Current Issues in Theoretical Philosophy: Volume 1 What is Truth? [Reprint 2012 ed.] 9783110886665, 9783110164411

Citation preview

What is Truth?

Current Issues in Theoretical Philosophy Edited by Richard Schantz

Vol. 1

w DE

Walter de Gruyter · Berlin · New York 2002

What is Truth? Edited by Richard Schantz

w DE

G_

Walter de Gruyter · Berlin · New York 2002

© Printed on acid-free paper which falls within the guidelines of the ANSI to ensure permanence and durability.

Die Deutsche Bibliothek — Cataloging-in-Publication Data What is truth? / ed. by Richard Schant2. - Berlin ; New York : de Gruyter, 2002 (Current issues in theoretical philosophy ; Vol. 1) ISBN 3-11-016441-8

© Copyright 2001 by Walter de Gruyter GmbH & Co. KG, D-10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording or any information storage and retrieval system, without permission in writing from the publisher. Printed in Germany Data conversion: DTP Johanna Boy, Brennberg Printing and binding: WB-Druck, Rieden/Allgäu Cover design: Christopher Schneider, Berlin

Contents Introduction

1

I The Correspondence Theory Truth: Concept and Property William P. Alston

11

Truths and Truthmakers David Μ. Armstrong

27

Truth Through Thick and Thin Richard Boyd

38

The Metaphysics of Deflationary Truth Michael Devitt

60

Truth, Meaning, and Reference Richard Schantz

79

II Deflationism Defended Explanatory vs. Expressive Deflationism about Truth Robert Brandom

103

On Locating Our Interest in Truth Dorothy Grover

120

Norms of Truth and Meaning Paul Horu/ich

133

On Some Critics of Deflationism Michael Williams

146

III Deflationism Attacked Minimalism and the Facts about Truth Marian David

161

VI

Contents

Disquotationalist Conceptions of Truth Wolfgang Künne

176

The Truth about Truth Colin McGinn

194

Generalizations of Homophonie Truth-sentences Peter van Inwagen

205

IV Tarski Challenged An Argument Against Tarski's Convention Τ Anil Gupta

225

What is Truth? Stay for an Answer Jaakko Hintikka

238

V Alternative Approaches The Two Faces of the Concept of Truth Michael Dummett

249

Truth: A Prolegomenon to a General Theory Lorenz Β. Puntel

263

How Not to Misunderstand Peirce - A Pragmatist Account of Truth Jay Rosenberg

283

A Problem about Truth Ralph Walker

299

An Indefinibilist cum Normative View of Truth and the Marks of Truth David Wiggins

316

Index of Subjects

333

Index of Names

334

Contributors

337

Introduction This book is the start of a series of three volumes dedicated to central debates in contemporary theoretical philosophy: What is Truth? (Volume I) The Externalist Challenge: New Studies on Cognition and Intentionality (Volume II) Prospects for Meaning (Volume III). The essays in this volume are new attempts at answering an old philosophical question: "What is truth?" This time-honoured question has become again a focal point of philosophical discussion. Competing answers have been given: truth is correspondence, truth is coherence, truth is pragmatical utility, truth is a primitive unanalyzable property, and truth is disquotation. At first glance, this plurality of answers might strike one as surprising. Is there not a rather simple answer to this venerable question, an answer already given by Aristotle, when he said: "To say of what is that it is not, or of what is not that it is, is false, while to say of what is that it is, or of what is not that it is not, is true." This suggests that a statement is true if and only if things in the world are as the statement says they are. Aristotle's dictum expresses the fundamental intuition underlying the classical correspondence theory of truth. According to this theory, a statement is true just in case it corresponds to a fact, and false just in case it does not correspond to a fact. The statement that grass is green is true because it corresponds to the fact that grass is green, and the statement that horses can fly is false because it does not correspond to any fact. Perhaps these are platitudes, but at least they seem to be faithful to our ordinary, pretheoretical concept of truth. It is no wonder, then, that almost every philosopher reflecting on the nature of truth before the eighteenth century accepted, explicitily or implicitly, the correspondence theory. To be sure, an adequate version of the correspondence theory has to go beyond truisms hardly anybody denies. The central concepts it employs must be explained, and, moreover, explained in a way that does not appeal to the notion of truth itself. What are facts, the extralinguistic and extramental relata of the correspondence relation, the truth makers of our statements? What are the constituents of facts? And what kind of relation is correspondence? Can correspondence be construed as congruence, as a relation of structural isomorphism between statement and fact, as Russell and early Wittgenstein once thought? Or are statement and fact merely conventionally correlated as whole units, as Austin was later to suggest? Or, and this a more modern view, can reference be explicated as a causal relationship between words and aspects of the world, and hence in a physicalistically respectable way (early Hartry Field)? Further, what are the bearers of truth value? Certainly, this is a question every, or almost every, theory of truth must face. In any case, it must be faced by those who believe that truth is a property, and that there really are things that are true or false. So the correspondence theory cannot allow itself to be silent on the nature of the truth bearers. Prima facie, there seems to be a multiplicity of things to which we apply "true"

2

Introduction

and "false". Among them are statements, propositions, sentences, utterances, and beliefs. So far I have assumed that it is statements that are true or false. The word "statement" is ambiguous, however. It may mean either the act of stating something or that which is stated. It seems that in declaring a statement to be true we are not, or at least not primarily, talking about an act of stating something but about what is stated about the content of the statement. A statement, in this sense, is often called a proposition. Since the content of a statement can also form the content of a belief and of other mental attitudes, the possibility seems to present itself of reducing the apparent multiplicity of potential truth value bearers to a single candidate, namely propositions. But what sorts of things are propositions? Are they abstract entities in a Platonic sense? Can they be identified with linguistic items — sentence types or sentence tokens? Are they sets of possible worlds? And how, so many philosophers ask, can the possession of content be integrated into a physicalist world view? So a decent correspondence theory has to develop accounts of the ontological status of facts, the status of correspondence, and the status of propositions. A good many philosophers, however, think that traditional attempts to explain the notions of fact and correspondence have generated nothing but empty pseudoexplanations. It is, so the critics of the correspondence theory maintain, a bad metaphysical theory because the central concepts it invokes possess no explanatory value at all. Rather, they can be understood only in terms of what they are supposed to explain. It is the concept of truth that wears the breeches; facts and correspondence being simply projected from truth, according to the very simple rule that the fact that p is that which corresponds to the true statement that p. Thus, all that we seem to gain by introducing facts and the relation of correspondence is the ability to say, in esoteric language, that a statement is true. Some of the most excellent philosophers, such as Strawson, Quine, Davidson, and Putnam, dismiss the correspondence theory for these or related reasons. Not surprisingly, however, their objections did not remain unanswered. The correspondence theory is also confronted with difficulties of a quite different kind. This theory is typically and naturally associated with metaphysical realism the view that there is an objective reality, a reality whose existence and structure are independent of our language and thought. There is a very popular objection to this combination of the correspondence theory and metaphysical realism, namely that it leads to epistemological scepticism. For the correspondence theorist, truth is an epistemically unconstrained concept; it is, in Hilary Putnam words, "radically nonepistemic". That is, whether a statement is true or not, does not depend on any epistemic virtue it displays. Hence, truth is not a matter of whether a statement is justified, warranted or rational. Truth is objective, hinging only on the way the world is. The charge, then, is that on classical correspondence realism we can never determine whether statements or beliefs are true because we cannot compare them with the facts to see whether they correspond to them. Statements and beliefs, so it is usually argued, may be compared with other statements or beliefs to see if they harmonize with each other or not. But we can never compare or confront statements or beliefs with the facts or with reality. There is, so it is often said, no way to get outside our language or outside the circle of our beliefs and explore the facts themselves (Brand Blanshard, Rorty).

Introduction

3

For many philosophers, this epistemological objection was the main motive to abjure the classical correspondence theory. Still convinced that truth has an inner nature, which it is the aim of a theory of truth to discover, they began to offer alternative substantive theories promising to be more faithful to our epistemic situation in the world. Since the correspondence theory was thought to lead to sceptical worries, to a radical separation between mind and world, between subject and object, attempts were now made to redefine truth so as to make it humanly accessible. Epistemic accounts began to flourish, claiming that the truth of a statement does not consist in an external relation to a feature of reality but in its possessing a positive epistemic status within our conceptual scheme or within our experience. Epistemic conceptions simply equate the truth of a statement with its verifiability, justifiability or warranted assertibility (Peirce, James, Dewey, Dummett, Putnam, Rorty). To this basic idea, coherence theories of truth add the further thesis that justification is, in general, a holistic affair. Accordingly, these theories, championed by the neoHegelian absolute idealists Η . H. Joachim, Ε Η. Bradley and Brand Blanshard as well as by Logical Positivists, such as Hempel and Neurath, hold that the truth of a judgment consists in its being a member of a comprehensive system of beliefs which is consistent and harmonious. It is noteworthy, that coherence theorists typically conceived of their theories as supplying not just a criterion or test of truth enabling us to determine whether a statement is true or false, but as supplying a definition, saying that truth is constituted by coherence between beliefs. Pragmatic theories of truth, developed in various forms by Peirce, James and Dewey, insist that there is a close connection between the concept of truth and our human experience and practice. According to the "pragmatic maxim", the meaning of a concept or an idea consists in the practical consequences of their use. In line with this, their approach to truth was to ask what difference it makes whether a belief is true. Peirce famously defined truth in epistemic terms as "the opinion which is fated to be ultimately agreed to by all who investigate". By definition, the final and common opinion reached by the use of the scientific method at the end of inquiry is the true opinion. So truth consists, primarily, in agreement amongst ourselves, not, or at least not primarily, in agreement with the world. For Peirce, a belief seems to be true just in case it is justified by the consensus of opinion we will attain at the end of scientific investigation. Putnam's former identification of truth with justification under ideal epistemic circumstances took up and refined Peirce's consensus theory. Pragmatists approach truth from an epistemic point of view which incorporates basic elements of coherence. But by additionally attaching epistemic significance to the dimension of sensory experience, they enrich pure versions of coherentism. That is what lies behind such typical pragmatist claims as that a belief is true if and only if it works, that is, if and only if it is a useful basis for action, which were often misunderstood, e.g. by Russell and Moore. Acting on a belief has to pay off, and it pays off if our future sensory experiences confirm the web of our beliefs. So for a belief to be true, it has to cohere not just with other beliefs but to be consonant with the deliverances of sensory experience as well. Whereas pragmatic theories tend to conflate the theory of truth with epistemological matters, Alfred Tarski accentuates the epistemological and metaphysical neutrality

4

Introduction

of the concept of truth. T h e goal of his pioneering "semantic conception of truth" was to explicitly define truth predicates for a n u m b e r of formalized languages, presumed to be adequate for science and mathematics. This limitation to formalized languages is the reason why he can treat sentences as bearers of truth value. Tarski was convinced that ordinary truth predicates occurring in natural languages inevitably give rise to paradoxes, above all to the Liar paradox. So his prime aim was to show that the concept of truth, if carefully employed, is a consistent concept, a concept which does not involve us in semantic paradoxes. Against the general truth scepticism of some Logical Positivists, Tarski wanted to rehabilitate the concept of truth as a scientifically and metaphysically respectable concept, as a concept which even conforms to the dictates of the doctrine of physicalism. Therefore, he had to define truth without recourse to unreduced semantic concepts. This is the background of his project to provide an exact definition of truth, a definition which must fulfill two conditions: firstly, it must be "materially adequate", and secondly, it must be "formally correct". T h e first of these conditions restricts the possible content, the second the possible form, of any acceptable definition. To fulfill the material adequacy condition Tarski introduces his famous Convention T: An acceptable definition of truth for a language must have among its consequences all instances of the following schema: s is true if and only if p where Ύ ' is to be replaced by a standardized name of any sentence of the object language, the language for which truth is being defined, and "/>" is to be replaced by that same sentence or its translation, depending on whether or not the object language is contained in the metalanguage, the language in which truth is defined, as a proper part. To use Tarski's example, an instance of this schema would be: "Snow is white" is true if and only if snow is white. Tarski calls every equivalence of form Τ a "partial definition of truth", which explains what the truth of the sentence in question consists in. A n d he often characterizes his project as providing a definition equivalent to the logical conjunction of all such instances of T, all such " J'-sentences" as they are usually called nowadays, for the language under study. Indeed, Tarski mentions that if the language consisted of only a finite number of sentences, Convention Τ could be satisfied by a definition that simply listed a /"-sentence for each sentence in the language. Any interesting language, however, has a potential infinity of sentences. For such languages the definition must take another form. In the case of languages whose only complex sentences are truth functions of their components, it is possible to define truth for complex sentences directly in terms of truth for elementary sentences. But with languages with quantificational structure this direct method founders. It is the most distinctive feature of Tarski's definition of truth that he ingeniously solved this problem with the concept of satisfaction. Satisfaction is a relation between open sentences and infinite sequences of objects which belong to the range of variables of the language. T h e essential idea of Tarski's solution is, firstly, to construct a recursive definition of satisfaction, and then to define truth on that ba-

Introduction

5

sis. Just as closed sentences are limiting cases of open sentences, so truth turns out to be a limiting case of satisfaction. Accordingly, Tarski defines a sentence as true just in case it is satisfied by all sequences of objects. While Tarski's formal method of defining truth has become a standard part of contemporary logic, there is considerable disagreement about the philosophical significance of his work. Some philosophers, e.g. Popper and Davidson, claim that Tarski established the legitimacy and fruitfulness of the concept of truth in philosophical investigations. Others, e.g. Putnam, deny that his formal results have much to do with our ordinary, intuitive concept of truth. One of the most hotly debated questions is whether Tarski's theory is a correspondence theory. He himself asserts that his semantic conception, by meeting the condition of material adequacy, does justice to the basic intuitions of the correspondence theory and of the classical Aristotelian conception of truth. Some philosophers share his assessment (Popper, Ayer, Field, early Davidson). Others take a more pessimistic view (J.L. Mackie, Susan Haack, late Davidson). Hartry Field made the stimulating proposal to interpret correspondence in terms of causal relations between uses of words and features of the world. He raised the objection that Tarski did not accomplish the aim he set himself, the aim to supply a definition that would satisfy the demands of physicalism. If, so Field argued, truth is defined in terms of satisfaction and reference, the problem of explaining satisfaction and reference still remains. Here we must go beyond Tarski's merely list-like definitions. Early Field hoped that the causal-historical theories of reference along the lines suggested by Saul Kripke, Hilary Putnam and Keith Donnellan might yield a plausible analysis. Deflationary or minimalist views of truth, some of them inspired by Tarski's seminal work, have provoked one of the most important debates in contemporary philosophy. According to this radical alternative to traditional views, the concept of truth is a clear and uncontentious concept, a concept which is philosophically much less interesting than proponents of robust theories think. Truth, deflationists claim, has no substantive role to play in philosophy. There are various brands of deflationism: there is Frank Ramsey's redundancy theory of truth, there is Peter Strawson's performative theory, there is the prosentential theory developed by Dorothy Grover, Nuel Belnap and John Camp, there is the sophisticated variation on it presented by Robert Brandom, there is the disquotationalist account advocated by W.V. Quine, Alfred Ayer, Richard Rorty, Stephen Leeds, and Michael Williams, there is Scott Soames's subtle defence of the general deflationary perspective, and there is last but not least Paul Horwich's "minimalism". What all these various deflationary views have in common is their conviction that substantive or robust theories of truth - such as correspondence or coherence or pragmatic theories - are all on the wrong track. According to deflationism, substantive theories share the assumption that truth has an inner nature, a nature which can be analyzed in epistemic or semantic or metaphysical terms. Deflationists categorically reject this assumption. There is no single substantive property all true statements share. Truth, they urge, has no underlying nature, no hidden essence. The concept of truth expresses neither a natural or real property nor a natural or real relation. For this reason, it cannot play a causal or an explanatory role in good systematic theories. Since there are no interesting connections between the concept of truth and fundamental

6

Introduction

philosophical concepts, such as meaning, belief, statement, translation, and synonymy, the concept of truth should not be given a central place in our philosophical reflections. Rather, truth is a purely formal or logical concept whose correct explanation requires far less extravagant conceptual resources than advocates of substantive theories believe. A radical and simple form of deflationism is the redundancy theory, which, although anticipated by some remarks of Gottlob Frege, was put forward by F.P. Ramsey. His basic idea was that statements of the form "p" have exactly the same meaning as statements of the form "It is true that p". This led Ramsey to think that the predicate "true" is redundant, eliminable by paraphrase from all contexts without semantic loss. "True" may have the pragmatic function of signaling agreement with a statement that has already been made. But it does not play a semantic role; it contributes nothing to the meaning of the sentences in which it occurs. It is a grammatical illusion to think that in asserting something of the form "It is true that p", we are attributing a property to a proposition; in reality, we are attributing nothing, but merely asserting that p. Naturally, Ramsey was well aware that not all uses of "true" involve an attribution to an explicitly given proposition. There are blind attributions of truth, e.g. "What the Pope said yesterday is true", and there are generalizations deploying "true", e.g. "Everything the Pope says is true". Ramsey tried to cope with such uses of "true" by using the resources of propositional quantification, leading him to paraphrase the last sentence as "For all p, if the Pope says p, then p". But, as soon was pointed out, both the standard objectual and the substitutional interpretation of the quantifier present serious problems. Another very influential deflationary view is disquotationalism, which derives primarily from the work of W.V. Quine. In contrast to the redundancy theory, disquotationalism takes sentences, not propositions, as the primary truth value bearers. In light of the difficulty of saying what they are, Quine, in particular, is suspicious of any notion of a language-independent proposition favouring instead sentence tokens as concrete, empirical entities. Disquotationalism is a great advance beyond the redundancy theory because it offers an illuminating answer to the pressing question as to why we have the word "true" in our language, which Ramsey remained silent about. Indeed, Quine approaches truth from considerations about the function of the truth predicate. The term "true" is presented, in the first place, as a device for semantic ascent and semantic descent. If we go up one level and attribute truth to the sentence "Snow is white", then we attribute whiteness to snow. The truth predicate enables us to return from the level of talk about language to the level of talk about the world. The use of "true" signalizes that, though a sentence is mentioned, and we are thus operating on a linguistic plane of reference, our interest is nevertheless not directed on language but on extralinguistic reality. The ascription of truth, as it were, sweeps away the quotation marks, producing a sentence suitable for saying that snow is white. In other words, it is the function of "true" to cancel the effect of semantic ascent. That is why Quine says quite succintly: "Truth is disquotation". Disquotationalists stress that the truth predicate is required primarily in those cases in which we want to generalize with respect to sentence positions (Leeds, Soames). We need "true" for saying in general that every sentence that has a certain form is true; or

Introduction

7

in deductive logic, to justify rules of inference, that is, to be able to say that in every inference of a certain kind truth is transmitted from the premises to the conclusion. The utility of the truth predicate is said to lie in its providing for a way of generalizing with respect to sentence positions. "True" is in the end merely a surrogate for infinite conjunctions and infinite disjunctions. If our language allowed the expression of infinite conjunctions and infinite disjunctions, the truth predicate would lose its most important function. So the truth predicate is not merely a device for disquotation; it is also a device for expressing or abbreviating infinite conjunctions and disjunctions by finite means. On the prosentential theory, developed by Dorothy Grover, John Camp and Nuel Belnap, there are expressions other than pronouns that can be used anaphorically, i.e., that can substitute for a previous primary expression of the same syntactical form. There are proverbs, proadjectives, and there are prosentences. Whereas pronouns occupy positions nouns occupy, prosentences occupy positions sentences occupy. The central claim of the proponents of this theory is that "that is true" and "it is true" function as prosentences. Thus "It is true that grass is green" can be construed as "Grass is green. That is true". It is important that prosentences are semantically unstructured units. All expressions in which "true" occurs outside of a presentence have a misleading surface grammar. In the deep structure "true" is not a genuine predicate with a separate meaning. On the contrary, "true" is always a syncategorematic fragment of a prosentence. A key claim of this account is that prosentences and other proforms can be used as variables of quantification, as well as they can be used to substitute for an antecedent. "Everything the Pope says is true", e.g., is construed as "For each proposition, if the Pope says that it is true, then it is true." This, very roughly, is the prosentential account of the role of the truth predicate in generalizations. The prosentential theory is similar to the disquotational theories in that both claim that the special usefulness of the truth predicate consists in providing us with certain kinds of expressibility — making available generalizations with respect to sentence positions. But their analyses differ in certain formal respects, in the logical mechanisms they employ. In contrast to disquotationalist and prosentential theories, Paul Horwich's minimalism treats propositions as the primary truth value bearers. From the outset, he points out that his account does not yield an explicit definition or analysis of "true". A satisfactory characterization of the meaning of a word does not have to take the form of a specification of necessary and sufficient conditions for its correct application Rather, he regards his account as an implicit or use definition of "true". On his view, the content of "true" is exhausted by his "minimal theory" which simply takes as axioms all nonparadoxical instances of the following equivalence schema: (T) The proposition that ρ is true if and only if p. He concedes that the theory cannot be explicitly formulated - owing to the infinity of the instances of his equivalence schema and owing to the inexpressibility of some propositions in any natural language. Of course, proponents of robust theories of truth will also accept the equivalence schema (T) or a similar one. What is characteristic of minimalism, however, is, as Horwich remarks, the further claim that the meaning of "true" is fixed by our commitment to (T). According to minimalism, there is no need to ex-

8

Introduction

plain why we regard the instances of (T) as true, no need to justify our acceptance of them by appealing to underlying facts about correspondence or coherence or pragmatical utility. Rather, our inclination to accept the instances of the equivalence schema is supposed to explain our overall use of the truth predicate. Deflationary accounts deny that truth has a nature. Adherents of substantive theories disagree. The essays in this volume bear witness to the urgency, significance and liveliness of the central philosophical issue of whether truth has a nature, and if so, what its nature consists in. Several people have contributed in various ways to the successful completion of this project. All of them are warmly thanked. Above all, I owe thanks to my colleague Oliver Scholz whose constant encouragement and advice were a great help to me. Special thanks go to the library of the Department of Philosophy at my former university, the Free University of Berlin, in particular to the librarian, Susan Hechler. Further thanks are due to Gertrud Grünkorn, Hans-Robert Cram and the editorial staff at Walter de Gruyter for their expert assistance. Finally, I would like to add a word of thanks to my student Daniel Seibel, who did a fine job with the computer.

I The Correspondence Theory

Truth : Concept and Property WILLIAM P. ALSTON

I In my A Realist Conception of Truth, 1996 (hereinafter RCT) I distinguished between the (ordinary) concept of propositional truth (truth as a property of propositions), and the properly of propositional truth. This was done in a hurried fashion in the context of a discussion of the relation between my realist conception of truth and a correspondence theory of truth. The suggestion was that here, as elsewhere, a property might have various features not reflected in our concept of that property. To choose a well worn example, heat (the property of a physical object's being more or less hot) is revealed by physics to be an average kinetic energy of constituent molecules, even though our ordinary concept of heat involves no such component (that's not the way we ordinarily identify heat). Similarly, it may be that the property of truth really is what is specified in one or another version of the correspondence theory, even though, on my view and other "minimalist" views, the concept we wield in ordinary thought and talk when saying of a proposition, statement, or belief that it is true does not specify any mode of "correspondence" with a fact, a truth maker. The minimalist concept is the one that is displayed by instances of the Truth Schema (T-schema): The proposition that ρ is true iff p. Let me take a moment to expand a bit on this minimalist version of the concept of truth. The idea is that what it takes for a proposition to be true is given in the content of the proposition itself. What it takes for the proposition that lemons are sour to be true is simply that lemons are sour, and so for any other instance of the T-schema. To specify the necessary and sufficient condition for a proposition to be true we simply repeat the propositional specifying phrase. Although this is not an explicit definition of 'true', even of a contextual sort, it suffices to pin down the meaning. Anyone who realizes that any substitution instance of the T-schema is a conceptually (analytically) necessary truth thereby has the concept of propositional truth (understands 'true' as applied to propositions). Please note that I am not suggesting that 'The proposition that lemons are sour is true' means the same as 'Lemons are sour'. That is the claim made by "redundancy" accounts of truth, a claim that leads straight to the denial that there is any property of truth, a denial that would undermine the problem dealt with in this paper. Two propositions can be analytically equivalent without being synonomous. Another example is 'Roses are red' and 'Roses exemplify the property of being red'. Just as 'The proposition that lemons are sour is true' differs in meaning from 'Lemons are sour' in expressing the concept of truth, unlike the latter, so 'Roses exemplify the property of being red' differs in meaning from 'Roses are red' by the fact that it, unlike the latter, expresses the concept of exemplification.

12

The Correspondence Theory

I also need to say something concerning the variety of truth bearers. Calling the concept of truth with which I am dealing 'prepositional truth' reflects the fact that I think of propositions as the primary bearers of truth value. I take it that when we attribute a truth value to a statement or belief we do so by virtue of the proposition expressed therein, what is stated or believed. Thus we can think of truth values of statements or beliefs as derivative from the truth values of the propositions they express.1 Hence I oscillate freely between speaking of propositions, statements, and beliefs as true or false. And this enables us to replace 'proposition with 'statement' or 'belief' in the T-schema. A statement (belief) that p is true iff p.1 To return to the central concern of this essay, given that the concept-property distinction was made only in passing in RCT, it is not surprising that reviews and other discussions of the book have paid little attention to it. I feel that it deserves more attention. Discussions of truth are led seriously astray by neglecting it. This paper is devoted to documenting this claim, to sketching out the way(s) in which a property of truth might go beyond the concept thereof, and to defending the project of seeking an extra-conceptual account of the property . The first order of business here is to lay out a general framework for the discussion. And that will involve, first of all, explaining my use of terms and making explicit how far I am and am not going in such explanations. Within the limits of this paper I am unable to do more than indicate the concept of concept I employ.3 My target will be sufficiently identified by the following stipulations. 1. One way of being in possession of a concept, the way that is germane to this discussion, is by attaching a certain meaning to a certain word or phrase (using the word or phrase with that meaning). So my concept of truth can be roughly identified with what I mean by 'true' in the relevant range of employments. And the latter can be roughly identified with what I say of something when I apply 'true' to it. 2. To have a concept is to be capable of certain cognitive discriminations. The cognition involved can be perceptual or "higher level" abstract thought. I have a visual concept of a tree if I can distinguish trees from other objects by the way they look, whether this discrimination involves applying a term to them, or reacting to them non-

1

2

3

One may wonder why the currendy most popular candidate for a truth-value bearer, sentences, is omitted from the above list. The reason is that sentences in natural languages, even those usable for making statements, do not, in general, express a unique proposition, even when used with the same meaning, and hence cannot be counted on to possess a stable truth value. For more on this, see Alston 1996, Ch. 1, sec.ii. Note that this way of relating propositions, statements, and beliefs, gives us a simple way of identifying propositions. A proposition is that which one states in making a statement, or that which one believes in having a belief. Alternatively put, it is the "content" of a statement or proposition. This does not amount to a choice between competing views as to the ontological status of propositions. But it does provide a working grasp of the notion of a proposition, and hence an identification of what the ontological question is a question about. For extended treatments see Peacocke 1992, Fodor 1998, Weitz 1988,

Truth: Concept and Property

13

verbally in some way. I have a higher level concept of trees if I am capable of forming a judgment which concerns trees in some way, e.g., the judgment that deciduous trees have leaves. And, of course, in making the discriminations I exercise such a capacity (employ the concept). 3. We may think of a concept as made up of "constituent features", representations of what one attributes to something when one applies the concept to it. Where we have an adequate explicit definition of a term that expresses a given concept, those features are the ones listed in that definition. But many meaningful terms are not susceptible of useful explicit definitions. Many theorists, including myself, hold that 'true' (as applied to propositions) is of that sort. But we can also identify constituent features in other ways. As pointed out above, I take the T-schema to exhibit the central feature of the concept of propositional truth. So much for 'concept'. For purposes of this paper I will take 'property' not to need even the degree of explication I have provided for 'concept'. It will suffice to say that we may take subject-predicate statements to involve the attribution or properties to "things" (to property-bearers), in the wide sense of 'property' used here, in which it includes kinds as well properties more narrowly so called. (For example, we can take the kind tree to be correlated with the property of being a tree.) And grammatical transforms of subject-predicate statements ('John's being tall', 'the fact that John is tall', 'were John tall', 'is John tall?', etc.) all involve the thought of attributing the property of tallness to John. The other preliminary task is to make some basic points about the relation of concept and property. In my opening remarks I indicated the point of central concern in this paper, viz., that a property can involve features not represented in our concept of that property. But this must be balanced by a recognition of other facets of the relationship. First, there can be a number of different concepts of the same property or other sort of conceptualized object. These concepts can differ in various ways, of which I will present a sample. (1) Most obviously, concepts can differ in degree of adequacy. One crude concept of a fish is exhausted by being an animate organism that lives in the water. But a more penetrating zoological taxonomy will reveal that some animals that live in the water are mammals, which differ in many important biological respects from most water denizens. One might step up the concept by adding the feature of being scaly and having fins. But that would exclude many species that are biologically similar to scaly finbearing fish, e.g., jelly fish. To go further into this example I would have to identify the most important distinguishing characteristics from the standpoint of general biological theory, and I will have to forego that. I will ask the reader to imagine a set of such features that distinguishes fish from other animals, and contrast the concept embodying these with the crude concept. When concepts with different extensions, as in the above case, are said to be concepts of the "same thing", we are obviously cutting corners a bit. The overlap in the extension is so large (and, perhaps, contains most of the most commonly observed members) that we can think of the different concepts as applying to the same set of items. But that is rough. And that leads us into the next dimension of difference.

14

The Correspondence Theory

(2) Two concepts with the same extension may differ in the selection of common and distinctive features of the members of that extension that they represent. An example made popular by Quine is 'creature with a heart' and 'creature with a kidney'. (I take it on faith that Quine knows his biology here.) Here, presumably there is no difference in adequacy of the two concepts. Both are of a low degree of scientific adequacy, having been constructed for the purpose of philosophical illustration. A more respectable example would be the concept of a three-sided plane rectilinear figure and of a three-angled plane rectilinear figure. Same extension again, but presumably neither is more adequate than the other. Here is an example like the above except for exhibiting degrees of adequacy. Say that my concept of a certain species of maple tree involves a distinctive leaf shape and a distinctive texture of bark. Contrast that with a concept in terms of DNA. The latter is certainly more fundamental for purposes of biological theory, even though they both cover the same extension. (3) The final mode of difference I will mention has to do with degree of articulation. We can contrast a purely perceptual concept of motion (one can in a wide variety of cases tell whether something is moving by looking at it) and more articulated concept of motion as a process of a material object coming to be in different places at different times, where the possessor of this latter concept can spell out the articulation verbally. To apply all this to truth, there could be different concepts of truth that pick out the same property. Here I am not speaking of the obvious point that 'true' is used with quite different meanings in, e.g., 'true friend', 'true likeness', true bill' and 'true proposition'. No, the point is that propositional truth itself may be conceptualized differently. We may contrast a "minimal" concept that is conveyed by the T-schema with a more developed correspondence concept employed by, e.g., Wittgenstein in the Tractatus. Here too there are questions of sameness and difference of extensions, but let's assume for purposes of discussion that there is a overlap in a large set of central cases, so that we can say that they are both concepts of (propositional) truth. This means that when we consider the possibility that the property of truth might go beyond the concept, we must be careful to specify which concept we are talking of. Just as the technical concept of heat embodies the feature of average kinetic energy of constituent molecules, while the ordinary concept does not, so the property of truth might go beyond the ordinary concept but not some technical philosophical concept. More generally, there is no specifiable feature of a property that could not be represented in a concept of that property. So far I have been concentrating on differences between concept and property and differences between concepts of (more or less) the same property. But we must also recognize connections between a concept and its object. Most importantly, if a concept of Ρ really deserves that title, it does at least a workable job of directing our thought to Ρ rather than to some other property. If it differs somewhat in extension from a completely adequate concept of P, like the above example of fish, it will not do an ideal job of this, but it will still serve for many purposes. And it can serve as a basis for refinement of that concept, as regularly happens in theoretical advances, which always begin, if we look back far enough, with ordinary concepts. And even if the ordinary concept is not completely satisfactory theoretically because it doesn't represent the

Truth: Concept and Property

15

essence of the property, as with the above example of heat (along with other "sensible" properties), it can guide us in our attempts to get at what underlies what is represented in that concept, in this case, sensed effects of the property itself. More on this when I come in section iv to discuss different ways in which a property can go beyond a concept of that property. Moreover, since the concept of Ρ we employ is our guide to thought and talk about Ρ and to further investigation of P, it puts constraints on an account of the nature of P. If an account of heat is to deserve that title, it must be an account of what feature of a perceivable object is responsible for sensations of heat when that object is perceived. Otherwise it is not an account of heat but of something else. When there is a "strong" extensional divergence (not just the property's being present when the commonsense concept has no application, as with extreme degrees of heat, but when the concept applies when the property is absent, as with fish), the constraint is weaker. But still a biological account of a kind that did not imply its inclusion of a goodly share of the objects that fail under the everyday concept of fish would not be properly termed an account of the nature of fish.4 This kind of constraint clearly applies to truth. Any adequate account of the nature of truth must honor the minimalist account of the concept in terms of the T-schema. If, as I believe to be the case with epistemic accounts of truth, the class of what are recognized as true propositions on that account does not even come close to matching the class of propositions that satisfy the criterion of truth provided by the T-schema, this shows that the account is inadequate.-5

II Before coming to the main substantive part of this paper, I want to document my initial suggestion that many contemporary discussions of truth are hampered in their comparisons of what are commonly called "deflationary" or "minimalist" accounts and "substantive" accounts 6 by ignoring the possibility that the former have to do with the (ordinary) concept of truth and the latter with the nature of the property of truth. One nice example is David 1994, which is devoted to a critical comparison of Disquotational and Correspondence theories of truth. 7

4

5 6 7

In the kind of case stressed in Putnam 1975, in which the "stereotype" of the kind featured in the concept does not apply universally, or in extreme cases at all, the concept containing that stereotype can still serve as a guide to the property, because either the stereotype applies in familiar cases, or it seems to most people to do so. Hence the concept can direct us, more or less well, to the right class of entities. For a development of this critique of epistemic theories of truth, see Alston 1996, Ch. 7. For examples of this terminology see the authors referred to in this and the following section. To be sure, a disquotational theory, and any other that denies that there is any property of truth, has no position on whether the property of truth can go beyond the concept of truth. But here I am concerned only with David's way of setting up the general problem of truth and his consequent assumption that disquotationalism and correspondence theories are competitors in (at least partially) a common enterprise.

