pointed at are different. The problem with this idea, according to Quine, is that even if the native does assent, rather than dissent, we still will not have the answer to our question; for as far as the evidence goes, the expression squiggle might be translated is identical with, is an undetached spatial part of the same extended whole as, is a temporal stage of the same enduring complex as, or various other phrases. According to Quine, the choice of a translation for the word gavagai is dependent on a choice of a translation for squiggle, and vice versa. We can, if we like, translate gavagai as rabbit and squiggle as is identical with, or we can translate gavagai as undetached spatial rabbit part and translate squiggle as is an undetached spatial part of the same extended whole as, or we can make any number of different choices, always adjusting our translations of gavagai and squiggle as a pair. Quine says that different translation theories that make different choices on these matters will be incompatible with each other, while still being equally compatible with all behavioral evidence about stimulus meaning, which, he thinks, is the totality of evidence that is relevant to theories of translation. Although it would be too much to say that he establishes this thesis, the case he makes for it does have some plausibility. At this point I would like to clarify something that has sometimes been the source of confusion. The underdetermination thesis tells us that different and incompatible theories of translation are equally supported by the set of all observational data. What does it mean to call two such theories incompatible? There is a potential problem here. It would be tempting to say that two theories that translate expressions of L1 onto expressions of L2 are incompatible if there is some expression ␣ of L1 such that the translation of ␣ given by one of the theories is an expression ␤ of L2 which means and refers to something different from the expression ␥ which is the translation of ␣ given by the other theory. However, there is a problem with this characterization: it takes for granted that there are definite facts about whether different expressions in L2 mean or refer to the same thing. Since this is something that Quine will later call into question, even when L2 is our own language, it would be desirable to have some other characterization of what it is for translation theories to be incompatible.

THE INDETERMINACY OF TRANSLATION

239

Often what Quine seems to have in mind is that the theories are logically incompatible.28 However, despite the obvious intuitive difference in meanings between ‘rabbit’, ‘undetached spatial rabbit part’, and ‘temporal stage of a rabbit’ as we use them now, the following claims are not logically incompatible: (i) The term ‘gavagai’ means the same as the term ‘rabbit’ (as we now use it). (ii) The term ‘gavagai’ means the same as the phrase ‘undetached spatial rabbit part’ (as we use now use it). (iii) The term ‘gavagai’ means the same as the phrase ‘temporal stage of a rabbit’ (as we now use it). These claims have the form  means the same as ,  means the same as ,  means the same as , which do not have the form of logically incompatible statements, any more than aRb, aRc, and aRd do. Statements (i–iii) may well be incompatible in some other way, but they are not logically incompatible. Consequently, translation theories making these different claims need not be logically incompatible with one another. However, logical incompatibility will result if these translation theories are embedded in larger background theories containing the following claims: (a) Rabbits are not undetached spatial rabbit parts, undetached spatial rabbit parts are not temporal stages of rabbits, and rabbits are not temporal stages of rabbits. (b) ‘Rabbit’ (as we use it now) refers to an object iff it is a rabbit & ‘undetached spatial rabbit part’ (as we use it now) refers to an object iff it is an undetached spatial rabbit part & ‘temporal stage of a rabbit’ (as we use it now) refers to an object iff it is a temporal stage of a rabbit. (c) If two words refer to different things, then they don’t mean the same thing.

28 See Quine, p. 179 of “On the Reasons for Indeterminacy,” and also p. 322 of “On Empirically Equivalent Systems of the World,” Erkenntnis 9 (1975): 313–28. By logical incompatibility I mean (and I assume Quine also means) the relation standardly defined (either model-theoretically or substitutionally) in logic books, as opposed to looser notions of conceptual incompatibility or analytic incompatibility.

240

QUINE’S PHILOSOPHICAL NATURALISM

(d) If a word or phrase w means the same as a word or phrase x & w means the same as a word or phrase y, then x means the same as y. Let T1 be a translation theory containing statement (i), and let T2 be a translation theory containing statement (ii). The union of T1, T2, and a set containing (a–d) is logically inconsistent. In light of this, we may take Quine’s talk about incompatible theories of translation to be talk about theories which, when augmented by background claims analogous to (a–d), are logically incompatible with one another. The justification for appealing to these auxiliary claims is that (a) states an obvious non-contentious fact about the world, and (b–d) are axiomatic to any overall theory that makes significant use of the concepts of meaning and reference. Let us sum up. Given Quine’s conception of the observational evidence relevant to translation, we have reached the conclusion that theories of translation are underdetermined by the set of all observational evidence for them. The next thing to notice is that this point is not restricted to theories that translate the native’s language onto our own. It also applies to our own attempts to understand each other. How do I know what you mean when you use the word rabbit? Naturally, I think that because you use the same word I do, you must mean the same by it as I do. However, there is no necessity in this. It may be, for all I know, that you use the word rabbit to mean what I mean by temporal rabbit stage. How can I find out which is your real meaning? Well, I have to find some basis for translating your words onto mine. But then all of Quine’s arguments about translation will come into play. My relationship to you, he will insist, is like my relationship to the native. In particular, the totality of behavioral data for translating your words into mine is equally compatible with a variety of different, incompatible systems of translation. That is not all; the same conclusions can be drawn about translations of my words, as I used them in the past, onto those same words, as I use them now. The behavioral data for all of these translations underdetermines the translations based on them.29 At this point, one might ask whether Quine’s conception of the data available for translation is accurate. My own view is that his identification of evidence for translation with claims about stimulus meaning is 29 Quine makes this argument (which we will take up in chapter 11) in “Ontological Relativity,” in Ontological Relativity and Other Essays (New York: Columbia University Press, 1969). See in particular pp. 47–48.

THE INDETERMINACY OF TRANSLATION

241

too restricted, and does not exhaust the range of relevant observational evidence. Other potentially relevant evidence for cases of real-life translation includes features of any number of different types of situations— e.g., situations in which words are introduced into the language (verbal definitions vs. ostensive illustrations), situations in which individuals acquire and become competent with a word (verbal explanations vs. paradigmatic examples), and situations in which words are spontaneously used without prompting (e.g., situations in which English speakers find it natural to use ‘rabbit’, which abound, vs. those in which they will spontaneously employ ‘temporal stage of a rabbit’, which are extremely rare). It seems to me that all of these considerations and more might properly influence the results of real-life translators. Would the addition of extra observational data of this sort change the overall result of Quine’s argument? It is not obvious that it would. To ask this question is to ask whether, by extending the set D of observational data statements beyond statements about stimulus meaning, we rule out the possibility that two incompatible theories of translation might be equally compatible with, and equally supported by, D. One problem posed by this question is deciding precisely what is meant by saying that a theory is compatible with a set of data statements D. A further issue is determining what it is for D to support T. Although Quine doesn’t go into much detail about this, his idea seems to be that D supports T when T can be used to make observational predictions shown true by D (and no predictions shown false by D).30 Although much more might be said about this, let us leave the matter of support there. As for compatibility, I have already given a Quine-friendly reconstruction of the claim that two theories are incompatible with one another. In the next section, when we move to his argument for the indeterminacy of translation, I will explore different ways of characterizing the determination relation between facts and theory. In general, for each such way, there is a corresponding sense of compatibility between theory and data that could be used here. So as not to preempt the coming discussion, for now let us simply treat the relation of compatibility between theory and data in the broadest possible way. Let us say that a translation theory T is compatible with our expanded set D of observational data statements iff the negation of T is not a necessary consequence of D—i.e., iff T could be true in a situation in which 30 For a useful discussion of how building more into the notion of support might affect Quine’s views, see Gilbert Harman, “Meaning and Theory,” Southwestern Journal of Philosophy 9 (1979): 9–20.

242

QUINE’S PHILOSOPHICAL NATURALISM

every member of D was true. Our question, then, is this: when D is understood in this way, are there incompatible theories of translation T and T⬘, each supported by D, such that (i) each could be true, given D, and hence (ii) neither T nor T⬘ is a necessary consequence of D? Well, what observation statements in addition to true statements about stimulus meanings should we take to be members of D? Recognizing that the answer to this question is to some extent up for grabs, we may here suppose that D contains additional observational truths of the sort already mentioned—namely, those describing (i) the history of uses of words in the language, including when and how they were introduced, plus how they were passed on to others, (ii) the situations in which they are commonly and spontaneously used, and (iii) all relevant observational facts about the non-linguistic environment. With this characterization of D, let T and T⬘ be two translation theories, each supported by D, that are incompatible with one another. Is it clear that at least one of them must be such that it couldn’t possibly be true, given the truth of D? I don’t think so. If, as we are supposing, D is limited to ordinary, run-of-the-mill observational statements, then, presumably, it can’t include any statements about (a) the beliefs, intentions, and other cognitive states of different language users, (b) the contents of the wishes, desires, and motivational states of such agents, (c) the contents of their perceptual experiences, (d) the causal relationships that hold between different cognitive, motivational, and perceptual states, or between those states and either the agent’s behavior or things in the environment, or (e) the internal neurological states of an agent, and their causal relations to each other, to the agent’s behavior, or to external things. Different philosophers do, of course, have different views about these matters and their relevance for facts about linguistic meaning and translation. However, it seems reasonable to think that without any factual statements of these types in our set D of data statements, there will be incompatible theories of translation each of which is capable of making correct predictions about relevant observational statements in D, even though the truth of the statements in D does not rule out the possible truth of either theory, and hence neither theory is a necessary consequence of D. For this reason, I am inclined to accept not only Quine’s claim that theories of translation are underdetermined by data about stimulus meanings, but also his implicit claim that they would remain so, even if one took a more relaxed and inclusive position about the totality of all

THE INDETERMINACY OF TRANSLATION

243

observational data relevant to translation. Although this position may initially sound rather worrisome, upon reflection, it is not obvious that it should be. For Quine, underdetermination of a theory by data amounts, roughly speaking, to the compatibility of the negation of the theory with the total set of data statements for the theory; a theory is underdetermined by the data just in case it is possible for all the data statements to be true even if the theory is not.31 In giving these characterizations, as well as in other discussions of underdetermination and indeterminacy, Quine himself is not always clear about precisely what he means by possible, compatible, and related notions. Although we will return to this shortly, let us put it aside for now. Suffice it to say that, when the relationship between observational evidence and theories is the issue, there are perfectly reasonable and natural understandings of these notions according to which not only theories of translation, but empirical theories of all sorts, are routinely underdetermined by the observational evidence bearing on them. Consequently, the claim that translation theories are also underdetermined by data is not a novel or radical idea. This is a point that Quine recognizes. What makes the doctrine of the underdetermination of translation by data striking and significant is his use of it as the basis for his truly radical doctrine of the indeterminacy of translation.

Quine’s Arguments for the Indeterminacy of Translation Recall our statement of Quine’s thesis of the indeterminacy of translation. the indeterminacy of translation Translation is not determined by the set N of all truths of nature, known and unknown. For any pair of languages and theory of translation T for those languages, there are alternative theories of translation, incompatible with T, that accord equally well with N. All such theories are equally true to the facts; there is no objective matter of fact on which they disagree, and no objective sense in which one is true and the other is not. See p. 179 of “On the Reasons for Indeterminacy,” and p. 313 of “On Empirically Equivalent Systems of the World.”

31

244

QUINE’S PHILOSOPHICAL NATURALISM

There are two main routes in Quine’s writings to the indeterminacy thesis. The first relies on Quine’s behaviorist premise that since we learn language by observing the linguistic behavior of others, the only facts relevant to determining linguistic meaning must be publicly observable behavioral facts—in particular, facts about stimulus meaning. We may take it that Quine’s discussion of the underdetermination of translation by data shows that these facts don’t determine which translations of our words are correct.32 Since only facts that determine meaning could determine translation, nothing else determines translation either. Thus, Quine is able to deduce the indeterminacy of translation from his behaviorism plus the underdetermination of translation by data. There are, however, problems with relying on behaviorism in this way. For one thing, behaviorism is highly controversial, and potentially problematic. In other domains of empirical investigation, we routinely countenance non-observational facts the existence of which is supported, but not logically or necessarily guaranteed, by the observations we make. To rule these out in the case of our theories of mind and language—in advance of establishing his indeterminacy theses— Quine would have to have a compelling independent argument that the only facts in these domains of inquiry are behavioral facts (which we may assume to be observable). Since, as far as I can see, he has no such argument, there is reason not to rest his case for the doctrine of the indeterminacy of translation on behaviorism.33 Another reason for not relying on behaviorism as a premise in the argument for the indeterminacy of translation is that Quine’s notion of stimulus meaning is problematic even from a behaviorist perspective. Stimulus meaning is, of course, defined in terms of assent and dissent. But what are assent and dissent, from a behaviorist point of view? To assent or dissent is not just to utter the English words ‘yes’ or ‘no’, for there are other ways of assenting and dissenting, and there are 32 If one is not convinced by Quine’s imaginative examples, think about the more mundane scenarios discussed in chapter 2, involving natives and islanders with different but identicalappearing rabbit-like species in both places. In scenarios like these, ‘Rabbit!’ might have the same stimulus meanings for the two groups, even though ‘rabbit’ applied to all and only members of one species when used by islanders, while applying to all and only members of a different species when used by those on the mainland. 33 For passages in Quine illustrating and explaining his commitment to behaviorism about language, see the first paragraph of the preface to Word and Object, pp. 26–29 of his essay “Ontological Relativity” and p. 81 of “Epistemology Naturalized,” both in Ontological Relativity.

THE INDETERMINACY OF TRANSLATION

245

other languages in which one can assent or dissent. One might think that one could define assent and dissent as uttering proper translations of the English words ‘yes’ and ‘no’. However, such a strategy is of no use to Quine, since, on his view, facts about what counts as a correct translation rest on prior facts about stimulus meaning, and hence on facts about assent and dissent.34 Intuitively, one thinks of assent to a sentence as expressing one’s belief that the sentence is true, and of dissent from a sentence as expressing one’s belief that the sentence is false. Of course, we also recognize that the correlations between assent and belief are not perfect, since all manner of factors may intrude in particular cases—e.g., conversational or conventional implicatures, considerations of politeness, the desire not to offend, or a desire to conceal one’s true opinion. Quine’s use of assent and dissent to characterize the evidence for theories of translation is effective only to the extent that we abstract away from these complicating factors. But when we do this, our notion of the stimulus meaning of a sentence for a speaker becomes essentially that of a pair consisting of the set of circumstances in which the speaker believes the sentence expresses a truth, and the set of circumstances in which the On pp. 29 and 30 of Word and Object, Quine discusses methods for generating “working hypotheses” about how to identify expressions of assent and dissent in the native’s language. Note that the “working hypothesis” model presupposes that the notions of assent and dissent have content that we already understand—the only question being which words express which. However, the methods don’t constitute definitions, behavioral or otherwise, and Quine does not regard them as conclusive. Hence they don’t answer the question What can assent and dissent mean for the behaviorist? and they don’t show how a thoroughgoing behaviorist can employ such notions in his theory. Quine could, of course, agree with this and maintain that what we have here is an extra indeterminacy—about what counts as assent and dissent, and hence about facts involving stimulus meaning—on top of the indeterminacy of translation given facts about stimulus meaning. However, if one goes this route, then one must admit that it is possible to make the argument for the indeterminacy of translation without invoking a general behaviorist premise that would falsify claims about stimulus meaning. Whatever the set D of genuine facts that Quine’s behaviorism really allows, he can argue that even if certain potentially controversial facts are added to D, facts about translation still will not be determined. This, I think, is the way that his argument is best reconstructed. If this is right, then, as Quine recognizes, there is no need to restrict the facts added to D to facts about stimulus meaning; he believes one can add any physical facts one likes without determining translation. Given this belief, he has no need to invoke behaviorism— which serves only to introduce needless controversy. Moreover, if even stimulus meaning goes beyond what can be acknowledged by behaviorism, then this in itself is a strong argument against it. Given the importance of the indeterminacy thesis to Quine, and the role of stimulus meaning in constructing his proposed replacement for our pretheoretic conception of meaning, he has little to gain and much to lose by insisting on behaviorism in any strict sense. 34

246

QUINE’S PHILOSOPHICAL NATURALISM

speaker believes the sentence expresses a falsehood. If we understand claims about stimulus meaning in this way, then it is reasonable to take them as providing evidence for theories of translation. But then, the evidence for such theories is not itself strictly behaviorist, but rather involves a certain kind of belief. One wonders why, if one species of belief is to be accepted and taken for granted, other beliefs shouldn’t be accorded a similar status. That, of course, is something that strict behaviorism doesn’t allow. In any case, I am not going to concentrate on Quine’s problematic behaviorism.35 Instead, I will focus on a more powerful and more widely influential route to his thesis of the indeterminacy of translation. Quine recognized that many philosophers might agree with his claim that the set D of (quasi-behavioral) facts about stimulus meanings does not resolve potential indeterminacies about meaning, while at the same time disagreeing with his contention that these are the only meaning-determining facts. To these philosophers he, in effect, issued a challenge—namely, to show how indeterminacies could be resolved by adding to D any other physical facts that one likes.36 Quine was convinced that even when the set of (quasi-behavioral) facts about stimulus meanings was enriched with all other physical facts, indeterminacies about whether a given word means the same as ‘rabbit’, ‘undetached rabbit part’, or ‘temporal stage of a rabbit’ would remain. It is understandable why he should think this. If I can’t deduce from the native’s behavior that his word ‘gavagai’ means the same as my word ‘rabbit’, how would it help to add that certain neurons in his brain fire when he uses the word? We can no more read off the contents of a person’s words from physiological claims about neurons than we can read off the contents of his words from statements about the noises he makes in certain environments. Consequently, it seems that if we cannot deduce a determinate meaning from a non-intentional description of linguistic behavior, adding facts about neurons won’t help. On Quine’s view, the same could be said about any other purely physical fact one might appeal to. We can now state his second route to the indeterminacy of translation. On this route, he derives it as a consequence of the subsidiary

35 For a further critique of Quinean arguments for the indeterminacy of translation that rely on his behaviorism, see section III of Michael Friedman, “Physicalism and the Indeterminacy of Translation,” Noûs 9 (1975): 353–73. 36 See Quine’s “Reply to Chomsky” and “On the Reasons for Indeterminacy of Translation.”

THE INDETERMINACY OF TRANSLATION

247

doctrines of physicalism and the underdetermination of translation by physics, which may be formulated roughly as follows:37 physicalism All genuine truths (facts) are determined by physical truths (facts). the underdetermination of translation by physics Translation is not determined by the set of all physical truths (facts), known and unknown. For any pair of languages and theory of translation T for those languages, there are alternative theories of translation, incompatible with T, that accord equally well with all physical truths (facts). In considering these doctrines, we need some criterion for what counts as a physical truth. The problem is not trivial, and is one to which we will return at the end of chapter 11. Until then, we will follow Quine’s lead. He often speaks of physics, or physical theory, as providing us with such truths. In general, he seems to take the physical truths to be roughly the class of all truths (known and unknown) formulated in the language of an ideal physics—something more or less like our actual physics with inconsistencies eliminated and gaps filled in.38 Given this rough and ready characterization, we can appreciate his argument for the indeterminacy of translation. If the two doctrines are correct, then claims about what our words mean—e.g., claims like ‘Rabbit’ as used by you means the same as ‘rabbit’ as used by me—are not determined by the truths of an ideal physics. Hence, these claims never state genuine facts, and never count as genuine truths. This, I believe, is Quine’s most influential and challenging argument for the indeterminacy of translation.

Evaluating Quine’s Case for Indeterminacy What should we say about this argument? The first point to make is that the contents of the premises and conclusion are unclear. In particular, it is unclear precisely what determination relation is invoked in See section IV of my “The Indeterminacy of Translation and the Inscrutability of Reference” for alternative formulations of physicalism and their relevance for Quine’s argument. 38 See, for example, Quine’s “Reply to Chomsky,” p. 303. See also his “On the Reasons for Indeterminacy of Translation.” 37

248

QUINE’S PHILOSOPHICAL NATURALISM

the doctrines of physicalism, the underdetermination of translation by physics, and the indeterminacy of translation. (This is something that, like Quine himself, I have skirted up to now.) Unfortunately, Quine never satisfactorily clarifies this. There are, however, several possibilities for spelling out what determination amounts to, and because of this, there is a danger of equivocation. In what follows, I will suggest that there are interpretations of the determination relation on which physicalism is plausible and interpretations on which the underdetermination of translation by physics is plausible, but there is no interpretation on which both physicalism and the underdetermination of translation by physics are plausible, and hence no interpretation on which the argument for the indeterminacy of translation can be sustained. Let’s look. What is it for one set of claims to determine another? It is not for the claims in the second set to be logical consequences of the claims in the first. Certainly, translation theories are not logical consequences of the set of all physical truths. But this is trivial, since whenever an empirical theory of any interest includes vocabulary not found in the truths of physics, it will fail to be a logical consequence of those truths. For example, not all the truths of biology are logical consequences of the set of true sentences of the language of an ideal physics. But biology is supposed by Quine to be determined by physics; so the determination relation cannot be that of logical consequence. If it were, physicalism would be self-evidently false. Let us examine a different conception of the determination relation. Someone might say that a set P of claims determines a set Q iff it is in principle possible, given the claims in P, for one to demonstrate the truth of the claims in Q, appealing only to logic and obvious apriori principles or definitions. The idea here is that for P to determine Q is for P to provide a theoretical basis for establishing Q that is absolutely conclusive, and that rules out any possibility of falsehood. In effect, determination is here construed as apriori consequence. On this interpretation, the underdetermination of translation by physics is both interesting and plausible. Given the total set of behavioral evidence about stimulus meanings of the sort Quine identifies, I cannot absolutely establish that what a speaker means by one of his terms is what I now mean by ‘rabbit’, as opposed to what I now mean by ‘undetached rabbit part’ or ‘temporal stage of a rabbit’. The claim that a speaker means one of these things rather than another is not an apriori consequence of the total set of claims about Quinean stimulus meanings. Moreover, it is hard to see how additional behavioral facts

THE INDETERMINACY OF TRANSLATION

249

or physical facts—about neurological events in our brains or physical interactions with our environments (past, present, and future)—would, by themselves, change the situation. Thus, it may turn out that theories of translation are underdetermined, in this sense, by the set of all physical truths. But this leads to the indeterminacy of translation only if physicalism is true, when the determination relation is taken to be apriori consequence. How plausible is that? Is there some reason to believe that all genuine truths must be purely apriori consequences of the set of all physical truths? As far as I can tell, this is not one of our pretheoretic convictions; nor has anyone given a satisfactory theoretical explanation of why it must be maintained. In addition, it is not at all clear that theories in scientific disciplines that Quine accepts as fully legitimate, and as capable of stating genuine truths about the world, are themselves apriori consequences of the set of all physical truths. In order to derive any empirical theory T from the set of truths of physics, one must appeal to theoretical identifications, or bridge principles, relating the vocabulary of T to the vocabulary of the underlying physics. If T is going to be an apriori consequence of the truths of physics, these identifications, or bridge principles, must themselves be apriori. However, the bridge principles needed by ordinary instances of theoretical reduction, such as the reduction of the biological concept of a gene to a physical construction involving the concept DNA, seem not to satisfy this condition. On the contrary, the relevant theoretical identity statement relating the notion of a gene to the concept of DNA seems to be an empirical, aposteriori truth. It is an empirical discovery that the two can be correlated in this way, not a matter of armchair conceptual analysis or philosophical speculation. Thus, if bridge principles relating the biological vocabulary to the physical vocabulary are restricted to apriori definitions, then our theoretical identity statement involving genes and DNA will be excluded, and the derivation of genetics from physical theory will be placed in jeopardy. Surely, no one would conclude from this that genetics in particular, or biology in general, fail to state genuine truths; rather, if a choice has to be made, we will reject the version of physicalism that results from interpreting the determination relation as apriori consequence. The same conclusion can be reached in a more mundane way by focusing on commonsense truths about the lives of ordinary people. Here is one such truth: I own a blue car. Could this truth be deduced from the set of truths of an ideal physics? Only if one could define

250

QUINE’S PHILOSOPHICAL NATURALISM

what it is to be me, to be a blue car, and to own something in terms of the theoretical vocabulary of an ideal physics. Needless to say, no one has the faintest idea how to do this, or any interest in it. The crucial problem here is that the required definitions or bridge principles would have to allow us to formulate conditional statements that were knowable apriori (without justification by observation and experience) even though their antecedents were physical truths and their consequents were ordinary English sentences like I exist, A car exists, I own a car, and I own a blue car, with their normal and customary meanings. It is hard enough to imagine any principles that could be used to derive ordinary talk about me, blue cars, and ownership from theoretical statements about elementary physical particles and other objects mentioned in an ideal physics. However, if such principles could be formulated, surely they would not be knowable apriori. Thus, the claim that I own a blue car is not an apriori consequence of the set of basic physical truths. Since the claim is nevertheless genuinely true, the version of physicalism that results from interpreting the determination relation as apriori consequence is clearly incorrect. In light of this, we might consider a still different—though also nonQuinean—way of understanding the determination relation.39 Suppose we say that a set P of statements determines a set Q of statements iff it would be impossible for all the statements in P to be true without all the statements in Q being true—i.e., iff Q is a (metaphysically) necessary consequence of P. On this interpretation, physicalism—in something close to the form we have stated it—is quite plausible; it states that all genuine truths (facts) supervene on the physical truths (facts). For example, if we have a set of biological facts, and if we consider a possible state of the world with respect to which some of those facts are different, the only way that can happen is if some of the underlying physical facts are also different with respect to that state of the world. Any two possible states of the world which have precisely the same physical facts are states of the world with the same biological facts. In general, there cannot be two possible states of the world which agree on all facts about every electron, proton, neutron, atom, molecule, and the like, throughout all space and time, but which differ in that in one world-state there are dogs, while in the other there are not, or 39 Quine, of course, rejects the notions of aprioricity and necessity. Alternative understandings of his theses that make use of these notions are worth considering because many of his readers do not reject them, and it is worth knowing whether they can be used successfully in the argument for the indeterminacy of translation.

THE INDETERMINACY OF TRANSLATION

251

which differ in that in one world-state I teach at Princeton, while in the other I teach at MIT, or which differ in that in one world-state my favorite color is green, while in the other it is blue. According to this interpretation of physicalism, any difference of these latter kinds would have to be accompanied by some differences at the underlying physical level. When interpreted in this way, the doctrine of physicalism says, roughly, that any two possible states of the world with the same physical facts are states of the world in which all the facts are the same—or, if that is too strong, any possible world-state in which the physical facts exactly match the physical facts in the actual state of the world is a world-state in which all facts match the actual facts. Since this seems overwhelmingly plausible, physicalism, when understood in this way, may well be acceptable. However, on this interpretation of the determination relation, the underdetermination of translation by physics becomes implausible. Whatever any of us means by rabbit, it is natural to suppose that our meaning what we do depends ultimately on the physical facts. For example, we may ask whether a physically identical twin—someone (in a physically identical possible world-state) whose utterances, behavior, brain states, causal and historical relations to the environment, and interactions with other speakers (who themselves are physically identical with speakers in the world as it actually is) completely and exactly match mine (in the world as it actually is)—could mean by ‘rabbit’ what I actually mean by, say, ‘undetached rabbit part’. It seems to me that the answer to this question must be “No”—for the very same reason that physicalism itself seems acceptable on this interpretation. This suggests that whatever meaning turns out to be, it is something which—like everything else—supervenes on the physical. But if this is right, then we have reason to reject the underdetermination of translation by physics. Might someone object to this judgment? I suppose so; however, in order for the objection to serve Quinean purposes, the objector would both have to defend that rejection and have to make a plausible argument for the truth of physicalism on the interpretation in which the determination relation is (metaphysically) necessary consequence. It is not easy to see how to do that. In order to establish the indeterminacy of translation, it is not enough for the skeptic about meaning simply to assume physicalism, deny that meaning supervenes on the physical, and demand that we prove otherwise (to his satisfaction). Rather, the objector must demonstrate, by some independent argument, either

252

QUINE’S PHILOSOPHICAL NATURALISM

that there is no such thing as meaning (in which case Quine’s whole discussion of the indeterminacy of translation becomes moot), or that there is such a thing, but it does not supervene on the physical (even though physicalism is true). Since no one has the remotest idea how to do this, this route to the indeterminacy of translation is distinctly unpromising. Better, I think, to accept physicalism, on the interpretation in which the determination relation is necessary consequence, and reject the thesis that translation is underdetermined by the physical truths. The thesis could, of course, be saved if it could be shown both that (i) theories of translation are not apriori, or conceptual, consequences of the set of physical truths, and that (ii) because of this, they are not necessary consequences of those truths, either. We have already seen that the first of these claims is quite plausible—it is plausible to suppose that theories of translation are not apriori, or conceptual, consequences of the set of physical truths. Thus, if one thought that necessity and aprioricity were the same, or at any rate coextensive, then one might wrongly conclude that both physicalism and the underdetermination of translation by physics are jointly true, when determination is taken to be necessary consequence. The error here lies in the identification of necessity and aprioricity. Quine himself accepted neither notion. However, at the time he wrote Word and Object, and for a number of years thereafter, many philosophers did accept these notions, though they typically identified the two. These philosophers were particularly vulnerable to his argument for the indeterminacy of translation. Looking at physicalism and interpreting the determination relation as necessary consequence, they correctly saw it as plausible. Looking at the underdetermination of translation by physics and viewing the determination relation as apriori consequence, they correctly saw it as plausible. Taking it for granted that necessary consequence and apriori consequence came to the same thing, they wrongly thought that Quine’s argument for the indeterminacy of translation was very powerful. The problem in their reasoning lay in the unargued, and I believe false, presupposition that statements about meaning and reference are necessary consequences of the physical truths only if they are apriori consequences of those truths. In Part 7, when we discuss Kripke’s Naming and Necessity, we will look more closely at how and why necessity and aprioricity should be distinguished. For now, the point to grasp is that if we don’t assume they are the same, then we have been given no plausible reason to accept the

THE INDETERMINACY OF TRANSLATION

253

underdetermination of translation by physics, on the interpretation in which determination is construed as necessary consequence. So far, we have not come up with an interpretation of the determination relation which supports the truth of both premises of Quine’s argument for the indeterminacy thesis. When determination is either logical consequence or apriori consequence, physicalism fails. When it is necessary consequence, it appears that the underdetermination of translation by physics fails, or at least becomes implausible and unsupported. On none of these interpretations of the determination relation does the argument for the indeterminacy of translation go through. Moreover, Quine himself would not be happy with the interpretations we have considered. Since his official view is that there are no apriori or necessary truths, he would not be content to characterize the determination relation in terms of either apriori or necessary consequence. Since he recognizes genuine cases of determination by the physical truths in which the determined truths contain vocabulary not found in the vocabulary of physics, he would not identify the determination relation with logical consequence, unsupplemented by “definitions” of non-physical vocabulary in terms of the physical vocabulary. Perhaps, then, we should think of the Quinean determination relation as involving logical consequence plus a set of definitions. In order for this to be acceptable to him, and not to collapse into apriori consequence, we must not require the definitions to express apriori truths. The proposal, then, is to characterize P as determining Q iff Q is a logical consequence of P together with bridge principles, or theoretical identifications, consisting of any truths—contingent, aposteriori, or otherwise—that relate the vocabulary of P to that of Q. This is essentially the classical concept of theoretical reduction (illustrated by Russell’s proposed reduction of arithmetic to logic discussed in chapter 6 of volume 1.)40 On this weakened conception of determination, the usual reduction of the notion of a gene to DNA will pose no difficulty for physicalism. Perhaps, then, Quine thinks both that physicalism is plausible when interpreted in this way, and that theories of translation are clearly not reducible to the set of physical truths. If so, he is wrong. Our weakened conception of determination is too weak for his purposes; according to it, even theories of translation may turn out to be trivially determined by the set of physical truths. 40 In “Physicalism and the Indeterminacy of Translation,” Michael Friedman uses familiar conceptions of reduction to characterize Quine’s theses of physicalism and the indeterminacy of translation. For discussion, see fn. 17, pp. 336–37, of my “The Indeterminacy of Translation and the Inscrutability of Reference.”

254

QUINE’S PHILOSOPHICAL NATURALISM

To see this, let Sx be some formula specifying a set of physical facts satisfied by me and only me—so that Sx is true relative to an assignment of me as value of x, and false relative to other assignments. It doesn’t matter what this formula looks like—whether it is complex, whether we can identify it as applying just to me, etc. It only matters that it exists (or could exist). For completeness, we may suppose that it includes a specification of all physical facts about me relevant to my use of language. Let Lx be a similar formula applying uniquely to a certain Spanish speaker, Luis. (For the sake of simplicity, suppose that I am a monolingual speaker of English and Luis is a monolingual speaker of Spanish.) Now imagine a claim of the following sort, exhaustively listing the translation of the finitely many individual words of my language into words and phrases of his language.41 TSL. ᭚x ᭚y [Sx & Ly & for all words w (of English) and words or phrases w* (of Spanish), w as used by x means the same as w* as used by y iff (i) w ⫽ ‘man’ and w* ⫽ ‘hombre’, or (ii) w ⫽ ‘headache’ and w* ⫽ ‘dolor de cabeza’, or . . . A corresponding claim lists the translation of the finitely many individual words of his language onto words and phrases of my language; similar claims may be imagined for each actual pair of language users— past, present, and future. Since it is plausible to suppose that there are only finitely many such pairs of speakers, it seems reasonable to assume that there exists (or could exist) some extremely long and complicated general formula of the following sort that encompasses all the individual cases. GT. For all speakers x and y, and for all words w in x’s language and words or phrases w* in y’s language, w as used by x means the same as w* as used by y iff (i) Sx & Ly & (w ⫽ ‘man’ & w* ⫽ ‘hombre’, or w ⫽ ‘headache’ & w* ⫽ ‘dolor de cabeza’, or . . .; or (ii) Lx & Sy &(w ⫽ ‘semaforo’ & w* ⫽ ‘traffic light’, or . . . ; or (iii) Sx & Gy & . . . ; or . . . GT may be regarded as a bridge principle providing a coextensive physicalistic counterpart of the predicate ‘means the same as’ applied In this discussion I ignore certain practical complications, such as the fact that some speakers speak more than one language, the fact that words of the language may be ambiguous, and the possibility that sometimes there may be no translation of a word in one language onto a word or phrase of the other language. Although these are real factors in translation, they are peripheral to Quine’s philosophical claims about translation.

41

THE INDETERMINACY OF TRANSLATION

255

to words and phrases in theories of translation. Next, we need a general bridge principle that encompasses the rules used in theories of translation to combine translations of parts into translations of whole sentences. Presumably, there are only finitely many such combinatorial rules for each pair of speakers. If, as we are assuming, there are only finitely many such pairs of actual speakers, these rules can, in principle, be exhaustively listed. This list, together with GT, can then be used to formulate another bridge principle that provides a coextensive physicalistic counterpart to the predicate ‘means the same as’ as used between sentences in theories of translation. But then, theories of translation will be derivable from the set of all physical truths together with these bridge principles, and so will count as determined by physics in our new weakened sense. Might Quine object to these derivations on the grounds that the bridge principles TSL and GT are not true? He certainly might, but on what grounds? As a matter of actual fact, the translational pairings between the expressions of English and Spanish that they encode are regarded as accurate by speakers who know both languages. Does Quine have a principled objection to these standardly accepted views? If so, what is it? He can’t appeal to the indeterminacy of translation to ground any such objection, since without an independent demonstration that these bridge principles are false, he has no argument for the indeterminacy doctrine in the first place. Thus, even on this friendly interpretation of the determination relation, Quine has not made his point. The purpose of this exercise has been to show that the interpretation of determination according to which P determines Q iff there are some truths T which together with P logically entail Q is too weak to support Quine’s indeterminacy thesis. Could stronger conditions be found that would serve his purposes? Several possibilities suggest themselves. First, it might be claimed that what we want is not just a reduction of theories of translation to physical truths, but rather a single reduction to the set of physical truths of all theories making use of semantic notions such as meaning and reference. Surely, if these notions are legitimate, they will have significant theoretical uses well beyond theories of translation in the narrow sense considered here. The fact that a trivial reduction is theoretically possible when translation theories are considered in isolation does not show that such a reduction is possible in a context that is properly more inclusive. The point is well taken. But the problem with this suggestion is that it takes us far beyond Quine’s own discussion and into uncharted waters.

256

QUINE’S PHILOSOPHICAL NATURALISM

Before we can make any progress on the question of whether a single physicalistic reduction of all legitimate uses of semantic notions is theoretically possible, we need a reasonably precise and exhaustive characterization of the range of theoretical uses of semantic notions. Quine has not attempted to provide such a characterization; until we have one, we can’t evaluate the case for skepticism about meaning (and reference) because no sufficiently articulated case for skepticism about these notions has been made. A second possible response to my trivial reduction is to stick to a physicalistic reduction simply of translation theories, but to add the restriction that the reduction must provide an explanation of the meaning facts reported by those theories. Whereas my trivial reduction doesn’t begin to do this, classical examples of theoretical reduction—such as the reduction of the biologist’s notion of a gene to facts about DNA—do. Part of Quine’s point may well have been that we shouldn’t expect the same sort of reductive physicalist explanation of purported facts about meaning and translation. That is certainly a reasonable thought. However, it is not enough to support his overall argumentative purposes. For one thing, we don’t have a clear, explicit, and general conception of what constitutes an explanation. Although we are pretty good at recognizing particular explanations when we see them, we don’t have the sort of precise and persuasive definition, or general theory, of what constitutes an explanation in all the contexts that would be needed if we were to appeal to the notion of explanation in defining the determination relation used in physicalism, the underdetermination theses, and the indeterminacy thesis. Suppose, for example, we were to impose the requirement that a set P of truths is to count as determining a set Q only if the reduction of Q to P explains the truths in Q. What, on this interpretation, should we then think of physicalism, and the thesis of the underdetermination of translation by physics? Here it is helpful to contrast facts about translation with domains of fact that no one wants to renounce—e.g., the facts of economics, or of common sense (such as the fact that I own a blue car). Is there any prospect of providing a genuine explanatory reduction of these facts to the facts of fundamental physics? Certainly, none is in the offing. Perhaps such an explanation is possible in principle, but it is hard to be sure. We may grant that the economic facts (and the commonsense facts) are necessary consequences of, and hence supervene on, the physical facts. Still, we have no way of ascertaining whether an explanatory reduction of the former to the latter is possible without knowing a great deal

THE INDETERMINACY OF TRANSLATION

257

more than we now do both about the relationship between the two sets of facts, and about what counts as a genuine explanation. It is sheer fantasy, at this stage, to speculate that economics (and commonsense facts about ownership of blue cars) will prove to be explainable in terms of purely physical facts, whereas facts about translation won’t. And if none of them proves to be explainable in this way, I certainly won’t give up my belief in economics, or my claim to my beloved blue Volvo. Why should facts about meaning and translation be treated differently? After all, if they are necessary consequences of the physical facts, then they too are facts, whether or not we are capable, even in principle, of giving explanatory physicalist reductions of them. Thus, although defining the determination relation in terms of explanation would save the thesis of the underdetermination of translation by physics from falsification by my trivial, non-explanatory reduction, it would not provide us with an interpretation of Quine’s central theses on which we have been given a good reason to accept them. A third possible response to my trivial reduction is to strengthen the determination relation by requiring the bridge principles used in deriving the set of claims Q from the physical truths P to provide definitions that link the terms in Q with physical formulas coextensive with them in all (metaphysically possible) counterfactual circumstances. The bridge principles in my trivial reduction do not satisfy this demand, and so do not satisfy this strengthened modal constraint on the determination relation, which is tantamount to the requirement that semantic predicates be necessarily coextensive with certain physical formulas. Although I have no quarrel with this strengthened requirement, two points should be noted. First, if, unlike Quine, we grant the legitimacy of modal notions, then we can characterize the determination relation directly in terms of necessary consequence. But then, as I have maintained above, we have reason to believe that theories of translation are determined by the physical truths. Second, if one insists on characterizing determination in terms of a strengthened reduction relation that requires physicalistic formulas that are necessarily coextensive with the predicates used in the theories undergoing reduction, then it is not clear what our attitude should be toward the resulting strengthened versions of physicalism and the underdetermination of translation by physics. Are there physicalistic formulas that are necessarily coextensive with the relevant semantic predicates employed by a theory of translation? It is not clear. Certainly Quine has given no compelling arguments to think that there aren’t. Thus, on this interpretation, the case

258

QUINE’S PHILOSOPHICAL NATURALISM

for the underdetermination of translation by physics has not been made. A similar point holds for physicalism. On the present strengthened interpretation of the determination relation, it is not evident what we should think of this doctrine. I have already granted the truth of a weaker version of physicalism, which states that all truths must be necessary consequences of the physical truths. In light of this, it is not obvious that we should add the further requirement that they be logical consequences of the physical truths plus definitions that provide necessarily coextensive physical translations of all vocabulary items used in stating any truth. Suppose certain claims are necessary consequences of the set of all physical truths. It is hard to see why we should deny that these claims state genuine facts if it turns out, for example, that there are no finitely long formulas in the language of an ideal physics that are necessarily coextensive with the relevant non-physical predicates. Since this stronger requirement is both questionable and unsupported by argument, the present interpretation of the determination relation is one in which neither of the two premises for the indeterminacy thesis can be regarded as secure. In light of this, the conclusion to be drawn is that Quine’s argument for the indeterminacy of translation fails to provide a compelling challenge to our fundamental convictions about meaning and translation. This does not, by itself, foreclose the possibility that something could be done to strengthen, or revive, his skeptical challenge. But surely, the burden of proof is on those who wish to persuade us to adopt a radically skeptical attitude toward our most basic, and seemingly indispensable, linguistic beliefs.

CHAPTER 11 QUINE’S RADICAL SEMANTIC ELIMINATIVISM chapter outline 1. Eliminativism and the Inscrutability of Reference Formulation and derivation of the inscrutability doctrine; an argument that no word refers to anything 2. Quine’s Own Eliminativist Argument Quine’s argument in “Ontological Relativity” for eliminativism about ordinary reference 3. Ersatz Reference: Quine’s Proposed Replacements for Our Ordinary Notion of Reference Disquotational reference in one’s own present language; translation plus disquotation for the reference of others 4. Unbridled Eliminativism: Belief, Assertion, Truth, and Other “Intentional Idioms” Quine’s eliminativism about belief, assertion, and other propositional attitudes in Word and Object; extension to our ordinary notion of truth and replacement of this notion truth with Tarski-truth in one’s own present language 5. Evaluating Quine’s Position: the Argument That it is SelfUndermining Quine’s dilemma: either his central theses state something about the totality of physical truths in the ordinary sense, in which case they are inconsistent with the radical eliminativism to which he is driven, or else they state something only about the totality of Tarski-truths in his present language, in which case they do not have their intended philosophical import 6. Quine, the Ordinary Language School, and the Fate of Meaning in Philosophy

In the last chapter, I maintained that Quine’s argument fails to establish his doctrine of the indeterminacy of translation. In this chapter, we will turn to the question of whether that doctrine is true. I will argue that it is not. My strategy will be first, to sketch a companion view

260

QUINE’S PHILOSOPHICAL NATURALISM

about reference that Quine accepts for essentially the same reasons he accepts the indeterminacy of translation, and second, to draw out consequences of the two views and the premises he uses to motivate them— consequences so radical as to be not only extremely implausible, but also self-undermining. If the strategy is successful, then we may safely conclude that the combination of Quine’s views about meaning and reference is unacceptable. Once this becomes clear, it will, I think, become plausible that both his doctrine of the indeterminacy of translation and his related doctrine asserting the indeterminacy of reference must go. Such, at any rate, will be my contention.1

Eliminativism and the Inscrutability of Reference We begin with a seemingly obvious consequence of Quine’s position on the indeterminacy of translation. It follows from his position that the claim that the native uses the term gavagai to mean the same thing as I mean by the word rabbit is not determined to be true by the set of all physical truths. More generally, it follows that for every expression ␣ in my language, the claim that the native uses the word gavagai to mean the same thing as I mean by ␣ is not determined to be true by the set of all physical truths. But then, since none of these claims is determined to be true by the physical truths, Quine’s doctrine of physicalism tells us that none of these claims expresses a genuine truth at all. Thus, from Quine’s position on indeterminacy plus his commitment to physicalism, we get C1. C1. No claim that the native uses ‘gavagai’ to mean the same as I mean by ␣ is true (where ␣ is any expression in my language). The same reasoning can be repeated to show that, on Quine’s view, the claim that the native uses the term gavagai to refer to what I refer to with the word rabbit is not determined to be true by the physical truths. In fact, as Quine himself recognizes, there is no need here to mention rather than use my word rabbit. On his view, the claim that the native uses the term gavagai to refer to rabbits is not determined to be true by physics. Moreover, it follows that for every expression ␣ of my language, the sentence The native uses ‘gavagai’ to refer to ␣ is not The material in this chapter expands and elaborates ideas originally presented in sections II and III of my “The Indeterminacy of Translation and the Inscrutability of Reference.”

1

RADICAL SEMANTIC ELIMINATIVISM

261

determined to be true by physics. But then, since none of these claims is determined to be true by physics, the doctrine of physicalism tells us that none of these sentences express truths. This gives us C2, which is an instance of a Quinean doctrine called the inscrutability of reference. C2. No sentence The native uses ‘gavagai’ to refer to ␣ expresses a truth. Quine’s understanding of and argument for the inscrutability of reference parallel his understanding of and argument for the indeterminacy of translation. Here is a passage from his article “Ontological Relativity” that gives the flavor of his view. It is philosophically interesting, moreover, that what is indeterminate in this artificial example [the gavagai case] is not just meaning, but extension; reference. My remarks on indeterminacy began as a challenge to likeness of meaning. I had us imagining “an expression that could be translated into English equally defensibly in either of two ways, unlike in meaning in English.” Certainly likeness of meaning is a dim notion, repeatedly challenged. Of two predicates which are alike in extension, it has never been clear when to say that they are alike in meaning and when not; it is the old matter of featherless bipeds and rational animals, or of equiangular and equilateral triangles. Reference, extension, has been the firm thing; meaning, intension, the infirm. The indeterminacy of translation now confronting us, however, cuts across extension and intension alike. The terms “rabbit,” “undetached rabbit part,” and “rabbit stage” differ not only in meaning; they are true of different things. Reference itself proves behaviorally inscrutable.2 There are two points to note here. First, Quine’s original argument involving rabbit and gavagai for the indeterminacy of translation can easily be recast as an argument for the following thesis. the indeterminacy of referential sameness Sameness of reference is not determined by the set N of all truths of nature, known and unknown. For any pair of languages and theory T of sameness of reference relating Pages 34–35 of “Ontological Relativity,” in Ontological Relativity and Other Essays (New York: Columbia University Press, 1969). The article was originally given as a lecture at Columbia University in 1968.

2

262

QUINE’S PHILOSOPHICAL NATURALISM

expressions of those languages, there are alternative theories of referential sameness, incompatible with T, that accord equally well with N. All such theories are equally true to the facts; there is no objective matter of fact on which they disagree, and no objective sense in which one is true and the other is not. Just as Quine’s argument for the indeterminacy of translation rests on the theses of physicalism and the underdetermination of translation by physics, so his argument for the indeterminacy of referential sameness rests on physicalism and thesis of the underdetermination of referential sameness by physics. the underdetermination of referential sameness by physics Referential sameness is not determined by the set of all physical truths (facts), known and unknown. For any pair of languages and theory T of referential sameness for those languages, there are alternative theories of referential sameness, incompatible with T, that accord equally well with all physical truths (facts). Clearly, the arguments for the indeterminacy of translation and the indeterminacy of referential sameness stand or fall together. Second, in the passage just cited, Quine does not content himself with asserting the indeterminacy (or inscrutability) of referential sameness, but rather speaks directly of the indeterminacy (inscrutability) of reference itself—which is characteristic of most of his discussion in the article. This suggests the following formulation of his doctrine of the inscrutability of reference. (When, in formulating the doctrine, I speak of theories of reference assignment for a language, I have in mind theories which ascribe referents to the expressions of the language, and which issue in claims of the same general sort as the following: The term ‘gavagai’ in the language of the native refers to an object o iff o is a rabbit.) the inscrutability of reference Reference assignment is not determined by the set N of all truths of nature, known and unknown. For any language L and theory T of reference assignment to expressions of L, there are alternative theories of reference assignment, incompatible with T, that accord equally well with N. All such theories are equally true to the facts; there is no objective matter of fact on which they disagree, and no objective sense in which one is true and the other is not.

RADICAL SEMANTIC ELIMINATIVISM

263

The inscrutability of reference should be seen as following from the indeterminacy of referential sameness plus the natural assumptions (i) that if reference is determinate (scrutable) for one language, then it is determinate (scrutable) for all, and (ii) that if the physical truths determine that an expression ␣ in L refers to an object o iff o is A and that an expression ␤ in L⬘ refers to an object o iff o is B, then it is determinate whether all and only A’s are B’s, and hence whether ␣ and ␤ refer to precisely the same things.3 Assumption (i) reflects the fact that there is nothing special about one language—e.g., English—that makes the reference of its expressions something that is determined by the totality of the physical truths, unless the reference of the expressions of other languages is so determined as well. Assumption (ii) expresses the idea that if reference is determinate, then the identities of the objects referred to are determinate as well. Whereas (i) is obviously correct, (ii) might be doubted. Since, however, Quine shows no sign of even noticing the assumption, let alone doubting it, and since he simply assumes that his argument for the indeterminacy of translation carries over without difficulty to the inscrutability of reference, we will follow his lead and here take it for granted as well. Given this assumption, one can deny the inscrutability of reference only if one also denies the indeterminacy of referential sameness. Since Quine thinks he has a good argument for the latter, he feels free to adopt the former. With his commitment to physicalism, the indeterminacy of translation, and the inscrutability of reference firmly in place, we now return to the task of drawing out the consequences of these claims. We have already derived C1 and C2. C1. No claim that the native uses ‘gavagai’ to mean the same as I mean by ␣ is true (where ␣ is any expression in my language). C2. No sentence The native uses ‘gavagai’ to refer to
expresses a truth. Reasoning analogous to that which gave us C2 gives us C3. C3. There is no object o and variable v such that The native uses ‘gavagai’ to refer to v is true relative to an assignment of o to v. 3 (ii) is not really an assumption, but an assumption schema with schematic letters ‘A’ and ‘B’ that stand in for arbitrary predicates. The schema stands for the class of assumptions that are its instances.

264

QUINE’S PHILOSOPHICAL NATURALISM

Having gotten this far, we next appeal to Quine’s standard semantics for the logical operators to reason as follows: C4. Since v ⌽ is true iff there is an object o such that ⌽ is true relative to an assignment of o to v, x (the native uses ‘gavagai’ to refer to x) is not true. (From C3) C5. Since ~S is true iff S is not true (provided that S is meaningful), ~x (the native uses ‘gavagai’ to refer to x) is true (provided that it is meaningful, which it surely is). (From C4) C6. Since we can infer S from ‘S’ is true, we may correctly assert: ~x (the native uses ‘gavagai’ to refer to x)—i.e., we may correctly assert that the native doesn’t use ‘gavagai’ to refer to anything. (From C5) C7. Since there is nothing special about the native, or the word gavagai, we must conclude that no one ever uses a word to refer to anything. (From C6) To my knowledge, Quine never explicitly says anything quite as radical and unequivocal as this. However, given our formulations of physicalism, the indeterminacy of translation, and the inscrutability of reference, plus the usual Quinean accounts of the logical operators, one is hard pressed to see how he can avoid this conclusion. Thus, it would seem that his position leads to the view that there is no such thing as reference; i.e., no one ever refers to anything. If this is right, then he is an eliminativist not only about meaning, in the sense in which we normally understand it; he is also an eliminativist about reference.

Quine’s Own Eliminativist Argument Although Quine doesn’t explicitly consider the argument just given, he does give a related argument in “Ontological Relativity.” There he considers whether his doctrine of the inscrutability of reference applies to ourselves, and whether, if it does, this makes “nonsense of the notion of reference.” Prior to this point in the essay, Quine has pointed out that the indeterminacy of translation is inextricably linked with the inscrutability of reference. He has argued that if it is indeterminate whether the native’s word gavagai means the same as my word rabbit,

RADICAL SEMANTIC ELIMINATIVISM

265

then it is indeterminate whether the native’s word gavagai refers to rabbits. He has also pointed out that if it is indeterminate whether the native’s word gavagai refers to rabbits, then it is indeterminate whether my neighbor uses the word rabbit to refer to rabbits. At this point Quine considers the possibility of making the same claim about himself. I have urged in defense of the behavioral philosophy of language, Dewey’s, that the inscrutability of reference is not the inscrutability of a fact; there is no fact of the matter. But if there is really no fact of the matter, then the inscrutability of reference can be brought even closer to home than the neighbor’s case; we can apply it to ourselves. If it is to make sense to say even of oneself that one is referring to rabbits and formulas and not to rabbit stages and Gödel numbers, then it should make sense equally to say it of someone else. After all, as Dewey stressed, there is no private language. We seem to be maneuvering ourselves into the absurd position that there is no difference on any terms, interlinguistic or intralinguistic, objective or subjective, between referring to rabbits and referring to rabbit parts or stages; or between referring to formulas and referring to their Godel numbers. Surely this is absurd, for it would imply that there is no difference between the rabbit and each of its parts or stages, and no difference between a formula and its Gödel number. Reference would seem now to become nonsense not just in radical translation but at home.4 The argument contained in this passage may be reconstructed as follows:

Quine’s Reductio in “Ontological Relativity”

R1. It is indeterminate (hence there is no fact of the matter) whether the native uses ‘gavagai’ to refer to rabbits, temporal stages of rabbits, undetached rabbit parts, etc. R2. If R1, then it is indeterminate (hence there is no fact of the matter) whether my neighbor uses ‘rabbit’ to refer to rabbits, temporal stages of rabbits, undetached rabbit parts, etc.

4

“Ontological Relativity,” pp. 47–48.

266

QUINE’S PHILOSOPHICAL NATURALISM

R3. Therefore, it is indeterminate (hence there is no fact of the matter) whether my neighbor uses ‘rabbit’ to refer to rabbits, temporal stages of rabbits, undetached rabbit parts, etc. R4. If R3, then it is indeterminate (hence there is no fact of the matter) whether I use ‘rabbit’ to refer to rabbits, temporal stages of rabbits, undetached rabbit parts, etc. R5. Therefore it is indeterminate (hence there is no fact of the matter) whether I use ‘rabbit’ to refer to rabbits, temporal stages of rabbits, undetached rabbit parts, etc. More generally (combining this result with R1 and R3, and generalizing), it is always indeterminate whether a word (as used by anyone) refers to all and only rabbits, as opposed to referring to all and only temporal stages of rabbits, or to all and only undetached rabbit parts, etc. R6. If it is always indeterminate whether a word refers to all and only rabbits, as opposed to all and only temporal stages of a rabbit, or to all and only undetached rabbit parts, then there is no difference between referring to all and only rabbits and referring to all and only temporal stages of rabbits, or to all and only undetached rabbit parts—i.e., it is true that a word w refers to all and only rabbits only if it is equally true that w refers to all and only temporal stages of rabbits, and to all and only undetached rabbit parts. R7. If there is no difference between referring to all and only rabbits and referring to all and only temporal stages of rabbits, or to all and only undetached rabbit parts—i.e., if it is true that a word w refers to all and only rabbits only if it is equally true that w refers to all and only temporal stages of rabbits, and to all and only undetached rabbit parts—then there is no difference between rabbits, temporal stages of rabbits, undetached rabbit parts—i.e., something is a rabbit iff it is a temporal stage of a rabbit iff it is an undetached rabbit part. R8. Therefore, there is no difference between rabbits, temporal stages of rabbits, undetached rabbit parts (and so on). Since the conclusion, R8, is clearly false, and the argument is logically valid, at least one of the premises R1, R2, R4, R6, or R7 must be incorrect. The problem is to determine which.