16

T h e Correspondence T h e o r y

The problem arises with Pilate's question "What is truth?" and it is tempting to say that it concerns primarily the nature of truth. But stating the problem in this way tends to beg the question against deflationism, for according to deflationary views, truth has no nature. Without begging the question against such views one can say that Pilate's question concerns the explanation of truth·, the problem is to explain what it is for something to be true or false. As I understand it, it is the task of a theory of truth to provide such an explanation. (7) It is natural to require from an adequate explanation of truth that it should provide us with a definition of truth. I shall, therefore, present theories of truth as at least aiming for a clause of the form χ is true =df. (8) Thinking of a definition of 'true' as giving an account of the concept of truth, we may say that David, by squeezing all "theories of truth" into such a format, ignores the possibility that while, e.g., disquotationalism is an account of the concept, correspondence theory is an account of the property, with the result that they are not alternative attempts to answer the same question. 8 Another mapping of the territory that freezes out my distinction is found in Wright 1992. After proposing a minimalist account of the concept of truth very similar to my version based on the T-schema, Wright considers the objection that this "fails to provide any substantive account of what truth is" (37). H e then continues: This line of objection betrays an important misunderstanding. It presupposes that minimalism is offering an account of the meaning of "true", in the traditional sense in which giving an account of the meaning of a word involves provision of an illuminating analysis of the concept it expresses...Traditional — "correspondence" and "coherence" - theories did hope for such an account. But minimalism has no such ambition...The proposal is simply that any predicate that exhibits certain very general features qualifies, just on that account, as a truth predicate. That is quite consistent with acknowledging that there may, perhaps must be more to say about the content of any predicate that does have these features. (37-38) Superficially this may look much like my suggestions that a correspondence theory can go beyond a minimalist account of the concept o f truth in such a way that they are compatible. But note that Wright's version o f this "going beyond" shares with what it goes beyond that it specifies features of a predicate. There is no recognition that it might consist of a description of a property, rather than of a term or concept. As a final exhibit consider Soames 1997. There in criticizing the minimalist account of the concept in Horwich 1990, again very close to my own, he writes: Our grasp of the property truth is not exhausted by our dispositions to accept T-propositions [roughly, instances of my T-schema] about ground-level propositions, which do not themselves contain the concept of truth. On the contrary, we find some T-propositions about "higher-level" (truth-attributing) propositions fully acceptable, even trivial; others we find puzzling and pathological; while still others are outright paradoxical. These reactions are not

8

To be sure, they w o u l d still not be compatible so long as disquotationalism is c o m m i t t e d to the denial that there is any property o f truth. But we might still say that this leaves r o o m for them to be compatible if disquotationalism would drop that denial and content itself with offering an account o f the concept.

Truth: Concept and Property

17

arbitrary. There is something about our notion of truth that guides them. N o account that leaves this out can be complete. (29-30)

Apart from the free substitution of notion for 'property' in the above, the point is that Soames takes the minimal account and what he thinks needs to be added to it to be in the same field of play. The larger context makes it clear that, in terms of my distinction, this field of play is the explication of a concept. Thus the property, as distinguished from the concept, is nowhere in sight. 9

Ill Thus far I have pointed out the general distinction between concept and property and have identified some areas in which it is clear that not every important feature of a property is represented in one or another concept of that property. The suggestion I extract from this is that something of the same sort could hold for truth, that there might be important, even fundamental features of the property of truth that are not captured by our ordinary concept of truth. And in the last section I have exhibited a few contemporary treatments that ignore this possibility. But all that the discussions up to this point indicates is a bare possibility. It remains to be considered what the prospects are of actualizing it. In addressing this issue I will take as my foil the views expressed in Horwich 1990. This is very well suited to my purpose because (1) Horwich's minimalist account of the concept of truth is very similar to mine, (2) unlike "deflationists" he too recognizes a property of truth, but (3) he argues against the view that there are important features of the property that go beyond the concept. Thus because of (1) and (2) Horwich and I are close enough to ground the possibility of a fruitful disagreement concerning (3). At the outset of his 1990 Horwich writes: Unlike most other predicates, 'is true' is not used to attribute to certain entities...an ordinary sort of property - a characteristic whose underlying nature will account for its relations to other ingredients of reality. Therefore, unlike most other predicates, 'is true' should not be expected to participate in some deep theory of that to which it refers - a theory that goes beyond a specification of what the word means. (2)

He goes on to say that the truth predicate exists solely to allow us to quantify in certain ways. If our truth predications were restricted to particular propositions one at a time, we could simply say 'The weather is cold' instead o f ' I t is true that the weather is cold'. But we often have occasion to say of a proposition that it is true without spelling it out as in 'What Jim just said is true'. And we often need to generalize over propositions, e.g., affirming whatever Jim believes or committing ourselves to any proposition of the form 'If ρ and q, then p'. The truth predicate gives us a handy way to make

9

One treatment that is not subject to this charge of massive assimilation is found in Kirkham 1992, Ch. 1. There he distinguishes six different "projects" concerned with truth. This scheme has more than enough room for my twofold distinction.

18

The Correspondence Theory

generalizations: 'Everything Jim believes is true' and 'If a conjunction is true, then so is each of this conjuncts', as well as to say that a certain proposition is true without spelling it out. And that's the whole story, says Horwich. The above quotation will not do as it stands, for it makes an unjustified claim about what predicates are ordinary. It restricts "ordinary predicates" to those that are used to attribute an property with an "underlying nature" that has powerful explanatory efficacy. But that just isn't so. Ordinary properties include congeniality, being a table or a chair, being a sophomore, being 5 feet tall, etc., none of which have an underlying nature with considerable explanatory or theoretical importance. But there are more serious issue involved than how to apply 'ordinary' to properties. Let's see what Horwich says by way of filling out the sketchy remarks above. According to minimalism, we should...beware against assimilating being true to such properties as being turquoise, being a tree, or being made of tin. Otherwise we will find ourselves looking for its constitutive structure, its causal behavior, and its typical manifestations - features peculiar to what I am calling ' c o m p l e x or 'naturalistic properties'. (39) Well, of course being true is very different from natural kinds like these. Here is a more extended and more sober argument. We can certainly entertain the possibility that the minimal theory is susceptible of explanation via some deeper account of truth. However there is excellent reason to suppose that in fact there is no such deeper theory. In the first place a scientific analysis of the usual sort is out of the question. For the axioms of the minimal theory are a priori, and therefore cannot be explained by a posteriori facts. In other words, any naturalistic reduction of truth - like the analysis of water as H 2 0 , heat as molecular motion, or gravity as curvature of space-time — would make an empirical claim about the coextensivity of truth and some property, F. Moreover, in order to yield the equivalence axioms [instances of the T-schema], this claim would have to be supplemented with further a posteriori claims of the form '(p) is F iff p'. And such empirical hypotheses could not be part of what explains an a priori theory. Secondly, the minimal theory of truth does not cry out for explanation in the way some theories do. Consider, for example, the account of chemical valence which consists in simply listing the valences of each element... In this case there is a reason to expect further reduction. For there are laws of nature about valence - laws about the relationship between the valences of elements and the proportions in which they combine - that are not explained by the list. And any lawlike generalization calls for explanation on pain of looking like a sheer coincidence. However the minimal theory of truth does not itself contain such laws... Thus there is nothing that should lead us either to expect or to desire a deeper explanation. (5051) There are further arguments, but I take these to be the more important ones. Note that Horwich is discussing the issue of whether there are important facts about the property of truth that go beyond the ordinary concept of truth, only in the form of the question "Is the minimal theory of truth susceptible of explanation via some deeper empirical (scientific) account of truth?". And, as we shall see, there can be forms of the former other than the latter. But as for Horwich's version of the issue, I am happy to concede that the minimal account of the concept of truth cannot be explained on the model of explaining the behavior of compounds in terms of their chemical constitution or explaining the laws of valence in terms of atomic structure. But if

Truth: Concept and Property

19

this is to be a strong reason for denying that the minimal account of the concept of truth can in any way be explained by a delineation of the nature of the property, along the lines of a correspondence theory for example, the additional premise is needed that the chemical analogy represents the only way in which a property can shed light on the concept of truth. And I shall provide reasons for rejecting that premise. More generally, by way of summary of the arguments we have found in Horwich on this point, they all consist in showing that the relation of the property and concept of truth fails to conform to certain familiar patterns of scientific explanation in which deeper lying features of a kind or a process explain more superficial features. And that leaves open the question of whether there are other ways in which facts about the property of truth might help us to understand the fact (basic to the minimalist account of the concept) that it is true that ρ iff p.

IV I now turn to the task of exploring ways in which a property Ρ might have important, even essential features that go beyond the ordinary or other concept of P. I will not aspire to give a complete catalogue of such features. It will suffice for present purposes if I cover both the way(s) that I think hold promise for the case of truth and the ways in which Horwich and others have pointed out not to be applicable to truth. Begin with the latter. Horwich concentrates on examples from scientific theorizing. It will be convenient to distinguish two groups into one or the other of which many of them fall. 1. First there is a deeper account of the nature of a kind of substance or of a state or process. Our ordinary concepts of a kind like gold or of a state like heat, turn out to be made up of relatively superficial features, many of them dispositional. Thus in Locke's classic account of the "nominal essence" of 'gold' (or, as he might just as well have said, of our everyday concept of gold), it is a body that is yellow, heavy, malleable, fusible, and soluble in aqua regia. But that is mere botanizing. A collection of such accounts of the natures of a large group of kinds of inanimate bodies would provide no general systematic theory of that division of nature. But when we get to the point of distinguishing the elements in terms of their atomic structures, we have an account of the natures of these kinds that constitutes a powerful general theory in terms of which we can explain an innumerable variety of surface features, as well as regularities in their behavior and interactions. We will have gone beyond the everyday concept of, e.g., gold by getting at the essence of the kind in terms of features not represented in the ordinary non-technical concept of gold. 2. In the above case we went beyond a concept of a thing or state that is in terms of relatively superficial, easily accessible features, by making explicit the underlying nature or micro-structure of that thing or state. These superficial features are often dispositional or functional. The second main way of "going beyond" the concept to reveal deeper features concerns dispositional and functional properties themselves. Lets say that the ordinary concept of solubility in water is the concept of some feature of an object, O, whereby if it is placed in water under conditions C, it will dissolve. We

20

The Correspondence Theory

might put this by saying that the concept contains a blank, and we go beyond the concept to something deeper by filling in the blank. To make the blank visible in the formulation, let's say that the concept of solubility is the concept of an property , such that if an object, O, is placed in water under conditions C, then if it possesses , it will dissolve. Or instead of the blank we could use a variable. It is the concept of a property, P, such that by virtue of having Ρ Ο is such that, if it were. . .The concept does not embody (represent) any more intrinsic nature of P. Ρ figures in the concept only as that by which Ο is disposed in a certain way. Obviously such a concept calls out for supplementation by an account of what property it is to which this relational description applies, where to answer this question we would need a (more) intrinsic characterization of the property, e.g. that it is a certain crystalline arrangement of molecules. This answer to the question would tell us something fundamental about the property that is not embodied in the common dispositional concept thereof. 10 A more complex example, which will be only sketchily presented here, concerns valences of chemical elements. The original concept of valence, and the one in terms of which it is still introduced to students, is a dispositional one. To say that sodium has a certain valence is to say that it has a disposition to combine with other elements, and compounds, in certain ways and in certain proportions to form compounds. But with the development of chemical research and chemical theory, it was discovered that these different reactive and combinatorial dispositions stemmed from differences in atomic structure. And so the property of having a certain valence turned out to be the property of exhibiting a certain atomic structure, which is construed in increasingly sophisticated ways at different stages of development. If we think of the original concept as the concept of some feature of an element by virtue of which it regularly combines with other substances in such-and-such proportions to form further compounds, then the theory of atomic structure tells us just what this feature is. Again it is a matter of filling a blank in the concept. Another important example is found in functional concepts. A functional concept is a concept of a thing, factor, state, process, or whatever, in terms of a function it performs. A loudspeaker is anything that has the function of converting electronic signals into sound. This function can be performed or "realized" by various sorts of structures and mechanisms. For each such way there is a deeper lying characterization of what makes the thing a loudspeaker, deeper than the functional characterization because the fact that the artifact is constituted in that way explains the performance of the function . Functional concepts have become prominent in the philosophy of mind and philosophical psychology, with the development of the idea that mental states and processes - belief, desire, intention, deliberation, attitudes - can be identified by the functions they perform in the psychic economy. Thus the belief that it will rain tomorrow

10 I don't mean to commit myself to the (surely mistaken) view that all dispositions are to be identified with an intrinsically characterizable property in this way. There can very well be some dispositions of fundamental particles that are not further reducible to intrinsic properties. But there are also many dispositions that fit the treatment given in the text, and it is of those that I am speaking.

Truth: Concept and Property

21

performs the function of, e.g., combining with an aversion to getting wet and a belief that one can avoid getting wet in the rain by carrying an umbrella, to yield a tendency to carry an umbrella if and when one goes out tomorrow. But, just as with the loudspeaker, what it is that performs the function is not represented in the purely functional concept. It is a further question whether that function is performed by some modification of an immaterial "soul" or psychic substance, or by some configuration of neurons or some neurophysiologies process in the brain. Again we find that a full story of the state or process in question requires us to go beyond the (initial) concept and specify the intrinsic nature of what performs the function. Now, as Horwich indicates, there is more than one reason why the property of truth cannot receive an extra-conceptual characterization in any of these ways. (1) Truth is not a substance, process, or state that has an underlying nature or micro-structure, or that undergoes micro-processes. (2) It is not a dispositional or functional property of anything that calls for a specification of the basis of the disposition or of the intrinsic nature of what carries out the function. (3) And, as Horwich points out, since the equivalencies that embody the minimal concept of truth are a priori (and, as I would say, analytic) truths, they are not susceptible of explanation in terms of an empirical theory. But this does not show that no further characterization is possible. It merely shows that the above ways of going beyond the concept do not provide a useful model for a further characterization of truth. But what other ways are there? Not surprisingly, philosophy will furnish a more appropriate hunting ground for models. To be sure, philosophy is such that we will not find conclusively established characterizations of the deeper nature of properties. But at least we can find promising, though multiple, hypotheses concerning the way in which one or another property displays features that go beyond what is encoded in the everyday concept of the property. Free action presents a nice example. Let's say that the commonsense concept of S's doing A freely at t, in the sense of that term that is common in philosophical discussions of free will, is the concept of S's doing A at t while being able to refrain from doing A. At the cost of some idealization we may say that this concept marks out the subject matter for the philosophical attempt to give a deeper characterization of what is involved in the ability to do otherwise. The "compatibilist" holds that this is simply the ability to do otherwise if one chooses (decides, wills) to do otherwise. But this does not satisfy the "libertarian" who holds that that this is not genuine freedom to do otherwise, unless one's doing otherwise is causally compatible with S's situation at A. Acting freely is a matter of having the power to refrain from doing A even if every influence on one's choice remains the same. And some compatibilists, in order to distinguish doing A freely from doing A's being a random event, add the stipulation that the act of doing A stems from a kind of causality, "agent causality", that differs from event causality. The doing of A is traceable only to S's deliberately exercising a power, not to any immediately preceding events that happen in or to S. As intimated above, I can't claim that one of these ways of going beyond the everyday concept of doing A freely is an established result of investigation on a par with the theory of atomic structure or the identification of heat with average kinetic energy of constituent molecules. O n the contrary, it is a matter of intense controversy what is

22

The Correspondence Theory

the right way of characterizing free action, and there is every prospect of its remaining so. But at least the controversy provides a clear example of two not horrendously implausible ways of doing that job. There are many other concepts that could be tapped for further examples — causality, meaning, reference, knowledge. For a quick example, take sentence meaning. Suppose the rudimentary concept that marks out the object of investigation is this. For a sentence to have a certain meaning is for it to have a certain content, to be standardly usable to convey a certain "message". But then the philosophical task is to determine what there is about a certain sentence that endows it with that content, that renders it so usable. It is clearly not any (relatively) obvious features — its phonological constitution or its grammatical structure. There are various attempts to answer this question. They include the following. (1) It is a certain function from contextual factors to a truth value. (2) It is the existence of a practice (or convention) in the society to use the sentence to affect hearers in a certain way. (3) It is the governance of the sentence by rules that specify conditions under which it is permissibly uttered. And there are many others, all of which go beyond anything that can plausibly be thought of as involved in an everyday, non-technical concept of what it is for a sentence to have a certain meaning. It is clear from these examples that philosophical ways of going beyond the concept differ widely from the examples from empirical theorizing that Horwich used to attack the idea that such "going beyond" is feasible for truth. Most crucially, they do not concern a natural kind or type of state or process that has an underlying nature we can find out about empirically. Nor is the enterprise of exploring the nature of the property one that involves the kind of empirical investigation that is typical of science. I will say more in a moment as to what is involved. But first I will point out a way in which these philosophical examples could be construed as exhibiting a more formal similarity with many of the scientific examples. This involves a feature that I pointed out above with regard to dispositional and functional concepts, viz., that the further account of the property involves filling in a "blank" or a "variable" in the concept. The concept of Ρ under discussion in those empirical cases is the concept of something which . . . But the intrinsic character of this something is left unspecified by the concept. What the concept spells out is something more external, more peripheral - a disposition, a function, a regular effect, a cause, or a position in some structure, sequence, or order. Thus if we think of the (ordinary) concept of elasticity as the concept of a property such that by virtue of having that property its possessor will stretch if pulled in a certain way, the further account of the property will specify its intrinsic nature, just what it is that gives a body that disposition. Again, the functional concept of a sense organ is the concept of something that is differentially sensitive to external stimulation and conveys the result of that differential sensitivity to some central processor. And the property of being a particular sense organ is further described by spelling out what kind of mechanism carries out this complex function. The philosophical examples can also be felicitously described in terms of the everyday concept's containing a blank that a further account of the property will seek to fill in. To act freely is to act in such a way that one has the power to act otherwise. And the further account of the property seeks to reveal what way that is. For a sentence to mean

Truth: Concept and Property

23

that the postman has just come is for it to be usable to, e.g., convey the information that the postman has just come. And the account of what it is to mean that is a specification of what makes a sentence usable to do that. But though there is this formal similarity, the investigation of the nature of the property will be carried out in quite different ways. Consider for a moment how a philosopher proceeds to look into what is involved in acting freely, once that phenomenon has been identified by the everyday concept thereof. She considers what careful reflection on the phenomenon reveals to be a necessary condition thereof. Or if that makes the procedure sound too susceptible of a generally agreed on outcome, we could say that she considers what hypothesis as to necessary conditions seems most plausible on careful review of everything that is relevant to deciding that issue, including of course what is entailed by the concept, but by no means limited to that. She asks herself whether one could still be said to have the power to do otherwise if the action was causally determined. And here is where there is a parting of the ways between compatibilist and libertarian. The former stresses a sense of'acting freely' on which there is a positive answer — having the power to do otherwise in the sense of being so disposed and situated that one would do otherwise if one decided to do otherwise — each of two or more choices would be effective if made. Nothing is preventing the subject from carrying out whatever choice is made. That is what being able to do otherwise consists in. But the libertarian points out that one doesn't really have the power to do otherwise unless one also has the power to choose otherwise. Without this latter power, the power the compatibilist insists on is a counterfactual power, one the agent would have if a certain condition were satisfied, where that condition is never satisfied. It's clear that these reflections lead in different directions. Philosophical reflection is notoriously bad at producing agreement among all practitioners thererof. But though the prospects for general agreement are dim, important aspects of the subject matter get revealed in the process. And however minute the prospects of definitive settlement of the issue, there is no doubt but that a complex intellectual enterprise is directed here to bringing out features of the property that goes beyond the initial concept that picks out the subject matter.

V I will now apply these general morals to the central issue of this paper - the question of whether there is a viable project of going beyond what is involved in a minimal reading of the ordinary concept of propositional truth (which I take to be the correct reading) to a further, deeper, more theoretical account of the nature of the property of truth. In the first part of this paper I canvassed a number of philosophers who take it that there is nothing to be said about truth beyond an analysis of the concept. The deflationists among them hold this because they deny that there is any property of truth to serve as the object of such an account. Others, like Horwich, are prepared to recognize a property but deny that anything important can be said about it that goes beyond the account of the concept, on the grounds that nothing analogous to a scientific theory is applicable here. But now that I have widened the sights to include distinc-

24

T h e Correspondence T h e o r y

tively philosophical accounts of properties that go beyond (everyday, nontechnical) concepts thereof, the possibility suggests itself of sidestepping Horwich's prohibitions and looking for a distinctively philosophical way of exploring the extra-conceptual nature of the property of truth. Even if such an exploration were to closely follow such "essentially contested" issues as the nature of free action or the nature of linguistic meaning, I believe that we could still make a convincing case for its viability, albeit without firm hopes of a definitive resolution. But it seems clear to me that the prospects are much brighter here, at least so long as we work within the minimalist reading of the concept of truth set out, in somewhat different ways, by Horwich and myself. To remind the reader of this, the idea is that the ordinary concept of prepositional truth is embodied in the T-schema: (1) The proposition that ρ is true i f f p. I now want to suggest that not only is it a sensible project to ask how the property of truth that is identified by this concept can be further characterized , but that one of the traditionally competing answers to this question is the one that is naturally suggested by the concept. Here the concept gives us much more guidance in delineating the property than is the case with our earlier philosophical examples. To begin the task of showing this, let's first note that the T-schema, (1), is substantially equivalent to (2) The proposition that ρ is true i f f it is a fact that p. The consequent of (2), 'it is a fact that ρ would seem to be necessarily equivalent to the consequent of (1), '/>'. "Lemons are yellow' is surely necessarily equivalent to 'It is a fact that lemons are yellow'. But if the fact that lemons are yellow constitutes a necessary and sufficient condition for it's being true that lemons are yellow, that puts us in a position to speak of the fact that lemons are yellow as a "truth-maker" for the proposition that lemons are yellow. That proposition is true just because it is a fact that lemons are yellow. So now the question becomes: How does a proposition have to relate to a fact in order that that fact be a truth-maker for the proposition? 11 Note that our minimal concept of truth embodies the identity of the propositional content of a statement or belief and the fact the obtaining of which would render it true. This stems from the use of the same propositional variable on both sides of the equivalence: It is true that ρ iff p. In saying this I am not suggesting that facts are to be identified with propositions, or with true propositions. In calling the content of a fact "propositional", I only mean that it is, so to say, propositionally "shaped". This is brought out by the fact that we use a propositional phrase to specify a particular fact — the fact that walnuts are hard to crack or the fact that Joan is thin. This naturally leads to the suggestion that the

11

N o d o u b t , at this point various difficulties will be raised concerning the ontology o f facts, but let's ignore them for the m o m e n t . O r rather, lets absorb them into what would have to be dealt with to get an adequate answer to the initial question.

Truth: Concept and Property

25

proposition-fact relationship that renders the proposition true is some sort of equivalence of the "content" of the two, that which makes the proposition the particular proposition it is and which makes the fact the fact it is. So long as we do not identify facts and (true) propositions, this can't be strict identity. In speaking above of an "identity "of content, I was engaging in a hyperbole to give an initial rough idea. But there must be some intimate "correspondence" between the two contents, intimate enough for one to be "read" off the other. Thus we have, under the guidance of the minimal concept, arrived at the threshold of a correspondence theory of truth. There are various paths we might take after stepping over the threshold. The most common one is to look for some sort of isomorphism between proposition and fact, involving a one-one correspondence between constituents and some kind of identity of structure. Just how this works out depends on how we think of propositions and facts, and here we come to apparently unending controversies that are analogous to our earlier philosophical cases. If the proposition Joan is thin is made up of concepts - an individual concept of Joan and the concept of thinness, bound together by a relation of inherence (property exemplification), then a fact to which this proposition corresponds would consist of what these concepts are concepts of, viz., Joan and thinness, bound together by the (same? corresponding?) relation of exemplification. Other accounts of the nature and structure of propositions and of facts would suggest other ways of spelling out the initial suggestion of some kind of isomorphism. It is obvious that I am not working out, much less defending, a particular version of correspondence theory in this paper. My aim is only to argue the following points. (1) The pursuit of such a theory is a serious project, one that, contrary to widespread contemporary views, is not fated to fail by the nature of the case. (2) By distinguishing between the (ordinary) concept of truth and aspects of the nature of the property of truth that go beyond that concept, we can reconcile a correspondence theory of truth with a minimalist account of the concept of truth that recognizes it to be innocent of any involvement with spelling out what is involved in proposition-fact correspondence. We can achieve this by taking the correspondence theory to be an account of the property, not of the concept. By ignoring this possibility and supposing that everything that can be said philosophically about truth concerns the concept, the philosophers cited in section ii have distorted the situation. (3) My third basic point is that a correspondence theory differs from, e.g., a coherence or pragmatist or other epistemic theory of truth by being a natural way of further spelling out what is (only) implicit in the minimal concept. My other philosophical examples of an armchair way of going beyond the concept to develop a theory of the property were not so fortunate in that respect. With free will and linguistic meaning, it is not clear that one theory, rather than its rivals, is definitely indicated by the everyday concept. There too we can think of the philosopher as seeking to make explicit what is implicit in the concept, but there are radically different ways of doing this, no one of which can plausibly claim to be uniquely marked out by the concept. With truth, on the other hand, I have just shown how, by natural, intuitively plausible steps one gets from the minimal concept of truth to a correspondence account of the property. And no such natural progression can be found for the other traditional theories of truth.

26

The Correspondence Theory

One would need a strong argument for the thesis that the truth of the proposition that grass is green could be conceptually equivalent to grass's being green only if truth consists of membership in an ideally coherent system or consists of justifiability in an ideal epistemic situation. And, to my knowledge, no such argument has been proffered. And even if it were, it would not offer the clear, natural, and plausible route from the concept to the account of the property that we have with the correspondence theory. With respect to the basic theses listed in the last two paragraphs, I have obviously given much more support to (2) and (3) than to (1). And for good reason. (1) could be supported to the same extent as (2) and (3) only by tackling all the thorny issues involved in working out a correspondence theory, as well as critically comparing different versions thereof. And there is no space for that in this paper. Hence I must content myself here with showing how to reconcile a correspondence theory of truth with the minimalist account of the concept of truth based on the T-schema, and merely suggesting that the possibility thus opened up is one that it is worthwhile to seek to actualize.

Bibliography Alston, William P. 1996, A Realist Conception of Truth, Ithaca: Cornell University Press David, Marian, 1994, Correspondence and Disquotation, New York: Oxford University Press Fodor, Jerry, 1998, Concepts, New York: Oxford University Press Horwich, Paul, 1990, Truth, Oxford: Basil Blackwell Kirkham, Richard L. 1992, Theories of Truth, Cambridge, MA: MIT Press Peacocke, Christopher, 1992, A Study of Concepts, Cambridge, MA: MIT Press Putnam, Hilary, 1975, "The Meaning of'Meaning'", In Language, Mind, and Knowledge, ed. Keith Gunderson. Minneapolis: University of Minnesota Press Soames, Scott, 1997 ,"The Truth About Deflationism", In Truth, ed. Enrique Villanueva. Atascadero, CA: Ridgeview Pub. Co. Weitz, Morris, 1988, Theories of Concepts, London: Routledge Wright, Crispin, 1992, Truth and Objectivity, Cambridge: Harvard University Press

Truths and Truthmakers DAVID. M . ARMSTRONG

I. Introduction I introduce the topic autobiographically. I first learnt of the notion of a truthmaker from C.B. (Charlie) Martin. The time was the late fifties, and he was a lecturer at the University of Adelaide. I was at Melbourne University. At the time we were both thinking about the doctrine of Phenomenalism, the claim that physical objects are constituted out of sense-data or sense-impressions. Neither of us had any sympathy for this view, but it was in the air at the time. The question for us was how it was best argued against. Phenomenalists had a problem about physical objects and events at times that they are not being perceived. The solution to the problem generally given is to be found in embryo in Berkeley and became firm doctrine in John Stuart Mill. It involved an appeal to certain contingent counterfactual truths. Contingent counterfactual claims are often to be found in ordinary discourse, for instance, 'If you had not put your foot on the brake so promptly just then, there would have been a nasty accident.' There can be rational discussion of such claims, and they can be true as well as false (or at least assertible and non-assertible). Perhaps, then, an account can be given of the physically unobserved in terms of what sort of perceptions would have been had if, contrary to fact, a suitable perceiver had actually perceived them. In Mills wonderful phrase, a physical object became a mere 'permanent possibility of sensation. Many prima facie difficulties for this line of defence using counterfactuals were known. But Martin asked a simple question that seemed to go to the heart of the problem. Suppose that the required counterfactual propositions are indeed true. What are the truthmakers for these truths? Must there not be some way that the world is in virtue of which these truths are true? What is it? How does the world make these truths true? A Realist about the physical world will have no difficulty in answering Martin's question. Berkeley had an answer, even if an obscure and difficult answer, in the archetype of the world that he supposed to exist in the eternal mind of God. A Realist about unfulfilled possibilities would perhaps have an answer. But what answer had the actual Phenomenalists got? All they had available for truthmakers were the actual sense-data or sense-impressions had by actual minds. Truthmakers for true counterfactuals about the perception of unobserved material reality would have to be found in the actual, bitty, sense-data. As a result, the unobserved physical reality cannot, for the Phenomenalism be what we all think it is in our unphilosophical moments: something ontologically additional to the observed portion of physical reality. A bad enough result, one would think. But worse follows. Consider a physical world without any minds in it. That seems to be a possibility, indeed, in view of the

28

The Correspondence Theory

delicacy of the initial conditions under which life evolved, it seems to be a physical possibility, one compatible with the actual laws of nature. What can the Phenomenalist say about such a world? Every physical truth about individual objects and processes must be given a counterfactual analysis in terms of perceptions not actually had. But what truthmakers in that world will there be for these truths? None, it would seem. The world is empty of perceptions and the minds that have them, so it is empty, period. So for a Phenomenalist there cannot be a physical world empty of minds. I do not want to claim that these arguments are absolutely conclusive against Phenomenalism. I deny that there are such arguments in metaphysics. In the present case, for instance, an Idealist, such as the contemporary Oxford Idealists John Foster and Howard Robinson, might even welcome them. But I claim that they are very powerful arguments, considerations that, in Mill's phrase, are capable of determining the intellect, and that they show, in a particular case, the power of the appeal to the need for truthmakers for accepted truths. Let us turn from Phenomenalism to Gilbert Ryle's account of the mind. Here I think that I grasped the point for myself, after having been instructed by Martin in the case of Phenomenalism. As is well known, Ryle bolstered his quasi-behaviouristic account of mental states, events and processes in The Concept of Mind by continual reference to dispositions. A mental state, such as a belief, he saw as fundamentally dispositional. It is a mark of dispositions that they need not be manifested, perhaps at any time during the existence of the thing that has the disposition, even although the physical possibility of that manifestation is involved in the very notion of a disposition. The brittle thing may never break, the elastic thing need never be first stretched and then allowed to return to its previous unstretched state. Similarly, a person might hold a belief, but never manifest that belief in behaviour during the whole of life. No problem for the Rylean account of mind. Unmanifested beliefs are no more than unmanifested dispositions. So, I think, Ryle saw it. But he could only so see the matter because he was working in a philosophical climate that saw little need to take up metaphysical questions, and in particular no need to consider the question of the truthmaker for dispositional truths about minds. I think he was quite right to claim an essential role for disposidonality in the elucidation of our notion of the mental. That was a great and lasting contribution. But we need then to go on to consider the question of the truthmaker for these dispositional truths. What is there in the world that makes these truths true? Ryle has no answer. Once we do raise the truthmaker question, then our view of the nature of mind will very likely be transformed and we will move in a quite un-Rylean direction. We will very likely identify a belief, say, with some inner state of the mind (materialist metaphysicians will identify it further with some state of the brain) which, in suitable circumstances, but only in suitable circumstances, will manifest itself in various ways, some of which ways may be outward behaviour. Of course, even if under the influence of the truthmaker question we do 'move inside' to the brain (or the soul), there will be plenty of room for disagreement about the exact nature of the inner state that should be postulated. For myself, I incline to a categorical state, a state involving non-dispositional properties, and, as I now under-

Truth and Truthmakers

29

stand the matter, a state that requires to be supplemented by the relevant laws of nature. (The laws of nature, in turn, cannot be mere true statements, but must be conceived ontologically.) Martin thinks of the state required as having a categorical 'side' but as also involving powers, powers that are not reducible to the categorical, and which serve as his substitute for laws of nature. Others take subtly different views. But the truthmaker insight, as I take it to be, prevents the metaphysician from letting dispositions 'hang on air' as they do in Ryle's philosophy of mind. For one who espouses truthmakers, such hanging on air is the ultimate sin in metaphysics.

2. The general theory of truthmaking 2.1. Historical. The notion of the truthmaker may be traced right back to Aristotle. (See, in particular, Categories, 14b, 14-22.) Aristotle's remarks were noted by a number of leading Scholastic philosophers, but the notion seems after this to have gone underground for some centuries, although intimations of it may be found here and there. The notion is present in Russell's thought, and in his later philosophizing he introduced a term, the somewhat unfortunate word 'verifier' (Russell, 1940, 1948, 1959). 1 Reference to truthmakers, and some development of truthmaking theory, is now quite widespread among philosophers working in Australia. I think that the source is always C.B. Martin, as certainly it was for me. But the very same notion, and the very same term, was introduced quite independently by Kevin Mulligan, Peter Simons and Barry Smith in a joint article 'Truth-Makers' published in 1984. They provide a suggestive quotation from Husserl, and mention Russell and the Tractatus Wittgenstein. 2.2. The Truthmaking relation. A truthmaker for a particular truth is some existent, some portion of reality, in virtue of which that truth is true. Philosophers and others who agree that a certain proposition is true may still be found disagreeing about what is the truthmaker for this supposed truth. I think that the truthmaker must necessitate, absolutely necessitate, any proposition for which it is a truthmaker. For consider a putative truthmaker Τ for ρ which exists, yet does not necessitate p. Putting the matter in terms of possible worlds, there will be worlds in which Τ exists but ρ is false. This means that there will be some further condition which must be satisfied which, together with the existence of T, ensures the truth of p. But will not this further condition have its truthmaker? If so, should not that truthmaker be added to Τ to give the true truthmaker for p? Here, however, I am assuming that every truth has a necessitating truthmaker, and, as we shall see, that assumption is controversial. How to spell out the necessitation, to say what its formal properties are, is also somewhat tricky. But I will leave that matter aside here.

1

I am indebted to the late George Molnar for pointing this out to me. Russell's later work has been amazingly neglected.

30

The Correspondence Theory

2.3. Truths. Those who have been trying to develop truthmaking theory have not said very much about the other terms of the truthmaking relations: truths. I wont say much here either. The ontological status of truths is not entirely clear. Entities that clearly exist are statements, on the linguistic side, and beliefs, judgments and thoughts, on the mental side. All these entities would seem to have the relational property of being true or being false, or, if we want to allow for truth-value gaps, being the sorts of things that are capable of truth or falsity. So perhaps we can say that truths are those statements, beliefs and so on, that have the relational property of being made true by some existent. Notice, however, that what is necessitated by a truthmaker is not the statings and the belitvings - that would be absurd - but only what is stati?^ or believed, the proposition stated or believed. The predicate 'true' applies properly to propositions. There is at least one problem, though. It seems that we want to countenance the notion of truths that, in the whole history of the world, are never expressed either verbally or mentally. We can call such truths 'truths without bearers'. Can we say that the notion of such a truth is analyzable as referring to the mere possibility of the corresponding statings, believings or whatever, the mere possibility of bearers? This pushes off the responsibility for giving an account of this sort of truth onto the theory of modality. I hope that this does not involve some hidden difficulty. 2.4. Truthmakers as correspondents for truths. Are truthmakers for truths just the 'correspondents' envisaged by the Correspondence theory of truth? I think the answer to this is a qualified 'Yes'. Truthmaker theory is a correspondence theory, but it does not burden itself with the dogma that the correspondence has to be one-one. This point was not, I think, ever explicitly stated by Russell, but he understood it well. The freeing-up that results from jettisoning the one-one dogma breathes new life into the Correspondence theory. A useful parallel here is with the relation between predicates and properties. In the past, many philosophers have assumed that to each different predicate (type, not token, is meant here) its own unique property corresponds. It is now quite widely realized that this view needs to be qualified. There is a conception of properties - the 'abundant' properties in David Lewis' terminology - for which this simple correlation holds. But there is another, metaphysically more important conception, of properties — called the 'sparse' properties by Lewis - for which this correlation breaks down. These sparse properties (and relations), many of us would argue, are not to be postulated a priori on the basis of semantic considerations, but a posteriori on the basis of our best science. Predicates may apply in virtue of a disjunction of properties, 'game' being a possible example, and two different predicates may apply in virtue of a single property, for instance 'inertial mass' and 'gravitational mass'. In the same general sort of way, truthmaker theory can happily accept that a single truth can have a multiplicity of truthmakers. It is a truth that at least one human being exists. Given an omnitemporal sense of exists', every human being that existed, exists now, or will exist is a truthmaker for this truth. Similarly, it seems, more than one truth can have the very same truthmaker. 'This surface is coloured' and 'this surface is red' seem to be made true by the very same surface.

Truth and Truthmakers

31

Notice, also, that the one truthmaker will make true a huge number of truths. A trivial example is that a truthmaker for p will equally be a truthmaker for any disjuntive proposition that has ρ as one of its disjuncts. 2.5. Minimal truthmakers. Suppose p to be a truth and Τ to be a truthmaker for p. There may well exist a T ' that is contained by T, and a T " that contains T, with both T ' and T " truthmakers for p. We may say that truthmakers for a particular truth may be more or less discerning. The more embracing the truthmaker, the less discerning it is. For every truth, the least discerning of all truthmakers is the world itself, the totality of being. The world makes every truth true, or at least every truth that has a truthmaker true. It is necessary, though, to think carefully about these relations of inclusion among truthmakers for a particular truth. The obvious relation to identify it with is mereological inclusion, the simplest relation of whole and proper part. It will be helpful, though by no means essential, to subscribe to the doctrine of Unrestricted Mereological Composition, the thesis that any plurality of entities, however heterogeneous, is a mereological whole. Then one can always have a single object as truthmaker. Mereology, however, may not be all that is needed. Some of us think that there are in the world facts (states of affairs), entities having such forms as as being F and a's having R to b. It is widely appreciated that these entities, if they exist, have a non-mereological form of composition. Yet the state of affairs of as being F seems clearly to include, have as constituents, the particular a and the property F. Furthermore, it seems that constituents can be truthmakers just as much as states of affairs. The notion that truthmakers for a particular truth may include truthmakers for that same truth, whether the inclusion be mereological or otherwise, raises the question of minimal truthmakers for that truth. A minimal truthmaker is a truthmaker that makes a particular truth true, but where there is no further truthmaker for that truth which is a proper part of the first truthmaker. If Τ is is a minimal truthmaker for p, then you cannot subtract anything from Τ and the remainder still be a truthmaker for p. Suppose, making some controversial assumptions for the sake of the example, that the truthmakers include properties, that these properties are universals, that universale are contingent existences, and that having rest-mass of one kilo is one such property. It is clear that this property is a truthmaker for the truth that this property exists. What is more, it is surely a minimal truthmaker. It will be seen that minimal truthmakers are of special interest and importance in truthmaking theory. Notice that a truth may have more than one minimal truthmaker. If it has only one, as is the case for the truth that the property having rest-mass of one kilo exists, then the truth may be said to have an unique minimal truthmaker. 2.6. Truths without minimal truthmakers. Is it the case that for every truth there exists at least one minimal truthmaker? It sounds very plausible, but Greg Restall has pointed out in unpublished work that, provided that the world contains at least a denumerable infinity of entities, there are truths that have no minimal truthmaker. Suppose, for instance, that the world contains a denumerable infinity of electrons, and consider the truth that there is an infinity of electrons. The totality of electrons is a truthmaker

32

The Correspondence Theory

for that truth. But consider every third electron. That totality makes the very same truth true, because it still contains an infinity of electrons. And so for the nth. electron, where η is any finite natural number. Hence a minimal truthmaker is never reached. Any infinity in nature will reproduce situations of this sort for certain truths. With this exception, though, it seems that any truth that has a truthmaker will have at least one minimal truthmaker. 2.7 The Transitivity principle. Suppose that Τ is a truthmaker for p, and that ρ entails q. Is it the case that Τ is then the truthmaker for g? If 'entails' here is classical entailment, and ρ is contingent, then ρ entails every necessary truth. Hence any contingent existent will be a truthmaker for all necessary truths. But if, as I do, we hope to find more or less discriminating truthmakers for necessary truths, this a discouraging result. But a more restrictive version of the Transitivity principle may be true. Frank Jackson has suggested that if ρ and q are both of them contingent, then Transitivity can be upheld (see Restall, 1996). To this Restall objects that the trouble is still there. He argues as follows. Let C be a contingent truth, s a truthmaker for C , and Ν a necessary truth. C & Ν is a contingent truth, but, given classical entailment, C entails C & N. Hence, given the Jackson Transitivity, s is a truthmaker for C & N . Restall then uses a (controversial) Conjunctivity priciple: if a truthmaker makes a conjunctive truth true, then it makes true each conjunct. So s is still a truthmaker for N . Can we save Jackson's idea by introducing the notion of a purely contingent truth, one that is not a conjunction of a contingent truth and a necessary truth? If so, Jackson's thesis can be restricted to 'purely contingent' truths. This will amount to saying that one of the entailments of C, its entailment of any N , is not to be allowed in applying the Transitivity principle within truthmaker theoretic arguments. Rather ad hoc, but perhaps it can be allowed. Restall himself, in his paper, argues for something more radical. H e interprets truthmaker necessitation itself as entailment, but as a restricted entailment, an entailment close to, but not quite as restricted as, relevance entailment. At any rate, if Jackson transitivity can be defended for contingent truths, this may cast light on the old, and attractive, idea that in a valid argument the conclusion is in some way included in its premiss. In such an argument, given true contingent premiss and purely contingent conclusion, it seems that any truthmaker for the premiss will include the truthmakers for the conclusion. And if the argument is valid, but its premiss false, the same result will obtain in 'worlds where the premiss is true'. 2.8. Does every truth have at least one truthmakeri I answer 'Yes' to this question, a position that may be called Truthmaker Maximalism. Some truthmaker theorists, however, are not maximalists. They may deny that contingent truths have truthmakers in the cases where they are negative and where they are general. Again, they may allow anything at all to be a truthmaker for necessary truths, thus robbing truthmaker theory of all interest in this sphere, and at least contradicting the spirit of Maximalism. In what follows I will sketch at least the outline of a robust maximalism in these disputed areas.

Truth and Truthmakers

2.9. The postulation

33

of truthmakers contrasted with 'quantifying over'. To postulate

certain truthmakers for certain truths is to admit those truthmakers to one's ontology. The complete range of truthmakers admitted constitutes a metaphysics, which alerts us to the important point that the hunt for truthmakers is as controversial and difficult as the enterprise of metaphysics. I think that proceeding by looking for truthmakers is a most illuminating and useful regimentation of the metaphysical enterprise, or at least the enterprise of realist metaphysics. But it is no easy and automatic road to the truth in such matters. But this raises the question of Quine, and the signalling of ontological commitment by what we are prepared to 'quantify over'. W h y should we desert Quine's proceedure for some other method? The great advantage, as I see it, of the search for truthmakers is that it focusses us not merely on the metaphysical implications of the subject terms of propositions but also on their predicates. For instance, the doing of ontological justice to the predicate leads us to consider whether we do not require at least selected properties and relations in our ontology. If properties and relations are admitted, we may think that some ontological connection between subjects and predicates is required, and thus, perhaps, be led to postulate facts or states of a f f a i r s among our truthmakers. The prepositional nature of truths will in any case push us in the same direction. The existence of negative truths and general (universally quantified) truths raises the question whether negative and general facts are required as truthmakers. All these difficult metaphysical issues tend to be swept under the carpet by correlating one's ontology with the subject term only of truths (or what one takes to be truths). Some may argue that what I see here as advantages of thinking in terms of truthmakers are actually disadvantages. The world is a world of things not of facts, it may be said, and so we do not want facts, and the nightmare of such entities as negative facts, in our ontology. This is an arguable position, although I reject it, but even if it is correct it can be accomodated by a doctrine of truthmakers. Let the world be a world of things. The fundamental truths will then have the form 'X exists' and the Xs, whatever they may be, will be truthmakers for these truths.

3. Some particular issues The really interesting questions, of course, arise when we come to consider in detail what truthmakers we should accept. But I think it is valuable to separate, in some degree at least, the general theory of truthmaking from particular postulations. The previous section 2 was intended as an introduction to the general theory. As already indicated, the question what truthmakers we should accept is the question what our ontology should be. I take up a few issues here, and these issues briefly only.

3.1. Properties, relations and states of affairs. We shall probably not want to treat every predicate that is predicated of an object with ontological seriousness. Consider, for instance, the predicate 'identical with itself. The object O, like any other object, will be identical with itself. The object itself seems all that is needed for truthmaker for this

34

The Correspondence Theory

predication. But suppose Ο has a mass of five kilograms exact. What is the truthmaker for this predication? Nominalists may point to O's membership of the class of five kilogram objects, and/or to the fact that the predicate 'five kilograms exact' applies to the object, or to the resemblance that all five kilogram objects have to each other. The matter deserves much lengthier consideration, of course, but intuitively the truthmaker is to be found in the object itself and not in any of these external matters. (I deliberately chose an example which is, prima facie, a non-relational property of O.) What then of Ο itself as truthmaker? It is indeed a truthmaker for the object's being five kilograms in mass, but is it a minimal truthmaker? We have to remember that Ο will have all sorts of other non-relational properties besides its mass: for instance, shape, size, temperature. Experiment will reveal that Ο acts causally (and differentially) in virtue of these different properties. It may depress a scale in virtue of its mass, burn fingers in virtue of its temperature. This suggests that we need selected properties in re as truthmakers for these predications. And since these properties must be properties of O, it will be quite plausible to go on to postulate facts or states of affairs, such as O's having five kibs mass as truthmakers for the truths involved. A similar line of thought leads to the postulation of relations, in particular the external relations of things, where the relations are not determined by the non-relational nature of the related things, and then states of affairs involving these relations. This is the briefest of sketches, and even if fully developed would not yield more than plausibilities, but I hope it shows how thinking in terms of truthmakers may direct one along certain paths in metaphysics. 3.2. Negative and general facts. One of Russell's great achievments in the lectures on Logical Atomism (1918) was to bring this problem into sharp relief. It is true, let us say, that it is not raining now, or that every mammal in the room is a human being. But what are the truthmakers for these truths? Suppose it is said that the truthmakers are, respectively, the present state of the environment and the collection of the mammals in the room now. Are we then referring to the present positive state of the environment, or do we include the absence of rain? Are we referring to the mere collection of the mammals, or, in the second case, to the fact that they are all the mammals? Many shrink from the second style of answer in both cases. They do not want to countenance negative or general states of affairs. But then they face a problem. The mere positive states of the environment do not necessitate the absence of rain, the collection of the mammals does not necessitate that they are the only mammals in the room. (In the first sort of case, something can perhaps be done by exclusion. That a surface is red necessitates that it is not green. But, even abstracting from other difficulties, not all cases of negative truths can be dealt with in this way.) A very unpleasant choice then looms. One must either settle for contingent truthmakers in such cases, or else deny that these truths have truthmakers at all. For myself, I go with Russell and draw the moral that negative truths and general truths must have truthmakers that necessitate these truths. I do think that Russell may have gone too far in admitting both negative and general facts. I think that we can get along with general facts alone, which I call totality states of affairs. But they are species

Truth and Truthmakers

35

of negative fact. That just these are all the mammals in the room is the fact that there are no more mammals in the room besides these. Here, briefly, is the argument for admitting totality states of affairs alone. Consider all the positive states of affairs that there are, the huge collection that exhausts positive reality. That collection fails to necessitate the state of affairs that it exhausts positive reality. But now add just one state of affairs: the (huge) fact that besides these states there are no further positive realities. Do not all further negative states of affairs supervene? And if this supervenience holds can we not say that all negative truths are necessitated? There remains the question what is the exact form of these totality states of affairs, as I call them. My suggestion is that it is a relation (I will here call it Tot, since Τ has already been appropriated for truthmakers) holding between an aggregate and a property. 'Property' is used here in its abundant sense, and the aggregate is a mereological whole. An aggregate, say the aggregate of all the electrons past, present and future, may be said to total a certain property, in this case the property of being an electron. The same relation holds between the aggregate of mammals in the room now and being a mammal in the room now. In general, at least, the Tot relation is contingent. See my 1997. Objections have been made to this account of totality states of affairs which I will not try to answer here (see Cox 1997). But I offer the following recommendation for the account. The relation I point to appears to exist. It holds between the aggregate of the electrons and being an electron, and so for other cases. If it does exist, then why might not the holding of this relation be the truthmaker for truths of totality? Opponents, I think, should be arguing that there is no such relation.

3.3 Modal Truths 3.3.1. Truths of possibility. If it is possible that p is true, then p itself may be necessary, or not necessary but still true. But for what we may call the mere possibilities p is false, though only contingently false. The mere possibilities seem to be the central cases of possibility. And it is for them that strange entities are postulated as truthmakers. Let not-p be contingently false, p will be contingently true. By Truthmaker maximalism, p will have at least one truthmaker. Given that p, and given that p is contingent, then it is necessitated that it is possible that not -p. This suggests that a transitivity principle of truthmaking may apply. The truthmakers for the two premisses 'p' and 'p is contingent', if taken together, are, I suggest, truthmakers for 'not-p is possible'. The latter truth is, presumably, itself necessary, so this will not be a case of Jackson transitivity. But it looks to be good candidate for transitivity, nevertheless. The truthmaker for ρ will be some contingent existent, C. That it is contingent is a necessity. So the truthmaker for 'it is possible that not-/»' is C plus whatever is the truthmaker for the necessary truth 'C is a contingent existent'. We have not yet discussed truthmakers for necessary truths, but it is plausible that the truthmaker in this particular case is no more than C itself. If so, a truthmaker for a contingent truth is also a truthmaker for the possibility of the truth of its contradictory.

36

The Correspondence Theory

If all this is correct, it seems clear, a fortiori, that a truthmaker for the contingent truth p will be a truthmaker for 'it is possible that p\ And if ρ is necessary, then whatever its truthmaker is ought to be a truthmaker for 'it is possible that p\ Plausible considerations in truthmaking theory seem to be delivering us a metaphysics of possibility. It is somewhat deflationary, but not as deflationary as making anything at all the truthmaker for necessary truths, 3.3.1.1. Aliens. The above treatment of truthmakers for truths of possibility, together with totality states of affairs (Russell's general facts), allows us to give a simple treatment of 'alien' possibilities. The term alien was introduced by David Lewis. An alien is for Lewis a property, object, or other entity which is not a being of this world, but is likely to be found in other worlds. For myself, as a one-worlder, there are no aliens, they are mere possibilities. But aliens, to be aliens, obey a further condition. Not only are they absent from the world, but they are not accessible by any combinatorial operation upon the entities of this world, that is, by any process of recombination of, reduplication or interpolation. (In using the word 'interpolation' I have in mind Hume's 'missing shade of blue'.) In the nature of the case no instance can be given. But, to take one case, suppose that there are simple properties in our world. An alien simple property would be a property quite other than these properties. For a one worlder there cannot be any aliens. But, perhaps contradicting Wittgenstein in the Tractatus, it does seem to be true that there might have been extra simple properties. The problem is to find a truthmaker for this apparent modal truth. The solution now to be given will serve, mutatis mutandis, for all aliens. The first thing required for the truthmaker is the collection of all the actual simple properties. But then is required the states of affairs that these are the totality of the simple properties. This is presumably a contingent truth. (If it is necessary, alien properties are impossible.) But given the view of mere possibilities given above, this particular totality state of affairs is the truthmaker for the possibility that the collection of all the simple properties is not all the simple properties. 3.3.2. Truths of Necessity. Consider the truth that 7 + 5 = 12. I suggest that the numbers 7, 5, and 12 will do as truthmakers here. We do not need + and = for the following reason. A triadic relation holds between the three numbers which may be symbolized thus: + = . This relation is an internal relation, that is to say, it holds in virtue of, and is necessitated by, the nature of the terms. But if there is this necessitation, what need of any truthmaker except the terms? Given the terms, the relation must hold, so can we not argue that we need only the terms as truthmakers? My hope is that this sort of analysis will apply to all necessary truths. (Including, of course, the necessary truths of truthmaker theory.) In each case the true proposition will be true in virtue of the entities that it involves. I don't know if every necessary truth sets up an internal relation between entities, or perhaps between an entity and

Truth and Truthmakers

37

itself. That would make things simple. But a necessary truth has as truthmakers just the entities it deals with. There is a contrast here with contingent truths, such as an object having a certain mass. T h e mere object and the mass are not enough: the object must instantiate the mass. This, unlike the case of necessary truths, demands states of affairs for truthmakers. This adumbration of a theory of truthmakers for necessary truths is not so deflationary as the account given of truthmakers for contingent possibilities. It would be compatible, for instance, with a Platonist view of the numbers 5, 7, and 12. But suppose we take what I think is the better view that these natural numbers are properties of classes, properties of five-numbered, seven-numbered and twelve-numbered classes. (We should reject Russell's view that they are the class of these classes.) T h e n truthmakers for the existence of these numbers are secured with relative ease. And where necessary truths deal with entities that do not exist (centaurs and suchlike), or entities that lie under suspicion of not being instantiated (very big infinite numbers), perhaps combinatorial principles will yield truthmakers among really existing entities. This account of truthmakers for nececessary truths, if satisfactory, enables us to go back to the necessary truth ' C is a contingent existent' where C is the truthmaker for contingent truth p. It seems that we needed a truthmaker for this truth in order to have a truthmaker for the modal truth 'it is possible that not-p'. ' C is a contingent existent' will, because it is a necessary truth, be true just in virtue of truthmakers C and the modal attribute, if there is such an attribute, of contingency. (A difficult point of metaphysics.) We do not require a state of affairs of C's having that attribute. Perhaps, indeed, we require no more than C.

References Armstrong, D.M. 1997. A World of States of Affairs, Cambridge: Cambridge University Press. Cox, Damian. 1997. 'The Trouble with Truthmakers'. Pacific Philosophical Quarterly, 78, 45-62. Mulligan, K., Simons, RM., Smith, B. 1984. 'Truth-Makers'. Philosophy and Phenomenological Research, 44, 287-321. German version as: 'Warmacher', in L.Bruno Puntel, ed., Der Wahrheitsbegriff. Neue Explikationsversuche (a collection of readings on modern theories of truth). Darmstadt: Wissenschaftliche Buchgesellschaft, 1987, 210-255. Restall, Greg. 1996. 'Truthmakers, Entailment and Necessity'. Australasian Journal of Philosophy, 74, 331340. Russell, Bertrand. 1918. The Philosophy of Logical Atomism. Reprinted in Russell's Logical Atomism, ed. David Pears, London: Fontana/Collins, 1972. — 1940. An Inquiry into Meaning and Truth. London: George Allen & Unwin. — 1948. Human Knowledge: Its Scope and Limits, London: George Allen and Unwin. — 1959. My Philosophical Development. London: George Allen & Unwin. Ryle, Gilbert. 1949. The Concept of Mind. London: Hutchinson.

Truth Through Thick and Thin 1 RICHARD BOYD

0. Truth: Thick and Thin Conceptions 0.0. Overview. "Thin" or deflationary conceptions of truth are tied to "translational" readings of the Tarski definition (Tarski 1951) according to which it reduces understanding the truth predicate over a language, L, to understanding sentences in L itself, or in some other language in which (a translation of) the truth predicate for L does not occur. Translational readings are often thought to eliminate the metaphysics of truth as correspondence between truth bearers and the world. My thesis is that, for domains of empirical discourse about natural kinds, properties, relations, magnitudes, etc., [henceforth: natural kinds] translational readings do not, even if they capture the notion of truth, underwrite a "thin," noncorrespondence conception of truth. 0.1. Two and a Half Readings of Tarski. Consider the basic clauses of a Tarski type truth definition which assign the primitive terms and predicates of a language, L, to expressions in a metalanguage, M. On the metaphysical reading each such clause asserts that there is a correspondence between the L term in question and the feature of the world to which the M-expression refers, so that the basic clauses underwrite a recursive definition of correspondence truth for L. According to the translational reading, those clauses instead merely underwrite translations of statements in Μ about truth in L into a sub-language of Μ in which there is no reference to truth in L. The concept of truth for L is completely given by instances of the Τ schema of the form S is true in L iff Q Where Q is the translation of S into M. On a thin interpretation, the reference to correspondence celebrated in the first reading is thereby eliminated. According to internal realism (Putnam 1978, 1980) and Fines Natural Ontological Attitude [henceforth: NOA] (Fine 1984) the semantics of scientific theories and other sorts of empirical discourse is provided by naturalistic ("causal") conceptions of reference for their primitive terms, of the sort ordinarily favored by friends of correspondence truth, but the sort of truth involved is not correspondence with reality. Roughly,

1

In formulating my views on the issues I discuss here I have benefitted greatly from conversations with Christopher Boyd, Eric Hiddleston, Barbara Koslowski, Ruth Millikan, Satya Mohanty, Sydney Shoemaker,Susanna Siegel, Jason Stanley, Zoltan Szabo and Quentin Wheeler.

Truth Through Thick and Thin

39

internal realist conceptions arise from the application of a thin conception of truth to the very statements which friends of thick truth offer as an explication of the nature of reference. In what follows I'll explore these two and a half readings with respect to the special case of empirical discourse.

0.2. Sorting Out the (.Anti-) Metaphysical Issues. I'll assume, for the sake of argument, that a Tarski type truth definition can be extended to natural language scientific and empirical discourses with devices for avoiding semantic paradoxes, and that each of the readings of Tarski is equally compatible with the required devices. Nothing I say will depend on how all of this is accomplished. Metaphysical readings of Tarski for empirical discourse face two challenges, each of them indicated in Putnam's defense of internal realism. First, for any natural kind term, t, there will be lots of different kinds, k, such that there is some interesting causal relation between uses of t and instances of k (Putnam 1978, 1980). Metaphysical readings must explain why all but one (or a few) or these kinds are not referents (or partial denotata in the sense of Field 1973) of t. The second challenge concerns the kinds to which terms in empirical discourse refer. Putnam (1983b) suggests that these entities are not "ready made" features of the natural world as, he maintains, a realist conception of reference, kinds and truth requires. Instead, we ourselves are implicated in the metaphysics of "natural" kinds ways that favor "internal realism" over metaphysical correspondence understandings in reference, kinds and truth. Thin translational interpretations of Tarski are designed to avoid such difficulties: reference is not a matter of correspondence metaphysics, nor is truth a genuine metaphysical property of sentences. The translations provided by the Τ schema reduce sentences about the truth in the object language to correspondence-metaphysics-free sentences within it (or within a suitable metalanguage).

0.3. Diagnosing Obesity. When is Dieting Ejfertive?T schema translations do not always eliminate commitments to a correspondence metaphysics. Let L be a fragment of English containing the vocabulary of philosophical semantics and the vocabulary of physics, L' the sublanguage of L containing just the vocabulary of physics, and Μ a metalanguage for L. Now consider the following sentence in L. s: Truth in L' is correspondence truth in the metaphysical sense. The Tarski definition, carried out in M, tells us is that s is true if and only if truth in L' is correspondence truth. Suppose that such equivalences fully analyze the truth predicate in M. Even so, the commitment to a correspondence-truth-metaphysics is not eliminated by the translational interpretation of the sentence in Μ which asserts the truth of s. This uncontroversial result does not challenge thin translational readings of Tarski. Applying the Tarskian translation to the claim that s is true is not supposed to eliminate /s metaphysical implications—even those involving correspondence-truth. The critic of correspondence truth will simply deny (or deny the intelligibility of) s. All the translational reduction is supposed to show is that there is no metaphysical commitment involved in predicating truth (or falsity) o f s .

40

The Correspondence Theory

Call a rationally acceptable sentence, s, susceptible to weight loss iff it satisfies the dietary constraint·, if the rational acceptance of s does not involve a commitment of the relevant sort (see sections 0.5, 0.6) to the sort of correspondence-truth metaphysics which it is the business of thin theories of truth to avoid, s is permanently obese otherwise. In general, for any sentence, s, which the defender of a thin translational conception agrees it is rational to accept, she must be able to argue that s satisfies this constraint. A successful translational reading ofTarski underwrites a thin theory of truth for a body of discourse iff all of the rationally acceptable sentences in that discourse are susceptible to weight loss. 0.4. The Metaphysics of Obesity and Weight Loss: Rational Acceptability and Metaphysical Commitments. Motives for adopting thick correspondence theories are numerous. They range from concerns about naturalistic conceptions of the semantics of natural kind terms to concerns about the foundations of epistemology, morals, or politics. Motives for thin conceptions are equally varied. For verificationists, a central concern was the semantics of "theoretical terms." Some philosophers have sought a metaphysically unobjectionable conception of mathematical truth. Others, like Putnam and Fine, have reservations about metaphysically realistic notions of reference and of kinds. Many friends of thin truth have epistemological, moral or political concerns about correspondence truth. Defenders of thin truth may thus differ about what sort of correspondence metaphysics is to be avoided. For some purposes it would be reasonable to classify domains of discourse as, e.g., "susceptible to weight loss by verificationist standards" or as "permanently obese by internal realist standards." Instead, I shall argue that—with respect to a conception of rationality which is all but uncontroversial—the metaphysical commitments involved in accepting the sentences which report rational achievements in the empirical domain include paradigmatic instances of the sort of correspondence between language and the world which it has been the aim of all or almost all thin theories of truth to avoid. 0.5. The Metaphysics of Rational Deployment, Contexts of Achievement·, and the Empirical Domain. Let's examine the dietary constraint by considering thin conceptions in metaethics. Imagine a noncognitivist moral anti-realist—Alice—who maintains that moral deliberations are part of a rational approach to the establishment of norms of social coordination, but that moral predicates do not correspond to genuine metaphysical items in the world. She maintains that there is a wide variety of quite different ethical systems in the implementation of which one might rationally employ the resources of moral discourse, and between which there are no non-question-begging rational grounds for choice. The correspondence between properties of actions, choices, policies, etc. which are recommended or condemned by the ethical practices of any one such ethical system and the terms of approbation and disapprobation in its moral discourse does not, in her opinion, represent a genuine referential correspondence, so she adopts a non-cognitivist metaethical position. 2 2

Here, and elsewhere, I attribute to this imaginary philosopher conceptions about options in metaethics

Truth Through Thick and Thin

41

One challenge she must meet is that her non-cognitivist semantics cannot account for the role in rational moral arguments of deployment of the truth predicate and of logical notions like implication—for example, of someone's saying (of another's moral statement) "What she said is true and it implies that such-and-such." Thin truth to the rescue! If Alice adopts a thin translational conception of truth she can respond by agreeing that the truth predicate is rationally applicable to some moral statements while insisting that she is not committed to a correspondence metaphysics. Rationally accepting "5 is true," for moral sentences S, involves no more metaphysical commitments than does accepting 5; and accepting S involves accepting the rationality of a certain form of social coordination, not some sort of correspondence metaphysics. Suppose that Alice's utilitarian colleague, Bill, claims that there is a systematic correspondence between moral terms and certain properties and relations defined in terms of aggregate happiness and that each moral term refers to the corresponding property or relation. He therefore defends a thick conception of moral truth. Under what conditions would Bill's conception pose a challenge to Alice's thin conception of moral truth ? First, Bill must argue that the epistemic access condition is satisfied: that there is a general tendency for the moral sentences which are rationally accepted to be (approximately) true under the proposed metaphysical interpretation (see Sections 1.3, 2.1). Suppose that Bill can show that this is so across a wide range of moral communities. This would pose an acute challenge to Alice's conception. Nevertheless, the satisfaction of the epistemic access condition with respect to a systematic correspondence does not always constitute even a prima facie case that the correspondence is the reference relation. The correspondence between a predicate, P, and the property of being believed within the relevant community to fall under Ρ will always satisfy the epistemic access condition, but in many domains of discourse it will not be the reference relation. Alice however already holds that rational moral discourse achieves something—a certain sort of social coordination involving human needs. The strength of Bill's position is that he can argue that the correspondence in question explains that achievement. In recognizing this advantage to Bill's position, we are recognizing that reference is a matter of epistemic access contributing to rational achievements within the relevant languageinvolving practices (Sections 1.3, 2.1.). Thus any discussion about susceptibility to weight loss must take place in a context determined by assessments of the nature and extent of the rational achievements within the relevant domain of discourse. I'll argue that on any reasonable assessment of the context of achievement for empirical discourse the dietary constraint is not satisfied. 0.6. Explanation, Weight Loss and Metaphysical Commitments. I've assumed that a commitment to correspondence truth can arise from the demand for a philosophical explanation of language-mediated rational achievements. A less metaphysically ambitious conception emphasizes conceptual understanding rather than philosophical ex-

to which I do not myself subscribe. See Boyd 1988 ,1995.

42

The Correspondence Theory

planation. Suppose that, for any rationally acceptable sentence, s in some domain of discourse, D, an understanding of the assertion that s is true which is adequate for a rational assessment of that assertion involves the deployment of no more metaphysical notions than does a similarly adequate understanding of the assertion that s. Suppose that, for rationally acceptable s in D, the latter sort of understanding does not require deploying the metaphysical notions that are at issue between fans of thick and of thin truth. A fan of thin truth like the newly converted Alice might then conclude that a thin conception of truth for D is vindicated, metaphysical excursions into philosophical explanation notwithstanding. This approach is inadequate to the issues involved. Thick conceptions capture the distinctly philosophical idea that rational successes within a domain involve a quite particular sort of recruitment of the cooperation of the world in theoretical or practical enterprises. It is hard to see why an adequate understanding of the sentences in a domain should require having an opinion—much less a true one—regarding this esoteric issue. The fact that no correspondence conceptions need be adopted for one to understand either s or "s is true" is metaphysically uninformative. Consider an analogy. Suppose that a Humean philosopher—intent on eliminating reference to supposed natural necessity—offers an analysis of the following form: a caused b iff there are laws, L, such that ...a, b, L,...., where ...a, b, L,... is a description of a relation between a, b, and L in which no casual terms are employed. Suppose that a critic objects that the notion of law is itself an irreducibly causal notion (Boyd 1985b). Would it suffice for the Humean to reply that it is possible to have a scientifically adequate understanding of the notion of a law—one adequate to underwrite rational assessment of proposed laws—without deploying the conceptual resources of a natural necessity metaphysics ? Plainly no. All sides agree that such an understanding is compatible with almost any opinion about (or indifference toward) esoteric issues concerning natural necessity. The question is not what concepts are required for a methodobgically adequate understanding of causal talk but what metaphysical commitments are required for a satisfactory explanation of causal knowledge. Analogously, the question of whether or not a translational reading of Tarski underwrite a thin conception of truth is a question about the metaphysical commitments which underwrite the explanation of the rationally achieved successes within the relevant domain rather than about the metaphysical notions necessary to understand or participate in those achievements. Bill need not hold that rationally competent moral reasoning requires deploying the resources of a thick conception of moral truth. 0.7. The World and We on Land and Sew. Rational Standards, Correspondence and "Social Construction." Suppose that Alice is convinced that the rational achievements of moral discourse depend on a referential correspondence between moral terms and aspects of aggregate human happiness. Must she then accept a thick conception of moral truth? No. She could rely on two responses suggested by Putnam's "Why there isn't a Ready Made world" (1983b). One concerns the extent to which the referents of natu-

Truth Through Thick and Thin

43

ral language terms are—in the relevant way—features of the world, rather than products of our own practices, or social constructions (my terminology, not Putnam's). Putnam claims that a "metaphysical realist" correspondence conception of truth entails that there is a single true theory whose metaphysically privileged vocabulary contains terms referring to the unique set of natural kinds, magnitudes, relations, etc. He (correctly) takes this alleged consequence of "metaphysical realism" to be refuted by the fact that there are scientific theories which are not reducible to each other or to any more basic theory, and by the fact that there are perfectly good natural kinds whose naturalness is (in part at least) a matter of their "fit" to particular human interests and projects. Since Bill's metaethical position treats moral properties as like project-specific natural kinds, Alice might adopt Putnam's social construction objection and deny that reference to them is the foundation for correspondence truth in the metaphysically relevant sense. In a related deflationary maneuver, Putnam considers responses to his reductionist conception of "metaphysical realism" including the suggestion of Boyd (1980, 1988, 1989) that the relation between the phenomena studied in the special sciences and the phenomena studied in fundamental physics is causal realization rather than reduction of vocabularies. Putnam denies that there is any metaphysically respectable notion of cause simpliciter, or of "total cause," in terms of which causal realization could be defined. Instead, the notion of cause is somehow derivative from the context-or-discipline specific notion of explanation. Apparently Putnam is raising the social construction objection with respect to the phenomenon of causation itself. Bill's utilitarianism makes moral properties matters of causal powers to affect net happiness, and Bill s conception of reference involves causal components. So Alice could deploy Putnam's social construction of causation objection to argue that neither the referents of moral terms which Bill has identified, nor the mechanisms of reference to them, are appropriate to the defense of a metaphysical correspondence conception of moral truth. Of course, either of these objections would, if successful, undermine the position I defend here so I'll argue that the involvement of human interests and practices in establishing reference relations, and in defining natural kinds, does not compromise the metaphysical conception of correspondence truth for empirical domains. 0.8. Strategy. I'll defend thick truth in three steps, relying heavily on the existing literature in the philosophy of science, providing brief arguments and indicating where more elaborate treatments can be found. I'll write in the style and language of scientific realism but the crucial considerations about the accommodation of language to causal structures are all implied by sophisticated empiricist conceptions as well (Boyd 1990a, 1992). Step One. I'll (a) rehearse arguments to the effect that in at least some of the empirical sciences researchers employ reliable inductive methods whose reliability can only be explained by positing a causally grounded relation of reference between terms and natural kinds. I'll then (b) defend the accommodation thesis according to which the role of reference to natural kinds (etc.) is to help establish an accommodation between inferential practices and relevant causal structures. Thus, prima facie, explaining epistemic successes in the empirical domain requires acknowledging a correspondence theory

44

The Correspondence Theory

of reference and of truth for that domain, independently of whether or not a successful translational reading of Tarski is available. Step Two: I'll next address concerns about the determinateness of reference, and about the social construction of natural kinds. I'll argue that natural kinds are social constructions but that the relevant sort of social construction would poses a challenge to a correspondence theory of truth only on a bizarre metaphysical conception according to which humans are not part of the natural world. Regarding determinateness, I'll argue that reference is a dialectically complex process, of which exactly determinate reference for particular terms is a special case. In fact, the reference relation is exactly as determinate {and exactly as indeterminate) as a metaphysical conception of correspondence truth for the empirical domain requires. Step Three. I'll next address social construction of causation objection. I'll very briefly defend the no-non-causal-contribution thesis that human social practices make no contribution to causal structures in the world except via the operation of ordinary causal mechanisms. There is no interesting metaphysical sense in which causal phenomena themselves are social constructions. This, too, follows from any metaphysical conception which takes humans to be ordinary natural phenomena.