RADICAL SEMANTIC ELIMINATIVISM

267

In attacking this problem it is important to be clear about the ontological nature of Quine’s indeterminacy and inscrutability theses. They claim that the set of all physical truths does not determine the truth of statements about meaning, translation, and reference. Now certainly the set of all physical truths does not distinguish in this respect between the native, my neighbor, and me. If physical truths determine reference for any of us, then physical truths determine reference for all of us. So, given the nature of the indeterminacy and inscrutability theses, one cannot avoid the argument’s conclusion by rejecting premises R2 or R4. If we claim that there is indeterminacy regarding the native, then we must accept the same indeterminacy for ourselves. In short, the only way to block the absurd conclusion R8 is to reject either R1, R6, or R7. Since R1 is essential to Quine’s position, his only option is to reject R6 or R7. What, then, from Quine’s point of view, is wrong with these assumptions? Recall the passage from “Ontological Relativity.” The first sentence of the final paragraph of the passage may be taken as a comment on R6: We seem to be maneuvering ourselves into the absurd position that there is no difference on any terms, interlinguistic or intralinguistic, objective or subjective, between referring to rabbits and referring to rabbit parts or stages; or between referring to formulas and referring to their Gödel numbers. Here, Quine indicates that the previous steps in the argument—R1 through R5—seem to leave us no choice but to accept the “absurd” consequent of R6. The next sentence in the passage is an expression of R7, which Quine uses to bring out this “absurdity.” Surely this is absurd, for it would imply that there is no difference between the rabbit and each of its parts or stages, and no difference between a formula and its Godel number. Finally, Quine summarizes the import of the argument in the last sentence of the passage. Reference would seem now to become nonsense not just in radical translation but at home. At this point we face an exegetical problem. The argument from R1 to R8 is a reductio ad absurdum. Quine has started with what he believes to be a truth, and proceeded by apparently undeniable steps to reach a conclusion that he clearly recognizes to be false. When a philosopher

268

QUINE’S PHILOSOPHICAL NATURALISM

does this he has an obligation to tell us which step or steps in the argument should be rejected, and to try to explain why. Quine does not do this. There is no doubt that he rejects R8. It is also clear that he accepts R1–R5, and that his only option is to reject either R6 or R7. However, he doesn’t tell us which premise is mistaken, nor does he explain how rejecting one of them would lead us to conclude that reference is “nonsense.” He does say, in the two paragraphs immediately following the quoted passage, that reference really is nonsense if considered “absolutely,” but it is not nonsense if thought of as relative to a background theory. This is far from transparent. However, there is a way of making sense of it. Consider again R6 and R7. R6. If it is always indeterminate whether a word refers to all and only rabbits, as opposed to all and only temporal stages of a rabbit, or to all and only undetached rabbit parts, then there is no difference between referring to all and only rabbits and referring to all and only temporal stages of rabbits, or to all and only undetached rabbit parts—i.e., it is true that a word w refers to all and only rabbits only if it is equally true that w refers to all and only temporal stages of rabbits, and to all and only undetached rabbit parts. R7. If there is no difference between referring to all and only rabbits and referring to all and only temporal stages of rabbits, or to all and only undetached rabbit parts—i.e., if it is true that a word w refers to all and only rabbits only if it is equally true that w refers to all and only temporal stages of rabbits, and to all and only undetached rabbit parts—then there is no difference between rabbits, temporal stages of rabbits, undetached rabbit parts—i.e., something is a rabbit iff it is a temporal stage of a rabbit iff it is an undetached rabbit part. Quine is committed to the antecedent of R6 by virtue of his acceptance of R5. What about the consequent of R6? Given his views about physicalism, indeterminacy, and inscrutability, he really can’t reject it. For Quine, all truths are determined by the physical truths. But since the claim that a word refers to rabbits bears the same relation to the physical truths as the claim that the word refers to temporal stages of rabbits, or to undetached rabbit parts, the only way that one of these reference claims could be true is if the others were true as well. Thus,

RADICAL SEMANTIC ELIMINATIVISM

269

Quine is committed not only to the antecedent of R6, but also to its consequent, and hence to R6 itself. If this is right, then his only hope of avoiding a reductio ad absurdum of his own position is to find a way of rejecting R7. Note that since the consequent of R7 is false, R7 will be false iff its antecedent is true. So now we have to ask ourselves, how could the antecedent of R7 be true? The antecedent of R7 is itself a conditional. antecedent of R7 It is true that a word w refers to all and only rabbits only if it is equally true that w refers to all and only temporal stages of a rabbit and to all and only undetached rabbit parts—i.e., if it is true that w refers to all and only rabbits, then it is equally true that w refers to all and only temporal stages of a rabbit, and to all and only undetached rabbit parts. We are trying to make the antecedent of R7 true, so we can reject R7 as a whole and block the derivation of the absurd conclusion R8. In principle, there are two ways that the antecedent of R7 could be true: (i) it could have a false antecedent, or (ii) it could have a true consequent. However, we know that the consequent can’t be true. For suppose it were. Then w would refer to all and only temporal stages of a rabbit, and it would also refer to all and only undetached (spatial) rabbit parts. With this in mind, let RP be some undetached rabbit part— say, a rabbit foot. Then w refers to RP. But since we know that w refers to all and only temporal stages of a rabbit, it follows that the enduring rabbit foot RP must be a temporal stage of a rabbit. But RP isn’t a temporal stage of a rabbit. Thus, the consequent of the antecedent of R7 can’t be true. It can’t be true because if it were true, then there would be no difference between undetached (spatial) rabbit parts and temporal stages of rabbits—every (spatial) rabbit part would be a temporal stage, and every temporal stage would be a (spatial) rabbit part. Since that is absurd, the consequent of the antecedent of R7 is false. This means that the antecedent of R7 can be true, and R7 as a whole can be false, only if the antecedent of the antecedent of R7 is false—i.e., only if it is false that w refers to all and only rabbits. In other words, Quine can block the derivation of the absurd conclusion R8 only by denying that there is any word that refers to all and only rabbits. So that is what he must do. He must deny that any word refers to rabbits (and only rabbits). Since we could repeat the same argument for any other class of things, Quine must deny that any word ever

270

QUINE’S PHILOSOPHICAL NATURALISM

refers to anything. Thus, what the argument shows is that Quine must be an eliminativist about reference. This, I believe, is the most accurate and straightforward interpretation of his position.5 It also fits nicely with the positive remarks he makes immediately following the quoted passage in which he sets out the reductio ad absurdum. As I have indicated, the lesson he explicitly draws from the argument is that “reference would seem now to become nonsense not just in radical translation but at home.” In the paragraphs immediately following this remark, he tries to explain what this means by propounding a doctrine of the relativity of reference and ontology. The key feature of the doctrine is supposed to be that reference is nonsense, if thought of “absolutely,” but it is not nonsense, if understood as relativized to some sort of background theory or language. One problem with this is that Quine never makes clear precisely what it means to think of reference “absolutely.” But whatever it means, he seems to be saying that if we understand reference in that way, then there is no such thing as reference—no expression ever refers (in that sense) to an object. In other words, he claims that the way to block the obviously false conclusion R8 is to embrace eliminativism about a certain notion of reference. However, this strategy will succeed only if the notion of reference about which one is an eliminativist is the one that appears in the earlier steps R1–R5 of the argument. Since the notion that appears there is reference as we ordinarily understand it, Quine is an eliminativist about our ordinary notion of reference.

Ersatz Reference What notion of reference does Quine propose as an alternative to the notion of “absolute,” i.e., ordinary, reference that he rejects? Although his discussion of this in “Ontological Relativity” is sketchy and somewhat opaque, he filled in some of the details in later work. As a result, it is 5 It is possible to get the same result by a slightly different interpretive route. On the above interpretation, R6 is true because its consequent—it is true that a word w refers to all and only rabbits only if it is equally true that w refers to all and only temporal stages of rabbits, and to all and only undetached rabbit parts—is vacuously true; R7 is false because its antecedent, which is the same as the consequent of R6, is vacuously true and its consequent is false. A slightly different rendering of the argument could be given by building into the consequent of R6 (and antecedent of R7) the existential claim that some expression does refer to all and only rabbits. On this reading, R6 turns out to be false, on Quine’s view, while R7 comes out true. A point in favor of this reading is that it fits Quine’s comment that the consequent of R6 is “absurd.”

RADICAL SEMANTIC ELIMINATIVISM

271

possible to give a reasonably clear and coherent account of his position. According to his positive account of reference, claims I make about the reference of others are relative to some background interpretation or translation manual. Roughly speaking, when I say, using Quine’s proposed substitute for our ordinary notion of reference, that x’s expression ␣ refers to F’s, I am saying that according to some underdetermined system T of translation that I am employing, x’s expression ␣ refers to F’s. Moreover, when I claim that x’s expression ␣ refers to F’s relative to a translation manual T, I am saying that there is some word or phrase p in my present language that satisfies two conditions: (i) according to T, ␣ means the same thing as p (T maps ␣ onto p), and (ii) p, as I now use it, refers to F’s.6 This invites a further question. What does it mean to say that an expression p in my present language refers to F’s—i.e., that p, as I now use it, refers to F’s? The answer is that by reference in my present language I mean, roughly, that which is given by the sum of all relevant disquotational axioms: ‘N’, as used by me now, refers to an object o iff o is (an) N. Examples of such axioms are: ‘Rabbit’ as used by me now refers to an object o iff o is a rabbit. ‘Dog’ as used by me now refers to an object o iff o is a dog. ‘Willard Van Orman Quine’ as used by me now refers to an object o iff o is Willard Van Orman Quine. In defining reference for my own present language, I use a disquotational axiom like this for each referential word in my language. For compound referring phrases, there are rules that give the reference of the compound expression in terms of the reference of the parts. Taken as a whole, these disquotational axioms and rules define what I mean by saying that a word or phrase of my language refers to an object or objects. So, on this view there are two kinds of reference: (i) disquotational reference for words and phrases of my own present language, and (ii) translation plus disquotational reference for all words and phrases other than those used by me now. Quine appears to explicitly endorse this view in Pursuit of Truth, where he clarifies and modifies his original discussion in “Ontological Relativity.” I can now say what ontological relativity is relative to, more succinctly than I did in the lectures, paper, and book of that title. It 6

Here, ‘p’ and ‘␣’ are metalinguistic variables, and ‘F’ is a schematic letter.

272

QUINE’S PHILOSOPHICAL NATURALISM

is relative to a manual of translation. To say that ‘gavagai’ denotes rabbits is to opt for a manual of translation in which ‘gavagai’ is translated as ‘rabbit’, instead of opting for any of the alternative manuals. . . . And does the indeterminacy or relativity also extend somehow to the home language? In “Ontological Relativity” I said it did, for the home language can be translated into itself by permutations that depart materially from the mere identity transformation. . . . But if we choose as our manual of translation the identity transformation, thus taking the home language at face value, the relativity is resolved. Reference is then explicated in disquotational paradigms analogous to Tarski’s truth paradigm (section 33); thus ‘rabbit’ denotes rabbits, whatever they are, and ‘Boston’ designates Boston.7 The best way to understand Quine’s view is, in my opinion, to take it as a proposal for replacing our ordinary notion of reference with two related notions—disquotational reference for one’s own present language, and translation plus disquotation for everything else.8 Technicalities aside, one can think of the disquotational reference of one’s present words as being defined in the following Tarski-like way:9 W. V. Quine, The Pursuit of Truth (Cambridge, MA: Harvard University Press, 1992), pp. 51–52. 8 The alternative—to take Quine to be using his two notions of reference to give an account of what we ordinarily mean by ‘refers’—is to attribute to him what, from an ordinary point of view, is an obvious falsehood. When I say that someone else’s expression refers to rabbits, I am not saying anything about myself or my present language. This is shown by the fact that (i) what I say could be true in a counterfactual situation in which I didn’t exist, or don’t speak a language at all, and (ii) a person could believe the proposition I express using ␣, as used by Maria, refers to rabbits without believing anything about me or my language, as evidenced by the fact that my remark John believes that , as used by Maria, refers to rabbits might be true even though John has no beliefs about me or my language. Even putting aside counterfactual situations, as well as considerations involving propositional attitude ascriptions, my remark that someone’s expression ␣ refers (in the ordinary sense) to rabbits might be true (in the actual context of utterance) in a case in which there is no expression in my language that means the same as ␣, so long as ‘rabbit’ in my language is coextensive with ␣. Conceivably, a skeptic like Quine might not be moved by these considerations, since from his point of view there may not seem to be a significant difference between analyzing what we ordinarily mean by ‘refers’ and providing a replacement for our ordinary notion of reference. However, there is no reason for those of us who are trying to understand and evaluate his view to take that position. (Later, more will be said about the difference between disquotational reference for my own case and claims I make about my own reference using the ordinary notion of reference.) 9 The idea of replacing our ordinary notion of reference with a trivial, disquotationally defined notion of reference is due to Alfred Tarski. For an explanation of this idea and Tarski’s use of it, see pp. 67–81 of chapter 3 of my Understanding Truth. 7

RADICAL SEMANTIC ELIMINATIVISM

273

disquotational reference of my present words For all names n of my own present language and objects o, n refers to o iff n ⫽ ‘Alfred’ and o is Alfred, or n ⫽ ‘Willard’ and o is Willard, or . . . (and so on for each name in my present language). For all predicates P of my own present language and objects o, P refers (applies) to o iff P ⫽ ‘rabbit’ and o is a rabbit, or P ⫽ ‘dog’ and o is a dog, or P ⫽ ‘white’ and o is white, or . . . (and so on for each predicate in my present language). When reference is defined in this way, the disquotational claim (i) For all objects o, ‘rabbit’ refers in my own present language to an object o iff o is a rabbit. has the content (ii) For all objects o, either ‘rabbit’ ⫽ ‘rabbit’ and o is a rabbit, or ‘rabbit’ ⫽ ‘dog’ and o is a dog, or ‘rabbit’ ⫽ ‘white’ and o is white, or . . . iff o is a rabbit. If we assume that the set of physical truths includes, or may be augmented by, trivial identity and nonidentity statements of elementary syntax like ‘rabbit’ ⫽ ‘rabbit’, ‘rabbit’ ⫽ ‘dog’, and ‘rabbit’ ⫽ ‘white’, then it is reasonable to suppose that (ii) is a consequence of the set of physical truths, and therefore is determinately true.10 So, commonplace claims I make about my own present reference— using Quine’s disquotational substitute for our ordinary notion—turn out to be determinately true. What about claims I make about the reference of others—using Quine’s substitute for these cases? Here the situation is not as clear-cut. Suppose I make a claim about someone else’s reference by assertively uttering (iii). (iii) ␣, as used by x, refers to (all and only) rabbits.

10 What about the claim expressed by x is a rabbit relative to an assignment of a particular rabbit to ‘x’? Is it determined to be true by the physical truths? I will here assume, for the sake of argument, that it is—either because we take an expansive view of what the physical truths are (which does not limit them to claims found in an ideal physics) or because we have settled on some notion of determination (for our purposes here it doesn’t matter which one) according to which the truths of an ideal physics do determine it to be true. The same points can be made about the claim I express by saying ‘rabbit’ refers to x relative to an assignment of a rabbit to ‘x’, when refers is defined disquotationally for my present language.

274

QUINE’S PHILOSOPHICAL NATURALISM

One way of construing this along Quinean lines would be to see it as making the claim (iv). (iv) ␣, as used by x, means the same as some expression in my language that (disquotationally) refers to (all and only) rabbits. On this construal, my claim about the reference of x’s term carries with it a claim about meaning (derived from a translation manual I implicitly employ), and therefore it is, on Quine’s view, indeterminate. As far as I can see, this might be perfectly all right with him. The advantage of his substitute notion of reference in this case would then lie in its making it clear and explicit why claims about the reference of others must be indeterminate. However, this is not the only alternative open to Quine. My remark (iii) could also be analyzed as having the content (v). (v) There is a translation theory or manual T that I have adopted for translating x’s words onto mine, and there is an expression r in my present language such that (a) according to T, ␣ means the same as r, and (b) r (disquotationally) refers to rabbits. Here, to say that I have adopted a translation manual T need not mean that I endorse claims of the form A means the same as B that arise from T’s mapping of A onto B. Presumably, Quine thinks that one can adopt a translation manual as a practical tool for dealing with other speakers, without being committed to the truth of the manual’s (implicit) claims about meaning. If this is right, then even a demonstration that these claims are indeterminate would not show that (v) is indeterminate.11 It should be noted that (v) is not very plausible, as it stands. As presently formulated, it does not require the translation manual T I have adopted to satisfy any conditions of empirical adequacy. For example, if I have adopted a translation manual T that translates your word ‘rabbit’ onto my phrase ‘prime number’, then according to (iii), my remark You use ‘rabbit’ to refer to prime numbers will be counted as true. But that seems unreasonable. One might address this problem by modifying (v) so that it requires the translation manual T that I have adopted to meet all Quinean conditions on empirical adequacy—i.e., to be compatible with all relevant facts about stimulus meaning. However, then a different problem arises. Suppose that T maps your word ‘rabbit’ onto my word ‘rabbit’, but also maps your word ‘dog’ onto my phrase ‘prime number’. Suppose further that because of this bizarre linking of your talk about dogs with my talk about numbers, T fails to meet Quinean conditions of empirical adequacy. Then, on the proposed modification of (v), this fact alone will be sufficient to guarantee that my claim Your word ‘rabbit’ refers to rabbits is false. But this also seems unreasonable. I leave it here as an open question whether these problems can be solved from a Quinean perspective. I am indebted to Anil Gupta for bring these issues to my attention.

11

RADICAL SEMANTIC ELIMINATIVISM

275

Whether or not (v) does turn out to be indeterminate, on Quine’s view, depends on whether or not he is willing to grant that the physical truths determine that it is true that I have adopted a translation manual according to which ␣ means the same as r, for some expression r of my language that disquotationally refers to rabbits. It is important to recognize in this connection that although in advancing (v) I need not have committed myself to any claim about what ␣ means, I have specified the meaning, or content, of one of the theorems extractable from the translation manual. This is itself a claim about meaning. Thus, (v) will qualify as determinate for Quine only if he grants that the physical truths determine the truth of certain claims about linguistic content. One cannot help but think that, from his perspective, it is just as doubtful that the physical truths will determine the precise content of what is said by a translation theory as it is that they will determine the content of what I say, or what the native says. Thus, on this way of understanding Quine’s construal (v) of my remark (iii), it seems likely that, on his view, claims like (iii), about the reference of expressions used by others, will turn out to be indeterminate, even when the notion of disquotational reference involved is the one that he proposes as a substitute for our ordinary notion. Conceivably, this result could be avoided by replacing (v) with (vi) as an analysis of (iii), thereby eliminating even implicit talk about meaning or content. (vi) I have adopted a system T for correlating x’s words with mine that satisfies all of Quine’s empirical constraints involving stimulus meaning, and for some expression r in my present language that disquotationally refers to rabbits, T maps ␣ onto r. One problem with this proposal is the bizarre result that if it turns out that the system TI have adopted doesn’t quite satisfy all of Quine’s constraints, then my remark that ␣, as used by x, refers to (all and only) rabbits, as well as every other remark I make about x’s reference, will turn out to be false. That seems extreme. However, if we put this worry aside, and focus on the issue of whether (iii), when analyzed as (vi), is the sort of thing that Quine might regard as determined by the physical truths, then the question comes down to whether or not he is willing to regard psychological claims about what system of correlation I have adopted as capable of being determined by the physical truths. Although the point is arguable, let us suppose that he would. On this interpretation, both claims about my own reference and claims about the reference of others

276

QUINE’S PHILOSOPHICAL NATURALISM

could be regarded as fully determinate—when the notion of reference involved is the Quinean substitute for our ordinary notion.12 We are now in a position to sum up our account of Quine’s position. We have seen that physicalism, the indeterminacy of translation, and the inscrutability of reference lead him to eliminativism about meaning and reference, as ordinarily understood. It must be remembered, however, that just as eliminativism about ordinary meaning is compatible with his attempt to construct a limited, behavioristically acceptable substitute notion of meaning (i.e., stimulus meaning), so eliminativism about ordinary reference is compatible with the construction of a limited, physicalistically acceptable substitute notion of reference—namely, disquotational (or Tarski) reference. Disquotational reference is much more restricted than ordinary reference. For Quine, it is applied only to one’s present language, whereas ordinary reference is a relation that purports to hold between expressions, objects, and languages L, for variable ‘L’. In addition, we regard the ordinary notion of reference as a relation that holds between certain expressions and certain corresponding objects (to which the expressions refer) in virtue of the ways in which the expressions are used by speakers. It is because speakers use a word in the ways they do that the word refers to what it does. By contrast, it is not because of how I use the word ‘rabbit’ that it disquotationally refers (i.e., Tarski-refers) to rabbits, and not dogs. On the disquotational definition, the fact The issue of whether remarks about the reference of others turn out to be determinate or indeterminate on the Quinean replacement analysis depends on the details of the analysis, and could in principle be decided either way, without substantially affecting his overall position. Suppose, for example, Quine were to adopt an analysis that rendered them indeterminate. On such a view, claims I make about the reference of others are indeterminate, whereas some claims I make about my own present reference are determinately true. Does this contradict Quine’s original statement that the indeterminacy of reference for me is the same as the indeterminacy of reference for others? No, it does not. Quine’s original indeterminacy claims were claims about statements involving the ordinary notion of reference. When what is meant by ‘refers’ is ordinary reference, Quine’s view is that no claim to the effect that a given word, as used by a given speaker, refers to a given object is determinate; hence no such claim is true, no matter whether the speaker is me or anyone else. However, when the ordinary interpretation of ‘refers’ is replaced by the disquotational, or Tarski, interpretation for my own present language, the claim made by ‘rabbit’ as used by me now refers to x, relative to an assignment of a particular rabbit to ‘x’, is not the claim made by that same formula relative to the same assignment, when ‘refers’ is given its ordinary meaning. Since the claims are different, there is no contradiction in supposing that one is trivial, and determined by the physical truths, whereas the other is not. Similarly, when my claims about my own reference are understood purely disquotationally, while my claims about the reference of others are analyzed as involving claims about translation from their words onto mine, there is no tension in holding that the former may be determined by the set of physical truths even if the latter are not. 12

RADICAL SEMANTIC ELIMINATIVISM

277

(i) that for all objects ‘rabbit’ as I now use it refers to an object o iff o is a rabbit, and it is not the case that for all objects o, ‘rabbit’ as I use it now refers to an object o iff o is a dog is, in essence, nothing more than the fact (ii) that for all objects o, ‘rabbit’ ⫽ ‘rabbit’ and o is a rabbit iff o is a rabbit and it is not the case that for all objects o ‘rabbit’ ⫽ ‘dog’ and o is a dog iff o is a rabbit. Since (ii) is a logical-syntactical-biological fact that doesn’t depend on any aspect of my use of language, the same is true of (i), when reference is understood disquotationally.13

Unbridled Eliminativism: Belief, Assertion, Truth, and Other “Intentional Idioms” We have seen how Quine’s theses of physicalism, the indeterminacy of translation, and the inscrutability of reference lead him to be an eliminativist both about our ordinary notion of meaning—including the notion of a sentence’s meaning that P—and about our ordinary notion of reference (including special types of reference, like denotation and application). That isn’t all. The same considerations that lead to eliminativism about meaning and reference lead to eliminativism about propositional attitudes such as believing that P, asserting that P, and wondering whether P.14 Quine indicates that this is so in section 45 of Word and Object, where he acknowledges that if ordinary intentional notions, including those expressed by the propositional attitude verbs believes and asserts, could be used to state objective facts about agents which were determined to be true by the physical truths, then translation itself would be determinate, which of course he takes it not to be. His conclusion is that claims about what agents believe and assert have the same status as claims about what they mean and refer to in the ordinary sense. No such claims state genuine truths. Just as there is, in reality, no such thing as ordinary meaning and reference, so there is no such thing as ordinary belief and assertion. Hence, there is no place for these notions in a truly objective and scientific description of the world. 13 14

For an explanation of these points, see chapter 4, pp. 107–16, of my Understanding Truth. ‘P’ is here used as a schematic sentence letter.

278

QUINE’S PHILOSOPHICAL NATURALISM

Here is an illustrative, though somewhat florid, passage from section 45. For, using the intentional words ‘believe’ and ‘ascribe’, one could say that a speaker’s term is to be construed as ‘rabbit’ if and only if the speaker is disposed to ascribe it to all and only the objects that he believes to be rabbits. Evidently, then, the relativity to non-unique systems of analytical hypotheses [which gives rise to indeterminacy] invests not only translational synonymy but intentional notions generally. Brentano’s thesis of the irreducibility of intentional idioms [to non-intentional notions] is of a piece with the thesis of indeterminacy of translation. One may accept the Brentano thesis either as showing the indispensability of intentional idioms and the importance of an autonomous science of intention, or as showing the baselessness of intentional idioms and the emptiness of a science of intention. My attitude, unlike Brentano’s, is the second. To accept intentional usage at face value is, we saw, to postulate translation relations as somehow objectively valid though indeterminate in principle relative to the totality of speech dispositions. Such postulation promises little gain in scientific insight if there is no better ground for it than that the supposed translation relations are presupposed by the vernacular of semantics and intention. Not that I would forswear daily use of intentional idioms, or maintain that they are practically dispensable. But they call, I think, for bifurcation in canonical notation. Which turning to take depends on which of the various purposes of a canonical notation happens to be motivating us at the time. If we are limning the true and ultimate structure of reality, the canonical scheme for us is the austere scheme that knows no quotation but direct quotation and no propositional attitudes but only the physical constitution and behavior of organisms.15 The basic idea is that if claims about propositional attitudes, including belief, were determinate, then translation would also be determinate. Since translation is not determinate, claims about propositional attitudes are themselves indeterminate. Strictly speaking, there are no facts about propositional attitudes like belief and assertion, just as there are no facts about translation. Whatever the practical utility of 15

Word and Object, pp. 220–21, my emphasis and additions.

RADICAL SEMANTIC ELIMINATIVISM

279

such notions, they have no place in a properly scientific description of the world. This is a truly breathtaking conclusion. But there is more to come; the final piece of the puzzle remains to be put in place. There is reason to think that all this Quinean eliminativism can be sustained only if he embraces a further eliminativism about our ordinary notion of truth—i.e., only if he rejects the ordinary notion of truth, and replaces it with a pared down and purified substitute. Admittedly, Quine is not explicit about this in Word and Object. However, it is clear that he cannot accept, and knows that he cannot accept, much of what we take to be axiomatic about truth as we ordinarily understand it. For example, he cannot accept commonplace observations about the sorts of things that are true. It might seem, at first glance, that the word true, as we ordinarily understand it, can be applied to a great variety of things. Among the things we ordinarily characterize as being true or false are statements, beliefs, claims, assertions, assumptions, hypotheses, propositions, sentences, and utterances. However, this impression of diversity may be misleading. To the extent that these entities can be identified, one needn’t worry about different kinds of truth bearers. Let us begin with statements. By a statement, I mean that which someone states, not his stating of it. The two are different. If someone states that all power corrupts, then what he states is that all power corrupts. It would be absurd to say that this, namely that all power corrupts, was what he did, or was his particular act of doing it. What he did was to state that all power corrupts, which is something that many others have done. His particular act of stating this was something that happened at a certain place and time. What he stated is neither something that is done nor something that happens. In ordinary speech, the word ‘statement’ is sometimes used to refer to what one states, as in: Mary and Martin made the same statement, which Bob contradicted. At other times it is used to refer to one’s stating something, as in: Mary’s statement lasted for more than a minute. When we speak of a statement as being true or false, we are using ‘statement’ in the first way, on a par with what is said or stated. In this sense, a statement is what many philosophers have called a proposition.16 To state that so and so is, of course, not the same as to assume, believe, conjecture, or claim that so and so. However, what is stated can For an illuminating discussion of these points about truth bearers, see Richard Cartwright’s classic paper, “Propositions,” in Analytical Philosophy, 1st series, ed. R. J. Butler (Oxford: Basil Blackwell, 1962); reprinted in his Philosophical Essays (Cambridge, MA: MIT Press, 1987). 16

280

QUINE’S PHILOSOPHICAL NATURALISM

also be assumed, believed, conjectured, claimed, or asserted. Thus, statements can be identified with assumptions, beliefs, conjectures, claims, and assertions, provided that these terms are taken to refer to what is assumed, believed, conjectured, claimed, or asserted, rather than to acts, or states, of assuming, believing, and the like. Since it is perfectly correct to speak of beliefs, assumptions, and the rest as being true, what might initially have seemed to be a multiplicity of different truth bearers can be reduced substantially. In all of these cases, we predicate truth of propositions. Since propositions are the meanings or contents of sentences, and the objects of belief and assertion, and since Quine is an eliminativist about ordinary meaning, belief, and assertion, he rejects propositions. Thus he rejects the primary bearers of truth and falsity, as we ordinarily understand those notions. Of course, we also apply our ordinary notion of truth to sentences and utterances, the existence of which he grants. Still, there are two reasons why he can’t accept our ordinary notion of truth as applied to sentences and utterances. First, when we say that a sentence or an utterance is true, ordinarily we mean that what it says, or what it is used to express, is true. If a sentence doesn’t mean anything, if it cannot be used to say or express anything, then it is not true in the ordinary sense. Since Quine holds that sentences don’t mean anything in the ordinary sense, since he holds that, strictly speaking, they are not used to say or assert anything, and since he further maintains that they don’t express any beliefs, his position leads to the conclusion that they are not true in the ordinary sense either. Second, our ordinary notion of the truth of a sentence is conceptually connected with our ordinary notion of reference, which Quine rejects. Let P be a predicate, let t be a singular term, and let S be the sentence that results from combining the two. Then S is true in the ordinary sense iff P is true of the object that t refers to, in the ordinary sense. But the ordinary relation of a predicate being true of an object is just the relation of the predicate referring or applying to an object that Quine rejects. Since he holds that no words refer or apply to objects in the ordinary sense, he must hold that no predicates are true of any objects in the ordinary sense, and hence that no sentences containing such predicates can be true in the ordinary sense of that term. One can’t be an eliminativist about the ordinary notion of a predicate being true of an object without being an eliminativist about the notion of a sentence containing the predicate being true in the ordinary sense. So, in addition to being an eliminativist about ordinary meaning and reference, Quine must be interpreted as being an eliminativist regarding

RADICAL SEMANTIC ELIMINATIVISM

281

our ordinary notion of truth. Strictly speaking, on his view, no sentences are true in the ordinary sense. Quine realizes this, and is not seriously troubled by it. One of the chief reasons for this is that he has a systematic strategy for softening the blow of all this eliminativism. His strategy is to suggest replacements of our ordinary notions of meaning, reference, and truth. Ordinary meaning is to be replaced by stimulus meaning, ordinary reference is to be replaced— in the special case of one’s own present language—by Tarski-reference, and ordinary truth is to be replaced—also in the case of one’s own present language—by Tarski-truth. Tarski-reference and Tarski-truth are disquotational in spirit, and can, in principle, be defined using completely formal, list-type definitions without appealing to any ordinary semantic or mentalistic notions. We will talk more about this in Part 6, in our discussion of Donald Davidson’s theory of truth and meaning.17 For now, we simply note that the claim that a sentence S of my own present language is Tarski-true is trivially equivalent to S itself. Consequently, the claim that S is Tarski-true will be determined to be true by the physical truths iff the original sentence S is determined to be true by the physical truths. Thus, the claim that a sentence of my language is Tarskitrue never introduces any new indeterminacy.

Evaluating Quine’s Position: The Argument That It Is Self-Undermining That is how Quine’s position should be understood. The next question is, what should we think about it? The first thing to remember is that Quine’s wholesale rejection of central tenets of our ordinary ways of understanding ourselves and our words is without any firm foundation. His radical skepticism about our ordinary intentional notions— meaning, reference, belief, and so on—is based on his argument for the indeterminacy of translation, and subsequent extensions of it. If my presentation of the argument in chapter 10 was correct, then once equivocation on the determination relation is resolved, the set of premises he appeals to loses its claim to independent plausibility. In effect, we are asked to give up central and vital elements of our conceptual scheme on the slenderest of grounds. That is bad enough; but it isn’t the worst of it. The worst of it is that Quine’s position has several 17

See also chapters 3 and 4 of my Understanding Truth.

282

QUINE’S PHILOSOPHICAL NATURALISM

consequences that are so unpalatable as to make it reasonable to regard it as self-undermining. The first, and perhaps least troubling, of these consequences involves Quine’s proposed elimination of all the normal semantic notions that one might use to identify and individuate languages, and to specify the language one speaks. Normally, we think of a language as a set of expressions with certain semantic properties. We identify languages (i) by the pronunciation of their expressions, (ii) by the syntactic and morphological structure of their words, phrases, and sentences, and (iii) by the meaning and reference of their words, the truth conditions of their sentences, and the propositions those sentences express. If the ordinary semantic notions of meaning, reference, truth, and propositions expressed are eliminated, then it is not clear that one can give characterizations of the languages spoken by oneself and others that are informative enough to make clear precisely what these languages are, and how they differ from one another (in ways that go beyond (i) and (ii)). I can give the conditions under which sentences are TARSKI-TRUE-IN-MY-PRESENT-LANGUAGE, but these conditions provide no information about what my words mean or refer to, no information about what propositions my sentences express, and no way of informatively identifying what language I speak.18 Even the ordinary notion of translation becomes difficult to grasp, in so far as it presupposes the identification of different languages, in part based on their semantic properties. For the Quinean, this may just be more eliminativism and revisionism; for the rest of us, it is another reason to resist that point of view. The second, and more serious, unpalatable consequence of Quine’s position is that his radical eliminativism runs the risk of rendering his own indeterminacy and inscrutability theses either false, uninteresting, or unstatable. Consider, for example, Quine’s claim that biology is determined by the physical truths, but semantics and non-behaviorist psychology are not. In speaking of truths here, Quine has to be speaking of Tarski-truths of the present-Quine-language. So we have a thesis to the effect that a certain set of physical sentences of present-Quinelanguage determines a certain other set of biological sentences in present-Quine-language, but does not determine a third set of semantic and psychological sentences in present-Quine-language. Presumably, for Quine, determination must involve the Tarski-entailment of one 18

For an explanation of this point, see pp. 102–7 of my Understanding Truth.

RADICAL SEMANTIC ELIMINATIVISM

283

set of sentences in present-Quine-language by physical sentences of that language, plus sentences that play the role of definitions in that language. Does this really capture Quine’s intention? There is reason to think that it doesn’t, since it may very well turn out that our present physics, biology, and so on, are expressively incomplete in the sense of lacking certain concepts needed for an accurate and comprehensive scientific description of the world. If the totality of physical concepts capable of being expressed in the present-Quine-language is incomplete in this way, then the fact that some set of claims is not determined by the Tarski-true physical sentences of Quine’s present language will not carry the desired ontological implications. Suppose that three hundred years ago, before we had atomic theory or modern biology, someone claimed that the set of Tarski-true physical sentences in his then present language failed to determine the set of Tarski-true biological sentences of his language. Surely no interesting philosophical conclusion would follow from this, even if it were true. There are two cases to consider: First, if the relevant physics is conceptually impoverished, then the failure of some claim Q to be determined by the set of true physical sentences won’t provide any compelling reason to doubt the genuine truth of Q, since the doctrine of physicalism, formulated in a way that appeals to a conceptually impoverished physical language, will not be something anyone would want to accept. (Note, the problem is not that the theorems of the physical theory adopted at a given time may leave out some physical truths, but that the physical language in which the theory is formulated is not capable of expressing all the physical facts.) Second, if the theory to be determined is impoverished, then even if Q does turn out to be determined, this won’t carry the needed guarantee that there aren’t further facts of the sort that a correct, unimpoverished theory would include, but that are nevertheless underdetermined by physics. For these reasons, Quine’s theses of physicalism, underdetermination, indeterminacy, and inscrutability cannot be given their intended force, if they are stated using only semantic notions restricted to his present language. One feels like saying that his theses concern claims about the relationship between the totality of physical truths—in the sense of the totality of true propositions about the subject matter of physics, or the totality of true physical sentences in some ideal language—on the one hand, and the totality of truths about anything else, on the other. But such formulations of Quine’s theses presuppose that the truths in

284

QUINE’S PHILOSOPHICAL NATURALISM

question are not limited to those expressible in one’s present language— and this is incompatible with his radical semantic eliminativism. The problem is that there seems to be no way to give his theses the needed generality without appealing to semantic or other intentional facts that those theses are meant to eliminate. If this is right, then his position can be criticized as being subject to a crippling dilemma. Either his theses of physicalism, underdetermination, indeterminacy, and inscrutability state something about the totality of physical truths in the ordinary sense, in which case they are inconsistent with the radical eliminativism to which they drive him, or else his theses state something only about the totality of Tarski-truths in his present language, in which case the underdetermination theses do not have their intended philosophical import, and the theses of physicalism, indeterminacy, and inscrutability are nonstarters. Faced with this dilemma, the die-hard Quinean could, of course, simply adopt his radical eliminativism by fiat. What he could not do is give Quinean reasons for so doing. Even if, per impossible, his physics ever were really complete, he could not formulate in his preferred physicalistic language the significant question of whether it was complete, without invoking the very semantic notions he had already sworn off. Since Quine himself obviously did not swear off them— even when attempting to state important truths—he should, I think, have accepted the implications of his implicit commitment, and sworn off his radical eliminativism. Finally, there is an additional way in which Quine’s position is selfundermining. Consider the following illustrations. (i) Quine says, or asserts, something—namely the conjunction of physicalism and his indeterminacy theses—that has the consequence that no one ever says or asserts anything. (ii) Similarly, Quine believes something that has the consequence that no one ever believes anything. (iii) Quine argues for something—his indeterminacy theses—by producing a sequence of meaningful sentences that has the consequence that there are no meaningful sentences. It follows from these illustrations that the very existence of Quine’s own assertions, his own beliefs, and his own arguments is sufficient to falsify that which he asserted, believed, and argued for. What he asserted,

RADICAL SEMANTIC ELIMINATIVISM

285

believed, and argued for has the character that the very act of asserting, believing, or arguing for it is itself sufficient to falsify it. Can Quine avoid this kind of self-undermining? Well, he could claim that he didn’t assert, believe, or argue for his theses. Though this might satisfy him, it can hardly satisfy us, since if he claims this, then he will have claimed something, and that will be enough to falsify his theses. For us, the problem is this: Quine writes a book that contains lots of sentences. We see the book and we want to know what it says. So we read it. In order to agree or disagree with him, we have to figure out what he said, and what he was trying to say. However, all the concepts we would ordinarily use to describe this are forbidden to us by his eliminativist theses. Could we invent some technical notions to replace those eliminated that would allow us to answer our questions, without reintroducing the allegedly objectionable features of our ordinary notions? I see no reason to think so. So it seems that we really have no choice but to regard Quine’s position as paradoxical and self-undermining.19 Is this, then, a refutation that even the Quinean skeptic would have to accept? Not, I suppose, if he is really prepared to accept the view that no one, himself included, ever means, says, believes, or claims anything. In fact, I don’t think that anyone ever has, or ever could, consistently accept such a thing.20 Still, the position explicitly stated by the skeptic is not self-contradictory, and it may be conceivable that a skeptic could adopt it (while of course denying that he did any such thing). However, even if one ignores the mismatch between what the skeptic states and his stating of it, there is considerable tension in the different things he must say. We have already seen that, as a Quinean, See section IV of my “The Indeterminacy of Translation and the Inscrutability of Reference,” for attempts to further refine such central Quinean theses as physicalism, and for an exploration of how much of his skeptical vision might be saved from refutation by a sympathetic reformulation of his basic position. 20 That is, if anyone accepted the view, he would, I think, be guilty of implicit inconsistency because, despite his protestations, I could not but regard him as also believing the negation of the view. One is reminded here of G. E. Moore’s classic comment on p. 44 of his “A Defense of Common Sense” (Contemporary British Philosophy, 2nd series, ed. J. H. Muirhead, 1925; reprinted in G. E. Moore, Philosophical Papers, Collier Books, 1962, 32–59): “I am one of those philosophers who have held that the ‘Common Sense view of the world’ is, in certain fundamental features, wholly true. But it must be remembered that, according to me, all philosophers, without exception, have agreed with me in holding this [i.e., they have all believed it to be true]: and that the real difference, which is commonly expressed in this way, is only a difference between those philosophers, who have also held views inconsistent with these features in ‘the Common Sense view of the world,’ and those who have not.” See chapter 1 of volume 1 for discussion. 19

286

QUINE’S PHILOSOPHICAL NATURALISM

he can’t get by without our ordinary non-behavioristic notions of assenting to and dissenting from a sentence. What sense, one wonders, could anyone attach to those notions who also maintained that no one could ever say, assert, claim, believe, or deny anything?21

Quine, The Ordinary Language School, and the Fate of Meaning in Philosophy As we have seen, linguistic meaning was a central preoccupation of the leading English-speaking philosophers in the first two decades after World War II. On the one hand, we had the ordinary language philosophers, who were convinced (i) that all philosophical problems arise from misunderstanding the meanings of our words, and (ii) that a proper understanding of meaning can be achieved by an informal investigation of how we normally use words, essentially unaided by any abstract logical or scientific theories of meaning. On the other hand, we had Quine, who thought (i) that philosophical problems are not problems about the meanings of words, and that philosophical theorizing should be continuous with logical and scientific theorizing about the world, and (ii) that our ordinary semantic notions, including meaning and reference, are essentially pre-scientific relics, with no scientific use, and corresponding to nothing in the real world. The contrast could hardly be greater, and the views could hardly be more extreme. I have argued that both went overboard. The ordinary language philosophers did not demonstrate that philosophical problems are all, at bottom, problems of meaning. In addition, their investigations of meaning—while valuable and illuminating in certain cases—were deeply flawed by the lack of any systematic theory of meaning to guide them. While Quine clearly and correctly recognized the need for theory, and while he was quick to recognize philosophically suspect appeals to meaning, his conception of the proper relationship between theories of mind and language, on the one hand, and theories of fundamental physical reality, on the other, was a piece of remarkably unrestrained speculation in the service of an extravagant philosophical vision. What was needed, after Quine and the ordinary language philosophers, was a way of taking meaning seriously, and studying it theoretically. It was to this fundamental need that our next philosopher, Donald Davidson, addressed himself. 21

Thanks to John Hawthorne for raising this final question.

SUGGESTED FURTHER READING FOR PART FIVE

Main Primary Sources Discussed Quine, W. V. “On the Reasons for Indeterminacy of Translation.” Journal of Philosophy 67 (1970): 178–83. ———. “Ontological Relativity.” In Ontological Relativity and Other Essays (New York: Columbia University Press, 1969). ———. “Reply to Chomsky.” D. Davidson and J. Hintikka, eds., In Words and Objections (Dordrecht: Reidel, 1969). ———. Word and Object. Cambridge, MA: MIT Press, 1960.

Additional Primary Sources Carnap, Rudolf. “Meaning and Synonymy in Natural Languages.” In Meaning and Necessity (Chicago: University of Chicago Press, 1956). Grice, H. P., and Peter Strawson. “In Defense of a Dogma.” Philosophical Review 65 (1956). Quine, W. V. “Epistemology Naturalized.” In Ontological Relativity and Other Essays (New York: Columbia University Press, 1969). ———. “On Empirically Equivalent Systems of the World.” Erkenntnis 9 (1975): 313–28. ———. The Pursuit of Truth. Cambridge, MA: Harvard University Press, 1992. ———. The Roots of Reference. La Salle, IL: Open Court, 1974. ———. “Two Dogmas of Empiricism.” Philosophical Review 60 (1951).

Additional Recommended Reading Berger, Alan. “Quine on ‘Alternative Logics’ and Verdict Tables.” Journal of Philosophy 77 (1980): 259–77. Cartwright, Richard. “Propositions.” In Philosophical Essays (Cambridge, MA: MIT Press, 1987); originally published in Analytical Philosophy, 1st series, ed. R. J. Butler (Oxford: Basil Blackwell, 1962).

288

SUGGESTED FURTHER READING

Friedman, Michael. “Physicalism and the Indeterminacy of Translation.” Noûs 9 (1975): 353–73. Harman, Gilbert. “Meaning and Theory.” Southwestern Journal of Philosophy 9, (1979): 9–20. Soames, Scott. “The Indeterminacy of Translation and the Inscrutability of Reference.” Canadian Journal of Philosophy 29 (1999): 321–70. ———. Chapters 3 and 4 of Understanding Truth. New York: Oxford University Press, 1999.

PA R T S I X DONALD DAVIDSON ON TRUTH AND MEANING

This page intentionally left blank

CHAPTER 12 THEORIES OF TRUTH AS THEORIES OF MEANING chapter outline 1. Davidson’s Place in Analytic Philosophy His role in bringing together formal and informal approaches to meaning in the analytic philosophy of language 2. Central Tenets of Davidson’s Philosophy and the “Davidsonian Program” Tarski-style truth theories for natural languages as theories of meaning that can be verified by comparing the truth conditions they assign to sentences with the conditions in which speakers hold their sentences to be true 3. What Justifies Taking Theories of Truth to be Theories of Meaning? Davidson’s first answer: knowledge of that which is stated by a finite, compositional truth theory of L is sufficient for understanding L; the incorrectness of this answer Davidson’s second answer: knowledge that a translational truth theory of L states that . . . is sufficient for understanding L (where the dots are filled in by the conjunction of axioms of a translational truth theory); a fallacy in the argument for this claim 4. Conclusions Doubts about Davidson’s contention that truth theories are theories of meaning, and a contemporary disconnect between technical work on the Davidsonian program and Davidsonian philosophical justifications of the program

Davidson’s Place in Analytic Philosophy Donald Davidson is well known for his important contributions to a number of philosophical topics, including (i) events, actions, the explanation of action, and the logical form of action sentences,1 (ii) the 1 See, for example, Donald Davidson, “Actions, Reasons, and Causes,” originally published in 1963, and “The Logical Form of Action Sentences,” originally published in 1967, both reprinted in Essays on Actions and Events (Oxford: Clarendon Press, 1980).

292

DAVIDSON ON TRUTH

nature of the mental and its relation to the physical,2 and (iii) the role of meaning in philosophy and the proper theoretical approach for studying it. Although his views had a major impact in each of these areas, his views about meaning were, arguably, the most central components of his philosophical point of view, as well as the most widely known and far-reaching. It is these with which we will be concerned. Davidson is famous for defending a systematic and widely influential conception of what a theory of meaning for a language is, how we should go about constructing such theories, what evidence justifies their acceptance, and what philosophical payoffs we may legitimately expect from them. He developed his views on these matters between the early 1960s and the late 1970s, during which time he was a professor first at Stanford University, then Princeton University, Rockefeller University, the University of Chicago, and finally the University of California at Berkeley. Though his views were developed and first presented during this period, they remained influential for a much longer time, and they are still with us today. Before getting down to details, I will to say a few words about the philosophical traditions he addressed, and some of the reasons why his work struck a responsive chord. First, in our study of post–World War II philosophy in England, from Wittgenstein through Austin and Grice, we have seen how linguistic meaning was placed at center stage. Philosophical problems were thought to arise from linguistic misunderstanding, and to be resolvable by careful attention to the meanings of words. However, as time went on, it became increasingly clear that what is, and what is not, part of the meaning of a word is not a transparent matter. Because of this, questions about meaning can be investigated profitably only with the help of some systematic theory. This was a problem for the philosophers of the ordinary language school, since they had little idea of what a theory of meaning for a language should look like, or how to go about constructing one. Thus, by the early 1960s there was a clear and urgently felt need, on the part of many philosophers, for a systematic theoretical approach to linguistic meaning. At the same time, philosophers working in a different analytic tradition had assembled an impressive set of logical tools for studying and interpreting the formal languages of logic and mathematics. In the See Donald Davidson, “Mental Events,” originally published in 1970, and “The Material Mind,” originally published in 1973, both reprinted in Essays on Actions and Events.

2

THEORIES OF TRUTH AS THEORIES OF MEANING

293

mid-1930s the Polish logician and philosopher Alfred Tarski showed how to give what he called a definition of truth for the languages of symbolic logic—languages which can be used to express standard mathematics, meta-mathematics, and (arguably) natural science. Tarski also showed how to define logical truth and logical consequence for these languages in a precise, mathematically tractable way.3 One of the main purposes to which his work was put was that of providing interpretations of different formal languages. To give an interpretation of one of these languages is to specify a domain of objects that the language is used to make claims about, to specify, for each name in the language, the object (in the domain) to which it refers, to assign to each function sign in the language a mapping from objects (in the domain) to objects (in the domain), to assign to each one-place predicate in the language the set of objects (in the domain) that the predicate is to be understood as being true of, to assign to each two-place predicate a set of pairs of objects from the domain that the predicate is to be understood as applying to, or being true of, and so on for n-place predicates generally. Doing all of this provides an interpretation of the vocabulary of the language. The interpretation of the sentences is then derived from the interpretation of the words via the clauses in Tarski’s so-called definition of truth. Actually, when Tarski’s logical machinery is used to provide languages with an interpretation, it should not be seen as giving us a definition of truth. When used in this way, it gives us a theory that, for each of the infinitely many sentences of the language, assigns a condition that obtains if and only if that sentence is true—where truth is something we take ourselves to understand antecedently, without definition. The end result of the interpretation is the derivation of at least one statement of the form (1) for each sentence of the language being interpreted—where the sentence replacing ‘P’ is supposed to be a paraphrase of the sentence replacing ‘S’. (The connective ‘iff’ is the material biconditional. A statement of this form is false when the sentences on the two sides of the biconditional operator differ in truth value. In all other cases—i.e., when both sentences are true, or when both are false—the biconditional statement is true.) 1. ‘S’ is a true sentence of L iff P. Alfred Tarski, “The Concept of Truth in Formalized Languages,” in Logic, Semantics, MetaMathematics, translated by J. H. Woodger, 2d edition (Indianapolis, IN: Hackett, 1983), originally published in Polish in 1933; “On the Concept of Logical Consequence,” in Logic, Semantics, Meta-Mathematics, originally published in Polish in 1936.

3

294

DAVIDSON ON TRUTH

Sentences of this form are called T-sentences. They specify the truth conditions of the sentences quoted on their left-hand sides. An example of a possible T-sentence is (2). 2. ‘᭚x (Loves fm(x), ff (x))’ is true in L iff there is at least one person whose mother loves that person’s father. (‘Loves’ applies to iff a loves b, ‘fm’ denotes a function that assigns to a person that person’s mother, ‘ff’ denotes a function that assigns to a person that person’s father.) An interpretation of the entire language is a finitely specifiable set of statements that allows us to derive something similar to (2) for each sentence in the language.4 This conception of an interpretation of a formal language was familiar throughout the ’40s, ’50s, and ’60s to logicians and philosophers who believed that the study of logic held important lessons for our understanding of language, and for philosophy in general. Among the philosophers working in this tradition were Rudolph Carnap, Alonzo Church, Ruth Marcus, Arthur Prior, Richard Montague, David Kaplan, and Saul Kripke. One of the main lines of research developed in this period involved enriching the formal languages of symbolic logic amenable to Tarski’s techniques, so that they could express more and more of the concepts that play central roles in natural language—for example, modal concepts expressed by words like ‘actual’, ‘necessary’, ‘possible’, ‘could’, ‘would’, and ‘might’, temporal concepts expressed by the different tense operators we find in natural language, indexical expressions like ‘I’, ‘we’, ‘you’ , ‘he’, ‘now’, ‘then’, ‘today’, and ‘yesterday’, and propositional attitude verbs, such as ‘believe’ and ‘know’. By the early sixties it had become possible to imagine the day in which natural languages like English would be treatable in something close to their entirety by descendants of the logical techniques initiated by Tarski. This is when Davidson came onto the scene. One of his main contributions was to make some of the simplest of the techniques employed with formal languages accessible and relevant to philosophers who had long been interested in natural language, but who were relative outsiders to the formal, logical tradition. Davidson took an approach that had proved fruitful in a specialized domain, and put it in the form of a program for constructing empirical theories of meaning For an explication of technical details and an explanation of the philosophical purposes to which Tarski’s formal theory of truth can be put, see chapters 3 and 4 of my Understanding Truth.

4

THEORIES OF TRUTH AS THEORIES OF MEANING

295

for natural languages. In addition, he argued that this approach could be expected to yield significant philosophical payoffs. In so doing, he helped to establish significant contact between two different philosophical subcultures in analytic philosophy. One of these consisted of linguistic philosophers who believed that questions about meaning lay at the center of all philosophy, but who had previously lacked a theoretical framework for thinking systematically about meaning. The second subculture consisted of formalists who had been busy using artificial languages to model various aspects of natural language, but who, with some exceptions, had not been centrally concerned with connecting this formal work to the broad range of philosophizing about traditional subjects. A big part of Davidson’s impact was due to the fact that he got these two groups to pay more attention to each other. He also appealed to those who were impressed with Quine’s naturalism, as well as his dismissal of modal notions like necessity and possibility, but who worried about his rejection of ordinary conceptions of meaning in favor of thoroughly sanitized behavioral substitutes. For philosophers with these concerns, Davidson’s seemingly hard-headed insistence on truth and reference as the central notions in a scientific theory of meaning, and his explanation of how such a theory could be viewed as an empirically testable hypothesis about the behavior of speakers, promised a way of vindicating and legitimizing semantic notions that, it seemed clear, we could not do without. For all these groups, Davidson seemed, at least for a time, to be just what the doctor ordered.

Central Tenets of Davidson’s Philosophy, and the “Davidsonian Program” With this background in place, we now turn to Davidson’s most important views, the three most important and central of which are the following: C1. It is possible to construct finitely axiomatizable theories of truth for natural languages that include among their logical consequences T-sentences that give the truth conditions for each sentence of the language under investigation. These T-sentences are derived from axioms of the truth theory that specify the referential properties of words and phrases that occur in the language. Thus, a statement of

296

DAVIDSON ON TRUTH

the truth conditions of each sentence is derived from more basic statements about the referential properties of the words that make it up. C2. A theory of truth for a language of this sort gives the truth conditions of every sentence of the language, and so qualifies as a theory of meaning, or interpretation, for a language. C3. We empirically verify that a proposed theory of truth for the language of a given group of speakers is correct by comparing the conditions, or situations, in which speakers hold particular sentences of their language to be true with the truth conditions assigned to those sentences by the theory being tested. All other things being equal, the correct theory of truth for the language of a given community is the theory according to which the conditions in which the speakers actually hold sentences to be true most closely matches the conditions in which the combination of the theory with our theory of the world predicts the sentences to be true. Roughly put, the correct theory is the theory according to which speakers of the language turn out to be truth tellers more frequently than on any other interpretation of the language.5 We will begin our examination of these claims with a few observations about the first—the claim that it is possible to construct theories of truth of the sort Davidson has in mind for natural languages. It is important to realize at the outset that this claim is not the statement of an already accomplished task. Natural languages are very complex, and no one has produced a Davidson-style theory of truth for the whole of any natural language. What Davidson began calling for in the early ’60s was a systematic effort to achieve that goal. To that end, he himself wrote articles attempting to explain how various linguistic constructions found in English could, in principle, be treated by the kind of truth theory he had in mind. The constructions he dealt with in his own articles included propositional attitude verbs like say and believe, certain adverbial modifiers used in the description of actions, and even the quotation construction.6 Other philosophers, inspired by his example, 5 The degree to which a theory satisfies this condition (of being maximally charitable to speakers) is, of course, not the only factor that goes into determining its correctness. For example, we also expect theories to be systematically compositional, optimally simple, and so on. Davidson would, I think, recognize a correct theory to be one that achieves an optimal balance of these virtues. 6 See, for example, Davidson, “On Saying That,” in D. Davidson and J. Hintikka, eds., Words and Objections (Dordrecht: Reidel, 1969), and “Quotation,” Theory and Decision 11 (1979): 27–40; both reprinted in Inquiries into Truth and Interpretation, 2nd ed. (Oxford: Clarendon Press, 2001). See also “The Logical Form of Action Sentences,” originally published in 1967, and reprinted in Essays on Actions and Events.

THEORIES OF TRUTH AS THEORIES OF MEANING

297

tried to show how different natural language constructions could be made to fit into the picture. The cooperative enterprise of Davidson and his followers came to be known as the Davidsonian program. Essentially, this program was an attempt to produce truth theories meeting his requirements for increasingly larger and more encompassing fragments of natural languages.7 This aspect of Davidson’s overall view can be seen as a sort of technical project. Viewed as such, it has had, to date, some substantial success, but it has certainly not achieved its ultimate goal. Moreover, the technical constraints that Davidson places on his theories are rather confining. They have the effect of excluding much of the more modern and sophisticated logical techniques that have been developed and used to model different features of natural language since the time of Tarski.8 For this reason, many philosophers of language and linguists who agree with Davidson that fruitful empirical theories of natural languages can be constructed using concepts and techniques initially developed for the formal languages of mathematics and symbolic logic, nevertheless disagree with his devotion to the first and most elementary logical concepts and techniques developed for this purpose. Since this is a matter that goes beyond what we can profitably discuss here, I will simply say that adherence to the particular technical constraints incorporated in the Davidsonian program is something that today represents more of a minority position than a consensus in the philosophy of language.

What Justifies Taking Theories of Truth to Be Theories of Meaning?9 Davidson’s First Answer

More important than the technical issues raised by the Davidsonian program are the philosophical claims he makes about his approach. Particularly important is C2—the claim that a Davidsonian theory of truth for a language qualifies as a theory of meaning, because it gives For one of the most systematic recent attempts to produce a Davidson-style theory for natural language, see Richard Larson and Gabriel Segal, Knowledge of Meaning (Cambridge, MA: MIT Press, 1995). 8 These are briefly touched on in the Epilogue. 9 The material in this section is, in substantial part, based on my “Truth, Meaning, and Understanding,” Philosophical Studies 65 (1992): 17–35. 7

298

DAVIDSON ON TRUTH

the truth conditions of every sentence of the language. Initially, this claim seems surprising. On the one hand, one would think that a theory of meaning for a language should tell us what the sentences of the language mean. On the other hand, a theory of truth only tells us what their truth conditions are. But statements of truth conditions are weaker than statements of meaning. For example, instances of schema M M. ‘S’ means in L that P. together with analytic, apriori instances of schema TM TM. If ‘S’ means in L that P, then ‘S’ is true in L iff P. entail the corresponding instances of schema T T. ‘S’ is true in L iff P. However, there is no true principle, let alone an analytic, apriori principle, which, together with instances of schema T, allows one to derive instances of schema M.10 Thus, statements about meaning that are instances of schema M are stronger, and more informative, than statements about truth conditions that are instances of schema T. This illustrates a problem inherent in taking theories of truth to be theories of meaning. If truth theories provide only statements of truth conditions, and if these do not entail statements indicating what sentences mean, then theories of truth don’t tell us what the sentences of a language mean. How, then, can one justify taking them to be theories of meaning? Davidson’s answer to this question is expressed in his articles “Truth and Meaning,” originally published in 1967, and “Radical Interpretation,” which first appeared in 1973.11 His initial answer was that a theory of truth for a language L qualifies as a theory of meaning for L if knowledge of that which it states is, in principle, sufficient for understanding the language. Here is a passage from “Radical Interpretation.” We are now in a position to say something more about what would serve to make interpretation possible. The interpreter must be able to understand any of the infinity of sentences the speaker 10 Instances are obtained by replacing the letter ‘S’ with a sentence of language L and replacing ‘P’ with a sentence of the language of the theory. 11 Donald Davidson, “Truth and Meaning,” originally published in Synthese 17 (1967), and “Radical Interpretation,” originally published in Dialectica 27 (1973), both in Inquiries into Truth and Interpretation. Page references will be to this anthology.

THEORIES OF TRUTH AS THEORIES OF MEANING

299

might utter. If we are to state explicitly what the interpreter might know that would enable him to do this, we must put it in finite form. If this requirement is to be met, any hope of a universal method of interpretation must be abandoned. The most that can be expected is to explain how an interpreter could interpret the utterances of speakers of a single language (or a finite number of languages): it makes no sense to ask for a theory that would yield an explicit interpretation for any utterance in any (possible) language. It is still not clear, of course, what it is for a theory to yield an explicit interpretation of an utterance. The formulation of the problem seems to invite us to think of the theory as the specification of a function taking utterances as arguments and having interpretations as values. But then interpretations would be no better than meanings and just as surely entities of some mysterious kind. So it seems wise to describe what is wanted of the theory without apparent reference to meanings or interpretations: someone who knows the theory can interpret the utterances to which the theory applies. . . . What follows is a defense of the claim that a theory of truth, modified to apply to a natural language, can be used as a theory of interpretation. The defense will consist in attempts to answer three questions: 1. Is it reasonable to think that a theory of truth of the sort described can be given for a natural language? 2. Would it be possible to tell that such a theory was correct on the basis of evidence plausibly available to an interpreter with no prior knowledge of the language to be interpreted? 3. If the theory were known to be true, would it be possible to interpret utterances of speakers of the language? The first question is addressed to the assumption that a theory of truth can be given for a natural language; the second and third questions ask whether such a theory would satisfy the further demands we have made on a theory of interpretation.12 12

Davidson, “Radical Interpretation,” pp. 127–31, my emphasis.

300

DAVIDSON ON TRUTH

So, a theory of truth will qualify as a theory of meaning, provided that it is the case that if one knew that which is stated by the theory, then one would be able to interpret, or understand, the language. This criterion is not without plausibility. If a theory tells you everything you need to know in order to understand the sentences of a language, then it may be defensible to count it as specifying all essential facts about meaning, even if it does not provide theorems, ‘S’ means in L that P, which state the meanings of individual sentences one by one. With this in mind, let us provisionally accept the claim that a theory which provides information knowledge of which is sufficient for understanding a language would qualify as a theory of meaning. The difficulty lies in seeing how a theory of truth could satisfy this condition. Davidson’s original idea, in “Truth and Meaning,” was that a truth theory of the proper sort would give what he calls a “holistic” account of meaning, if it derived an appropriate statement of the truth conditions for each sentence from an account of the semantically significant structure of the sentence, including the reference of its semantically significant parts. On Davidson’s picture, the meaning of a sentence depends on the meanings, or, more properly, the referents of its ultimate parts. These in turn (that is, the referents of the parts) are regarded by him as nothing more than the contributions they make to the meanings (truth conditions) of all the sentences in which they occur. From this, he concludes that meaning, namely that which a competent speaker grasps, is supposed to be a structure that can be revealed only in the whole. What we want from a theory of meaning is a specification of the complex network of relationships mastery of which is sufficient to endow a speaker with semantic competence. This, I take it, was to be Davidson’s excuse for regarding the appropriate sort of theory of truth to be a theory of meaning, even though it fails to provide theorems stating what any individual sentence means. Davidson expresses essentially this view in the following passage form “Truth and Meaning.” We decided a while back not to assume that parts of sentences have meanings except in the ontologically neutral sense of making a systematic contribution to the meaning of the sentences in which they occur. . . . One direction in which it [this insight] points is a certain holistic view of meaning. If sentences depend for their meaning on their structure, and we understand the meaning of each item in the structure only as an abstraction from

THEORIES OF TRUTH AS THEORIES OF MEANING

301

the totality of sentences in which it figures, then we can give the meaning of any sentence (or word) only by giving the meaning of every sentence (and word) in the language.13 The resulting picture seems to go something like this: We know the meaning of a sentence on the basis of knowing the meaning or reference of its parts. This is reflected in the truth theory by the derivation of T-sentences, giving the truth conditions of sentences, from axioms about the reference of their parts. But these axioms about what individual words refer to do not state facts that correspond to anything in reality. There is no reference relation “out there” in the world connecting words with objects. Rather, the axioms of the truth theory that specify the reference of individual words are simply pieces of theoretical machinery that are instrumentally useful in deriving the T-sentences, which are the theorems of the theory. One of these referential axioms gets its content, not from corresponding to anything in the world, but from its role in connecting all the theorems about truth conditions of sentences containing the word mentioned in the axiom to one another. For example, the content of the axiom that says that the word azul in Spanish applies to an object iff the object is blue lies simply in the role played by this axiom in deriving the T-sentences for different Spanish sentences containing the word azul. Consequently, what you know when you know the truth conditions of the sentence La camisa es azul on the basis of knowing the reference of its parts are the ways in which the truth conditions for this sentence are systematically related to the truth conditions of all other sentences that contain either the word azul or the phrase la camisa. What a speaker of a language knows about meaning is not a set of unconnected truths, but rather an intricate system of relationships that extends to every word, phrase, and sentence in the language. Davidson thinks that this is just what one knows when one knows a truth theory for a language. He further elaborates this view in a passage from “Truth and Meaning” in which he discusses the deviant, non-translational, but true T-sentence S. S. ‘Snow is white’ is true iff grass is green. In considering S, it is important to remember that the connective iff in the T-sentences generated by a Davidsonian truth theory is the material biconditional. Thus, A iff B is true when, and only when, both A and B have the same truth value—i.e., when both are true, or both are 13

Davidson, “Truth and Meaning,” p. 22.

302

DAVIDSON ON TRUTH

false. Since both ‘Snow is white’ is true and grass is green are true, the non-translational T-sentence S is true, even though there is no necessary, or conceptual, connection between grass being green and the sentence ‘Snow is white’ being true. This illustrates the weak sense in which Davidson’s theories of truth specify the truth conditions of sentences. They do not specify what would and must be the case if the sentences they allegedly interpret are to be true; rather, they merely specify for each true sentence some state of affairs that in fact obtains, and for each false sentence some state of affairs that does not.14 Because the T-sentences that are the theorems of his truth theories are so weak, Davidson is forced to consider the possibility that a truth theory for a language might entail true but deviant T-sentences like S rather than standard T-sentences like ST. ST. ‘Snow is white’ is true iff snow is white. What should we think of such theories? Davidson answers that we ought not to think a theory that entails ST any more correct than a theory that entails S instead: provided, of course, we are as sure of the truth of (S) as we are of that of its more celebrated predecessor [ST]. Yet (S) may not encourage the same confidence that a theory that entails it deserves to be called a theory of meaning. The threatened failure of nerve may be counteracted as follows. The grotesqueness of (S) is in itself nothing against a theory of which it is a consequence, provided the theory gives the correct results for every sentence (on the basis of its structure, there being no other way). It is not easy to see how (S) could be party to such an enterprise, but if it were—if, that is, (S) followed from a characterization of the predicate ‘is true’ that led to the invariable pairing of truths with truths and falsehoods with falsehoods— then there would not, I think, be anything essential to the idea of meaning that remained to be captured. What appears to the right of the biconditional in sentences of the form ‘s is true iff p’ when such sentences are consequences of a theory of truth, plays its role in determining the meaning of s 14 A different approach to theories of meaning—so-called possible worlds semantics (briefly discussed in the Epilogue)—aims to provide truth conditions in the stronger sense. Davidson, who shared much of Quine’s skepticism about modal notions, thought it a virtue to get by without them.

THEORIES OF TRUTH AS THEORIES OF MEANING

303

not by pretending synonymy but by adding one more brushstroke to the picture which, taken as a whole, tells what there is to know of the meaning of s; this stroke is added by virtue of the fact that the sentence that replaces ‘p’ is true iff s is.15 The view expressed here is, I take it, that there is nothing more to be known about the meanings of the sentences of a language than is stated by a theory of truth that systematically derives a true T-sentence for each object-language sentence on the basis of its structurally significant parts. Davidson suggests that the requirement that the derivations be systematic, and based on the structure of sentences, may eliminate grotesque theories that issue in true, but deviant, T-sentences like S. The idea, presumably, is that in order to derive S from a compositional account of the structure of the sentence ‘Snow is white’, a theory would have to contain axioms specifying grass as the reference of ‘snow’, and green things as the objects to which the predicate ‘is white’ applies. But with such axioms, one would end up deriving false T-sentences like ‘Snow is grass’ is true iff grass is grass and ‘The chalk is green’ is true iff the chalk is white, in addition to deriving “accidentally true” T-sentences like S. Thus, it is thought that truth theories that are both true and appropriately structural will end up deriving instances of schema T in which the sentences on the right-hand side are close enough paraphrases of the sentence quoted on the left-hand side so that nothing essential to meaning would fail to be grasped by a speaker who knew the totality of that which is stated by the theory. Because of this, Davidson seems to have thought that knowledge of that which is stated by an acceptable theory of truth for a language really would be sufficient to understand the language. This, then, was his initial answer to the challenge How can theories of truth be regarded as theories of meaning? The Incorrectness of This Answer

Before long it became obvious, even to Davidson himself, that he was wrong about this. The requirement that a theory of truth derive its Tsentences on the basis of semantically significant structure does not block the derivation of what one might call non-translational T-sentences like S. To derive such sentences, all one needs to do, given an appropriately “Truth and Meaning,” p. 26. In this passage, Davidson uses lowercase ‘s’ as a schematic letter instances of which are names of sentences, and ‘p’ as a schematic letter instances of which are sentences.

15

304

DAVIDSON ON TRUTH

simple language, is to take the clauses in one’s favorite truth theory that specify the reference of individual words of the language and replace them with other clauses that end up assigning the same referents to the words through the use of very different concepts. If the theorems of the original truth theory are true, then the theorems of the modified truth theory will also be true, despite the fact that knowledge of that which they state clearly will not suffice for understanding a single sentence of the object-language. For example, the original, or standard, truth theory for an extensional fragment of Spanish might include the theorem (3).16 3. ‘Los pantalones son verdes’ is true in Spanish iff the pants are green. Here, the English sentence on the right-hand side is a meaningpreserving paraphrase of the Spanish sentence on the left. Suppose we have a truth theory for an extensional fragment of Spanish that generates (3), plus similar T-sentences for all the other sentences in the fragment. Suppose further that all of the theorems of our truth theory are true, and that each T-sentence is derived from true axioms about the reference of the words contained in the sentence it gives the truth conditions of. Suppose, in short, that we have just the kind of truth theory that Davidson is looking for. In any case of this sort, it is always possible to construct an alternative truth theory that satisfies all of Davidson’s requirements, including the requirement of generating only theorems that are true, but that derives a different T-sentence for the Spanish sentence ‘Los pantalones son verdes’. For example, we can construct that alternative truth theory so that instead of generating (3), it generates (4). 4. ‘Los pantalones son verdes’ is true in Spanish iff the pants are green and all consistent, axiomatizable, first-order theories of arithmetic are incomplete. In addition, we can construct the alternative theory so that a similarly true, but bizarre, T-sentence is derived for every sentence in the fragment of Spanish that the theory applies to. Clearly, an arbitrary truth theory that yields only non-translational T-sentences like (4) cannot be regarded as a correct theory of meaning.17 A fragment of a language is extensional iff all sentences in the fragment are such that substitution in them of terms that refer to the same thing never changes truth value. 17 For an important discussion of this and related points, see J. Foster, “Meaning and Truth Theory” in Truth and Meaning, ed. G. Evans and J. McDowell (Oxford: Clarendon Press, 1976). 16

THEORIES OF TRUTH AS THEORIES OF MEANING

305

When this fact became widely realized, it raised an obvious question. What constraints does one need to impose on truth theories for a language—beyond the truth of their theorems, and the systematic derivation of those theorems from axioms specifying the reference of the words of the language—in order to guarantee that the theories end up generating translational T-sentences like (3), instead of bizarre ones like (4)? For a time, Davidson and his followers were concerned with this question, and they tried to come up with constraints that would narrow down the class of truth theories for a language to those the logical consequences of which always included T-sentences in which the sentence appearing on the right-hand side of the biconditional was a paraphrase of the sentence quoted on the left. We need not go into the details of these suggested constraints, or evaluate how successful or unsuccessful they were. For our purposes, we may simply assume that the problem has been solved, and that we have somehow managed to limit the class of truth theories that are candidates for being theories of meaning to those that generate what we might call translational T-sentences. The point to emphasize is that even if we limit our attention to translational truth theories—i.e., theories that generate T-sentences like (3)—knowledge of that which is stated by these truth theories is not sufficient for understanding the language. Roughly speaking, the reason such knowledge isn’t sufficient is that one might know that which is stated by a translational truth theory without knowing that the theory is translational. For example, someone who knows a translational truth theory that issues (3) as a theorem, might still believe (5), provided that he believes that all consistent, axiomatizable, firstorder theories of arithmetic are incomplete. 5. ‘Los pantalones son verdes’ means in Spanish that the pants are green and all consistent, axiomatizable, first-order theories of arithmetic are incomplete. In addition, such a person could be in the same position regarding every other sentence in the fragment of Spanish we are concerned with. This person would combine true beliefs about truth conditions with false beliefs about meaning, and so would not understand the language. Consequently, knowledge of that which is stated by even the most promising of truth theories does not suffice for understanding a language. This means that Davidson’s original justification for taking theories of truth to be theories of meaning simply doesn’t work.

306

DAVIDSON ON TRUTH

Davidson’s Second Attempt to Justify Taking Theories of Truth to Be Theories of Meaning

By the mid-1970s this problem was widely recognized, and attempts were made to come up with a different criterion that would justify treating certain kinds of truth theories as theories of meaning. I will skip over most of the unnecessary detail, and concentrate on one attempt that both Davidson and some of his main critics ended up accepting. According to this proposal, a translational truth theory T for a language L is a theory of meaning for L by virtue of the fact that one can use it in a certain way to specify knowledge that is sufficient for understanding L. The knowledge that is supposed to be sufficient for understanding L is knowledge of the claim expressed by filling in the dots in the sentence Some truth theory for L, meeting proper constraints, states that . . . with the conjunction of the axioms of T.18 In considering this proposal, it is crucial that we understand the claim that a theory meets proper constraints as guaranteeing, for each sentence S of L, that among the logical consequences of the theory is at least one translational T-sentence—i.e., at least one T-sentence ‘S’ is true in L iff p— in which p expresses the same proposition as, and hence is a paraphrase of, S.19 In the case of a translational T-sentence like (3), the idea is that we should know that which it states, and we should also know that the proposition so stated is expressed by a biconditional sentence the righthand side of which is a paraphrase of, and hence expresses the same proposition as, the Spanish sentence quoted on the left. From this we are supposed to be able to deduce (6). 6. ‘Los pantalones son verdes’ means in Spanish that the pants are green. The idea behind this deduction is roughly as follows: Suppose we know that some truth theory meeting proper constraints states that . . . (where in place of the dots, we fill in the conjunction of the axioms of theory T). Suppose further that from these axioms we can derive (3) as a logical consequence. Then we know that some truth theory states something that entails the proposition that ‘Los pantalones son verdes’ This proposal is discussed by Foster in “Meaning and Truth Theory” and endorsed by Davidson in his “Reply to Foster,” in Truth and Meaning. 19 ‘S’ and ‘p’ are here used as metalinguistic variables. 18

THEORIES OF TRUTH AS THEORIES OF MEANING

307

is true in Spanish iff the pants are green. In addition, since we know that the truth theory meets proper constraints, we know that this proposition is expressed by a translational T-sentence, call it TT. Since TT is translational, we know that the sentence quoted on its left, namely Los pantalones son verdes, expresses the same proposition as the proposition expressed by the sentence on its right. But the proposition expressed by the sentence on its right must be the proposition that the pants are green, because TT is a biconditional that expresses the proposition that the Spanish sentence is true iff the pants are green. Consequently, the Spanish sentence means—i.e., expresses the proposition—that the pants are green. The main stages in this derivation can be expressed as a series of six steps. Step 1. Some translational truth theory states that . . . (where the dots are filled in by the conjunction of the axioms of a truth theory). Step 2. So, some translational truth theory states something that entails that ‘Los pantalones son verdes’ is true in Spanish iff the pants are green. Step 3. So, some translational truth theory has as a logical consequence a T-sentence that states that ‘Los pantalones son verdes’ is true in Spanish iff the pants are green. Step 4. Since the truth theory is translational, the T-sentence mentioned in step 3 is one the right-hand side of which means the same as the sentence ‘Los pantalones son verdes’ mentioned on its left-hand side. Step 5. Since the T-sentence states that ‘Los pantalones son verdes’ is true in Spanish iff the pants are green, both the righthand side of that T-sentence and the Spanish sentence mentioned on the left must mean that the pants are green. Step 6. So, ‘Los pantalones son verdes’ means in Spanish that the pants are green. Two points about this derivation stand out.20 First, the fact that truth theories derive their T-sentences compositionally by paying attention to The problems with this derivation are discussed in more detail in my “Truth, Meaning, and Understanding,” Philosophical Studies 65 (1992).

20

308

DAVIDSON ON TRUTH

the structure of sentences plays no essential role in this derivation. Neither does the truth of the claims made by the T-sentences themselves. All we need for the derivation is to be provided with sentences ‘S’ is F iff p, where p is guaranteed to be a translation of S. Beyond this, it is not important what the predicate F used in the theory is, and it is not important how the sentences are derived, what they say, or even whether they are true. One could, in fact, systematically replace the predicate is true in a translational truth theory with the predicate is false without affecting the derivation. Thus, the role of truth theories in specifying the knowledge that is supposed to be sufficient for understanding sentences is, on this proposal, essentially heuristic, and in principle dispensable. To me, this suggests that the criterion Davidson here employs for counting something as a theory of meaning is too weak. If truth theories satisfy the criterion, so will a variety of other sorts of theories that, intuitively, should not be counted as theories of meaning.21 The second point about the derivation is even more fundamental. The derivation is fallacious, and so Davidson’s truth theories do not satisfy his revised criterion for counting as theories of meaning. The main problem involves step 4. At step 3 it is supposed to be established that a certain T-sentence is a consequence of a translational truth theory—that is, a truth theory that includes among its logical consequences T-sentences the right-hand sides of which are translations of the sentences quoted on the left. In order to get to step 4, we need to establish that the particular T-sentence which states that ‘Los pantalones son verdes’ is true in Spanish iff the pants are green is translational in this sense. But the proposal says nothing to guarantee this. The fact that the truth theory is translational ensures that one of the T-sentences it generates for the Spanish sentence ‘Los pantalones son verdes’ is translational, but it does not tell us which T-sentence that is. If we knew that for each Spanish sentence in our fragment, the truth theory would allow the derivation of exactly one T-sentence, then we 21 There is a stronger and a weaker version of this point. The weak version, which is what I intend, is that even if a theory T were such that knowing that which is expressed by Some theory for L, meeting proper constraints, states that T were sufficient for understanding L, this result would not be sufficient to show that T should be regarded as a theory of meaning. This, I believe, is clearly correct. The stronger version of the point is that even if the knowledge in question were sufficient for understanding L, this result would not be sufficient to show that Some theory for L, meeting proper constraints, states that T should itself properly be regarded as a theory of meaning for L. Although I believe this also to be correct, it may, potentially, be more controversial.

THEORIES OF TRUTH AS THEORIES OF MEANING

309

could be sure that the T-sentence we are interested in was translational, and we could move directly to step 4. However, we cannot be sure of this, since, in general, truth theories provide many T-sentences for each sentence in the fragment. For example, suppose that some T-sentence ‘S’ is true iff P is a logical consequence of a truth theory. Suppose further that Q is any logical consequence of P, or that Q is any logical consequence of the truth theory itself, or that Q is any logical consequence of the conjunction of P and the truth theory. Then the T-sentence ‘S’ is true iff P & Q is also a logical consequence of the truth theory. However, for many choices of Q, the conjunction of P and Q will not mean the same as P. Hence, it cannot be that all these T-sentences are translational, even though they are all logical consequences of the theory. But this means that we have no guarantee that the particular T-sentence which states that ‘Los pantalones son verdes’ is true in Spanish iff the pants are green is translational. As a result, we cannot move to step 4, and the derivation is blocked.

Conclusions For these reasons, we cannot accept Davidson’s final proposal for justifying the claim that truth theories of the sort he favors qualify as theories of meaning. This leaves us with no justification for the claim. There are two main possibilities to consider. Either there is no satisfactory justification for the claim, and it should be rejected, or the claim is correct, and has some other justification that we haven’t considered. A natural thought is that such a justification might be found by restricting one’s attention to what we might call canonical T-sentences generated by a truth theory—where these include not all T-sentences that are logical consequences of the theory, but only those that are derived in some restricted and specified way. The idea would be to specify a pattern of derivation which, when applied to the right sort of truth theory, would generate only translational T-sentences. If such a specification were made precise and put in place, the reasoning from step 1 to step 6 would no longer be blocked. Supposing this could be done, what, really, would we have gained? Questions of technical feasibility aside, one wonders what the rationale for such a move might be. More and more, it begins to appear that the actual claims made by truth theories themselves are unimportant, and that their role in theories of meaning

310

DAVIDSON ON TRUTH

is nothing more than heuristic. All one needs to employ reasoning of the general sort illustrated by steps 1–6 above is some mechanism that pairs sentences in the language being investigated with paraphrases of them in the language of the theory. If the mechanism can be formulated precisely and labeled translational, it can be used in the derivation of theorems stating the meanings of sentences. But if there is one such mechanism, presumably there are many. Hence, even if truth theories supplemented with specifications of canonical derivations prove to be among them, presumably we need some further criterion for selecting genuine theories of meaning from the available candidates. One thought might be that the right theory of meaning is the one that normal speakers of the language unconsciously employ in producing and understanding utterances. But to claim that Davidsonian truth theories satisfy this criterion would be to engage in unfounded psychological speculation of the sort that Davidson himself would, I think, rightly regard as extravagant. Hence, the claim that truth theories of the type he advocates can properly be regarded as theories of meaning remains unproven. Although Davidsonians have continued to try to provide a persuasive rationale for this claim, it continues to be controversial. Consequently, we will leave it as an unresolved issue.22 There is, however, one further point to be made before we leave this topic. Earlier I mentioned that there is a technical project, the Davidsonian program, that continues to this day. The project is that of trying to write Davidsonian truth theories for larger and larger fragments of natural language. Much of the work that has been done takes the goal to be construction of finitely axiomatizable, compositional truth theories, all of the theorems of which are true. This is a much weaker goal than the goal of producing translational truth theories. Moreover, when one looks at the work that has been produced, one finds that a great deal of it has involved analyses that are not translational.23 As a result, there is something of a disconnect in the field. When pursuing the technical project, many researchers tacitly presuppose Davidson’s original statement of the problem—which requires only the 22 Strategies for defending the idea that Davidsonian theories of truth can play the role of theories of meaning are to be found in Martin Davies, Meaning, Quantification, and Necessity (New York: Routledge, 1981), and in Richard Larson and Gabriel Segal, Knowledge of Meaning (Cambridge: MIT Press, 1995). 23 An instance of this problem is discussed in my “Truth and Meaning: The Role of Truth in the Semantics of Propositional Attitude Ascriptions,” in Truth, Rationality, Cognition, and Music: Proceedings of the Seventh International Colloquium on Cognitive Science, ed. Kepa Korta and Jesus M. Larrazabal (Dordrecht: Kluwer, 2003).

THEORIES OF TRUTH AS THEORIES OF MEANING

311

production of true truth theories. However, when asked to confront the philosophical question of what justifies one’s taking a mere theory of truth to be a theory of meaning, Davidson and others have moved to the strong requirement of translational theories. But if that, or something like it, is going to be the requirement, then there has been very little technical progress toward meeting it.

CHAPTER 13 TRUTH, INTERPRETATION, AND THE ALLEGED UNINTELLIGIBILITY OF ALTERNATIVE CONCEPTUAL SCHEMES chapter outline 1. The Structure of Davidson’s Argument against Alternative Conceptual Schemes Translation into English requires agreement with our world view. So, if there are alternative conceptual schemes, the languages in which they are expressed must not be translatable into English. But such languages are impossible. Since sentences in any conceivable language must have truth conditions, our notion of truth must be applicable to them; but it cannot intelligibly be applied to sentences of a language that is not translatable into English 2. Translation and Agreement Why correct theories of interpretation assign meanings to sentences that make the sentences speakers accept true, except when we can give plausible explanations of why they should be mistaken; a telltale misconception about truth in Davidson’s characterization of the evidence for theories of interpretation and translation 3. Truth, Intelligibility, and Untranslatable Sentences A fundamental confusion about the notion of truth required by Davidson’s program, and its relation to those definable by Tarskian techniques; why our ordinary notion of truth can be applied to sentences that are not translatable into English 4. Summary and Final Assessment

The Structure of Davidson’s Argument against Alternative Conceptual Schemes In the previous chapter, we scrutinized Davidson’s attempts to justify the claim that certain theories of truth can play the role of theories of meaning. We now turn to a different but related topic, namely, his argument

ALTERNATIVE CONCEPTUAL SCHEMES

313

in “On the Very Idea of a Conceptual Scheme” that, given our conception of truth and its role in providing interpretations of the speech of other linguistic communities, we have no alternative but to reject the very possibility that there could be speakers whose views about the world are profoundly and systematically at variance with our own.1 Davidson’s argument is based on two things: his conception of what makes such a theory of interpretation correct, and his view of the concept of truth employed in such theories. In chapter 12, I stated his view about how we determine a proposed theory of interpretation to be correct essentially as follows. Theories of interpretation are truth theories. We empirically verify that a proposed theory of truth for the language of a given group of speakers is correct by comparing the conditions, or situations, in which speakers hold particular sentences of their language to be true with the truth conditions assigned to those sentences by the truth theory. All other things being equal, the correct truth theory for the language is the theory according to which the conditions in which the speakers actually hold sentences to be true most closely matches the conditions in which the combination of the theory with our background theory of the world predicts the sentences to be true. Roughly put, the correct theory for the language is the theory according to which speakers of the language turn out to be truth tellers more frequently than on any other interpretation of the language. Davidson takes it to follow from this criterion that our ability to interpret the speech of another group involves viewing them as sharing with us all sorts of what we take to be true beliefs. Roughly, he takes it to involve viewing them as sharing with us more true beliefs than any alternative interpretation would attribute to them (provided that all the interpretations under consideration are appropriately compositional, and conform to routine principles of simplicity and good scientific methodology). The relevance of this for his argument is the following: If other conceptual schemes (fundamental ways of looking at the world) radically at variance with our own are possible, then—supposing that they are capable of being expressed in some languages—either those Donald Davidson, “On the Very Idea of a Conceptual Scheme,” Proceedings and Addresses of the American Philosophical Association 47 (1974); reprinted in Inquiries into Truth and Interpretation; citations will be to that volume.

1

314

DAVIDSON ON TRUTH

languages are translatable into our language, or they aren’t. If they are translatable into ours, then there must be massive agreement between us and the speakers of those languages, in which case they can’t have conceptual schemes radically at variance with ours. So, if there are conceptual schemes radically at variance with our own, then we won’t be able to interpret or translate the languages of those who share them. Let L be such a language. Like all languages rich enough to express a systematic view of the world, L will contain multitudes of true sentences—many of which will be translatable into ours only if L as a whole is (for the most part) translatable. It follows that, since L is not so translatable, it must contain many truths that we cannot translate. But, Davidson maintains, this idea is incoherent. He argues that our grasp of the notion of truth, as applied to sentences of another language, is tied to our ability to translate that language into our own. Thus, he thinks, a proper understanding of the notion of truth reveals that it is incoherent to suppose that there are many true sentences of any language that is not (for the most part) translatable into ours. He therefore concludes that the idea that it is possible for rational agents to have systems of fundamental beliefs—alternative conceptual schemes— radically at variance with our own is incoherent. This argument has two basic steps. Step 1: Translation into English involves fundamental agreement with our world view. So if something can be translated into English, it cannot be the language of speakers with a radically different conceptual scheme. Step 2: If it were possible for speakers to have a conceptual scheme radically different from ours, which they expressed in a language that couldn’t (for the most part) be translated into English, then we would have to be able to make sense of the idea that many of their sentences possess truth conditions even though we have no idea what those sentences mean, and hence no idea either of how to translate them, or what would have to be established in order to show that they are true. But, Davidson thinks, we have no conception of what the word true might mean when applied to a situation like this. Thus, he believes, our grasp of the notion of truth makes it incoherent to suppose that there could be a conceptual scheme so different from ours as to make translation into our language impossible. Since he also dismisses the idea that there could be a conceptual scheme which couldn’t be expressed in any language at all, he concludes that the very notion of conceptual schemes or systems of beliefs fundamentally different from ours makes no sense. Both steps in the argument are interesting, and both raise serious questions. We

ALTERNATIVE CONCEPTUAL SCHEMES

315

will start with the first step—where it is supposed to be shown that translation requires massive agreement with us.

Translation and Agreement The Argument

Earlier, I said that Davidson’s criterion for empirically verifying theories of interpretation requires such theories to maximize the degree to which speakers of other languages hold beliefs that agree with ours. That is a little too strong. It would be more accurate to say that his criterion requires us to accept a theory of interpretation that leads us to ascribe what we take to be clearly false beliefs to speakers only when we can give a reasonable explanation of why they hold these false beliefs. It is worthwhile quoting a few passages from Davidson about this. We have still to say what evidence is available to an interpreter— evidence, we now see, that T-sentences are true. The evidence cannot consist in detailed descriptions of the speaker’s beliefs and intentions, since attributions of attitudes, at least where subtlety is required, demand a theory that must rest upon much the same evidence as interpretation. The interdependence of belief and meaning is evident in this way: a speaker holds a sentence to be true because of what the sentence (in his language) means, and because of what he believes. Knowing that he holds the sentence to be true, and knowing the meaning, we can infer his belief; given enough information about his beliefs, we could perhaps infer the meaning. But radical interpretation should rest on evidence that does not assume knowledge of meanings or detailed knowledge of beliefs. A good place to begin is with the attitude of holding a sentence true, of accepting it as true. This is, of course, a belief, but it is a single attitude applicable to all sentences, and so does not ask us to be able to make finely discriminated distinctions among beliefs. It is an attitude an interpreter may plausibly be taken to be able to identify before he can interpret, since he may know that a person intends to express a truth in uttering a sentence without having any idea what truth. . . . Suppose, then, that the evidence available is just that speakers of the language to be interpreted hold various sentences to be true at certain times and under specified circumstances. How can

316

DAVIDSON ON TRUTH

this evidence be used to support a theory of truth? On the one hand, we have T-sentences, in the form: (T) ‘Es regnet’ is true-in-German when spoken by x at time t if and only if it is raining near x at time t. On the other hand, we have evidence, in the form: (E) Kurt belongs to the German speech community and Kurt holds true ‘Es regnet’ on Saturday at noon and it is raining near Kurt on Saturday at noon. We should, I think, consider (E) as evidence that (T) is true. Since (T) is a universally quantified conditional, the first step would be to gather more evidence to support the claim that: (GE) (x)(t) (if x belongs to the German speech community then (x holds true ‘Es regnet’ at t if and only if it is raining near x at t)).2 Here Davidson gives an example of the kind of match he is looking for between predictions about truth conditions made by the truth theory, and observed conditions in which speakers of the language hold various sentences to be true.3 The match is illustrated by the relationship between principles (T), (E), and (GE). (T)

‘Es regnet’ is true-in-German when spoken by x at time t iff it is raining near x at t.

(E)

Kurt belongs to the German speech community and Kurt holds true ‘Es regnet’ on Saturday at noon and it is raining near Kurt on Saturday at noon.

(GE) For all x and for all t, if x belongs to the German speech community then x holds true ‘Es regnet’ at t iff it is raining near x at t. Davidson continues by considering an obvious objection. “Radical Interpretation,” pp. 134–135. It may be worth noting that Davidson’s notion of accepting or holding a sentence true is somewhat more theoretical than he gives the impression of recognizing. When one takes seriously Grice’s observation that the information conveyed by an utterance of the sentence may include implicatures that are not part of the meaning or truth conditions of the sentence, once realizes that accepting a sentence, in the ordinary sense, does not always involve believing it to be true (or vice versa). In the discussion that follows, I abstract away from this apparent shortcoming in Davidson’s methodology. 2 3

ALTERNATIVE CONCEPTUAL SCHEMES

317

The obvious objection is that Kurt, or anyone else, may be wrong about whether it is raining near him. And this is of course a reason for not taking (E) as conclusive evidence for (GE) or for (T); and a reason not to expect generalizations like (GE) to be more than generally true. The method is rather one of getting the best fit. We want a theory that satisfies the formal constraints on a theory of truth, and that maximizes agreement, in the sense of making Kurt (and others) right, as far as we can tell, as often as possible. The concept of maximization cannot be taken literally here, since sentences are infinite in number, and anyway once the theory begins to take shape it makes sense to accept intelligible error and to make allowance for the relative likelihood of various kinds of mistake. . . . This method is intended to solve the problem of the interdependence of belief and meaning by holding belief constant as far as possible while solving for meaning. This is accomplished by assigning truth conditions to alien sentences that make native speakers right when plausibly possible, according, of course, to our own view of what is right.4 The lesson I draw from this is that Davidson’s conception of the empirical evidence that would verify a theory of interpretation directs us to assign meanings to sentences which would make the sentences the speakers accept true, according to our beliefs, as far as possible, except for cases in which we can give plausible explanations of why they should be ignorant or mistaken. Our task is to understand why this principle, or something close to it, seems reasonable. Davidson insists that the principle reflects the interdependence of meaning and belief. Confronted with a new culture with agents speaking a language different from ours, we need to come up with a way of interpreting not just the meanings of their words, but also the contents of their beliefs. These tasks are inseparable from one another. We use the same evidence for both tasks, and the decisions we make about one have direct implications for the other. Davidson addresses the problem of simultaneously constructing a workable theory of meaning and an acceptable theory of belief for a new culture in the following passage: The way this problem is solved is best appreciated from undramatic examples. If you see a ketch sailing by and your companion 4

“Radical Interpretation,” pp. 136–137, my emphasis.

318

DAVIDSON ON TRUTH

says, ‘Look at that handsome yawl,’ you may be faced with a problem of interpretation. One natural possibility is that your friend has mistaken a ketch for a yawl, and has formed a false belief. But if his vision is good and his line of sight favorable it is even more plausible that he does not use the word ‘yawl’ quite as you do, and has made no mistake at all about the position of the jigger on the passing yacht. We do this sort of off the cuff interpretation all the time, deciding in favor of reinterpretation of words in order to preserve a reasonable theory of belief. As philosophers we are peculiarly tolerant of systematic malapropism, and practiced at interpreting the result. The process is that of constructing a viable theory of belief and meaning from sentences held true. . . . But the principles involved must be the same in less trivial cases. What matters is this: if all we know is what sentences a speaker holds true, and we cannot assume that his language is our own, then we cannot take even a first step towards interpretation without knowing or assuming a great deal about the speaker’s beliefs. Since knowledge of beliefs comes only with the ability to interpret words, the only possibility at the start is to assume general agreement on beliefs. We get a first approximation to a finished theory by assigning to sentences of a speaker conditions of truth that actually obtain (in our own opinion) just when the speaker holds those sentences true. The guiding policy is to do this as far as possible, subject to considerations of simplicity, hunches about the effects of social conditioning, and of course our commonsense, or scientific, knowledge of explicable error.5 There is a modest point here that is simple and correct. (Note how highly qualified the thesis in the passage is.) Suppose we have evidence of sentences held true by speakers. If we are too ready to ascribe meanings to those sentences that make them false, we run the risk of being unable to give any plausible theory of what factors lead to speakers’ false beliefs. The more obvious the falsehoods attributed by a certain interpretation, and the more trouble finding an explanation for error, the more likely it becomes that we have interpreted the foreigner incorrectly. This doesn’t lead to any hard and fast rule about how much false belief we can attribute to the foreigner. However, it is reasonable to think that the need to view the speaker as a rational agent imposes 5

“On the Very Idea of a Conceptual Scheme,” p. 196.

ALTERNATIVE CONCEPTUAL SCHEMES

319

some pressure to interpret him in a way that makes his beliefs reasonable, and, to that extent, as having at least some likelihood of being true.6 This defensible, and fairly weak, principle is one that Davidson can, I think, safely assert. One of its virtues is that it leaves room for interpreting speakers of other languages as having a large number of what we regard as false beliefs, provided we can give an account of why they have those beliefs. In practice, this may often mean that we view the mistakes of other speakers against a background that includes a certain amount of agreement with us. Davidson emphasizes the amount of agreement involved. He thinks that the need to view other speakers as having reasonable beliefs, if we are to be able to formulate and verify any theory of interpretation of their language, ensures that we must view them as expressing truths, and thereby agreeing with us, in an overwhelmingly large range of cases. I see nothing in his discussion that establishes this, and I believe he exaggerates the amount of truth telling we have to attribute to speakers in order to find their beliefs and other attitudes explainable. Certainly, we can accept the principle that our ability to interpret the speech of another group requires some agreement in belief between them and us. How much is an open question. Moreover, the underlying idea doesn’t have as much to do with agreement as one might initially think. What is important is not that interpretees believe what interpreters believe, but that interpreters can give plausible explanations of the interpretees’ beliefs. Since it is often difficult to explain beliefs in obvious falsehoods, one should expect to find some agreement. But there is no general rule about this; how much agreement is reasonable to expect is something that, presumably, has to be decided on a case-by-case basis. In light of this, it is hard to draw any strong conclusions about how different from us speakers can be, while still being interpretable by us. The short answer, I think, is that they can be pretty different. Different enough so that they don’t “share our conceptual scheme”? Since the idea of sharing a conceptual scheme is vague, the question may 6 The plausibility of this principle is not marred by the imperfect example involving yawls and ketches that Davidson uses to motivate it. One can understand the word yawl in the sense of satisfying minimum standards for linguistic competence, and still not be able to distinguish yawls from ketches or other sailboats. In Davidson’s example, the speaker most likely does falsely believe of some ketch that it is a yawl, no matter how good his vision and how closely he attends to the details. Since he is a speaker of the common language, the word yawl, as he uses it, has its standard meaning and reference, even though on some occasions what he, the speaker, refers to when he uses the word is not a yawl, but a ketch. (Those who, wrongly in my opinion, reject the notion of a common language may, like Davidson, fail to be moved by this observation.)

320

DAVIDSON ON TRUTH

have no determinate answer. But I do think that there could be very substantial differences between us and the speakers of another language that are relatively independent of how much we agree or disagree in belief. Here is an example. Consider the beliefs of speakers whose natural and unreflective way of looking at the world does not involve the recognition of objects that persist through time, but does involve the recognition of temporal stages of objects that bear proximity relations to one another. Do such speakers have a conceptual scheme different from ours? Is it conceivable that we could interpret them? So long as speakers in the alien culture are willing to grant that there are certain kinds of sums of temporal stages over time that could be regarded as enduring objects—as strange and unnatural as those objects seem to them (“Of course such objects exist, but why would anyone be interested in them?”) —and so long as, given our enduring objects as starting points, together with moments of time, we are willing to grant the existence of temporal stages—as strange and unnatural as they seem to us—there may be no fundamental disagreement between the two cultures about what kinds of objects exist. Neither culture need view the central claims made by the other as false. Rather, each may view the entities that the other takes as most worthy of attention as strange, unnatural, and derivative; and, because of this, each might find it difficult to explain or understand how the other could take such unnatural entities as basic. But speakers of the two cultures need not disagree over the kinds of objects that exist, or even over their properties. Thus, the need to assure agreement in order to construct a Davidsonian theory of interpretation does not have any significant bearing on the possibility that such an alien culture might exist, or even that we could interpret them. Of course, one might wonder what facts would determine whether an interpretation of members of the alien culture as talking about temporal stages was correct, as opposed to an interpretation that saw them as talking about enduring objects. This is the Quinean problem of the indeterminacy of translation and the inscrutability of reference recurring for Davidsonian theories of interpretation. Though Davidson himself would presumably classify the case as one involving genuine indeterminacy, so far as I can tell, nothing in his discussion either establishes or requires this; hence nothing precludes the idea that a temporal-stage culture could exist and be interpretable by us, despite the fact that their “conceptual scheme” seems rather different from ours.

ALTERNATIVE CONCEPTUAL SCHEMES

321

A Telltale Misconception about Truth

At this point I want to digress for a moment about a matter of detail that I think is indicative of a larger confusion on Davidson’s part. The matter concerns his discussion of the evidence for theories of interpretation and translation. As we saw, in that discussion he illustrated his view with the examples (T), (E), and (GE). (T)

‘Es regnet’ is true-in-German when spoken by x at time t iff it is raining near x at t.

(E)

Kurt belongs to the German speech community and Kurt holds true ‘Es regnet’ on Saturday at noon and it is raining near Kurt on Saturday at noon.

(GE) For all x and for all t, if x belongs to the German speech community then x holds true ‘Es regnet’ at t iff it is raining near x at t. Notice the truth predicate in (T). Davidson hyphenates it—true-inGerman. The result of this is to turn a phrase into something that is supposed to function as a single word. Semantically, this has the effect of turning talk about a two-place relation that a sentence S bears to a language L, when S is a true sentence in, or of, L, into talk about a class of one-place predicates true-in-L1, true-in-L2, true-in-L3, and so on. There are three points to make about this: First, when our ordinary word true is used in English to talk about sentences—as opposed to talking about what sentences say or express—it is used to express a relation. When we say, that S is a true sentence of English, Japanese, Hindi, etc., we are using a construction, ——— is a true sentence of ——— that is genuinely relational in the sense that both of its argument places can be filled by arbitrary singular terms—including names, definite descriptions, or bound variables. We can talk about true sentences of a language spoken by a billion people, true sentences of a language no one remembers, true sentences of the native language of the first U.S. president of the twenty-second century, whoever he or she may be, and so on. The only way to give a satisfactory account of this is to see is a true sentence of as being relational—on a par with is north of, is a father of, and so on. Second, the origin in philosophy of the hyphenated truth-in-L locution seems to be Alfred Tarski’s use of it in formulating his formally defined truth predicates. His use of the hyphenated form

322

DAVIDSON ON TRUTH

was motivated by his view that the unrestrictedness of our ordinary truth predicate—the fact that it can apply to sentences of any language, including a language containing the very truth predicate itself—is the source of the Liar paradox, which, he believed, shows that our ordinary truth predicate is incoherent and must be replaced by a class of restricted truth predicates, each applying to a single language (free of its own semantic predicates). These restricted truth predicates for different languages are conceptually independent of one another in a way that genuinely relational predicates are not. Tarski’s definitions of these predicates are forced to mention every expression in the object language to which they apply, while making use of every concept expressed in that language, plus logical and set theoretic notions.7 It follows that the contents of the truth predicates defined for different languages have to be different—since any two different languages will always either have different expressions, or use the same expressions with different meanings. So for the Tarskian, the predicate true-in-L1 has to mean something very different from true-in-L2. Grasping what one expresses has to be different from, and largely independent of, grasping what the other expresses, and the two predicates will standardly apply to different classes of things. The third and final point to notice about Davidson’s statement of claims like (T) is that his use of the Tarski-inspired hyphenated notation suggests that he thinks that his theories employ notions of truth analogous to Tarski’s defined notions. This, I believe, betrays a serious confusion. If knowledge of truth conditions is to contribute to how we can come to understand a language, then the notion of truth involved can’t be Tarski’s.8 Certainly, different notions of truth should not be required by theories of interpretation for different languages. The whole point of a theory of interpretation is to provide us with a way of coming to understand an arbitrary language that we may know little or nothing about initially. If we are in this position, we won’t know which restricted truth predicate to pick out before we know which language we are dealing with, and what its semantic properties are. But that is what a theory of interpretation is supposed to tell us. To avoid this difficulty, we need already to possess some general notion of truth that can be applied to any language, in order to frame our interpretative hypotheses. Surely that is just our ordinary, relational notion of truth. But if that is so, then Davidson shouldn’t be 7 8

See chapters 3 and 4 of my Understanding Truth. See Understanding Truth, chapter 4, pp. 102–107.

ALTERNATIVE CONCEPTUAL SCHEMES

323

hyphenating it. In itself this confusion can, of course, be easily rectified—just take out the hyphens in (T). However, there is reason to think that the error is connected to a larger confusion that has an impact on the second step of Davidson’s argument about alternative conceptual schemes, to which we now turn.

Truth, Intelligibility, and Untranslatable Sentences The second step in Davidson’s argument attempts to rule out the possibility of there being a culture whose beliefs are so different from ours that we can’t translate the words of its speakers. The idea is to show that we can’t coherently characterize agents as having beliefs, if we can’t translate their language into ours. Well, let’s see. Imagine that a group of aliens from outer space lands on earth, and we try to interact with them. We notice that they produce sounds that seem to allow them to communicate with one another. We further notice that they appear to gain information from looking at screens with symbols on them, and that doing what appear to be calculations with their symbols, and exchanging the results with others of their kind, seems to allow them to coordinate their activities over long distances, and to interact with natural forces in ways that are remarkably successful. In spite of all this, their symbols remain inexplicable to us. They seem to have a language, but it appears to be one that we can’t interpret. Not being able to interpret their language, we can’t be sure whether they deny most of our basic beliefs about the world, or whether they just have a system of concepts so different from ours that we can’t match up the two. Is this sort of scenario possible or not? On the face of it, it doesn’t seem to be out of the question. Nevertheless, Davidson gives an argument that is designed to rule it out. Our attempt to characterize languages or conceptual schemes in terms of the notion of fitting some entity has come down, then, to the simple thought that something is an acceptable scheme or theory if it is true. Perhaps we better say largely true in order to allow sharers of a scheme to differ on details. And the criterion of a conceptual scheme different from our own now becomes: largely true but not translatable. The question whether this is a useful criterion is just the question how well we understand the notion of truth, as applied to language, independent of the notion

324

DAVIDSON ON TRUTH

of translation. The answer is, I think, that we do not understand it independently at all.9 Davidson thinks that any conceptual scheme must be one in which the sentences that express its tenets are largely true. Again, the issue of how much of it must be true arises, and it is vague what answer should be given. Let us be cautious, and simply assume that some substantial amount of it must be true. Thus, what we need to make sense of is the possibility of a conceptual scheme some significant parts of which are true, but the sentences of which are (for the most part) untranslatable into English. Davidson thinks that he can expose this idea as incoherent. We recognize sentences like ‘ “Snow is white” is true if and only if snow is white’ to be trivially true. Yet the totality of such English sentences uniquely determines the extension of the concept of truth for English. Tarski generalized this observation and made it a test of theories of truth: according to Tarski’s Convention T, a satisfactory theory of truth for a language L must entail, for every sentence s of L, a theorem of the form ‘s is true if and only if p’ where ‘s’ is replaced by a description of s and ‘p’ by s itself if L is English, and by a translation of s into English if L is not English. This isn’t, of course, a definition of truth, and it doesn’t hint that there is a single definition or theory that applies to languages generally. Nevertheless, Convention T suggests, though it cannot state, an important feature common to all the specialized concepts of truth. It succeeds in doing this by making essential use of the notion of translation into a language we know. Since Convention T embodies our best intuition as to how the concept of truth is used, there does not seem to be much hope for a test that a conceptual scheme is radically different from ours if that test depends on the assumption that we can divorce the notion of truth from that of translation.10 Davidson’s point seems to be that we understand the notion of truth as it applies to English sentences by virtue of our accepting instances of schema T, like ‘Snow is white’ is true iff snow is white. He seems also to suggest that we extend the concept of truth to include sentences of another language L by coming up with translations of sentences of L into English, and then using those translations to construct instances 9

“On the Very Idea of a Conceptual Scheme,” p. 194. Ibid., pp. 194–195.

10

ALTERNATIVE CONCEPTUAL SCHEMES

325

of Tarski’s schema T to fix the specialized interpretation of the truth predicate that applies to sentences of L. Having gotten this far, he observes that this makes no room for the idea of true sentences not translatable into English. There is something truly bizarre about this idea. Suppose that speakers hundreds or thousands of years ago had come up with this argument in support of the view that it is conceptually incoherent to suppose that future scientists would have a scientific world view, expressed in a specialized language most sentences of which could not (for the most part) be translated into the language of earlier times. Had they thought this, they would have been wrong. Similarly, if we were now to argue that it is conceptually incoherent to suppose that future scientists might invent theories so radically different from our present ones that we couldn’t translate the new theories into our present language, we would probably be wrong. But that seems to be the direction in which Davidson’s argument leads. So it looks as if there must be something wrong with the argument. We need to find out what. Davidson mentions Tarski’s Convention T. He says that “according to Tarski’s Convention T, a satisfactory theory of truth for a language L must entail, for every sentence s of L, a theorem of the form ‘s is true if and only if p’ where ‘s’ is replaced by a description of s and ‘p’ by s itself if L is English, and by a translation of s into English if L is not English” (my emphasis). There are two reasons why this remark is out of place. First, Tarski’s task, unlike Davidson’s, was not to give theories of interpretation employing our ordinary, antecedently grasped notion of truth. Instead, Tarski’s task was to show how, given a formal language L, we could define, in a metalanguage M, a special restricted truth predicate that could replace our ordinary notion of truth, when it is applied to sentences of L. The special role played by Convention T in Tarski’s project has no analogue for the use of truth in Davidson’s program of constructing theories of interpretation. Tarski used Convention T to determine whether the notion one had defined in the metalanguage really applied to all and only the true sentences of the object language. If one’s definition does logically entail for each sentence s of L an instance of schema T in which the metalanguage sentence on the right-hand side is a translation of s, then one can be sure that one has defined a predicate that applies to all and only the truths of L. The reason for this is that it is an essential feature of our ordinary notion of truth that whenever P means the same as some (non-indexical) sentence S of a language L, the claim ‘S’ is a true sentence of L iff P will

326

DAVIDSON ON TRUTH

be true. Thus, if T-in-L is a restricted truth predicate for L introduced by a formal Tarskian definition that satisfies Convention T, then when P means the same as some sentence S of L, we will have both ‘S’ is a true sentence of L iff P and ‘S’ is T-in-L iff P, from which it follows that ‘S’ is a true sentence of L iff ‘S’ is T-in-L. Because of this, one can be sure that T-in-L applies to all and only the true sentences of L, and hence, for Tarski’s purposes, that it can replace the ordinary notion of truth when talking about sentences of L.11 That was Tarski. By contrast, Davidsonian theories of interpretation are not intended to define special restricted notions of truth that apply to languages other than English. Rather, they employ our already understood, ordinary notion of truth to interpret those languages. The fact that we couldn’t construct a Tarskian definition, and verify that it applied to all and only the truths of the language, if we had no way of translating the language into one we understood, shows at most that we couldn’t replace the ordinary notion with a Tarskian one, restricted to that language. It doesn’t show that the ordinary one fails to apply to that language, and it doesn’t show that we don’t know what it would mean to say of sentences we can’t translate that they are true. We don’t use Convention T to figure out what true means when it is applied either to our own language or to other languages. On the contrary, it is because we already know what true means that we know in advance that any instance of schema T in which the sentence used on the right paraphrases the sentence mentioned on the left must be true. This brings us to Davidson’s second misstep in the passage— namely, misstating the content of Convention T, as Tarski understood it. According to Tarski, a proposed definition in the metalanguage M of a predicate T satisfies Convention T, and so counts as a definition of truth for L, iff the definition logically entails s is T-in-L iff p for each sentence s of L, where p is some translation of s into the metalanguage in which the definition is constructed. Note, what is required is a translation into the metalanguage, not necessarily into English. Tarski’s method of constructing definitions of restricted truth predicates has the consequence that one can construct a formal definition of truth for a language L in a metalanguage M, and show that it satisfies Convention T, only if L is translatable into the metalanguage. There is no requirement in Tarski that there be a single metalanguage that could be used to define truth for all object languages. 11

See my Understanding Truth, chapter 3.

ALTERNATIVE CONCEPTUAL SCHEMES

327

Suppose, however, that one thought that English was such a universal metalanguage.12 One might be tempted to think that our word true in English is just the logical sum of all the Tarski-style truth predicates for different languages that we can define in English and show to satisfy Convention T. According to this view, the notion of truth we express in English is the combination of the notions true-sentence-of-English, truesentence-of-Spanish, true-sentence-of-Italian, and so on, where all of these restricted notions of truth are given Tarski-like definitions in English that satisfy Convention T. If one thought this, then one would find the idea that there was a language containing many true sentences that could not be translated into English incoherent. Davidson himself may have believed something along these lines when he wrote “On the Very Idea of a Conceptual Scheme.” But if he did, he should have known better. He, more than anyone else, should have realized that the notion of truth we express in English cannot be just the sum of Tarski-style truth predicates definable in English. Take a T-sentence like ‘Los pantalones son verdes’ is true in Spanish iff the pants are green. If the truth predicate in this sentence is one given by a Tarskian definition that logically entails this instance of schema T, then, presumably, the T-sentence is a trivially obvious, apriori, and necessary truth. It is so because it is logically entailed by a definition, which is all of these things. However, the T-sentence can’t be necessary, since it is a contingent fact of Spanish that those words have the meaning and truth conditions they do; and it can’t be apriori, since, if it were, it would be knowable without appeal to any empirical evidence. But one can’t learn the meanings or truth conditions of Spanish sentences without being given information about how they are used by Spanish speakers. It is central to Davidson’s view that statements about truth conditions provide information about meaning. Such information is empirical. To discover what words or sentences mean in different languages, you must gather empirical information about the language and Tarski himself, notoriously, believed something along these lines. Thus, in section 1 of “The Concept of Truth in Formalized Languages” (page 164), he says, “a characteristic feature of colloquial language (in contrast to various scientific languages) is its universality. It would not be in harmony with the spirit of this language if in some other language a word occurred which could not be translated into it; it could be claimed that ‘if we can speak meaningfully about anything at all, we can also speak about it in colloquial language’.” However, (i) it is not fully clear what Tarski meant by this (surely new words are invented all the time that are not definable in terms of the old words); (ii) whatever he may have meant, it is completely separable from his use of Convention T to state adequacy conditions on his formal definitions of truth; and (iii) he did not go down the road of viewing our ordinary word true as being, in effect, the sum of all the restricted notions definable using English as a metalanguage. 12

328

DAVIDSON ON TRUTH

its speakers. Thus, if knowledge of truth conditions is to provide information about meaning in the way that Davidson assumes, then the truth predicate cannot be defined in such a way as to have the T-sentences as logical consequences. Hence, our ordinary truth predicate cannot be anything like the sum of Tarskian truth predicates definable in English. I will illustrate this point about Tarskian truth predicates in more detail in a moment. But the central point of which this is an illustration can be grasped at once. If, as Davidson supposes, we are going to use instances of schema T to provide information about the meanings of object language sentences, we cannot claim that those instances are what we use to give content to the truth predicate. We can do one or the other, but not both. If our project is Tarski’s, we can take metalanguage paraphrases of object language sentences for granted and use them to give content to formally defined truth predicates. If our project is Davidson’s, we must presuppose an antecedently grasped notion of truth that can be applied to an arbitrary language L prior to being given translational instances of schema T for L.13 I will illustrate the things I have been saying about Tarski with a simple example. Though the example is artificial, the point it makes carries over in full force to the usual, more complicated, Tarskian definitions. Let L be a language with just two sentences—Los pantalones son verdes, and La camisa es azul. We can define a Tarskian truth predicate T for L that satisfies Convention T as follows: For all sentences s of L, s is T iff s ⫽ ‘Los pantalones son verdes’ and the pants are green or s ⫽ ‘La camisa es azul’ and the shirt is blue. Now consider the T-sentence ‘La camisa es azul’ is T iff the shirt is blue. Since we have a definition of T, we can replace the truth predicate with its definition, while preserving the meaning of the T-sentence. Doing this gives us: [(‘La camisa es azul’ ⫽ ‘Los pantalones son verdes’ and the pants are green, or ‘La camisa es azul’ ⫽ ‘La camisa es azul’ and the shirt is blue) iff the shirt is blue] Given the obvious falsity of the first disjunct of the left-hand side of the biconditional, we can simplify this as follows: Davidson now, I think, recognizes this. See his “The Structure and Content of Truth” (The Dewey Lectures, 1989), Journal of Philosophy 87 (1990): 279–328. 13

ALTERNATIVE CONCEPTUAL SCHEMES

329

[‘La camisa es azul’ ⫽ ‘La camisa es azul’ and the shirt is blue] iff the shirt is blue. This triviality is what someone knows about the sentence La camisa es azul when he knows that which is expressed by the Tarskian Tsentence. Clearly, this information tells one nothing at all about meaning. Thus, if knowledge of truth conditions is to provide any information whatsoever about meaning, then the notion of truth involved cannot be one that is defined in the manner of Tarski. So our ordinary notion of truth is not the sum of Tarski-like restricted notions of truth. What is the content of our ordinary notion, and how do we acquire it? Here is a rough picture. We learn a bit of language. Having learned some language, we are introduced to the notion of truth using the language we have learned. Somebody tells us that if one says or believes that Mommy is working, and Mommy is working, then what one says or believes is true. Similarly, if one says or believes that Daddy is asleep, and Daddy is asleep, then what one says or believes is true. Moreover, if one says or believes that Daddy is asleep, but Daddy isn’t asleep, then what one says or believes isn’t true. And so on, and so on. On the basis of instructions like these, we come to recognize that whenever one accepts or believes a proposition, one is entitled to infer the claim that the proposition is true, and vice versa. Whenever one denies or disbelieves a proposition, one is entitled to infer that the proposition isn’t true. Similarly for other attitudes one may take toward a proposition. We acquire the concept of truth as property applied to propositions— namely that which we assert and believe when we sincerely and assertively utter sentences. At the time we acquire the concept of truth, we recognize that we are not familiar with every proposition there is. If we think about it, we may realize that we will never be familiar with every proposition. But we are prepared to grant that some propositions we have not yet encountered are true. If we do come across a new proposition, we apply the concept of truth to it in the same manner that we apply it to propositions we are already familiar with. Since we know that whatever attitude we are warranted in taking toward a proposition, we are similarly warranted in taking toward the claim that it is true, we will be prepared to accept and assert a new proposition just in case we are prepared to accept and assert that it is true. Similarly for other attitudes.14 This idea is discussed in more detail in my “Understanding Deflationism,” Philosophical Perspectives (forthcoming).

14

330

DAVIDSON ON TRUTH

We also apply the notion of truth to sentences. We regard a sentence as true iff it expresses a true proposition. What now becomes of the idea that there could be a language containing true sentences that are not translatable into English? This is just the idea that there could be a language that expresses true propositions that are not expressed by any sentence of English. This is no more incoherent than the claim that there are true propositions one has not yet encountered. Since neither is incoherent, Davidson’s argument that our notion of truth dictates that there could not be any such language fails.