1. Kinds and Accommodation 1.0. "Grue," Induction and all that. Locke (1689) speculates at several places (e.g., IV, iii, 25) that when kinds of substances are defined by purely conventional "nominal essences," as he thinks they must be, it will be impossible to have a general science of, say, chemistry. Nominal essences define kinds of substance in terms of sensible properties, but the factors which govern the behavior of substances are the fundamental properties of their insensible corpuscles. Since there is no reason to suppose that our conventional nominal essences correspond to uniformities in microstructure, there is no reason to believe that kinds defined by nominal essences will be apt for the formulation of general knowledge of substances. Only if we are able to sort substances according to their hidden real essences will systematic general knowledge of them be possible. Scientific realists agree. Terms in inductively successful scientific disciplines must be defined by a posteriori (often "unobservable") real essences, just as Locke thought, but contra Locke knowledge of real essences is possible (Putnam 1972, 1975a, 1975b; Boyd 1983, 1985a, 1985b, 1989, 1999a). Goodman 1973 and Quine 1969 reflect an empiricist route to the same basic conclusion: that reference to kinds with natural a posteriori definitions is crucial to the formulation of projectible hypotheses and thus to successful induction in the empirical sciences (Boyd 1992). 1.1. Accommodation and Reliable Induction. The philosophical theory of natural kinds thus addresses the question of how classificatory schemes contribute to the epistemic reliability of inductive and explanatory practices. The naturalness of natural kinds consists in their aptness for induction and explanation, and their definitions reflect the properties of their members which contribute to that aptness. According to the accommodation thesis what establishes the reliability of inferential practices—what

Truth Through Thick and Thin

45

reference to natural kinds makes possible—is the accommodation of inferential practices to relevant causal structures. Suppose that you have been conducting experiments, exposing various salts of sodium to flames. In each case, the flame turned yellow. [Recall the "flame test" for sodium.] You conclude that always if a salt of sodium is heated in a flame, then a yellow flame results. You are right and your inductive success is not an accident; it is a reflection of the fact that the categories salt of sodium, flame, and yellow (as a property of flames) are natural categories in chemistry, and of the fact that the hypothesis you formulated with the aid of reference to these categories is a projectible one. 3 There are indefinitely many false and unprojectible generalizations which equally well fit all your data. You discerned the true one because your inductive practices, including reference to these natural categories, allowed you to identify a generalization which was appropriately related to the causal structures of the relevant phenomena. What distinguishes the generalization you accepted from the unprojectible generalizations which also fit all your data is that in any situation in which its antecedent is true, the state of affairs described by the antecedent will (in the relevant environment) cause the effect described by its consequent. Your practice of deploying natural categories in formulating projectible generalizations allowed you to identify a causally sustained generalization. Other instances of inductive success are more complicated, but, in general, we can reliably identify true (or approximately true) generalizations just to the extent that we can identify generalizations which are sustained by relevant causal structures. In order to frame such generalizations at all we require a vocabulary, with terms like "sodium salt" and "flame", which is itself accommodated to those structures (Boyd 1990a, 1990b, 1991, 1992, 1999a, 1999b). Thus to explain the reliability of inductive methods in successful scientific disciplines one must posit a systematic correspondence between the use of natural kind terms, on the one hand, and causal structures, on the other. I propose that this correspondence is the reference relation upon which a correspondence conception of empirical truth should be grounded. 1.2. Disciplinary Matrices and a Kind of Relativism. The naturalness of a natural kind depends on the inferential architecture within which representations of it are embedded. The kind salt of sodium is a natural kind just because reference to it contributes to the accommodation of the inductive and explanatory practices of workers in chemistry and related disciplines to causal structures. It is irrelevant whether or not reference to salts of sodium contributes to accommodation in other disciplinary settings. In the final analysis there are not natural kinds simpliciter, but instead kinds which are natural with respect to the inferential architectures of various disciplinary matrices·, families of inferential aims and practices, together with associated conceptual resources (Boyd 1999a, 1999b). Thus any reference to natural kinds involves either (perhaps tacit or

3

I shamelessly lift this example from Boyd 1999b.

46

The Correspondence Theory

indirect) reference to systems of human inferential practices or (perhaps tacit or indirect) quantification over such systems. Such systems are paradigmatic cases of social constructions, so we need to inquire about the metaphysical implications of their intimate connection to natural kinds. A helpful analogy is our next topic. 1.3. Accommodation

Demands,

Epistemic Access, and Evolutionary

Function.

The

accommodation demands of a disciplinary matrix, M, are the dimensions of accommodation between M s conceptual and classificatory resources and relevant causal structures which would be required for the inductive and explanatory aims characteristic of Μ to be achieved. Even in largely successful matrices some accommodation demands may not be satisfiable: there may not exist the sorts of causal structures which could sustain the sought after generalizations or explanations. We may now state more precisely the conclusion of Section 0.5: A natural kind term, t, refers, within a disciplinary matrix, M, to a natural kind, k, only if (a) the use of the term t in Μ satisfies the

epistemic access condition with respect to k, and (b) (a) helps to explain how the accom-

modation demands of Μ are satisfied. There are important analogies between assigning a posteriori natural definitions to natural kind terms and assigning evolutionary fitnctions to phenotypic features of organisms. In a famous paper ("What the Frog's Eye Tells the Frog's Brain") Lettvin et al 1959 identify cells in the optic system of certain frogs which respond to small objects in the frog's visual field, but only if those objects are moving in a particular sort of trajectory. These cells and their particular response pattern constitute an adaptation. Their evolutionary fiinction is to contribute to the detection of insect prey: they represent a component in the frog's "solution" to the evolutionary problem of food gathering. Adaptive explanations for phenotypic features always have this form. Causal powers of the phenotypic trait characterized in terms of propositional content (e.g., the capacity "to contribute to the detection of insect prey") are identified as the evolutionary function of the trait because the trait's having those powers contributes to the solution of an evolutionary problem (like food gathering). Talk of "evolutionary problems" is a metaphorical way of talking about the demands which the environment places on adaptations (or accommodations) in respect of reproductive fitness. Among the many analogies between evolutionary explanations and semantic ones (see, e.g., Millikan 1984), the following is especially important for our purposes: evolutionary function: evolutionary problems :: reference : accommodation demands. Of course, evolutionary functions are adduced to explain how the relation of evolutionary lineages to particular causal structures in their environment contributes to the satisfaction of the accommodation demands of natural selection. There are other important respects of analogy. Recall that naturalistic theories of reference and truth must explain how the real referent of a natural kind term differs from the other interesting categories with which its use is systematically connected. An analogous question arises for the cells discovered by Lettvin et al. They respond to (some) insect prey in the frog's visual field, but this is not the only biologically interesting characterization of their re-

Truth Through Thick and Thin

47

sponses. In the wild they do not respond to all and only insect prey, so for some behavioral studies of frogs, or of the psychology and physiology of frog vision, a different characterization of their response profile might be appropriate. Nevertheless, these other biologically interesting characterizations are not competitors as characterizations of evolutionary function precisely because the explanatory role of reference to evolutionary functions is to identify those causal powers of the phenotypic trait which explain the solution to evolutionary problems facing the relevant lineage, and for that purpose reference to insect prey detection is appropriate. The analogy between evolutionary problems and accommodation demands points to a similar solution to the problem of determinateness of reference The referent of a natural kind term is determined by the pattern of interactions between language users and the world which explains how solutions are achieved to the quite particular set of problems associated with the accommodation demands of a disciplinary matrix. I'll argue that the task of explaining accommodation demand satisfaction constrains attributions of reference, and that reference is, and should be, exactly as determinate (and exactly as indeterminate) as this constraint requires. I'll deploy the same analogy in exploring the social construction objection with respect to natural kinds. Any theory of adaptation in general, or of any particular adaptation (like, e.g., an adaptation for prey detection), involves (perhaps tacit or indirect) reference to, or quantification over, (lineages of) organisms and the evolutionary problems facing them. Moreover, the categories invoked in explaining adaptations are themselves defined in terms of the characteristics of particular lineages ("insect prey" in the example we are considering really refers to insects which can be prey of the frogs in the lineage in question). Neither the evolutionary functions of phenotypic traits of frogs, nor the categories deployed in describing them, are thereby shown to be amphibian constructions in any sense that makes them less than real. This remains true even when the evolutionary functions in question must be specified in terms which literally refer to social construction, as in the case of traits like nest construction in social insects. The metaphysical reality of natural kinds and of reference to them is, I'll argue, no more impugned by relations to human projects than the reality of evolutionary functions and the categories required to describe them is impugned by relations to the evolutionary problems facing lineages.

2. The Metaphysics of Natural Kinds: Definitions, Reference and "Social Constructions" 2.0. A Picture and Three Problems. According to naturalistic conceptions, the reference relation between a natural kind term, t, and a kind k, is a matter of epistemically relevant causal relations connecting uses of t with instances of k. This suggests (but does not entail) a metaphysical picture: reference is a relation between linguistic entities and entirely extra-linguistic (and in that sense independently existing) natural kinds. Natural kinds are, somehow or other, in the world, and available for discovery and naming, independently of human practices . If one accepts this picture of semantic naturalism as many philosophers (including,, e.g., Putnam 1978, 1980, 1983b) seem to do, then it is natural to think that what

48

The Correspondence Theory

makes the associated naturalistic conception of truth a correspondence conception is that references is seen as a relation between terms and such independently existing kinds. The correspondence conception, so conceived, is subject to three challenges. First, if we think of natural kinds as things somehow independent of linguistic and methodological practices, then there are lots of natural kinds out there, and it is difficult to see how the causal conception of reference fixing could explain how a natural kind term could ever have a unique referent (this seems to be the basis of the "model theoretic" arguments in Putnam 1978, 1980). This problem is exacerbated by the fact that causal theories of reference arose as criticisms of descriptivist (and conventionalist) conceptions of reference in the empiricist tradition. This has led some defenders of causal theories, and some critics (e.g., Putnam 1978, 1980) to conclude that naturalistic conceptions of reference, at least if they are to underwrite correspondence conceptions of truth, must be pure causal theories in the sense that they do not invoke descriptions or other conceptual elements, like referential intentions, in explaining reference. On this conception challenges regarding determinateness will seem even more acute. Finally, reference to natural kinds is supposed to explain the inductive successes of scientific practice, so there must be some quite intimate connection between natural kinds and the conceptual machinery of the sciences. If one thinks of naturalistic theories as entailing that natural kinds are independent of that machinery, it is hard to see how the explanation could work unless it rested on some sort of objective idealist theory according to which natural kinds are somehow metaphysically "fitted" for explanation and induction independently of the relevant practices. But, that's not consistent with any sort of naturalism. So, as Putnam and NOAers like Fine suggest, naturalistic correspondence theories seem to be in trouble. I reject the picture in question. It is reductionist·, it understands the naturalness of kinds and of reference as matters of appropriateness for induction and explanation, but it imports to the conception of naturalness a contrast between the natural and the social or conceptual. It is committed to reducing apparent reference to social or conceptual matters in the theory of natural kind terms and their referents to "pure" causal talk uninfected with reference to the social or conceptual. Instead, I propose a nonreductionist conception. To offer a naturalistic account of kinds and reference (or any phenomena for that matter) is to situate them entirely within the natural causal order, but—since we, our psychological states, and our social structures, are as much parts of the natural world as are other animals and their psychological states and social structures—reference to these phenomena as causal phenomena does not compromise a proposed naturalistic account of reference and kinds. 2.1. Non-Reductionist Naturalism, I: Reference and Kind Definitions. Ignoring issues like partial denotation (Field 1973) to which we'll return later, I advocate the following accommodationist conception of natural kind terms and their referents. Let Μ be a disciplinary matrix and let tj,...t be the natural kind terms deployed within M. Then the families Fj,..,F of properties provide the natural definitions of the kinds referred to by tj,...t , and determine their extensions, just in case:

Truth Through Thick and Thin

49

1. (Epistemic access condition) There is a systematic, causally sustained, tendency—established by the causal relations between practices in Μ and causal structures in the world—for what is predicated of t within Μ to be approximately true of things which satisfy F;, i=l,...,n. 2. (Accommodation condition) This fact, together with the causal powers of things satisfying the F;'s, causally explains how the use of tj,...,t in Μ contributes to accommodation of the inferential practices of Μ to relevant causal structures, and thus to the epistemic success of M. [This condition plays a role similar to that played by reference to "conceptual roles" in other accounts of reference, with the difference that the constraints imposed by it are profoundly a posteriori. See Boyd 1999a, 1999b, 1999c on programmatic definitions of natural kind terms.] This conception is anything but reductionist. It specifies the semantics of natural kind terms by reference to the epistemic reliability of certain human practices. In mature sciences, that epistemic reliability depends on the acceptance by scientists of relevantly approximately true theories (Boyd 1983, 1985a, 1989; for discussions of approximation see Boyd 1990b, Psillos 1999). Thus, the accommodationist conception acknowledges a grain of truth in descriptivism: ordinarily our capacity to refer to a kind depends on our having some significant approximately true beliefs about it. Similarly, it acknowledges a role for speakers' (and communities') intentions which help to determine the methodological practices and accommodation demands within disciplinary matrices. Reference to natural kinds is not "purely causal," in the reductionist sense. Consider the challenges regarding determinateness of reference discussed earlier. In so far as some causal theories of reference face such challenges, they do so because they treat reference and natural kinds reductively: as independent of human practices and concepts. The accommodationist conception avoids this pitfall. It simultaneously defines the reference relation, and the natural definitions and extensions of natural kind terms, in terms of the contributions which the deployment of those terms make to the satisfaction of accommodation demands. For a natural kind term within a particular disciplinary matrix, there may be lots of kinds, natural with respect to some disciplinary matrices or other, which satisfy epistemic access condition with respect to the term. The accommodation clause picks out the term's referent by identifying, as its defining properties, those which play the indicated role in explaining the satisfaction of the accommodation demands of the particular disciplinary matrix. With respect to issues of determinateness, the analogy between attributions of reference and attributions of evolutionary functions is thus exact. In each case the demands of accommodation narrow the class of candidates. Of course, the epistemic access and accommodation conditions will not always pick out a unique referent even for a kind term fruitfully employed in a disciplinary matrix (just as the constraints on evolutionary explanations do not always assign a unique evolutionary function to an adaptive trait). I'll argue later that the relation defined by those conditions is exactly as determinate as reference is and exactly as determinate as it as it should be to underwrite a correspondence conception of truth.

50

The Correspondence Theory

2.2. Non-Reductionist Naturalism, II: Natural Kinds as Social Constructions. Locke (1689) said that, while Nature makes things similar and different, kinds are "the workmanship of men." Gender bias aside he was right. The theory of natural kinds just is the theory of how accommodation is (sometimes) achieved between the workmanship of our linguistic, classificatory and inferential practices and causal structures in the world. Locke said that "...each abstract idea, with a name to it, makes a distinct Species": kinds are established by a sort of unicameral linguistic legislation in which people get to establish kinds by whatever nominal conventions for the use of general terms they choose to adopt. According to accommodationism natural kinds are products of bicameral legislation in which the world plays a heavy legislative role. A natural kind is nothing (much) over and above a natural kind term together with its use in the satisfaction of the accommodation demands of a disciplinary matrix. [ What else? Whatever is necessary to accommodate translations which preserve satisfaction of accommodation demands and to accommodate phenomena like reference failure and partial denotation.] Natural kinds are features, not of the world outside our practice, but of the ways in which that practice engages with the rest of the world; they are the workmanship of women and men. I'll argue in Part Three that the causal structures to which accommodation is required are not themselves social constructions. Still, natural kinds are social artifacts.4 The kind natural kind is itself a natural kind in the theory of our inferential practice. Does this compromise a correspondence conception of truth? I'll next begin to argue that it does not. 2.3. Accommodation, Deference and Metaphysical Correspondence. Here, in broad outline, is why an accommodationist conception of natural kinds underwrites a correspondence conception of truth. 1. The causal structures to which our use of natural kind terms is accommodated are, in the metaphysically relevant sense, independent of human practices (see Part Three). 2. Therefore, the bicameral linguistic legislation which establishes natural kinds and our reference to them reflects a pattern of deference to (causal structures in) the world on the part of users of natural kind terms. 3. Since reference to natural kinds reflects deference to independently existing causal structures, truth for sentences about natural kinds is appropriately understood as a matter, not just of some sort of referential correspondence, but of correspondence with reality in the sense appropriate to thick theories of truth.

4

O f course, there are natural kinds for which there are no associated terms. New species are discovered every day and they are natural kinds even before they are discovered or named. So are species which will never be discovered and named. When we speak of unknown natural kinds we are, at least roughly, tacitly or explicitly referring to, or quantifying over, disciplinary matrices and speaking of possible accommodation enhancing augmentations of their vocabularies.

Truth Through Thick and Thin

51

In Section 2.4. I'll address questions about the determinacy of reference. Let's first examine the question of the social construction of natural kinds along the lines indicated by step 3, above. Does the metaphysical relationship between natural kinds and human practice undermines the correspondence conception of truth for sentences about natural kinds? What the accommodationist conception of natural kinds and reference indicates is

that the analogy evolutionary functions : evolutionary problems :: reference : accommodation demands is exact in so far as the metaphysics of correspondence is concerned. Just as the notions of evolutionary function and of evolutionary problems are defined entirely by their role in the explanation of evolutionary adaptation (=accommodation) to features of the environment(s) of lineages, the notions of reference and accommodation demands are defined entirely by their role in explaining the epistemic success of human practices. Thus the fact that (perhaps tacit or indirect) reference to, or quantification over, human projects is involved in any discussion of natural kinds, and that the a posteriori definitions of particular natural kinds are determined by facts about human practices, is no more metaphysically surprising than the fact that reference to evolutionary functions involves reference to, or quantification over, lineages, and that evolutionary functions must be specified in terms of categories (like prey) which are defined in terms of those lineages. In general, when we seek to explain facts about natural phenomena we need (in order to achieve accommodation) to refer to natural kinds defined in terms of the causal powers and dispositions of those phenomenon and their constituents. If the phenomena and their constituents are real natural phenomena, then so are their various powers and dispositions, and so are the kinds defined in terms of those powers and dispositions. Evolutionary function attributions, when true, reflect a metaphysically respectable correspondence between phenotypic traits of organisms and real features of the natural world even though those attributions involve (tacit or indirect) reference to those organisms and to kinds defined in terms of them. The same thing is true for semantic explanations in terms of accommodation demands and reference. In order to explain the social and linguistic evolution of inductively and explanatorily successful schemes of classification, one must refer to phenomena defined in terms of human inductive and explanatory practices, and the causal powers of their constituents. Since we, LIKE FROGS, are real features of the natural world, no metaphysical unreality attaches to the categories, like reference and like particular natural kinds, which are thus defined partly in terms of features of our practices. Truth for sentences about natural kinds, defined recursively in terms of reference, is thus "correspondence truth" on any understanding which gets the metaphysics of natural phenomena and their causal powers right. [I have here assumed that, LIKE FROGS, we are natural phenomena, and in particular that our relation to causal structures is no different metaphysically from that enjoyed by FROGS. See Part Three.] Let's make precise the claim that natural kinds are defined in terms of our practices and that reference to them involves tacit or indirect reference to or quantification over such practices.

52

The Correspondence Theory

1. The natural definitions of natural kinds do not always include social properties (they don't in chemistry for example), but, 2. even in such cases the constraints identified in the accommodation condition, which help determine the kind's natural definition, reflect the role of the relevant term in the practices of a disciplinary matrix. 3. The natural definition of the kind natural kind does thus involve properties of human practices. So, when we discuss natural kinds generally, or when we classify a kind as a natural kind, we are (explicitly) referring to a category defined (but not analytically defined) in terms of human practices. 4. In using a natural kind term like "water" we are explicitly referring only to H 2 0 and not also to some social properties, so when chemist talk about water they are not

talking about their own practices, but, 5. in deploying the term "water" as we do in our inductive and explanatory practices, we are tacitly treating it as a natural kind term, and in that sense making a kind of indirect reference to the relevant practices, in much the same way that in using the term "dollar" we are indirectly referring to complex social and economic practices even though we are not talking about them. The relation between talk about natural kinds and reference to human practices is thus complex and deep but, as the analogy with evolutionary functions indicates, it is not metaphysically deflating. 2 . 4 . Accommodation,

Precision and Referential

Determinateness.

O n e objection to

naturalistic correspondence conceptions of reference and truth is that they fail to explain how determinate reference can obtain. The accommodationist conception addresses this question by identifying the referent of a natural kind term, used within a disciplinary matrix, in terms of the satisfaction of the accommodation demands of that particular matrix. Still, there will sometimes be some indeterminacy of the reference relation even if it is so characterized. Does this residual indeterminacy undermine a correspondence conception of truth? No. Whatever indeterminacy escapes the resources of these conditions is really there; reference (understood as metaphysical correspondence) is no more determinate than those conditions indicate. To see this we need to a refine the accommodationist conception to cover "partial denotation" and "denotational refinement" (Field 1973). 2 . 4 . 0 . Partial Denotation,

and Reference as a Dialectical Phenomenon. A natural k i n d

term, t, within a disciplinary matrix, M, partially denotes two different kinds^ if there are two families, Fj and F 2 of properties such that: 1. The satisfaction of the accommodation demands of Μ would be enhanced by the use of two terms, one with natural definition Fj and one with natural definition F 9 (call these k, and k 7 ).

5

This analysis generalizes in obvious ways to cases in which there are more than two partial denotata.

Truth Through Thick and Thin

53

2. The epistemic access condition is satisfied to a significant extent by t and Fj and by t and F 2 . 3. Fj and F 2 are similar enough that the epistemic access which uses of t afford practitioners to k j and k 2 contributes to the epistemic reliability of practices in the disciplinary matrix, so that something like the accommodation condition is satisfied by t and Fj and by t and F 2 . Denotational refinement takes place when 1-3 are recognized and separate terms come to be deployed to refer to k j and k 2 (ordinarily t will come to be used to denote one of those kinds). The paradigm cases of these phenomena involve the refinement of chemical terminology occasioned by the discovery of isotopes. What the accommodation thesis indicates is that reference (at least for natural kind terms) is the relation between language use and the world which explains how the accommodation of language to relevant causal structures is achieved. I have argued elsewhere (Boyd 1993) that reference should thus be seen as the dialectical process of accommodation between the use of such terms and causal structures which is achieved by reliable inductive and explanatory practice. Both partial denotation and denotational refinement are aspects of the ongoing process of reference. The achievement of a referential situation in which a natural kind term enjoys a determinate natural definition is

a special case of the phenomenon of reference. When the referential situation of a natural kind term involves partial denotation, sentences containing that term will be, in a certain sense, ambiguous. They may be true on one reading and false on another. More importantly, each of the two readings may reflect a different aspect of approximation between scientific representations and actual causal structures, and denotational refinement will ordinarily be necessary in order to achieve more accurate approximate representations of them. However offensive such situations may be to the sensibilities of logicians, working with terms that are ambiguous in this way—and refining their use when the ambiguity is recognized—is central to reliable scientific (and everyday) practice. Any adequate conception of the growth of approximate knowledge in chemistry must, for example, say something very much like this with respect to uses of terms like "carbon" and "element" before and after the discovery of isotopes. [In general, an adequate understanding of approximate truth regarding natural kinds requires specifying dimensions of approximation in terms of a correspondence metaphysics (Boyd 1990b, Psillos 1999).] Reference and truth, properly understood, are thus no more determinate than the proposed account of partial denotation and denotational refinement indicates. There is a metaphysically real referential correspondence between kind terms and the features of the world which underwrites our capacity to represent causal relations, but that correspondence has a dialectical complexity which attention to the formal apparatus of the Tarski definition renders obscure. It is a virtue, not a vice, of naturalistic causal correspondence conceptions of reference and truth that they explain this dialectical element in reference. 2.4.1. " Vague" Kinds and "Vague" Natural Definitions. Some scientifically important predicates (species' names, for example) exhibit a different sort of indeterminacy. They have ineliminably vague extensions. Putnam 1983a apparently thinks that this refutes

54

The Correspondence Theory

metaphysical correspondence conceptions of truth, apparently because he thinks that correspondence conceptions must entail that natural kinds always have determinate extensions. They needn't. Suppose that a term, t, determinately denotes, t will be defined by some family, F, of properties satisfying the epistemic access and accommodation conditions. F need not, however, be a set of properties which are necessary and sufficient for membership in the extension of t. Disciplinary practices in matrices in which complex phenomena are studied often take advantage of (imperfectly) homeostatically united clusters of properties in order to define explanatorily/inductively relevant kinds. The natural definition of such a kind (a homeostatic property cluster, or HPC, kind) is an (imperfectly) homeostatically united family of properties together with the homeostatic mechanisms which unite them. In the most general case, a homeostatic property cluster definition is a process-like historically individuated property cluster/«^, so that the membership conditions it specifies may vary over space and time (for a fuller exposition see Boyd 1988, 1989, 1993). All that is required for an H P C kind to be the referent of a natural kind term under such circumstances is that there be no kinds, natural in the relevant matrix, corresponding to precisifications of the homeostatic property cluster. Darwin saw that this was so with respect to species level taxa. The 'precision that such precisifications would introduce fails to correspond to any biologically important causal phenomena, so reference to precisifications would not contribute to accommodation. The accommodationist conception thus explains the ineliminable "vagueness" of H P C kinds like species. Reference to H P C kinds has all the hallmarks of metaphysical correspondence: assuming (see Part 3) that causal structures are not themselves social constructions in a deflationary sense, the homeostatic property clusterings which constitute the definitions of H P C kinds are metaphysically respectable phenomena in nature, and reference to them is crucial to the satisfaction of the accommodation demands of the relevant disciplinary matrices. It might seem otherwise. Correspondence truth is supposed to be correspondence with the facts. Some sentences about H P C kinds have indeterminate truth values, so are there suppose to be indeterminate facts to which they correspond? No. The H P C conception does not imply that there are indeterminate facts, only that the satisfaction of the accommodation demands of lots of disciplinary matrices involves reference to phenomena with (to some extent) indeterminate boundaries. Biological species are H P C kinds defined by the homeostatic mechanisms which ensure genetic and phenotypic similarities between their populations. 6 Let S be a species, and Ρ a population, such that it is indeterminate whether or not Ρ lies in S. For

6

It is controversial that species are kinds rather than individuals (see Ereshefsky 1992, and Wilson 1999); for a defense of the thesis that they are natural kinds, but that the kind-individual distinction is not especially important, see Boyd 1999b.

Truth Through Thick and Thin

55

each of the sorts of homeostatic mechanisms which operate to preserve the integrity of S, and for each of the inductive or explanatorily relevant similarities between populations in S, there will be a fact of the matter (at any given time) about the extent to which the mechanism in question operates in the case of P, and about the extent to which the organisms in Ρ exhibit the similarity in question. These facts, in aggregate, constitute the (perfectly determinate) fact (at the relevant time) about P's relation to S. What follows from the fact that the sentence "P lies in S" lacks a determinate truth value is just that this relation is not characterizable in the sub-language of biology in which the only predicates are species names. This is hardly surprising since the contribution of species names to accommodation is to permit accurate descriptions of species level stabilities rather than of departures from such stabilities. Scientists (and others) routinely refer to individuals whose boundaries (either spatial or temporal) are not entirely determinate: individual animals and plants, populations, rivers, islands, clouds, planets, stars.... Philosophers are not inclined to think that this fact compromises a correspondence conception of truth even if they reject it for other reasons: they recognize that there are macroscopic stabilities in nature for which no exact boundaries exist, and that this fact raises no special metaphysical questions about their ontological status or about reference to them. It is the precisified versions of such entities about which there might be ontological worries. The H P C conception of kinds simply extends this insight to (some) natural kinds as well. In truth, we live in a world with interesting but vague things and kinds in it, so a language apt for formulating representations which correspond with reality must exhibit just the sort of "vagueness" we have been discussing. 2.4.2. Homeostasis and the Limits of Linguistic Legislation. An additional advantage to identifying the natural definitions of HPC terms with process-like clusterings of properties involves the semantics of HPC terms with respect to counterfactual situations. Consider a situation in which an actual world H P C term, t, has as its definition a property clustering, C. What it is for something to lie within the extension of t, in a possible world, W, is that in W it possess sufficiently many of the properties represented by the property clustering C as that clustering is manifested in W. The individuation of property clusterings across possible worlds is like the individuation of, e.g., persons or nations across possible worlds. For significantly distant possible worlds such individuation breaks down, and homeostatic property cluster terms cease to be defined. The bicameral legislation which establishes them has authority limited to nearby possible worlds. Most of the key terms in philosophy ("knowledge," "reference," "justification," "rationality," "justice," "goodness," etc.) refer to H P C phenomena and have determinate application only with respect to nearby possible worlds. Our practice is to assess the truth values, in non-actual possible worlds, for statements about these phenomena by consulting our "linguistic" or "philosophical" intuitions. As it happens, our intuitions deliver determinate answers about a far wider range of possible worlds than those over which the relevant bicameral legislation has any authority whatsoever. We hold philosophical theories accountable for giving determinate answers regarding counterfactual cases for which there is in fact no answer. Our methodology thus rewards philosophi-

56

The Correspondence Theory

cal theories only if they get wrong the fundamental metaphysics of the phenomena we study (Boyd 1988, 1999b, 1999c). The remedy is the recognition that, for a wide class of terms, definitions are provided by a metaphysical correspondence between use of the terms and process-like phenomena of property clustering in the actual world.

3. Causation and Human Practices. 3.0. Why Botheri We have seen that the establishment of natural kinds is accomplished by (bicameral) social construction but that this fact provides prima facie evidence for, rather than against, a correspondence conception of truth. The only remaining question is whether causal relations and causal processes themselves are always social constructions: whether, for example, the attraction between a suitably rubbed rubber comb and some pieces of paper—not the paper, or the comb, or the categories paper, comb and attraction, but the attraction itself—is a social construction. This might seem a foolish question. Isn't the thesis that causation itself is a social construction a "straw »ι man ? Probably few philosophers really mean to advocate this thesis but some seem to do so. Kuhn (1970) claims that a paradigm provides quasi-metaphysical knowledge of the underlying causes of the relevant phenomena while also apparently maintaining (in arguing for the incommensurability of successive paradigms) that the fundamental laws of a paradigm are analytic or conventional truths. Similarly, Putnam 1983b responds to the non-reductionist conceptions that materialism is a doctrine about the causal composition of natural phenomena by maintaining that the notion of causation is conceptually linked to particular programs for explanation. It is hard to see how this response could have its intended force if it simply made the standard anti-reductionist point that the resources for the individuation and description of causes will always arise in some explanatory project or other. Putnam's response would appear to be cogent only if it involved the proposal that the phenomenon of causation itself is somehow reducible to the social phenomenon of explanation giving. Surely neither Kuhn or Putnam really intended to approach these levels of metaphysical relativism but not everyone influenced by them is committed to avoiding relativism. So, it is reasonable to respond to the social constructivism about causation they appear to have advocated, without speculating about their own philosophical intentions. 3.1. Metaphysical Innocence. According to the constructivism apparently defended by Kuhn, the fundamental laws within a paradigm are truths by convention which nevertheless provide quasi-metaphysical knowledge of fundamental causes. Can this be so? Can we make causal claims true or false by adopting rules of language, or conceptual schemes, or paradigms, or graduate programs? The metaphysical innocence thesis (Boyd 1990a, 1992, 1999a) is the common sense view that human social practices are metaphysically innocent: that we cannot make metaphysical propositions (in particular, causal propositions) true or false simply be adopting particular conventions or practices. Of course linguistic conventions and other

Truth Through Thick and Thin

57

social practices contribute to establishing the truth conditions for causal claims, and we are participants in the bicameral legislation which establishes the conceptual resources with which we describe causal phenomena. What is at issue is whether or not social or linguistic practices can somehow establish causal (or other metaphysical) facts or states of affairs. A version (see below) of the negative answer is presupposed whenever philosophers argue that some apparently metaphysical issue is better thought of instead as being settled by matters of linguistic convention, or whenever scientists treat the choice between different ways of representing natural phenomena as matters of pedagogical or cognitive or computational convenience rather than as matters of substance. 3 . 2 . Clarification:

The No Non-Causal

Contribution

Thesis. Strictly speaking, social

practices such as the adoption of linguistic conventions or the establishment of paradigms do make a contribution to establishing causal states of affairs. Social practices are themselves causal phenomena, so they have a causal impact on other causal relations and properties. The version of the metaphysical innocence thesis which is presupposed by philosophical and scientific common sense regarding causation is the no non-causal contribution thesis (2N2C): that human social practice make no non-causal contribution to causal properties and relations. 3.3. Causation, Frogs and us. Like all or most of philosophical and scientific common sense, 2N2C is not analytic. I have elsewhere (Boyd 1990a, 1992, 1999) indicated some of the empirical and theoretical arguments in its favor. For present purposes, let us just note that 2N2C amounts to the claim that, in so far as our impact on causal

phenomena is concerned, we are natural phenomena like FROGS, or trees or rocks. However they have been understood, surely Kuhn and Putnam did not mean to deny this. Even so, 2 N 2 C is the doctrine we need to examine here. If 2N2C is true then the accommodationist conception of natural kinds supports a metaphysically thick correspondence conception of truth, even though it entails that natural kinds are, in an important metaphysical sense, social constructions. As promised the two social construction objections would poses a challenge to a correspondence theory of empirical truth only on a bizarre metaphysical conception according to which humans are not part of the natural world. 3.4. Concluding Naturalist Postscript. We are animals. Like other animals, we have non-linguistic representational capacities—perceptual and memorial ones, for example. For none of these capacities, human or non-human, is a "thin" translational conception of representation possible. Explaining the representational semantics of these capacities requires positing some sort of correspondence between internal representational structures and features of the empirical world. Since our linguistic representational capacities are evolutionarily continuous with the non-linguistic representational capacities of our ancestors, and since our linguistic representations are psychologically integrated with many of our non-linguistic representations, it is to be expected that, at least in the empirical domain, our linguistic representations too should have a correspondence semantics. They do. Whatever may be

58

The Correspondence Theory

true of other domains of discourse, empirical discourse about natural kinds is permanently obese.