Summary and Final Assessment Let us sum up the results of our evaluation of Davidson’s argument about alternative conceptual schemes. First, the fact that we can interpret the speech of another group does not guarantee as much agreement between them and us as Davidson seems to assume. So long as it is possible for us to explain why the other speakers hold beliefs different from ours, we can make sense of a great deal of disagreement. Second, we can make sense of big differences between ourselves and speakers of another culture that don’t involve disagreements—e.g., differences regarding which objects are basic, and most worthy of attention. These two points suggest that, contrary to Davidson, even those whose utterances we can interpret and translate may have views different enough from ours to warrant the attribution of a different conceptual scheme. Finally, we found no reason to believe that there couldn’t be speakers whose conceptual schemes were so different from ours that we couldn’t translate their speech. In light of this, we have little alternative but to conclude that Davidson’s case against alternative conceptual schemes is a failure. Not so, his overall philosophical program. Whatever the shortcomings in conception and execution, his overall truth-theoretic approach to meaning in natural language represented a major advance over both the barren semantic skepticism of Quine, and the anti-theoretical, yet philosophically overreaching, linguistic methodology of Wittgenstein and the ordinary language philosophers.15 No judgment is here intended about the overall merits of the philosophical achievements made by these philosophers—just their conceptions of the importance of linguistic meaning for philosophy and their views about how it may fruitfully be studied.

15

SUGGESTED FURTHER READING FOR PART SIX

Main Primary Sources Discussed Davidson, Donald. “On the Very Idea of a Conceptual Scheme.” In Inquiries into Truth and Interpretation, 2d ed. (Oxford: Clarendon Press, 2001); originally published in Proceedings and Addresses of the American Philosophical Association, 47, 1974. ———. “Radical Interpretation.” In Inquiries into Truth and Interpretation; originally published in Dialectica 27, 1973. ———. “Reply to Foster.” In Inquiries into Truth and Interpretation; originally published in G. Evans and J. McDowell, eds., Truth and Meaning (Oxford: Clarendon Press, 1976). ———. “Truth and Meaning.” In Inquiries into Truth and Interpretation; originally published in Synthese 17, 1967.

Additional Primary Sources Davidson, Donald. “Actions Reasons, and Causes.” In Essays on Actions and Events (Oxford: Clarendon Press, 1980; originally published in Journal of Philosophy 60 (1963): 685–700. ———. “The Logical Form of Action Sentences.” In Essays on Actions and Events; originally published in N. Rescher, ed., The Logic of Decision and Action (Pittsburgh: University of Pittsburgh Press, 1967). ———. “The Material Mind.” In Essays on Actions and Events; originally published in P. Suppes et al., eds., Logic, Methodology, and Philosophy of Science, vol. 4 (Amsterdam: North Holland, 1973). ———. “Mental Events.” In Essays on Actions and Events; originally published in L. Foster and J. Swanson, eds., Experience and Theory (Amherst, MA: University of Massachusetts Press, 1970). ———. “On Saying That.” In Inquiries into Truth and Interpretation; originally published in D. Davidson and J. Hintikka, eds., Words and Objections (Dordrecht: Reidel, 1969). ———. “Quotation.” In Inquiries into Truth and Interpretation; originally published in Theory and Decision 11 (1979): 27–40.

332

SUGGESTED FURTHER READING

———. “The Structure and Content of Truth.” The Dewey Lectures 1989. Journal of Philosophy 87 (1990): 279–328. Tarski, Alfred. “The Concept of Truth in Formalized Languages.” In Logic, Semantics, Meta-Mathematics, translated by J. H. Woodger, 2d edition (Indianpolis, In Hackett, 1983); originally published in Polish in 1933. ———. “On the Concept of Logical Consequences.” In Logic, Semantics, Meta-Mathematics; originally published in Polish in 1936.

Additional Recommended Reading Davies, Martin, Meaning, Quantification, and Necessity. New York: Routledge, 1981. Foster, J. “Meaning and Truth Theory.” In G. Evans and J. McDowell, eds., Truth and Meaning (Oxford: Calrendon Press, 1976). Larson, Richard, and Peter Ludlow. “Interpreted Logical Forms.” Synthese 95 (1993): 305–56; reprinted in Peter Ludlow, ed., Readings in the Philosophy of Language (Cambridge, MA: MIT Press, 1997). Larson, Richard, and Gabriel Segal. Knowledge of Meaning. Cambridge, MA: MIT Press, 1995. Soames, Scott. “Truth and Meaning: The Role of Truth in the Semantics of Propositional Attitude Ascriptions.” In Kepa Korta and Jesus M. Larrazabal, eds., Truth, Rationality, Cognition, and Music: Proceedings of the Seventh International Colloquium on Cognitive Science (Dordrecht: Kluwer, 2003). ———. “Truth, Meaning, and Understanding.” Philosophical Studies 65 (1992): 17–35. ———. “Understanding Deflationism.” Philosophical Perspectives (forthcoming). ———. Chapters 3 and 4 of Understanding Truth. New York: Oxford University Press, 1999.

PA R T S E V E N SAUL KRIPKE ON NAMING AND NECESSITY

This page intentionally left blank

CHAPTER 14 NAMES, ESSENCE, AND POSSIBILITY chapter outline 1. The Significance of Naming and Necessity 2. Why Descriptions Do Not Give the Meanings of Names The modal argument that the meanings of names are not given by descriptions speakers associate with them 3. Rigid Designation Rigid Designation and the Modal Argument Definition of rigid designation; Kripke’s argument that names are rigid, whereas descriptions typically associated with them are not; the attempt to avoid this argument by appealing to rigidified descriptions; extension of the modal argument to rebut this proposal

A confusion to be avoided How it is that the claim that n designates o may fail to be necessary, even though n designates o with respect to all possible world-states

Rigid designation and essentialism The use of rigid designation to rebut Quine’s objection that it makes no sense to say of o, independently of how it is described, that it has a property essentially

Rigid designation, possible worlds, and“transworld identification” Possible worlds are not alternate universes requiring criteria for identifying counterparts of actually existing objects; they are possible states the universe could have been in; the sense in which rigid designators are used to “stipulate” these states

4. Why Descriptions Normally Do Not Fix the Referents of Names Arguments against the weak, reference-fixing, version of the description theory Kripke’s historical chain conception of reference determination Do historical chains provide reference-fixing descriptions? Why they do not; resolution of an unclarity about reference-fixing

The Significance of Naming and Necessity In this chapter, we begin our discussion of Saul Kripke’s book Naming and Necessity, which was originally presented as three long public lectures at Princeton University in January of 1970, when Kripke was 29

336

KRIPKE

years old.1 A tape recording was made of the lectures, and two professors in the Princeton philosophy department at the time, Gilbert Harman and Thomas Nagel, produced a transcript from the tapes. Kripke added footnotes, and later wrote a preface for the book version. The impact of the lectures was profound and immediate, and over the years their influence has grown. In the philosophy of language, Naming and Necessity is among the most important works ever, ranking with the classical work of Frege in the late nineteenth century, and of Russell and Tarski in the first half of the twentieth century. Beyond the philosophy of language, it fundamentally changed the way in which much philosophy is done. The most important aspects of the work are (i) a set of theses about the meaning and reference of proper names, (ii) a corresponding set of theses about the meaning and reference of natural kind terms such as heat, light, gold, water, and tiger, (iii) a compelling defense of the metaphysical concepts of necessity and possibility, (iv) a sharp distinction between the metaphysical notion of necessity and the epistemological notion of aprioricity, (v) forceful arguments that there are necessary truths that are knowable only aposteriori, and apriori truths that are contingent, and hence not necessary, and (vi) a persuasive defense of the intelligibility of essentialism—i.e., the claim that it makes sense to characterize objects as having some of their properties essentially, and others accidentally. In addition to these explicit aspects of the work, the discussion in Naming and Necessity had far-reaching implications for what has come to be known as externalism about meaning and belief—roughly, the view that the meanings of one’s words, as well as the contents of one’s beliefs, are partly constituted by facts entirely outside oneself. Finally, Naming and Necessity played a large role in the implicit, but widespread, rejection of the view—so popular among ordinary language philosophers—that philosophy is nothing more than the analysis of language.

Why Descriptions Do Not Give the Meanings of Names We will begin our investigation of Kripke’s work by looking at his discussion of the description theory of proper names, of which he distinguishes two different versions. According to the first, proper names have the same meanings as descriptions that speakers associate with Saul Kripke, Naming and Necessity, Cambridge, MA: Harvard, 1980, originally published in D. Davidson and G. Harman, eds., Semantics of Natural Languages (Dordrecht: Reidel, 1972); citations will be to the 1980 text.

1

NAMES, ESSENCE, AND POSSIBILITY

337

them. According to the second, although names may not be synonymous with descriptions, the referent of a proper name n, as used by a speaker at a given time, is determined, as a matter of linguistic rule, to be the one and only object that satisfies the descriptions associated with n by the speaker at that time. Since the meaning of a term is assumed to determine its reference, the first version of the description theory is understood as entailing the second. However, the converse does not hold; it may turn out that the referent of a name is, as a matter of semantic rule, determined by a description, even though the name is not synonymous with the description. These two versions of the description theory are expressed by theses 1 and 2.2 two theses about names Thesis 1: The meaning of a name n (for a speaker at a time) is given by a description, a conjunction of descriptions, or a cluster of descriptions D that the speaker associates with n at the time. If D gives the meaning of n, then substitution of one for another preserves both meaning and proposition expressed. Thus, if S’ results from S by substitution of D for one or more occurrences of n in S, then S and S’ mean the same thing, and express the same proposition. Thesis 2: The referent of a name n (for a speaker at a time) is semantically fixed (determined) by a description, a conjunction of descriptions, or a cluster of descriptions D that the speaker associates with n at the time. If D fixes the referent of n, then: (i) the speaker believes that D applies to a unique individual; (ii) if D does apply to a unique individual o, then o is the referent of n; (iii) if D does not apply to a unique individual, then n has no referent. (iv) the speaker knows (or is capable of knowing) apriori that if n exists, then n is D expresses a truth. In lecture 1, Kripke gives an argument against thesis 1, which has come to be known as the modal argument. We illustrate the argument 2 These are reconstructions of the two theses as Kripke understands them. For Kripke’s own formulations, see pp. 71–80 of Naming and Necessity.

338

KRIPKE

with an example. Let n be the name Aristotle. Let (1–8) be candidates for the description D that gives the meaning of the name. 1. the founder of formal logic 2. the greatest student of Plato 3. the teacher of Alexander 4. the famous Greek philosopher named ‘Aristotle’ 5. the last great philosopher of antiquity 6. the conjunction of 1–5 7. the conjunction of all descriptions the speaker associates with n 8. a cluster of descriptions including 1–5 that the speaker associates with n For our purposes, the claim that the meaning of a name is given by a cluster of descriptions D1 . . . Dn will be regarded as equivalent to the claim that the meaning of the name is given by the description the thing of which most (or a sufficient number) of the following are true: it is D1, it is D2, . . . , it is Dn. We may now test the claim that the meaning of Aristotle is given by one or more of the descriptions (1–8), by applying the following, modal test. the modal test If D gives the meaning of n, then the proposition (truth/ statement/claim) expressed by the sentence If n existed, then n was D. is a necessary truth. The rationale behind this test is that if D has the same meaning as n, then substitution of one for the other in a sentence won’t change the proposition expressed (or the statement made). But that means that the sentence If n existed, then n was D expresses the same proposition (says the same thing) as the sentence If D existed, then D was D. Since the latter sentence expresses a necessary truth, so does the former sentence. In using the terminology necessary truth we mean the following: A proposition is a necessary truth iff (i) it is true given the way the world actually is and (ii) it would have been true had the world been in any other possible state it could have been in.

NAMES, ESSENCE, AND POSSIBILITY

339

If thesis 1 is correct, then there must be some description D which speakers associate with the name ‘Aristotle’ such that the proposition expressed by If Aristotle existed, then Aristotle was D is a necessary truth. In fact, since D gives the meaning of ‘Aristotle’, the proposition expressed by that sentence should be both necessary and knowable apriori. But, Kripke argues, there is no such description D. For example, consider the description the founder of formal logic as a possible candidate. To apply the modal test, we ask whether the statement If Aristotle existed, then Aristotle was the founder of formal logic is a necessary truth. To say that it is a necessary truth is to say that there is no possible way that the world could have been which would make the antecedent, Aristotle existed, true, but the consequent, Aristotle was the founder of formal logic, false. But that doesn’t seem right. On the contrary, it seems that the world could have been in a state in which Aristotle existed, but didn’t do any logic at all. Since Aristotle could have existed without being the founder of formal logic, the conditional statement If Aristotle existed, then Aristotle was the founder of formal logic is not a necessary truth. Thus, the name Aristotle does not mean the same thing as the description the founder of formal logic. This result is not an isolated one. The same argument could be given for the other candidate descriptions (1–8), or for other descriptions with which one would most naturally think of replacing the name Aristotle. The reason for this is that nearly all descriptions people nowadays associate with the name have to do with Aristotle’s famous accomplishments, especially those involving his philosophy. However, as Kripke rightly points out, none of those accomplishments were necessary conditions for Aristotle to exist.3 Aristotle could have existed even if he hadn’t gone into philosophy, or done anything important. Thus, Kripke is prepared to run the argument we have just given using any description or cluster of descriptions based on Aristotle’s famous achievements, or well-known characteristics. But since these provide the main descriptive content that most of us associate with the name, he concludes that the description theory of meaning is incorrect as a theory of how most of us use the name. The force of this argument against the description theory of the meaning of proper names can be brought out by asking what conditions a description would have to satisfy in order to block the argument. At a minimum, the description D would have to be such that the proposition expressed by If n existed, then n was D was both necessary and apriori. 3

Even being named ‘Aristotle’ was not a necessary condition for Aristotle to exist.

340

KRIPKE

The problem is that for nearly all names one can think of, one can’t come up with such descriptions. If this is right, then thesis 1 is false. The argument we have just given against thesis 1 is not directed against thesis 2, which expresses the view that descriptions semantically fix the referents of names. The reason for this is that thesis 2 does not claim that names and descriptions are synonymous. It was only the claim that names have the same meanings as certain descriptions that allowed us to view thesis 1 as committed to the claim that when we substitute the names and descriptions for one another in a sentence, we do not change the proposition expressed, and so do not change the modal or epistemological status of the sentence. Suppose, however, that we had a theory that told us not that D gives the meaning of n, but only that D semantically fixes the referent of n. Such a fix-the-referent theory would not claim that n is synonymous with D, and so would not be committed to the claim that substitution of one for another must preserve the proposition expressed. But if substitution could change the proposition expressed, then we have no reason to think that it couldn’t also change the modal or epistemic status of the sentence. Thus, although the modal argument may be seen as refuting the view that names are synonymous with the descriptions that speakers associate with them, it does not, by itself, refute the view that these descriptions semantically determine the referents of names. So, if one wants to criticize the fix-the-referent version of the description theory, one must come up with additional arguments. In lectures 2 and 3 Kripke does this. But before we look at those arguments, we need to examine some of the key concepts he employs.

Rigid Designation Rigid Designation and the Modal Argument

The first such concept is that of a rigid designator.4 rigid designation A singular term t is a rigid designator of an object o iff t designates o with respect to all possible states of the world (in This is a simplified definition that abstracts away from various complications. For example, we here leave indexicals and variables out of consideration, and so need not relativize reference to contexts and assignments of values to variables. For a definition taking these complications into

4

NAMES, ESSENCE, AND POSSIBILITY

341

which o exists); and, moreover, t never designates anything other than o (with respect to any possible-world-state). If a singular term t is a rigid designator of an object o, then sentences containing t are true when taken as descriptions of alternative possible states w1, w2, w3 of the world iff one and the same object, o, has the relevant properties in those alternative states. For example, if t is a rigid designator of an object o, F expresses the property ␾, and p is the proposition expressed by Ft, then (i) p is true relative to the actual state of the world iff as things actually are in the world, o has the property ␾, and (ii) p is true with respect to any other possible state w of the world iff relative to w, o has property ␾ (i.e., if the world were in state w, then o would have property ␾). If t is a non-rigid designator of o, then although (i) remains as above, (ii) does not. If t is non-rigid, then there are sentences Ft, properties ␾, propositions p, and possible states of the world w and w*, such that p is expressed by Ft, and either p is true with respect to w even though o does not have ␾ with respect to w, or p is false with respect to w* even though o has ␾ with respect to w* (or both). For example, consider the pair (9) and (10). 9. The winner of the 1996 U.S. presidential election was a Democrat. 10. Bill Clinton was a Democrat. The description the winner of the 1996 U.S. presidential election and the name Bill Clinton designate the same individual o. Since o was a Democrat, both (9) and (10) are true, with respect to the world as it actually is (was). This is not so with respect to a possible state w of the world in which Clinton runs as a Democrat but the Republican Bob Dole wins the 1996 election. The individual whose party affiliation determines the truth value of (10) relative to the world-state w is Bill Clinton, the same individual who is relevant to determining the truth value of (10) in the actual world-state. Hence, (10) is true with respect to w. However, the individual whose party affiliation determines the truth value of (9) relative to w is not Bill Clinton, but rather is Bob Dole. Hence (9) is false with respect to w. This example shows that the account, see my Reference and Description: The Two-Dimensionalist Attempt to Revive Descriptivism (in preparation). For present purposes, we take singular terms to include names and singular definite descriptions like the so and so.

342

KRIPKE

description the winner of the 1996 U.S. presidential election is nonrigid, whereas the name Bill Clinton is rigid. How does it happen that for a rigid designator the same object o is relevant with respect to all possible states of the world; whereas for a non-rigid designator the relevant objects differ from world-state to world-state? The answer may be expressed as follows: If t is rigid, then whatever object is actually designated by t is designated by t with respect to all possible states of the world in which that object exists, and nothing other than that object is designated by t with respect to any world-state; but if t is non-rigid, then either what is actually designated by t fails to be designated by t with respect to other possible worldstates (in which that object exists), or something other than that object is designated by t with respect to some world-state.5 This suggests a linguistic test for determining whether an arbitrary singular term in English is a rigid designator. a linguistic test t is a rigid designator iff the sentence The individual that is (was) actually t could not have existed without being t, and nothing other than the individual that is (was) actually t could have been t expresses a truth. Alternatively put: A singular term t of English is a rigid designator iff the relevant sentences of the form (11) and (12) are false. A singular term is non-rigid iff either (11) or (12) is true. 11. The individual who is (was) actually t could have existed without being t. 12. It could have been the case that someone other than the individual who is (was) actually t was t. Kripke maintains that if we apply this test we will find that proper names are rigid designators, whereas most ordinary descriptions are not. This is not to say that he thinks that no descriptions are rigid; for example, he would recognize the positive square root of 25 and the individual who is identical with Saul Kripke to be rigid. However, he believes that where most ordinary names are concerned, descriptions like these are not good candidates for being ones used by speakers to give This idea is a rough approximation which ignores certain abstruse complications that arise in special cases when indexicals are considered and the notion of rigid designation is relativized to context. (The same is true of the linguistic test that follows.)

5

NAMES, ESSENCE, AND POSSIBILITY

343

meanings or establish reference.6 So, in the case of the vast majority of ordinary names, he would maintain that the descriptions associated with them are non-rigid. In particular, the descriptions (1–8) associated with the name Aristotle in our example are non-rigid. With this in mind, we can restate Kripke’s modal argument. In essence, the argument is the following: P1. Names are rigid designators. P2. Standardly, the descriptions associated by speakers with names are not rigid designators. C. So, names are standardly not synonymous with descriptions associated with them by speakers. This argument was immediately recognized to be a powerful challenge to descriptivism about the meanings of proper names, and it continues to be accepted by a great many philosophers to this day. In recent years however, a response to the argument has gained a certain amount of currency among some theorists who continue to be attracted to descriptivism. The response is based on the observation that for any non-rigid description the F we can form a rigid description the actual F that designates, when we are talking about any possible world-state, the individual that is F in the actual world-state.7 Consider, for example, sentence (13). 13. It could have been the case that the actual winner of the 1996 presidential election didn’t win the 1996 presidential election. Intuitively, what this sentence says is true. This means that there must be a possible world-state w such that sentence (14), as used by us, is true when taken as a description of w. 14. The actual winner of the 1996 presidential election didn’t win the 1996 presidential election. This, in turn, means that the individual who counts as the referent of our use of the term the actual winner of the 1996 presidential election, We here leave aside special cases like the numeral ‘2’, which conceivably might be defined the successor of 1. 7 This statement is somewhat of an exaggeration, and ignores the complications mentioned in footnotes 4 and 5. Those interested in a fuller story should see the discussion of the use of the actuality operator to rigidify descriptions in the work cited in footnote 4. 6

344

KRIPKE

when used to say something about w, is the individual who won the election, not with respect to w, but with respect to the world as it actually is. Sentence (14) is true with respect to w because that individual—the person who was the winner with respect to the actual world-state—didn’t win, with respect to w. The lesson here is that the result of adding the actuality operator to a description the F results in a new description, the actual F, which rigidly designates the object that the former description designates in the actual world-state (if such an object is uniquely designated by the F relative to the actual state of the world). This idea has been used by some post-Kripkean descriptivists to suggest that names are synonymous, not with ordinary descriptions, but with descriptions rigidified using the actuality operator. This claim in effect attacks premise P2 above, and is not refuted by the modal argument that Kripke gives. Nevertheless, the proposal fails for other reasons. First, if the proposal were correct, then the proposition expressed by If n existed, then n was D would be the same as the proposition expressed by If the actual D existed, then the actual D was D. This latter proposition is something knowable apriori, independent of any empirical evidence. However, when n is an ordinary proper name, the proposition expressed by If n existed then n was D typically is not knowable apriori. Thus, the two propositions are different, and the proposal for saving thesis 1 founders. (This point is implicit in Naming and Necessity. We will return to it later, when we discuss the material in Kripke’s lecture 2.) There is also a second problem with the proposal that Kripke did not address, but which is discussed at length in chapter 2 of my Beyond Rigidity.8 I will mention only the gist of it. The proposition that the actual F is G is a proposition that says of the actual state of the world that the individual who is uniquely F with respect to it is also G.9 Consequently, it is possible to believe that the actual F is G only if one is in some kind of epistemic contact with the actual state of the world, and, by virtue of this, believes a certain proposition about it. Let us suppose, for the sake of argument, (i) that all of us, living in the world as it actually is, do have such contact with this state of the world, and (ii) that because of this, when we believe that the F is G we automatically qualify as also believing, of the actual state of the world, that the individual who is uniquely F with respect to it is G as well. Even if we grant this, however, we must also recognize that things change when 8 9

Scott Soames, Beyond Rigidity (New York: Oxford University Press, 2002). In this paragraph, I use ‘F’ and ‘G’ as schematic letters. ‘D’ is used as a metalinguistic variable.

NAMES, ESSENCE, AND POSSIBILITY

345

we consider, not what agents actually believe, but what it is possible for agents to believe. Certainly, it is possible to believe that Aristotle was a philosopher without being in any epistemic contact with the actual world-state, and hence without believing anything about it. For example, if various facts irrelevant to Aristotle had been different, I might still have believed that Aristotle was a philosopher. To say this is just to say that there is some possible world-state w, different from the actual world-state, such that with respect to w I believe that Aristotle was a philosopher, even though with respect to that world-state I may not believe anything about the actual world-state. I may not believe anything about the actual world-state, since the actual world-state is a total or maximal property which represents the way things actually are. Had the world been in state w, I may have been familiar with the way things are with respect to w, and so have been epistemically acquainted with w, but I need not have been familiar with other maximally possible states of the world, and so I need not have been familiar with the actual world-state. If this is right, it shows that for any description the actual D, it is possible to believe the proposition expressed by Aristotle was a philosopher without believing the proposition expressed by The actual D was a philosopher, and hence that the two propositions are different.10 Thus, the proposal that names are synonymous with descriptions rigidified using the actuality operator is false. A Confusion to Be Avoided

Before going further, we pause to dispel a confusion that is all too easy to fall into, but very important to avoid. The confusion arises from a puzzle generated by the following two claims. (i) The name ‘Aristotle’ is a rigid designator. Thus, for all possible states of the world w, it refers to the same individual—the man Aristotle—with respect to w. (ii) It is not a necessary truth that Aristotle was named ‘Aristotle’; it could have been the case that the name ‘Aristotle’ did not This step in the argument tacitly assumes, as did the previous argument against the proposal, that x believes that S reports a relation between the believer and the proposition expressed by S. Some proponents of analyzing names as descriptions rigidified using the actuality operator dispute this assumption. For rejoinders to the most systematic attempt to develop such a point of view, see my “Saul Kripke, the Necessary Aposteriori, and the Two-Dimensionalist Heresy,” in M. Garcia-Carpintero and J. Maciá, eds. The Two-Dimensional Framework: Foundations and Applications (Oxford: Oxford University Press, forthcoming), and my Reference and Description: The Two-Dimensionalist Attempt to Revive Descriptivism, in preparation. 10

346

KRIPKE

refer to Aristotle. Thus, there must be some world-state w such that the claim that ‘Aristotle’ does not refer to Aristotle is true with respect to w. Both claims are true, and both would be endorsed by Kripke. However, this might seem puzzling, since (i) and (ii) may appear to be inconsistent. What makes (i) and (ii) seem inconsistent is the tendency to tacitly accept (iii) as something so obvious as to go without saying. (iii) The three-place relation ——— refers to ——— with respect to ——— (tacitly invoked in (i)) holds between the name ‘Aristotle’, the man Aristotle, and a world-state w iff it is true with respect to w that the two-place relation ——— refers to ——— (invoked in (ii)) holds between ‘Aristotle’ and Aristotle—i.e., iff the claim that ‘Aristotle’ refers to Aristotle is true when taken as a description of w. Although principle (iii) might seem undeniable at first glance, in fact it is false. Throughout this chapter we have been following Kripke in taking the three-place relation ——— refers to ——— with respect to ——— to hold between a name n, object o, and world-state w iff n, as used by us here and now in the world as it actually is, refers to the object o, when our words are taken as descriptions of w. Because of this, n may refer to o with respect to w even if (a) the name n doesn’t exist with respect to w, or (b) with respect to w, the name n is not used by speakers to refer to anything, or (c) with respect to w, the name n is used by speakers to refer to something other than o. What, if anything, speakers would have used the name n to refer to had the world been in state w is irrelevant to whether n refers to o with respect to w. However, what speakers would have used n to refer to, had the world been in state w, is crucial to determining which pairs of names and objects the two-place relation ——— refers to ——— applies to with respect to w. It is true with respect to w that the name n refers to the object o iff had the world been in state w, speakers would have used n to refer to o. Thus, what (ii) says is that there are world-states such that had the world been in those states, speakers would not have used ‘Aristotle’ to refer to Aristotle. This is compatible with the claim made by (i)—namely that, here and now in the world as it actually is, we use the name ‘Aristotle’ to refer to the man Aristotle when our words are taken as descriptions of any world-state whatsoever.

NAMES, ESSENCE, AND POSSIBILITY

347

Rigid Designation and Essentialism

Throughout Kripke’s discussion of names in lecture 1 of Naming and Necessity, he takes it for granted that the distinction between the essential properties of an object and its contingent properties is a legitimate one. An essential property of an object is a property the object could not lack in any circumstance in which it existed at all. A contingent, or accidental, property is one that the object has, but could have existed without. Examples of contingent properties of mine are the property of living in Princeton, the property of being a father, and the property of being a philosopher. Uncontroversial examples of essential properties of mine are rarer, but the following seem to be good candidates: the property of being a human being, the property of having a brain, the property of having a body made up of molecules, the property of being mortal, the property of not being identical with Saul Kripke. There is a close connection between the notion of a rigid designator and the claim that an object has a property essentially. This connection is expressed by (i).11 (i) If n is a rigid designator of o, and F is a predicate expressing the property P, then the claim that P is an essential property of o is equivalent to the claim it is necessary that if n exists, then n is F. The equivalence mentioned in (i) is relevant to a puzzle that had been used by Quine for more than two decades prior to Naming and Necessity to cast doubt on the intelligibility of essentialism.12 Quine had contended first, that the notion of an essential property of an object is Some explanation of my terminology may be in order. Predicates express properties while being true of objects with respect to world-states. Note, what property a predicate expresses is not relativized to different world-states. If F expresses P, then for any world-state w, F is true of an object o with respect to w iff o has P with respect to w. 12 See Quine, “Notes on Existence and Necessity,” Journal of Philosophy 40 (1943): 113–27; “The Problem of Interpreting Modal Logic,” Journal of Symbolic Logic 12 (1947): 43–48; “Reference and Modality,” in From a Logical Point of View (Cambridge, MA: Harvard University Press, 1953, 1961, 1980); and “Three Grades of Modal Involvement,” originally published in 1953, reprinted in Quine, The Ways of Paradox (New York: Random House, 1966). For explication and critique of Quine, each excellent in its way, see David Kaplan, “Opacity,” in E. Hahn and P. A. Schilpp, eds., The Philosophy of W. V. Quine (La Salle, IL: Open Court, 1986), and John Burgess, “Quinus ab omni naevo vindicatus,” in Ali A. Kazmi, ed., Meaning and Reference (Calgary: University of Calgary Press, 1998). Burgess makes the point that if one (wrongly) regards necessity simply as analyticity, and takes analyticity to be a property of sentences—things which both Quine and those he was criticizing at the time did—then his skepticism about “analytically essential properties of an object” makes much more sense (though it no longer has anything to do with genuine metaphysical 11

348

KRIPKE

defined in terms of the notion of necessity, and second, that whatever grasp we have of necessity is expressed by our use of the predicate is a necessary truth, which applies to sentences, or by our use of the operator it is necessary that . . . , which attaches to sentences. Thus, he thought, if we are to make sense of the idea that is essentially F applies to a certain object o, we have no choice but to view this claim as based on the assumption that the sentence it is necessary that if t exists, then t is F is true, for some designated choice of a term t that refers to, or describes, o. However, Quine also observed that for any object o, there will be some terms t that refer to o which make the sentence It is necessary that if t exists, then t is F false, even if there are other terms t that refer to o which make it true. Thus, he thought, relative to one way of describing o it may turn out that the property expressed by F is an essential property of o, whereas relative to a different way of describing o it will turn out that the property expressed by F is not an essential property of o. But what if we consider o on its own, independent of any description? Is the property expressed by F one of o’s essential properties or not? It would seem that there is no saying. Here is a representative example of Quine’s presentation of this allegedly puzzling problem. Perhaps I can evoke the appropriate sense of bewilderment as follows. Mathematicians may conceivably be said to be necessarily rational and not necessarily two-legged; and cyclists necessarily two-legged and not necessarily rational. But what of an individual who counts among his eccentricities both mathematics and cycling? Is this concrete individual necessarily rational and contingently two-legged or vice versa? Just insofar as we are talking referentially of the object, with no special bias toward a background grouping of mathematicians as against cyclists or vice versa, there is no semblance of sense in rating some of his attributes as necessary and others as contingent. Some of his attributes count as important and others as unimportant, yes; some as enduring and others as fleeting; but none as necessary or contingent.13 necessity, or with metaphysically essential properties). Kaplan argues (i) that Quine’s position on the matter was tied to a larger mistaken opposition to “quantifying in” to all nonextensional constructions, (ii) that anyone who accepts logical truth can accept benign versions of essentialism, and (iii) that though the intelligibility of more robust essentialist claims should not be in doubt, the truth, or the falsity, of such claims does raise substantive metaphysical questions. 13 Quine, Word and Object, p. 199.

NAMES, ESSENCE, AND POSSIBILITY

349

Let i be some individual who is both a brilliant mathematician and a champion cyclist, and suppose that the world’s greatest mathematician and the world’s greatest cyclist both designate i. Then, since (15a) is, arguably, true, while (15b) is not, 15a. It is necessary that: if the world’s greatest mathematician exists (i.e., if there is such an individual as the world’s greatest mathematician), then the world’s greatest mathematician is rational. b. It is necessary that if the world’s greatest mathematician exists (i.e., if there is such an individual as the world’s greatest mathematician), then the world’s greatest mathematician is two-legged. it follows, on Quine’s view, that relative to the choice of describing i as the world’s greatest mathematician, being rational is one of i’s essential properties, but being two-legged is not. However, if we choose to describe i as the world’s greatest cyclist, we get the opposite result. Since (16a) is, arguably, true, while (16b) is not, 16a. It is necessary that: if the world’s greatest cyclist exists (i.e., if there is such an individual as the world’s greatest cyclist), then the world’s greatest cyclist is two-legged. b. It is necessary that: if the world’s greatest cyclist exists (i.e., if there is such an individual as the world’s greatest cyclist), then the world’s greatest cyclist is rational. it follows, on Quine’s view, that relative to the choice of describing i as the world’s greatest cyclist, being two-legged is one of i’s essential properties, but being rational is not. Accordingly, Quine thinks, it makes no sense to ask of i himself, independent of any way of describing him, which of his properties are essential and which are not. In general, Quine assumed that there was no principled, non-arbitrary way of selecting, for an arbitrary object o and property P, what sort of term t should be used to underwrite claims to the effect that o did, or did not, have P essentially. Thus, his doctrine was that it makes no sense to ask whether an object has a property essentially, independently of how it is described. Rather, objects have or lack properties essentially only relative to ways of describing them. The relevance of all this to Kripke is that if, as he maintains, there is a genuine distinction between rigid and non-rigid designators, then

350

KRIPKE

rigid designators provide a principled connection between claims about the essential properties of objects and claims about which sentences do, and which sentences do not, express necessary truths. It is only sentences of the sort mentioned in (i) that contain a rigid designator of the object o that are relevant to the question of whether o has the property P essentially or not. When we consider whether an object has a property essentially, we use a rigid designator to talk about one and the same object with respect to all possible world-states. Because the designator is rigid, the question of whether the object has the property with respect to all the world-states is equivalent to the question of whether the sentence or formula attributing the property to the referent of the rigid designator is true relative to all those world-states. The truth values of other sentences containing non-rigid designators of the object are simply irrelevant.14 In this way, Kripke rebuts Quine’s objection to the intelligibility of essentialism. The dialectical situation is this: We begin with an intuitive distinction. Although I am correctly described as the Princeton philosopher who was raised in Seattle, being a philosopher, working at Princeton, and being raised in Seattle are contingent properties of mine—I could have existed even if I hadn’t been raised in Seattle, gone into philosophy, or worked at Princeton. By contrast, being a sentient being and not being identical with Saul Kripke seem to be essential properties of mine— there seem to be no possible scenarios in which I exist but am not a sentient being, or in which I am Saul Kripke. Everyone understands these claims. Although there may be disagreements and uncertainties about which properties fall into which categories, we all recognize the intelligibility of claims of this sort prior to receiving any instruction in philosophy. Then Quine comes along with an objection. He gives an argument that is designed to show that we have all been unwittingly talking nonsense. However, his objection relies on a false premise— namely, that there is no non-arbitrary way of selecting, for a given object o and property P expressed by a predicate F, what sort of term t designating o should be used to construct the statements, It is necessary that if t exists, then t is F, upon which the truth or falsity of essentialist claims about o depend. Kripke refutes this premise by showing that rigid designators and only rigid designators provide the connection between claims about the necessity of statements, on the one Note, since the world’s greatest mathematician and the world’s greatest cyclist are both nonrigid, the sentences in (15) and (16) are irrelevant to the question of whether or not the individual denoted by them both is essentially rational, or is essentially two-legged.

14

NAMES, ESSENCE, AND POSSIBILITY

351

hand, and the essential properties of objects, on the other. With Quine’s objection out of the way, our pre-philosophical conviction that essentialist claims are intelligible stands unchallenged. At this point, one must be on guard against a certain all too familiar response from unreconstructed Quineans. Of course, they will say, if rigid designation makes sense, then essentialism also makes sense. But does rigid designation make sense? Take the name Aristotle, for example. To say that it is a rigid designator is to say that the claim made by our use of a sentence like Aristotle was a philosopher is true when evaluated with respect to a possible world-state w (e.g., one in which the man m we actually call ‘Aristotle’ never met or studied with Plato) iff m was a philosopher (i.e., had the property of being a philosopher) with respect to w. But this makes sense only if it makes sense to ask of the individual m, independent of any description, whether m had a certain property relative to some merely possible world-state. Surely, this is the sort of thing that Quine questioned. Hence, the Quinean skeptic maintains, Kripke’s appeal to rigid designation is question-begging, and Quine’s powerful objection remains intact. In my view, and in Kripke’s, this signature move of the Quinean skeptic is entirely misguided. We began with an intuitive, pre-philosophical distinction the intelligibility of which is recognized by nearly everyone. Quine offers an objection. He purports to show that there is something incoherent in the way we have all been looking at things. The burden of proof is then on him to demonstrate that there is some internal incoherence in our thinking, some conflict between different aspects of our view that we had not noticed. Kripke’s rebuttal of Quine’s objection shows that he failed to do this. At this point, it is no response to reply that a real Quinean skeptic—one determined, at any cost, not to accept the intelligibility of essentialist claims—would not grant the presuppositions of Kripke’s rebuttal. It is not up to Kripke to prove the intelligibility of essentialism from premises that would have to be accepted by even the most determined skeptic, any more than it is the job of the opponent of radical skepticism about the external world to prove the existence of objects other than oneself and one’s ideas to the satisfaction of a philosopher determined to take a solipsistic stand. It is sufficient to rebut any reasoned objections that such skeptics may raise in an attempt to persuade us that, by standards that even we must recognize, our ordinary, commonsense views are in error. In light of this, it is not surprising that Kripke’s final response to Quine on this point has a distinctly Moorean flavor. Immediately after

352

KRIPKE

a paragraph summarizing Quine’s objection to the intelligibility of the claim that objects have both essential and accidental properties, independent of how they are described, Kripke says the following: It is even suggested in the literature, that though a notion of necessity may have some sort of intuition behind it (we do think some things could have been otherwise; other things we don’t think could have been otherwise), this notion [of a distinction between necessary and contingent properties] is just a doctrine made up by some bad philosopher, who (I guess) didn’t realize that there are several ways of referring to the same thing. I don’t know if some philosophers have not realized this; but at any rate it is very far from being true that this idea [that a property can meaningfully be held to be essential or accidental to an object independently of its description] is a notion which has no intuitive content, which means nothing to the ordinary man. Suppose that someone said, pointing to Nixon, ‘That’s the guy who might have lost’. Someone else says ‘Oh no, if you describe him as ‘Nixon’, then he might have lost; but, of course, describing him as the winner, then it is not true that he might have lost’. Now which one is being the philosopher, here, the unintuitive man? It seems to me obviously to be the second. The second man has a philosophical theory. The first man would say, and with great conviction, ‘Well, of course, the winner of the election might have been someone else. The actual winner, had the course of the campaign been different, might have been the loser, and someone else the winner; or there might have been no election at all. So such terms as “the winner” and “the loser” don’t designate the same objects in all possible worlds. On the other hand, the term “Nixon” is just a name of this man’. When you ask whether it is necessary or contingent that Nixon won the election, you are asking the intuitive question whether in some counterfactual situation, this man would in fact have lost the election. If someone thinks that the notion of a necessary or contingent property (forget whether there are any nontrivial necessary properties [and consider] just the meaningfulness of the notion) is a philosopher’s notion with no intuitive content, he is wrong. Of course, some philosophers think that something’s having intuitive content is very inconclusive evidence in favor of it. I think it is very heavy evidence in favor of anything, myself. I really don’t know, in a way, what more conclusive evidence one can have about anything, ultimately speaking. But, in

NAMES, ESSENCE, AND POSSIBILITY

353

any event, people who think the notion of accidental property unintuitive have intuition reversed, I think.15 At bottom, Kripke’s position in the face of the Quinean skeptic about the intelligibility of essentialism is rather like Moore’s position in the face of the skeptics he faced.16 For Kripke, (i) there is a strong initial presumption that both our ordinary counterfactual talk and the distinction between essential and accidental properties that goes along with it are intelligible, and (ii) in order to maintain the distinction, it is enough to rebut skeptical arguments designed to demonstrate that no such distinction could be coherent. If, as I believe, this is the right way to view the situation, why was Quine’s skepticism on this point so influential for so long? In my opinion, three main factors played important roles. First, for many years Kripke’s technical apparatus of possible world semantics, including his concept of rigid designation, was either nonexistent, not widely understood, or imperfectly grasped (and sometimes entangled with extraneous and implausible views). Without a useful and readily applicable concept of rigid designation, it was not entirely clear what the response to Quine’s skeptical objection should be.17 Second, as in many discussions of skepticism, issues about who bears the burden of proof became thoroughly confused from very early on, with defenders of Quine (perversely) refusing to grant the initial presumption of intelligibility to the commonplace counterfactual talk appealed to by Kripke and other proponents of the distinction between essential and accidental properties. Third, the by now familiar confusion of necessity with analyticity played a large role in obscuring the central issues at stake. Naming and Necessity, pp. 41–42. The square brackets are Kripke’s. See chapters 1 and 2 of volume 1 for a discussion of Moore’s response to skepticism. 17 There were, to be sure, historical anticipations of Kripke’s doctrines about rigid designators, and their potential use in replying to Quine-style objections, including important discussions by Bertrand Russell, Raymond Smullyan, Frederick Fitch, Ruth Barcan Marcus, Paul Ziff, Arthur Prior, Dagfinn Follesdal, Keith Donnellan, Peter Geach, and others. For historical background, see my “Revisionism about Reference” and “More Revisionism about Reference,” along with John Burgess’s “Marcus, Kripke, and Names” and “How Not to Write History of Philosophy,” all of which can be found in Paul W. Humphreys and James H. Fetzer, eds., The New Theory of Reference (Dordrecht, Boston and London: Kluwer, 1998). Although written in response to a nasty controversy involving proper historical credit, these papers contain significant information about Kripke’s precursors, and early attempts to reply to Quine. (A different aspect of the controversy is covered by Burgess in his unpublished manuscript, “Geach, Donnellan, Kripke, and Names.”) For a useful and extensive survey of a number of relevant topics, see also Stephen Neale, “On a Milestone of Empiricism,” in Alex Orenstein and Petr Kotatko, eds., Knowledge, Language, and Logic (Dordrecht and London: Kluwer, 2000). 15 16

354

KRIPKE

Quine’s discussions make clear both that he identified analyticity with necessity, and that he took analyticity to be a property of sentences. Given his view that claims about which properties an object has essentially or accidentally are claims about which statements about the object are necessary, he naturally concluded that claims about the essential or accidental properties of objects must in the end be claims about which sentences containing terms designating those objects are true in virtue of meaning. Since there is no direct and natural link tying explicitly linguistic claims about the meanings of sentences with equivalent claims about the essential or accidental properties of objects, he naturally concluded that such claims about objects must be confused.18 Prior to Kripke’s clear articulation and defense of a metaphysical conception of necessity—not tied to or dependent upon linguistic concepts like analyticity—there was no way that philosophers concerned with these issues could have seen them clearly. Once the distinction was made, the clarity and utility of the notion of rigid designation and the intelligibility of essentialist claims became inseparable, and all but irresistible. Rigid Designation, Possible Worlds, and Criteria of “Transworld Identification”

We next connect the notion of rigidity with the nature of possible worlds, and with another issue, or pseudo-issue, that Kripke takes up in lecture 1—the need for criteria of “transworld identification.” It is sometimes said that before we can evaluate the truth or falsity of the claim that it could have been the case that Nixon, say, was so and so, we have to settle the question of who counts as Nixon in different possible worlds. There are several different ideas connected with this that Kripke rejects. First, if—for some reason (perhaps simply because you are taken in by the world-terminology)—you think that possible worlds are large concrete objects—alternative universes that really exist, but do so in a part of space and time inaccessible to us—then it may seem obvious to you that each of us inhabits only one world (just as each of us inhabits only one location at a time on Earth). On this picture, none of the individuals who exist in the other worlds can be identical with any of us, Nixon included. According to this picture, the The story of this line of reasoning is very well told by John Burgess in “Quinus ab omni naevo vindicatus.” As Burgess points out, it did not help the dialectical situation any that Quine’s main pre-Kripkean opponents—who defended de re modality, essentialism, and quantifying in to modal constructions—standardly explained their conception of necessity by identifying it with analyticity or logical truth.

18

NAMES, ESSENCE, AND POSSIBILITY

355

most one can do is establish criteria for who in those other worlds is similar enough to our Nixon so that they can appropriately be described as playing the “Nixon role.” (As if one could ever learn about such worlds at all.) This is not Kripke’s conception of possible worlds. For Kripke, a possible world is a possible world-state—a way that everything could have been. It is, in effect, a maximal property that the universe could have had.19 To say that there are possible worlds in which Nixon lost the election is just to say that there are properties that the universe could have had which are such that, if the universe had them, Nixon would have lost the election. In specifying these properties—i.e., these world-states—we can refer directly to Nixon himself. We don’t have to come up with descriptive criteria that must be fulfilled if one is to play the “Nixon role.” We may put this another way. In specifying possible world-states, we are not restricted to using general descriptive terms. We are not restricted to saying things like the world-states are those which are such that had the universe been in any one of them, then someone who graduated from a small college in the most populous state in the country and who was a former vice president, who later became president but was forced to resign, would have been so and so. If those were the only kinds of descriptions we could give of possible world-states, then we would need descriptive criteria of identity to find out which individual, if any, could be taken to be Nixon with respect to a given world-state. However, we are not restricted in this way. There is no reason why pure descriptions, not containing any names or other rigid designators, should have preferred status in specifying world-states. We can, if we like, specify a class of world-states as those in which Nixon has a certain property P. Since ‘Nixon’ is a rigid designator, in doing this we are specifying the world-states as those in which a certain individual has P. Given these world-states, we don’t, then, have to decide who Nixon is. Kripke addresses this point in the following passage: The tendency to demand purely qualitative descriptions of counterfactual situations has many sources. One, perhaps, is the confusion of the epistemological and the metaphysical, between a prioricity and necessity. If someone identifies necessity with a prioricity, and thinks that objects are named by means of uniquely identifying 19 For useful discussions of essentially this conception of possible worlds, see Robert Stalnaker, “Possible Worlds,” Nous 10 (1976): 65–75, reprinted in Michael Loux, ed., The Actual and the Possible (Ithaca, NY: Cornell University Press, 1979); and Nathan Salmon, “On the Logic of What Might Have Been,” The Philosophical Review 98 (1989): 3–34.

356

KRIPKE

properties, he may think that it is the properties used to identify the object which, being known about it a priori, must be used to identify it in all possible worlds, to find out which object is Nixon. As against this, I repeat: (1) Generally, things aren’t ‘found out’ about a counterfactual situation, they are stipulated; (2) possible worlds need not be given purely qualitatively, as if we were looking at them through a telescope.20 Although the chief contention of this passage seems, clearly, to be correct, a final point of clarification is in order. When Kripke talks of stipulating a possible world, he doesn’t mean that which things are possible is a matter of our stipulation. He means that which possibilities we choose to single out, or talk about, is a matter that is up to us. It is up to us to stipulate, or specify, which of the possible states that the world genuinely could have been in we are interested in, and wish to make claims about. Further, the fact that we can use the name ‘Nixon’ when we specify a class of possible world-states doesn’t mean that no such specifications can fail. We might try to specify possible world-states in which Nixon has a certain property P, but fail to do so because, in fact, Nixon couldn’t have had that property. For example, we can’t successfully stipulate a possible situation in which Nixon is an inanimate object. In this sort of case there is no such possible world-state corresponding to our specification. That is just to say that our specifications are fallible; indeed, the observation that they are fallible shows that facts about possibility are not created or determined by our stipulations. The same point holds for attempted specifications involving descriptions. In general, descriptions have no priority over names or other rigid designators in specifying possible world-states.

Why Descriptions Normally Do Not Fix the Referents of Names Arguments against the Weak, Reference-Fixing Version of the Description Theory

Let us review where we are. So far in our discussion of Naming and Necessity we have done the following things: (i) We defined the notions of rigid and non-rigid designators, and argued that proper names are rigid, whereas most descriptions associated with names by speakers are not. 20

Naming and Necessity, pp. 49–50.

NAMES, ESSENCE, AND POSSIBILITY

357

(ii) We discussed Kripke’s modal argument that names do not mean the same as non-rigid descriptions that speakers associate with them. In addition, we extended this argument to show that names do not mean the same as descriptions rigidified using the actuality operator. (iii) We clarified the distinction between essential and nonessential properties, and attempted to explain what Kripke means by talk of possible worlds. Having done this, we will next examine a weaker theory about the relationship between names and descriptions. In lecture 2, Kripke considers the possibility that descriptions may, as he says, fix the referents of names, without giving their meanings. The idea is something like this: Even though proper names don’t have meanings in the usual sense, something must be responsible for establishing and maintaining the link between a name and what it refers to. Something must determine what one’s utterance of ‘Aristotle’ refers to. Perhaps descriptions semantically associated with names do this after all. That is, it may be that each proper name is associated with certain descriptions that provide the criteria for determining what it refers to—descriptions that are part of its meaning, and that are mastered by competent speakers, even though they do not provide synonyms for the name.21 Once the descriptions have fixed what the name designates with respect to the actual world-state, the connection between the name and the referent becomes rigid. Thus, with respect to any possible world-state w, the name designates the individual that satisfies the descriptions with respect to the actual world-state—whether or not that individual satisfies those descriptions with respect to w. For example, suppose that we have a sentence Fn, in which the referent of n is semantically fixed by a set D of descriptions. On this theory, the truth conditions of the proposition expressed by this sentence can be thought of as being determined as follows. First, it is determined which object o uniquely 21 In saying that the descriptive reference-fixing conditions are semantically associated with the name, that they are part of the meaning of the name, and that they are mastered by competent speakers, we distinguish the interesting, though contentious, claim that the referent of n is semantically fixed by a description from the trivial and uninteresting claim that it is possible to describe the process by which n got its referent. For any word whatsoever—and, necessarily, if, therefore, obviously, Aristotle, 3, etc.—there is some description that correctly describes the process by which the word acquired its meaning, or its reference. However, that does not mean that all words have their meanings, or referents, semantically fixed by descriptions. It is one thing to describe how words got to have the meanings and referents they do; it is another to say that descriptions are parts of the meanings of words in the manner contemplated by the description theory.

358

KRIPKE

satisfies D with respect to the actual state of the world. Then, given this object, we can see that the proposition expressed by the sentence is true with respect to an arbitrary possible world-state w iff Fx is true of o relative to w. In the process of testing this theory, Kripke isolates corollaries (i–iv) of the theory. The referent of a name n (for a speaker at a time) is semantically fixed by a description, a conjunction of descriptions, or a cluster of descriptions D. If D fixes the referent of n, then: (i) the speaker believes that D applies to a unique individual; (ii) if D does apply to a unique individual o, then o is the referent of n; (iii) if D does not apply to a unique individual, then n has no referent; (iv) the speaker knows (or is capable of knowing) a priori that the sentence If n exists, then n is D expresses a truth. In lecture 2, Kripke goes through these theses one by one and gives what he takes to be counterexamples to them. I won’t belabor this, but will try to say enough to indicate the main idea. In so doing, I will employ a strategy that is implicit in Kripke’s discussion. I will take it that the descriptions that are candidates for fixing the reference of a name n for a speaker are more or less those that the speaker would give, either initially or subject to some reasonable idealizations, if asked Who or what do you use n to refer to?. So, in giving these counterexamples, I will assume that if a speaker uses descriptions to semantically fix the referent of a name, then he would normally either already be aware, or easily be capable of becoming aware, of those descriptions, and be able to supply them if asked.22 Later, when we get to Kripke’s own positive theory about how reference is determined, we 22 Some post-Kripkean descriptivists do not accept this limitation on candidates for descriptions that semantically fix referents. My own view is that once this limitation is abandoned, it becomes difficult to make the important distinction indicated in the previous footnote between descriptions of the pre-semantic causal process by which words acquire meaning and reference, and descriptions which are part of the meanings of terms that are mastered by competent speakers. For a critique of recent versions of descriptivism that, in my view, run afoul of this distinction, see the works cited in footnote 10.

NAMES, ESSENCE, AND POSSIBILITY

359

will consider the question of whether, by relaxing this condition, his own positive theory could be put in the form of a description associated with the name by a speaker. We now consider the corollaries. First, corollary (i). Kripke observes that in the case of many names, the descriptive information speakers associate with the name is too impoverished to pick out an individual uniquely. One example of this is provided by the name Cicero. What do most of us know about Cicero? Many don’t know much more than that he was a famous Roman, perhaps a statesman and orator of some sort. Presumably, however, there was more than one famous Roman statesman and orator. Most of us recognize this. So, Kripke would say, we don’t even believe that the description we associate with the name picks out one individual uniquely. Hence, this is a counterexample to corollary (i). Nevertheless, our use of the name Cicero does refer to a unique person. Kripke concludes from this that the linguistic mechanism that determines the referent of our use of the name must be something other than that maintained by the descriptive theory. There is a point here that is worth noting, which shows that this sort of example is more common than one might first think. Imagine that we have a speaker who knows more about Cicero than that he was a famous Roman statesman and orator. Suppose he knows a certain fact that Quine is fond of stating—namely, that Cicero was the famous Roman statesman who first publicly denounced Catiline. Now this description really does pick out the man Cicero uniquely. So you might think that, for this particular speaker, Kripke’s objection to corollary (i) doesn’t work. However, one should be cautious about drawing this conclusion. For the description in question itself contains a proper name—Catiline. And one might ask the speaker to give his description of Catiline. If the speaker is like many of us, the best he could do would be to say that Catiline is the Roman leader first publicly denounced by Cicero. So what we have is a pair of names, Cicero and Catiline, each associated with a description that determines a unique individual, but only if the name contained within the description already has a reference independently. If the speaker’s information is exhausted by these descriptions, then the description theory will not be able to explain how the referent of either name is determined. This example brings out a strong requirement imposed by the pure form of the description theory, when it is taken to be a theory about how the referents of all names are fixed. What the theory requires is that each name be associated with purely descriptive properties that

360

KRIPKE

are sufficient to uniquely determine its referent. But this is highly counterintuitive. If you ask yourself whether you have such properties associated with every name you use, you will, I am confident, come to the conclusion that you don’t. If this is right, then speakers do not even take themselves to have the sort of descriptive information associated with each name that the pure form of the theory requires. So much for corollary (i). On to corollaries (ii) and (iii). The Cicero-type examples we have been discussing are cases in which we refer to an individual, even though the purely descriptive information at our disposal fails to pick out any individual uniquely. Another kind of case Kripke considers is one in which the problem is not a lack of information, but rather the existence of misinformation. Consider the name Thales. About all that I know about Thales is that he was a pre-Socratic philosopher who held that all is water. Suppose, however, that there was a certain individual called Thales by his contemporaries, or at any rate called by some name which when translated and passed on to us has come down as Thales. Suppose further that the man’s contemporaries attributed a view to him that he never held. Suppose he never held that all was water, but rather believed something more sensible. Nevertheless, the story about him spread and changed, and all that has come down to us now is that Thales held that all is water. In this imaginary case, the description we associate with the name doesn’t designate the person the name really refers to. It could even be that there was some other preSocratic philosopher who was a hermit no one knew about. Even if by some chance he really did hold that all was water, and so satisfied the description we associate with the name, that would not make him Thales. The name we use would not refer to him, but rather to the originally misunderstood philosopher. These points are prima facie indications that corollaries (ii) and (iii) of the description theory are false. Another case of this type that Kripke cites is that of Peano. The main thing that most people who have heard of Peano believe about him is that he was the originator of the now standard axioms of elementary arithmetic—the so-called Peano axioms. In fact, he did publish those axioms, and people remember him for it. However, in a footnote he credited the axioms to another mathematician—Dedekind. The footnote was mostly forgotten, and Peano ended up being credited by most people with the axioms. Let us take it that the axioms really did originate with Dedekind. Before I read Naming and Necessity I don’t think I knew that, though I had heard of the Peano axioms. If I had

NAMES, ESSENCE, AND POSSIBILITY

361

any description associated with the name Peano at that time, it was, I think, the discoverer of the axioms of arithmetic. Still, my use of the name Peano did not refer to Dedekind. Perhaps, though, it will be objected that I had some other description associated with the name that really did refer to Peano. Let’s consider some candidates for such a description. The first candidate is the parasitic description: the person to whom most people refer when they use the name ‘Peano’ The idea behind this suggestion is that a person who doesn’t know enough to describe Peano uniquely on his own can fix the referent of his use of the name via this parasitic description, provided that most other people themselves have the resources to refer correctly and uniquely to Peano. The problem with this idea is that it runs the risk either of failure or of circularity. If most people don’t have other descriptions that independently succeed in referring to Peano, then the description theorist is stuck with the result that those who use our parasitic description fail to refer. But we can imagine cases in which most people lack such non-parasitic descriptions; in fact this may still be true in the case of the name Peano. Nevertheless, people in these situations do refer to someone, and the person they refer to is Peano. What about other candidates for the reference-fixing description? Consider: the person to whom most experts refer when they use the name ‘Peano’ This is no good, since it doesn’t specify what sort of experts—in Italian opera? Renaissance painting? etc.—we are talking about. Do we do any better with the person to whom most mathematicians refer when they use the name ‘Peano’ ? Well, mathematics is a large field, and it could easily happen that most mathematicians have only the Dedekinddescription associated with the name. Nevertheless, they still use the name to refer to Peano, not Dedekind. How about the description the person to whom most Peano-experts refer when they use the name ‘Peano’ ? But this will only take us in a circle, since in order to find out who Peano is, we have first to locate the Peano experts, whereas to find out who is a Peano-expert—i.e., who has expert knowledge of the man Peano—we first must find out who Peano is. A similar circularity affects the person to whom the axioms of arithmetic are standardly attributed

362

KRIPKE

What is it to attribute the axioms of arithmetic to someone? I suppose it is to say of that person that he discovered them. Well, what would people say if we asked them, Who discovered the axioms of arithmetic? A lot would probably answer Peano, thereby saying of whomever that name refers to that he discovered the arithmetical axioms. So who does the name refer to? According to the theory, it refers to the person who satisfies the description the person to whom the axioms of arithmetic are standardly attributed. Thus, in order to determine the person who satisfies the description, we first have to get the referent of the name Peano, but, if this version of the description theory is correct, we can’t do that until we first determine who satisfies the description. Hence, we have gone in a circle again. The upshot of all of this is that it is hard to come up with any non-circular, reference-fixing description that is not subject to clear and obvious counterexamples. On the basis of this, Kripke concludes that corollaries (ii) and (iii) of the reference-fixing version of the description theory are false. This brings us to the final corollary of the description theory. (iv) The speaker knows (or is capable of knowing) a priori that if n exists (existed), then n is (was) D expresses a truth. Why is this corollary part of the theory? Well, if the linguistic rule by which the referent of a name is fixed is that the name is to refer to whoever or whatever is designated by a certain specific description D, then one knows (or is capable of knowing), simply by virtue of knowing this rule, that the sentence If n exists (existed), then n is (was) D cannot fail to be true. Why? Because if D fails to designate anything, then n doesn’t either, and the sentence is trivially true by virtue of the falsity of its antecedent. On the other hand, if D does designate something, then n designates the same thing, and the consequent is guaranteed to be true. Either way, the conditional sentence as a whole is guaranteed to be true.23 All of this one knows (or is capable of knowing) just by knowing the rules of the language, without doing any extra empirical investigation. Though one might certainly raise the question of whether this is what genuine apriori knowledge really amounts to, for the moment we will follow Kripke’s lead in discussing his examples If one is worried that a sentence containing a name that fails to refer may not express a proposition, and so fail to be true, then one may change the statement of the corollary to read, The speaker knows (or is capable of knowing) apriori that if ‘n exists’ expresses a truth, then ‘n is (was) D’ also expresses a truth. Since this issue does not affect the outcome of the present discussion, I will leave it to one side.

23

NAMES, ESSENCE, AND POSSIBILITY

363

and mean by apriori nothing more than that which can be known solely on the basis of understanding one’s language.24 Hence, if the reference-fixing version of the description theory is correct, corollary (iv) should always hold. However, when we consider particular names, we see that it typically fails. Consider the name Columbus, for example. Here, the most salient description that many people associate with the name is something like the European who sailed from Spain in 1492 looking for a new route to Asia, but ended up discovering America instead. However, the linguistic rule determining the referent of the name is surely not that it is, by definition, whoever satisfies this description. For if we ask ourselves how we know that if Columbus really existed, then he was the European who sailed from Spain in 1492 looking for a new route to Asia, but ended up discovering America instead, we surely will not say that we know this apriori, simply by understanding the language. Rather, we know this because we read it in textbooks, and because our teachers told us. And how did our sources know this? Presumably by similar means, involving various kinds of scholarship, testimony, surviving documents, artifacts, and the like. But if that is so, then our knowledge rests on empirical evidence, and so is aposteriori. Moreover, our belief about Columbus remains subject to revision, if some of the historical evidence we rely on is shown to be fake, faulty, inaccurate, or seriously incomplete. Although we don’t expect it to happen, it is certainly conceivable that new evidence might be discovered showing that Columbus never left Spain, but sent someone else in his place. This shows that neither the proposition expressed by (15), nor the claim that sentence (15) expresses a truth, is knowable apriori in the sense we have indicated. 15. If Columbus existed, then Columbus was the European who sailed from Spain in 1492 looking for a new route to Asia, but ended up discovering America instead. Thus it looks like all the major corollaries of the reference-fixing version of the description theory are false—when taken to express universal generalizations covering all proper names. On the basis of this, Kripke concludes that there is no general semantic rule specifying that the referent of a name is the individual designated by descriptions associated with it by speakers. However, this does not mean that we 24

This conception of the apriori will be revisited and critically evaluated in chapter 16.

364

KRIPKE

never fix the referent of a name descriptively. On the contrary, Kripke thinks we can and, in rare cases, do semantically fix the referent of a name by stipulating that it is to be whatever satisfies a certain description. He gives the example of the name Neptune, which he suggests may originally have been introduced as a proper name for whatever caused certain perturbations in the expected orbit of the planet Uranus. Kripke insists that even if the name is introduced by a stipulation that it is to refer to whatever turns out to satisfy a certain description, still the name is not synonymous with the description, but rather is a rigid designator. Moreover, even if the name is first introduced by a reference-fixing description, later, when it is passed from speaker to speaker, the description may be lost and the name may come to be understood without reference to the description. If Kripke is right, such may have been the history of the name Neptune. Surely, however, it is no longer apriori for us that the sentence If Neptune existed, then Neptune caused perturbations in the expected orbit of Uranus expresses a truth (if it ever was). Kripke’s Historical Chain Conception of Reference Determination

So according to Kripke, the reference of a name is occasionally, semantically fixed by a description. But in most cases that is not how the reference is initially determined; and even when it is, the semantic association of the name with the reference-fixing description is likely to be short-lived. This raises the question of how, in the vast majority of cases, the reference of a name is determined. Kripke presents a positive picture of reference-determination that attempts to answer this question. His idea is extremely simple and commonsensical. A particular name is introduced for an object or person. After the name has been introduced by some sort of linguistic baptism, those introducing the name begin to use it in conversation to refer to its bearer. New people hear the name, and begin to use it themselves, intending to refer to the same individual that their sources used it to refer to. This process continues, with the name being passed from one user to the next, with each user forming a link in a chain of reference transmission, or inheritance. Often, some descriptive content will accompany the passing of the name, but typically this content will vary from speaker to speaker, and as the chain of use gets longer, it may end up that for many speakers the name has more misinformation than accurate information

NAMES, ESSENCE, AND POSSIBILITY

365

about its referent associated with it. No matter. The information that a person ends up associating with the name is not what determines its referent. Rather, the referent of a use of a name by a speaker x is determined by the historical chain connecting x’s use to speakers from whom x acquired the name, connecting those speakers to their sources, and leading ultimately back to the individual baptized with the name. According to this picture, if a person picking up a name intends to use it to refer to whatever individual is referred to by his sources for the name, then it usually doesn’t matter very much which other beliefs he may have about the referent. What determines reference is not a speaker’s beliefs, but the chain of use in which the speaker stands. For Kripke, referring is typically not something that one does in isolation; it is a community effort. That is the general picture. It should be noticed, however, that Kripke is not explicit about the exact nature of the chain of use, or about just which facts have to hold in order for a use of a name to stand as one link in a chain that determines later reference. Suppose, for example, that the person from whom I first picked up the name Plato was talking about his neighbor, whom he believed to be very wise. Suppose that after speaking to this person I had many other conversations in which the name Plato was used to describe Socrates’ famous biographer. At a certain point, I read about Plato, and so discovered the name in print. I even read translations of Plato’s work. All of this could be true, even if I wrongly assumed that the person from whom I first heard the name was talking about the same individual as everyone else. In this sort of case, to whom do I now refer when I use the name Plato? Do I refer to the wise neighbor of my ultimate source, as he himself did? Or do I refer to the ancient philosopher, as do most of my other sources for the name? Surely, the latter answer is correct. However, Kripke nowhere presents a precise and explicit theory which gives a clear verdict about potentially problematic cases like this. His aim, he says, is not to provide such a theory, but rather to sketch an alternative picture of how reference standardly works—a picture in which a historical chain of uses somehow links the speaker to the referent, rather than a picture in which the link between speaker and referent is provided by descriptions the speaker associates with the name. The incompleteness of Kripke’s picture is also illustrated by the phenomenon of reference change. Sometimes a name can start off referring to one thing, then go through a period in which that thing is confused with something else, and still later it will count as referring only to the

366

KRIPKE

second thing—even if no one ever consciously tried to redefine the name, or change its reference. Such a case is presented by Gareth Evans in an article called “The Causal Theory of Names.”25 He gives the example of the name Madagascar. He maintains that a version of the name was originally used to designate some part of the mainland of Africa. However, when the Arabs and Europeans arrived, they picked up the name from the inhabitants, mistakenly thinking it was a name of the big island off the southeastern coast of Africa. As Evans tells the story, they used it intending both to refer to the island and to refer to the area that the locals used it to designate, thinking that the two were one and the same. After a period of time, the referent of the name became, unambiguously, the island. If genuine, examples of this sort do not discredit Kripke’s historical conception of how the referent of a name is fixed. However, they do show that there is a substantive, nontrivial question of what goes into establishing the links in the historical chain. Does the Historical Chain Theory Provide Descriptivists with Reference-Fixing Descriptions?

With this in mind, we may turn to the suggestion sometimes made by die-hard descriptivists that what Kripke has done is simply to give a specific kind of reference-fixing version of the description theory of names. Their idea, in simplest terms, is that the reference of a name n for a particular speaker is semantically determined by some description extractable from Kripke’s historical chain theory of reference transmission and inheritance. David Lewis states the idea as follows in his 1997 paper, “Naming the Colours”: Did not Kripke and his allies refute the description theory of reference, at least for names of people and places? Then why should we expect descriptivism to work any better for names of colours and colour experiences? . . . I disagree. What was well and truly refuted was a version of descriptivism in which the descriptive senses were supposed to be a matter of famous deeds and other distinctive peculiarities. A better version survives the attack: causal descriptivism. The descriptive sense associated with a name might for instance be ‘the place I have heard of under the name “Taromeo” ’ or maybe Proceedings of the Aristotelian Society, supplementary volume 47 (1973), 187–208; reprinted in his Collected Papers (Oxford: Clarendon Press, 1985).

25

NAMES, ESSENCE, AND POSSIBILITY

367

‘the causal source of this token: Taromeo’, and for an account of the relation being invoked here, just consult the writings of causal theorists of reference.26 To keep things simple, we may illustrate this idea using the description the individual to which the person or persons from whom I acquired the name referred when they used it. One of the things that makes the idea that a person’s use of a name is fixed by such a description seem plausible is the requirement, recognized by Kripke, that in order for a chain of reference transmission to be created by passing a name from one speaker to the next, the person acquiring the name must intend his reference to be parasitic on the reference of his sources. The descriptivist may be seen as proposing to put this requirement in the form of a description that semantically fixes the referent of the name. Although the idea may appear reasonable, there are, I think, several reasons to be cautious. First, it is not clear that speakers invariably have implicitly in mind, among all the different descriptions they associate with a given name, some precise reference-fixing description for it. We know that the description the individual to which the person or persons from whom I first acquired the name referred when they used it does not always pick out my referent for n. Moreover, it is not fully clear even to theorists precisely which parasitic description of this sort will be sufficient to handle all the different problematic cases. At present there simply is no precise and explicit historical chain theory that one can turn into a description that is adequate for every case. Moreover, even if theorists were to come up with such a description, it is far from obvious that ordinary speakers must have had it implicitly at their disposal all along, whenever they used a proper name. Second, even if we assume that speakers always have an appropriate parasitic description associated with a name, it would have to be shown that they somehow accord this description priority in determining the referent of the name over all other descriptions that they associate with it. This might not be easy. If we were to ask ordinary speakers to give us the descriptions that most reliably specify the referents of the different names they use, it is clear that they would not spontaneously volunteer the relevant, non-circular, parasitic descriptions. P. 353, footnote 22 of “Naming the Colours,” in his Papers in Metaphysics and Epistemology (Cambridge University Press, 1999), originally published in Australasian Journal of Philosophy 75 (1997). Other contemporary descriptivists express similar ideas. For example, see David Chalmers, “On Sense and Intension,” Philosophical Perspectives 16, Language and Mind (Oxford: Blackwell, 2002), 135–82. 26

368

KRIPKE

Perhaps, if we were to guide them through enough Kripke-style thought experiments about reference, we would have some success in eliciting parasitic descriptions that approximate correct results for many cases. The descriptivist might then postulate that, like the slave boy in Plato’s Meno, these ordinary speakers must have had, and unconsciously given priority to, the relevant reference-fixing descriptions all along. Surely, however, this story must be regarded as speculative in the extreme. Third, throughout all of this, it is important to keep in mind the distinction mentioned in footnotes 21 and 22 between descriptions of the pre-semantic causal processes by which words acquire and retain their meanings and/or referents, and descriptions which are parts of the meanings thereby acquired, and hence grasped by competent speakers who have mastered the words. Although there are clearly descriptions of the first sort for every word in our language—including if, and, and but—this does not show that that there are descriptions of the second type for all of them. The doctrine that the referents of names are semantically fixed by descriptions associated with them by speakers is an ambitious and problematic thesis that asserts the existence of descriptions of the second type for all names. Finally, one shouldn’t think that the reference-fixing version of the description theory—thought of as a semantic theory implicitly mastered by competent speakers—somehow has to be correct. It could be that there is a certain process by which reference is passed from speaker to speaker. It could also be that, standardly, when one picks up a name, one intends to use it to refer to whatever the person one picked it up from uses it to refer to, or to whatever it is properly used to refer to in one’s linguistic community as a whole. Once one has acquired the name, one can start using it to express beliefs about the object it stands for. Later, one may forget completely about all aspects of how one acquired the name, while retaining the relevant beliefs one uses the name to express. If, at this later time, one asks what determines the referent of one’s use of the name, then the answer might be that it simply inherits its reference from the beliefs one has been using it to express—whether or not one can come up with any correct description about how one originally acquired the name, or any correct description of one’s later uses of the name that may have changed its original reference. (Think of the Plato case.) If this is right, then there may be a natural process of reference inheritance by which later uses of a name inherit their reference from earlier uses, even if speakers themselves

NAMES, ESSENCE, AND POSSIBILITY

369

don’t have to have any complete and accurate understanding of the process. There is nothing incoherent about this idea. So there is nothing outlandish about the idea that there might be no correct reference-fixing version of the description theory that is part of what one learns when one masters a language.27 Finally, there is, I think, an unclarity in Naming and Necessity about the kind of question we are asking when we ask how the reference of a name is fixed, for a speaker at a time. Often, Kripke writes as if this were a semantic question about a rule that speakers master when they learn a name—a rule of the sort “n” as used in context C refers to (a) whatever satisfies such and such descriptions, or (b) whatever stands at the end of such and such chain of reference transmission. On this way of looking at things, the question How is the referent of a name fixed? is used to express is a request for a semantic rule which is implicitly grasped by speakers, and which could be used by them in specifying the truth conditions of sentences containing the name. But this is not the only type of question that the sentence could be used to asked. Sometimes, for example, it may be used to ask a pragmatic question about which of its various contents a term is used with on a particular occasion. For example, one might wonder what determines whether, on a certain occasion, ‘David’ is used to refer to David Kaplan or to David Lewis, just as one might wonder what determines whether, on a particular occasion, the word ‘bank’ is used to talk about a river bank or a financial institution. The causal sources of these particular uses of the words may very well be relevant to answering these questions. However, the questions are not semantic ones about which meanings are assigned to words; rather, they are questions about how, when a word carries multiple meanings, it is determined which meaning a speaker is using it with. There is still another type of question that How is the referent of a name fixed? may be used to ask. As theorists, we may want to know (i) how a 27 It should also be remembered that even if some weak version of the description theory were correct, according to which the referents of names were semantically fixed by Lewisstyle descriptions, the theory would provide no help in solving the problem that descriptivists have traditionally been most concerned with—namely, specifying the content of the propositions expressed by sentences containing names, and believed and asserted by speakers who accept, or assertively utter, those sentences. When I say that the ancient Babylonians believed that Venus was a star, I am not saying that they believed that the object which is connected by an appropriate historical chain of use to my utterance of the word Venus was a star. So even if, as seems extremely unlikely, some such version of descriptivism could be made to work as a semantic account of reference-fixing, it would not solve the problem of content that historically has been of most concern to descriptivists.

370

KRIPKE

name, or any other expression, initially came to have the meaning and/or reference it does and (ii) in virtue of what aspects of its use by speakers it retains that meaning and reference in the common language of the community. These are theoretical questions that may arise even after it has been settled precisely what the meaning and reference of a given expression is, either in the language of the community as a whole, or as used by a speaker on a particular occasion. These questions are foundational questions about the causal processes that originally endowed expressions with their semantic properties, and that maintain them in the language with the properties they have. When Kripke discusses the fix-the-referent version of the description theory, he is clearly discussing a proposal about the semantics of names. Because of this, his way of framing the general discussion— What fixes reference?—may seem to suggest that the two answers he gives to this question—Descriptions in certain unusual cases and Historical chains of reference transmission for the great mass of proper names—are on a par. Since the claim about descriptions clearly may be understood to be semantic, and hence about the linguistic rules that speakers must master in learning various names, this encourages the unwary reader wrongly to think of the claim about historical chains in the same way. This, I believe, is the source in Kripke’s discussion of the descriptivists’ last-gasp attempt to interpret his historical chain theory of reference transmission as providing the descriptions needed for a correct descriptive theory of the semantics of names. This is what leads them to their view that, details aside, the reference of n for a particular speaker is semantically determined by a description such as the individual to which my sources referred when they used n. In my opinion, this whole way of thinking is wrongheaded. There is nothing special here about the semantics of names. Standardly, when one uses any word in the language of one’s community, one does so with the intention that it should carry whatever meaning and reference it has already acquired. This is a fact about the use of all expressions, not about the semantics of any of them. To the extent that there are additional questions about names, they are pragmatic and foundational questions. Kripke’s comments about initial baptisms should be seen as answers to the foundational question In virtue of what did these terms originally come to refer in the language to what they do? His comments about causal chains of reference transmission should be understood as providing information relevant to answering the pragmatic question How is it determined which of many bearers of the name a particular

NAMES, ESSENCE, AND POSSIBILITY

371

utterance of it refers to? and the foundational question In virtue of what does the name continue to refer in the language to this object? 28 If one then asks, And what should a semantic theory of such names tell us?, the most reasonable answer, I believe, is that it should tell us what the these terms refer to in the common language of the community, and nothing more. However, this takes us beyond what Kripke explicitly committed himself to, and to the brink of one of the fundamental unsolved problems raised by Naming and Necessity. This problem, which is centrally relevant to Kripke’s groundbreaking discussion of the distinction between necessary truth and apriori truth, is one we will take up in the next chapter. 28 This way of looking at Kripke’s discussion of reference-fixing via baptism and historical chains of reference transmission is explained and forcefully defended by my former student Jonathan McKeown-Green in his unpublished Princeton doctoral dissertation, The Primacy of Public Language (2002).

CHAPTER 15 THE NECESSARY APOSTERIORI chapter outline 1. A Framework for Discussing the Necessary Aposteriori and the Contingent Apriori Propositions as contents of sentences, objects of belief, and bearers of truth 2. Genuine Instances of the Necessary Aposteriori: Explanation and Consequences for Our Conception of Inquiry Inquiry as locating the actual world-state within the space of metaphysically possible world-states; the incompatibility of this conception with the necessary aposteriori. Why Kripke’s explanation of genuine instances of the necessary apriori shows that the actual is sometimes epistemologically prior to the possible 3. Identity Statements and the Necessary Aposteriori The gap in Kripke’s arguments that true identity statements involving names are often knowable only aposteriori, and a plausible reconstruction of his reasoning using disquotational principles relating acceptance of sentences to belief in the propositions they express; the apparent falsity of strong disquotational principles; the ineffectiveness of plausible weak disquotational principles; the source of these problems in the nontransparency of meaning; final assessment: Kripke’s arguments concerning names present us with a dilemma 4. A Further Note on Inquiry

A Framework for Our Discussion The claim that there are genuine examples of the necessary aposteriori, and the corresponding claim that there are genuine examples of contingent apriori, are among the most important and far-reaching doctrines of Naming and Necessity. In this chapter, I will explain and evaluate the first of these claims, and the arguments that Kripke gives for it. In chapter 16, I will do the same for the second claim. Kripke’s discussion of these topics may properly be regarded as ground-breaking, and I

THE NECESSARY APOSTERIORI

373

will argue that many of his examples and arguments are illuminating. However, I will also indicate ways in which I believe he muddied the waters by saying certain things that are puzzling at best, and defective at worst. In the case of the necessary aposteriori, much of what he says can, I think, be taken to be straightforwardly correct, though, as we will see both in this chapter and in chapter 17, certain parts of his discussion unnecessarily weaken his case. In the case of the contingent apriori, considerably more reconstruction and revision is needed. Our aim in this chapter and the next will be to clarify and correct these matters, in order to arrive at clear and defensible versions of the two doctrines. In bringing out both the puzzling and the illuminating aspects of Kripke’s discussion, I will, in these two chapters, make use of a modest theoretical framework that goes beyond what he explicitly commits himself to in Naming and Necessity. The central assumptions of this expository framework are the following: A1. Some things are asserted, believed, and known. Propositional attitudes like assertion, belief, and knowledge are relations that hold between agents and the things that they assert, believe, and know. A2. The things asserted, believed, and known may be expressed by sentences, and designated by clauses such as the statement that S, the assertion that S, the belief that S, the claim that S, the proposition that S, or simply that S. I will call the things designated by these clauses propositions. A3. Propositions are bearers of (contingent or necessary) truth and falsity. A4. Propositions are not identical with sentences used to express them. Whatever they turn out to be, propositions are, roughly speaking, things which different sentences that “say” or express the same thing have in common. A5. Propositional attitude ascriptions—x asserts/believes/knows/ knows apriori/knows aposteriori that S—report that an agent asserts, believes, knows, knows apriori, or knows aposteriori the proposition designated by that S. Kripke’s discussion of the necessary aposteriori and the contingent apriori indicates his commitment to the view that (i) there are genuine

374

KRIPKE

cases in which some one thing is both necessarily true and knowable only aposteriori (on the basis of empirical evidence), and (ii) there are genuine cases in which some one thing is both contingent and knowable apriori (without appeal to such evidence). From the point of view of our modest theoretical framework, this is the view that there are some propositions—i.e., things capable of being asserted, believed, and known—that are both necessarily true and knowable only aposteriori, and there are other propositions that are both contingently true and knowable apriori. Although I will argue that this view is correct, I will also raise serious questions about Kripke’s treatment of a substantial range of cases.

Genuine Instances of the Necessary Aposteriori: Explanation and Consequences for Our Conception of Inquiry There is a natural and initially attractive conception of inquiry according to which ignorance about a given subject is a matter of lacking information about which, of certain relevantly different possible states the world could be in, it is actually in; and complete ignorance is a condition in which one doesn’t know which, of all the possible states that the world could be in, it is actually in. According to this conception, when an agent is in this condition, (i) all metaphysically possible states of the world are epistemically possible—i.e., every way that the world could possibly be is a way that, for all the agent knows, it might actually be, and (ii) every epistemic possibility is a metaphysical possibility—i.e., every way that, for all the agent knows, the world might be is a way that the world really could be. Inquiry is the process of escaping from this position of ignorance. By investigating the world or relying on the testimony of others, the agent learns contingent truths that distinguish the way the world actually is from other ways it might possibly be, but isn’t. Each time the agent learns one of these truths, he narrows down the class of metaphysical/epistemic possibilities compatible with what he knows, and within which he locates the way the world actually is. According to this conception, acquiring information is equated with narrowing down the range of metaphysically possible world-states that are compatible with what one knows. We may also speak of the truth of one proposition as providing information supporting the truth of another. On this conception, the truth of a proposition

THE NECESSARY APOSTERIORI

375

p provides information supporting the truth of a proposition q by ruling out certain possible ways in which q might fail to be true. Thus, the truth of p supports the truth of q only if the set of possible world-states with respect to which both p and the negation of q are true is non-empty. There are two immediate consequences of this conception of inquiry. The first is that necessary truths are uninformative. Since they are true with respect to all possible world-states, knowledge of them provides no information, and is irrelevant to locating the way the world actually is within the range of possible ways it might be. Second, there are no necessary truths which, though knowable, are knowable only aposteriori. To say that a proposition q is knowable only aposteriori is to say that one can have the justification required to know q only if one has empirical evidence supporting its truth. However, according to the conception of inquiry just sketched, this is impossible. In order for the truth of any proposition p to support the truth of q, and hence to provide evidence for it, there must be possible world-states with respect to which q is untrue, which are ruled out by the truth of p. Since q is necessary, there are no possible world-states with respect to which it is untrue; hence there can be no evidence for q. This means that, on the conception of inquiry just sketched, there can be no necessary truths which, though knowable, are knowable only aposteriori. Although a number of philosophers have taken this conception of inquiry, and the consequences that follow from it, to be plausible and even axiomatic, the conception is directly challenged by the framework developed by Kripke in Naming and Necessity.1 This challenge is illustrated by the following examples. 1. Gregory Soames is not identical with (i.e., is not the same individual as) Brian Soames. 2. If Saul Kripke exists, then Saul Kripke is a human being. 3. This table is not made out of clay. 4. If this table exists, then this table is made of molecules. It seems evident that each of these sentences expresses a proposition that is knowable only aposteriori, on the basis of some sort of empirical A good example of a philosopher who accepts this conception of inquiry, and extends it to provide a model of discourse, is Robert Stalnaker. See his “Assertion,” Syntax and Semantics 9 (1978): 315–32, reprinted in his Context and Content (New York: Oxford University Press, 1999), and his Inquiry (Cambridge, MA: MIT Press, 1984).

1

376

KRIPKE

evidence. In the case of (1), one needs to find out who Gregory and Brian are, and to assure oneself that they are different. In the case of (2), if the question were ever raised as to whether Kripke was a sophisticated robot, or an alien sent from another world, one would need empirical evidence to rule out these possibilities—though, of course, given their fanciful nature, not much evidence would be required. A similar point is true of (3), the justification of which might be provided by a cursory examination of the table. In the case of (4), the evidence required to know the truth that it expresses is much greater, and more sophisticated. Still, since in all four cases empirical evidence is required to know the truths expressed, all four propositions are knowable only aposteriori. They are also necessary. In each case, the subject expression is a rigid designator—the names Gregory Soames and Saul Kripke, plus the demonstrative phrase this table.2 Because of this, the sentences express necessary truths iff the properties they attribute to the referents of their subjects are essential properties—the properties of being nonidentical with Brian Soames, of being human, of being not made out of clay, and of being made of molecules. These do seem to be essential properties; in fact they seem to be essential properties of anything that has them. For example, it is plausible to think that any individual who really is not the same individual as Brian Soames could not have existed while being the very same individual as Brian Soames. Because the name Brian Soames is itself a rigid designator, we can also make the point in another way: since the property of being non-identical is an essential property of any pair of things that have it, if two individuals (such as my two sons) really are non-identical, then there is no possible circumstance in which they are one and the same individual. Similar points hold for the other properties mentioned in these examples—anything that really is human could not have existed without being human, any object not made out of clay could not have existed while being (originally and entirely) made out of clay, and anything that really is made up of 2 To say that an indexical phrase, like this table, is a rigid designator is to say, roughly, that when one uses it in context of utterance to refer to a particular thing, that thing remains its referent with respect to all possible world-states with respect to which we might evaluate a sentence containing it. So if I say, This table is made of wood, referring to the table t now directly in front of me, then I say something (express some proposition) which is true, when taken as a description of any possible world-state w, iff had the world been in state w, t (that very object) would have been made of wood. See David Kaplan, “Demonstratives,” in Joseph Almog, John Perry, and Howard Wettstein, eds., Themes From Kaplan (New York and Oxford: Oxford University Press, 1989).

THE NECESSARY APOSTERIORI

377

molecules could not have existed without being made up of molecules. Thus, sentences (1–4) all express necessary truths. Since they are also knowable only aposteriori, they are examples of the necessary aposteriori. How can this be? How can a proposition that is necessary (and known to be necessary) be knowable only aposteriori? Kripke’s answer appeals to our knowledge of which properties are essential. He argues, quite plausibly, that we know apriori that properties like non-identity, being human, being not made out of clay, and being made out of molecules are essential properties of the things that have them. So we know apriori that if things have these properties, then they have them necessarily. This means that the propositions expressed by (1–4) are such that we know apriori that if they are true, then they are necessarily true. Still, finding out that they are, in fact, true requires empirical investigation. This means that sometimes, in order to find out whether certain things are true with respect to all possible states of the world, and other things are true with respect to no possible states of the world, we first must find out what is true with respect to the actual state of the world. Sometimes in order to find out what could and could not be, one first must find out what is. This insight is incompatible with acceptance of the conception of inquiry sketched above. On that conception, necessary truths are uninformative, epistemic possibility is restricted to metaphysical possibility, and all inquiry is a matter of narrowing down the range of (metaphysically) possible states that the world could genuinely be in. Kripke’s arguments may plausibly be taken to show that these views, and the conception on which they are based, are simply mistaken. The central problem with the conception is its restriction of ways things could conceivably be to ways things could really be—i.e., its restriction of epistemic possibility to metaphysical possibility. This is something that Kripke rejects. Instead of identifying these two kinds of possibility, he sharply distinguishes them.3 Once this is done, and both rigid designation and the existence of nontrivial essential properties of objects are accepted, the necessary aposteriori follows unproblematically.4 As we will see in chapter 17, this point requires further discussion. Near the end of lecture 3 of Naming and Necessity, Kripke responds to an objection to his conception of the necessary aposteriori in a way that may seem to cast doubt on his commitment to a sharp and robust distinction between epistemic and metaphysical possibility. In the end, I will argue that the passage is both misleading and something of an anomaly. However, since it occurs in his discussion of natural kind terms, we will wait until then to analyze it. 4 Essentially this explanation of the necessary aposteriori is given by Kripke on pp. 151–53 of “Identity and Necessity,” in Milton Munitz, ed. (New York: NYU Press, 1971). 3

378

KRIPKE

How, then, should we think of inquiry? Is there any way of modifying the original conception that retains its attractive features, while avoiding its errors? Although Kripke doesn’t address this question explicitly, there is a natural strategy for doing this that makes use of materials he provides. Remember, for Kripke, possible states of the world are not alternate concrete universes, but abstract objects. They are maximally complete ways the real concrete universe could have been—maximally complete properties that the universe could have instantiated. Thinking of them in this way suggests an obvious generalization. Just as there are properties that certain objects could possibly have and other properties that they couldn’t possibly have, so there are certain maximally complete properties that the universe could have had—possible states of the world—and other maximally complete properties that the universe could not have had—impossible states of the world.5 If some of the properties that objects couldn’t have had are, as our examples (1–4) indicate, properties that one can conceive them as having, then surely some maximally complete properties that the universe could not have had (some impossible states of the world) are properties one could conceive it as having. Given this, one could explain the informativeness of certain necessary truths as resulting from the fact that learning them allows one to rule out certain impossible, but nevertheless conceivable, states of the world. Moreover, one could explain the function played by empirical evidence in providing the justification needed for knowledge of the necessary propositions expressed by sentences (1–4); empirical evidence is required to rule out certain impossible, but nevertheless conceivable and epistemologically relevant, world-states with respect to which the propositions are false.6 Thus, by expanding the range of epistemically conceivable states of the world to include some that are metaphysically impossible, one can modify the original conception of inquiry so as to accommodate Kripkean examples like (1–4). Whether or not the modification is sufficient to save the conception from further problems is a question to which we will return later.

For an explanation and defense of this conception of world-states (in response to a different problem), see Nathan Salmon, “The Logic of What Might Have Been,” The Philosophical Review 98, 1 (1989): 3–34. 6 Here, I assume that names and indexicals (unlike certain definite descriptions such as the stuff out of which this table, if it exists, is constituted) rigidly designate the same thing with respect to all world-states, possible or not. 5

THE NECESSARY APOSTERIORI

379

Identity Statements and the Necessary Aposteriori The Gap in Kripke’s Argument

I now turn to what may be the most well-known examples that Kripke gives of the necessary aposteriori—identity statements involving rigid designators. 5. Hesperus is Phosphorus. 6. Hesperus is Hesperus. 7. a ⫽ b. If a sentence of the form (7) is true, then the term occupying the position of ‘a’ and the term occupying the position of ‘b’ refer to the same thing. If they are rigid designators, this means that they refer to the same thing with respect to all world-states in which that thing exists (and never refer to anything else). To keep things simple, let’s ignore world-states in which the thing doesn’t exist. Then we know that any true sentence of the form (7) in which the terms are rigid designators is a necessary truth; so (5) and (6) are necessary truths. (We take is in (5) and (6) to be the is of identity.) Of course, if we replace one of the names with a non-rigid designator, as in (8), then the example may be true without being necessary. 8. The planet seen at a certain place in the evening sky in certain seasons is Hesperus. In lecture 2 of Naming and Necessity, Kripke considers an objection to this view which he attributes to Quine. There’s a dispute about this between Quine and Ruth Barcan Marcus. Marcus says that identities between names are necessary. If someone thinks that Cicero is Tully, and really uses ‘Cicero’ and ‘Tully’ as names, he is thereby committed to holding that his belief is a necessary truth. She uses the term ‘mere tag’. Quine replies as follows, ‘We may tag the planet Venus, some fine evening, with the proper name “Hesperus”. We may tag the same planet again, some day before sunrise, with the proper name “Phosphorus”. When we discover that we have tagged the same planet twice our discovery is empirical.7 7

Naming and Necessity, p. 100.

380

KRIPKE

Quine concludes from this that the statement expressed by (5) must be contingent—presumably because it is not knowable apriori. Kripke’s reply is, essentially, that this objection rests on the incorrect assumption that necessity and aprioricity come to the same thing. Once this is recognized, the objection collapses. There is, however, a slightly different objection that could be read into Quine’s comment about our “discovering that we have tagged the same planet twice.” This phrase suggests that perhaps what he has in mind is (9). 9. ‘Hesperus’ and ‘Phosphorus’ refer (in our language) to the same thing. The proposition expressed by (9) really is contingent, in addition to being knowable only aposteriori. However, the contingency of (9) is no argument against the necessity of (5), since the two sentences clearly express different propositions. Thus, Kripke’s position remains intact. Having considered these objections to his view that true identity statements involving names are necessary and aposteriori, Kripke devotes the last four pages of lecture 2 (the bottom of page 101 to the top of page 105) to explaining in detail why his view is correct. The section is too long to quote, so I will summarize it. The view Kripke presents goes essentially as follows: Let a ⫽ b be a true identity sentence involving proper names. These names may either be ordinary names like Cicero and Tully, or special names like Hesperus and Phosphorus the understanding of which may involve associating them with reference-fixing descriptions. In either case, Kripke argues, the evidence available to a competent user of the names is insufficient to determine that they are coreferential. He illustrates this point by noting that there is a possible state of the world in which speakers are in an evidentiary situation that is qualitatively identical with the one which we actual speakers are in, and yet, in the merely possible situation, the names are used to refer to different things. For example, there is a world-state with respect to which speakers fix the referent of the name Hesperus just as we do in the actual world—by pointing to a bright object that appears in the evening in a certain part of the sky in certain seasons. Furthermore, speakers in that possible world-state fix the referent of the name Phosphorus by pointing to a bright object that appears in the morning in certain seasons. From a qualitative point of view, these speakers are in the same evidentiary situation regarding their uses of

THE NECESSARY APOSTERIORI

381

the names as we are. Yet their uses of the names refer to different things. Kripke describes the case as follows: The evidence I have before I know that Hesperus is Phosphorus is that I see a certain star or certain heavenly body in the evening and call it ‘Hesperus’, and in the morning and call it ‘Phosphorus’. I know these things. There certainly is a possible world in which a man should have seen a certain star at a certain position in the evening and called it ‘Hesperus’ and a certain star in the morning and called it ‘Phosphorus’; and should have concluded—should have found out by empirical investigation—that he names two different stars, or two different heavenly bodies. At least one of these stars or heavenly bodies was not Phosphorus, otherwise it couldn’t have come out that way. But that’s true. And so it’s true that given the evidence that someone has antecedent to his empirical investigation, he can be placed in a sense in exactly the same situation, that is a qualitatively identical epistemic situation, and call two heavenly bodies ‘Hesperus’ and ‘Phosphorus’, without their being identical. So in that sense we can say that it might have turned out either way.8 Kripke intends this example to show that the evidence available to us in the actual state of the world, as well as to agents in qualitatively similar world-states, simply by virtue of being competent users of the names, is insufficient to show that the names are coreferential. We may express this idea as follows: Let E be the collection of metaphysically possible world-states in which the epistemic situation of agents regarding their uses of the terms Hesperus and Phosphorus is qualitatively identical with our actual epistemic situation. Kripke may have been thinking that for any proposition p, if it is not the case that p is true in all members of E, then p is a proposition which is not determined to be true by the qualitative evidence available to us, and so is one that we do not, and cannot, know apriori, simply on the basis of our mastery of the relevant terms or concepts. Let us grant this for the sake of argument. Well, one proposition that fails to be true in all members of E is the proposition that the names Hesperus and Phosphorus are coreferential; another closely related proposition is the one expressed by (10).

8

Ibid., pp. 103–4.

382

KRIPKE

10. The sentence ‘Hesperus is Phosphorus’ expresses a truth in our language. Thus, Kripke is in a position to conclude that the metalinguistic claim which says that the sentence ‘Hesperus is Phosphorus’ expresses a truth in our language is something that we cannot know apriori, but rather can come to be known only on the basis of empirical investigation. So far so good. However, there is a problem. The lesson Kripke explicitly draws from this example is not that a certain metalinguistic claim is knowable by us only aposteriori, but rather that the claim that Hesperus is Phosphorus is knowable by us only aposteriori. So two things are true: first, that we do not know a priori that Hesperus is Phosphorus, and are in no position to find out the answer except empirically. Second, this is so because we could have evidence qualitatively indistinguishable from the evidence we have and determine the reference of the two names by the positions of the two planets in the sky, without the planets being the same.9 The problem is that Kripke’s conclusion does not follow from his apparent premises. The proposition that Hesperus is Phosphorus is, as he insists, true in all metaphysically possible world-states. So it is true in all members of the class E of such world-states in which agents are in an epistemic situation qualitatively identical to ours. Since it is true with respect to those world-states, the principle that only propositions true in all members of E are knowable by us apriori does not rule out that we may know it apriori, even though it does rule out that we can know apriori that ‘Hesperus is Phosphorus’ expresses a truth in our language. Since the proposition expressed by (5) is not the same as the proposition expressed by (10), showing the latter to be knowable by us only aposteriori is not enough to establish that the former is knowable only in the same way. The point I am making depends on sharply distinguishing (5), on the one hand, from (10), on the other. When explaining the necessity of (5), Kripke uses his example of possible world-states in which agents are in an epistemic situation qualitatively identical to ours to remind us that the contingency of (10) is irrelevant to the necessity of (5). According to Kripke, the agents in these world-states use the sentence ‘Hesperus is Phosphorus’ to express a different proposition from the proposition we actually use it to express. The fact that the proposition 9

Ibid., p. 104.

THE NECESSARY APOSTERIORI

383

they use it to express is false with respect to their states of the world does not show that the proposition we actually use it to express is false at any world-state. What Kripke fails to point out is that the same reasoning can be applied to the epistemic status of the two examples. The proposition expressed by sentence (10) is knowable only aposteriori. But how does that bear on the question of whether proposition (5) is knowable apriori? The agents of Kripke’s imagined world-states do not know the proposition they use the sentence ‘Hesperus is Phosphorus’ to express, for the simple reason that the proposition that they use it to express is false with respect to their world-states. But how does this show that the different proposition that we use the sentence to express isn’t known by us, or that it isn’t known by us independent of empirical investigation? Until we can answer this question, we have no way of viewing Kripke’s discussion as supporting his conclusion that the claim that Hesperus is Phosphorus is not knowable apriori and so qualifies as an example of the necessary aposteriori. Kripke’s Informal Strategy for Filling the Gap

Although there is undoubtedly a gap in Kripke’s argument, it is not an inexplicable one. Throughout the passage, he exploits a very familiar and highly intuitive connection between speakers’ understanding and acceptance of sentences and our ability to use those sentences to report what they believe. For example, if you know that I fully understand and sincerely accept the sentence Trenton is a town in central New Jersey, then normally you will feel justified in reporting that I believe that Trenton is a town in central New Jersey. Similarly, if you know that I fully understand the sentence but do not accept it—either because I am uncertain of its truth or because I believe it to be false— then normally, and all other things being equal, you will feel justified in reporting that I don’t believe that Trenton is a town in central New Jersey. The same is true of the sentence Hesperus is Phosphorus. In Kripke’s example, before we made or learned of the astronomical discovery, we understood but did not accept the sentence; hence it is natural, and all but inevitable, to conclude that at that time we did not believe that Hesperus is Phosphorus. Moreover, as Kripke emphasizes with his evocation of agents in circumstances qualitatively identical to ours, we would not have been justified in accepting the sentence Hesperus is Phosphorus based on the evidence we had at that time. Because of this, it is natural to conclude that we wouldn’t have been justified in believing

384

KRIPKE

that Hesperus is Phosphorus based on the evidence we then had. But if that is so, then knowledge that Hesperus is Phosphorus must require empirical justification, in which case it must not be knowable apriori that Hesperus is Phosphorus—exactly as Kripke concludes. This, I believe, is what Kripke had in mind. Although there is much to be said for this pattern of reasoning, there is a potential puzzle lurking within it. In order to bring out the puzzle, I will first use our modest theoretical apparatus of propositions as objects of belief and knowledge to formulate explicit and generalized premises to fill the gap in Kripke’s argument. Next, I will scrutinize these new premises and show that either their truth is highly doubtful, or they are not strong enough to allow Kripke to derive his conclusion. Having isolated sources of potential difficulty, I will return to the informal reasoning just sketched, expose the puzzle that lies behind it, and evaluate Kripke’s conclusion about this putative example of the necessary aposteriori. A Formal and Explicit Reconstruction of Kripke’s Argument

The informal pattern of reasoning I have attributed to Kripke posits a close connection between our attitudes toward sentences and the beliefs those sentences are used to express. Using the terminology of propositions, we may express this thought as follows: Since sentences are vehicles for expressing propositions, the cognitive attitudes (belief, knowledge, etc.) that we bear to propositions expressed by sentences are mediated by the attitudes we bear to sentences that express them. Often, our believing a certain proposition goes hand in hand with understanding and accepting a sentence that expresses it. One initially plausible conception of the systematic connection between understanding and accepting a sentence, on the one hand, and believing the proposition it expresses, on the other, is stated by the following “strong disquotation” principle.10 strong disquotation A sincere, reflective, rational individual i who understands a sentence S is disposed to accept S, and believe S to be true, iff i believes the proposition semantically expressed by S. Thus, if S is a sentence of English, then a reflective, rational To avoid complications, we will understand this principle as restricted to sentences that don’t contain indexicals like ‘I’, ‘now’, and so on.

10

THE NECESSARY APOSTERIORI

385

individual i who understands S satisfies the formula x believes that S iff i accepts S, and believes S to be true; if S is not a sentence of English, but is translatable into English as P (where S and P express the same proposition), then such an individual i satisfies x believes that P iff i accepts S, and believes S to be true. Agents in an epistemic situation similar to ours before the astronomical discovery don’t accept the sentence Hesperus is Phosphorus, and so they don’t believe what they express by the sentence. Similarly, prior to the astronomical discovery we didn’t accept the sentence, so, according to the principle, at that time, we didn’t believe that Hesperus is Phosphorus. Moreover, the evidence available to both of us by virtue of our understanding the sentence is such that we would not have been justified in accepting the identity sentence on the basis of that evidence. With this in mind, one might formulate the following principle involving disquotation and justification: strong disquotation and justification A sincere, reflective, rational individual i who understands S and is in possession of evidence e would be justified in accepting S, and believing it to be true, on the basis of e iff i’s possession of e is sufficient to ensure that i would be justified in believing the proposition semantically expressed by S. So, if S is a sentence of English, then i’s possession of e is sufficient to ensure that i satisfies the formula x would be justified in believing that S iff i would be justified in accepting S, and believing S to be true, on the basis of e; if S is not an English sentence, but is translatable into English as P (where S and P express the same proposition), then i’s possession of e is sufficient to ensure that i satisfies x would be justified in believing that P iff i would be justified in accepting S, and believing it to be true, on the basis of e. If these two principles are accepted, then Kripke’s argument at the end of lecture 2 can be reconstructed as follows: (i) Since there are possible situations in which Hesperus is Phosphorus expresses something false, even though the agents in those situations are perfect reasoners who have evidence qualitatively identical with the evidence available to us simply on the basis of our linguistic competence,

386

KRIPKE

the evidence available to us simply on the basis of our linguistic competence does not justify our accepting the sentence. (ii) So, by the strong disquotation and justification principle, the evidence available to us simply by virtue of our competence, plus our reasoning correctly about it, is not enough to justify us in believing that Hesperus is Phosphorus. (iii) If the belief that Hesperus is Phosphorus were justifiable apriori, then it would be justifiable by the evidence available to us by virtue of our linguistic competence, plus our reasoning correctly about it. (iv) So that belief is not justifiable apriori. Hence, it is not knowable apriori that Hesperus is Phosphorus. This reconstruction of Kripke’s argument has the virtue of being logically valid; the conclusion that it is not knowable apriori that Hesperus is Phosphorus is a logical consequence of Kripke’s premises about the attitudes of agents toward the sentence Hesperus is Phosphorus plus the supplementary premise of strong disquotation and justification. Nevertheless, the argument is problematic—in part because there are reasons to suspect that the principles of strong disquotation and strong disquotation plus justification are untrue. One such reason is that these principles have the consequence that in order to believe a proposition, one must be disposed to accept every sentence one understands that semantically expresses it. Thus, the principles leave no room for the possibility that an individual might understand two sentences that semantically express the same proposition, without knowing that they do, and so might accept one of the sentences while not accepting the other. (In such a case the strong disquotation principle leads to the contradictory conclusion that the agent both believes and does not believe one and the same proposition.) Since there is reason to think that such possibilities are genuine, there is reason to reject the strong disquotation principles.11 One example of the problem with the strong disquotation principle is brought out by Kripke himself in his example of puzzling Pierre, discussed in his paper, “A Puzzle about Belief,” published in 1979, For further discussion, see pp. 10–13 and chapter 3 of Beyond Rigidity. See also Nathan Salmon, “A Millian Heir Rejects the Wages of Sinn,” in C. A. Anderson and J. Owens, eds., Propositional Attitudes: The Role of Content in Logic, Language, and Mind (Stanford, CA: CSLI, 1990), 215–47, at pp. 220–22. Also, Stephen Rieber, “Understanding Synonyms without Knowing That They Are Synonymous,” Analysis 52 (1992): 224–28. 11

THE NECESSARY APOSTERIORI

387

nine years after he delivered the lectures that became Naming and Necessity.12 Kripke’s Pierre is a Frenchman who has grown up in Paris speaking French, who sees picture postcards of London and forms the belief that London is pretty, which he expresses by saying Londres est jolie. Later, he moves to London, learns English, not by translation but by immersion, and comes to live in a poor and unattractive part of the city. On the basis of his experience, he forms a belief that he expresses by saying London is not pretty. It is not that he has given up the belief he formed in Paris on the basis of the picture postcards. He still affirms Londres est jolie when speaking French to old friends, even though he does not accept the English sentence London is pretty. The reason for this disparity is that he doesn’t realize that Londres and London name the same city. This doesn’t mean that he fails to understand the two sentences. He understands the former as well as he and his Frenchspeaking friends did while he was living in France, which was certainly well enough to assert and communicate his belief that London is pretty, and he understands the latter as well as his monolingual English-speaking neighbors in London do, who surely count as competent speakers. Moreover, the sentences are translations of one another; they mean the same thing and, it would seem, express the same proposition. But now we have a problem. By the strong disquotation principle (right to left direction) we get the result that Pierre does not believe the proposition expressed by London is pretty, since he understands but does not accept London is pretty. By a corresponding application of the strong disquotational principle (left to right direction) we get the result that he does believe the proposition expressed by Londres est jolie. Since the standard translation from French to English tells us that these sentences express the same proposition, we get the conclusion that Pierre both does and does not believe one and the same proposition. Since this is a contradiction, we have a reductio ad absurdum of the conjunction of the principle of strong disquotation with the standard (meaningpreserving) translation of French to English. This provides reason to doubt the strong disquotational principles. 12 Saul Kripke, “A Puzzle about Belief,” in A. Margalit, ed., Meaning and Use (Dordrecht: Reidel, 1979); reprinted in N. Salmon and S. Soames, eds., Propositions and Attitudes (Oxford: Oxford University Press, 1988). Kripke’s formulation of the strong disquotational principle in that article is more informal and less theoretically loaded than mine. Although these differences are not without consequence, and are well worth studying, I will not discuss them here. Suffice it to say that they do not affect the difficulties we are about to uncover.

388

KRIPKE

One natural response to this problem is to replace the strong disquotational principles—which are biconditionals—by the following weak disquotational principles—which are merely conditionals. weak disquotation If a sincere, reflective, rational individual i who understands a sentence S is disposed to accept S, and believe it to be true, then i believes the proposition semantically expressed by S. If S is a sentence of English, i thereby satisfies the formula x believes that S; if S is not a sentence of English, but is translatable into English as P (where S and P express the same proposition), then i satisfies the formula x believes that P. weak disquotation and justification If a sincere, reflective, rational individual i who understands S and is in possession of evidence e would be justified in accepting S, and believing S to be true, on the basis of e, then i’s possession of e is sufficient to ensure that i would be justified in believing the proposition semantically expressed by S. So, if S is a sentence of English, then i’s possession of e is sufficient to ensure that i satisfies the formula x would be justified in believing that S, if i would be justified in accepting S, and believing S to be true, on the basis of e; if S is not an English sentence, but is translatable into English as P (where S and P express the same proposition), then i’s possession of e is sufficient to ensure that i satisfies x would be justified in believing that P, if i would be justified in accepting S, and believing S to be true, on the basis of e. When the principle of weak disquotation is substituted for that of strong disquotation in the story about Pierre, we no longer get the contradictory result that Pierre believes and does not believe the same thing. Instead, we get the weaker, and potentially defensible, result that he has contradictory beliefs; he believes that London is pretty by virtue of understanding and accepting the French sentence Londres est jolie, and he believes that London is not pretty (by virtue of understanding and accepting the English sentence London is not pretty).13

Although Kripke himself draws no firm conclusion about the correctness or incorrectness of the principle of weak disquotation based on this case, he does find it puzzling and problematic that the principle leads to the result that Pierre—who may be fully rational and logical—has

13

THE NECESSARY APOSTERIORI

389

However, even if this result is acceptable, and the principle of weak disquotation and justification is substituted for that of strong disquotation and justification in our reconstructed version of Kripke’s argument, the argument still does not go through. The problem now is that the conclusion no longer follows from the premises. In this argumentative scenario, all that follows from our not being justified in accepting the sentence (5), Hesperus is Phosphorus, solely on the basis of our understanding it, is that if the proposition it expresses can be known, or justifiably believed, apriori, then such knowledge or belief must arise from something other than understanding and accepting that very sentence. This leaves open the possibility that there might be some sentence other than (5) which both expresses the proposition that Hesperus is Phosphorus, and is such that one can be justified in accepting it, and believing it to be true, solely on the basis of one’s understanding of it. Since nothing in our new reconstruction of Kripke’s argument rules out this possibility, the version of the argument employing the weak disquotational principles does not entail his conclusion. Although the conclusion is entailed by the version of the argument employing the strong disquotational principles, there is little comfort in this, since, as we have seen, they appear to be false. The Lesson to be Learned: The Nontransparency of Meaning

Although the two formal reconstructions of Kripke’s arguments fail in different ways, the source of the failure is the same in both cases—the nontransparency of meaning. Two sentences may mean the same thing, and hence semantically express the same proposition, even though a competent speaker who understands both sentences does not realize this, and so accepts one, and believes it to be true, while refusing to accept the other, and either believing it to be false, or suspending judgment on it. This was Pierre’s situation with the sentences Londres est jolie and London is pretty. Since the principle of strong disquotation is incompatible with this seemingly obvious truth, it should, I believe, be rejected—in which case the first reconstruction of Kripke’s argument must be judged unsound.

contradictory beliefs. For a defense of both weak disquotation and this result, see Nathan Salmon, Frege’s Puzzle (Cambridge, MA: MIT Press, 1986), and my “Direct Reference, Propositional Attitudes, and Semantic Content,” originally published in Philosophical Topics 15 (1987): 47–87; reprinted in Salmon and Soames, Propositions and Attitudes.

390

KRIPKE

The second reconstruction of the argument—relying on weak disquotation—fails because it does not rule out the possibility that there might be some sentence S other than Hesperus is Phosphorus which semantically expresses the same proposition as it does, and which is such that understanding S is sufficient for accepting it and knowing it to be true. How, one might wonder, could there be such a sentence? Well, think again about names. If, as Kripke contends, names do not have descriptive meanings or senses, what do they mean? One natural thought is that the meaning of a name is its referent. If this thought is correct, then coreferential names like Hesperus and Phosphorus mean the same thing. But surely if two names mean the same thing, then substitution of one for another in a simple sentence like (5) should preserve the meaning of the sentence. By this reasoning, we may reach the conclusion that the sentences Hesperus is Phosphorus and Hesperus is Hesperus mean the same thing, and hence semantically express the same proposition. Surely, it is reasonable to think that understanding Hesperus is Hesperus is sufficient for accepting it, and coming to know that it expresses a truth. Thus, it is natural to conclude that it is knowable apriori that Hesperus is Hesperus. But then, if the proposition that Hesperus is Phosphorus is nothing more than the proposition that Hesperus is Hesperus, it must be knowable apriori, contrary to Kripke. Although this line of reasoning may initially seem far-fetched, and although it certainly goes beyond Kripke, there is nothing in Naming and Necessity to refute it. Moreover, once it has been recognized that meaning is not always transparent, one cannot object to the claim that Hesperus is Hesperus and Hesperus is Phosphorus mean the same thing—and hence semantically express the same proposition—simply by observing that it is possible for someone to understand both sentences, and yet not know that they mean the same thing. This surely is possible, but since meaning isn’t transparent, this possibility doesn’t establish that the sentences mean different things. Without establishing this, there is no way to use our disquotational principles to get Kripke’s desired result. Of course, Kripke didn’t state any principles of strong or weak disquotation in Naming and Necessity, nor did he invoke our modest theoretical apparatus of propositions as objects of belief and knowledge. Hence, one might wonder about the relevance of our findings for his informal argument at the end of lecture 2. As we shall see, there is cause for concern.

THE NECESSARY APOSTERIORI

391

His informal argument that it is not knowable apriori that Hesperus is Phosphorus is roughly the following: (i) For someone who understands the sentence Hesperus is Phosphorus, sincerely accepting it and believing it to be true goes hand in hand with believing that Hesperus is Phosphorus—i.e., one who understands Hesperus is Phosphorus accepts it and believes it to be true iff one believes that Hesperus is Phosphorus. (ii) Similarly, one who understands the sentence Hesperus is Phosphorus would be justified in accepting it and believing it to be true iff one would be justified in believing that Hesperus is Phosphorus. (iii) In order to be justified in accepting Hesperus is Phosphorus and believing it to be true, it is not sufficient for one simply to understand it; in addition, one needs empirical evidence that the two names refer to the same thing. (iv) Therefore, understanding the sentence Hesperus is Phosphorus is not sufficient for one to be justified in believing that Hesperus is Phosphorus; in order for one who understands the sentence to be justified in believing that Hesperus is Phosphorus, one must have empirical evidence that the two names refer to the same thing. (v) Therefore, the statement that Hesperus is Phosphorus is not knowable apriori. Principles (i) and (ii) are informal Kripkean counterparts of specific instances of the principles of strong disquotation and strong disquotation and justification. Thus, in order to accept Kripke’s informal argument, one must have good reason to believe that these principles are true.14 I do think that this much strong disquotation was implicit in Kripke’s argument in Naming and Necessity. With this in mind, consider the following passage from footnote 44 of “A Puzzle About Belief ”: “some earlier formulations expressed disquotationally such as ‘It was once unknown that Hesperus is Phosphorus’ are questionable in light of the present paper (but see the previous note for this case).” The point here is that although it is unquestionable that it was once unknown that Hesperus is Phosphorus expressed a truth, we now see that we should not jump to the conclusion that it was once unknown that Hesperus is Phosphorus. This is questionable because it may just be that to know that Hesperus is Hesperus is to know that Hesperus is Phosphorus. But if that is so, then the argument at the end of lecture 2 of Naming and Necessity is undermined. I believe Kripke did not see this possibility at the time of the lectures. Nevertheless, footnote 44 continues: “I was aware of this question at the time Naming and Necessity was written, but I did not wish to muddy the waters further than necessary at that time. I regarded the distinction between epistemic and metaphysical necessity as valid in any case and

14

392

KRIPKE

Let’s just concentrate on (i), which is a particular instance of the following more general principle. (i*) Let S be an non-indexical sentence of English and let i be a competent speaker of English who understands S. Then, the attitude ascription x believes that S is true of i iff i accepts S. Here is an example scenario that creates problems for (i*): A student at the local college, Martin Martin, is both the quarterback of the football team and the best math student in school. His math teacher, Professor McX, has graded his work and consulted other professors about him. On that basis, she forms the opinion that Martin is a brilliant mathematician, which she expresses by saying Martin is a brilliant mathematician. Since her class is large, she doesn’t know him by sight, so when she decides to attend one of the football games on Saturday, and sees him turn in an astounding performance on the field, she does not realize that the talented quarterback she is watching is her student, Martin. Sitting in the stands with her friend Harriet, she makes numerous comments about Martin—pointing him out and saying things like He is a wonderful athlete. Later Harriet, who knows Martin’s girlfriend, reports that Professor McX thinks that Martin is a wonderful athlete. Surely, Harriet’s report is true; the professor does believe this. Yet, if one were to ask her: Do you think that Martin is a wonderful athlete? she would understand the question, but she would not assent. Although she understands the sentence Martin is a wonderful athlete, she has no basis for accepting it, since she doesn’t realize that Martin is the name of the quarterback who impressed her so much. So, although she believes that Martin is a wonderful athlete, and although she understands the sentence Martin is a wonderful athlete, she does not accept it or believe it to be true. adequate for the distinctions I wished to make.” Three points: (i) although the distinction between different types of necessity (and possibility) is relevant for explaining many instances of the necessary aposteriori, it won’t help here if knowing that Hesperus is Hesperus turns out to be the same as knowing that Hesperus is Phosphorus; (ii) in Kripke’s argument at the end of lecture 2, he does not invoke the distinction between different kinds of necessity (or possibility), and (iii) although he may have been generally aware of the difficulties posed by Pierre-type examples for strong disquotational principles at the time of Naming and Necessity, it seems evident that he did not focus on their implications for his argument at the end of lecture 2, probably because at that time he did not take seriously the idea that Hesperus is Hesperus and Hesperus is Phosphorus might mean the same thing.