Bibliography Boyd, R„ 1980. "Materialism Without Reductionism: What Physicalism Does Not Entail." In N. Block (ed.), Readings In Philosophy of Psychology, vol.1. Cambridge: Harvard University Press. — . 1985a. Lex Orendi est Lex Credendi. in Churchland and Hooker (eds.) Images of Science: Scientific Realism Versus Constructive Empiricism. Chicago: University of Chicago Press. — . 1985b. "Observations, Explanatory Power, and Simplicity." In P. Achinstein and O. Hannaway (eds.) Observation,Experiment, and Hypothesis In Modern Physical Science. Cambridge: M I T Press. — . 1988. "How to be a Moral Realist." in G. Sayre McCord (ed) Moral Realism. Ithaca: Cornell University Press. — . 1989. "What Realism Implies and What It Does Not" Dialectica. — . 1990a. "Realism, Conventionality, and 'Realism About'" in Boolos, ed. Meaning and Method. Cambridge: Cambridge University Press. — . 1990b. "Realism, Approximate Truth and Philosophical Method" in Wade Savage, ed. Scientific Theories, Minnesota Studies in the Philosophy of Science vol. 14. Minneapolis: University of Minnesota Press — . 1991. "Realism, Anti-Foundationalism and the Enthusiasm for Natural Kinds." Philosophical Studies 61: 127-148. —. 1992. "Constructivism, Realism, and Philosophical Method." in John Earman, ed. Inference , Explanation and Other Philosophical Frustrations. —. 1993. "Metaphor and Theory Change" (second version) in A. Ortony (ed.) Metaphor and Thought, 2nd Edition. New York: Cambridge University Press. — . 1995. "Postscript: Materialism and Realism in Metaethics" (postscript to "How to be a Moral Realist") in Paul K. Moser and J.D. Trout, eds. Contemporary Materialism A Reader. London and New York: Routledge. — . 1999a. "Kinds as the "Workmanship of Men": Realism, Constructivism, and Natural Kinds." Proceedingsm of the Third International Congress, Gesellschaft fur Analytische Philosophie. Berlin: de Gruyter. — . 1999b. "Homeostasis, Species, and Higher Taxa," in R. Wilson, ed. Species·. New Interdisciplinary Essays. Cambridge: M I T Press. —. 1999c. "Kinds, Complexity and Multiple Realization: Comments on Millikan's 'Historical Kinds and the Special Sciences'" Philosophical Studies. Ereshefsky, M., ed. 1992. The Units of Evolutiow.Essays on the Nature of Species. Cambridge: M I T Press. Field, H . 1973. "Theory Change and the Indeterminacy of Reference." Journal of Philosophy (70): 462 481. Fine, A. 1984. "The Natural Ontological Attitude." i n j . Leplin (ed.) Scientific Realism. Berkeley: University of California Press. Goodman, N. 1973. Fact Fiction and Forecast, 3rd edition. Indianapolis and New York: Bobbs Merrill. Kuhn, Τ. 1970. The Structure of Scientific Revolutions, 2nd edition. Chicago: University of Chicago Press. Lettvin, Maturana, McCulloch and Pitts. 1959. "What the Frog's Eye Tells the Frog's Brain." Proceedings of the IRE (47), 1940-1951. Locke, J. 1689/1975. An Essay Concerning Human Understanding. Oxford: Oxford University Press. Millikan, R. G. 1984. Language, Thought, and Other Biological Categories. Cambridge: M I T Press. Psillos, S. 1999. Scientific Realism·. How Science Tracks Truth. New York and London: Routledge. Putnam, H. 1972. "Explanation and Reference." in G. Pearce and P. Maynard, eds. Conceptual Change. Dordrecht: Reidel. — . 1975a. "The Meaning of 'Meaning'." in H. Putnam, Mind, Language and Reality. Cambridge: Cambridge University Press. — . 1975b. "Language and Reality." in H. Putnam, Mind, Language and Reality. Cambridge: Cambridge University Press. — . 1978. "Realism and Reason," in Putnam, Meaning and The Moral Sciences. London and New York: Routledge and Kegan Paul. — . 1980. "Models and Reality" Journal of Symbolic Logic 45, 464-482. — , 1983a. "Vagueness and Alternative Logic." in H. Putnam, Realism and Reason. Cambridge: Cambridge University Press.

Truth Through Thick and Thin

59

—, 1983b, "Why There Isn't a Ready Made World." in H. Putnam, Realism and Reason. Cambridge: Cambridge University Press. Quine, W. V. O. 1969. "Natural Kinds." in W.V.O. Quine, Ontological Relativity and Other Essays. New York: Columbia University Press. Tarski, A. 1951. "The Concept of Truth in Formalized Languages." in Tarski, A. Logic, Semantics and Metamathematics. New York: Oxford University Press. Wilson, R. 1999. Species·. New Interdisciplinary Essays. Cambridge: MIT Press.

The Metaphysics of Deflationary Truth MICHAEL DEVITT

1. Introduction What exactly is deflationary truth and how does it differ from correspondence truth? This question has been much discussed and yet, I shall argue, the answer remains unclear. That unclarity arises from insufficient attention to the distinction between the metaphysics of truth and the linguistics of the truth term·, hence from insufficient attention to what deflationary theories say, or should say, about the metaphysical issue. In arguing this, I shall emphasize that deflationism is similar to a sort of "nonfactualism".

2. Four Problems There are four related problems in locating the difference between the deflationary theory and the correspondence theory. The first is that the two theories have opposite focuses. Whereas the focus of the correspondence theory is on the nature and role of truth, the focus of the deflationary theory is on the nature and role of the truth term; for example, of 'true'. The former focus is metaphysical, the latter, linguistic. So, an awful lot of what deflationists say does not bear directly on what the correspondence theorists say, and vice versa. A simple explanation of this difference in focus - too simple as we shall soon see is as follows. Deflationism is really a sort of eliminativism, or antirealism, about truth: it deflates truth itself. We might say, very roughly, that according to deflationism, there is no reality to truth. Since there is no reality to truth there is nothing positive to be said about the nature of truth. However, unlike some early eliminativists, deflationists have no objection to the use of the truth term. Indeed, they are enthusiastic about the term and have a great deal to say about its linguistic role and semantics.1 In contrast, correspondence theorists are realists about truth and therefore struggle to explain its nature. But, for them, the truth term is just another one-place relational predicate like, say, 'warranted' or 'patriotic' - with the standard sort of semantics of such predicates. This semantics is likely to start from the assumption that the term denotes the

1

T h e "semantics" of a term concerns its meaning. So also does its "linguistics" but the latter may also concern other aspects of the term's nature and role.

The Metaphysics of Deflationary Truth

61

property truth or applies to all true things. This is so unexciting as to be hardly worth saying and the theorists are not usually inclined to say anything more exciting. In sum, the deflationist has little to say about the metaphysics of truth but much to say about the linguistic role of 'true', whereas the correspondence theorist has a lot to say about the metaphysics of truth but little to say about the linguistics of 'true'. I am here describing a real difference in focus between the two theories. The second problem in distinguishing the theories is that this difference is often not apparent. Discussions of deflationism tend to blur the distinction between the linguistic and the metaphysical. In particular, remarks that should be about the truth term are often presented as being about truth: there is use/mention sloppiness, even confusion. So it can often seem that discussions are talking about truth when they are not really. The third problem is that when discussions of deflationism do address the metaphysical issue, rather than merely appearing to when addressing the linguistic issue, what is said is often unsatisfactory. And this is not surprising because it turns out to be rather hard to capture the deflationary metaphysics of truth. That is the fourth problem. A sign of this problem is that my characterization of the metaphysics of deflationism in describing the first problem really is very rough. To appreciate the third and fourth problems it helps to notice that deflationism is similar to "nonfactualism" in ways to be explained (section 4). Despite these four problems, the difference between deflationary and correspondence theories on the linguistic issue of the truth term is relatively clear. Not so, the difference on the metaphysical issue of truth. As a result of the four problems there is a good deal of uncertainty, if not confusion, over the difference between deflationary and correspondence views of the nature of truth. This is serious because this metaphysical difference is deeper than the linguistic one: it is explanatorily prior. Section 3 will be concerned with the linguistics of the truth term. Sections 4 to 6 will be concerned with the deflationary metaphysics of truth and the third and fourth problems. A standard characterization of the deflationary metaphysics will be criticized and a better one attempted. These accounts of the linguistic and metaphysical issues are necessary background for appreciating evidence in section 7 of the very tricky second problem: a use/mention sloppiness that obscures the real metaphysics of deflationism. The uncertainties and confusions arising from these four problems will emerge as we go along but will be particularly prominent in section 7. In talking of uncertainty and confusion I speak from bitter experience, for I am only too well aware that my own writings on truth have provided some examples.2

3. The Truth Term The deflationists have some very interesting things to say about the truth term. They have persuasively demonstrated that the term has an extremely useful "logical" or "ex-

2

For example, 1991; 1997: 3.4.

62

The Correspondence Theory

pressive" role. Thus, suppose that Jack says, "We all lie about our sex lives," and Jill replies, "That is true." Intuitively, the role of the truth term here is to enable Jill to "say the same thing" as Jack without repeating his words (and whilst admitting his priority) . Attention to such examples encourages the simplest deflationary theory, the "redundancy" theory, for they make it seem as if we could dispense with the truth term altogether. However, other examples show that the term is very useful. It enables us to assert briefly something that may otherwise be tedious, if not impossible, to assert. Suppose that Imogen wishes to express general but qualified agreement with a certain article. She can say simply, "Most of what that article says is true." Consider what would be required to say this without using 'true'. Her claim entails that at least half the claims in the article are true, but is not specific about which half. So her claim is equivalent to a long disjunction of conjuncts, each conjunct consisting of a different set of more than half the claims in the article. If she could remember all the claims, she could, in time, manage to express this disjunction. If not, she needs the truth term. So does a person who has forgotten Goldbach's Conjecture but nevertheless wants to express agreement with it. He can say, "Goldbach's Conjecture is true." A person who has lost track of all the utterances of the Great Helmsman can nevertheless express her commitment, 'Everything Chairman Mao said was true'. Without the truth term, she faces the impossible task of asserting an infinite conjunction. So also does a logician in asserting each instance of a schema that has an infinite number of instances. 3 The truth term can play its logical role because it yields equivalences like the classic one between "Snow is white is true' and 'Snow is white'. When the term is attached to the quotation name of a statement it yields a statement that is equivalent to that statement: it undoes the effect of quotation marks. (Attention to this led to the name, 'the disquotational theory of truth'.) Indeed, when the truth term is attached to any device for referring to a statement it yields a similar equivalence; it is a "denominalizing" device. Thus Jill's remark, "That is true," is equivalent to Jack's, "We all lie about our sex lives." If I were to say, "Jack's remark is true," my remark would be equivalent to Jack's. The person who said, "Goldbach's Conjecture is true," said something equivalent to the Conjecture. In general, the deflationary view supports "the equivalence thesis": all appropriate instances of the "equivalence schema" s is true iff ρ hold, where an appropriate instance substitutes for 'p' a translation of the statement referred to by what is substituted for 's\ 4 What is the "meaning" of the truth term? The deflationists have offered a variety of answers. Thus, Paul Horwich (1990) proposes a "minimal" theory according to which 'true' is an unusual "logical" predicate implicitly defined by its use in the appropriate

3 4

The expressive power that we get from the truth term could also be obtained by introducing sentence variables into our language; see Horwich 1990, pp. 4-5n for an interesting discussion. This is rough; for example, we need to guard against the semantic paradoxes and allow for indexicals. The formulation talks of translation because an appropriate instance might refer to a statement that is not in the language of the instance; e.g. 'Schnee ist weiss' is true iff snow is white.

63

The Metaphysics of Deflationary Truth

instances of the equivalence schema. Dorothy Grover (1992) urges a "prosentential" theory according to which 'true' is not a predicate at all. Rather it is a syncategorematic part of an anaphoric "presentence," where presentences are to sentences as pronouns are to nouns. This ingenious theory has the unhappy consequence that 'that' in 'that is true' does not refer to some statement, as one would naturally suppose (and as I supposed in introducing the equivalence thesis). This led Robert Brandom to propose a variation on the prosentential theory that avoided this consequence: the truth term should be treated as a presentence forming operator (1988: 88-90). So much for deflationary views of the meaning and role of the truth term. What are correspondence theorists to make of this? The important thing to notice is that they can, and should, go along with most it. Certainly, they cannot go along with a deflationary theory of meaning of the truth term, whether of the Horwich, Grover, Brandom, or any other variety. They think that the term has the standard semantics of a one-place relational predicate, very likely explained in terms of reference to the truth property or to true statements, as I noted. Still they can and should accommodate the rest of the deflationary story. In particular, they should accept the equivalence thesis: that is a constraint on any theory of the truth term. And if the correspondence theory meets that constraint it can account for the logical role of 'true' that the deflationists have so persuasively demonstrated.5 So although the correspondence theorist disagrees with the deflationist over the meaning of the truth term, he should agree that the term has the logical role explained by the deflationist. There will probably be one other important disagreement. The deflationist will insist that the truth term does not have any role other than the logical one; in particular it does not have the "descriptive" role of a normal predicate. The correspondence theorist is likely to think that the term has a substantial descriptive role in some theory of the world. These linguistic differences between a deflationary and a correspondence theory over the truth term are striking and obvious. The metaphysical differences between the two theories over the nature of truth are much less so. Yet the metaphysical differences are explanatorily prior because they largely motivate the linguistic ones.

4. The Deflationary

Metaphysics

of Truth:

the

Problem

We have seen what the deflationists say about the truth term, but what is their view of truth? Where do they stand on the metaphysical issue? I have said that deflationism is a sort of eliminativism or antirealism, and roughly characterized it as denying that there is any reality to truth (section 2). But the inadequacy of this is apparent when we note that deflationists are as ready to talk about statements being true as correspondence theorists; thus many deflationists will say that Jacks remark is true, because to say this is just to express the common belief that we all lie about our sex lives; and they will all

5

As Mark Lance in effect, points out (1997: 184-5).

64

The Correspondence Theory

agree that 'Snow is white' is true, because to say this is just to say that snow is white. So what does their antirealism consist in? The focus of deflationist literature is not on answering this question and the little the literature says is often unsatisfactory; that was our third problem in section 2. The question turns out to be rather hard to answer; that was our fourth problem. To appreciate these two problems it helps to realize that analogous problems arise elsewhere. For, deflationism about truth is similar to the "nonfactualism" exemplified by "noncognitivism" about morals, "projectivism" about causality, positivistic instrumentalism about science, and Simon Blackburn's "quasi-realism" (1984, 1993a, 1993b). 6 Characterizations of the metaphysics of nonfactualism also tend to be unsatisfactory and it is difficult to give a satisfactory one. I have discussed these problems for nonfactualism elsewhere and will draw on those discussions in what follows. 7 Deflationism has two defining features of this kind of nonfactualism. The first is at the linguistic level and is very explicit in the literature. Nonfactualism in an area has a revisionist view of the language in that area: the language is not "descriptive" as we would naturally take it to be. This view is expressed in a variety of ways, some rather unsatisfactory, but the key idea is clear: terms that appear to be predicates in the area do not have the standard semantics of a normal predicate; perhaps they are not predicates at all. Because these terms are in this way "nondescriptive" they are not like a normal predicate in purporting to "describe reality"; they have some other role. Thus, the most famous nonfactualism, noncognitivism about morals, has a revisionist view of the semantics of 'good' as a result of which indicative sentences containing it are not assertions or statements. Rather, those sentences express attitudes or emotions, or prescribe norms or rules. 8 The deflationist view of the truth term, discussed in the last section, is a similar sort of revisionism. The truth term does not have the standard semantics of a normal predicate. And its role is not to describe sentences; its only role is logical or expressive. What is meant by "the standard semantics"? Typically a philosopher's standard semantics will be truth-referential but it need not be: it might be verificationist, for example. And the standard semantics of a deflationist cannot be truth-referential because, for her, truth is not explanatory (section 5). So the non-truth-referential meaning that she attributes to a normal predicate — for example, a certain sort of use condition - she does not attribute to the truth term. Despite the linguistic similarities between deflationism and nonfactualism, there are important differences which should make one reluctant to treat deflationism as a spe-

6

7 8

Boghossian 1990a and 1990b argue that deflationism is inconsistent with nonfactualism about an area and that deflationism itself is incoherent. For some responses, see Devitt 1990, Devitt and Rey 1991, and Soames 1997. 1996; 1997: 307-20. The former includes a brief, and it now seems to me, somewhat mistaken discussion of deflationary truth (pp. 169-70). See, e.g., Ayer 1952: 103, 107; Sayre-McCord 1988: 4-5, 8; Boghossian 1990a: 160-1, 164; Blackburn 1993a: 3, 60; 1993b: 365; Haldane and Wright 1993b: 11-12; Hale 1993: 337, 340.

The Metaphysics of Deflationary Truth

65

cies o f nonfactualism. First, the deflationist's "expressive" role for 'true' is nothing like the noncognitivist's "expressive" role for 'good': the former is logical, the latter emotive. Second, the noncognitivist holds that because 'good' is nondescriptive, sentences o f the form 'x is good' are not factual. In contrast, the deflationist does not hold that because 'true' is nondescriptive, sentences o f the form 'S is true' are not factual. For her, whether these sentences are factual depends on whether 5 is factual. So, if S is 'x is good', it is factual for a deflationist who is a cognitivist but not factual for one who is a noncognitivist. T h e second defining feature o f nonfactualism is at the metaphysical level and is often more implicit than explicit in the literature. Nonfactualism in an area is antirealist about that area. Thus noncognitivists are antirealist about goodness. Deflationists are similarly antirealist about truth. Consider the problem o f characterizing the antirealism o f nonfactualism. T h e most straightforward way o f characterizing antirealism in general, using the ordinary language for denying ontic commitment, obviously does not capture the metaphysics o f nonfactualism. Thus the noncognitivist does not claim that there are no good people, right actions, and so on. She is as ready as the realist to say, "This person is good" and "That action is right," for she thinks that these utterances express appropriate emotions or prescriptions. We have already made the analogous point about deflationism: the deflationist does not claim that there are no true statements. T h e nonfactualist and the deflationist talk like a realist but give that talk a revisionist interpretation. 9 This is what poses the problem o f distinguishing these doctrines from realism at the metaphysical level; cf. our fourth problem. But perhaps the problem is illusory. Maybe my confident claim that nonfactualism and deflationism are antirealist is mistaken. Perhaps the focus o f these doctrines is so linguistic because they really have no commitment one way or the other on the metaphysical issue. I f this were so our enterprise o f attempting to characterize their antirealism would be misguided. There are two reasons why this dissolution o f the problem must be rejected. T h e first is that the doctrines are presented in opposition to realist views; thus, deflationists oppose correspondence truth. And, despite the linguistic focus, the doctrines are accompanied by claims that are clearly intended to be antirealist even if, as we shall see, the claims are often not adequate to the intention. T h e second reason is that an antirealist metaphysics is needed to motivate the revisionist view o f language urged by these doctrines. I f there were not something problematic or defective about the area o f reality that 'true' or 'good' appear to concern why suppose that they do not have the standard semantics o f a descriptive predicate? O f course, the semantic revisionism is typically supported by some purely linguistic considerations: evidence o f a nondescriptive role for the language in question. Thus, deflationists are motivated by the logical role o f the truth term and noncognitivists by the action-guiding role o f moral language. But what is to stop language covered by the standard se-

9

Note that the nonfactualist is not speaking a different language from the factualist. Rather, she has a different theory o f the language that they both speak.

66

The Correspondence Theory

mantics from playing these roles? Indeed, we have already suggested that a truth term with the standard semantics could play the logical role (section 3). So the antirealist metaphysics is still needed to make the standard semantics for this language unattractive. It is needed to show that the language does not have a descriptive role as well as the role emphasized by nonfactualism and deflationism. Behind the linguistic facade of these doctrines must lie an antirealist metaphysics. It is because deflationism's metaphysics is needed to motivate its semantics that the metaphysical difference between deflationism and the correspondence theory is explanatorily prior to the linguistic difference. 10 So the problem of characterizing the antirealism of nonfactualism and deflationism remains. The most straightforward characterization is obviously hopeless and may have no adherents. However, another simple characterization is popular: the doctrines are said to deny that there are any properties in the area in question. 11 Thus noncognitivism denies that there is a property of goodness and deflationism that there is one of truth. As soon as we look carefully at this popular characterization, we see that it cannot be satisfactory. This is our third problem with deflationism and the analogous problem with nonfactualism. The characterization is unsatisfactory because it overlooks the extent to which a philosopher's attitude to the metaphysics characterized might reflect a position on the general issue of realism about properties rather than on the particular problematic area of reality that is the concern of the nonfactualist or deflationist; for example, rather than a position on morality or truth. Thus, consider a nominalist. She will agree that there are no properties of goodness and truth because she thinks that there are no properties at all! Yet, manifestly, this alone does not commit her to nonfactualism and deflationism; to thinking that there is something especially defective about the realms of morals and truth, something that motivates a revisionist semantics. She might be as realist as could be about morality and truth. Or, consider someone like David Armstrong (1978) who is a selective realist about properties. Armstrong thinks that empty predicates, disjunctive predicates, and negative predicates have no corresponding properties. He thinks that some predicates apply to the world in virtue of many properties. Most importantly, he looks to science to tell us which properties there are. Such a person might well be a reductive realist, thinking that 'good' or 'true' apply to an object in virtue of properties none of which are goodness or truth; they apply in virtue of scientifically acceptable properties. So the popular characterization fits his views even though his metaphysics of goodness and truth is quite contrary to the

10 Concerning realism issues in general, I have argued that, from a naturalistic perspective, we should always "put metaphysics first" by establishing a metaphysical base with near enough no appeal to semantics and by arguing from that base for a semantics. For we know far more about the world than we do about meanings (1997; Devitt and Sterelny 1999: 11.4, 12.4). 11 Or that there are any facts in the area. That characterization has similar problems to the one about properties. For examples of these characterizations, see Ayer 1952: 89; Wright 1988: 29-30; Sayre-McCord 1988: ix-x, 4; Brandom 1988: 90-1; Boghossian 1990a: 157-9, 161-2; Grover 1992: 14; Blackburn 1993a: 3, 52, 57; Hale 1993: 337; Railton 1993: 280; Soames 1997: 4; Lynch 1998: 112.

The Metaphysics of Deflationary Truth

67

antirealist one that we are attempting to characterize. Finally, consider the unselective realist who thinks that there is a property for each predicate. A nonfactualist might accept that 'good' is a predicate, as indeed Blackburn does (1993a: 206), and a deflationist might accept that 'true' is, as indeed Horwich does. If such a person is an unselective realist she will think that there is a property of goodness or truth, thus ^«agreeing with the popular characterization. And even if the nonfactualist denies that 'good' is a predicate, and the deflationist that 'true' is, hence that there are properties of goodness and truth, the popular characterization of their antirealism is dubious: it 'runs the wrong way'. It finds a defect in reality because of something special about language where we need to find a defect in reality to motivate the view that the language is special. 12 The general issue of realism about properties is independent of the issues of nonfactualism and deflationism. It should be possible for someone to embrace or reject the metaphysics of these doctrines whatever her position on this general issue. There should be a way of stating that metaphysics that is appropriate whatever the truth of the matter about the reality of properties. So far, then, we have made no progress characterizing the antirealism of nonfactualism and deflationism. The most straightforward statements of realism, using the ordinary language of ontic commitment, are not denied by these doctrines because they are reinterpreted so that they have no such commitment. We have just seen the failure of a characterization using more "philosophical" talk of properties. In general, the nonfactualist/deflationist practice of talking like a realist while giving that talk a revisionist interpretation makes progress hard. We are attempting a characterization of the metaphysics that must motivate the special semantic treatment that the doctrines give to a certain area of language. Yet our attempts seem doomed to vitiation by that very semantic treatment. Nonfactualism and deflationism are supposed to be a sort of antirealism and yet it seems impossible to give a metaphysical statement of their antirealism. Realism issues begin to evaporate. Indeed, Blackburn sometimes comes very close to claiming that they have evaporated (1993a: 4, 15-34, 55-9; 1993b: 368).

5. The Deflationary Metaphysics of Truth: the Solution To avoid the evaporation of realism as a metaphysical issue and to characterize the metaphysics of nonfactualism in an area, we must first find some language in that area that is not just apparently descriptive but is treated by the nonfactualist as really descriptive. We must then examine her statements using that language to find ones that disagree with realist statements about the area.

12

Richard Kirkham notes the problem that the realism issue about properties poses for the popular characterization. His solution is to characterize deflationism as the thesis "that 'true' is not a genuine predicate" (1992: 3 1 1 ) . One objection to this is that some deflationists - for example, Horwich - think that 'true' is a genuine predicate. A more serious objection is that it is a linguistic characterization and we need a metaphysical one.

68

The Correspondence Theory

I have argued elsewhere (1996: 165-70; 1997: 313-18) that two sorts of realist claim are the most promising candidates for denial by the nonfactualist. First, the typical realist offers explanations of the nature of the problematic reality in language that the nonfactualist should agree is factual. For, the realist thinks that the problematic reality is constituted by, or supervenes on, a reality that should be ««problematic for the nonfactualist. Even though the nonfactualist claims to be able to accept many sentences that seem to describe the problematic reality, taking them as expressive, prescriptive, or whatever, she does not accept the need for, or possibility of, these substantial "broadly reductive" explanations. Thus, moral realists claim that there are things about a person in virtue of which she is good, that make her good; for example, being kind, considerate, generous, honest, etc. And there are things about an action in virtue of which it is wrong, that make it wrong; for example, leading to unhappiness, being contrary to socially accepted rules, and so on. The noncognitivist must reject all such "in virtue of" claims as totally misconceived. The deflationist has a similar disagreement with the typical realist about truth. The realist will claim that there is something common and peculiar to true statements: a statement is true in virtue of some sort of correspondence relation to the world; this relation makes it true. A substantial theory is then needed to describe and explain this correspondence, a theory that may include, for example, causal theories of reference. Deflationists should reject any such reductive explanation of truth. Horwich does so in denying that truth has an "underlying nature" or some "hidden structure awaiting our discovery" (1990: 2): "being true is insusceptible to...scientific analysis" (p. 6). Grover claims that "truth talk...can be explained without appeal to any kind of analysis of the nature of truth" (1992: 3) This is not to say that the deflationist rejects all statements of the form 'p explains that S is true'. The deflationist, like everyone else, accepts the need for, and possibility of, explanations of "worldly facts" such as that snow is white, explanations that appeal to laws of nature. Suppose that Ε explains that snow is white. So, given the deflationary theory, Ε explains that 'snow is white' is true. But this sort of explanation, varying from truth to truth, is not what the correspondence theorist seeks. He seeks an explanation of what all true statements have in common, an account of "correspondence to the world." That is the sort of explanation that the deflationist must reject. The second sort of realist claim that the nonfactualist should deny concerns causal role. The typical realist thinks that the problematic reality is the cause or effect of some unproblematic reality. The nonfactualist should not accept these claims about the role of the problematic reality because on her view there is no reality that could play such a role. Thus, the typical moral realist thinks that there are causes and effects of a person being good or bad. He thinks that it is because Hitler and his associates were depraved that we believe that they were depraved. And it is because they were depraved that they behaved as they did and that millions of people died in concentration camps. The noncognitivist must reject all such explanations. Once again, the deflationist has a similar disagreement with the realist about truth. The typical realist will give truth important explanatory roles; for example, to explain the success of science or the success of people in meeting their goals; or to explain meaning, where meaning itself plays a role in the explanation of behavior. A deflation-

The Metaphysics of Deflationary Truth

69

ist must reject all such explanations and Brandom clearly does reject them all (1988: 91-2). 13 This is not to say that the deflationist cannot use the truth term in explanatory statements. The logical role of the truth term makes an explanation of the form 'p because it is true that q equivalent to one of the form /' > because q. But the appearance of 'true' in the former sort of explanation does not make truth explanatory of p. Consider an example: 'Clinton was impeached because he is hated by the religious right' can be rewritten as 'Clinton was impeached because it is true that he is hated by the religious right'. Manifestly, what is explanatory here is hatred not truth. Even where the expressibility provided by the truth predicate is essential to an explanation - because without it the explanation would be infinite - it is not truth that is explanatory. In sum, the typical realist thinks that there is a reality to truth which, like any other reality, has a nature and causal role; and that this nature and role need explanations. The deflationist reveals her antirealism by rejecting the need for and possibility of such explanations. Although she can join with the realist in accepting ordinary truth claims, she cannot join with him in his explanation of the reality which he takes those claims to describe. The deflationist should have nothing that is positive and substantial to say about truth. Sadly, this account of the distinction between realism and nonfactualism/deflationism has a flaw, reflected in the frequent uses of 'typical'. There are doubtless some philosophers who claim to be moral realists and yet join the noncognitivists in denying the need for an explanation of moral reality and in denying that this reality has any causal role: it is inexplicable and epiphenomenal. One can imagine an analogous claim from someone who sees himself as a realist about truth. Such positions are deeply antinaturalist, of course. They are also hard to motivate: Why believe in a truth or goodness that can do nothing and cannot be explained? Still, the positions are possible. And if they have a standard semantics for 'true' and 'good' they surely are realist, for they accept the straightforward statements of realism without interpreting away the ontic commitment of the statements. So, the flaw in my account is that it does not distinguish this atypical realism from nonfactualism and deflationism at the strictly metaphysical level. I suspect that this realism cannot be so distinguished. If not, we must conclude, disappointingly, that xa f u l l y capture the antirealism of nonfactualism/deflationism we have to add a little semantics: what makes these doctrines antirealist is not only their denial of explicable nature and causal role but also their adoption of a nonstandard semantics that removes the commitment from apparently straightforward statements of realism. The nonfactualist/deflationist and the atypical realist agree that in a certain area there is no reality with an explicable nature and a causal role. Despite this failure, the atypical realist holds that there is a reality in that area: the reality is simply inexplicable and epiphenomenal. In contrast, the failure motivates the nonfactualist/deflationist to

13

See also Grover 1 9 9 2 : 14. However, Horwich does claim an explanatory role for truth ( 1 9 9 0 : 45). I have argued against this ( 1 9 9 1 : 2 7 8 - 8 0 ) .

70

The Correspondence Theory

reject the reality altogether by revising the semantics for what would otherwise be straightforward statements of realism.

6. The Equivalence Thesis The difference between deflationism and the correspondence theory should emerge in their responses to a demand for an explanation of the equivalence thesis. Let us take the most famous instance of the equivalence schema as our example: 'Snow is white' is true iff snow is white. In virtue of what is this so?14 The core of the correspondence theorist's answer is a reductive theory of the nature of true statements, of what is common and peculiar to these statements. This will be an account of the relation that true statements stand in to the world. When this theory is applied to 'Snow is white' it shows that this statement is related to the world in such a way that the statement is true iff snow is white. So the theory of truth, together with facts about the statement 'Snow is white', explain why 'Snow is white' is true iff snow is white. The deflationist, in contrast, cannot accept any appeal to a theory of the nature of truth in her explanation because she dismisses the possibility of saying anything substantial about that nature. So, what explanation does she offer? Basically, none. She thinks the demand for an explanation here is misguided: that 'Snow is white' is true iff snow is white is a "brute fact" needing no explanation. However, she has something further to say to make this provocative claim palatable: a diagnosis of the error of thinking that we need an explanation here. The diagnosis moves up to "the semantic level," considering the way the brute fact is expressed. Although the deflationist denies the need to explain why 'snow is white' is true iff snow is white, she accepts the need to explain why people wrongly think that the statement "snow is white' is true iff snow is white' expresses something that needs explaining. The error arises from treating 'true' as if it were a normal descriptive relational predicate, thus taking the truth of 'Snow is white' to depend on some relation that statement has to snow being white. Once the nondescriptive meaning of 'true' is appreciated, we see that to say that 'Snow is white' is true is not to relate the statement in some way to the world but simply to say that snow is white. So, of course 'Snow is white' is true iff snow is white, just as snow is white iff snow is white. 15 No more needs to be said (unless "the logical structure of the world" is to be explained). Similarly, it might be claimed that the following are brute facts needing no explanation: that Schnee ist weiss iff snow is white; that all bachelors are unmarried; and that Hesperus is Phosphorus. However, someone might think otherwise because she failed to appreciate the relevant semantic facts: that 'Schnee is weiss' is synonymous

Note that we are asking in virtue of what 'Snow is white is true iff snow is white, not in virtue of what "Snow is white' is true iff snow is white' is true. Deflationism and the correspondence theory would give similarly different responses to the latter question, but the responses would be more complicated. Compare: "All the anaphoric [prosentential] theory of truth tells us about what it is for 'Snow is white' to be true, is that it is for snow to be white" (Lance 1997: 188).

The Metaphysics of Deflationary Truth

71

with 'snow is white'; that the meaning of 'bachelor' includes the meaning of 'unmarried'; and that 'Hesperus' and 'Phosphorus' rigidly designate the same object. The contrast between the two theories should not be that the correspondence theory must offer a substantial explanation of why 'Snow is white' is true iff snow is white where the deflationary theory offers a trivial one appealing to the meaning of'true'. 1 6 The contrast should be that the correspondence theory must offer an explanation where the deflationary theory appeals to the meaning of 'true' to explain why no explanation is necessary. The position I am attributing to the deflationist on this matter is undoubtledly hard to grasp. The position is developed and modified a little in the next section. I started this paper by mentioning four related problems in distinguishing deflationism from the correspondence theory. The first problem was a difference in focus: the focus of deflationism is on the linguistics of the truth term, the focus of the correspondence theory on the metaphysics of truth. The third problem was the unsatisfactory nature of attempts to characterize the metaphysics of deflationism and the fourth was the difficulty of such a characterization. I have said a lot about these three but nothing yet about the second problem. We now have the background to discuss it.

7. Use/Mention Sloppiness The second problem was use/mention sloppiness, even confusion, in the literature: deflationist remarks that should concern the linguistics of the truth term are often misrepresented as being about the metaphysics of truth, thus obscuring the real metaphysics of deflationism. In giving examples of this sloppiness I do not mean to suggest that all of them amount to real confusions in thinking. Some surely are just insignificant carelessness or convenient rhetoric. Still I want to show, first, how pervasive the sloppiness is. Second, I want to make it plausible that there are some cases of real confusion: that what should be a theory of the truth term is really being taken as a theory of truth, not simply carelessly expressed. This shows, it seems to me, how very difficult it is to handle the use/mention distinction in discussing truth. In the light of our discussion so far, it is easy to spot the sloppiness. The deflationist is talking about truth itself, and saying something appropriate, when she denies that truth has a nature or causal role that needs or can have an explanation (section 5). And she is talking about truth itself, but saying something inappropriate, when she denies that truth is a property (section 4). Anything else she says, particularly anything positive, that is represented as being about truth should very likely be about the truth term. 1. The problem starts with the very names of some deflationary theories. The name 'the redundancy theory of truth' implies that truth is redundant yet what is really redundant according to the theory is the truth term. Similarly, what is really disquota-

Which is what I have suggested in previous discussions (e.g. 1991: 276-7; 1997: 32-3). My mistake arose from a use/mention confusion of the sort discussed in the next section.

72

The Correspondence Theory

tional according to 'the disquotational theory of truth' is the truth term not truth. What is prosentential according to the 'the prosentential theory of truth' is not truth but a linguistic expression including the truth term. The generic name, 'the deflationary theory of truth', does refer to theories that deflate truth not the truth term, and so the name does not confuse use and mention. Still, the name is a bit misleading because only a small part of what deflationary theories actually say concerns truth. What they say mostly concerns the truth term. They deny a descriptive role for the term but emphasize other roles that were largely unnoticed or ignored by correspondence theories. On balance, deflationary theories zwflate the truth term. 2. Consider next an historically important but notoriously difficult case of deflationism: Alfred Tarski. A special difficulty is that Tarski does not see himself as a deflationist but rather, it seems, as a correspondence theorist (1956: 153, 404). On the opening page of "The Concept of Truth in Formalized Languages," Tarski variously describes his enterprise as the definition of truth, of the term 'true sentence, and of the concept of truth (1956: 152). The last two can be taken to be the same 17 but, prima facie, they are different from the first. Defining truth is a matter of explaining its nature, a metaphysical matter,18 whereas defining the term and the concept are linguistic matters. We have use/mention sloppiness. What does Tarski actually do? He defines the meaning of 'true-in-Z', where L is any of a certain range of formal languages. Does this have anything to do with explaining the nature of truth? Set aside worries arising from the fact that he has defined 'truein-Z' not 'true' and suppose that he had defined 'true'. Could that have shown anything about truth? It depends on the definition. In certain cases we can move straight from a definition of a word's meaning to an explanation of the nature of the reality that the word concerns. For example, we can move straight from defining 'bachelor' as 'adult unmarried male' to the explanation: to be a bachelor is to be an adult unmarried male. So moving back and forth between talk of defining 'bachelor' and defining bachelorhood would be an insignificant use/mention sloppiness, of interest only to pedants. But a linguistic definition licenses this move to metaphysics only if it treats the word in question as a normal descriptive predicate. Where the definition amounts to a revisionist view of the word's meaning, the definition cannot yield a substantial explanation of the nature of the reality that the word appears to concern.19 Indeed, an antire-

17

18 19

In general, I take it that our concept or notion of truth can near enough be identified with the meaning of the truth term. In "The Semantic Conception Conception of Truth," after a similar variety of descriptions of his enterprise, Tarski has this to say about his usage: "The words 'notion' and 'concept' are used in this paper with all of the vagueness and ambiguity with which they occur in philosophical literature. Thus, sometimes they refer simply to a term, sometimes to what is meant by a term, and in other cases to what is denoted by a term" (1949: 80n). As in scientifically defining water as H 2 0 ; cf. the linguistic definition of 'vixen' as 'female fox'. If a predicate is covered by a description theory, as our example takes 'bachelor' to be, it will have a definition. If it is covered by a causal theory, as words like 'tiger' very likely are, it will not have a definition. So an explanation of the nature of Fs can be derived from a theory of 'F only if 'F is both a normal descriptive predicate and covered by a description theory. (Even where the explanation of nature can be derived in this way, it should not be, in my view. W e should start with metaphysics not semantics because we know more about the world than about meanings; see note 10.)

The Metaphysics of Deflationary Truth

73

alist view of that reality is necessary to motivate the revisionist semantics (section 4). Consider an example: we could not move from a noncognitivist definition of 'x is good' as 'hoorah for x!' to an explanation of goodness as "hurrahness"; and noncognitivism is partly motivated by an antirealist view of goodness. One lesson I think that we should draw from Hartry Field's classic article, "Tarski's Theory of Truth" (1972), is that Tarski's definition of 'true' is of the revisionist sort and so, as it stands, does not show us anything substantial about truth. Tarski's use/mention sloppiness is of more than pedantic interest. Tarski's definition of 'true-in-Z' rests on list-like definitions of various referential words along the lines of the following definition of 'designate': ' N designates χ =df 'TV is 'France' and χ is France or...or Ν is 'Germany' and χ is Germany' By comparing such definitions with a similar one for 'valence', Field brings out dramatically that the definitions do not yield satisfactory reductive explanations of the nature of reference. 20 So, the definition of 'true' in terms of the referential words does not yield a satisfactory reductive explanation of the nature of truth. In the light of subsequent discussions, we can see why: the list-like definitions are essentially deflationary 21 and so could not yield anything substantial about reference. Indeed, in offering these definitions Tarski is implicitly committed to antirealism about reference: only if there were something problematic about reference would there be adequate justification for not treating the referential terms as ordinary two-place relational predicates; for not saying, for example, that 'designate' designates the relation designation., or applies pair-wise to all ordered pairs where the first member designates the second. Tarski shared the physicalism of the positivists and clearly did think that there was something problematic about both reference and truth. And that was the thought that drove his enterprise. Although Tarski seemed to view himself as a correspondence theorist about truth, the theory he actually presented is deflationary, as I think is now generally agreed. So there is a far from innocent use/mention confusion in representing Tarski's definition as a theory of truth, as Tarski and others do. 2 2 Tarski's definition tells us a lot about 'true-in-Z'. It tells us nothing about truth-in-Z, 23 because it is implicitly committed to the view that there is nothing to tell.

20

21 22

23

The definition of 'designate' does yield: for Ν to designate χ is for Ν to be 'France' and χ to be France or...or Ν to be 'Germany' and χ to be Germany'. Perhaps we could count this as an explanation of the nature of reference: talk of "nature" is not clear enough to rule this out; cf. section 6. But we cannot count it as substantial reductive explanation of the sort indicated in section 5. Also implausible. Brandom 1984 is a more appealing deflationary theory. For example, Rudolf Carnap talks ofTarskis "definition of truth" (1963: 60); Kirkham, in his impressively thorough introduction to theories of truth, nicely distinguishes the metaphysical project from the linguistic one (1992: 20-1) but then places Tarski "firmly with the metaphysical project" (p. 33). As it stands, but if we revised it by dropping its list-like definitions, then we could see it as yielding an explanation of truth in terms of reference, as Field points out. If this were then supplemented by a substantial theory of reference, we would have a correspondence theory of truth.

74

The Correspondence Theory

3. Scott Soames begins "What is a Theory of Truth?" (1984), an important defense ofTarski from Fields criticisms, with the report: "Alfred Tarski's theory of truth and its successors...are commonly believed by philosophers to provide analyses of the nature of truth" (p. 411). If Soames is right that this belief is common — and I think he is the use/mention confusion I have just noted is widespread. Soames does not share my view that this belief misrepresents Tarski's achievement but he notes that "there is considerable doubt about whether, or in what sense, [Tarksi's theory] is a theory of truth." He goes on: One main reason for this uncertainty is the difficulty of determining what a theory of truth ought to be. Generally, theories of truth have tried to do one or the other of three main things: (i) (ii) (iii)

to give the meaning of natural-language truth predicates; to replace such predicates with substitutes, often formerly defined, designed to further some reductionist program; or to use some antecedently understood notion of truth for broader philosophical purposes... (p. 4 1 1 )

This is striking. Suppose that we wondered what a theory of, say, genes tries to do. Two things occur: (a) it tries to describe the role of genes - state the laws about genes - which is what Mendelian genetics does; (b) it tries to say what genes are — explain their nature - which is what molecular genetics does. Now explaining the role of genes is, near enough, analogous to Soames' (iii). But explaining what genes are has no analogue on Soames' list! The metaphysical task of explaining what truth is, which is surely what correspondence theorists and many others were trying to do, has become one or other of the two linguistic tasks, (i) and (ii). Use has become mention. 4. The theory Horwich proposes in his influential book, Truth (1990), is explicitly deflationist. Yet he talks positively of "the minimalist function" of truth (p. xii), of "the entire conceptual and theoretical role of truth" (p. 6), of "the properties of truth" (p. 26), and of "all the facts involving truth" (p. 7). Strictly speaking, on his antirealist theory, truth can have no function, role, or (nontrivial) property, and cannot be involved in any facts. The truth term is what has the function, role, properties and involvement. Horwich claims that his minimal "theory of truth... involves nothing more than the equivalence schema" (p. 12); it is "what is expressed by [the schema's] uncontroversial instances" (p. 7). 2 4 But this is misleading at best for his theory of truth is really to be found in various negative remarks about the nature of truth, some of which I have quoted (section 5). O f course, the word 'true' is used not mentioned in each instance of the equivalence schema and this might suggest that these instances explain the nature of truth. But that suggestion treats 'true' as a normal descriptive predicate which is precisely what Horwich, like all deflationists, denies: he has a revisionist the-

24

Consider also: " T h e basic idea for deflationary theories o f truth...is roughly that there is no more to truth than the equivalence thesis' (Devitt 1997: 30); "The deflationist tells us...: Truth's nature', such as it is, is (pretty much) exhausted by the equivalence of a claim ρ with the claim ρ is true" (Richard 1 9 9 7 : 57).

The Metaphysics of Deflationary Truth

75

ory of the truth term. 25 This theory of the truth term is what really "involves nothing more than the equivalence schema": it holds that every fact about the role of the term can be explained simply by taking the meaning of the term to be implicitly defined by its use in the appropriate instances of the equivalence schema. 26 The equivalence schema has nothing to do with his theory of truth, everything to do with his theory of the truth term. 27 An analogy with goodness may help. Suppose that a noncognitivist were to talk positively of the function, role and properties of goodness, and of the facts involving goodness. Suppose that she claimed to be giving a theory of goodness that was quite clearly based on her views about the nondescriptive meaning and expressive role of 'good'; for example, she claimed that her theory of goodness involves nothing more than the view that to say that χ is good is just to express a pro-attitude toward x. It would be obvious that she was misdescribing her position: on her view, such remarks should really apply only to 'good' not goodness. For her theory of goodness is, roughly, that there isn't any; more precisely, it is the view suggested in section 5. 5. Stephen Leeds' "Theories of Reference and Truth" (1978), and Robert Brandom's "Pragmatism, Phenomenalism, and Truth Talk" (1988) 28 are two of the best brief presentations of deflationary truth. Leeds sketchs a disquotational theory of the sort famously suggested by Quine (1970). He comments: "What we have sketched is not a theory of truth...but a theory of the concept of truth" (1978: 122). But then he spoils this assessment by claiming that his account explains "facts about truth-in-English" and "what we ordinarily say about truth" (p. 123). It doesn't. But it might explain facts about 'true' in English and ordinary uses of'true'. Brandom takes the "central theoretical focus" of deflationism to be "on what one is doing when one takes something to be true, that is, our use of 'true'." He goes on: "It is then denied that there is more to the phenomenon of truth than the proprieties of such takings" (1988: 77). But, strict-

25

Horwich is happy to go along with the unselective realist about properties, holding that 'true' is a predicate referring to a "logical" property (p. 38). So, in that respect, instances of the schema are "about truth." But it is still a mistake to think that the instances say anything substantial about the nature of truth. Truth as a logical property has no nature open to reductive explanation. Indeed, it has no properties except trivial ones like being logical and being a property. And although we might perhaps take the equivalence schema to yield an explanation of truth, it does not yield a substantial reductive one; cf note 20. 26 On this see pp. 34-7 (abstracting from the conflation of meaning with a speaker's knowledge of meaning; cf. Devitt and Sterelny 1999: ch. 8). 27 Kirkham takes Horwich at his word and so sees his remarks about the equivalence schema as an answer to the metaphysical question about the nature of truth (1992: 339). As a result of this, and Horwich's acceptance that truth is a property, Kirkham does not classify Horwich as a deflationist. Soames takes "the leading idea" of deflationism to be that the equivalence schema "is in some sense definitional of the notion of truth" (1997: 4). The talk of "notion" makes this appropriately linguistic. But the talk immediately follows the inappropriately metaphysical: "the equivalences...are crucial in explaining what truth consists in" (p. 3). And it is immediately followed by the claim that the statement, "there is no such property as truth," which is straightforwardly metaphysical, is a variation of it (p. 4). 28 Brandom's excellent paper is sadly neglected; it gets no mention, for example, in Kirkham's encyclopedic discussion (1992).

76

The Correspondence Theory

ly speaking, on the deflationary view the proprieties are not any part of the phenomenon of truth because, roughly, there is no such phenomenon. The only phenomena are truth takings. 6. Finally, consider Marian David's Correspondence and Disquotation (1994), the most detailed and informed critique of the disquotational theory of truth available. David starts his description of the disquotational theory by claiming that it unlike, say, the correspondence theory, is "an antitheory of truth": its view is, "Truth has no nature." So far, so good. But then he continues: "The correct explanation of truth... requires less extravagant resources." The correct explanation is that "truth is disquotation" (pp. 3-4). But the disquotational view does not require less extravagant resources to explain truth, it does not require any because, properly understood, it is the view that truth does not need and cannot have an explanation. That is the respect in which it "has no nature." And disquotation does not explain truth, it explains the truth term. Not surprisingly, when David sets out to find the unextravagant disquotational theory of what it is for a sentence to be true, he finds the theory "a bit elusive" (p. 62). The core of the disquotational theory is, of course, the equivalence schema. David worries away at the schema trying unsuccessfully to find in it a theory of truth other than a correspondence theory. Sometimes he comes close to realizing that he is seeking something that the disquotationalists think is not there to be found: they think that "sentence-truth is in a sense 'nothing'" (p. 65); Strictly speaking, the question [about truth] will not even receive a response with the right logical form to count as an answer to this question, for the grammatical truth predicate does not function like an ordinary predicate... Given that the standard way of answering "What is F?"-questions does not work when it comes to truth all one can do is describe the linguistic role that the term true' plays in our language, (pp. 68-9)

Just so. Still he remains puzzled: "where does the deflationary idea that truth is nothing but disquotation come from?" (p. 69). Deflationists have given him reason to be puzzled, as we have seen. Despite what is often suggested, the disquotational view should not be that truth is nothing but disquotation. The view should be that truth is nothing. In this section, I have indicated how pervasive use/mention sloppiness is in the discussion of deflationary truth. Some of this sloppiness is surely insignificant. Yet I hope to have shown that some of it is not: a theory of the truth term is really being taken as a theory of truth. This helps to obscure the metaphysics of deflationary truth and hence the difference between deflationism and the correspondence theory described in sections 5 and 6.

8. Summary I have attempted to bring out the real difference between deflationism and the correspondence theory by emphasizing the similarity between deflationism and nonfactualism. At the linguistic level, the real difference is fairly apparent. The correspondence theorist can, and should, grant that the truth term has the logical role emphasized by

The Metaphysics of Deflationary Truth

77

the deflationist. But the correspondence theorist does not accept the deflationary view that the term has no other role: he holds that it has a descriptive role. Furthermore, he thinks that the term has the standard semantics of a one-place descriptive predicate, a view that the deflationist rejects. At the metaphysical level, the real difference between the two theories is much harder to discern. The typical correspondence theorist thinks that truth has a nature and causal role that need explaining. T h e deflationist should reveal her antirealism in the characteristic nonfactualist way by rejecting the need for and possibility of any such explanation. Finally, the metaphysical difference motivates the linguistic one, implicitly if not explicitly: it is largely because of her antirealism that the deflationist rejects a standard semantics and a descriptive role for the truth term. I have located the difficulty in discerning the metaphysical difference in four problems. T h e first problem is a difference in focus: the focus of deflationism is on the linguistics of the truth term, the focus of the correspondence theory on the metaphysics of truth. T h e second problem, just illustrated in some detail, is that use/mention sloppiness in discussions of deflationism tends to obscure the real metaphysics of deflationism. T h e third problem is that when discussions do address the metaphysical issue, rather than merely appearing to when addressing the linguistic issue, what is said is often unsatisfactory. And this is not surprising because it turns out to be rather hard to capture the deflationary metaphysics of truth, as it is to capture the metaphysics of nonfactualism. That is the fourth problem. 2 9

References Armstrong, D. M. 1078. A Theory of Universals: Universals and Scientific Realism Volume II. New York: Cambridge University Press. Ayer, A. J. 1952. Language, Truth and Logic. New York: Dover Publications, Inc. Blackburn, Simon 1984. Spreading the Word. Oxford: Clarendon Press. Blackburn, Simon 1993a. Essays in Quasi-Realism. New York: Oxford University Press. Blackburn, Simon 1993b. "Realism, Quasi, or Queasy?" In Haldane and Wright 1993a: 365-83. Boghossian, Paul A. 1990a. " T h e Status of Content." Philosophical Review, 99, 157-84. Boghossian, Paul A. 1990b. " T h e Status of Content Revisited." Pacific Philosophical Quarterly 71: 264-78. Brandom, Robert 1984. "Reference Explained Away." Journal of Philosophy 81: 469-92. Brandom, Robert 1988. "Pragmatism, Phenomenalism, and Truth Talk." In Midwest Studies in Philosophy, Volume XII: Realism and Antirealism, eds Peter A. French, Theodore E. Uehling, Jr., and Howard K. Wettstein. Minneapolis: University of Minnesota Press: 75-94. Carnap, Rudolf 1963. "Intellectual Autobiography." In The Philosophy of Rudolf Carnap, ed. P. A. Schilpp. La Salle: Open Court: 1-84. David, Marian 1994. Correspondence and Disquotation: An Essay on the Nature of Truth. Oxford: Oxford University Press. Devitt, Michael 1990. "Transcendentalism about Content." Pacific Philosophical Quarterly 71: 247-63.

29

Special thanks to Hartry Field whose doubts about the main theses of this paper have led to many changes. I am grateful for comments at the University of Sydney and the Graduate Center of City University of New York when versions of the paper were presented. My thanks also to Marian David, Paul Horwich, Mark Lance, Bill Lycan, Paul Pietroski and Georges Rey for helpful comments. Devitt 2001 is an expanded version of this paper.

78

The Correspondence Theory

Devitt, Michael 1991. "Minimal Truth: A Critical Notice of Paul Horwich's Truth." Mind and Language 6: 273-83. Devitt, Michael 1996. "The Metaphysics of Nonfactualism." In Philosophical Persectives, 10: Metaphysics, 1996, ed. James Tomberlin. Atascadero, CA: Ridgeview Publishing Company: 159-76 Devitt, Michael 1997. Realism and Truth, 2nd edn with a new Afterword. (1st edn 1984; 2nd edn 1991.) Princeton: Princeton University Press. Devitt, Michael 2001. "The Metaphysics of Truth." In The Nature of Truth, ed. Michael Lynch. Cambridge, MA: MIT Press. Devitt, Michael, and Georges Rey. 1991. "Transcending Transcendentalism: A Response to Boghossian." Pacific Philosophical Quarterly 72: 87-100. Devitt, Michael, and Kim Sterelny 1999. Language and Reality: An Introduction to the Philosophy of Language, 2nd edn. (1st edn 1987). Oxford: Blaclcwell Publishers. Field, Hartry 1972. Tarskis Theory of Truth. Journal of Philosophy 69, 347-75. Grover, Dorothy 1992. A Prosentential Theory of Truth. Princeton: Princeton University Press. Haldane, John, and Crispin Wright, eds. 1993a. Reality, Representation, and Projection. New York: Oxford University Press. Haldane, John, and Crispin Wright 1993b. "Introduction." In Haldane and Wright 1993a: 3-12. Hale, Bob 1993. "Can There Be a Logic of Attitudes?" In Haldane and Wright 1993a: 337-63. Horwich, Paul 1990. Truth. Oxford: Basil Blackwell. Kirkham, Richard 1992. Theories of Truth: A Critical Introduction. Cambridge MA: M I T Press. Lance, Mark 1997. "The Significance of Anaphoric Theories of Truth and Reference." In Villanueva 1997: 181-98. Leeds, Stephen 1978. "Theories of Reference and Truth." Erkenntnis 13: 111-29. Lynch, Michael P. 1998. Truth in Context: An Essay on Pluralism and Objectivity. Cambridge, MA: MIT Press. Popper, Karl 1968. Logic of Scientific Discovery. New York: Basic Books. Quine, W. V. 1970. Philosophy of Logic. Englewood Cliffs, NJ: Prentice-Hall. Railton, Peter 1993. "What the Non-Cognitivist Helps Us to See the Naturalist Must Help Us to Explain." In Haldane and Wright 1993: 279-300. Richard, Mark 1997. "Deflating Truth." In Villanueva 1997: 57-78. Sayre-McCord, Geoffrey 1988. "Preface" and "Introduction: The Many Moral Realisms." In Essays on Moral Realism, ed. Sayre-McCord. Ithaca: Cornell University Press: ix-xii, 1-23. Soames, Scott. 1984. "What is a Theory ο(Ίτ\ιύ\Γ Journal of Philosophy 81: 411-29. Soames, Scott 1997. "The Truth About Deflationism." In Villanueva 1997: 1-44. Tarski, Alfred 1949. "The Semantic Conception of Truth and the Foundations of Semantics." In Readings in Philosophical Analysis eds H. Feigl and W. Sellars. New York: Appleton-Century-Crofts: 52-84. Tarski, Alfred 1956. Logic, Semantics, Metamathematics: Papers from 1923-1938. Oxford: Clarendon Press. Villanueva, Enrique, ed. 1997. Truth: Philosophical Issues 8, 1997. Atascadero: Ridgeview Publishing Company. Wright, Crispin 1988. "Realism, Antirealism, Irrealism, Quasi-Realism." In Midwest Studies in Philosophy, Volume XII: Realism and Antirealism, eds Peter A. French, Theodore E. Uehling, Jr., and Howard K. Wettstein. Minneapolis: University of Minnesota Press: 25-49.

Truth, Meaning, and Reference RICHARD SCHANTZ

I "What is Truth?" asked Pilate. "Truth is correspondence, a relation to reality" answered the adherents of the most natural and most venerable of all accounts o f truth, the correspondence theory. It is based on a fundamental intuition: the statement or belief that tigers are striped is true because it corresponds to the fact that tigers are striped, and the statement that dogs can fly is false because it does not correspond to any fact. Such examples lead to the central thesis of the classical correspondence theory: a statement or belief is true just in case it corresponds to a fact, and it is false just in case it does not correspond to a fact. I am absolutely sure that Pilate would have liked this answer. Almost everybody will accept it - at least as far as it goes, that is, before the details are filled in. I am less sure what he would have said about such alternative conceptions of truth as: truth is coherence within a harmonious system o f beliefs, truth is rational acceptability in epistemic ideal conditions, truth is pragmatical utility, truth is a primitive unanalyzable property. Anyway, I will not discuss here these competing explanations of what it is for something to be true. I have argued elsewhere that all these approaches are untenable. 1 As I see it, there is only one serious competitor to the correspondence theory: deflationist or minimalist theories. There are various brands of deflationism on the philosophical market: there are Gottlob Frege's, Frank Ramsey's, Peter Strawson's, and C.J.F. Williams' variants of the redundancy theory of truth; there is the "prosentential" theory of Dorothy Grover, Nuel Belnap and John C a m p ; there is the disquotationalist account advocated by Alfred Ayer, W.V. Q u i n e , Richard Rorty, Stephen Leeds, and Michael Williams; there is Scott Soames' subtle defense of the general deflationary perspective; there is Hartry Field's version of "pure disquotational truth"; and there is last but not least Paul Horwich's "minimalism". 2 What all these various deflationary views have in c o m m o n is their conviction that what has come to be called substantive or robust theories o f truth - such as correspondence or coherence or pragmatic theories - are all on the wrong track. According to deflationism, substantive theories share the assumption that truth has an inner nature, a nature which can be analyzed in epistemic or semantic or metaphysical terms. Deflationists reject this assumption. Truth, they urge, has no underlying nature.

1 2

See Schantz 1996 See Frege 1966, [1892]; Ramsey 1927; Strawson 1950; C.J.F. Williams 1976; Grover, Belnap, Camp 1975; Grover 1992; Ayer 1936; Quine 1970, 1990; Rorty 1986; Leeds 1978; M . Williams 1986, 1999; Soames 1999; Horwich 1990

80

The Correspondence Theory

The concept of truth does not express an interesting property or an interesting relation; it has no important connections with other concepts such as reference or meaning or belief. Consequently, it should not be given a central place in our philosophical reflections. Rather, truth is a purely formal or logical concept whose correct explanation requires far less extravagant conceptual resources than the proponents of substantive theories suppose. The ambitious aim of deflationism is to deflate the overinflated theories developed by their substantivist opponents — in particular by the correspondence theory. And, indeed, a satisfactory form of the correspondence theory must go beyond the expression of fundamental intuitions which nobody denies. The key concepts it invokes must be fleshed out. In the first instance, it should provide an account of three basic aspects: the truth bearer, the correspondence relation, and the truth maker - the fact or the reality to which the truth bearer corresponds. In my view, it is statements, in the sense of the contents of declarative sentences, that are the primary truth bearers - the sort of thing that can be true or false. Other popular candidates such as utterances, i.e. sentence tokens, are true only in a derivative sense; they are true just in case the statement made by the production of the sentence is true. Further, I take it that truth makers are rather familiar entities: facts, or states of affairs, are simply objects having certain properties and standing in certain relations to each other at various spatiotemporal locations. For example, the statement that the cat is on the mat is made true by the fact that the cat is on the mat, or, more simply, by the cat's being on the mat. So the acknowledgment of facts does not involve introducing any new entities. We have the best reasons for believing that facts, so conceived, are parts of the natural world because we can perceive them; for example, we can see that the cat is on the mat. Since facts do not inflate our ontology, even a philosopher with a taste for desert landscapes need have no qualms about accepting them. And to accept facts as parts of the natural world is also not to commit the "logically fundamental type-mistake" of turning facts into things or complexes of things, as Strawson once maintained. 3 I insist that facts are categorially different from things, and yet, like things, they are to be found in the world out there. But I need not say anything more about facts here. 4 They will not loom large in my discussion. What about the correspondence relation? What is it for a statement to correspond to a fact? Correspondence does not require something as pretentious as a relation of picturing or mirroring between statements and facts, as early Wittgenstein und Russell once thought. 5 Ordinary reference is enough. Despite the many objections that have been raised against the idea, I will argue that the notion of correspondence can be spelled out as reference. Reference itself I take to be a causal, and hence a naturalistically respectable relation. So a semantic theory is required to show how the statementto-fact correspondence can be constructed from referential relations between the con-

3 4 5

Strawson 1950, 134 See my 1996 for an attempt to rehabilitate facts, or states of affairs, as the ontological relatum of the correspondence relation. Wittgenstein 1922; Russell 1912

Truth, Meaning, and Reference

81

stituents of statements, i.e. what is expressed by singular and general terms, and the things, properties and relations they refer to or apply to. Thus, in my estimation, an adequate version of the correspondence theory ought to explain the truth of a statement in terms of referential relations between its component parts and features of the external world.

II When truth is our subject, an instructive, modern starting point is Alfred Tarski s semantic conception of truth. The more so, since Tarski himself regarded his theory as an elaboration of the traditional correspondence theory of truth. For example, in "The Establishment of Scientific Semantics" he says: We shall understand by semantics the totality of considerations concerning those concepts which, roughly speaking, express certain connexions between the expressions of a language and the objects and states of affairs referred to by these expressions. [...] The concept of truth also - and this is not commonly recognized - is to be included here, at least in its classical interpretation, according to which "true" signifies the same as "corresponding with reality".^

And again: We regard the truth of a sentence as its "corresponding with reality".7

In "The Semantic Conception of Truth" he tells us that he does not aim to assign a new meaning to an old word, but "to catch hold of the actual meaning of an old notion". 8 He hopes his definition to do justice to the intuitions of the "classical Aristotelian conception of truth", 9 as expressed in Aristotle's famous dictum: To say of what is that it is not, or of what is not that it is, is false, while to say of what is that it is, or of what is not that it is not, is true.

And in "The Concept of Truth in Formalized Languages" he expresses the semantic definition, informally, in the following words: A true sentence is one which says that the state of affairs is so and so, and the state of affairs indeed is so and so. 10

To be sure, Tarski regards none of these formulations as "sufficiently precise and clear". He thinks that such expressions as "agreement" or "correspondence with reality" are vague and metaphorical. But he sees it as the chief task of his semantic definition to make the intuitive meaning of these formulations clearer and to give them a correct

6 7 8 9 10

Tarski 1956b, 401 Ibid., 404 Tarski 1944, 341 Ibid., 342 Tarski 1956a, 155

82

The Correspondence Theory

form. In particular, he wants to preserve the correspondence intuition that a sentence which claims that things are so-and-so is true if and only if things really are so-and-so. Let us look a bit closer at Tarki's semantic point of view. As is well-known, it was his prime aim to show that the concept of truth, if carefully employed, is a consistent concept, a concept which does not involve us in semantic paradoxes. Therefrom resulted his project to provide an exact definition of truth, a definition which must fulfill two conditions: firstly, it must be "materially adequate", and secondly, it must be "formally correct". The first of these conditions restricts the possible content, the second the possible form, of any acceptable definition. To fulfill the material adequacy condition Tarski introduces his famous Convention T: An acceptable definition of truth for a language must have among its consequences all instances of the following schema: s is true if and only if p where "s" is to be replaced by a standardized name of any sentence of the object language, the language for which truth is being defined, and "p" is to be replaced by that same sentence or its translation, depending on whether or not the object language is contained in the metalanguage as a proper part. To use Tarski's example, an instance of this schema would be: "Snow is white" is true if and only if snow is white. Tarski calls every equivalence of form Τ a "partial definition of truth", which explains, as he says, "wherein the truth of this one individual sentence consists". 11 And he often characterizes his project as providing a definition equivalent to the logical conjunction of all such instances of T, all such " T'-sentences" as they are usually called nowadays, for the language under study. Indeed, Tarski mentions that if the language consisted of only a finite number of sentences, Convention 7*could be satisfied by a definition that simply listed a 7"-sentence for each sentence in the language. Any interesting language, however, has a potential infinity of sentences; for such a language a list-like definition cannot be given. We cannot compile an infinite list. For such languages the definition must take another form. In the case of languages whose only complex sentences are truth functions of their components, it is possible to define truth for complex sentences directly in terms of truth for elementary sentences. But with languages with quantificational structure this direct method founders because the components of complex sentences are no longer themselves necessarily closed sentences. The use of variables, which is necessary for quantificational languages, has the consequence that open sentences and value assignments to free variables must be taken into account. In such languages, complex sentences may be produced from open sentences by binding the variables occurring in them. It is the most distinctive feature of Tarski's definition of truth that he ingeniously solved this problem with the concept of satisfaction. Satisfaction is a relation between sentences, open or closed, and infinite sequences of objects which belong to the

11 Tarski 1944, 344

Truth, Meaning, and Reference

83

range of variables of the language. Since, in the case of quantificational languages, sentence-forming operations operate on closed as well as open sentences, it is not possible to give a recursive definition of truth itself. The essential idea of Tarski's solution is to take a roundabout course: he first constructs a recursive definition of satisfaction, and then defines truth on that basis. Just as closed sentences are limiting cases of open sentences, so truth is a limiting case of satisfaction. Accordingly, Tarski defines a sentence as true just in case it is satisfied by all sequences of objects. With the technical resources of set theory, the recursive definition can, finally, be converted into an explicit or eliminative definition.

Ill Tarski thought he had shown how truth can be eliminated via an explicit definition, that is, a definition containing no semantic expressions at all. By avoiding any semantic vocabulary in the definiens, he hoped to satisfy the demands of the doctrine of physicalism and the unity of science which, under the influence of Logical Positivism, dominated the philosophical scene of these days. In a brilliant article, early Hartry Field has argued, however, that Tarski did not accomplish the ambitious aim he set himself. 1 2 According to Field, Tarski took an important first step when he succeeded in reducing the concept of truth to the concept of satisfaction, or, as Field prefers to say, to the concept of primitive denotation, i.e., the semantic relation holding between the primitive descriptive expressions of the language and their referents. But the requisite second part of the reduction, the explication of primitive denotation, fails, Field claims, because it simply takes the form of a definition by enumeration - a finite and exhaustive list of assignments of terms to objects giving us no idea of how the list is generated. The non-projectible and non-explanatory character of the list-like definition of primitive denotation is reflected in the fact that if a new word were added to a language the original definition would give us no clue as to how this new case should be treated. Consequently, Field maintains that a physicalistically acceptable reduction of the primitive denotation relation requires more than an extensionally correct characterization; what is needed in addition is a characterization of the non-semantic relations between words and objects in virtue of which the former denote the latter. A genuine reduction must explain the supervenience of semantic facts upon physical facts about speakers and their relations to their environments. Tarski's definitions do not satisfy this demand. We do not have to tackle the questions, here, of what Tarski's and Field's versions of physicalism come to, and whether or not they are acceptable, in order to sympathize with the guiding idea that a satisfactory account of truth ought to break out of the circle of semantic concepts. As I see it, an account of truth in non-semantic terms is not necessarily an account of truth in physicalist terms.

12 Field 1972

84

The Correspondence Theory

Field himself thinks that something like the causal theory of reference in the style of Saul Kripke, sufficiently well worked out, can be used to explain the primitive denotation relation. The central idea of the latter theory is that a name "a" refers to an object χ if and only if our use of "a is caused by an appropriate chain of uses of "a" initiated by a baptism of χ by "a". And similarly, a predicate " F ' applies to a set of objects f if and only if our use of "F' is appropriately causally connected to members of f . Field is confident that a definite causal account of reference along these lines can provide the reductive analysis which he thinks is needed to complete the physicalist explanation of truth. And the latter, in turn, he thinks is needed if one intends to develop a genuine correspondence theory of truth. Despite the deep philosophical tasks they face, I am still quite sympathetic with causal theories of reference. Elsewhere I have given my reasons for believing that there is such a relation as reference, a robust relation, and that there is nothing in nature to constitute it apart from our causal interactions with the world. 1 3 But we have to go beyond theories in the style of Kripke. Besides the historical cause, the reliable cause of uses of words, and presumably even their teleological cause, must be taken into account. 1 4 I favour an informational approach to the problem of naturalizing intentionality - the problem of explaining how the semantic or representational properties of thoughts, and derivatively of linguistic symbols, can arise out of non-semantic, nonintentional properties and relations. Information - reliable correlation - is the stuff out of which intentionality and reference are made.

IV Modern correspondence theorists often rely in one way or another on Tarski's theory of truth. But so do modern deflationists as well. Let us start looking at the disquotational account which is surely still the most prominent deflationary view on the contemporary scene. The general idea on which it is based is that a sentence of the form "s is true" has the same meaning or the same cognitive content as the sentence s. So, according to this theory, the two sentences '"Snow is white' is true" and "Snow is white" are semantically equivalent or convey essentially the same information. But what exactly does "disquotation" mean? Let us keep to Quine, upon whose canonical formulations most adherents of this conception implicitly or explicitly fall back. Quine says: T h e truth predicate is a reminder that, despite a technical ascent to talk of sentences, our eye is on the world. This cancellatory force of the truth predicate is explicit in Tarski's paradigm: "Snow is white" is true if and only if snow is white.