THE NECESSARY APOSTERIORI

393

This scenario casts doubt on (i*). But if (i*) is in doubt, then there may be reason to doubt (i) and (ii) as well. We need not here try to confirm or resolve those doubts, or to decide precisely what to say about potentially problematic cases like the one just sketched. It is enough to have shown that the premises needed, and tacitly used, in Kripke’s argument are insecure, and cannot, without further investigation, be taken for granted in establishing his conclusion that it is not knowable apriori that Hesperus is Phosphorus.15 As we have seen, this is true whether the premises are formally stated using the theoretical apparatus of propositions as semantic contents of sentences and objects of propositional attitudes, or whether they are more informally and less abstractly formulated as (i) and (ii). Hence, we have little choice but to conclude that, despite the fact that there are other quite legitimate examples of the necessary aposteriori given in Naming and Necessity, Kripke’s discussion at the end of lecture 2 does not establish that the statements made by identity sentences involving coreferential names are among them.16 Final Assessment: A Dilemma

Despite this negative result, I suspect that many may be prepared to take it simply to be a datum that one can know that Cicero is Cicero, or that Hesperus is Hesperus, without knowing that Cicero is Tully, or that Hesperus is Phosphorus, and that this shows that sentences which differ only in the substitution of coreferential proper names may mean

My own view about this issue is that the sentence Is it knowable apriori that Hesperus is Phosphorus? may be used to ask different questions. If one uses it to ask whether the proposition semantically expressed by Hesperus is Phosphorus is knowable apriori, then the answer is yes. (For more, see below.) However, if one uses it to ask a question that could more explicitly be paraphrased as Is it knowable apriori that the planet, Hesperus, seen in the evening, is the planet, Phosphorus, seen in the morning? then the answer is: that proposition is not knowable apriori, but it is also not necessary. In other work, I develop the idea that sentences are often used to assert more than their semantic contents, and sometimes these contents are not asserted at all. See Beyond Rigidity, and also “Naming and Asserting,” in A. Szabo, ed., Semantics and Pragmatics (Oxford University Press, forthcoming). Following in the footsteps of my Ph.D. student Michael McGlone, I have come to believe that if this general idea is correct, then much of what is confusing about Kripke’s puzzle about belief is due to the difficulty of keeping track of which propositions are being questioned and asserted in which contexts. For a recent note on this, see my “Saul Kripke, the Necessary Aposteriori, and the Two-Dimensionalist Heresy.” 16 We will return in chapter 17 to Kripke’s thoughts about the example presented at the end of lecture 2 when we take up his problematic response to a general objection to the necessary aposteriori that he discusses in lecture 3. 15

394

KRIPKE

different things and semantically express different propositions. And if the sentences a ⫽ a and a ⫽ b semantically express different propositions, then at least the observation that the proposition expressed by the former is knowable apriori won’t force the conclusion that the proposition expressed by the latter is too. So, even if we don’t have an argument for the truth of The proposition that a ⫽ b is knowable only aposteriori, accepting the alleged datum would allow one to block the most obvious line of argument for the falsity of that claim. Fair enough. However, there is still a difficulty to be faced. We need some positive theory of the contributions made by proper names to the propositions semantically expressed by sentences containing them. If the alleged datum is to be accepted, then this account must make clear precisely in what respect the propositions expressed by sentences containing different but coreferential proper names can, and in certain cases do, differ. What makes this task daunting is that the old solution to this problem—namely, the view that names have descriptive semantic contents—seems to have been thoroughly discredited by Kripke’s arguments. If this is right, if the idea that names have descriptive semantic contents really has been discredited, then, given the alleged datum, one cannot identify the semantic contents of names either with their referents, or with descriptive semantic information that may vary from one coreferential name to another. In what other way do coreferential names differ? Often, they have different spellings, pronunciations, or syntactic structures, and in theory one could appeal to these differences to distinguish the different propositions semantically expressed by sentences containing different names. Surely, however, we don’t want to say that whenever speakers use words with different spellings, pronunciations, or syntactic structures, they must, thereby, differ in the propositions they assert or believe. Consequently, one who accepts Kripke’s anti-descriptivist arguments is left with a dilemma. There are two main options. According to the first option, one may accept the alleged datum that, typically, when sentences differ only in the substitution of one coreferential proper name for another, it is possible to assert, believe, or know the proposition semantically expressed by one of the sentences without asserting, believing, or knowing the proposition semantically expressed by the other. If one does take this option, then one must give some positive account of propositions and propositional attitudes that explains how this is possible. That doesn’t seem easy. According to the second option, one may reject the alleged datum

THE NECESSARY APOSTERIORI

395

and identify the semantic contents of names with their referents. If one does this, then one will be led to maintain that sentences which differ only in the substitution of coreferential names semantically express the same propositions, and, given this, one will find it very natural to conclude that attitude ascriptions involving such sentences are semantically equivalent. On this view, the proposition expressed by Hesperus is Phosphorus is identical with the proposition expressed by Hesperus is Hesperus. But then if (as many assume) the latter is knowable apriori, the former is too—in which case the claim that (5) is an example of the necessary aposteriori is false. This is the view defended by Nathan Salmon in Frege’s Puzzle.17 It is also the alternative I favor. However, it too faces a serious difficulty. The difficulty is to explain how, if the view is correct, speakers succeed in using sentences P and Q that differ only in the substitution of one coreferential proper name for another to assert and convey different information, and to express different beliefs—which they clearly do. In addition, it must be explained why speakers often do not regard an attitude ascription n asserts/believes/ knows that P to be truth-conditionally equivalent to the corresponding attitude ascription n asserts/believes/knows that Q—which they often do not. So the dilemma is either (i) to take it for granted that the sentences Hesperus is Hesperus and Hesperus is Phosphorus semantically express different propositions, and try to explain what this difference consists in, or (ii) to treat the two sentences as semantically expressing the same proposition, and try to explain how speakers use them to assert and convey different information, and to express and report different beliefs. The resolution of this dilemma is one of the most important unfinished legacies left to us by Naming and Necessity.18

A Further Note on Inquiry At the beginning of the chapter, we considered a conception of inquiry according to which the point of rational belief formation is to locate the actual state of the world by eliminating merely possible

Salmon, Frege’s Puzzle. In chapters 3 and 8 of Beyond Rigidity, I opt for (ii) and try to provide the basis for the needed explanations; these ideas are revised and extended in “Naming and Asserting.” See

17 18

396

KRIPKE

world-states—i.e., maximally complete properties that the universe could have instantiated—incompatible with propositions one has come to know or believe on the basis of evidence. A revision of this conception was suggested in order to accommodate necessary aposteriori propositions expressed by sentences like (1–4). According to the revised conception, conceivable world-states—maximally complete properties that one can conceive the universe as instantiating (whether or not they could in fact be instantiated)—replace genuinely possible world-states in the model of inquiry. On this conception, the point of rational belief formation is to locate the actual state of the world by eliminating merely conceivable world-states incompatible with propositions one has come to believe on the basis of evidence. It is important to notice that our discussion in the previous section calls even this revised conception of inquiry into question. Suppose that it is possible to rationally believe a proposition p—like the proposition that London is pretty—by virtue of understanding and accepting a sentence P that expresses it, while also rationally believing the negation of p, by virtue of understanding and accepting the negation, ~Q , of a different sentence—without being in a position, or having the evidence necessary, to recognize the inconsistency. In this sort of case, it would seem that the fact that the agent believes p does not eliminate all conceivable world-states with respect to which the negation of p holds. On the contrary, when the agent considers such states, as they are presented by the sentence ~Q , they remain for him genuine epistemic possibilities. But if believing p doesn’t eliminate the epistemic possibility that the negation of p may be true, then the notion of eliminating conceivable states incompatible with what one believes becomes problematic, and the model of inquiry based upon it breaks down. Thus, if inconsistent beliefs of the sort discussed here are genuinely possible, then a conception of inquiry substantially different from those we have considered may be needed.19 Salmon, Frege’s Puzzle, for a different defense of (ii). For defenses of (i), see Mark Richard, Propositional Attitudes (Cambridge: Cambridge University Press, 1990), and Larson and Ludlow, “Interpreted Logical Forms.” 19 This point is related to arguments designed to show that propositions cannot be identified with sets of circumstances (metaphysically possible world-states, epistemically conceivable world-states, logically possible world-states, situations, etc.) with respect to which they are true. For such arguments, see my “Direct Reference, Propositional Attitudes, and Semantic Content.”

CHAPTER 16 THE CONTINGENT APRIORI chapter outline 1. Names Introduced by Reference-Fixing Descriptions The way Kripke’s examples of the contingent apriori depend on names the referents of which are semantically fixed by descriptions; the argument that sentences containing such names express singular propositions about their referents 2. Prerequisites of Reference-Fixing and the Contingent Apriori Scenario 1: Restrictions on Reference Fixing. Problematic consequences for the contingent apriori of requiring that speakers know of a certain object that it satisfies a description prior to using the description to fix reference Scenario 2: Unrestricted Reference Fixing. Consequences for the contingent apriori of allowing unrestricted use of descriptions to introduce and semantically fix the referents of names Problems with scenario 2. Why putative examples of the contingent apriori based on the unrestricted use of descriptions to fix the referents of names are not compelling Weak disquotation vs. unrestricted reference-fixing. The conflict between two principles needed to make Kripke’s examples of the contingent apriori work; why reference-fixing by description must be restricted Why no Kripke-style examples can be genuine instances of the contingent apriori 3. Are There Examples of the Contingent Apriori? The importance of Kripke’s insight over and above the details of his execution; how to use the actuality operator to construct genuine instances of the contingent apriori

The Semantics of Names Introduced by Reference-Fixing Descriptions In this chapter, we turn to Kripke’s discussion of the contingent apriori. As with our discussion of the necessary aposteriori, we adopt the expository framework given at the beginning of chapter 15, according to which propositions are the things expressed by sentences, the bearers of

398

KRIPKE

(contingent or necessary) truth, as well as the objects of belief, knowledge, and assertion. In addition, we continue to assume that an attitude ascription x asserts/believes/knows that S reports a relation—assertion/ belief/knowledge—between an agent and the proposition expressed by S. Kripke’s central doctrine about the contingent apriori may then be understood to be the claim that there are a number of cases in which a single proposition is contingent (i.e., true with respect to the actual state of the world, and false with respect to some other possible world-states) while nevertheless being knowable apriori (without appeal to empirical evidence). Kripke’s examples of the contingent apriori in Naming and Necessity involve cases in which a name has its referent semantically fixed by a non-rigid description. In these cases, he contends, we introduce a name n by stipulating that it is to rigidly designate the unique thing that satisfies a certain descriptive condition D. If there is such a thing, and the stipulation is successful, then the referent of n with respect to an arbitrary world-state w will be the actual denotation of the D, even though n will typically not be synonymous with the description, and sentences containing n will typically differ in meaning from corresponding sentences containing the D. An example of this is provided by the sentences in (1) and (2). 1a. If n exists, then n is the D. b. If there is a unique thing which is D, then n is the D. 2a. If the D exists, then the D is the D. b. If there is a unique thing which is D, then the D is the D. Although the sentences in (2) are necessary truths, (1a) is contingent whenever the D is non-rigid, and (1b) is contingent whenever the way in which the D is non-rigid is by denoting, with respect to certain merely possible world-states, objects other than the entity it actually denotes. However, despite being contingent, (1a) and (1b) are taken to be apriori, on the grounds that understanding them is supposed to be sufficient for concluding that they are true. Here, it is worth noting that although Kripke often talks about sentences of this sort being both contingent and apriori, he typically feels free to express his view using the language of indirect discourse, as well. For example, he is willing to say, It is possible for one to know apriori that if there is a unique thing which is D, then n is the D, even though it is not a necessary truth that

THE CONTINGENT APRIORI

399

if there is a unique thing which is D, then n is the D, or to say, The claim that if there is a unique thing that is D, then n is the D is knowable apriori even though it is a contingent rather than a necessary truth. The most prominent example of this sort in Naming and Necessity concerns the standard meter. Here, Kripke imagines using the description the length of stick s in Paris at time t to fix the referent of the term one meter. He suggests that one who does this can know, solely on the basis of this reference-fixing stipulation, that stick s is one meter long at t (if s exists at t, and hence has a unique length at t). He takes this to be an example of the contingent apriori, since clearly, s (if it had been heated or cooled) could have been some length other than this length (imagine a visual demonstration of its actual length), which is the length that the term one meter rigidly designates. Here is what he says. What then, is the epistemological status of the statement ‘Stick S is one meter long at t 0’, for someone who has fixed the metric system by reference to stick S? It would seem that he knows it a priori. For if he used stick S to fix the reference of the term ‘one meter’, then as a result of this kind of ‘definition’ (which is not an abbreviative or synonymous definition), he knows automatically, without further investigation, that S is one meter long. On the other hand, even if S is used as the standard of a meter, the metaphysical status of ‘S is one meter long’ will be that of a contingent statement, provided that ‘one meter’ is regarded as a rigid designator: under appropriate stresses and strains, heatings or coolings, S would have had a length other than one meter even at t 0’. (Such statements as ‘Water boils at 100°C at sea level’ can have a similar status.) So in this sense, there are contingent a priori truths.1 Although there is something quite plausible about what Kripke says, there is a puzzle about this example, and all others of the same type. In each of these cases, Kripke produces an example sentence S, and goes on to argue that there is, in effect, some one claim (statement or proposition) designated by the clause that S which is both contingently true and knowable apriori. But it is unclear precisely what claim (statement or proposition) this is. For example, consider the claim that if stick s exists at t, then its length at t is one meter. We are told that this claim 1

Naming and Necessity, p. 56, my boldface emphasis.

400

KRIPKE

is both contingent and knowable apriori. Since it is contingent, we know that it cannot be the claim that if stick s exists at t, its length at t is the length of stick s at t. This is an example of an elementary point that Kripke insists on: When the reference of n is semantically fixed by a description the D, then typically n is not synonymous with the D, and sentences that differ only in the substitution of one for the other typically mean different things, and express different propositions (make different claims). So what is this claim that is supposed to be both contingent and apriori? What exactly is the proposition expressed by If stick s exists at t, then its length at t is one meter? Perhaps it is the claim that if stick s exists at t, then its length at t is this length (imagine a visual demonstration of the length that is the actual length of the stick). Perhaps the claim in question is a singular proposition that says of a certain length l—the length rigidly designated by the name one meter—that if stick s exists at t, then its length at t is l.2 The question at issue is what names the referents of which have been fixed by descriptions contribute to the propositions expressed by sentences containing them. Though Kripke doesn’t himself use this theoretical language, he has already implicitly told us that they don’t contribute the senses of their reference-fixing descriptions. So perhaps what they contribute are just the objects they rigidly designate— a certain length l, in the case of one meter. Some support for this idea can be found in the paragraph immediately following the passage just cited. In the case of names one might make this distinction too [between ‘definitions’ which fix a reference and those which give a synonym]. Suppose the reference of a name is given by a description or a cluster of descriptions. If the name means the same as that description or cluster of descriptions, it will not be a rigid designator. It will not necessarily designate the same object in all possible worlds, since other objects might have had the given properties in other possible worlds, unless (of course) we happened to use essential properties in our description. So suppose we say, ‘Aristotle is the greatest man who studied with Plato’. If we used that as a definition, the name ‘Aristotle’ is to mean ‘the greatest man who studied with Plato’. The relationship between these two conjectures is the following: the proposition expressed by If stick s exists at t, then its length at t is this length, said demonstrating the precise length l in question, either is identical with, or trivially entails, the singular proposition that says of l that that if stick s exists at t, then its length at t is l. For purposes of our discussion, it doesn’t matter whether the relationship is identity of entailment. Hence, to simplify the presentation, I will often speak as if it were identity.

2

THE CONTINGENT APRIORI

401

Then of course in some other possible world that man might not have studied with Plato and some other man would have been Aristotle. If, on the other hand, we merely use the description to fix the referent then that man will be the referent of ‘Aristotle’ in all possible worlds. The only use of the description will have been to pick out to which man we mean to refer. But then when we say counterfactually ‘suppose Aristotle had never gone into philosophy at all’, we need not mean ‘suppose a man who studied with Plato, and taught Alexander the Great, and wrote this and that, and so on, had never gone into philosophy at all’, which might seem like a contradiction. We need only mean, ‘suppose that that man had never gone into philosophy at all’.3 Note, Kripke is quite explicit that if a description the D merely fixes the referent of a name n, then uses of n is F and the D is F may mean quite different things, formulate different claims, and so, it would seem, express different propositions. Moreover, he suggests that in these cases if n refers to some man, then someone who uses n is F means and says something like that man is F, where the term that man is used demonstratively to pick out the referent of n. This is interesting. Although he is not explicit about the significance of this point, his remarks invite a pair of inferences. (i) Even when n is a name the referent of which is semantically fixed by a description, the content of one’s use of n is F is a singular proposition that says, of the referent of n, that it has a certain property (the one expressed by F). (ii) In order for one to know this proposition, and hence to satisfy the ascription x knows that n is F, one must know, of the particular object which is the referent of n, that it has a certain property (the one expressed by F).4 Ibid., p. 57, my boldface emphasis. Here, and throughout, I use the locution believe of an object that it is so and so. To say that an agent believes of o that it is F is to say that the agent satisfies the formula believes that x is F relative to an assignment of o to ‘x’. This will be the case iff the agent believes the singular proposition that predicates the property expressed by F of o. Of course, in order to believe this proposition, one must think about o in a certain way, which may involve attributing various descriptive characteristics to o. Since these are not part of the singular proposition expressed by x is F relative to an assignment of o to ‘x’, belief in that proposition is often accompanied by belief in other, descriptively enriched propositions. However, the truth of the claim The agent believes of o that it is F does not depend on which other descriptively enriched propositions the agent may also believe. For more on believing of—often called de re belief—see pp. 149–153 of my “Donnellan’s Referential/Attributive Distinction,” Philosophical Studies 73 (1994): 149–168. 3 4

402

KRIPKE

These claims—(i) and (ii)—gain further plausibility from the observation that corresponding claims also seem to hold in the case of ordinary names, the referents of which are not semantically fixed by descriptions. For example, consider the name Saul Kripke, the referent of which is not simply stipulated to be the same as that of some reference-fixing description. Corresponding to (i), we observe that someone who assertively utters the sentence Saul Kripke is F says of a certain man, Saul Kripke, that he has a certain property (the one expressed by F). Corresponding to (ii), we notice that if Mary knows that Saul Kripke is a famous philosopher who was a native of Omaha, Nebraska, then there is a particular individual, Saul Kripke, such that she knows that he is a famous philosopher who was a native of Omaha, Nebraska—Mary knows of this particular man that a certain fact about him holds. By contrast, principles corresponding to (i) and (ii) do not hold when descriptions are substituted for names. For example, suppose that Mary knows that there are spies, without knowing of any particular individual that he or she is a spy. Suppose further that she correctly believes that some one among the spies must be the shortest spy, again without having any idea who that person may be.5 Then, contrary to (i), if Mary assertively utters the sentence The shortest spy is a spy, she has not said, of the individual Boris (who, unknown to her, is the shortest spy), that he is a spy, and, contrary to (ii), although Mary knows that the shortest spy is a spy, there is no particular person p such that Mary knows that p is a spy—i.e., there is no individual such that Mary knows, of that individual, that he or she is a spy. Thus, whereas (i) and (ii) hold for ordinary names, corresponding principles do not hold for many descriptions. In light of this, Kripke’s implicit claim that (i) and (ii) also hold for special names the referents of which are semantically fixed by descriptions is a plausible and useful reminder that, for him, the propositions semantically expressed (and asserted) by (utterances of) sentences containing these special names are to be understood to be on a par with the singular propositions semantically expressed (and asserted) by (utterances of ) sentences containing ordinary names.6 5 This example is based on a distinction originally discussed by Quine in “Quantifiers and Propositional Attitudes,” Journal of Philosophy 53 (1956), reprinted in Quine’s The Ways of Paradox. 6 On p. 41 of Naming and Necessity Kripke gives the same account of what is meant by an utterance of a sentence containing an ordinary name—Nixon—as he gives of what is meant by an utterance of a sentence containing a name the reference of which is fixed by a description. He says, “When you ask whether it is necessary or contingent that Nixon won the election, you are asking the intuitive question whether in some counterfactual situation, this man would in fact have lost the election.” Note the demonstrative, and compare with the passage from p. 57 already cited.

THE CONTINGENT APRIORI

403

In light of this, it is reasonable to think that, for Kripke, the proposition that if stick s exists at t, then its length at t is one meter is the very same proposition as the proposition that if stick x exists at t, then its length at t is this length (imagine a visual demonstration of the length which one meter rigidly designates). This proposition says of a certain length l that l is the length of stick s at t, if s exists at t. This, presumably, is the proposition that is supposed to be contingent, yet knowable apriori. More generally, Kripke’s hypothesis may be taken to be the claim that all examples of the contingent apriori involving names introduced by reference-fixing descriptions are cases in which a single proposition is both contingent yet knowable apriori—where this proposition is a singular proposition, which says something of the particular individual that is the referent of the name.

Prerequisites of Reference-Fixing and Their Relevance for the Contingent Apriori Although I believe this to be the correct reading of Kripke’s text, it raises difficult interpretive and philosophical problems. In his discussion of the meter-stick example, which he takes to provide a genuine instance of the contingent apriori, he nowhere indicates whether, in fixing the referent of the term one meter, we have had independent acquaintance with stick s. This may encourage the reader to think that it doesn’t matter for the philosophical points he is making whether or not we have had such contact. However, this is not so. In fact, a resolution of this issue is necessary, if we are to understand and evaluate the central points in Kripke’s discussion of the contingent apriori.7 Scenario 1: Restrictions on Reference-Fixing

First, consider the scenario in which we have seen stick s at t, and formed an idea of its length, prior to introducing the term one meter with the description the length of stick s at t. In this scenario, we may be taken to have a perceptually justified true belief of a certain length, that l is the length of stick s at t. Consequently, when we introduce the term one meter as a rigid designator of that length, our understanding This issue is raised, and its significance insightfully discussed, by Nathan Salmon in “How to Measure the Standard Metre,” Proceedings of the Aristotelian Society 88 (1987/88): 193–217.

7

404

KRIPKE

of the sentence the length of stick s at t is one meter (if s exists at t) guarantees that we know both that it expresses a truth, and precisely which truth it expresses—we know that it expresses the true proposition that l is the length of stick s at t (if s exists at t). The only remaining question about this scenario is whether this result is sufficient for our knowledge of this proposition to qualify as apriori. It is not clear that it is. After all, in this scenario, we have a perceptually justified belief in this proposition even prior to the introduction of one meter. That belief is certainly not apriori; and it is hard to see why merely performing a linguistic ceremony should transform it into one. Hence it is hard to see how, in this scenario, we have a genuine example of the contingent apriori. We will return to this point later. Before we do, however, we need to examine a different way of understanding Kripke’s example. Scenario 2: Unrestricted Reference-Fixing

In this scenario, we are portrayed as attempting to introduce the term one meter as a rigid designator of whatever satisfies the description the length of stick s at t in a situation in which we have never seen stick s, and have no idea how long it is. Suppose that it is possible to successfully introduce the term in this sort of case. Then the situation is one in which the following facts obtain. F1. There is a proposition p which truly says of a certain length— namely, this length (imagine a visual demonstration of the length l which is the actual length of the stick at t)—that it is the length of stick s at t, if stick s exists at t. F2. Prior to introducing the term one meter, we do not believe p, since we have no idea how long s is at t. F3. We introduce the term one meter by stipulating that it is to rigidly designate whatever length is the actual length of s at t. F4. As a result of this stipulation, the sentence The length of stick s at t is one meter, if s exists at t comes to express p. F5. We understand the sentence in the following sense: we understand the process of introducing a name by a description; we understand how the referent of the name is thereby determined; and we understand how the proposition expressed by the sentence is determined, given the referent of the name.

THE CONTINGENT APRIORI

405

Understanding all this is sufficient for us to know that the sentence expresses a truth—i.e., that its meaning, which may be taken to be a function from contexts of use of the sentence to propositions expressed, assigns a true proposition as value to our context as argument. At this point, two questions arise. Q1. How, if at all, does the knowledge mentioned in F5 result in our knowing p, and hence in our knowing that the length of stick s at t is one meter, if s exists at t? Q2. If the knowledge mentioned in F5 does result in our knowing p, is this knowledge properly characterized as apriori knowledge? Since Naming and Necessity is not explicitly committed to F1–F5, it does not provide explicit answers to Q1 and Q2. There is simply nothing in Kripke’s discussion of the meter-stick example that definitively settles the matter of whether one can introduce a name by a referencefixing description, without acquaintance with its denotation and independent evidence that the description applies to it. However, other parts of his discussion have encouraged some readers to think that one can use a description to introduce and semantically fix the referent of a name, even when this condition is not satisfied. One of these passages is the following, in which Kripke gives the example of the “name” Jack the Ripper as an instance of a real-life case in which the referent of a name may be semantically fixed by a description. What picture of meaning do these Theses ((1)–(5)) give you? [The theses are meant to characterize cases in which the referents of names are semantically fixed by descriptions.] The picture is this. I want to name an object. I think of some way of describing it uniquely and then I go through, so to speak, a sort of mental ceremony: By ‘Cicero’ I shall mean the man who denounced Catiline; and that’s what the reference of ‘Cicero’ will be. I will use ‘Cicero’ to designate rigidly the man who (in fact) denounced Catiline, so I can speak of possible worlds in which he did not. But still my intentions are given by first, giving some condition which uniquely determines an object, then using a certain word as a name for the object determined by this condition. Now there may be some cases in which we actually do this. Maybe, if you want to stretch

406

KRIPKE

and call it description, when you say: I shall call that heavenly body over there ‘Hesperus’. That is really a case where the theses not only are true but really even give a correct picture of how reference is determined. Another case, if you want to call this a name, might be when the police in London use the name ‘Jack’ or ‘Jack the Ripper’ to refer to the man, whoever he is, who committed all these murders, or most of them. Then they are giving the reference of the name by a description.8 Here Kripke explains what it is to use a description to fix the referent of a name, and indicates that (although most names, like ‘Cicero’, do not have their referents fixed in this way), ‘Hesperus’ and “Jack the Ripper’ may really fit the picture. Since the situation he imagines in the case of ‘Jack the Ripper’ is one in which the police had little or no direct contact with the man denoted by the description allegedly used to fix reference, and since they seem not to have had independent evidence to believe of any specific man that the description applied to him, this passage might be read as suggesting that, according to Kripke, one can use a description to semantically fix the referent of a name, even in cases in which one is not acquainted with the denotation of the description, and has no independent evidence that the description applies to that individual. Next, we apply this idea to the scenario in which we use the length of stick s at t to fix the referent of one meter, even though we have never seen s and have no idea how long s is. This brings us back to F1–F5, and the two questions raised about those alleged facts. F5 tells us that we understand the sentence The length of stick s at t is one meter, if s exists at t, and that this understanding is sufficient to conclude that it expresses a truth (in our language). The next question is, how, if at all, this ensures that we know the truth it expresses. Here I hypothesize implicit reliance (by some readers and perhaps even Kripke himself ) on some version of the principle of weak disquotation cited in the discussion in chapter 15 of the necessary aposteriori, and repeated here in abbreviated form.9 weak disquotation If a sincere, reflective, rational individual i who understands a sentence S is disposed to accept S, and believe S to be true, Naming and Necessity, pp. 79–80, my emphasis and additions in brackets. A different version of the principle might speak of accepting the meaning of S and believing that meaning to assign a true proposition to one’s present context. Our discussion is intended to be neutral as to variations such as this. 8 9

THE CONTINGENT APRIORI

407

then i believes the proposition semantically expressed by S, and thereby satisfies the formula x believes that S. The idea is that the knowledge mentioned in F5 leads us both to understand and to accept the sentence The length of stick s at t is one meter, if s exists at t, and believe it to be true; weak disquotation then ensures that we believe the contingent proposition p that it expresses. How, then, is it determined that this knowledge qualifies as apriori? Here we appeal to the principle of weak disquotation and justification cited in chapter 15, and repeated here in abbreviated form. weak disquotation and justification If a sincere, reflective, rational individual i who understands S and is in possession of evidence e would be justified in accepting S, and believing it to be true, on the basis of e, then i’s possession of e is sufficient to ensure that i would be justified in believing the proposition semantically expressed by S, and hence it is sufficient to ensure that i satisfies the formula x would be justified in believing that S. The knowledge mentioned in F5 not only leads us to accept the sentence The length of stick s at t is one meter, if s exists at t, and believe it to be true, it also justifies this acceptance and belief. Since this knowledge is simply knowledge of meaning, the principal of weak disquotation and justification guarantees that our knowledge of semantic facts about the sentence justifies us in believing the proposition p that it expresses, and hence in believing that the length of stick s at t is one meter, if s exists at t. Given that p is true and that we are justified in believing it, one may conclude that we know p—we know that the length of stick s at t is one meter, if s exists at t—solely by virtue of our understanding the sentence that expresses this knowledge, and knowing semantic facts about it.10 What does this have to do with the apriori? The idea, I suppose, is something like the following: weak linguisticism about the apriori If one knows a proposition p solely by virtue of understanding a sentence that expresses p, and knowing semantic facts about it, then one knows p apriori. For our purposes, I here assimilate knowledge to justified true belief. Although such an analysis fails in certain special cases, none of these are in the offing here, and we can ignore unnecessary complications. 10

408

KRIPKE

Given all this, one can characterize the proposition that the length of stick s at t is one meter, if stick s exists at t, as a legitimate example of the contingent apriori—where one’s apriori knowledge of this proposition is seen as arising from the use of the description the length of stick s at t to fix the referent of one meter, in a case in which one has never seen s, and one has no idea, prior to introducing the term, how long s is. Problems with Scenario 2 Problem 1

Although the story we have told yields a putative example of the contingent apriori that fits much of what Kripke says in Naming and Necessity, in my opinion, the story is not believable, and in the end, I will argue, even Kripke doesn’t believe it. The first problem is that the reasoning employed in scenario 2 mischaracterizes the apriori, and eviscerates the distinction between propositions knowable apriori and propositions knowable only aposteriori. Part of the problem lies in the principle of weak linguisticism about the apriori. If this principle is not taken to be part of a new stipulative definition of what we are to mean by apriori, but rather is understood as a hypothesis about the traditional notion of knowledge the justification of which does not rest on empirical evidence, then it is extremely doubtful that it is true. This point is illustrated by the examples in (3) and (4). 3a. ‘Rabbits are animals’ is a true sentence of English iff rabbits are animals. b. The word ‘rabbit’ refers in English to an object iff that object is a rabbit. 4a. ‘Rabbits are animals’ is a true sentence of my language iff rabbits are animals. b. The word ‘rabbit’ refers in my language to an object iff that object is a rabbit. The propositions expressed by these sentences are standardly not regarded as capable of being known apriori. First consider those in (3). Anyone who understands their meanings, and knows that they are sentences of English, has all the information needed to determine their truth. However, the propositions they express cannot be known independent of empirical evidence. If they could be so known, then

THE CONTINGENT APRIORI

409

non-English speakers should be able to know them simply by reflecting on and reasoning about them—which of course they cannot. What separates English from non-English speakers is that the former are in possession of empirical evidence about what English words and sentences mean. This evidence plays two roles in examples like (3a,b)—(i) it allows English speakers to understand the sentences, and (ii) it justifies their belief in the truth of the sentences, and the propositions they express. It is because of this second role played by empirical evidence that these propositions are not knowable apriori. The same holds for the propositions expressed by (4a,b). The only difference between this case and the one involving (3) is that whereas an English speaker might doubt whether the language that he speaks is “English” (and so be in doubt about (3a) and (3b)), I can hardly doubt that the language I speak is my language. Thus, in the case of (4a,b), all that is required in order for me to know that these sentences (and the propositions they express) are true is that I understand them. However, since the evidence that goes into this understanding also justifies my belief in the propositions they express, my knowledge of these propositions is not apriori in the traditional sense. What has been said about (3) and (4) carries over to the account of the sentence The length of stick s at t is one meter, if s exists at t given in scenario 2. The reference-fixer’s knowledge that the sentence expresses a truth in his language is based on his understanding of it, which in turn is rooted in his knowledge of certain empirical facts, including facts about the use of the description the length of stick s at t to introduce and semantically fix the referent of the term one meter. Since his knowledge of the truth of this sentence is not apriori, and since, according to the weak disquotation principles, his justification for believing the proposition p that the sentence expresses is inherited from his justification for believing the sentence to be true, his knowledge of p—i.e., his knowledge that the length of stick s at t is one meter, if s exists at t—is not apriori either. In all these cases—(3), (4), and the scenario 2 treatment of the meter-stick example—the contingent propositions known to be true are (or are supposed to be) knowable solely on the basis of understanding sentences that express them. If they are knowable in this way, then they may properly be regarded as trivial in an interesting way. However, it is doubtful that they should be counted as apriori in the traditional sense. Thus, it is doubtful that weak linguisticism about the apriori is true. The problem is compounded when this principle is combined, as it is in scenario 2, with weak disquotation, weak disquotation and justification, and the unrestricted ability to use descriptions to introduce and

410

KRIPKE

semantically fix the referents of names. When this is done, the class of propositions characterized as knowable apriori becomes far too expansive to be theoretically interesting. Let o be any object whatsoever, and let {P1, . . . Pn} be any set of properties the conjunction of which uniquely applies to o. Given the unrestricted ability to use descriptions to introduce and semantically fix the referents of names, one could always form a description the D denoting the object uniquely designated by the conjunction of the Pi’s, and use it to rigidly fix the reference of a name n—which would then refer to o. Finally, let p be any proposition that says of o that, if it exists, then it has one or more of the Pi’s. Using n, we could formulate a sentence expressing p which would be guaranteed to express a truth, and would be known by linguistically competent speakers to do so. Under the assumptions of scenario 2, this is sufficient for p to be characterized as capable of being known apriori. In other words, if the assumptions of scenario 2 were correct, then virtually every true proposition that predicates one or more properties of any object would qualify as capable of being known apriori. Surely, this cannot be. Problem 2

I have already expressed doubt about the principle of weak linguisticism about the apriori. In addition, the difficulty just noted points to a further problem with the combination of weak disquotation and the unrestricted ability to use descriptions to introduce and semantically fix the referents of names. Thus, there is reason to think that one or the other of these two doctrines must be false. Recall the five alleged facts which, in scenario 2, arise from the unrestricted ability to use descriptions to semantically fix reference. F1. There is a proposition p which truly says of a certain length— namely, this length (imagine a visual demonstration of the length l which is the actual length of the stick at t)—that it is the length of stick s at t, if stick s exists at t. F2. Prior to introducing the term one meter, we do not believe p, since we have no idea how long s is at t. F3. We introduce the term one meter by stipulating that it is to rigidly designate whatever length is the actual length of s at t. F4. As a result of this stipulation, the sentence The length of stick s at t is one meter, if s exists at t comes to express p.

THE CONTINGENT APRIORI

411

F5. We understand the sentence in the following sense: we understand the process of introducing a name by a description; we understand how the referent of the name is thereby determined; and we understand how the proposition expressed by the sentence is determined, given the referent of the name. Understanding all this is sufficient for us to know (i) that the sentence expresses a truth, and (ii) that its meaning, which may be taken to be a function from contexts of use of the sentence to propositions expressed, assigns a true proposition as value to our context as argument. Adding weak disquotation to this collection gives the result that we believe p, and hence believe that the length of stick s at t is one meter, if s exists at t. But notice how mysterious this is. Priori to going through a little verbal ceremony, we are ignorant of a certain entirely nonlinguistic, nontrivial, empirical fact—namely that stick s, if it exists at t, is this long (I visually demonstrate the length l, which is its actual length at t). We say a few formulaic words—let ‘one meter’ rigidly designate the length of stick s at t, whatever it might be—and, presto, we know the fact that we were previously ignorant of. Wait until students taking exams hear about this. When asked Who did such and such?, they may answer N did such and such, where ‘N’ is a name I hereby stipulate has its reference rigidly fixed by the description, ‘the individual who did such and such’. Such a performance would be absurd. True belief is not that easy to come by, which suggests either (i) that there are substantial restrictions on when descriptions can be used to introduce and semantically fix the referents of names, or (ii) that there are no such restrictions, but weak disquotation fails when applied to sentences containing such names. Another indication that this is the case is provided by a persuasive argument given by Kripke himself, but, as far as I know, one that he never published.11 Let A be any agent who has at least one false belief, which is expressed by the sentence S. Next, consider the description (5).12 5. The x: (if S, then x ⫽ Princeton University) & (if ~S, then x ⫽ Saul Kripke’s left thumbnail). 11 I am familiar with the argument from his seminars at Princeton University in the ’80s and ’90s, as well as from some of his public lectures elsewhere. I do not know whether he was aware of the argument at the time he gave the lectures in Naming and Necessity. 12 The x: . . . x . . . is a singular definite description formed by attaching the description operator the x to a formula containing at least one free occurrence of ‘x’. It is paraphrased the unique individual x which is such that. . . .

412

KRIPKE

Since S is false, this description refers to Kripke’s left thumbnail. However, since A believes that S is true, he will take it to refer to Princeton University, and hence will accept sentence (6a), and wrongly believe it to express a truth. 6a. The x: (if S, then x ⫽ Princeton University) & (if ~S, then x ⫽ Saul Kripke’s left thumbnail) is an institution of higher learning. So far, there is nothing paradoxical about this. Having one false belief naturally leads to having other closely related false beliefs, and this is just an instance of that. A problem arises, however, if it is possible for A to use (5) to semantically fix the referent of a name ‘PU’, even though A has no independent evidence that its actual denotation, Saul Kripke’s left thumbnail, satisfies the description. Given this latitude, which is assumed in scenario 2, we get the result that A understands, accepts, and believes true the sentence (6b), which expresses a singular proposition p which says of Kripke’s thumbnail that it is an institution of higher learning. 6b. PU is an institution of higher learning. Combining this result with weak disquotation gives the result that A believes p, and hence that (7a) is true. 7a. A believes that PU is an institution of higher learning. Since PU is Saul Kripke’s left thumbnail, and since (7a) tells us that A believes of it that it is an institution of higher learning, we get (7b). 7b. There is a certain object o such that o is Saul Kripke’s left thumbnail, and A believes that o is an institution of higher learning. Kripke rightly regards this result—which could be repeated for virtually any agent, object, and property—to be a reductio ad absurdum of the conjunction of weak disquotation with the unrestricted ability to use descriptions to introduce and semantically fix the referents’ names. Thus, it seems, at least one of these principles must go. This is a result that ought to have been expected all along. The principle of weak disquotation is plausible, and makes good sense, so long as there is a close relationship between understanding a sentence and knowing what proposition it expresses. When there is such a relationship, someone who understands the sentence may be thought of as entertaining the proposition it expresses, and someone accepting the

THE CONTINGENT APRIORI

413

sentence is plausibly taken as believing the proposition. If, however, one opens a substantial gap between understanding a sentence s and knowing of a certain proposition p, that s expresses p, then the original rationale for weak disquotation is undermined. This is just what happens when one assumes both that sentences containing names express singular propositions that predicate various things of their referents, and that there are essentially no restrictions on the use of descriptions to semantically fix the referents of names. Under these assumptions, such names constitute a kind of indexical, with their referents at a given context of utterance being determined to be whatever satisfies their reference-fixing descriptions with respect to the possible world-state of the context of utterance. On this picture, knowing the referent of such a name n at a context, and knowing of a proposition p that it is expressed by the sentence containing n, requires knowing which object satisfies the description associated with n at the possible world-state of the context. As a result, an agent who lacks this knowledge will no more know which proposition is expressed by his own words than would an agent at Yankee Stadium who assertively uttered He is a Yankees fan, with his eyes closed while pointing blindly at me in a crowd, not having any idea at whom he was pointing. Note, the agent with his eyes closed understands the sentence he utters in the sense that he knows what it means. He may also believe that his use of it expresses a truth, since he may regard it as very likely that the person he is pointing at is a Yankee’s fan. However, it would be wrong to suppose that he thereby acquires a (false) belief about me. Since he has no idea at whom he is pointing, he does not believe that I am a Yankees fan. By the same token, he does not believe apriori the contingent truth that if I exist, then I am the person he is pointing at—which he would express by uttering If he exists, then he is the person I am pointing at while pointing blindly at me. Thus, it would be a mistake to suppose that weak disquotation correctly applied to such a case. By the same token, if one thought that there were essentially no restrictions on the use of descriptions to semantically fix the referents of names, then it would be a mistake to suppose that weak disquotation applied to sentences containing such names.13 13 Though the case seems clear to me, not everyone agrees. For example, Gil Harman supports what is essentially the opposite point of view in his “How to Use Propositions,” American Philosophical Quarterly 14 (1977): 173–76. See also section XVII of David Kaplan, “Demonstratives,” in Almog, Perry, and Wettstein, Themes From Kaplan, and Robin Jeshion, “Acquaintanceless De Re Belief,” in M. O’Rourke et al. eds., Truth and Meaning: Investigations in Philosophical Semantics (New York: Seven Bridges Press 2001).

414

KRIPKE

Weak Disquotation vs. Unrestricted Reference-Fixing

At this point, we are left with two possible alternatives. According to view 1, descriptions may freely be used to semantically fix the referents of names, whether or not one has independent knowledge of which objects satisfy the descriptions, or which propositions are expressed by sentences containing such names. However, if one lacks such knowledge, then accepting these sentences, and believing them to be true, will not be sufficient for one to believe the propositions they express. On this view, names the referents of which are fixed by descriptions are easy to come by, but sentences containing them are not of much use for expressing beliefs, because weak disquotation fails for such sentences. Thus, when the D semantically fixes the referent of n, one may know, solely on the basis of one’s understanding of the language, that the sentences If there is a unique D, then n is the D and If n exists, then n is the D express truths; however, one cannot know the propositions expressed by these sentences on this basis, and in many cases one may fail to know, or believe, them at all. According to view 2, a description cannot semantically fix the referent of a name for a speaker unless the speaker independently believes of the object which is denoted by the description that the description applies to it. However, if this condition is satisfied, then a speaker who understands, accepts, and believes the sentence to be true will believe the proposition the sentence expresses. Thus, when the D is successfully used to fix the referent of n, a speaker may know, solely on the basis of his understanding the language, the propositions expressed by the sentences If there is a unique D, then n is the D and If n exists, then n is the D, in addition to knowing that these sentences express truths. However, in this case, the knowledge required to understand the language will include the independent, non-linguistic knowledge reported by x knows of y that it is D relative to an assignment of the speaker to ‘x’, and the object denoted by the D to ‘y’. Faced with this choice, we have reason to prefer view 2. As I see it, view 1 simply creates too large a gap between the proposition expressed by a sentence in a context, and the information with which competent speakers are presented when they understand the sentence, as used in the context. According to view 1, speakers can routinely understand a sentence S that contains a name the reference of which is fixed by a description, and they can know that S expresses a truth, and be prepared to accept and sincerely assent to S, but not be in any position to believe, or

THE CONTINGENT APRIORI

415

even to entertain, the proposition that S semantically expresses. This is puzzling: if one understands S, and S expresses p, then, unless there is something quite special and out of the ordinary about the context, entertaining S should count as entertaining p; and if one comes to accept S, or judge it to be true, then, typically, this should count as believing p, or taking it to be true. If this doesn’t, standardly, count as believing the proposition expressed by S, then S will not be of much use in expressing one’s beliefs. But if that is so, then it is hard to avoid concluding that S is seriously defective, and perhaps not fully meaningful after all, since it can’t be used to perform one of the primary functions of language. This argument can be strengthened by extending it to the notion of assertion. Someone who believes that the shortest spy is a spy may not believe of anyone that he (or she) is a spy. Similarly, someone who asserts that the shortest spy is a spy may not have said of anyone that he (or she) is a spy. Suppose now that such a speaker goes through a little linguistic ceremony; he uses the description the shortest spy to fix the referent of the name Shifty—without having any idea who the shortest spy is. If the speaker is successful in introducing the name in this way, then, the sentence Shifty is a spy semantically expresses a proposition that can be asserted only by one who says or asserts, of the person who is the shortest spy, that he or she is a spy. But it hardly seems that our speaker would be correctly described as doing this, if he were to assertively utter Shifty is a spy. Surely, the speaker wouldn’t thereby assert that he or she (demonstrating the individual who in reality is the shortest spy) is a spy. We may suppose that prior to introducing the name, the speaker was not in a position to say anything of this person. But then, going though a formal linguistic ceremony is not going to help. Thus, I believe we must conclude that in assertively uttering Shifty is a spy, the speaker does not assert the proposition that, according to view 1, it semantically expresses. The same point could be made about the speaker in Kripke’s example who tries to use the description (5) to introduce and semantically fix the referent of PU, which he wrongly takes to designate Princeton University, rather than Kripke’s left thumbnail. Just as his utterances of (6b) don’t count as expressions of a bizarre belief that the thumbnail is an institution of higher learning, so they don’t count as assertions that it is, either. In my opinion, these results constitute a strong reason for rejecting view 1.14 If it were correct, then there would be many cases in which 14 I believe that Kripke now takes the same position, though it is not expressed in Naming and Necessity, or in any other of his published writings that I know of.

416

KRIPKE

the referent of a name was semantically fixed by a description for speakers and hearers, even though they did not believe, of the object denoted by the description, that the description applied to it. Where S is a sentence containing such a name, we would then have a situation in which S means (expresses the proposition) p, speakers and hearers understand S, but they cannot use S to assert p, entertain p, or express a belief in p. That is hard to swallow. If speakers can’t use the sentence to assert, entertain, or express a belief in p, then it is highly doubtful that the sentence really does mean (semantically express) p. All of this argues against the view that we may succeed in introducing a name with a description that semantically fixes its referent, but does not give its content, using any description we like—even one which denotes an object of which we have no independent belief that it satisfies the description. As I see it, this extremely relaxed view about what is required in order to introduce a name by a reference-fixing description should be given up. In its place, we should adopt a view that insists that the introduction of a genuine name by a reference-fixing description (as opposed to the stipulation that the “name” is to be short for the description, or some variant of it) requires us to be in sufficient contact with the object denoted by the description to have independent beliefs about it—including the belief that the description applies to it. With this view in place, we can no longer regard scenario 2 as providing an example of the contingent apriori. Consequences for Kripke’s Examples of the Apriori

Let us return, then, to scenario 1. In this scenario, we have seen stick s at t, and formed an idea of its length, prior to introducing the term one meter with the reference-fixing description the length of stick s at t. As a result, we have a perceptually justified true belief, of a certain length l, that it is the length of stick s at t. Call this proposition about l that we believe on the basis of perceptual evidence p. When we introduce the term one meter with the reference-fixing description, our knowledge of p is part of the knowledge required to understand the sentence The length of stick s at t is one meter (if s exists at t). Since we have this knowledge, we understand the sentence. Moreover, our knowledge of the sentence guarantees not only that we know that it expresses a truth, but also that we grasp the proposition it expresses, and know it to be true. We know, on the basis of the knowledge required to understand the sentence The length of stick s at t is one meter,

THE CONTINGENT APRIORI

417

if s exists at t, that the length of stick s at t is one meter, if s exists at t. Thus, nothing more than understanding the sentence is required in order for us to know the proposition it expresses. Nevertheless, this knowledge is not apriori knowledge, since it rests on our knowledge of p, which in turn is justified perceptually. Note, it is not just that we happened to acquire our belief in p as a result of perception; p is a proposition that can only be known on the basis of such evidence. The same is true of the closely related proposition that the length of stick s at t is one meter, if s exists at t. Although it is conceivable that one could know this proposition without knowing the slightly stronger proposition p, there is no way one could know it that doesn’t rely on empirical evidence for justification. Thus, Kripke’s example of the meter stick is not a genuine case of the contingent apriori; nor are any of his other examples.15 Because of requirements on what it takes to successfully introduce a name with a description that semantically fixes its referent, there are no examples of the contingent apriori which depend upon genuine names the referents of which are semantically fixed by descriptions.

Are There Examples of the Contingent Apriori? I believe there are, the best examples being those that involve the actuality operator discussed in chapter 14.16 Briefly put, this operator works as follows: Let S be a sentence that expresses a proposition p in a context of utterance C. Then, in C, Actually S expresses a proposition that says, of the actual world-state of C, that it is a world-state in which p is true—i.e., it says that p is true with respect to (or given) the way the world actually is. Hence, the proposition expressed by Actually S in C is true when evaluated with respect to an arbitrary possible world-state w iff S is true with respect to the actual world-state of C; since this never changes—i.e., since whether or not S is true with respect to the fixed actual world-state never changes as one moves from 15 Nathan Salmon reports that Kripke told him, in personal communication, that in his discussion of the meter-stick example in Naming and Necessity he had in mind “a case in which the reference-fixer sees S there in front of him and uses the description referentially to refer to that length” (p. 200 of Salmon’s “How to Measure the Standard Metre”). Nevertheless, Kripke did not conclude, as Salmon and I do, that in that case the example cannot properly be regarded as an instance of the contingent apriori. 16 Examples of this type were pointed out by David Kaplan in “On The Logic of Demonstratives,” Journal of Philosophical Logic 8 (1979): 81–98, and in “Demonstratives.”

418

KRIPKE

one merely possible world-state to another—whenever S expresses a truth in C, Actually S expresses a necessary truth. The corresponding fact about descriptions is that whenever the x: Fx successfully denotes a unique individual o in the actual world-state of a context of utterance, the x: actually Fx, as used in that context, denotes o with respect to all possible world-states in which o exists, and never denotes anything else. Hence actually is a rigidifier which, when added to a description that denotes a unique individual with respect to the context of utterance, results in a new description that rigidly denotes that individual with respect to all possible world-states in which that individual exists. The description the x: actually Fx may also be paraphrased the actual F. If asked what this description means, we answer that it may be paraphrased the unique object which is F in @, where ‘@’ designates the way that the world actually is—i.e., the maximally complete property that the world actually instantiates. Thus, a sentence The actual F is G says, in effect, that which is said by The unique object which is F in @ is also G. With this in mind, we may consider the following examples. 8a. Princeton University has a philosophy department iff actually, Princeton University has a philosophy department. b. If stick s exists at t (and hence has a unique length at t), then the length of stick s at t is the actual length of stick s at t. These are genuine examples of the contingent apriori. First contingency. Since Princeton has a philosophy department, the right-hand side of the biconditional (8a) is a necessary truth and the left-hand side is a contingent truth; hence (8a) is itself contingent. (8b) is also contingent, since although it is true with respect to the actual state of the world, it is false with respect to possible states of the world w in which s exists at t, but either its length at t is greater than the length it has with respect to the actual state of the world, due to its having been heated and stretched in w, or its length at t is less than the length it has with respect to the actual state of the world, due to its having been cooled and compressed in w. Next aprioricity. The basic principle to which we need to appeal is (9a), and its corollary (9b). 9a. For any proposition p and possible world-state w, one may know p in w on the basis of evidence e iff in w, one may know, of w, that it is a world-state in which p is true, on the basis of that same evidence e.

THE CONTINGENT APRIORI

419

b. For any proposition p and possible world-state w, one may know p apriori in w iff in w, one may know apriori, of w, that it is a world-state with respect to which p is true. Taking these principles plus certain obvious subordinate principles as given, we can establish that the proposition expressed by (8a) is knowable apriori as follows.17 10(i) The proposition that Princeton University has a philosophy department iff Princeton University has a philosophy department is knowable apriori (in the actual world-state). Call this the proposition that P iff P. (ii) From (i) plus (9b), it follows that in the actual world-state @, it is knowable apriori, of @, that it is a world-state with respect to which the proposition that P iff P is true—i.e., that it is a world-state w which is such that the proposition that P iff P is true with respect to w. (iii) So, in (or with respect to) @, it is knowable apriori, of @, that it is a world-state w which is such that the proposition that P is true with respect to w iff the proposition that P is true with respect to w. (iv) So, in (with respect to) @, it is knowable apriori, of @, that it is a world-state w which is such that the proposition that P is true with respect to w iff the proposition that P is true with respect to @. (v) So, in (with respect to) @, it is knowable apriori, of @, that it is a world-state w which is such that the proposition that P iff the proposition that P is true with respect to @ is true with respect to w—i.e., in @, it is knowable apriori, of @, that it is a world-state with respect to which the proposition that P iff the proposition that P is true in @ is true. (vi) From (v) and (9b) it follows that, in (with respect to) @, the proposition that P iff the proposition that P is true in @ is knowable apriori.

17 In (10), I use ‘P’ as an abbreviation of the sentence ‘Princeton University has a philosophy department’.

420

KRIPKE

(vii) Since @ is the actual world, the proposition that the proposition that P is true in @ is the proposition expressed by Actually P. Thus, it follows from (vi) that the proposition expressed by P iff actually P—i.e., the proposition expressed by (8a)—is knowable apriori (in the actual world). It may be helpful to think of this argument schematically, along the lines of (11).18 11(i) P iff P (ii) So, ␭w[(P iff P) w] @ (iii) So, ␭w [P(w) iff P(w)] @ (iv) So, ␭w [P(w) iff P(@)] @ (v) So, ␭w [(P iff P@)w] @ (vi) So, P iff P@ (vii) So, P iff actually P In (i) of (11), we start with a claim that is obviously knowable apriori. We then use (9b) to assure us that (ii) is apriori, where (ii) tells us that @ has the property of being a world-state w with respect to which it is true that P iff P. (iii) is an apriori consequence of (ii); it tells us that @ has the property of being a world-state with respect to which the proposition that P is true iff it is a world-state with respect to which the proposition that P is true. The key move is to (iv).19 Consider an analogy. Suppose that m has the property that any man has just in case he loves Anna iff he loves Betty—i.e., ␭x[Lxa iff Lxb] m. Then m also has the property that any man has just in case he loves Anna iff m loves In (11) I continue to use ‘P’ as an abbreviation for ‘Princeton University has a philosophy department’. I use ‘@’ to designate the actual state of the world, and I use ‘w’ as a variable over world-states. When S is any sentence, I use S@ to express the claim that attributes the property of being true with respect to the actual world-state to the proposition that S expresses, and I use the formula S(w) to express the claim that that proposition is true with respect to the world-state which is the value of ‘w’. Where . . . v . . . is any formula containing one or more free occurrences of the variable v, ␭v [. . . v . . .] is a predicate expressing a property that is true of an object o iff . . . v . . . is true of o relative to an assignment of o to ‘v’. 19 The idea for this move can be found on pp. 210–11 of Gareth Evans, “Reference and Contingency,” in his Collected Papers (Oxford: Oxford University Press, 1985). Despite this insight, in my opinion Evans misconstrues examples involving names the referents of which are fixed by description. See pp. 148–50 of my review of his Collected Papers, Journal of Philosophy 86 (1989). 18

THE CONTINGENT APRIORI

421

Betty—i.e., ␭x[Lxa iff Lmb] m. Similarly, (iii) tells us that @ has the property that any world-state has just in case the proposition that P is true with respect to that world-state iff the proposition that P is true with respect to that world-state—i.e., ␭w [P(w) iff P(w)] @. It follows that @ also has the property that any world-state has just in case the proposition that P is true with respect to that world-state iff the proposition that P is true with respect to @—i.e., ␭w [P(w) iff P(@)] @. (v) is an apriori consequence of (iv); it tells us that @ has the property of being a world-state with respect to which the proposition that P iff P is true with respect to @ is true. We then use (9b) again (this time in the right-to-left direction) to assure us that the proposition expressed by (vi) is knowable apriori, which is the same as the proposition expressed by (vii)—namely, the proposition expressed by (8a).20 Thus, in addition to being contingent, the proposition expressed by (8a) is knowable apriori. A similar demonstration can be given that the proposition expressed by (8b) is knowable apriori, on the basis of the obvious fact that the proposition expressed by (8c) is knowable apriori. 8c. If stick s exists at t (and hence has a unique length at t), then the length of stick s at t is the length of stick s at t. Since the proposition expressed by (8b) is also contingent, it too is a genuine example of the contingent apriori. In general, for every There is a potential objection to principles (9a) and (9b) (in the right-to-left direction) which suggests that they may not be true strictly as stated. However, the objection does not affect our use of them in the above derivation. The objection is based on the following scenario. Suppose one describes a possible world-state w to oneself in some detail, and concludes, correctly, that it is a possible world-state with respect to which a certain proposition q is true. This may count as knowledge, of w, that p is true with respect to it. Let us suppose that it does. Suppose further that w is, in fact, the actual world-state—the world-state with respect to which one is doing the relevant describing and thinking—but that one doesn’t realize that the world-state one has been describing and thinking about is the state that the world is actually in. In such a case, one might believe of the actual world-state that q is true with respect to it, without believing q, and without having any justification for believing q. If there are possible scenarios like this, then they threaten the right-to-left direction of (9a), and perhaps even (9b). However, even if they falsify (9a) and (9b) as stated, they do not invalidate the above derivation, and in particular, they do not invalidate the move from (v) to (vi). In the scenario relevant to the derivation, the agent starts with apriori knowledge of a certain proposition, and moves to knowledge of the actual world-state that the proposition is true with respect to it—which he can express using the demonstrative this world-state with the understanding that it refers to the world-state he directly experiences. Throughout the entire derivation, where we use ‘@’ the agent can use this world-state with that understanding. If the agent does this, then all the argumentative moves, including the move from (v) to (vi), will be transparent.

20

422

KRIPKE

Kripke-style example of a sentence . . . n . . . containing a name n the referent of which is fixed by a description the D that is supposed to be an instance of the contingent apriori (but really isn’t), the corresponding sentence in which the actual D is substituted for n is a genuine example of the contingent apriori—i.e., it is a sentence that expresses a proposition which, though true, and knowable apriori, is not a necessary truth.21 In short, although none of Kripke’s putative examples of the contingent apriori are correct, and although names the referents of which are semantically fixed by descriptions in his sense never give rise to genuine instances of the contingent apriori, Kripke showed us what to look for, and put us on the track to finding the examples that really do make his point. Surely, that is the thing to remember. 21 Here, I rely on our earlier discussion in chapter 14 which supports the claim that for Kripke, even when a name has its referent semantically fixed by a description the D, it is not synonymous with the actual D.

CHAPTER 17 NATURAL KIND TERMS AND THEORETICAL IDENTIFICATION STATEMENTS chapter outline 1. Extending the Analysis of Names to Natural Kind Terms Proper names and natural kind predicates: similarities and differences Rigid designation is not defined for predicates 2. Identity Sentences Containing Natural Kind Predicates The logical forms of such sentences 3. The Modal Status of Theoretical Identities Involving Simple Natural Kind Predicates Non-descriptionality and the explanation of why some of these sentences are necessary, if true 4. Extending the Account The necessity of certain theoretical identities involving semantically complex, theoretically interesting predicates 5. The Epistemic Status of Theoretical Identities Involving Natural Kind Predicates Why some are knowable only aposteriori 6. A Final Challenge to the Necessary Aposteriori The hypothetical objector and Kripke’s unsatisfying reply 7. A Proper Response to the Challenge: Essential Properties, Impossible World-States, and the Necessary Aposteriori The difference between epistemic and metaphysical possibility

Extending the Analysis of Names to Natural Kind Terms In this chapter, we will discuss how Kripke extends his central theses about proper names to the broader, more heterogeneous, and more philosophically significant class of natural kind terms. His claim that true theoretical identity sentences involving such terms are examples of the necessary aposteriori is particularly important in this connection.