13 14

Compare Schantz 1996, passim My standpoint on the nature of intentionality is strongly influenced by the seminal work of Dretske 1 9 8 1 , Fodor 1987, and Millikan 1 9 8 4

Truth, Meaning, and Reference

85

Quotation marks make all the difference between talking about words and talking about snow. T h e quotation is a name of a sentence that contains a name, namely "snow", of snow. By calling the sentence true, we call snow white. T h e truth predicate is a device of disquotation. 1 5

For Quine, the term "true" is a device for semantic ascent and semantic descent. If we go up one level and attribute truth to the sentence "Snow is white", then we attribute whiteness to snow. The truth predicate enables us to return from the level of talk about language to the level of talk about the world. The use of "true" signalizes that, though a sentence is mentioned, and we are thus operating on a linguistic plane of reference, our interest is nevertheless not directed on language but on extralinguistic reality. The ascription of truth, as it were, sweeps away the quotation marks, producing a sentence suitable for saying that snow is white. In other words, it is the function of "true" to cancel the effect of semantic ascent. That is why Quine says quite succinctly: "Truth is disquotation". 16 So the truth predicate seems to be superfluous when it is applied to a particular sentence because we could just as well use that very sentence. Quine is well aware that not every use of the truth predicate can be dispensed with in this simple way. Disquotationalists typically concede that the term "true" is not exclusively used in connection with explicitly given sentences, that is, with sentences in quotation marks; it also occurs in other types of contexts. Sometimes we want to attribute truth to sentences without having the sentences at hand. But, if that is the case, the simple disquotational device by itself does not enable us to eliminate the truth predicate from all contexts in which it occurs. Therefore, a strict disquotational account is not general enough and must be supplemented somehow. Deflationists usually maintain that the truth predicate shows its special utility or importance just in those contexts where certain technical complications compel us to mention sentences instead of directly talking about the world. Following Tarski, 17 disquotationalists emphasize that the truth predicate is required primarily in those cases in which we want to generalize with respect to sentence positions. We need "true" for saying in general that every sentence that has a certain form is true; or in deductive logic, to justify rules of inference, that is, to be able to say that in every inference of a certain kind truth is transmitted from the premises to the conclusion. The utility of the truth predicate is said to lie in its providing for a way of generalizing with respect to sentence positions. "True" is in the end merely a surrogate for infinite conjunctions and infinite disjunctions. If our language allowed the expression of infinite conjunctions and infinite disjunctions, the truth predicate would lose its most important function. 1 8 So the truth predicate is not merely a device for dis-

15 Quine 1970, 16 Quine 1990, 17 Tarski 1944, 18 Leeds 1978,

12 80 358-9 121

86

The Correspondence Theory

quotation; it is also a device for expressing or abbreviating infinite conjunctions and disjunctions by finite means. The raison d'etre of our concept of truth is its role as a device of generalization.

V I have pointed out that most deflationists tend to connect their accounts of truth to Tarski's formal work. The question poses itself as to how seriously they take Tarski s estimation of his own theory as an elaboration of the traditional correspondence theory of truth. I have characterized 2 correspondence theory as a theory which explains truth by certain relations between linguistic expressions and extralinguistic entities, between language and reality. In the sense of this characterization, Tarski's theory is a correspondence theory because it explains truth in terms of satisfaction, and satisfaction is a relation between open sentences and objects which belong to the range of variables of those open sentences. On the other hand, some non-deflationist philosophers - among them Davidson, shortly before abjuring the correspondence theory 19 - have asserted that Tarski's theory is a correspondence theory because it implies all equivalences T, all /-sentences. And indeed, although these equivalences do not mention the notion of correspondence at all, they nonetheless bring to light that the truth of a sentence depends on only two things: on its meaning, and on the way the world is. Unlike others, however, Davidson added that to fully understand why truth is correspondence with the way things are a detour must be made through the concept of satisfaction.20 This addition is indeed essential. The view, that any theory of truth that meets the criterion of Convention Τ is already a correspondence theory, is untenable. For deflationists typically endorse the equivalences Τ but vigorously renounce the possibility of developing a full-blooded correspondence theory. Even Quine says that ^-sentences are the "significant residue" of the correspondence theory,21 that the "underlying validity" of the correspondence theory finds expression in /-sentences,22 or that "truth consists in the world's being as the sentence says".23 But, on his view, /"-sentences are the only thing that can in the end be saved from the correspondence theory. /"-sentences preserve the strong intuition, on which the various versions of this theory are based, the intuition, that true statements agree with the world, or that it is the world which makes our statements true. Quine leaves no doubt that he categorically dismisses a correspondence theory that tries to go beyond the disquotational account of truth. In his eyes, the attempts to develop such a theory typi-

19 20 21 22 23

Davidson 1986, 309; for a critical examination of Davidsons views on truth, see Schantz 1 9 9 6 and 2 0 0 1 Davidson 1986, 3 0 9 Quine 1 9 8 7 , 2 1 3 Quine 1 9 9 0 , 8 0 Ibid., 81

Truth, Meaning, and Reference

87

cally lead its adherents to postulate mysterious pseudoentities like facts, states of affairs or propositions devoid of any explanatory value. Dorothy Grover, too, maintains that her variant of deflationism, the prosentential theory of truth, is capable of capturing some significant aspects of the correspondence theory.24 On the prosentential account, she developed together with John Camp and Nuel Belnap,25 there are expressions other than pronouns that can be used anaphorically, i.e., that can substitute for a previous primary expression of the same syntactical form. There are proverbs, proadjectives, and there are prosentences. Whereas pronouns occupy positions nouns occupy, prosentences occupy positions sentences occupy. The central claim of the proponents of this theory is that "that is true" and "it is true" function as prosentences. Thus "It is true that grass is green" can be construed as "Grass is green. That is true". It is important that prosentences are semantically unstructured units. All expressions in which "true" occurs outside of a prosentence have a misleading surface grammar. In the deep structure "true" is not a genuine predicate with a separate meaning. On the contrary, "true" is always a fragment of a prosentence. A key claim of this account is that prosentences and other proforms can be used as variables of quantification, as well as they can be used to substitute for an antecedent. "Everything Bill says is true", for example, is construed as "For each proposition, if Bill says that it is true, then it is true." This, very roughly, is the prosentential account of the role of the truth predicate in generalizations. The prosentential theory is similar to the disquotational theories in that both claim that the special usefulness of the truth predicate consists in providing us with certain kinds of expressibility — making available generalizations with respect to sentence positions. But, as we have seen, their analyses differ in certain formal respects. Since on the prosentential theory all /"-sentences are affirmed, Grover thinks she can justifiably say - in line with the correspondence theory - that it is the world that makes our sentences and beliefs true. Further, in contrast to many other deflationists, she acknowledges that there actually are interesting, explanatorily relevant, referential relations between language and the world. Such language-world relations play an essential role in the explanation of our various linguistic activities. Thus, in her estimation, reference is a substantive relation. Truth, on the other hand, is neither a substantive relation nor a substantive property. Grover emphatically dismisses the idea of explaining truth in terms of reference or satisfaction. Rather, she argues for keeping the issue of truth separate from the issue of reference, and urgently warns against the attempt to incorporate relations between linguistic expressions and extralinguistic entities into the account of truth. Truth talk can be explicated without appeal to a semantic theory that explains how names and predicates refer to things, properties and relations.

24 Grover 1992, 6-9, 29-34 25 Grover, Camp and Belnap 1975

88

The Correspondence Theory

VI Unlike Grover, Paul Horwich combines deflationism about truth with deflationism about reference. Nonetheless, he is convinced that his "minimalist conception of truth" can accomodate the central intuition on which correspondence theories are based. He maintains: The correspondence conception of truth involves two claims: (a) that truths correspond to reality; and (b) that such correspondence is what truth essentially is. And the minimalist response [...] is to concede the first of these theses but to deny the second.26 What truth essentially is, is captured in the equivalence schema T, or, since for Horwich - in contrast to Quine and, of course, to Tarski — propositions are the primary bearers of truth values, in the analogous schema "The proposition that p is true if and only if p". Horwich regards this schema as an axiom schema so that the totality of its instances constitute the axioms of the theory. The theory simply consists of all equivalence axioms. He concedes that the theory cannot be explicitly formulated - owing to the infinity of the instances of his equivalence schema and owing to the inexpressibility of some propositions in any natural language. In this regard, Horwich turns his back on Tarski who insisted that a definition of the truth predicate has to be finite. According to Horwich, we should neither expect nor desire a deeper explanation of why it is that the equivalence schema holds. 27 The search for a finite body of general principles that explain the minimal theory is a forlorn hope. The propositions of the minimal theory are much too simple to require any explanation. And yet, he asserts that the minimal theory provides, in combination with theories of other phenomena, a satisfactory explanation of all facts involving truth. 28 Moreover, the minimal theory is supposed to give an account of the meaning, and of our understanding, of the word "true". 29 The meaning of "true" - its use in our language — is determined by our disposition to stipulate instances of the equivalence schema, and, correspondingly, our understanding of "true" consists in our disposition to accept a priori, without evidence, any instantiation of this schema. Horwich recognizes that the minimal theory does not yield an explicit, eliminative analysis of "true", one that permits us to eliminate "true" from every context in which it occurs. A satisfactory characterization of the meaning of "true", he maintains, does not have to take the form of a specification of necessary and sufficient conditions for its correct application. Rather, he regards the minimal theory as an implicit or use def• · · orc« true ». 30 mition Though his minimal theory of truth involves nothing more than the instances of the equivalence schema, Horwich is prepared to admit that truth has further properties. And he maintains that the property of corresponding to the world belongs to these

26 27 28 29 30

Horwich 1 9 9 0 , 1 2 4 Ibid., 50-53 Ibid., 7, 2 5 - 2 6 Ibid., 3 6 - 3 8 Ibid., 3 4 - 3 7

Truth, Meaning, and Reference

89

further properties. His minimalist position does not deny, then, that true propositions correspond to the world; it does not even deny that each true proposition is made true by the existence of a corresponding fact. 31 All of these properties, so Horwich asserts confidently, are compatible with his minimalism. What is not compatible with his minimalism, however, is an explanation of truth in terms of correspondence or reference or satisfaction. There are, he claims, no constitutive relations between truth on the one hand and reference, satisfaction or correspondence on the other. 32 It is a remarkable fact that, in this important respect, he distances himself so plainly from Tarski, for, in Tarski's compositional approach, truth is defined just in terms of satisfaction. Horwich explicitly reproaches theories in Tarski's style as being "unnecessarily complex" and for fostering the misleading impression that truth, reference and satisfaction are inextricably intertwined with one another. Truth, Horwich maintains, "has a certain purity", and by that he means that our understanding of truth is independent of those other notions. It is not that he wants to deny that there are semantic principles relating those concepts, such as the principle: "Fa" is true iff there exists an object χ such that "a" refers to χ and "F' is satisfied by x. But he does not treat such principles as explanatorily basic. Instead, he proclaims that they should be derived from a conjunction of the theory of truth and separate minimalist theories of reference and satisfaction. Thus, he subscribes to a full-fledged minimalism. Not only does he advocate a minimalist theory of truth, but also minimalist theories of reference and satisfaction or being true of. Reference and satisfaction, he argues, are just as "non-naturalistic, and in need of infinite, deflationary theories, as truth is".33 Horwich, we see, is an out and out minimalist. Thus, the conceptually fundamental principle explaining our use of "refers" is the schema (y) [Singular concept Ν refers to ^ i f f « = y].^ To know the meaning of "refers" consists simply in our disposition to accept instances of the disquotational schema. A reductive, naturalistic analysis of the reference relation along the lines of (x)(y) [x refers to y iff c(xy)] where is a causal relation of some sort, is not to be expected. And, similarly, the conceptually fundamental principle governing our use of "is true of" is the schema (y) [Predicative concept F is true of jy iff fly)}, and not anything of the form (x)(y) [x is true of y iff r(xy)].

31 32 33 34

Ibid., 110-112 Ibid., 117-120 Ibid., 125 Horwich 1998, 108

90

The Correspondence Theory

Horwich maintains that minimalism about truth-theoretic notions enables us to dissolve the traditional problem of intentionality. Trivialities such as "Paris" refers to Paris, "Aristotle" refers to Aristotle, and so on, are supposed to exhaust the nature of reference. No more than truth does reference have an underlying nature. Although he calls his own position "semantic deflationism", he is not a deflationist about meaning in the ordinary sense of that term. On the contrary, he advocates a use theory of meaning, according to which words possess substantive meaning properties, properties which have an underlying, non-semantic nature. The naturalistic reduction basis of the meaning properties of words is the property of their use being governed by certain fundamental regularites. In short, meaning facts are analysed in terms of facts about use. What Horwich's deflationism about meaning, in effect, amounts to is the view that once the phenomena of truth and reference are properly "deflated" by the minimalist account, they no longer stand in the way of developing an attractive use theory of meaning. The main difficulty in determining the nature of meaning, so he argues, has been thought to derive from the fact that the meaning of a general term determines its extension. By this general fact, we are committed to certain meaning-totruth conditionals such as: X means D O G —> χ is true of all and only dogs. 35 Inflationists claim that such conditionals place a substantive, relational constraint on what the meaning-constituting properties must be; convinced that "x is true of y" is analysable as "x bears relation r to y" they suppose that χ is true of all and only dogs = (y)(r(xy) iff y is a dog). However, from the deflationary perspective such meaning-to-truth conditionals hold without imposing any such substantive, relational requirement on the meaning-constituting property. Horwich declares these conditionals to be "conceptually fundamental", and he says that they define the relation of being true of. 36 We need not assume that the property in virtue of which "dog" possesses its particular meaning is some sort of non-semantic relation between that term and dogs. Ultimately, it is nothing but our commitment to the meaning-to-truth conditionals that accomodates the phenomenon of representation or aboutness, the determination of extension by meaning. Now, I agree with Horwich that a decent theory of meaning ought to acknowledge that meaning determines extension, in the sense that any two expressions with the same meaning must have the same extension. Let us accept this as an adequacy condition on a theory of meaning. After all, it is because of their meaning what they do that words represent diverse aspects of the world. The representational power of words has been thought to raise the difficult question of how meanings must be constituted for this determination of extension to take place. Horwich, however, dismisses this question believing that it rests on mistaken inflationary presuppositions. In contrast, he holds

35

Horwich adopts the convention of capitalizing English expressions in order to designate their meanings. Thus "DOG" is a name for the meaning of the word "dog". 36 Ibid., 227

Truth, Meaning, and Reference

91

that the determination of extension by meaning is trivially explained by the meaningto-truth-conditionals — regardless of what we assume about the nature of meaning-constituting properties; in particular, without having to assume that meaning-constituting properties have a relational form. But are things really so simple? With the help of a famous science-fiction example, due to Hilary Putnam, I will show that Horwich's own use theory of meaning does not satisfy the constraint that meaning fixes extension. 37 Imagine that somewhere in the galaxy there is a planet, called "Twin Earth". Twin Earth is very much like Earth. Many of our molecule-for-molecule doppelgänger on Twin Earth even speak a language that seems like English. There is only one difference: the liquid called "water" on Twin Earth, although it has all the superficial properties that water does, is not really water that is, H20 - but a different liquid which Putnam calls "XYZ\ So when an Earthling and his doppelgänger use the same term "water", they nonetheless refer to different substances. The extension of the word "water" differs as between Earth and Twin Earth, and since extension is a function of meaning, its meaning differs as well. What this shows is that both the extension and the meaning of at least certain sorts of terms, the natural-kind terms, are at least partly determined by the physical environments in which they are used. The lesson is that in order to understand meaning we must take account of the relation between language and things outside itself. As Putnam memorably puts it: '"meanings' just ain't in the head".38 Reference, he concludes, is a component of meaning. That is the reason why meaning determines extension. Although use theories themselves typically reject the view that meanings are in the head, the Twin-earth fantasy nevertheless faces them with a serious problem. The problem is that while the term "water" has different meanings on Earth and on Twin Earth, the corresponding patterns of usage in the two speech communities apparently do not diverge in the least. After all, Earthlings and their doppelgänger are exactly alike in their total bodily states and in their total public linguistic behaviour. And since this is so, the meaning of a term does not consist exclusively in a fundamental use regularity. A proponent of the use theory might reply that this argument begs the question. It simply takes for granted that an Earthlings use of a term is the same as that of his doppelgänger. But their pattern of usage will be the same only if we abstract from the different physical environments in which words are deployed. If we individuate use more broadly, in terms of the particular substances or objects or properties towards which it is directed, then Earthlings' use is not the same as that of their Twins. Earthlings' use of "water" is related to water, that is H20, while Twin Earthlings' use of this word is related to XYZ. But this reply, which seems to me to be correct, is not available to Horwich. For him, meanings do not essentially involve referential relations between words and aspects of the environment. Again and again he urges that the meaning-constituting regularities of use need have no relational form. Reference is no part of meaning. It can-

37 38

Putnam 1975 Ibid., 227

92

The Correspondence Theory

not be. Given its deflationary nature, reference cannot play a role in explaining meaning. It is noteworthy that Horwich himself mentions "Putnam's problem", as one of many problems that a satisfactory theory of meaning should resolve, the problem namely: "to do justice to the 'twin earth' thought experiments suggesting that what a given speaker means by a term is not 'in his head' but depends upon aspects of his physical and social environment". 39 It is indeed a point in favour of use theories in general that they can smoothly accomodate the social character of meaning, the fact that to speak a language is to engage in a rule-governed social practice. It is the contribution of the non-social, physical or chemical environment that Horwich's version of the use theory does not properly take into account. Admittedly, he does allow that characterizations of use refer to the environment of the speaker,40 and he concedes that the use regularity constituting "x means D O G " may advert to dogs in some way.41 Furthermore, he informs us that perhaps the property underlying the meaning of "red" is the disposition to accept "This is red" in the presence of red things. 42 With claims such as these, he seems to be on the right track. But the plausibility of these claims is due to their agreement with the tenets of representational theories of meaning, according to which meanings are constituted by representational properties, by direct or indirect causal links to reality. Horwich reminds us that there are words whose meanings are not constituted by any relations to members of their extension; the logical connective "and" is one of his favourite examples. This suggests that he merely claims that some meanings - paradigmatically, the meanings of logical words - are not engendered by any external relations to the world. Nobody, I hope, will deny that such meanings have a non-relational nature. But with respect to the interesting and controversial cases, to our basic descriptive repertoire, to terms such as "dog", "red", and "water", which do represent aspects of the environment, his attitude is less clear. The explanatorily basic use facts of such terms "may", as he cautiously puts it, make reference to the environment. My position is that they have to. There is no way to explain the meanings of such terms without appealing to objects, substances or properties in the surroundings of the speaker, without invoking some causal relation between the terms and the things to which they apply. Horwich fails to give us a plausible idea of how the meanings of basic descriptive terms, those that establish immediate contact to reality, might consist in a monadic use property. It should be clear that my objection is not an objection to use theories of meaning in general. Indeed, a decent use theory - one which, like Horwich's, does not entail an extreme form of holism, but one which, unlike Horwich's, respects the insight of externalism that the meaning properties of basic descriptive terms are relational properties - seems to me to be promising. It is, after all, the way we use words that makes them mean what they do.

39 Horwich 1998, 2 40 Ibid., 41 41 Ibid., 66 42 Ibid., 6

Truth, Meaning, and Reference

93

But, in contrast to what Horwich thinks, a use theory of meaning is not the natural ally of deflationism about reference or representation. By endorsing a naturalistic account of meaning, and by holding that meaning fixes extension, even if only in the sense that any two expressions with the same meaning must have the same extension, he seems to have given his inflationist opponents all they need to reduce representation and aboutness, too, to a non-semantic property. The account they propose will have this form X bears R to β iff χ means F. The intimate connection between meaning and the relation of being true of, plus the non-semantic nature of meaning, seem to commit Horwich to a reductive analysis of this form. Evidently, he does not accept this proposal. In his view, meaning-constituting properties are not reducible to non-semantic relations. Nevertheless, as we have seen, he admits that the property of meaning DOG might be engendered by some relation to dogs. But this relation, he continues, might be completely different from the relation to tables that engenders the property of meaning TABLE. 43 This line of thought suggests that he requires an analysis of representation to be uniform, not allowing that there might be a plurality of different underlying, non-semantic relations Rj, R2, ..., one for each different kind of general term. 44 There is, however, no need for naturalists about representation to accept such a severe requirement. Contrary to what Horwich believes, they are not committed to the claim that there is such a thing as the character of representation or reference that, in conjunction with the meaningconstituting property of a given term, enables us to deduce what the term is about or refers to. But a deduction of this sort, he thinks, is just what inflationists' attempts at naturalizing semanticity or intentionality are aiming at. It is revealing that Horwich does not impose a similarly strong requirement on his own naturalism about meaning. What he offers us instead is a series of reductive analyses: there is a substantive property underlying the meaning of "dog" and another substantive property underlying the meaning of "table", and so on. But there is no uniform or general analysis of the schema "x means F\ Apparently, naturalism about meaning seems to be far less demanding than naturalism about representation. On the one hand, Horwich demands too much from naturalistic approaches to the problem of representation. On the other hand, he underestimates their explanatory power. This becomes obvious when we turn to his criticism of the causal theory of reference, advocated in various forms by Kripke, Evans, and Devitt. According to this view, a given name refers to what it does in virtue of a certain sort of causal chain of utterances linking the name and its referent. So my present use of "Aristotle" is a recent link in a causal-historic network of reference-borrowings, whose first link is the ceremony in which the infant Aristotle was given that name. I acquire the name from others who acquired it from others, and so on, until we reach the dubbing. Thus, my

43 44

Ibid., 2 4 But see ibid., 128, F n l l

94

The Correspondence Theory

use of "Aristotle" refers to Aristotle because a chain of communication has been established in our linguistic community by transmitting the name from person to person, a chain which is ultimately grounded in Aristotle himself. It seems reasonable to say that reference is preserved in this way. Horwich objects that causal theories do not even provide a first approximation to a theory of reference. This can be seen, he urges, when we distinguish between the facts in virtue of which a word has its meaning and reference in a given language, and the facts in virtue of which members of the linguistic community use the word with that meaning and reference.^ Causal theories, he admits, offer a satisfactory solution of the second issue; they can indeed account for the social transmission of a name or any other term within a linguistic community. Thus he appreciates the phenomenon of the "division of linguistic labour" or "semantic anti-individualism". But he thinks that it is concerning the relevant first issue - the facts that bestow on a word its meaning and reference — that causal theories leave us entirely in the lurch. Is this a fair evaluation? Note, firstly, that, contrary to what Horwich thinks, causal theorists need not deny that names have meanings. They only have to deny that the meaning of a name is given by a set of definite descriptions associated with it which fixes its reference. Let us look in this light at his own proposal regarding the meaning of the name "Aristotle": "Aristotle"'s meaning what it does consists in the fact that the basic feature of its use is the (conditional) holding true of "This is Aristotle" when pointing at Aristotie. 46 This looks promising indeed. And Horwich rightly stresses the special role of the experts, to whom the rest of us have to defer. It is the experts who were "acquainted" with Aristotle and who could point at him. But pointing at Aristotle, and being acquainted with him, involve perceiving him, and to perceive something is to be causally stimulated by it. It seems that the chances are not so dim that both the social transmission of a name and the meaning-constituting process of fixing its reference can be explicated along causal lines. Anyway, Horwich has given us no good reason for giving up the view that the meaning of a name is constituted by causal links to the world. Rather, a consideration of names buttresses the representationalist view that the meanings of words are exhausted by properties that play a role in determining their reference.

VII Hartry Field meanwhile defends a variant of the deflationary conception of truth he calls "pure disquotational truth". It is a striking feature of pure disquotational truth

45 46

Ibid., 117 Ibid., 129

Truth, Meaning, and Reference

95

that it is restricted to the idiolect of the speaker. According to this position, a person can meaningfully apply "true" only to utterances she understands. Field suggests, as a rough heuristic, that when a person calls an utterance true in the pure disquotational sense she is saying that it is true-as-she-understands it. The central thesis of the doctrine of pure disquotational truth is that a person's claim that an utterance u is true (true-as-she-understands-it) is cognitively equivalent to u itself (as-she-understands-it). Obviously, pure disquotational truth, as characterized so far, cannot be the last word. It is an integral part of this view that a person can apply a pure disquotational truth predicate only to utterances that she understands. However, our use of "true" is not restricted to utterances within our actual idiolect. We are ready to acknowledge that there are true sentences in other languages, sentences we do not understand. So disquotational truth must be extended somehow. Field dispenses with the notion of interlinguistic synonymy because it seems to involve the notion of sameness of truth conditions. Instead, he favours the "more flexible" option of defining what it is for a foreign sentence to be true relative to a correlation of it to one of our languages. As he says: "what we are doing when we conjecture whether some utterance we don't understand is true is conjecturing whether a good translation of the utterance will map it into a disquotationally true sentence we do understand." 4 7 But a deflationist who takes this approach still faces the serious challenge of having to give an account of translation or meaning - as that which is preserved in a good translation - that does not appeal to truth-theoretic notions. Antideflationists doubt that this can be done. O n the contrary, we are convinced that translation and meaning just have to be explained in terms of truth conditions. Since it is a defining characteristic of deflationism that truth conditions play no central role in semantics and the theory of mind, meaning and content must be explained in a different way. Merely pointing out that translation is a "highly context-sensitive and interest-relative notion" 4 8 is not to offer a satisfactory theoretical alternative. Moreover, Field, like Horwich, maintains that deflationism about truth conditions goes hand in hand with deflationism about reference. Reference, too, is a purely logical concept, not a semantic one. Hence, reference and truth turn out to be use-independent notions. A main motivation for deflationism about reference, Field tells us, is that it avoids a grave problem inflationism finds itself confronted with. An inflationist, trying to determine what a general term such as "rabbit" refers to, has three options: either she can find some naturalistic facts, facts about our employment of language, to answer this question; or she can claim that there simply is no fact of the matter, that we have to accept a surprising level of referential indeterminacy; or she can say that it is a non-naturalistic fact about us that our word "rabbit" refers to what it does. As mentioned earlier, I hold on to the first choice, the project of naturalizing reference or aboutness, as difficult to carry out as it may be in the end. Field, however, thinks that it is a point in favour of deflationism that it does not have to choose be-

47 Field 1994, 273 48 Ibid., 273

96

The Correspondence Theory

tween these alternatives afflicted with philosophical difficulties as they all are. There is nothing at all to explain here since "it is simply part of the logic of 'refers' (or 'is true of') that 'rabbit' refers to (or 'is true of') rabbits and nothing else."49 Reference does not involve more than such entirely trivial facts. Now, of course, the sentence "'rabbit' refers to rabbits" is trivially true. But, evidently, knowledge of such simple disquotational truths amounts to knowledge of what the quoted term refers to only if we already know what the used term refers to. Remember, both Earthlings and Twin Earthlings will accept the sentence "'water' refers to water". On Twin Earth, however, "water" does not refer to water. Or think of Quine's problem. Suppose we were trying to find out what the foreign term "gavagai" stands for: whether for whole rabbits, or for rabbit parts, or for rabbit phases. In this situation, no disquotation schema is to be had to help us out of the fix. If we want to answer the question of what "gavagai" does indeed refer to, we have no choice but to find the facts in virtue of which this term refers to what it does refer to - or to accept referential indeterminacy. Apropos Quine. Wrestling with the problem of whether the indeterminacy of reference extends also to the home language, Quine came to think that the decision to translate our own language into itself homophonically, thus taking it at face value, would somehow resolve the indeterminacy. In this context, he made a revealing remark: "'rabbit' refers to rabbits, whatever they are".^° With this remark, Quine, in effect, concedes that the reference of the used occurence of "rabbit" remains indeterminate. This is no wonder - given his philosophical assumptions. If there really were no naturalistic fact of the matter as to what "gavagai" refers to, then there would be no naturalistic fact of the matter as to what "rabbit" refers to. This is not to deny that there is a difference as concerns our relations to our own language and to a foreign one. There is one. We, as normal speakers of English, will reply affirmatively to the query "Does 'rabbit' refer to rabbits?", while in the case of radical translation no corresponding disposition is available. But simply from the fact that we accept the sentence "'rabbit' refers to rabbits", and reject the sentence "'rabbit' refers to rabbit parts", it does not follow that there is a determinate answer to the question of what "rabbit" refers to. Sentences of the form "'F refers to Fs" are trivially correct - a fact rightly accentuated by deflationists. Whatever the term "rabbit" may refer to, whatever entities in the word it may be correlated with, we will always give the same trivial answer to the question of what it does refer to, namely: "rabbit" refers to rabbits. Quine was very fond of disquotational paradigms. But this did not entice him into thinking that they enable us to solve all difficulties about reference. Field would probably reply that the problems I try to highlight for the deflationary perspective on reference have more to do with translation and meaning than with reference. In analogy with extending disquotational truth to foreign languages, he might say that if "rabbit" is a good translation of "gavagai", then "gavagai" refers to what "rab-

49 50

Ibid., 260 Q u i n e 1990, 52

Truth, Meaning, and Reference

97

bit" refers to, namely, rabbits. But this proposal assumes that we are able to discover what "gavagai" means in the foreign language, or that "rabbit" is a reasonably good translation of "gavagai", without having to investigate what "gavagai" refers to. This is hopeless, I think, for the reasons already given. With a term like "gavagai", at least part of the task of finding out what it means is finding out what it stands for; its referential character seems to be essential to its meaning what it does. For Field, however, there is nothing to find out about reference; reference is not an element of meaning. Field's position becomes clearer when we throw a glance at his new perspective on the causal theory of reference he himself formerly championed. Field, now, has to deny that the causal chain emanating from Aristotle and leading to my use of the name "Aristotle" explains how my use of the name could refer to Aristotle. After all, the central tenet of deflationism is that the only explanation we need as to why my use of the name refers to Aristotle is given by the disquotation schema. Nevertheless, he still agrees that the causal theory does explain something. But in order to see what this theory does explain it must be inverted somehow. Only then we will realize that what the causal story really explains is the correlation between a persons beliefs involving the name "Aristotle" and the facts about Aristotle. Probably a large part of our beliefs about Aristotle are true. If we believe "Aristotle was F', then the conditional probability that Aristotle was F is pretty high. It is, Field claims, this reliability of our beliefs that is the new explanandum for the causal story: our beliefs about Aristotle are reliable because they are ultimately connected via a causal chain to the beliefs of experts who had a direct access to Aristotle himself. But this inversion of the causal theory - shifting the focus from explaining reference to explaining our reliability - seems to be rather artificial. For surely, beliefs are semantically structured entities; they have semantically significant parts. Thus, beliefs about Aristotle involve the name "Aristotle" or an internal analog of it. And for beliefs involving "Aristotle" to be reliably correlated with the facts about Aristotle, the name has to be reliably correlated with Aristotle. But isn't reliable correlation just the kind of stuff out of which the relation of reference or aboutness is made? Aren't we here well on our way to a reduction of reference to a naturalistically respectable relation between words and things? Field's strong version of deflationism seems to turn itself into a substantive form of reductionist inflationism. He himself sees the problem; he even admits "that there is no sharp line between reduction and elimination".-'1 Nevertheless, deflationism requires him to urge that - though uses of names and the things they refer to are reliably correlated - reliable correlation has nothing to do with reference. The general problem for deflationism results rather directly from its crucial tenet that the concepts of truth and reference cannot play a central role in the philosophy of language and the theory of the contents of intentional states. They are purely logical devices. The challenge confronting the deflationist, then, is to explain basic semantic concepts, especiallly the notions of meaning and content, without invoking the concepts of truth and reference. Field proposes to construct meaning and content from

51

Field 1994, 253

The Correspondence Theory

98

several elements including above all "conceptual or computational role" and "indication relations". The latter of course, amount to nothing but reliable correlations between belief states and states of the world outside. As we have seen, he is well aware of the possible objection that one might plausibly come to regard these indication relations as constituting the truth conditions relation. In other words, the theory of meaning and content he is about to develop seems to employ certain physicalistically respectable terms that could be considered as a reduction of truth conditions. Field replies that we are far away from such a reduction because there are many examples "where the indication relations don't reflect what we would intuitively regard as truth conditions". 52 But it is what we would intuitively regard as truth conditions that we have to advert to in order to form a reasonable judgement on whether or not the account of meaning and content in terms of conceptual role, indication relations and perhaps certain other elements can be recognized as adequate. Bluntly put, we know a prion that a proposed theory cannot be correct so long as it does not allow us to infer instances of the following schema: if s means that p, then s is true if and only if p - which must, of course, not be confused with the different and more controversial schema: if s is true if and only if p, then s means that p. And once the theory does allow us to infer instances of the first schema, then it can be regarded as a believable reduction of truth conditions to physicalistic terms.

VIII We have seen that Horwich's and Fields attempts at explaining meaning and content without overt appeal to reference and truth are faced with serious dificulties. Reference seems to be essential to meaning and thus does not seem to be a purely formal or logical concept. Since reference proved to be quite a robust notion, it must be explained in a wholly different manner. As I see it, there is no alternative to analyzing reference and representation in naturalistic terms. Such an analysis can then be used to to vindicate the idea of explaining truth in terms of reference, that is, to vindicate the venerable correspondence theory of truth.

Bibliography Ayer, A. (1936), Language, Truth and Lope, London Davidson, D . (1984): Inquiries into Truth and Interpretation, Oxford — (1986): "A Coherence Theory ofTruth and Knowledge", in: LePore, E. (ed.): Truth and Perspectives on the Philosophy of Donald Davidson, Oxford, 307-319 — (1990): " T h e Structure and Content ofTruth", Journal of Philosophy 87, 279-328 — (1996): " T h e Folly of Trying to Define Truth", Journal of Philosophy 93, 263-78 Dretske, F. (1981): Knowledge and the Flow of Information, Cambridge/MA Field, H . (1972): "Tarski's Theory ofTruth", Journal of Philosophy 69, 347-375 — (1994): "Deflationist Views of Meaning and Content", Mind 103, 249-285

52

Ibid., 255

Interpretation.