424

KRIPKE

This claim captured the imagination of philosophers, and held out the promise of important and widespread applications in philosophy as a whole, in a way that went well beyond the perceived implications of his similar claim about identity sentences containing proper names. Still, his analysis of proper names laid the foundation for his treatment of natural kind terms, and it is important to understand how his views about the latter grew out of his views about the former. The central theses about names defended in the first two lectures of Naming and Necessity are summarized by T1–T4. theses about proper names T1. Proper names are non-descriptional: (i) they are not synonymous with descriptions or clusters of descriptions associated with them by speakers; (ii) the referent of a name with respect to an arbitrary world-state w is not determined semantically via the satisfaction of any description or descriptive condition with respect to w; instead, (iii) the referent of a name is initially fixed at the actual world-state, and, once fixed, is stipulated to remain the same with respect to all other world-states. T2. The referent of a proper name is initially fixed in one or the other of two ways—by an ostensive baptism, or by a stipulation that it is to be whatever satisfies a certain description. Later, when the name is passed from speaker to speaker, the way in which the reference was initially fixed usually doesn’t matter. Typically, speakers further down the historical chain use the name to refer to the initial referent whether or not they associate properties with the name that (uniquely) apply to it. T3. Proper names are rigid designators—i.e., a proper name that designates an object o does so with respect to all world-states in which o exists, and never designates anything else. T4. Identity sentences involving different but coreferential names (or other rigid designators) express necessary truths. Nevertheless, often these truths are knowable only aposteriori. In lecture 3 of Naming and Necessity, similar theses are defended for natural kind terms. For example, Kripke argues at length that natural

NATURAL KIND TERMS

425

kind terms like gold, tiger, cat, water, heat, and light are not synonymous with clusters of descriptions standardly associated with them by speakers. As in the case of proper names, two ways are given by which the reference of a term may be fixed. One way involves direct presentation of samples of the putative kind, together with the stipulation that the term is to be understood as applying to all and only instances of the unique natural kind (of a certain sort) of which nearly all members of the sample are instances. The other way of fixing the reference of a natural kind term involves the use of a description that picks out the kind, or members of the kind, by some, usually contingent, properties. Later, when the kind term is passed from speaker to speaker, the way in which the reference was initially established normally doesn’t matter—just as with proper names. As a result, speakers further down the chain may use the term to apply to instances of the given kind whether or not the descriptive properties they associate with the term really pick out members of that kind. In addition, scientific investigation may lead to the discovery of properties that are necessary and sufficient for membership in the kind. These properties are expressed in theoretical identity (or identification) sentences that are necessary but aposteriori. Examples of such theoretical identity sentences specifically discussed in Naming and Necessity are: Water is H2O. Flashes of lightning are flashes of electricity. Light is a stream of photons. Gold is the element with atomic number 79. Cats are animals. Whales are mammals. Heat is the motion of molecules. The parallels between Kripke’s treatment of proper names and his discussion of natural kind terms are evident. However, there are special complications that arise in the discussion of natural kind terms. Among the most important of these are questions about rigid designation, and related questions about the modal properties of certain identity sentences. As in the case of proper names, natural kind terms are said to be rigid, and the putative rigidity of these terms is used to support the corollary that theoretical identity sentences involving them are necessary, if true. For example, in discussing theoretical identifications involving natural kind terms, Kripke says, “Theoretical identities, according to the conception I advocate, are generally identities involving two rigid designators and therefore are examples of the necessary aposteriori.”1 However, there is a potential difficulty here that has not 1

Naming and Necessity, p. 140, my emphasis.

426

KRIPKE

been widely appreciated. Kripke gives no separate definition of what it means to say that a natural kind term is rigid; nor does he provide distinct arguments to show that they are rigid. This is a problem because his explicit definition of rigidity tells us only what it is for a singular term to be rigid. If all natural kind terms were just ordinary singular terms, each purporting to designate a single object, then this definition could be applied directly to them, without qualification. However, as Kripke recognizes, natural kind terms fall into a variety of syntactic and semantic categories. For example, he says, According to the view I advocate, then, terms for natural kinds are much closer to proper names than is ordinarily supposed. The old term ‘common name’ is thus quite appropriate for predicates marking out species or natural kinds, such as ‘cow’ or ‘tiger’. My considerations apply also, however, to certain mass terms for natural kinds, such as ‘gold’, ‘water’ and the like.2 A little later, summing up his views, he adds, my argument implicitly concludes that certain general terms, those for natural kinds, have a greater kinship with proper names than is generally realized. This conclusion holds for certain for various species names, whether they are count nouns, such as ‘cat’, ‘tiger’, ‘chunk of gold’, or mass terms such as ‘gold’, ‘water’, ‘iron pyrites’. It also applies to certain terms for natural phenomena, such as ‘heat’, ‘light’, ‘sound’, ‘lightning’, and, presumably, suitably elaborated, to corresponding adjectives—‘hot’, ‘loud’, ‘red’.3 It appears from these passages that Kripke intends his general theses about natural kind terms to apply, at least in some form, to terms of various syntactic and semantic categories. This raises obvious questions, among them, “What is it for a predicate to be a rigid designator?”, “Are natural kind predicates involving general terms like cow, tiger, animal, chunk of gold, and flash of lightning rigid?”, “What is it for a sentence containing a pair of natural kind predicates to count as a theoretical identity sentence?”, and “Are such sentences necessary if true? If so, do some of them express claims that are knowable only 2 3

Ibid., p. 127, my emphasis. Ibid., p. 134.

NATURAL KIND TERMS

427

aposteriori?” Since Kripke nowhere explicitly answers these questions, it is a matter of speculation how he would deal with them. Although these questions are certainly important, it might seem as if they can easily be answered. In particular, it might seem straightforward how to define rigidity for predicates. Think again about rigidity for singular terms. In the case of a singular term, t, the object designated by t with respect to a world-state w is the extension of t with respect to w, and the claim that t is rigid entails that it has the same extension with respect to every world-state at which it has an extension at all. In the case of a predicate, P, the extension of P with respect to a world-state w is the set of objects that P applies to, or is true of, with respect to w. One idea for characterizing a notion of rigidity for predicates parallel to the notion of rigidity for singular terms would be to stipulate that P is rigid only if P has the same extension with respect to any world-state at which it has a non-null extension. However, this clearly won’t do. Consider, for example, the natural kind predicate is an animal. Its extension with respect to a world-state is the set of all things that are animals at that world-state. Since the set of individual animals varies from one world-state to the next, this definition of rigidity would classify it as non-rigid. Since the same point could be made for virtually every predicate of contingently existing objects, rigidity for predicates shouldn’t be defined as requiring sameness of extension at different world-states (if it is to be a theoretically significant notion). Nor will it do to say that a predicate is rigid iff there is a unique property which it expresses that determines its extension at each possible world. It could, of course, be argued that there is such a property in the case of natural kind predicates involving terms like cow and animal—namely, the property of being a cow and the property of being an animal. However, the same could be said for virtually any predicate. For example, the predicates is a bachelor and is a philosopher also express properties that determine their extensions with respect to arbitrary possible world-states. In general, for virtually any predicate F one can think of, and any world-state w, the extension of F with respect to w is the set of things that have, at w, the property expressed by being an F. But there is no point in defining a notion of rigidity for predicates according to which virtually all predicates turn out, trivially, to be rigid. There is, it should be noted, an obvious alternative definition which does not have this consequence, and which is a natural extension of Kripke’s definition of rigidity for singular terms. The idea is that a

428

KRIPKE

predicate is rigid iff it is an essentialist predicate, where the latter is characterized as follows: EP. A predicate P is essentialist iff for all possible worlds w and objects o, if P applies to o with respect to w, then P applies to o in all worlds in which o exists. Two other ways of expressing the same idea are: EPa. A predicate P (of English) is essentialist iff Necessarily any individual that is (was) P could not have existed without being P expresses a truth. EPb. A predicate P is essentialist iff the property it expresses is an essential property of anything that has it. The parallels between this interpretation of rigidity for predicates and the corresponding theses for rigid singular terms are obvious. For example, the linguistic test for the essentiality/rigidity of English predicates provided by EPa is similar to Kripke’s rigidity test for singular terms. RT. A singular term t (of English) is a rigid designator of an object iff The individual that is (was) t could not have existed without being t (and no one other than that individual could have been t) expresses a truth. According to this test, the name Aristotle is rigid, since necessarily, the individual who was Aristotle could not have existed without being Aristotle (and no one else could have been Aristotle); whereas the definite description the teacher of Alexander is not rigid. Similarly, the predicate is an animal is (arguably) essentialist, and therefore, on this interpretation, rigid, since necessarily anything that is an animal could not have existed without being an animal. By contrast, is a philosopher is not essentialist/rigid. Although this definition of rigidity for predicates is more plausible than the other two, it also won’t work. As I demonstrate in chapter 9 of Beyond Rigidity, it suffers from two crushing defects. First, not all natural kind predicates covered by Kripke’s theory are essentialist. Since these predicates occur in necessary aposteriori statements of theoretical identity, the presence of essentialist predicates is not crucial to Kripke’s explanation of these cases.4 Second, even though many natural kind 4 The example (10b) discussed below, which contains the two-place natural kind predicate hotter than, is an illustration of this point.

NATURAL KIND TERMS

429

predicates that Kripke discusses are essentialist, the fact that they are essentialist does not explain the necessity of true statements of theoretical identity involving them.5 Since it is precisely an explanation of such cases of the necessary aposteriori that we are most interested in, extending the notion of rigid designation to predicates in this way is of no help. At this point, there are two different ways in which we might proceed. One way would be to ignore predicates and restrict ourselves to those uses of natural kind terms in which they can naturally be understood to be functioning as names of natural kinds themselves—where each kind is taken to be an abstract object which has concrete individuals as instances (or members).6 On this approach, we would appeal to the concept of rigidity as defined for singular terms, and try to apply Kripke’s doctrines about identity sentences of the form  ⫽  to reach results about necessary aposteriori identity statements formulated using names of abstract kinds.7 A different way of proceeding would be to focus on natural kind predicates directly, to put aside the concept of rigidity, and to identify other ways in which, according to Kripke, natural kind predicates are similar to names. On this approach, we first have to decide what counts as a theoretical identity sentence involving natural kind predicates; next, we have to identify features of these predicates which they share with proper names that are responsible for (i) the fact that true theoretical identification sentences involving them are necessary, if true, and (ii) the fact that sometimes these sentences express claims that are knowable only aposteriori. There are two reasons for favoring the second approach, which we will adopt here. First, it is more general; for some natural kind terms, See pp. 257–59 of Beyond Rigidity. These uses are explained in chapter 11 of Beyond Rigidity. 7 There is a different way of thinking of some natural kinds which takes them to be large, scattered concrete objects. For example, gold is sometimes thought to be a scattered object—the gold-fusion—of which each individual gold thing is a part. However, if ‘gold’ is treated as a singular term naming this unusual object, then it is hard to think of it as a rigid designator—for it would seem that gold could have existed, even if none of the chunks of matter that are gold in the world as it actually is existed. Let w be a possible world-state with respect to which that is so. With respect to w, the proposal that ‘gold’ is a singular term that designates the sum of all existing gold in the world would have to treat ‘gold’ as designating a gold-fusion with no parts in common with the gold-fusion it designates with respect to the actual state of the world. Since it is hard to see how the gold-fusion in w could be the very same concrete object as the gold-fusion in the world as it actually is, it is hard to see how, on this proposal, ‘gold’ could be a rigid designator. I will therefore put this possible interpretation aside. For further discussion, see chapter 11, footnote 19 of Beyond Rigidity; see also Kathrin Koslicki, “The Semantics of Mass Predicates,” Noûs 38 (1999):46–91. 5 6

430

KRIPKE

e.g., star and electron, their only natural use is as predicates rather than names, and it is awkward and artificial to try to come up with complex singular terms appropriately related to them that can be used to play the naming role required by the first approach. Second, in the case of many natural kind terms—e.g., the adjective red—which are capable of functioning both as predicates of concrete individuals (The ball is red) and as names of abstract kinds (Red is a color), we seem to understand the predicate prior to understanding the associated singular term, which suggests that the semantic properties of predicates should be specifiable without having to appeal to the related singular term.

Identity Sentences Containing Natural Kind Predicates8 In order to understand and evaluate the doctrine that theoretical identity sentences containing natural kind predicates are necessary, if true, we must first understand what sentences we are talking about. Normally, when we talk about identity sentences, we have in mind sentences of the form (1), where ␣ and ␤ are names or other singular terms. 1. ␣ ⫽ ␤ If (1) is true, then these singular terms refer to the same thing. This means that if ␣ and ␤ are rigid, they will refer to the same thing with respect to every possible world-state in which the (actual) referent of ␣ (and ␤) exists. So, if (1) is true and the terms are rigid, then (1) is true with respect to all possible world-states, and hence is necessary.9 However, when natural kind predicates are involved, the question of what sort of sentence is to count as an identity is more complicated. At various points throughout lecture 3, Kripke discusses a class of statements involving natural kind terms that he calls “theoretical identifications.” The initial examples of these sentences are (2–5).10 2. Light is a stream of photons. 3. Water is H2O. 4. Lightning is an electrical discharge. 5. Gold is the element with the atomic number 79. The material in this section is also discussed in chapter 9 of Beyond Rigidity. For simplicity, I ignore possible world-states in which the referent of the terms doesn’t exist. 10 Naming and Necessity, p. 116. 8 9

NATURAL KIND TERMS

431

It is worth noting that (2) and (4) do not even appear to be sentences in which a pair of singular terms flank the identity sign. Later in the lecture, Kripke considers other statements that he seems to place in the same category as the original examples (2–5). One such statement is (6). 6. Cats are animals. Clearly, this is not a sentence the logical form of which is given by (1). Rather, its logical form is normally regarded by philosophers as involving a pair of predicates, and as being representable (at least for many purposes) by something like (7). 7. ᭙x (x is a cat  x is an animal) For all x, if x is a cat, then x is an animal. Although it doesn’t contain the identity predicate, (7) might broadly be classified as an identity statement on the grounds that it identifies each cat with some animal. In general, English sentences of the form 8a. (All) A’s are B’s, An A is a B (on one of its uses) and 8b. All and only A’s are B’s, Something is an A iff it is B (on one use) may naturally be counted as statements that identify A’s with B’s, even though they are routinely represented by formulas of the form (9), which do not contain the identity sign. 9a. ᭙x (Ax  Bx) b. ᭙x (Ax ↔ Bx) Next consider (4), from Kripke’s original list of theoretical identifications. Here, the expression is an electrical discharge seems to function as a predicate, just as the expression is a cat does in the sentence Felix is a cat. Thus (4) is not the sort of identity statement in which a pair of singular terms flanks the identity sign. Rather, it says that any individual instance of lightning is an electrical discharge, in which case it is to be understood on the model of (9a). Once this is recognized, one might analyze (2) as also being of this form. The lesson here is that theoretical identity statements need not involve singular terms, or the identity sign, but instead may have the

432

KRIPKE

form of universally quantified conditionals or biconditionals. This point is further illustrated by the following passage from Naming and Necessity. [I]t it turns out that a material object is (pure) gold if and only if the only element contained therein is that with atomic number 79. Here, the ‘if and only if’ can be taken to be strict (necessary). In general, science attempts, by investigating basic structural traits, to find the nature, and thus the essence (in the philosophical sense) of the kind. The case of natural phenomena is similar; such theoretical identifications as ‘heat is molecular motion’ are necessary, though not apriori. The type of property identity used in science seems to be associated with necessity, not with aprioricity, or analyticity: For all bodies x and y, x is hotter than y if and only if x has a higher mean molecular kinetic energy than y. Here the coextensiveness of the predicates is necessary, but not a priori.11 In this passage, Kripke gives two examples of necessary aposteriori statements of theoretical identification that have the form of universally quantified biconditionals. Moreover, he seems to suggest that the doubly quantified biconditional (10b) may be an analysis of the identity (10a). 10a. Heat is molecular motion. b. For all bodies x and y, x is hotter than y if and only if x has a higher mean molecular kinetic energy than y. In light of all this, it seems clear that Kripke’s claims about theoretical identity sentences involving natural kind terms should be seen as encompassing cases in which the terms function as predicates, and the identity sentences have analyses which are universally quantified conditionals or biconditionals.

The Modal Status of Theoretical Identities Involving Simple Natural Kind Predicates12 What is it about natural kind predicates that makes Kripke confident that theoretical identity sentences containing them are (often) necessary, Ibid., p. 138, my emphasis. The material in this section and the next draws heavily on the first half of chapter 10 of Beyond Rigidity.

11 12

NATURAL KIND TERMS

433

if true? On page 140 he seems to tell us when he announces that “theoretical identities, according to the conception I advocate, are generally identities involving two rigid designators and therefore are examples of the necessary a posteriori.” However, I have already noted that this answer is problematic and insufficiently general, since the notion of a rigid designator has not been defined for predicates. Fortunately, Kripke has a better answer available. Like names, natural kind predicates are non-descriptional—i.e., they are not synonymous with descriptions associated with them by speakers, and their reference (the set of things they correctly apply to) is not semantically determined by such descriptions. This, I suggest, is the important parallel with proper names. In the first section of this chapter, I quoted passages from pages 127 and 134 which indicate that Kripke intended his analysis of the parallels between proper names and natural kind terms to apply to kind terms of several grammatical categories, including those that function as predicates. It is striking that, although he sometimes mentioned rigidity as being included among the parallels between names and natural kind terms, it is non-descriptionality that occupies center stage in the discussion immediately following the second of these two passages. [M]y argument implicitly concludes that certain general terms, those for natural kinds, have a greater kinship with proper names than is generally realized. This conclusion holds for certain for various species names, whether they are count nouns, such as ‘cat’, ‘tiger’, ‘chunk of gold’, or mass terms such as ‘gold’, ‘water’, ‘iron pyrites’. It also applies to certain terms for natural phenomena, such as ‘heat’, ‘light’, ‘sound’, ‘lightning’, and, presumably, suitably elaborated, to corresponding adjectives—‘hot’, ‘loud’, ‘red’. Mill, as I have recalled, held that although some ‘singular names’, the definite descriptions, have both denotation and connotation, others, the genuine proper names, had denotation but not connotation. Mill further maintained that ‘general names’, or general terms, had connotation. Such terms as ‘cow’ or ‘human’ are defined by the conjunction of certain properties which pick out their extension—a human being, for example, is a rational animal with certain physical characteristics. The hoary tradition of definition by genus and differentia is of a piece with such a conception. If Kant did, indeed, suppose that ‘gold’ could be defined as ‘yellow metal’, it may well be this tradition which led him to the definition. . . .

434

KRIPKE

The modern logical tradition, as represented by Frege and Russell, disputed Mill on the issue of singular names, but endorsed him on that of general names. Thus all terms, both singular and general, have a ‘connotation’ or Fregean sense. More recent theorists have followed Frege and Russell, modifying their views only by replacing the notion of a sense as given by a particular conjunction of properties with that of a sense as given by a ‘cluster’ of properties, only enough of which need apply. The present view, directly reversing Frege and Russell, (more or less) endorses Mill’s view of singular terms, but disputes his view of general terms.13 This passage occurs in the middle of the third lecture, where Kripke says he is “recapitulating” his main theses about natural kind terms before going on to a discussion of the mind-body identity theory. His recapitulation involves the point just made about the non-descriptionality of natural kind terms (of various grammatical categories), plus his account of how the reference of these terms is fixed. It is evident that for Kripke the objects to which natural kind terms correctly apply with respect to different possible states of the world are not determined by whatever descriptive properties, if any, speakers associate with the terms. It is true, of course, that speakers often associate them with descriptive properties that they use to identify particular instances of the kinds. However, these properties typically fail to provide necessary and sufficient conditions for something to be a member of a kind (with respect to a world-state), and sometimes the properties associated with such a term are not even true of actual instances of the kind (as when speakers think of a whale as a kind of fish). Just as someone can successfully use a proper name to refer to an object without associating the name with descriptive properties that uniquely apply to its referent, so a speaker can successfully use a natural kind predicate to say something about the members of a kind, even if he lacks the ability to accurately describe either the kind itself or its instances. Next, we turn to Kripke’s positive account of how the extension of a natural kind predicate (the class of things to which it correctly applies) with respect to an arbitrary world-state is semantically determined. According to that account, the predicate is first associated by speakers with a kind—either ostensively or via a description. In the ostensive case, speakers directly associate the predicate with a sample of individuals, which they presume to be instances of a single natural kind k of a given 13

Naming and Necessity, pp. 134–35.

NATURAL KIND TERMS

435

type (e.g., a single substance, or a single species). Since kinds are hierarchically arranged in terms of genus and species, there is never a single natural kind of which members of the sample are instances. Thus, the presumption that k be of some definite type is crucial when a natural kind term is introduced in this way. (For example, when we introduce gold ostensively in terms of a sample, we think of the kind as being a type of metal.) In the case in which a predicate is associated with a kind descriptively, speakers employ a description that picks out a unique kind, often by appeal to contingent properties of the kind or its instances. Once the kind k has been determined, either ostensively or by description, it is understood that for any world-state w, the extension of the predicate with respect to w is to be the set of instances of k with respect to w.14 With this in mind, let us return to the question of whether theoretical identity sentences involving natural kind predicates are necessary, if true. The explanation that follows, though constructed from central aspects of Kripke’s account, is not explicit in his text. Moreover, it does not cover every “theoretical identification sentence” mentioned there. However, it does apply to many of the most important examples. Since I don’t know of any other explanation that covers all of his examples, and since the examples themselves seem to fall rather naturally into different categories, I do not regard the incompleteness of the explanation that follows to be a crippling defect.15 The explanation concerns theoretical identity sentences of the form (8a,b), which are analyzed along the lines of (9a,b). 8a. (All) A’s are B’s, An A is a B (on one of its uses) 8b. All and only A’s are B’s, Something is an A iff it is B (on one use) 9a. ᭙x (Ax  Bx) b. ᭙x (Ax ↔ Bx) This picture should not be credited to Kripke alone. Substantially the same account was developed independently by Hilary Putnam. See in particular Putnam’s “Is Semantics Possible?” in H. Kiefer and M. Munitz, eds., Language, Belief and Metaphysics (Albany: State University of New York Press, 1970); “Explanation and Reference,” in G. Pearce and P. Maynard, eds., Conceptual Change (Dordrecht: Reidel, 1973); “Meaning and Reference,” Journal of Philosophy 70 (1973):699–711; “The Meaning of ‘Meaning’,” in K. Gunderson, ed., Language, Mind, and Knowledge, Minnesota Studies in the Philosophy of Science, no. 7 (Minneapolis: University of Minnesota Press, 1975). All of these papers are reprinted in Hilary Putnam, Philosophical Papers, vols. 1 and 2 (Cambridge: Cambridge University Press, 1975). 15 For further discussion of the scope of this explanation, as well as the relationship between it and Kripke’s text, see chapters 9 and 10 of Beyond Rigidity. 14

436

KRIPKE

To begin with, I will restrict attention to those sentences of this form in which the predicates A and B are different, semantically simple, natural kind predicates (often, single words). In addition, it will be convenient, initially, to further restrict the case by assuming that the reference of A is fixed ostensively. That is, we assume that the reference of A is fixed by stipulating that it is to apply to all and only instances of the unique natural kind of a certain type T (e.g., a substance or species) of which nearly all elements of the sample used in introducing it are instances. Unlike A, the predicate B is not required to be an ostensive natural kind term; rather, we allow its extension to be fixed either ostensively or by description. However, we do require that A and B be predicates associated with kinds of the same type T—for instance, both may be substance terms, both may be species terms, or both may be of some other category.16 When A and B are related in this way, it may turn out that the relationship between their extensions is not evident to competent speakers, but is discoverable only by empirical investigation. Let us suppose that this is so, and that, as the result of empirical investigation, it has now been discovered that every object in the extension of A is also in the extension of B. This discovery establishes that sentence (8a) is true. What needs to be shown is how this apparently modest result, combined with the semantic character of the predicates, suffices to guarantee that they designate the same natural kind, and hence to guarantee that (8a) is necessary (and (8b) is too). To show this, we reason as follows: (i) From the assumption that the ostensive natural kind predicate A has successfully been introduced, it follows that there is a unique natural kind kA (of a given type T) of which nearly all members of the sample associated with A are instances, and A applies (with respect to a world-state) to all and only instances of kA (with respect to that world-state). (ii) From the assumption that the natural kind predicate B has successfully been introduced, it follows that there is a natural kind kB which is such that B applies (with respect to a world-state) to all and only members of kB (with respect to that This requirement imposes an important limitation on the explanation being proposed— namely, that it does not apply to examples like Cats are animals or Whales are mammals. In these cases, although the A and B predicates are natural kind terms, they are not intended to designate kind of the same type; cats and whales are species terms, whereas animals and mammals designate kinds of a higher type. Thus, if, as seems plausible, these sentences are necessary if true, some different and independent explanation must be given. The most likely explanation is that it is a metaphysically essential property of the pair consisting of the species cat and the kind animal that any instance of the former is an instance of the latter; ditto for the pair consisting of the species whale and the kind mammal. 16

NATURAL KIND TERMS

437

world-state). (iii) By hypothesis, the two predicates designate kinds of the same type (type T); thus both kA and kB are species, or both are substances, or both are kinds of some other category. (iv) If the theoretical identity sentence (8a) is true, then (since nearly all objects in the A-sample are A’s) nearly all of the objects in the A-sample are B’s, and hence they are instances of kind kB as well as kind kA. (v) Since, by hypothesis, the A-sample determines a single kind (of the given type T—a single species, substance, etc.), it follows that kind kA ⫽ kind kB. (vi) But this means that in addition to (8a), (8b) must also be true. (vii) Moreover, both must be necessary truths, since from steps (i), (ii), and (v) it follows that for all world-states w, the extension of A with respect to w ⫽ the set of instances of kA with respect to w ⫽ the set of instances of kB with respect to w ⫽ the extension of B with respect to w. (viii) So, if A and B are natural kind predicates of the sorts indicated, then a theoretical identity sentence (8b) involving those predicates is a necessary truth, if the corresponding sentence (8a) is true. This provides us with the paradigm (11). 11a. x (Ax  Bx) is true, when understood as intended. b. A and B are simple natural kind predicates of the same type (e.g., both species predicates, both substance predicates, etc.). Moreover, A “designates” the unique natural kind (of the specified type) instantiated by nearly all members of its associated sample. Where kA and kB are the kinds associated with (“designated by”) A and B, respectively, the extensions of A and B with respect to a world-state w are the sets of individuals that are members, with respect to w, of kA and kB, respectively. c. x (Ax  Bx) and x (Ax ↔ Bx) are both necessary truths. Although the argument may initially seem surprising, it is easy to see why it works. On Kripke’s account, the linguistic mechanism that fixes the extension of ostensive natural kind predicates guarantees that, if the term has been introduced successfully, then there can be only one natural kind, of the appropriate sort, of which nearly all members of the sample are instances. So, when A is ostensive, and the predicate B is related to A in the appropriate way, finding out that everything in the extension of A is in the extension of B is tantamount to finding out that the kind associated with A is identical with the kind associated with B. This powerful conclusion does not come from nowhere.

438

KRIPKE

Rather, it comes from the fact that the semantic presuppositions associated with the introduction of the natural kind predicates include very substantial non-linguistic claims—including the claim that the individuals in the sample associated with the ostensive predicate A are members of a single natural kind of a given type, and the claim that the predicates A and B stand for kinds of the same type.17 Because of this, one should be careful about drawing the moral from this example that it is easy to know or establish certain necessary aposteriori truths. One could equally well conclude that it is rather daunting to establish certain non-modal identification statements.18

Extending the Account So far, we have restricted paradigm (11) to universally quantified conditionals and biconditionals involving semantically simple natural kind predicates. However, we know that Kripke was willing to characterize a wider class of theoretical identity sentences, including (12) and (13), as being necessary, if true. 12. For all bodies x and y, x is hotter than y iff x has a higher mean molecular kinetic energy than y. 13. For all x, x is a drop of water iff x is a drop of a substance molecules of which contain two hydrogen atoms and one oxygen atom. Example (13) is a theoretical identification sentence of the form (9b) in which one of the predicates, is a drop of water, may, for present purposes, be taken to be a simple natural kind predicate, whereas the other predicate is a semantically compound expression that is necessarily For a brief discussion of what happens when these semantical presuppositions fail, see the section “When Semantic Presuppositions Fail,” in chapter 10 of Beyond Rigidity. 18 One should also be careful about taking this demonstration of how the necessity of certain statements can be inferred from their truth to be a story about how we go about discovering the truth and necessity of statements of this sort in practice. A reasonable idealized model of this might include (i) investigation of the A-sample to determine that most of its members are instances of a unique kind k of the relevant type T, (ii) projection of the extension of A from this finding, (iii) similar determination of the kind B and the extension of B, and (iv) comparison of kinds and extensions of the two predicates, (v) determination of truth of the claim All A’s are B’s, and (vi) use of relevant elements of the above model to derive the necessity of this claim. 17

NATURAL KIND TERMS

439

coextensive with the kind designated by that simple predicate.19 Because this compound predicate is not semantically simple, (13) is not an instance of paradigm (11). Thus, even though (13) is an example that is now (after Kripke) widely regarded as both necessary and aposteriori, nothing we have said up to now shows that its necessity has anything to do with the non-descriptive semantics assigned to the simple natural kind predicate is a drop of water by Kripke’s theory. But why should the complexity of the predicate on the right-hand side of (13) make a difference? Shouldn’t we be able to apply the reasoning behind (11) directly to (13), despite the fact that the B-predicate is semantically complex? The restriction in (11) to cases in which the Bpredicate is semantically simple seems inessential. Surely, it might be suggested, it is enough that B be a natural kind predicate. Isn’t the predicate is a drop of a substance molecules of which contain two hydrogen atoms and one oxygen atom such a predicate? At this point, before jumping to any conclusions, we would do well to compare (13) to (14) and (15). 14. For all x, x is a drop of water iff x is a drop of the substance instances of which fall from the sky in rain and fill the lakes and rivers. 15. For all x, x is a drop of water iff x is a drop of any one of the liquids that fall from the sky in rain and fill the lakes and rivers. These examples are, we may assume, true but contingent. Thus the reasoning in (11) cannot apply to them. But why not? First consider (14). The predicate on the right-hand side of this example contains a singular definite description that non-rigidly designates a certain substance. With respect to the actual state of the world, that substance is water; with respect to another world-state, a different substance may be designated. Since the substance designated by this description with respect to an arbitrary world-state is the kind that determines the extension of the compound predicate at the world-state, different kinds determine the extension of the predicate with respect to different world-states. As a result, the reasoning behind paradigm (11) fails at step (ii) (and step (vii)). Hence (14) is not characterized as necessary. This involves a simplification. In reality, the predicate is a drop of water is compound, in virtue of individuating water by the drop. I ignore this here, since it does not affect our central point. For a discussion of this issue, including an account of the semantics of simple mass predicates like is water, see chapter 11 of Beyond Rigidity.

19

440

KRIPKE

Next consider (15). In this case, the compound predicate on the right-hand side doesn’t contain any singular definite description, or any singular term, that either rigidly or non-rigidly designates any substance. Although, in fact, there is only one kind of liquid that falls from the sky in rain and fills the lakes and rivers, this substance—water—is not semantically associated with the predicate in the way required by paradigm (11). Rather, the compound predicate expresses a property which, as it happens, is instantiated by all and only instances of the substance water in the world as it actually is, but which could have been instantiated by a different substance, or even by a motley collection of different liquid substances, mixtures, and the like in different possible circumstances. Thus, the reasoning behind paradigm (11) again fails at step (ii) (and (viii)), and (15) is not characterized as necessary. What distinguishes (13) from (14) and (15)? Why are we inclined to think that the former is necessary, whereas the latter pair are not? The answer has to do with what we believe about substances. As Nathan Salmon pointed out many years ago, we believe that it is a feature of any genuine substance S that whatever its molecular structure turns out to be, all possible instances of S share that structure, and all possible instances of that structure are instances of S.20 From this plus the truth of (13) and the semantics of ‘water’, it will follow that (13) is necessary. But what is a substance, and why do we believe molecular structure to be constitutive of it? Here is a speculative proposal (which goes beyond anything explicitly indicated by Kripke). A substance is one type of natural kind. Natural kinds are abstract objects which have individual concrete entities as instances. Whatever these kinds are, they are individuated by their possible instances in the following (weak) sense: for all natural kinds x and y, x 苷 y iff there is some possible state of the world w such that the class of things that are instances of x with respect to w 苷the class of things that are instances of y with respect to w. In other words, kinds with the same possible instances are identical. By contrast, properties— particularly those expressed by semantically compound predicates like is an equiangular triangle and is an equilateral triangle—may differ even if they are necessarily coextensive. Even though properties are more fine-grained than natural kinds, they may determine kinds. We will say that a property P determines a natural kind k iff for all possible states of the world w, the extension of P with respect to w is just the extension 20

Nathan Salmon, Reference and Essence (Princeton, NJ: Princeton University Press, 1981).

NATURAL KIND TERMS

441

of k with respect to w. (Note that properties may determine at most one natural kind, whereas a given natural kind may have many properties that determine it.) Substances are natural kinds determined by properties of a special type—namely, properties of an individual, or sample of matter, x that specify how x is made, or constituted, out of basic physical stuff. With this in mind, consider the following idealized story about the introduction of the simple natural kind term water. Let us suppose that it is introduced with the intention that it is to be a substance term that applies to everything that shares the same physical constitution as the individual samples of water that led to its introduction.21 On this view, when we introduce the term, we may know neither what physical constituents make up these samples, nor how they are combined. Nevertheless, we intend the term to apply (with respect to a possible world-state) to all and only those things that share the basic physical constitution (whatever it may turn out to be) exhibited (in the world as it actually is) by all, or nearly all, elements in our sample. We may put this by saying that when we introduce the term water, we, in effect, stipulate that it is to apply to instances of the substance—the unique physically constitutive kind—of which nearly all members of the sample are instances.22 On this idealized model, it may well be that no one knows much about the nature of this physically constitutive kind when the predicate is a drop of water is introduced. Consequently, it is no threat to the model that the predicate was introduced long before the development of chemical theories of molecular structure. When these theories were developed, we acquired new families of concepts to describe these structures, and we came to understand claims about the molecular structures of macroscopic objects as claims about how those objects are constituted out of basic physical constituents. In short, these theories For a discussion of a somewhat less idealized scenario concerning the introduction of water, which does not require those introducing the term to have explicit intentions about physical constitution, see the section entitled “The Role of Intention in Determining the Reference of Natural Kind Terms” in chapter 10 of Beyond Rigidity. 22 Strictly speaking, this explanation applies to one, expansive, sense of the term water. This term also has a restricted sense in which it applies to all and only instances of the substance water that are in liquid form. In this sense, it contrasts with ice, which applies to all and only instances of the substance water in frozen form. Water, when used with its expansive sense, is a simple substance term; when used with its restricted sense, it, like ice, is partially descriptive. In order to explain these distinctions properly, one must deal with the use of the terms water and ice as simple mass predicates. This is done in chapter 11 of Beyond Rigidity, where a more complete theory is given, and various puzzles and problems are dealt with. 21

442

KRIPKE

brought with them a class of formulas expressing properties which, by virtue of specifying different possible molecular structures, were understood as determining different possible physically constitutive kinds or substances. After empirical investigation, certain formulas—e.g., molecules of x contain two hydrogen atoms and one oxygen atom—were taken to be candidates for giving the basic physical constitution of standard water samples. In this way, the predicates is a drop of water and is a drop of a substance molecules of which contain two hydrogen atoms and one oxygen atom came to be understood as designating physically constitutive kinds, and hence as designating kinds of the same type. This had the consequence that when it was discovered empirically that water samples have the molecular structure H2O, this was sufficient to guarantee that the two predicates designated the same kind—in the case of water, the kind is designated by virtue of being the very meaning of the term, whereas in the case of the compound predicate, the kind is designated by virtue of being determined by the property it expresses. Since this kind may be seen as determining the extensions of both predicates with respect to all possible world-states, it follows, without need of further assumptions, that (13) is a necessary truth. The same sort of explanation might be given for related examples, such as (16). 16. Chunks of gold are chunks of an element the atomic number of which is 79. What about other natural kind predicates that are not substance terms? One familiar sort of case involves what might be called explanatory kinds, where these are thought of as kinds determined by properties the possession of which causally explains certain observed characteristics. Heat and molecular motion, in Kripke’s discussion, seem to fall into this category. To see how this is supposed to go, imagine that the abstract singular term heat is introduced by an ostensive definition stipulating that it is to designate the unique kind of physical state or process present in a certain class of samples that causes certain effects, including, among other things, certain reactions in us. Imagine the further stipulation that the related predicate is hotter than is to apply to pairs in which the relevant physical state or process in one object is more pronounced than it is in the other object. When the kinetic theory is formulated as a hypothesis to be tested, it will be antecedently understood that a relational predicate characterizing different levels of mean molecular kinetic energy is a candidate for specifying the natural

NATURAL KIND TERMS

443

kind designated by is hotter than. Once it is discovered that certain facts about kinetic energy causally explain certain effects, it will follow that the kind designated by is hotter than just is the kind determined by the relation (i.e., the 2-place property) having-a-higher-meanmolecular-kinetic-energy-than, and Kripke’s example (12) will be characterized not only as true, but also as necessary. This explanation of the necessity of (12) is slightly different from the explanation of the necessity of (13), which was built on paradigm (11). When an explanatory kind is involved, the explanation runs roughly as follows: (i) A simple natural kind term E is introduced with the stipulation that it applies (with respect to any possible state of the world) to all and only instances of the kind determined by those properties possession of which (causally) explains certain phenomena in the world as it actually is. (ii) It is then discovered scientifically that the property of being such and such (causally) explains the phenomena. (iii) From this it follows that the kind directly designated by the simple predicate E is the kind determined by the property of being such and such. (iv) This is sufficient to establish the necessity of the claim that says that an individual (or a pair) “is E” iff it is (or they are) such and such.

The Epistemic Status of Theoretical Identities Involving Natural Kind Predicates So far we have constructed an account, using Kripke’s analysis of natural kind terms, to explain the necessity of certain theoretical identification sentences containing natural kind predicates. It remains to determine whether the necessary truths expressed by these sentences are knowable apriori, or only aposteriori. Let us begin with (13). In explaining why it is necessary, we noted that the two predicates it contains—is a drop of water and is a drop of a substance molecules of which contain two hydrogen atoms and one oxygen atom—designate the same natural kind. However, we also noted in passing that these predicates have different meanings. Whereas the meaning of water is simply the natural kind that it designates, the meaning of substance molecules of which contain two hydrogen atoms and one oxygen atom is a (complex) property that determines that kind. Since the two predicates mean different things, (13) is not synonymous with, and does not express the same proposition as, the trivial identity (13a). 13a. For all x, x is a drop of water iff x is a drop of water.

444

KRIPKE

Thus, the fact that the necessary proposition expressed by (13a) is knowable apriori does not show that the necessary proposition expressed by (13) is knowable apriori. That is all to the good. If, in addition, the kind water is such that one cannot know of it, except by empirical investigation, that for all x, x is an instance of it iff x is made up of molecules containing two hydrogen atoms and one oxygen atom, then the proposition expressed by (13) is not knowable apriori, but rather is a genuine example of the necessary aposteriori. This supposition is not unreasonable. After all, we don’t know the kind water by mere reflection, or by virtue of being given a verbal definition of any sort. Rather, our acquaintance with it is as something shared by a variety of concrete particulars which we use to identify it. This being so, one’s knowledge that drops of water are drops of H2O must be grounded, either in one’s own knowledge that particular water samples have the molecular structure H2O, or in the knowledge of others, passed on to one, which they are able to ground in their knowledge of particular water samples. This knowledge can only be aposteriori. An important part of this explanation is the account of why the two predicates in (13) do not mean the same thing. In giving this account, I have moved well beyond doctrines that Kripke explicitly endorses. In effect, I have suggested that we adopt a position that may justly be called Extended Millianism. Simple Millianism, which was inspired in its modern form by Kripke, but not endorsed by him, is the view that the meaning of a proper name is its referent. On this view, coreferential names mean the same thing, and true identity sentences involving them semantically express propositions that are necessary and apriori, rather than necessary and aposteriori (though such sentences may nevertheless be used in different contexts to assert or convey various aposteriori truths as well).23 Extended Millianism holds that the meaning of a simple natural kind predicate is the natural kind it designates; as a result, simple predicates that designate the same natural kind—such as groundhog and woodchuck—mean the same thing. This has the immediate consequence that the theoretical identity sentence 17a. For all x, x is a woodchuck iff x is a groundhog. which fits paradigm (11), and hence is a necessary truth, semantically expresses the same proposition as 23 This view is defended by Nathan Salmon in Frege’s Puzzle (Cambridge, MA: MIT Press, 1986), and by me in chapters 1–8 of Beyond Rigidity.

NATURAL KIND TERMS

445

17b. For all x, x is a woodchuck iff x is a woodchuck. and hence is knowable apriori (though it too can be used to assert or convey various aposteriori truths in different contexts). Things change when we consider compound expressions. Just as in the case of noun phrases, singular definite descriptions have meanings that are not identical with the objects they denote, but rather are properties in virtue of which they denote what they do, so a semantically compound natural kind predicate, such as is a drop of a substance molecules of which contain two hydrogen atoms and one oxygen atom, has a meaning that is not identical with the natural kind it designates, but rather is a property that determines that kind. On this picture, simple natural kind predicates are analogous to proper names, and compound descriptive predicates are analogous to singular definite descriptions. Because of this, it is possible for some sentences that have the logical form ᭙x (Ax ↔ Bx) to be both necessary, if true, and knowable only aposteriori—where one of the predicates is a simple natural kind predicate, and the other is a compound descriptive phrase.

A Final Challenge to the Necessary Aposteriori Before leaving Naming and Necessity, there is one final discussion, not far from the end of the work, that merits our close attention. After summing up his treatment of natural kind terms, and illustrating their role in generating surprising and significant examples of the necessary aposteriori, Kripke takes up a fundamental challenge to his conception of this class of truths. Up until this point in the lectures, when describing what he takes to be instances of the necessary aposteriori, he emphasizes that although they are necessary, and hence true with respect to every possible state of the world, nonetheless, for all we knew prior to empirically discovering their truth, “they could have turned out otherwise.” Realizing that this may sound puzzling and problematic, he gives voice in lecture 3 to the following objection. Theoretical identities, according to the conception I advocate, are generally identities involving two rigid designators, and therefore are examples of the necessary a posteriori. Now in spite of the arguments I gave before for the distinction between necessary and a priori truth, the notion of a posteriori necessary truth may still be somewhat puzzling. Someone may well be inclined to argue as

446

KRIPKE

follows: ‘You have admitted that heat might have turned out not to have been molecular motion, and that gold might have turned out not to have been the element with the atomic number 79. For that matter, you also have acknowledged that . . . this table might have turned out to be made from ice from water from the Thames. I gather that Hesperus might have turned out not to be Phosphorus. What then can you mean when you say that such eventualities are impossible? If Hesperus might have turned out not to be Phosphorus, then Hesperus might not have been Phosphorus. And similarly for the other cases: if the world could have turned out otherwise, it could have been otherwise.24 The problem here starts out being about theoretical identity sentences involving natural kind predicates, but quickly expands to cover all instances of the necessary aposteriori. Let p be such an instance. Since p is aposteriori, its falsity must be conceivable, and we need empirical evidence to rule out that possibility. Without such evidence it could turn out that, for all we know, p is false. But, the objector maintains, if p is necessary, there are no possibilities to be ruled out in which p is false, since no matter what possible state the world could be in, it is a state in which p is true. Thus, if p really is necessary, we don’t require empirical evidence to know p after all, and if p really is aposteriori, then p isn’t necessary; either way, the necessary aposteriori is an illusion. Kripke begins his reply to this objection with the following passage. The objector is correct when he argues that if I hold that this table could not have been made of ice, then I must also hold that it could not have turned out to be made of ice; it could have turned out that P entails that P could have been the case. What, then, does the intuition that the table might have turned out to have been made of ice or of anything else, that it might even have turned out not to be made of molecules, amount to? I think that it means simply that there might have been a table looking and feeling just like this one and placed in this very position in the room, which was in fact made of ice. In other words, I (or some conscious being) could have been qualitatively in the same epistemic situation that in fact obtains, I could have the same sensory experience that I in fact have, about a table which was made of ice.25

24 25

Naming and Necessity, pp. 140–41. Here and throughout, italic emphasis is Kripke’s. Ibid., pp. 141–42.

NATURAL KIND TERMS

447

In discussing what Kripke says, it is helpful to imagine the following scenario: A table has been brought into my office; I have examined it and determined that it is made out of wood, and not ice. I point to the table and truly say that I know that it is not made out of ice. I know that it is not made out of ice because I have used empirical observation and investigation to rule out what otherwise would have been an epistemologically relevant possibility. Prior to that investigation, it could have turned out, for all I knew, that the table was made of ice. What does Kripke tell us about this? He tells us that the intuition that it could have turned out that this table was made of ice is nothing more than the judgment that it is genuinely possible for me, or some other agent, to be in a situation qualitatively identical to the one I am in now, and be pointing at a table that is made out of ice. He generalizes this point in the next paragraph. The general answer to the objector can be stated, then, as follows: Any necessary truth, whether a priori or a posteriori, could not have turned out otherwise. In the case of some necessary a posteriori truths, however, we can say that under appropriate qualitatively identical evidential situations, an appropriate corresponding qualitative statement might have been false. The loose and inaccurate statement that gold might have turned out to be a compound should be replaced (roughly) by the statement that it is logically possible that there should have been a compound with all the properties originally known to hold of gold. The inaccurate statement that Hesperus might have turned out not to be Phosphorus should be replaced by the true contingency mentioned earlier in these lectures: two distinct bodies might have occupied, in the morning and the evening, respectively, the very positions actually occupied by Hesperus-Phosphorus-Venus.26 This paragraph, together with the one just preceding it, marks the beginning of what, in my opinion, is potentially the most misleading, and even disastrous, passage in Naming and Necessity. Two main issues are addressed: the necessity of certain propositions, and the fact that they are knowable only aposteriori. Regarding the former, Kripke makes the following points: (i) that there is a natural and correct way of understanding the locution It could have turned out that ~S in which it entails It is not necessary that S, 26

Ibid., 142–43, my boldface emphasis.

448

KRIPKE

(ii) that when the locution is understood in this way, his previous remarks—to the effect that empirical evidence is needed to rule out possible circumstances responsible for the truth of It could have turned out that ~S in cases in which S is an example of the necessary aposteriori—were strictly speaking, inaccurate, but (iii) that even when it could have turned out that is understood as in (i), and S is a genuine example of the necessary aposteriori, there often are different, descriptive propositions which are both contingent and knowable only aposteriori, and which are also easily confused with the proposition expressed by S; these are the propositions the negations of which could genuinely have turned out to be true. These points provide the basis for Kripke’s answer to the objector. He maintains that when the objector protests that his examples of the necessary aposteriori can’t be necessary if they are really aposteriori, he is confusing the propositions expressed by the examples with other, related propositions that, although knowable only aposteriori, are really contingent. In the case of my claim that this table is made out of ice, the objector is confusing it with the proposition that the table in front of me that appears so and so is made out of ice, or some closely related proposition. In the case of Kripke’s claim that the statement that Hesperus is Phosphorus is both necessary and aposteriori, the objector is confusing it with the statement that the heavenly body that appears in the evening sky (at certain places and times) is the heavenly body that appears in the morning sky (at other places and times), and so on for the other examples. Although this response is unobjectionable, as far as it goes, it does not go far enough. Two difficulties stand out. First, there is a natural way of understanding the worry raised by the objector in which it is general, and not dependent on any particular example about which one might be confused. If, as seems plausible, the function of the empirical evidence needed to justify aposteriori knowledge is to rule out possible circumstances in which the proposition known is false, then it would seem to follow that there simply can’t be any necessary truth for which empirical evidence is necessary in order for one to know it.27 27 A general objection of this sort can easily be extracted from the definition of knowledge proposed by David Lewis in “Elusive Knowledge,” Australasian Journal of Philosophy 74 (1996), reprinted in his Papers in Metaphysics and Epistemology (Cambridge: Cambridge University Press, 1999). In addition, that paper contains another basis that Lewis and some other philosophers have used to object to the idea that there are necessary truths that are knowable

NATURAL KIND TERMS

449

Since this problem naturally springs to mind when considering the position of the objector, one would expect Kripke to address it and provide a solution. That he does not do so here is striking. In focusing so much on the necessity of his examples, he appears to neglect the question of how these necessary truths are known, and in particular, the role of empirical evidence in coming to know them. Second, to the extent that he does suggest an answer to this question, it is puzzling and inadequate. In considering the Hesperus/Phosphorus example, he directs us to his earlier discussion at the end of lecture 2. However, as we saw in chapter 15, the explanation he gives there of the aposteriori character of this example is seriously flawed. As I argued earlier, it appears that his argument that the proposition that Hesperus is Phosphorus is knowable only aposteriori implicitly relies on some version of what I called principles of strong disquotation and justification. Details aside, the idea expressed by these principles is that if I am a competent speaker who understands S, then I satisfy the formula x is justified in believing that S on the basis of evidence e iff I accept S, believe it to be true, and am justified in so doing on the basis of e. In the Hesperus/Phosphorus case, if I understand the sentence Hesperus is Phosphorus, while associating the two names with the descriptions the heavenly body seen in the evening sky (at certain places and times), and the heavenly body seen in the morning sky (at other places and times), then I will justifiably accept and believe the sentence Hesperus is Phosphorus to be true only if I justifiably believe that the heavenly body seen in the evening sky (at certain places and times) is the heavenly body seen in the morning sky (at other places and times). Since my justification for this descriptive belief can only be empirical, my justification for accepting the sentence Hesperus is Phosphorus must also be empirical. Strong disquotation and justification will then tell us that my belief that Hesperus is Phosphorus is one that is justified empirically, and hence that my knowledge of this proposition is aposteriori, rather than apriori. Apparently believing that that this result generalizes to other agents, times, and sentences expressing the proposition that Hesperus is Phosphorus, Kripke arrives at the conclusion that this proposition can be known only aposteriori. only aposteriori—namely, the identification of propositions with sets of metaphysically possible worlds-states. For arguments against any such identification, see my “Direct Reference, Propositional Attitudes, and Semantic Content,” originally published in Philosophical Topics 15 (1987):47–87; reprinted in N. Salmon and S. Soames, eds., Propositions and Attitudes (Oxford: Oxford University Press, 1988).

450

KRIPKE

As I indicated in chapter 15, the problem with this argument is that, depending on the specific versions in question, principles of strong disquotation and justification are false, or at least highly questionable. Thus, Kripke’s explanation of the alleged aposterioricity of the necessary truth that Hesperus is Phosphorus cannot be accepted. That was bad enough at the end of lecture 2, when one putative example of the necessary aposteriori was in question. Here in lecture 3, when he is responding to a general objection to all examples of the necessary aposteriori, it is disastrous. Here, the suggestion implicit in his discussion is that the earlier explanation of the alleged aposterioricity of our knowledge that Hesperus is Phosphorus is to be generalized to all cases of the necessary aposteriori. To see how this might go, consider the example of the table that has been brought into my office. In pointing to the table and saying This table is not made out of ice, I am expressing a proposition that is a necessary truth—since the particular table is one that could not have existed if it were made out of ice. Nevertheless, in the circumstances in which I find myself, I would not accept, and would not be justified in accepting, the sentence This table is not made out of ice (where this refers to the table in question) unless I also believed, and was justified in believing, the general, descriptive proposition that the (or a) table that has just been brought into my office, which appears so and so, is not made out of ice. This proposition is, of course, contingent rather than necessary, and hence not to be confused with the proposition expressed by the indexical sentence I uttered. However, it is also one that I am justified in believing only on the basis of empirical evidence. Since this evidence is included in the evidence on which I base my utterance, my evidence for accepting the sentence uttered, and believing it to be true, must also be empirical. Appealing to principles of strong disquotation and justification, one gets the result that although it is a necessary truth that this table is not made out of ice, and although I know that this table is not made out of ice, my knowledge is based on empirical evidence, and hence is aposteriori.28 Generalizing to other agents, times, and alternative ways of expressing the same proposition, one might naturally conclude that the proposition that this table is

28 For this one needs context-relativized principles of strong disquotation, and strong disquotation and justification. Kripke, of course, does not explicitly formulate any such principles. For an indication of one way to formulate a context-relativized version of strong disquotation, and a discussion of problems with such principles, see pp. 12–13 of Beyond Rigidity.

NATURAL KIND TERMS

451

made out of ice is one that is both necessary and knowable only aposteriori. On this view, there may be no single contingent, aposteriori, and descriptive proposition q which is such that for every agent a, in order for a to know that this table is not made out of ice, a must know q; nevertheless, in order for any agent a to know that this table is not made out of ice, a must know some contingent, aposteriori, and descriptive proposition q that a associates with this proposition. This, it might be thought, is sufficient to guarantee that knowledge that this table is not made out of ice can only be aposteriori. One piece of evidence that Kripke intended to convey some such view as this is that he seemed to see his response to the objector as generalizing his earlier treatment of the Hesperus/Phosphorus example to all of his other examples of the necessary aposteriori. This is suggested in the paragraph immediately following the one last cited. I have not given any general paradigm for the appropriate corresponding qualitative contingent statement. Since we are concerned with how things might have turned out otherwise, our general paradigm is to redescribe both our prior evidence and the statement qualitatively and claim that they are only contingently related. In the case of identities, using two rigid designators, such as the Hesperus-Phosphorus case above, there is a simpler paradigm which is often usable to at least approximately the same effect. Let ‘R1’ and ‘R2’ be the two rigid designators that flank the identity sign. Then ‘R1 ⫽ R2’ is necessary if true. The references of ‘R1’ and ‘R2’, respectively, may well be fixed by nonrigid designators ‘D1’ and ‘D2’, in the Hesperus and Phosphorus cases these have the form ‘the heavenly body in such-and-such position in the sky in the evening (morning)’. Then although ‘R1 ⫽ R2’ is necessary, ‘D1 ⫽ D2’ may well be contingent, and this is often what leads to the erroneous view that ‘R1 ⫽ R2’ might have turned out otherwise.29 In the case of the Hesperus/Phosphorus example, Kripke suggests that the related contingent proposition with which the objector confuses the necessary truth that Hesperus is Phosphorus is also a proposition that the agent must know in order to be counted as knowing that Hesperus is Phosphorus. Here, he seems to suggest that the same point can be made for all of his other examples of the necessary aposteriori as well. This, I believe, is Kripke’s final answer to the objector in lecture 3. 29

Naming and Necessity, pp. 143–44.

452

KRIPKE

Unfortunately, it is highly problematic. Although a number of lesser worries might be raised, the most serious problem is that Kripke’s answer extends the inadequacies found in chapter 15 with his account of the Hesperus/Phosphorus example to all of his examples of the necessary aposteriori—thereby threatening to undermine one of the most important general philosophical theses in Naming and Necessity. Fortunately, this result can be avoided. Although the answer that Kripke actually gives to the objector is problematic, there is another, much more plausible answer he could have given that grows naturally out of his general philosophical position. Recall the objector’s point. If p is knowable only aposteriori, then empirical evidence is needed to rule out certain possible circumstances in which p is false. However, if p is necessary, then there are no such circumstances to rule out, since p is true with respect to every possible state that the world could be in. Thus, no proposition can be both necessary and knowable only aposteriori. Faced with this argument, there are three main responses that could be offered by a defender of the necessary aposteriori: (i) he could reject the idea that when empirical evidence is required for knowledge, its function is to rule out possibilities; (ii) he could reject the idea that all conceptual, or epistemic, possibilities are genuine, metaphysical possibilities—i.e., he could reject the claim that every way that, for all we know, the world might be is a way that the world genuinely could be; or, (iii) he could maintain that examples of the necessary aposteriori are those in which even though a sentence S expresses a necessary truth p, what we standardly call knowing that S always requires knowing some contingent, aposteriori proposition q that is related to p in a certain way. As I have argued, Kripke seems to have chosen (iii).30 However, given his other views, his most natural response should have been to choose (ii)—to reject the idea that all epistemic possibilities are genuine metaphysical possibilities. He is committed to this anyway; and since it is all he needs to block the objection, no other response is necessary.31 30 The fact that Kripke’s discussion can, without much strain, be read as eventuating in (iii) has lent some support to those who maintain that no single proposition can be both necessary and knowable only aposteriori. See, for example, Robert Stalnaker, “Assertion,” Syntax and Semantics 9 (1978):315–32, reprinted in his Context and Content (New York: Oxford University Press, 1999); David Lewis, “Elusive Knowledge,” Papers in Metaphysics and Epistemology (Cambridge: Cambridge University Press, 1999); and Frank Jackson, From Metaphysics to Ethics (New York: Oxford University Press, 1998). Criticism of option (iii), in its various manifestations, can be found in my Reference and Description: The Two-Dimensionalist Attempt to Revive Descriptivism (in preparation), as well as in my “Saul Kripke, the Necessary Aposteriori, and the Two-Dimensionalist Heresy.” 31 I am indebted to Ali Kazmi for very helpful discussions of this section.

NATURAL KIND TERMS

453

A Proper Response to the Challenge: Essential Properties, Impossible World-States, and the Necessary Aposteriori As I indicated in chapter 15, Kripke’s views about the necessary aposteriori are connected to his views about essential properties.32 He argues that we know apriori that various properties and relations—such as non-identity, being human, not being made out of ice, and being made out of molecules—are essential to anything that has them.33 Thus, we know apriori that if things have these properties and relations, then they have them necessarily. This means that various propositions which predicate these properties and relations of objects are such that we know apriori that if they are true, then they are necessarily true. Still, finding out that they are true requires empirical investigation. If this is right, then sometimes in order to find out whether certain things are true with respect to all possible states of the world, and other things are true with respect to no possible states of the world, we first must find out what is true with respect to the actual state of the world. Sometimes in order to find out what could or could not be, one first must find out what is. This will seem problematic only if you have restricted the ways things could conceivably be to the ways things could really be—i.e., only if you have restricted epistemic possibility to metaphysical possibility. Although the passage in lecture 3 of Naming and Necessity that we have been talking about may seem to show Kripke ignoring or backsliding on this point, it doesn’t, in my opinion, negate the central lesson of his work that one must sharply distinguish these two kinds of possibility.34 Once this is See in particular in his essay “Identity and Necessity.” There are also properties that are essential to some, but not all, things that have them. These too give rise to instances of the necessary aposteriori that can be treated in roughly the manner suggested in the text. In the interest of simplicity, I put these to one side in the text. 34 Kripke’s footnote 72, toward the end of the passage under discussion, shows that even there he was aware of the importance of the distinction between epistemic and metaphysical possibility. He says, referring to some of his remarks in the passage: “Some of the statements I myself make above may be loose and inaccurate in this sense. If I say, ‘Gold might turn out not to be an element,’ I speak correctly; ‘might’ here is epistemic and expresses the fact that the evidence does not justify a priori (Cartesian) certainty that gold is an element. I am also strictly correct when I say that the elementhood of gold was discovered a posteriori. If I say, ‘Gold might have turned out not to be an element,’ I seem to mean this metaphysically and my statement is subject to the correction noted in the text.” To understand the relation between this footnote and the text, it is important to remember that the footnotes were added to the lectures by Kripke a considerable time after they were given and a written transcript had been produced. I believe that when writing the footnote, he noticed that his discussion in the text had neglected the important and relevant distinction between epistemic and metaphysical necessity, and he wished— without changing the text—to call the reader’s attention to his commitment to this distinction. 32 33

454

KRIPKE

done, and both rigid designation and the existence of nontrivial essential properties of objects is accepted, the necessary aposteriori follows unproblematically. This point is connected to the initial idealized conception of inquiry discussed in chapter 15, which is presupposed by Kripke’s hypothetical objector in lecture 3. On that conception, ignorance is a matter of lacking information about which, of certain different possible states that the world could be in, it is actually in; and complete ignorance is a condition in which one doesn’t know which, of all the possible states that the world could be in, it is actually in. According to this conception, when an agent is in this condition, (i) all metaphysically possible states of the world are epistemically possible—i.e., every way that the world could possibly be is a way that, for all the agent knows, it actually is, and (ii) every epistemic possibility is a metaphysical possibility— i.e., every way that, for all the agent knows, the world might be is a way that the world genuinely could be. Inquiry is the process of escaping from ignorance. By investigating the world, the agent learns contingent truths that distinguish the way the world actually is from other ways it might possibly be, but isn’t. Each time the agent learns one of these truths, he narrows down the class of possibilities compatible with what he knows, and within which he locates the way the world actually is. Thus, acquiring information is equated with narrowing down the range of genuinely possible world-states that are compatible with what one knows. On this picture, a proposition q provides evidence supporting the truth of a proposition p by ruling out certain possible ways in which p might fail to be true. This has the immediate consequence that the idea of necessary truths which are knowable only aposteriori becomes problematic. It is problematic because to say that p is knowable only aposteriori is to say that one can have the justification required to know p only if one has empirical evidence supporting its truth. Since p is necessary, however, there are no possible world-states with respect to which it is untrue; hence there can be no empirical evidence supporting p. Therefore, there is no such thing as the necessary aposteriori. The objector in lecture 3 is naturally read as having something like this in mind. Hence, it is strange that Kripke didn’t challenge the presuppositions underlying that conception of inquiry. Instead, he ended up appearing to suggest that what makes it possible for a necessary truth to be knowable only aposteriori is that, in the end, knowledge of it always involves knowledge of something else—something contingent

NATURAL KIND TERMS

455

which requires empirical support. However, there was no need for this detour. Here, it is helpful to remember that for Kripke, possible states of the world are not alternate concrete universes, but abstract objects. As I pointed out in chapter 15, they can be thought of as maximally complete ways the real concrete universe could have been—maximally complete properties that the universe could have instantiated. Thinking of them in this way suggests an obvious generalization. Just as there are properties that certain objects could possibly have and other properties that they couldn’t possibly have, so there are certain maximally complete properties that the universe could have had—possible states of the world—and other maximally complete properties that the universe could not have had—impossible states of the world. Just as some of the properties that objects couldn’t have had are properties that one can conceive them as having, so some maximally complete properties that the universe could not have had (some impossible states of the world) are properties one could conceive it as having. Given this, one can explain the informativeness of certain necessary truths as resulting from the fact that learning them allows one to rule out certain impossible, but nevertheless conceivable, states of the world. Moreover, one can explain the function played by empirical evidence in providing the justification needed for knowledge of necessary aposteriori truths; empirical evidence is required to rule out certain impossible, but nevertheless conceivable and epistemologically relevant, worldstates with respect to which the necessary propositions are false.35 Thus, by expanding the range of epistemically conceivable states of the world to include some that are metaphysically impossible, one can modify the original conception of inquiry so as to accommodate Kripkean examples of the necessary aposteriori. Whether or not this is the final word on the subject is another question. As indicated at the end of chapter 15, some further modification even of this expanded conception of inquiry may well be needed. However, the central idea that not all epistemic possibilities are metaphysical possibilities seems both to be solid and to provide the key to responding to Kripke’s hypothetical objector to the necessary aposteriori. Once one dispels the air of mystery surrounding this category of truth, the motivation for Kripke’s problematic response to the objector in lecture 3 falls away, and the full force of his discovery of the necessary As in chapter 15, I here assume that names (unlike definite descriptions such as the stuff out of which this table, if it exists, is constituted) rigidly designate the same thing with respect to all world-states, possible or not.

35

456

KRIPKE

aposteriori can be appreciated. One of the great philosophical achievements of the twentieth century, it has transformed the philosophical landscape, recalibrated our sense of what is possible, and reshaped our understanding of our own philosophical past. No single insight has been more important than this in gaining the perspective needed to understand and critically evaluate the philosophical tradition stretching from Moore, Russell, and Wittgenstein, through logical positivism and the ordinary language school, to Quine, Davidson, and Kripke himself. Without Kripke’s discovery, the history told in these pages would have been very different; indeed, these volumes would scarcely have been possible.

SUGGESTED FURTHER READING FOR PART SEVEN

Main Primary Sources Discussed Kripke, Saul. Naming and Necessity. Cambridge, MA: Harvard, 1980; originally published in D. Davidson and G. Harman, eds., Semantics of Natural Languages (Dordrecht: Reidel 1972), pp. 253–355.

Additional Primary Sources Kripke, Saul. “Identity and Necessity.” In Milton Munitz, ed., Identity and Individuation (New York: NYU Press, 1971). ———. “A Puzzle about Belief.” In A. Margalit, ed., Meaning and Use, (Dordrecht: Reidel, 1979); reprinted in N. Salmon and S. Soames, eds., Propositions and Attitudes (Oxford: Oxford University Press, 1988).

Additional Recommended Reading Burgess, John. “How Not to Write History of Philosophy.” In The New Theory of Reference (Dordrecht, Boston, and London: Kluwer, 1998). ———. “Marcus, Kripke, and Names.” In Paul W. Humphreys and James H. Fetzer, eds., The New Theory of Reference. ———. “Quinus ab omni naevo vindicatus.” In Ali A. Kazmi, ed., Meaning and Reference (Calgary: University of Calgary Press, 1998). Chalmers, David. “On Sense and Intension.” Philosophical Perspectives 16, Language and Mind (Oxford: Blackwell, 2002), 135–82. Evans, Gareth. “The Causal Theory of Names.” Proceedings of the Aristotelian Society, supplementary vol. 47, 1973, 187–208; reprinted in his Collected Papers (Oxford: Oxford University Press, 1985). ———. “Reference and Contingency.” In his Collected Papers (Oxford: Oxford University Press, 1985). Harman, Gil. “How to Use Propositions.” American Philosophical Quarterly 14 (1977):173–176.

458

SUGGESTED FURTHER READING

Jackson, Frank. From Metaphysics to Ethics. Oxford: Oxford University Press, 1998. Jeshion, Robin. “Acquaintanceless De Re Belief.” In M. O’Rourke et al., eds., Truth and Meaning: Investigations in Philosophical Semantics (New York: Seven Bridges Press, 2001). Kaplan, David. “Demonstratives.” In Joseph Almog, John Perry, and Howard Wettstein, eds., Themes from Kaplan (New York and Oxford: Oxford University Press, 1989). ———. “On The Logic of Demonstratives.” Journal of Philosophical Logic 8 (1979):81–98. ———. “Opacity.” In E. Hahn and P. A. Schilpp, eds., The Philosophy of W. V. Quine (La Salle, IL: Open Court, 1986). Koslicki, Kathrin. “The Semantics of Mass Predicates.” Noûs 38 (1999): 46–91. Larson, Richard, and Peter Ludlow. “Interpreted Logical Forms.” Synthese 95 (1993):305–56; reprinted in Peter Ludlow, ed., Readings in the Philosophy of Language (Cambridge, MA: MIT Press, 1997). Lewis, David. “Elusive Knowledge.” In Lewis, Papers in Metaphysics and Epistemology (Cambridge: Cambridge University Press, 1999); originally published in The Australasian Journal of Philosophy 74 (1996). ———. “Naming the Colours.” In his Papers in Metaphysics and Epistemology, (Cambridge: Cambridge University Press, 1999); originally published in The Australasian Journal of Philosophy 75 (1997). ———. On the Plurality of Worlds. Oxford: Blackwell, 1986. McKeown-Green, Jonathan. The Primacy of Public Language. Unpublished Princeton University Ph.D. dissertation, 2002. Neale, Stephen. “On a Milestone of Empiricism.” In Alex Orenstein and Petr Kotatko, eds., Knowledge, Language and Logic (Dordrecht, Boston, and London: Kluwer, 2000). Putnam, Hilary. “Explanation and Reference.” In G. Pearce and P. Maynard, eds., Conceptual Change (Dordrecht: Reidel, 1973); reprinted in Hilary Putnam, Philosophical Papers, vols. 1 and 2 (Cambridge: Cambridge University Press, 1975). ———. “Is Semantics Possible?” In H. Kiefer and M. Munitz, eds., Language, Belief and Metaphysics (Albany: State University of New York Press, 1970); reprinted in Putnam, Philosophical Papers. ———. “Meaning and Reference.” Journal of Philosophy 70 (1973): 699–711; reprinted in Putnam’s Philosophical Papers. ———. “The Meaning of ‘Meaning’.” In K. Gunderson, ed., Lan-

SUGGESTED FURTHER READING

459

guage, Mind, and Knowledge, Minnesota Studies in the Philosophy of Science, no. 7 (Minneapolis: University of Minnesota Press, 1975); reprinted in Putnam, Philosophical Papers. Quine, W. V. “Notes on Existence and Necessity.” Journal of Philosophy 40 (1943):113–27. ———. “The Problem of Interpreting Modal Logic.” Journal of Symbolic Logic 12 (1947):43–48. ———. “Reference and Modality.” In From a Logical Point of View (Cambridge, MA: Harvard University Press, 1953, 1961, 1980). ———. “Three Grades of Modal Involvement.” In Quine, The Ways of Paradox (New York: Random House, 1966); originally published in 1953. Richard, Mark. Propositional Attitudes. Cambridge: Cambridge University Press, 1990. Rieber, Stephen. “Understanding Synonyms without Knowing That They Are Synonymous.” Analysis 52 (1992):224–28. Salmon, Nathan. Frege’s Puzzle. Cambridge, MA: MIT Press, 1986. ———. “How to Measure the Standard Metre.” Proceedings of the Aristotelian Society, 88 (1987–88):193–217. ———. “A Millian Heir Rejects the Wages of Sinn.” In C. A. Anderson and J. Owens, eds., Propositional Attitudes: The Role of Content in Logic, Language, and Mind (Stanford, CA: CSLI, 1990), 215–47. ———. “On the Logic of What Might Have Been.” Philosophical Review 98 (1989):3–34. ———. Reference and Essence. Princeton, NJ: Princeton University Press, 1981. Soames, Scott. Beyond Rigidity. New York: Oxford University Press, 2002. ———. “Direct Reference, Propositional Attitudes, and Semantic Content,” originally published in Philosophical Topics, 15, 1987, 47–87; reprinted in N. Salmon and S. Soames, eds., Propositions and Attitudes (Oxford University Press), 1988. ———. “Donnellan’s Referential/Attributive Distinction.” Philosophical Studies 73 (1994):149–68. ———. “More Revisionism about Reference.” In The New Theory of Reference (Dordrecht, Boston, and London: Kluwer, 1998). ———. “Naming and Asserting.” In A. Szabo, ed., Semantics and Pragmatics, (Oxford University Press, forthcoming). ———. Reference and Description: The Two-Dimensionalist Attempt to Revive Descriptivism. In preparation.

460

SUGGESTED FURTHER READING

———. “Review of Gareth Evans, Collected Papers.” Journal of Philosophy 86 (1989). ———. “Revisionism about Reference.” Synthese 104 (1995):191–216; reprinted in The New Theory of Reference (Dordrecht, Boston, and London: Kluwer, 1998). ———. “Saul Kripke, the Necessary Aposteriori, and the Two-Dimensionalist Heresy.” In M. Garcia-Carpintero and J. Maciá, eds., The Two-Dimensional Framework: Foundations and Applications (Oxford: Oxford University Press, forthcoming). Stalnaker Robert. “Assertion.” Syntax and Semantics 9 (1978):315–32; reprinted in his Context and Content (New York: Oxford University Press, 1999). ———. Inquiry. Cambridge, MA: MIT Press, 1984. ———. “Possible Worlds.” Noûs 10 (1976):65–75; reprinted in Michael Loux, ed., The Actual and the Possible (Ithaca, NY: Cornell University Press, 1979).

EPILOGUE THE ERA OF SPECIALIZATION

Specialization and Fragmentation In these volumes we have looked at many of the most important developments in analytic philosophy during the period from the turn of the twentieth century through the early 1970s. As I indicated in the “Introduction to the Two Volumes,” it was necessary for me to be highly selective in what to present, with the result that a great deal of valuable work had to be left out. One main criterion for selecting philosophers, schools of philosophy, and philosophical issues to be treated in depth was the scope of their impact, not just on specialists, but on philosophy as a whole. I have tried to focus on work that was, and was widely recognized to be, significant for the broad spectrum of philosophy in the analytic tradition. Everyone working in this tradition knows a bit about, and appreciates something of the significance of, Moore, Russell, Wittgenstein, the logical positivists, the ordinary language philosophers, Grice, Quine, Davidson, and Kripke. Although many other outstanding philosophers in the period did work of lasting importance—Alfred Tarski, Rudolf Carnap, Frank Ramsey, Carl Hempel, Nelson Goodman, Hilary Putnam, and David Lewis, to name just a few—their influence on analytic philosophers across the board was not as great as those on which we have focused. There are, however, two exceptions to this— two philosophers not discussed here who, to my mind, fit naturally into the group we have focused on. They are Gottlob Frege and John Rawls. Frege, whom I mentioned in the “Introduction to the Two Volumes,” was perhaps the most important philosophical pioneer in the development of modern symbolic logic, the philosophy of logic, the philosophy of mathematics, and philosophical semantics. Although his work is undeniably technical, and although most of it falls outside our official period—being done, for the most part, in the last twenty-five years of the nineteenth century—its influence on philosophy in the analytic tradition grew throughout the period, until it reached the proportions of

462

EPILOGUE

the other major figures we have concentrated on. Consequently, his omission amounts to an undeniable gap in the story I have told. However, as I spelled out more fully in the earlier introduction, it is one I hope, in the future, to be able to fill with a study of a historically integrated line of research starting with Frege’s formalization of logic and his philosophical conception of semantics, continuing through Tarski’s work on truth and logical consequence in formalized languages, moving on to the halting and painstaking development of modal logic and modal semantics by Carnap, C. I. Lewis, Marcus, Kripke, and others, and culminating in important applications and extensions of these ideas by such philosophers such Montague, Kaplan, Stalnaker, and David Lewis. Not only fascinating in itself, this line of work has important connections to other, broader, less technical issues in philosophy, and ultimately to the singular achievement of separating and understanding the different modalities: logical truth, necessary truth, apriori truth, and analytic truth. John Rawls is a different story. His major work, A Theory of Justice, was published near the end of our period, in 1971, to be followed in later years by refinements and extensions.1 Although it could be argued that political philosophy does not have the centrality for analytic philosophy as a whole as does metaphysics and the philosophy of language, the impact of A Theory of Justice was extraordinary. Among other things, it provoked an extraordinary response from Rawls’s Harvard colleague, Robert Nozick, whose great work in political philosophy, Anarchy, State, and Utopia, was published in 1974.2 Together, these books redefined and reinvigorated political philosophy in the English-speaking world, and, in the judgment of many, constituted the high point of normative philosophy in the analytic tradition throughout the entire twentieth century. Their omission from our discussion might reasonably be regarded as the most glaring gap in our story. The reasons for this gap are purely autobiographical. Although I used to teach this material in one of the courses out of which these volumes grew, eventually the syllabus became so overcrowded that they had to be dropped. Since adding a discussion of them now would take considerable time, and result in substantial lengthening of a work that is already quite long—to say nothing of the fact that, undoubtedly, many others could cover this material better than I could—I have decided to leave the gap unfilled. 1 2

John Rawls, A Theory of Justice (Cambridge, MA: Harvard University Press, 1971). Robert Nozick, Anarchy, State, and Utopia (New York: Basic Books, 1974).

THE ERA OF SPECIALIZATION

463

Apart from this, there are, I believe, no other comparable gaps in the story of twentieth-century philosophical analysis that I have been able to tell. I say this despite the fact that I have had very little to say about the work done in the last twenty-five years of the century. My judgment about the effect of the absence of this material on my story does not reflect the view that we are still too close to that period to be able to reasonably assess it, though there is some truth in that observation. Nor does it reflect the view that these years have been barren of philosophical achievement. On the contrary, there has been so much philosophical achievement, of so many different sorts, that I doubt that it can be encompassed in the kind of history I have tried to write. In my opinion, philosophy has changed substantially in the last thirty or so years. Gone are the days of large, central figures, whose work is accessible and relevant to, as well as read by, nearly all analytic philosophers. Philosophy has become a highly organized discipline, done by specialists primarily for other specialists. The number of philosophers has exploded, the volume of publication has swelled, and the subfields of serious philosophical investigation have multiplied. Not only is the broad field of philosophy today far too vast to be embraced by one mind, something similar is true even of many highly specialized subfields. For example, philosophy of science as practiced today includes not just the traditional topics of induction, confirmation, prediction, observation, empirical adequacy, and the nature of scientific explanation, but also highly specific questions in the philosophy of physics, of quantum mechanics, of biology, of psychology, and even of linguistics. Since experts in these areas need to be proficient in the special sciences they deal with, as well as in related subfields of philosophy, there is no way for anyone to have a solid grasp on the philosophy of science as a whole. The same is true of other subfields, the history of philosophy no less than others. There is a myth in some quarters that the history of philosophy is neglected in the analytically dominated philosophy departments of much of the English-speaking world. The facts are quite otherwise. There may, of course, be disagreements about the proper balance of different fields within any discipline. However, the volume of work done today in the history of philosophy, as well as the depth, thoroughness, and penetration of the studies of particular historical figures, is unprecedented—as are the number of different periods, and different figures within those periods, receiving serious attention from more or less analytically trained philosophers. As with other subfields

464

EPILOGUE

of philosophy, this wealth of new work is not an unmixed blessing. Since many of the easy advances have already been made, it is understandable that further progress requires greater specialization, and more intense focus. The results of this are not just advances in historical knowledge, but also the creation of a certain distance between experts in one area of the history of philosophy and almost everyone else. It is, I think, fair to say that much of the advanced work of today’s historians of philosophy is less accessible to the general run of philosophers, and even to some historians of other eras, than ever before. So what, in light of this, would a history of the last thirty-plus years of the twentieth-century look like? It would look not like one linear and integrated story, but like many distinct and overlapping stories— of the philosophy of language and the philosophy of logic during the period, of the philosophy of mind and the philosophy of language, of moral psychology and the philosophy of mind, of cognitive science and the philosophy of psychology, of epistemology, of metaphysics, of the semantics and metaphysics of normative discourse, of ethics and political philosophy, and more. It is, I think, a mistake to look for one big, unified picture of analytic philosophy in this era. What we need is a collection of more focused pictures, each giving a view of the major developments of related lines of work, and each drawn with an eye to illuminating the larger lessons for work in neighboring subfields.

An Illustrative, Somewhat Technical, and Specialized Example: The Philosophy of Logic and Language I will illustrate what I have in mind with a very brief and programmatic overview of one specialized subfield: the philosophy of logic and language. From roughly the mid-1960s to the end of the century, several highly specialized and productive lines of research flourished, yielding a number of real advances. These include (i) work on truth and the Liar paradox, and (ii) work on vagueness and the Sorites paradox, (iii) the development of intensional logics and their application to natural language, and (iv) work on reference, propositional attitudes, propositional attitude ascriptions, and semantic content. Although these topics don’t come close to exhausting the categories of significant work in the philosophy of logic and language at the end of the twentieth century, they are useful in illustrating the richness and complexity of recent developments. I will say a brief word about each.

THE ERA OF SPECIALIZATION

465

First, the Liar paradox, which is illustrated by sentence L. L. The sentence labeled ‘L’ in the epilogue of volume 2 of Philosophical Analysis in the Twentieth Century is not true. L is a perfectly meaningful sentence of English, as is illustrated by the fact that someone who hadn’t read the epilogue would have no difficulty comprehending what it says. Moreover, what it says (the proposition it actually expresses) would clearly have been true if the only sentence labeled ‘L’ in this section of the book had been the sentence Modern symbolic logic is a branch of psychology. But if L says something that would have been true had certain circumstances obtained, then it says something (expresses some proposition), and so is meaningful. What makes the sentence paradoxical is that a contradiction can be derived from apparently incontrovertible assumptions about it, the truth predicate it contains, and other expressive resources of English. This fact gives rise to a set of related questions: What does the paradox show about English? Does it show that the rules of the language are inconsistent? If it does show this, what is the significance of this fact? If it doesn’t show that the rules of English are inconsistent, does it show that the assumptions about English used in the paradox to derive the contradiction are not jointly correct? Supposing that it does show this, which assumption (or assumptions) should be rejected, and what should replace it (them)? Putting English aside, can we define a notion of truth, sufficient for the study of mathematical, scientific, and other theories, which is paradoxfree, and does not lead to contradiction? If so, what is the best way of doing this? 3 Modern work on the Liar paradox attempts to answer these (and related) questions. For all intents and purposes, this work began in the ’30s with Kurt Gödel and Alfred Tarski, who ingeniously devised ways to put the paradox to productive use in proving their famous meta-mathematical theorems of the incompleteness of arithmetic and the indefinability (within arithmetic) of arithmetical truth.4 The principles involved in the proofs of these results lead to the conclusion that languages with a certain minimal richness satisfying certain minimal conditions cannot contain their own truth predicates. Although this conclusion seems perfectly justifiable in the case of arithmetic, it is highly questionable that it generalizes to natural languages like English. To think that it 3 For an introduction to the discussion of some of these problems, see the final section of chapter 2 my Understanding Truth (New York: Oxford University Press, 1999), pp. 49–56. 4 See Understanding Truth, chapter 3, pp. 82–86 for an informal explanation of the use of the Liar in obtaining these results.

466

EPILOGUE

does is, in effect, to think that English is not really a single language, but an infinite hierarchy of languages, each with a truth predicate for lower languages in the hierarchy, but none for itself. Not only does this picture strike many as far-fetched, it can be shown to lead to serious difficulties of more specialized sorts.5 These difficulties constitute the starting point for the explosion of philosophical and logical work on the Liar done in the last thirty years. The variety of different approaches is staggering, and their complexity and technical sophistication is daunting to nonexperts. Since the topic has become both vast and extremely specialized, I will not attempt a description here, but will instead simply list some of the most important work.6 It includes Charles Parsons’s 1974 “The Liar Paradox,”7 Saul Kripke’s 1975 “Outline of a Theory of Truth,”8 Tyler Burge’s 1979 “Semantical Paradox,”9 Han Herzberger’s 1982 “Notes on Naive Semantics,”10 Anil Gupta’s 1982 “Truth and Paradox,”11 Jon Barwise and John Etchemendy’s 1987 The Liar,12 Graham Priest’s 1987 In Contradiction,13 Van McGee’s 1990 Truth, Vagueness, and Paradox,14 Haim Gaifman’s 1992 “Pointers to Truth,”15 Noel Belnap and Anil Gupta’s 1993 The Revision Theory of Truth,16 Keith Simmons’s 1993 Universality and the Liar,17 and Hartry Field’s forthcoming “A Revenge-Immune Solution to the Semantic Paradoxes.”18 When one looks at this work, two things emerge—first, how far we have advanced since the time of Tarski in our understanding of this collection of philosophical issues crucially involving the notion of truth, and second, how great the specialization and level of technical detail had become in order to get us this far. When a subfield (truth and the Liar paradox) of a subfield (the philosophy of See chapter 5 of Understanding Truth. An extensive description of the philosophical underpinnings and elementary technical details of one of the leading approaches—Kripke’s—can be found in chapter 6 of Understanding Truth. 7 Journal of Philosophical Logic 3 (1974):381–412; reprinted in Robert Martin, ed., Truth and the Liar Paradox (Oxford and New York: Clarendon and Oxford University Press, 1984). 8 Journal of Philosophy 72 (1975):690–716; reprinted in Martin. 9 Journal of Philosophy 76 (1979):169–98; reprinted in Martin. 10 Journal of Philosophical Logic 11 (1982):61–102; reprinted in Martin. 11 Journal of Philosophical Logic 11 (1982):1–60; reprinted in Martin. 12 Oxford and New York: Clarendon Press, 1987. 13 Dordrecht: Martinus Nijhoff, 1987. 14 Indianapolis: Hackett, 1990. 15 Journal of Philosophy 89 (1992):223–61. 16 Cambridge, MA: MIT Press, 1993. 17 Cambridge: Cambridge University Press, 1993. 18 Journal of Philosophical Logic, forthcoming. 5 6

THE ERA OF SPECIALIZATION

467

logic) becomes this complex and specialized, it can only mean that the discipline itself—philosophy as a whole—has become an aggregate of related but semi-independent investigations, very much like other academic disciplines. This is the environment that has emerged at the end of the twentieth century. In it, there can be no single, unified story of the major developments in philosophy as a whole—even to the limited extent that we were able to approximate in our study of the first threequarters of the century. A second thread in the development in the philosophy of logic and language in the last part of the twentieth century is the investigation of vague predicates, and in particular, those vague predicates that figure in standard versions of the Sorites paradox. This paradox, also known as the paradox of the heap, is sometimes studied in connection with the Liar paradox and sometimes not—depending on whether or not the theorist believes their solutions share important common features. In its classical form, the Sorites starts from the premise that a single grain of sand is not itself a heap of sand. It is next asserted that if one has something which is not a heap of sand, and one adds a single grain of sand to it, the result is still not a heap of sand. In defense of this assertion, it is frequently pointed out that its negation is logically equivalent to the claim that there is some precise number n such that no collections containing less than n grains of sand count as heaps, whereas all collections containing n grains or greater are heaps. But, it is typically observed, it is extremely implausible that there is an absolutely precise line dividing the heaps from the nonheaps. After all, one wants to protest, the notion of a heap is vague, and because it is vague, no such precise dividing line can exist. In this way, one is led to accept the second premise of the paradox—namely, that if n grains of sand are not sufficient to make a heap of sand, then n ⫹ 1 grains aren’t either. But now the trouble is obvious. Though each of the two premises appears plausible when considered by itself, together they seem to entail the absurd result that there are no heaps of sand—i.e., no matter how many grains of sand may be gathered together, they are not sufficient to make a heap of sand. Surely that can’t be right. But if it isn’t, then something in our reasoning has gone wrong. The problem posed by the Sorites paradox is to identify the misstep, to explain how, despite its initial plausibility, that step could be wrong, and to replace it with something equally plausible that will not lead to paradox. As in the case of the Liar, work on this ancient paradox has exploded in the last thirty years. Also as in the earlier case, the variety and complexity of different proposed solutions are daunting. At one end of the

468

EPILOGUE

spectrum are inconsistency approaches, the basis for which may be seen as laid in Crispin Wright’s classic 1976 paper, “Language-Mastery and the Sorites Paradox.”19 Roughly put, these approaches maintain that the rules governing ordinary vague predicates simply do not allow for sharp and precise lines dividing objects to which the predicates apply from objects of any other sort, and for that reason these rules are inconsistent with even the seemingly innocuous judgment that vague predicates clearly do apply to some objects and clearly do not apply to others. Having gotten this far, the inconsistency theorist’s next thought is likely to be that vague predicates are defective, and not genuinely applicable at all—a position illustrated by Peter Unger’s 1979 article, “There are no ordinary things.”20 At the other end of the spectrum is a position known as epistemicism, masterfully defended by Timothy Williamson in his 1994 book, Vagueness.21 On his view, vague predicates are in fact perfectly precise—in the sense that there are sharp and precise lines dividing objects to which they truly apply from objects to which they truly do not—but that it is impossible for us ever to know where these lines are. In between these two positions are a variety of less extreme views which seek to maintain most of our pre-theoretic intuitions about vagueness, while avoiding paradox by positing special features of the meanings of vague predicates that explain how they differ from other, nonvague, predicates. These views include (a) degree-of-truth-andapplication views, according to which application of a vague predicate to an object is not an all-or-nothing affair, but comes in degrees, as does the truth of claims made using such a predicate;22 (b) partial-definition views, according to which vague predicates are defined so as to clearly apply to some things, and clearly not apply to others, while being undefined for objects in between;23 and (c) contextualist views, according to which vague predicates do allow for sharp boundaries, but boundaries In G. Evans and J. McDowell, eds., Truth and Meaning (Oxford: Clarendon Press, 1976). Synthese 41 (1979): 117–54. 21 London: Routledge, 1994. 22 See J. A. Goguen, “The Logic of Inexact Concepts,” Synthese 19 (1969): 325–73; K. Machina, “Truth, Belief, and Vagueness,” Journal of Philosophical Logic 5 (1976): 47–78; and (for an interesting application) G. Forbes, “Thisness and Vagueness,” Synthese 54 (1983): 235–59. Also see L. A. Zadeh, “Fuzzy Sets,” Information and Control 8 (1965): 338–53, and “Fuzzy Logic and Approximate Reasoning,” Synthese 30 (1975): 407–28. 23 Two important classic papers here are K. Fine, “Vagueness, Truth, and Logic,” Synthese 30 (1975): 265–300, reprinted in R. Keefe and P. Smith, eds., Vagueness: A Reader (Cambridge, MA: MIT Press, 1997), and H. Kamp, “The Paradox of the Heap,” in U. Monnich, ed., Aspects of Philosophical Logic (Dordrecht: Reidel, 1981). Other important, more recent works are J. Tappenden, “The Liar and Sorites Paradoxes,” Journal of Philosophy 90 (1993): 551–77, and R. Keefe, Theories of Vagueness (Cambridge: Cambridge University Press, 2000). 19 20

THE ERA OF SPECIALIZATION

469

that are constantly moving due to conversational pressures. These last, contextualist approaches come in different varieties depending on how they are combined with other views. For example, Delia Graff combines contextualism with epistemicism, whereas I combine it with views that treat vague predicates as partially defined.24 As in the case of the study of truth and the Liar, so in the case of the study of vagueness and the Sorites, the unprecedented volume of new philosophical work has led to undoubted advances in our understanding. Also as before, the price of these advances has been a high degree of complexity and specialization. Just as mastering the expanding literature on truth and the Liar can be a full-time job, so can mastering the growing literature on vagueness and the Sorites. Although the two problems (and two literatures) are related, and can usefully be studied together, they are also semi-independent—with certain approaches to one problem having little to do with various approaches to the other. The new study of vagueness also has connections with other aspects of the philosophy of logic and language. The growing interest in this topic reflects one aspect of the steady expansion of philosophical logic beyond its roots in formal logic and mathematics to the study of central questions about the meanings of expressions in natural language, and the principles employed in reasoning with them. This brings us to the third broad strand of thought in the philosophy of logic and language that has grown up at the end of the twentieth century—the development of intensional logics and their application to natural language. Central to this work is so-called possible-worlds semantics, the model theory of which was pioneered by Saul Kripke in the late ’50s and early ’60s.25 The most significant extensions of this framework and applications of it to the description of natural language grew out of Richard Montague’s innovations in the ’60s.26 Those innovations inspired 24 Delia Graff, “Shifting Sands: An Interest-Relative Theory of Vagueness,” Philosophical Topics 28 (2000): 45–81; Scott Soames, chapter 7 of Understanding Truth, and Soames, “HigherOrder Vagueness for Partially-Defined Predicates,” in J. C. Beall and M. Glanzberg, eds., Liars and Heaps: New Essays on Paradox (Oxford: Oxford University Press, forthcoming). Both of these views are influenced by Diana Raffman’s important paper, “Vagueness without Paradox,” Philosophical Review 103 (1994): 41–74. 25 Saul Kripke, “A Completeness Theorem in Modal Logic,” Journal of Symbolic Logic 24 (1959): 1–14; also, “Semantical Analysis of Modal Logic,” Zeitschrift für mathematische Logik und Grundlagen der Mathematik 9 (1963): 67–96; and “Semantical Considerations on Modal Logic,” Acta Philosophica Fennica 16 (1963): 83–94, reprinted in Leonard Linsky, ed., Reference and Modality (Oxford: Oxford University Press, 1971). 26 Richard Montague, Formal Philosophy: Selected Papers of Richard Montague (New Haven: Yale University Press, 1974).

470

EPILOGUE

a generation of philosophical logicians, philosophers of language, and theoretical linguists, whose insights are now firmly entrenched in the formal and scientific study of language in both linguistics and philosophy. One of the earliest and most significant advances occurred in the late sixties and early seventies, when Robert Stalnaker and David Lewis applied the insights of intensional logic and possible-worlds semantics to the topic of counterfactual conditionals, with profound impact on philosophers working in different areas of the subject.27 Later, the philosopher Hans Kamp and the theoretical linguist Irena Heim developed what has come to be known as Discourse Representation Theory, by extending the Montague approach so as to capture what they took to be semantic dependencies among different sentences in a discourse, and to provide Montague-style interpretations of discourses as wholes.28 Progress continued in the seventies and eighties, when David Kaplan extended the reach of intensional logic and semantics to indexical expressions like I, you, he, she, here, now, and actually, and succeeded in using work in formal semantics to illuminate broader philosophical questions about meaning, reference, assertion, and belief.29 This brings us to the topic of reference, propositional attitudes, and propositional attitude ascriptions, the fourth and final strand in our overview of contemporary philosophy of logic and language. We begin with a leading idea of possible-worlds semantics. In addition to providing a semantics for modal operators like it is a necessary truth that, it could have been the case that, and if it had been the case that ——— , then it would have been the case that ——— , which cannot be handled in standard systems of extensional logic, the possible-worlds approach gave us Robert Stalnaker, “A Theory of Conditionals,” Studies in Logical Theory, American Philosophical Quarterly, Monograph Series, no. 2 (Oxford: Blackwell, 1968), reprinted in Ernest Sosa, ed., Causation and Conditionals (Oxford: Oxford University Press, 1975) and in W. L. Harper, R. Stalnaker, and G. Pearce, eds., Ifs (Dordrecht, Boston, and London: Reidel, 1981); also, Stalnaker, “Indicative Conditionals,” Philosophia 5 (1975), reprinted in Ifs. See also David Lewis, Counterfactuals (Cambridge, MA: Harvard University Press, 1973). Lewis dedicated Counterfactuals to the memory of Richard Montague. 28 Hans Kamp, “A Theory of Truth and Semantic Representation,” in J. Groenendijk et al., eds., Formal Methods in the Study of Language (Amsterdam: Mathematical Centre, 1981). Irene Heim, The Semantics of Definite and Indefinite Noun Phrases (doctoral dissertation, MIT, 1982; New York: Garland, 1988). 29 David Kaplan, “On the Logic of Demonstratives,” Journal of Philosophical Logic 8 (1979): 81–98, reprinted in Nathan Salmon and Scott Soames, eds., Propositions and Attitudes (Oxford: Oxford University Press, 1988); “Demonstratives: An Essay on the Semantics, Logic, Metaphysics, and Epistemology of Demonstratives and Other Indexicals,” and “Afterthoughts,” both in Joseph Almog, John Perry, and Howard Wettstein, eds., Themes from Kaplan (New York and Oxford: Oxford University Press, 1989). 27

THE ERA OF SPECIALIZATION

471

a rich and sophisticated conception of the truth conditions of sentences that many theorists found to be a plausible basis for a theory of meaning. The basic idea was this: What makes a sentence meaningful is the way it is used to represent the world. A sentence that represents the world as being a certain way imposes conditions that the world must satisfy if it is to be the way it is represented to be. These are conditions in which the sentence is true. Since someone who understands the sentence knows its truth conditions, it was thought by many that a semantic theory that assigned to each sentence of a language the set of possible states of the world with respect to which that sentence was true, could be regarded as a theory of meaning. This led naturally to the view that the proposition semantically expressed by a sentence S can be identified with the set of possible states of the world with respect to which S is true. Although for some purposes this way of thinking of the proposition expressed by a sentence is useful, for others it is highly problematic. As bearers of necessary or contingent truth value, sets of possible worldstates function very well. However, this is not the only role that propositions play in our thought and language. The following seem to be not just commitments of traditional philosophical discussions of propositions, but also natural and minimal extensions of our ordinary ways of thinking and speaking: (a) that when one asserts or believes that so and so, there is something (the claim or proposition that so and so) that one asserts or believes, (b) that the things (propositions) asserted or believed are bearers of (necessary or contingent) truth or falsity, (c) that these things (propositions) are also the semantic contents expressed by sentences (relative to contexts), and (d) that typically when one assertively utters or sincerely accepts a sentence S in a context c, one asserts or believes (perhaps among other things) that which S says, or semantically expresses, in c (i.e., one asserts or believes the proposition semantically encoded by S relative to c). Propositions is simply the name we give to whatever fills all these roles. However, if each role is taken seriously, then the view that the proposition expressed by a sentence (in a context c) is nothing more than the set of possible world-states with respect to which the sentence (as used

472

EPILOGUE

in c) is true runs into trouble. For example, it seems patently false that, when P and Q are necessarily equivalent sentences (relative to c), to assert or believe what one expresses is thereby to assert or believe what the other expresses. Someone can say or believe that 1 ⫽ 1 without saying or believing that arithmetic is incomplete, or that water is H2O (even though the relevant sentences are each true with respect to all possible world-states, and so are necessarily equivalent). Problems like these raise two fundamental questions: What are the semantic contents of (propositions expressed by) sentences relative to contexts? and What are the conditions under which a propositional attitude ascription x asserts/believes that S is true of an agent? Some theorists, most notably Robert Stalnaker, have argued that by giving up the view that these attitude ascriptions always report relations between an agent and the proposition semantically expressed by the complement clause S, one can continue to identify propositions with sets of possible world-states.30 Other theorists have not been convinced. For example, in the mid-eighties Jon Barwise and John Perry argued that the problems posed by propositional attitude ascriptions (along with other constructions) required replacing semantic theories that characterize both the truth conditions and semantic contents of sentences in terms of possible world-states with semantic theories that characterize them in terms of much more fine-grained truth-supporting circumstances called (abstract) situations. Whereas possible world-states are naturally thought of as maximally complete (and consistent) properties of the entire world, the situations proposed by Barwise and Perry could naturally be seen as partial (and sometimes even inconsistent) properties of parts of the world. Having replaced possible states of the world with abstract situations, Barwise and Perry proposed to treat propositional attitude ascriptions x asserts/believes that S as reporting relations between an agent and the semantic content of S, conceived of as the set of abstract situations that would support the truth of S.31 Their hope 30 See Robert Stalnaker, “Propositions,” in A. MacKay and D. Merrill, eds., Issues in the Philosophy of Language (New Haven, CT: Yale University Press, 1976); “Assertion,” in P. Cole, ed., Syntax and Semantics, p. vol. 9 (New York: Academic Press, 1978, reprinted in Stalnaker, Context and Content (New York: Oxford University Press, 1999); and Inquiry (Cambridge, MA: MIT Press, 1984). For a very useful discussion of related issues in possible world semantics, see David Lewis, “General Semantics,” Synthese 22 (1970):18–67, reprinted in Philosophical Papers, vol. 1. Though Lewis was always content to let the objects of propositional attitudes be simple intensions of sentences—i.e., sets of possible worlds (for us, world-states), in “General Semantics” he sketches a view in which the meanings of sentences are structured intensions. 31 Jon Barwise and John Perry, Situations and Attitudes (Cambridge, MA: MIT Press, 1983).

THE ERA OF SPECIALIZATION

473

was that in doing this, they could preserve much of the structure of possible-world semantics while avoiding its notorious problems with propositional attitude ascriptions. A few years later, however, this promising program suffered a serious setback when it was shown how, using a relatively austere set of semantic assumptions shared by many different theorists (including Barwise and Perry), one can reconstruct analogs of all the usual problems posed by propositional attitude ascriptions for standard systems of possible-worlds semantics within any structurally similar theory, no matter how fine-grained the truth-supporting circumstances are taken to be.32 The lesson drawn from this by many was that propositions cannot be sets of truth-supporting circumstances of any kind; rather, they should be understood in the old Russellian way as structured complexes of objects and properties. By this time, a strong move to Russellian structured propositions was already well underway, led by the pathbreaking work of David Kaplan and Nathan Salmon. Although in “Demonstratives,” Kaplan employed the familiar technology of possible-worlds semantics in constructing his formal logic of demonstratives, he made it clear that the proper underlying philosophical picture of meaning and of propositional attitudes required Russellian propositions. In “Opacity,” he made the picture more explicit, and applied it in capturing Quine’s “relational sense of belief ”33 and in answering Quine’s long-standing objections to quantifying into modal sentences, propositional attitude ascriptions, and other referentially opaque contexts.34 In 1986, extending the insights of Kripke and Kaplan, Nathan Salmon argued in Frege’s Puzzle for a semantic theory incorporating a Millian conception of proper names (according to which the meaning of a name is its referent), a neo-Russellian conception of structured propositions (constructed out of objects and properties), and a relational treatment of attitude ascriptions (which treats them as reporting that agents stand in Scott Soames, “Direct Reference, Propositional Attitudes, and Semantic Content,” Philosophical Topics 15 (1987):47–87; reprinted in Salmon and Soames, Propositions and Attitudes, and in Peter Ludlow, ed., Readings in the Philosophy of Language (Cambridge, MA: MIT Press, 1997). 33 W. V. Quine, “Quantifiers and Propositional Attitudes,” Journal of Philosophy 53 (1956), reprinted in L. Linsky, ed., Reference and Modality (Oxford: Oxford University Press, 1971). 34 David Kaplan, “Opacity,” in Lewis Edwin Hahn and Paul Arthur Schilpp, eds., The Philosophy of W. V. Quine (La Salle, IL: Open Court, 1986). For Quine’s objections to quantifyingin, see “Reference and Modality,” in From a Logical Point of View (Cambridge, MA: Harvard University Press, 1953, 1961, 1980), and his “Notes on Existence and Necessity, “Journal of Philosophy 40 (1943), reprinted in Leonard Linsky, ed., Semantics and the Philosophy of Language (Urbana, Chicago, London: University of Illinois Press, 1952). 32

474

EPILOGUE

certain relations to the propositions semantically expressed by their complement clauses).35 In this system, the standard technology of possibleworlds semantics remains essentially in place, but it is supplemented with a systematic assignment of structured propositions to sentences relative to contexts. With widespread acceptance of the idea of structured objects of belief and assertion, later years would see intense debate about the problem of substitution of coreferential names and indexicals within attitude ascriptions. Given the Kaplan-Salmon picture, one would expect such substitution never to change truth value. Nevertheless, it has seemed to many that it is perfectly possible to believe that Hesperus is a planet without believing that Phosphorus is, despite the fact that the two names are coreferential. Different theorists, including Salmon himself (in Frege’s Puzzle), Saul Kripke,36 Mark Richard,37 Mark Crimmins and John Perry,38 Richard Larson and Peter Ludlow,39 and I40 have all weighed in on this issue—which remains controversial. It also remains important, lying at the center of an intense and many-faceted debate about the relationship between meaning and language use (semantics and pragmatics)—a debate that has become the cutting edge in the effort to develop precise theories of meaning that apply to natural language.

Broader Philosophical Lessons of the Specialized Work We ended our survey of contemporary philosophy of logic and language by mentioning a highly specialized controversy about the effects of substitution of certain kinds of expressions in propositional attitude Nathan Salmon, Frege’s Puzzle (Cambridge, MA: MIT Press, 1986). Saul Kripke, “A Puzzle About Belief,” in A. Margalit, ed., Meaning and Use (Dordrecht: Reidel, 1979); reprinted in Salmon and Soames, Propositions and Attitudes. 37 Mark Richard, Propositional Attitudes (Cambridge: Cambridge University Press, 1990), and “Defective Contexts, Accommodation, and Normalization,” Canadian Journal of Philosophy, 25 (1995):551–70. 38 John Perry, “Frege on Demonstratives,” Philosophical Review 86 (1977):474–97; Perry, “The Problem of the Essential Indexical,” Noûs, 13 (1979):3–21, reprinted Salmon and Soames, Propositions and Attitudes; Mark Crimmins and John Perry, “The Prince and the Phone Booth: Reporting Puzzling Beliefs,” Journal of Philosophy 86 (1989): 685–711; and Crimmins, Talk about Beliefs (Cambridge, MA: MIT Press, 1992). 39 Richard Larson and Peter Ludlow, “Interpreted Logical Forms,” Synthese 95 (1993): 305–56, reprinted in P. Ludlow, Readings in the Philosophy of Language. 40 Scott Soames, Beyond Rigidity (New York: Oxford University Press, 2002); Soames, “Naming and Asserting,” in A. Szabo, ed., Semantics and Pragmatics (Oxford University Press, forthcoming). 35 36

THE ERA OF SPECIALIZATION

475

ascriptions—i.e., sentences that report the assertions, beliefs, intentions, etc. of agents. As I have indicated, this controversy lies at the center of a broader debate among philosophers of language about the relationship between meaning proper and a wide variety of different aspects of language use. At issue is the relationship between the meaning of a sentence, the assertions one makes when assertively uttering it, and the information conveyed to hearers by such utterances. For too long, discussions of meaning in philosophy (and linguistics) have either ignored the differences between these categories, lumped them together, or confused one with another, often with disastrous results. One example of this involves a principle I called (in chapter 3 of volume 1) the transparency of meaning. This principle asserts that if two expressions mean the same thing, then a competent speaker who understands them both will know that they are synonymous. Plausible though it may sound in the abstract, this principle has been severely challenged by recent work on reference and propositional attitude ascriptions.41 This has significance not just for debates about where semantic facts about meaning end and pragmatic facts about language use begin, but also for broader issues about the role of meaning in philosophy. We have already seen Moore’s implicit and uncritical reliance on the transparency of meaning in his famous “open question” argument (chapter 3 of volume 1), as well as Russell’s similar use of the principle in his epistemological restrictions on logically proper names (discussed in chapter 5 of that volume). These examples are illustrations of a broad and pervasive tendency within much theorizing about language in twentieth-century analytic philosophy—a tendency to take the meanings of words, phrases, and sentences to be clear and obvious to competent speakers, and to be easily separable from other information associated with their use. Throughout these volumes, we have seen manifestations of this tendency embedded in a variety of different positions and philosophical points of view—including the linguistic theory of the apriori (chapter 12 of volume 1), the rampant appeals to analyticity decried by Quine (chapter 16 of volume 1), the later Wittgenstein’s deflationary 41 See Nathan Salmon, Frege’s Puzzle; his “How to Become a Millian Heir,” Noûs 23 (1989): 211–20; and his “A Millian Heir Rejects the Wages of Sinn,” in C. A. Anderson and J. Owens, eds., Propositional Attitudes: The Role of Content in Logic, Language, and Mind (Stanford, CA: CSLI, 1990). Also see Stephen Rieber, “Understanding Synonyms without Knowing That They Are Synonymous,” Analysis 52 (1992):224–28; and sections III, IV, and IX of my “Substitutivity,” in J. J. Thomson, ed., On Being and Saying: Essays for Richard L. Cartwright (Cambridge, MA: MIT Press, 1986), and chapter 3 of my Beyond Rigidity.

476

EPILOGUE

insistence that philosophical theses could not be genuinely informative (chapter 1 of volume 2), the methodology of the ordinary language school (discussed throughout Parts 2 through 4 of volume 2), and the discussion of whether knowing that Hesperus is Hesperus is sufficient for knowing that Hesperus is Phosphorus (discussed in chapter 15 of volume 2). What we now see at the end of the century is, in my opinion, the beginnings within the philosophy of logic and language of a less introspective, more theoretical and scientific, perspective on meaning. Though facts about the meanings of our words and the information semantically encoded by our sentences are, from this perspective, real and important, we have no privileged epistemological access to them. If this is right, it means that, as philosophers, we have no privileged and secure linguistic starting point of the sort imagined by so many of our analytic predecessors. Meaning is neither the source of all philosophical problems, nor the key to solving them all. As most of today’s analytic philosophers instinctively know, whether they have articulated the idea or not, there is no one way that philosophical problems arise. They arise everywhere—in mathematics, the sciences, the arts, and in our commonsense ways of thinking. There is also no one way they are solved; theory construction, logical analysis, conceptual clarification, even informal observation of our cognitive and linguistic practices, all play a role. There is another lesson here for the state of philosophy today, as well. If the picture I have sketched is accurate, then what seems to be the fragmentation in philosophy found at the end of the twentieth century may be due to more than the institutional imperatives of specialization and professionalization. It may be inherent in the subject itself.

INDEX

argument from illusion, 178–87 Armstrong, David, 56 Austin, John L., xiii, xiv, 68, 119, 128, 171–74, 292; and the argument that empirical knowledge does not always equire evidence, 174–78; on the argument from illusion, 178–87; on the role of ‘appearance statements’, 187–92 Ayer, A. J., 36, 180–82, 185–87; and the defense of sense datum theory, 176–78; the view of perception of, 172–73 Barwise, Jon, 466, 472–73 Belnap, Noel, 466 Berger, Alan, 233 Berkeley, George, 93, 189 Boghossian, Paul, 60 Brentano, Franz, 278 Burge, Tyler, 466 Burgess, John, 347, 353 Byrne, Alex, 56

phy of, 291–97; problems with his theory of meaning, 303–5, 307–11 Davies, Martin, 310 Descartes, René, 95–97 Donnellan, Keith, 353 Entwhistle, Rebecca, 137 Etchemendy, John, 466 Evans, Gareth, 366, 420 Field, Hartry, 466 Fine, Kit, 468 Fitch, Frederick, 353 Follesdal, Dagfinn, 353 Forbes, Graeme, 468 Foster, John, 304, 306 Frege, Gottlob, 336, 434, 461–62 Friedman, Michael, 246, 253

Carnap, Rudolf, 130, 225, 229, 294, 461, 462 Cartwright, Richard, 279 Chalmers, David, 367 Chomsky, Noam, 102 Church, Alonzo, 294 conversational implicature, 201–3; cancelability of, 208–9; contrasted with conventional implicature, 203–4; nondetachability of, 209–10; particularized and generalized, 206–7; and the transparency of meaning, 210–14; use of by Ayer in response to Austin, 176–78 Crimmins, Mark, 474

Gaifman, Haim, 466 Geach, Peter, 124, 144–46, 353; his view of good as a descriptive predicate modifier 150–52 Geach-Searle objection, 144–46, 151 Gödel, Kurt, 465 Goguen, A. J., 468 Goodman, Nelson, 236, 461 Graff, Delia, 469 Grice, Paul, xiii, xiv, 22, 68, 91, 134, 177, 225, 234–35, 292, 461; and the critique of ordinary language philosophy, 197–200; on implicatures, 201–14; on the performative theory of truth, 214–16; on principles governing communication, 200–201. See also conversational implicature Gupta, Anil, 466

Davidson, Donald, xiii, xiv–xv, 129, 217–18, 286, 456, 461; against the possibility of alternative conceptual schemes, 312–19; on evaluating his position regarding alternative conceptual schemes, 319–30; on theories of truth as theories of meaning, 297–303, 306–7; place in the history of analytic philoso-

Hare, Richard, xiii; his argument that goodness is not a property, 147–50; his performative theory of goodness, 136–46; objections to his theory of goodness, 144–47 Harman, Gilbert, 68, 241, 336, 413 Hawthorne, John, 167, 234, 286 Heim, Irena, 470

478 Hempel, Carl, 461 Herzberger, Hans, 466 Jackson, Frank, 452 Jeshion, Robin, 413 Kamp, Hans, 468, 470 Kaplan, David, 294, 347–48, 369, 376, 413, 417, 462, 470, 473–74 Kazmi, Ali, 14, 452 Keefe, R., 468 Koslicki, Karen, 429 Kripke, Saul, xiii, xv, 21, 91, 218, 252, 294, 461, 462, 466, 469, 473, 474; his arguments against the description theory of proper names, 336–40, 356–71; the dilemma posed by his discussion of names, 393–95; the epistemological consequences of his discovery of the necessary a posteriori, 374–78, 395–96; on essentialism, 347–54; his examples of the contingent a priori, 403–17; on historical chains as fixing reference, 364–71; and the importance of the discovery of the contingent a priori, 417–22; on an objection to the necessary a posteriori, 445–56; on names introduced by description and their importance for the contingent a priori, 397–403; on the necessity of theoretical identities involving natural kinds, 432–43; the place in analytic philosophy of, 335–36, 456; on possible worlds, 354–56; on rigid designation and its philosophical consequences, 340–56; on the similarities between names and natural kind terms, 423–30; his view that true identity statements involving names are knowable only a posteriori, 379–93; on Wittgenstein, 34, 39, 41, 42, 60 Larson, Richard, 297, 310, 395, 474 Lemmon, E. J., 128 Lewis, C. I., 462 Lewis, David, 59, 128, 366–67, 369, 448, 452, 461, 462, 470, 472 logical positivism, xix Ludlow, Peter, 395, 474 Machina, K., 468 Malcolm, Norman, xiii, 157; the limitations of his anti-skeptical argument,

INDEX

166–70; his paradigm case argument, 85 n. 16, 161–66, 190–91 Marcus, Ruth, 294, 353, 379, 462 McGee, Van, 466 McGlone, Michael, 393 McKeown-Green, Jonathan, 371 Mill, John Stuart, 433–34 Montague, Richard, 294, 462, 469 Moore, G. E., xv–xvi, 55, 70, 75, 285, 456, 461, 475; anti-skeptical arguments by, 157–61 Nagel, Thomas, 336 Nozick, Robert, 462 paradigm case argument. See Malcolm, Norman Parsons, Charles, 466 performative sentences, 116–28 Perry, John, 472–73, 474 Pitcher, George, 56 Plato, 368 Priest, Graham, 466 Prior, Arthur, 294, 353 private language argument. See Wittgenstein, Ludwig Pryor, Jim, 192 Putnam, Hilary, 435, 461 Quine, W. V., xiii, xiv, xx, 216–18, 295, 302, 320, 330, 359, 379–80, 402, 456, 461, 473, 475; on arguments for the indeterminacy of translation, 243–58; development of the philosophy of, 224–28; Kripke’s response to his objections to essentialism, 347–54; on the inscrutability of reference, 260–77; paradoxical views of, 277–86; on the underdetermination of semantic theories by data, 228–43 Raffman, Diana, 469 Ramsey, Frank, 461 Rawls, John, 462 Richard, Mark, 395, 474 Rieber, Stephen, 475 rigid designation. See Kripke, Saul Russell, Bertrand, 74, 90, 91, 253, 336, 353, 434, 456, 461, 475; on logical forms, xvi–xviii; on proper names, 11–15 Ryle, Gilbert, xiii, 45, 116, 189; on acting skillfully, 102–6; assessment of the phi-

INDEX

losophy of, 114; on ‘Descartes’s myth’, 93–99; on fatalism, 71–81; on knowing how and knowing that, 99–102; life of, 67–68; on the nature of philosophical problems, 68–70, 75–81; objections to his philosophy of mind, 106–11; on perception and skepticism, 82–90; on science and commonsense, 81–82; on selfknowledge, 111–13 Salmon, Nathan, 15, 355, 378, 389, 395, 403, 417, 440, 444, 473–74, 475 Searle, John, 20, 124, 128, 144–46, 151 Segal, Gabriel, 297, 310 sense data, 44–59, 172–74, 178–87 Smullyan, Raymond, 353 Speaks, Jeff, 8 Stalnaker, Robert, 109, 355, 375, 452, 462, 470, 472 Strawson, Peter, xiii, 36, 135, 198, 225; his confusion of meaning and use, 128–34; Grice’s critique of his theory of truth, 214–16; objections to his theory of truth, 120–26; his performative theory of truth, 116–20; on the private language argument, 49

479 Tappenden, J., 468 Tarski, Alfred, 217, 272, 282, 293–94, 321–22, 325–28, 336, 461, 462, 465, 466 Thau, Mike, 56 Unger, Peter, 468 Warnock, G. J., 189 Wilson, George, 60 Wittgenstein, Ludwig, xiii, xiv, xviii–xix, xxi, 68, 70, 74, 75, 90, 116, 129, 292, 330, 456, 461, 475; on communal agreement, 38–44; a critique of the Tractatus by, 4–24; on family resemblance, 15–17; on the nature of philosophy, 27–31; his new conception of language, 24–27; on ostensive definition, 6–8; the private language argument of, 44–59; on proper names, 11–15, 17–24; on reference, 8–11; on rule following, 33–44, 51–57 Wright, Crispin, 60, 468 Zadeh, L. A., 468 Ziff, Paul, 353