Truth, Meaning, and Reference

99

Fodor, J. (1987): Psychosemantics, Cambridge/MA Frege, G. (1966): "Uber Sinn und Bedeutung", in: Patzig, G. (ed.), Funktion, B e g r i f f , Bedeutung, Göttingen, 40-65 Grover, D. (1992): A Prosentential Theory of Truth, Princeton Grover, D., Belnap, N., Camp, J., (1975): "A Prosentential Theory of Truth", Philosophical Studies TJ, 73125 Horwich, P. (1990): Truth, Oxford — 1998): Meaning, Oxford Putnam, H. (1975), "The Meaning of'Meaning"', in his Mind, Language and Reality: Philosophical Papers, vol. 2, Cambridge, 215-271 Leeds, S. (1978): "Theories of Reference and Truth", Erkenntnis 13, 111-127 Millikan, R. (1984): Language, Thought and Other Biological Categories, Cambridge/MA Quine, W.V.O. (1970): Philosophy of Logic, Englewood Cliffs — (1987): Quiddities, Harvard — (1990): Pursuit of Truth, Cambridge, MA Ramsey, F. (1927): "Facts and Propositions", Proceedings of the Aristotelilan Society, suppl. vol. 7, 153-170 Rorty, R. (1986): "Pragmatism, Davidson and Truth", in: LePore, E. (ed.): Truth and Interpretation. Perspectives on the Philosophy of Donald Davidson, Oxford, 333-355 Russell, B. (1912), The Problems of Philosophy, Oxford Schantz, R. (1996): Wahrheit, Referenz und Realismus. Eine Studie zur Sprachphibsophie und Metaphysik, Berlin & New York — (2001), "Truth and Reference", Synthese 126, 261-281 Soames, S. (1984): "What is aTheory of Truth?", Journal of Philosophy 81, 411-429 Strawson, P. (1950), "Truth", Proceedings of the Aristotelilan Society, suppl. vol. 24, 129-156 Tarski, A. (1956a): "The Concept of Truth in Formalized Languages" in: Logic, Semantics, Metamathematics, Oxford, 152-278 — (1956b): "The Establishment of Scientific Semantics", in: Logic, Semantics, Metamathematics, Oxford, 401-408 — (1944) "The Semantic Conception of Truth and the Foundations of Semantics", Philosophy and Phenomenological Research 4, 341-75 Williams, C.J.F. (1976): What is Truth?, Cambridge Williams, M. (1986): "Do we (Epistemologists) need aTheory of Truth?", Philosophical Topics 4, 223-242 — (1999): "Meaning and Deflationary Truth", Journal of Philosophy 96, 545-564 Wittgenstein, L. (1922), Tractatus Logico-philosophicus, London

II Deflationism Defended

Explanatory vs. Expressive Deflationism About Truth ROBERT B . BRANDOM

It has become customary to refer to a class of theoretical approaches to truth as 'deflationary'. Broadly disquotational theories are typically taken as paradigms.1 In this paper, I offer three suggestions concerning deflationism. First, I want to recommend a particular form of deflationary theory of the use of the word 'true' and its cognates, which I have developed in more detail elsewhere: the anaphoric approach. I will describe that approach in general terms, and rehearse some of the considerations that lead me to see it as both technically and philosophically more satisfying than standard disquotational approaches. Next, I argue that, so understood, 'true' plays a crucial expressive role. Adding such a locution to a language substantially increases its overall expressive resources and capabilities. Thus one should not take a deflationary attitude toward the expressive role of 'true'. Finally, I describe the sense in which I think one should take a deflationary attitude toward the explanatory role of 'true'. Playing the expressive role distinctive of truth locutions disqualifies them from being understood as expressing concepts on which to base certain kinds of global explanations of propositional contentfulness in general. In particular, one is debarred from pursuing an order of explanation that seeks to render the notion of propositional contentfulness intelligible in terms of a prior concept of truth conditions. This is not, however, to say that the notion of truth conditions can be of no explanatory use whatsoever. I will discuss some of the more localized explanatory projects in which that concept can serve. I close by pointing out a direction in which such an explanatory (but not expressive) deflationist about truth might look for some fundamental semantic concepts to use in global explanations of contentfulness, once truth is ruled out.

I. The Anaphoric Account of the Expressive Role of'True' The most sophisticated and successful account I know of the expressive role of the concept of truth - of what one is doing in deploying truth talk - is an anaphoric theory. Such theories originate with Grover, Camp, and Belnap's prosentential theory of truth. The version I favor understands locutions such as "... is true" and its relatives as proform-forming operators. In the simplest case, "That is true," is a presentence, which relates to, and inherits its content from, an anaphoric antecedent - for instance someone else's tokening of "Snow is white," - in the same way that a pro noun such as 'he' relates

1

The most complete presentation of a disquotational theory is Paul Horwich Truth [Basil Blackwell, Oxford, 1990], For an interesting discussion, see Marian David Correspondence and disquotation: an essay on the nature of truth [Oxford University Press, 1994]

104

Deflationism Defended

to and inherits its content from an anaphoric antecedent - for instance, someone else's tokening of 'Tarski'. As the authors of the original theory introduce them by analogy to pronouns, presentences are defined by four conditions: •

They occupy all grammatical positions that can be occupied by declarative sentences, whether free-standing or embedded.

•

They are generic, in that any declarative sentence can be the antecedent of some prosentence.

•

They can be used anaphorically either in the lazy way or in the quantificational way.

•

In each use, a prosentence will have an anaphoric antecedent that determines a class of admissible sentential substituends for the prosentence (in the lazy case, a singleton). This class of substituends determines the significance of the prosentence associated with it.

Anaphora is a relation according to which the content of one tokening is determined by its relation to another tokening or class of tokenings: its anaphoric antecedent(s). The anaphoric dependent is not in general replaceable by its antecedent. The cases where it is are what Geach calls the 'lazy' cases. Thus in 1) # Have I read the book? I haven't even taught it yet! # 2 the anaphorically dependent expression tokening 'it' can be replaced by another tokening of the same type as its anaphoric antecedent tokening 'the book' without altering the sense of the remark. By contrast, in 2) # Yesterday I met an economist. The economist told me that he believes the Chinese will be forced to devalue the renminbi. # the anaphoric dependents that form the later elements of the anaphoric chain cannot be replaced by their antecedents without altering the sense of the discourse. Saying 3) # Yesterday I met an economist. An economist told me that an economist believes the Chinese will be forced to devalue the renminbi. # does not - as the original does - commit one to its being the same economist one met, was told by, and who has views about devaluation of the Chinese currency. The anaphoric dependents inherit their content from their antecedents, but some expressions (such as an economist') can grammatically play the role only of initiating anaphoric

2

In discussing anaphoric connections across sentences, it is convenient to follow Charles Chastain in his seminal work "Reference and Context" (in Keith Gunderson, ed. Language, Mind, and Context, Minnesota Studies in the Philosophy of Science, vol. 7 [University of Minnesota Press, Minneapolis, 1975], pp. 194-269.) in using '#' quotes to mark discourses containing multiple sentences, perhaps uttered by different interlocutors.

Explanatory vs. Expressive Deflationism about Truth

105

chains, while others (such as 'he') can grammatically play the role only of continuing them. This is true even when the anaphoric dependent precedes its antecedent in the discourse, as in 4) # Although she didn't want to, the mathematician proof. #

was obliged to rework her

In the category of pro sentences, instead of pro nouns, a case involving lazy anaphora corresponding to (1) might be 5) # Hegel said "Truth is a vast Bacchanalian revel, with not a soul sober," and I believe it is true. # According to the prosentential theory in its original form, the prosentence "it is true," in (5) functions so as to give the second conjunct the sense of 6) ...and I believe truth is a vast Bacchanalian revel, with not a soul sober. A case like (2) might be something like 7) # One of Hegel's notorious remarks about truth is hard to understand, but I believe it is true. # This is not equivalent to 8) # One of Hegel's notorious remarks about truth is hard to understand, but I believe one of Hegel's notorious remarks about truth. # For just as the anaphoric relation in (2) does, and the mere repetition in (3) does not, settle it that the same economist is being discussed throughout, (7) does and (8) does not settle it that the same notorious remark of Hegel about truth is both hard to understand and endorsed by the speaker. Once again, backwards anaphora is possible: 9) # Even though for all I know, it is true, I will never admit that I understand that remark of Hegel's about truth. # The authors of the original version of the prosentential theory wrestled all sentences involving 'true' into a form in which their single prosentence "it (or that) is true," appears - typically by seeing a disguised propositional quantification. So 10) "Snow is white," is true, is read as 11) For any sentence, if that sentence is "Snow is white," then it is true. I have urged elsewhere3 that it is preferable to understand ".. .is true" as aprosentenceforming operator, which applies to a noun phrase specifying an anaphoric antecedent,

3

"Pragmatism, Phenomenalism, and Truth Talk," Midwest Studies in Philosophy vol. XII: Realism; 1988 pp. 7 5 - 9 3 . , and in Chapter Five of Making It Explicit [Harvard University Press, 1994],

106

Deflationism Defended

and yields a presentence anaphorically dependent on that specified antecedent. According to this approach, understanding a sentence in which the word 'true' (or one of its cognates) appears is a two stage stage process. First one must process the noun phrase to determine what sentence tokening (or class of such tokenings) it picks out as anaphoric antecedent(s). Then one determines the sense of the sentence that is anaphorically dependent on the antecedent(s). The full expressive resources of the language may be brought to bear in specifying the antecedent, so computing it from the noun phrase is not always done in the same way. Sometimes the noun-phrase to which the prosentence-forming operator "...is true," is applied specifies its antecedent by naming it. Where quotation marks are used to form a quote name, the result is the sort of case that disquotational theories treat as paradigmatic. In (10), " "Snow is white," " is a quote name of the sentence "Snow is white," and the anaphora is lazy, so (10) is equivalent to 12) Snow is white. But the antecedent can also be specified by describing it, as in 13) Tarski's favorite sentence is true. which under suitable assumptions is also equivalent to (12). The antecedent can also be paraphrased or put in indirect discourse. Then indexicals (and choice of language) are referred to the speaker of the paraphrase, rather than the one to whom the original antecedent is attributed: 14) John said that he is not confused on this point, and what he said is true. Again, a demonstrative can be used to indicate the anaphoric antecedent of the presentence that results from applying "...is true" to it. 15) # Hegel said that a hero is not a hero to his valet, but that is not because the hero is not a hero, but because the valet is a valet. That is true. # Looking carefully, one will see that there are actually two presentences in this little discourse, since the second 'that' is elliptical for "that is true." In this case the anaphoric chain is extended, as when one tokening of 'he' or 'it' has another such tokening as its immediate antecedent, but is thereby linked to the antecedent of that anaphor. The antecedent of the presentence can also be specified by a noun phrase that is itself an anaphoric dependent - now a pro noun whose antecedent is a sentence specification, perhaps a name or a description. Thus (7) can be understood as involving the application of the prosentence-forming operator "...is true," to the pronoun 'it'. Computing the antecedent of the resulting presentence is now itself a two stage process. First one must find the noun phrase that is the antecedent of'it', namely a tokening of "one of Hegel's notorious remarks about truth." This is a description of a sentence uttering or inscription - perhaps a tokening of "Truth is a vast Bacchanalian revel, with not a soul sober." Understanding the description in this way commits one to understanding the assertion of "it is true" in (7) as having the sense of an endorsement of

Explanatory vs. Expressive Deflationism about Truth

107

the claim that truth is a vast Bacchanalian revel with not a soul sober. According to this reading, understanding the "it is true," in (7) requires discerning and processing two anaphoric chains, one linking noun phrases and ending in the anaphorically dependent pro noun 'it', and the other linking sentences and ending in the anaphorically dependent prosentence "it is true." The second stage in interpreting a truth claim is determining the sense of the prosentence, after an antecedent for it has been settled on. In what we can call 'strictly', 'directly', or 'syntactically' lazy cases, the prosentence can simply be replaced by its antecedent, as in (5) and (6), and (10) and (12), which will preserve all relevant semantic properties. In what could be called 'broadly', 'indirectly', or 'semantically' lazy cases, the prosentence can be replaced (again preserving all relevant semantic properties) by any sentence that has the same content as the antecedent. Doing this can require the same sorts of transformation of indexicals and of language as is required in indirect discourse in general. So in the direct discourse equivalent reported in indirect discourse in (14) 16) # John: "I am not confused on this point." Bob: "What John says is true." # Bob's remark is not equivalent to his saying "I am not confused on this point." It is equivalent, in his mouth, to "John (or he) is not confused on this point." And in (8), (9), and (10), we should keep in mind that Hegel's remarks were made in German, and will need to be translated into English equivalents. (This point was fudged in relating (12) and (13), since Tarski's favorite sentence — even according to the fantasy being pursued — would not have been (12), but its Polish equivalent.) As in the pronominal case, the interpretation of prosentences bound by quantificational antecedents is yet more complex. 17) Every sentence Hegel wrote is true. This is usefully thought of in the expanded, explicitly conditional form 18) For any sentence, if Hegel wrote it, then it was true The immediate anaphoric antecedent of the prosentence is picked out by the pronoun 'it' that occurs in it, which is linked to the 'it' in "Hegel wrote it." This link determines the instances of the quantification, such as 19) If Hegel wrote "Die Vernunft ist Bewusstseins Gewissheit, alle Realität zu sein," then it is true. By combining various considerations advanced above, we can determine the sense of claims like this. By uttering (17), the speaker commits himself to all substitution instances of (19) — all the claims that have this form. There is one further sort of complication in settling the sense of the prosentence at the second stage - after one has picked out an anaphoric antecedent at the first stage. Be-

108

Deflationism Defended

sides taking into account the significance of the aforementioned distinctions between syntactically lazy, semantically lazy, and quantificational anaphoric connections to the antecedent, one must look at verbal modifications of the prosentence itself. 20) Before Weierstrass, mathematicians believed that every continuous curve must be somewhere differentiable, but he showed that that is not true. Here the crucial point is that such uses of 'true' be construed as having sentential operators applied to the underlying prosentence. So the final clause of (20) is understood as 21) Not ( it is true ). The whole thing then has the sense of 22) Not (every continuous curve must be somewhere differentiable). The verbal modifications indicating the application of sentential operators to presentences must be handled the same way in sentences involving tense and modality, as in 23) What Bismarck said about France in 1870 was true then. and 24) The sentence at the top of p. 23 of this book might be true. In each case, the modifier is to be thought of as applied after the antecedent has been determined, to the content inherited from that antecedent. From the point of view of this analysis, orthodox disquotationalist accounts have a number of deficiencies: •

They lose the anaphoric link between the prosentence formed using 'true' and its antecedent(s). It is not in general enough for a theory to entail simply that the two sentences have the same sense. That one inherits its sense from the other can also make a difference, just as we saw at the level of pro nouns in examples (2) and (3). I'll say a bit more about this below while discussing the role played by anaphora in securing interpersonal communication.

•

The only cases that are literally disquotational are those in which the anaphor picks out its antecedent by offering a quote name of it, as in (5) and (10). Even the shift from direct (quotational) to indirect discourse — from something like (10) to something like (14) - requires more than just disquotation. For here the paraphrase relation must be invoked to acknowledge that there is really a class of anaphoric antecedents to be taken into account, since there can be tokenings of many types that all count as sayings that-p. As one moves further away from quote-naming, for instance to picking out the antecedent tokenings by describing them (as in (13), (23), and (24)) the model of disquotation becomes correspondingly less useful in guiding us through the computation of antecedents. Here disquotation simply offers a bad theory of the process of determining the anaphoric antecedent. For in fact, prosentences can use all the referential apparatus of the language to do that job.

Explanatory vs. Expressive Deflationism about Truth

109

•

This disability leads directly to another, which concerns the next stage of interpretation. For one can no more more 'disquote' the demonstrative 'that' in "That is true," than one gets to a statement of Goldbach's conjecture by disquoting the expression "Goldbach's conjecture" in "Goldbach's conjecture is true."

•

Treating disquotation as a paradigm depends on a repetition model of anaphora: one in which the expression containing the anaphor is to be understood by replacing it with (another tokening of the same type as) its antecedent. But not all pronouns should be understood as working in the narrowly or syntactically lazy way, and the same goes for the pro sentences formed using 'true'. This fact is perhaps most evident when the proform is functioning quantificationally, but it appears already where the anaphorically dependent and antecedent tokenings are uttered by different speakers (or differ in some other index, such as time) and the antecedent contains indexical or token-reflexive expressions such as T , 'now', or 'that'. (And again if different languages are involved.) Since anaphora is a relation between tokenings, the use of tokenings of types such as 'That is true,' as a response to a tokening of Ί am hungry,' can be construed correctly - just as 'he' can have Τ as its antecedent without thereby referring to whoever uttered 'he'. An incautiously stated disquotational theory would get these indexical cases wrong.

•

Disquotational theories do not sufficiently articulate the process of computation of an antecedent and inheritance of content from it to indicate the role in that process of sentential modifiers applied to the prosentence formed using 'true': talk about what is not true, or was or will be true, or about what might or must be true.

In sum, disquotational theories ignore three crucial dimensions of fine structure that are integral to the anaphoric approach: the different ways an antecedent can be picked out (not just by quote names), the different sorts of content inheritance (not just lazy), and the different ways in which the content of the prosentence can be related to the content of the antecedent (verbal modifications may be needed). Along all these dimensions the account of 'true' as a prosentence-forming operator is more detailed and articulated, and offers more step-by-step guidance for actually determining the sense of the whole range of expressions in which 'true' can occur. Another advantage, which I believe has no analogue on the disquotational side, concerns the relation between 'true' and the corresponding semantic vocabulary that applies to essentially «^sentential expressions: terms such as 'refers', and 'denotes'. The theory that construes 'true' as a prosentence-forming operator generalizes smoothly and naturally to a treatment of'refers' as a pronoun-forming operator Its basic employment is in the construction of what may be called anaphorically indirect definite descriptions. These are expressions such as "the one the chairman referred to [represented, described, talked about] as 'almost a third-rate intellect'", understood as a pronoun whose anaphoric antecedent is some utterance by the chairman. A full-fledged pronominal or anaphoric theory of 'refers' talk can be generated by first showing how other uses of 'refers' and its cognates can be paraphrased so that 'refers' appears only inside indirect descriptions, and then explaining the use of these descriptions as pronouns formed by

110

Deflationism Defended

applying the 'refers' operator to some antecedent-specifying locution.4 Specifying the expressive role of 'refers' or 'denotes' in this way then permits the recursive generation of the Tarski biconditionals in a straightforward fashion. So treating 'true' as an operator that applies to a sentence nominalization and produces a prosentence anaphorically dependent upon the nominalized sentence token, and 'refers' as an operator that applies to an expression picking out a term tokening and produces a pronoun anaphorically dependent upon it permits a single theory form to explain the use of all legitimate semantic talk about truth and reference in purely anaphoric terms.

II. Why One Ought Not Take a Deflationary Attitude Toward the Expressive Role of 'True' Here, then, we have seen a sketch of the expressive role that is characteristic of the expression '...is true'. It is a verbally modifiable operator that applies to a singular term that picks out a sentence tokening (or class thereof), and forms a prosentence that anaphorically depends upon that sentence tokening (or class thereof) as its antecedent(s). Its content is to be computed on the basis of its relation to that antecedent, in any of the standard anaphoric ways, including quantificational ones. This specification of the functional role of this fundamental semantic vocabulary is sufficient both to identify expressions playing this role in alien languages, and to say what must be done to add their expressive power to languages that lack it. It would be a travesty to say that on this view truth locutions were redundant or eliminable. On the contrary, it is evident that the availability of such idioms contributes substantial expressive power to a language. In general, this contribution is just the extension to the level of whole sentences of the expressive power provided by anaphoric mechanisms already at the level of singular terms. The most obvious dimension of surplus expressive power contributed by anaphoric mechanisms is the quantificational. Anaphora is how natural languages achieve the effects secured by variable binding in formal languages such as the first order predicate calculus. Absent such a mechanism, there is no way to express what (17) says, any more than at the subsentential level one could express 25) Everybody loves somebody sometime. Tarski proved that the expressive power of formal languages containing "...is true" operators exceeds that of the corresponding semantically impoverished languages. This is due in no small part to the quantificational use of the prosentences such vocabulary introduces.5 But anaphora extends the expressive power of natural languages in substantial ways that have nothing to do with quantificational uses, as well.

4 5

I elaborate such a theory in "Reference Explained Away," Journal of Philosophy, LXXX1 no. 9, September 1984, pp. 469-492., and in Chapter Five of Making It Explicit. The best treatment I know of these matters is Anil Gupta and Nuel Belnaps The Revision Theory of Truth [MIT Press, Cambridge, Mass; 1993].

Explanatory vs. Expressive Deflationism about Truth

111

For one thing, anaphoric mechanisms are what make it possible to incorporate into the language otherwise unrepeatable expressions, paradigmatically demonstratives (and some uses of indexicals). In 26) # Look at that1 I wonder what it was. From the glimpse I got of the animal, it looked like a fox. But I'll bet it actually was a rabbit. # the original use of the demonstrative acquires its content from an essentially fleeting event. The glimpse it reports is not repeatable, not available to lend content in that sort of way to other speech (and thought) acts. What makes that content available for further thought and talk is the fact that it can be picked up and preserved anaphorically, as the initiator of a chain of anaphorically dependent expressions. No language can contain deictic mechanisms without also containing anaphoric ones. For apart from their capacity to anchor anaphoric chains, and so give rise to repeatable anaphoric chains, deictic tokenings would be linguistically idle, wheels that did not engage with the conceptual machinery of thinking and talking.6 Similarly, the paradigmatically indexical expression 'now' is a usable expression only because the content it introduces can be made available for further use, for instance in inferences, by anaphorically dependent tokenings of 'then, at that time', and so on. 7 In this way contents available to one person on one occasion even become available to other interlocutors. And this fact points to a second nonquantificational expressive function of anaphora: its role in communication. This role extends beyond generating repeatable structures (anaphoric chains) anchored by unrepeatable deictic and indexical tokenings. Suppose Β comes late into a conversation A is having: 27) A: # ...This comment by the policemanj makes him2 very angry. So then the guy2 jumps out of his2 car, and takes a swing at the copj! # Β might then jump into the conversation, saying something like 28) B: # I'll bet that the copx saw to it that that idiot2 spent the night in jail. No police officer could let his2 behavior go unpunished. # Here B, in a literal sense, does not know who he is talking about. Having missed the beginning of the conversation, which introduced the characters, he doesn't know whether A is talking about something he witnessed, something that was described to him, or recounting a piece of fiction he read. He has no idea who the impulsive motorist is. Yet by anaphorically picking up the chains A has displayed, Β settles it that he is talking (and thinking) about whoever it is that A was talking (and thinking) about. If A's claims have truth conditions and inferential consequences, then so do B's. Com-

6

7

I have argued in further detail for this conceptual dependence of deixis on anaphora in Section IV of Chapter Seven of Making It Explicit. On the general points being made here, see also Sections III and V. This is the central point I take Hegel to be making in the opening "Sense Certainty" section of the Phenomenology.

112

Deflationism Defended

munication in a fundamental sense is secured independently of what is going on in B's head, in that Β is in a position to undertake definite commitments, to talk about definite individuals (in the sense that the truth of his claims depends on how things are with those individuals), even though he is not capable of specifying who it is he is talking about other than by appeal to A. This capacity to talk and think without knowing what we are talking and thinking about is an essential aspect of interpersonal communication. Indeed, I think that this is the right way to think about what we are doing when we use proper names generally - that our tokenings continue anaphoric chains initiated by others, perhaps others long dead. That is, I think that the phenomena that causal or historical theories of proper name reference are getting at are best understood in terms of a more general notion of anaphoric links among expression tokenings. 8 The crucial expressive role played by /'«imentential, indeed, interpersonal anaphoric links in securing communication across gaps created by differences in information and belief is reflected in a specialized /«frasentential use of anaphora in ascriptions of propositional attitudes. Such ascriptions come in (at least) two forms, which can be syntactically regimented as: 29) De dicto B: A believes that the inventor of bifocals invented the lightning rod. 30) De re

B: A believes of Benjamin Franklin that he invented the lightning rod.

In Quine's terminology, in the de re form, a singular term has been exported from within the 'that' clause, where it resides in the de dicto form. The exported term becomes the anaphoric antecedent of a pronoun that marks its place in the scope of the 'that'. What does this ascription-structural anaphora have to do with interpersonal anaphora? Suppose that the original remark was 31) A:

The inventor of bifocals invented the lightning rod.

Then B's utterance of (29) will be fully warranted as a correct report of the belief expressed by A's claim. But suppose A does not (at least, according to B) believe that Benjamin Franklin is the inventor of bifocals. Then it would be /'«correct for Β to assert the de dicto 32) A believes that Benjamin Franklin invented the lightning rod. For if you asked A whether he believed what is expressed by the sentence used here to characterize his beliefs, he would deny it. Where it is B, and not A, who believes 33) Benjamin Franklin is the inventor of bifocals. Β should mark this divergence of belief by using the term 'Benjamin Franklin' in his ascription outside the scope of the 'that' clause, which specifies the ascribed belief in terms that the one to whom it is ascribed should acknowledge. This shows that the use

8

I have argued this point at greater length in Section V of Chapter Eight of Making It Explicit.

Explanatory vs. Expressive Deflationism about Truth

113

of that term is part of the commitment Β undertakes in producing the ascription, not part of the commitment he attributes - that is, that it is B, not A, who is responsible for using that term to express the content of the attributed belief. (This use of 'of is a syntactic regimentation of what is a much messier practice in natural languages. But the distinction between de dicto and de re ascriptions that it regiments is real and important. For present purposes it does not matter that often de re locutions are used to indicate more than just the difference of perspective I've pointed to here.^) Ascriptionstructural anaphora in de re ascriptions of prepositional attitude lets us keep our books straight on who is responsible for what in specifications by one individual of the content of the states and utterances of another. All three of these substantial nonquantificational expressive functions performed by .«^sentential anaphorically dependent expressions such as 'he' are also performed by sentential ones formed using 'true'. Thus someone might continue the discourse in (26) by saying 34) If what you said yesterday is true then it will be the first time anyone has ever seen a rabbit around here in the middle of the day. Here (we may suppose) the whole content of the tokening of "It was a rabbit" in (26) is being picked up as the antecedent of a conditional, so that its consequences can by explored hypothetically. And the conversation in (27) and (28) might be continued by another latecomer, who heard only B's remark 35) # C: That might not be truev What did the guy2 actually do?# Here the 'that' is picking up (either anaphorically or deictically) the final sentence-tokening of (28), and 'the guy' is anaphorically picking up the tokening of 'he' it contains. C's whole first sentence then is anaphorically dependent on the final sentence Β uttered in (28) (though it would be a delicate matter to make this out in terms of replaceability, since these utterances come out of different mouths, against the background of different information sets). The most striking parallel, however, concerns the ascription-structural anaphora. For expressions of other syntactic categories besides singular terms can be exported from de dicto ascriptions to form de re ones. Thus one can have: 36) B: A believes of the largest marine mammals that they will soon be extinct. where A would assent to believing this about whales, but is not sure whether they are the largest marine mammals, and 37) B: A j believes of Buster Crabbe's favorite form ofactivity2 that that2 is what heλ should be doing three times a week.

9

I discuss it at greater length in Sections I-V of Chapter Eight of Making It Explicit.

114

Deflationism Defended

if A does not know that Buster Crabbe's favorite form of activity is swimming. But besides common nouns and predicates, it is possible to export whole sentences. The anaphoric trace left inside the scope of the 'that' to mark the exportation is then a prosentence. So suppose that in 1951 Senator McCarthy would have assented to 38) The spectre of communism is haunting Europe. Someone else who knows - as McCarthy undoubtedly did not - that (38) is the first sentence of the Communist Manifesto, could report the belief McCarthy endorses in (38) by the de re·.

39) Senator McCarthy believed of the first sentence of the Communist Manifesto that it is true. (And on that basis, that McCarthy believed of some of the Communist Manifesto that it was true — horrified as the senator would have been by that allegation.) Thus when whole sentences are exported into de re position, one uses sentences formed from 'true' in the position of ascription-structural anaphors. I think this fact is as compelling evidence as there well could be for construing such sentences as anaphoric presentences a striking confirmation of the analysis recommended above. Thus the presence of 'true' and its cognates in a language adds at the sentential level all of the crucial expressive power added by anaphorically dependent expressions at the subsentential level: •

The capacity to make new quantificationally complex claims,

•

The capacity to pick up deictic and other otherwise unrepeatable expressions and use them in further conceptual endeavors, paradigmatically as premises in inference,

•

The capacity to secure interpersonal communication across substantial differences in belief and information among the interlocutors, and

•

The capacity to make explicit who is responsible for what when one interlocutor characterizes the beliefs of another.

Anaphora generally plays an essential and ineliminable expressive role. In making possible the formation of prosentences 'true' adds correspondingly significant expressive resources to the language. Though anaphora is about redundancy in the sense of repeatability, as a linguistic mechanism it is itself anything but a redundancy. I conclude that one should not be a deflationist about the expressive role of 'true'.

III. The Sense in which One Ought to Take a Deflationary Attitude Toward the Explanatory Role of 'True' Theories of truth are often thought of as 'deflationist' in an ontological sense. Here the question is whether truth is a property, or perhaps, whether it is a substantive property. A feature dear to the hearts of the originators of the prosentential theory, as to disquo-

Explanatory vs. Expressive Deflationist!! about Truth

115

tationalists, is the metaphysical parsimony of the approach. For what in the past were explained as attributions of a special and mysterious property (truth) are exhibited instead as uses of grammatical proforms anaphorically referring only to the sentence tokenings that are their antecedents. The approach is intended to be, as one might say, ontologically deflating - or at least unexciting. In an influential article, Paul Boghossian has pointed out the potential for instability in an ontologically deflationary view that sees the ontological issue as a question that goes beyond asking whether'... is true' is a predicate. 10 In that case, he argues, this sort of parsimony must undercut itself and lapse into incoherence. The general worry Boghossian raises is that the force of deflationist claims depends on the contrast between predicates (such as '...has a mass of more than ten grams') that do, and those (such as '...is true') that do not, correspond to properties. Such contrasts seem to presuppose a robust correspondence theory of the contents of some predicates - at least those the semantic deflationist finds unproblematic, paradigmatically those of natural science. But consistently following out the rejection of robust correspondence theories of content requires treating using an expression as a predicate as all there is to expressing a property, and using a declarative sentence to make a true claim to be all there is to stating a fact. So on a deflationary construal, one is forbidden to deny that the predicate '...is true' denotes a property. In this way, theories that deny that truth is a property can be seen to be conceptually unstable. Notice, however, that this argument depends on treating "...is true" as a predicate. If it is, then since that expression is used to make claims and state facts, it must, on deflationary accounts, be taken to express a property. But the essence of the anaphoric approach to truth talk is precisely to take issue with this grammatical presupposition. According to those accounts, "...is true" expresses a prosentence-forming operator. Its syntax and grammar are quite distinct from those of predicates, to which it bears only the sort of surface similarity that quantificational expressions bear to genuine singular terms. The part of speech "...is true" is assimilated to by these theories does not have a directly denotational semantics. Rather, tokenings formed using "...is true", but inherit their significance anaphorically, by an entirely distinct mechanism. So when it is claimed here that "...is true" does not express a property, this means that it is not even of the right grammatical form to do so - any more than 'no-one' is of the right form to pick out an individual, although there are some features of its use that could mislead one on this point. Further, this claim is not made ad hoc, to avoid the sort of theoretical circularity Boghossian points out, but is motivated by ground-level considerations having to do with the unifying a variety of uses of 'true' and 'refers' in a theo-

10

Paul Boghossian, "The Status of Content", The Philosophical Review April, 1990. 1 have in mind the argument epitomized on p. 181 in the claim that: "...the denial that a given predicate refers to or expresses a property only makes sense on a robust construal of predicate reference...But if this is correct, the denial...that the truth predicate refers to a property, must itself be understood as framed in terms of a robust notion of reference..."

116

Deflationism Defended

retically perspicuous way. Thus the anaphoric understanding of the expressive role of 'true' is immune to arguments of the sort Boghossian deploys. According to the anaphoric approach, "...is true" is a prosentence-forming operator, and no more expresses a property than 'it' does. But the issue that people are after when the deny that "... is true" expresses a substantive property is not really addressed by this grammatical point. I think that issue is best understood as concerning the proper explanatory role that truth locutions can be called on to play. Although one who endorses the anaphoric account of the use of'true' (and 'refers') cannot put the issue in ontological terms of properties (and relations) — and is to that extent an ontological deflationist - such a theorist is committed to various consequences concerning the suitability of prosentences formed using 'true' for various sorts of explanatory project. In particular, telling the anaphoric story about the expressive role of truth commits one to seeing it as capable of playing an important role in local explanations of meaning, and as precluded from playing an important role in global explanations of meaningfulness in general. In particular, if the anaphoric account of the expressive role of 'true' is correct, then it is a fundamental mistake to understand propositional contentfulness by appealing to a prior notion of truth conditions. For the uses of 'true' that one would make in such an explanation themselves presuppose a notion of propositional contentfulness. By "local explanations of meaning" I mean explanations of the meaning of particular expressions. It follows from the expressive role of 'true' that it is often usefully appealed to in such explanations. So we can say things like 40) Any claim of the form -p is true just in case ρ is not true, to explain the use of the tilde, and 41) ρ entails q just in case whenever ρ is true, q is true, to explain the notion of entailment. And because we can do that, we can understand a definition such as 42) Any natural number η is a prime number if and only if it is only evenly divisible by itself and 1, as explaining the concept prime number by offering truth conditions for it. For a quantified biconditional like (42) is true just in case if one side of the biconditional is true, then the other is also true. These are all truth claims that can be parsed prosententially. And thinking about the sort of quantification that is implicitly involved in such explanations of the meanings of particular expressions shows why prosentences are useful in expressing them. Thus we can see, according to the anaphoric approach to the expressive role of 'true', why explanations of meaning can naturally take the form of specifications of truth conditions: claims to the effect that sentences containing the expression whose meaning is to be explained are true just in case Even if, as in (42) the explanation does not itself use the word 'true', in explaining what we are doing in offering such explanations, we will need to generalize in a way that requires using that term (or one of its cognates, such as 'holds', 'obtains', 'is the case' and so on). Here

Explanatory vs. Expressive Deflationist!! about Truth

117

'true' plays an essential role in expressing claims (especially general ones) about meaning. O n the other hand, if one understands the expressive role of 'true' in the way recommended here, then one is precluded from making certain other sorts of fundamental explanatory appeals to the notion of truth, and hence of truth conditions. In particular, I think that one cannot explain the notion of anaphora that is relied upon by broadly prosentential theories without appealing to an antecedent notion of propositional content - what in the simplest cases is inherited by a presentence from its anaphoric antecedent. That is, one cannot entitle oneself to employ a notion of anaphora in one's semantic theory unless one is already entitled to use a notion of propositional content. Thus if one's explanation of 'true', and hence of truth conditions, is dependent upon a notion of anaphora, one cannot without circularity explain the notion of propositional contents in terms of truth or truth conditions. 11 For those notions cannot be made available for explanatory use in advance of an account of propositional content. This consequence is not special to the anaphoric account of the expressive role of'true'. Orthodox disquotational accounts equally preclude one from treating the notion of truth, and hence of truth conditions, as explanatory raw materials suitable for use in explaining what it is for a sentence to mean something. For they evidently take for granted the meanings of the sentences that are the results of disquotation. So disquotational and anaphoric accounts are alike in their global explanatory deflationism. This is what I propose one ought to mean by 'deflationism', when it is unqualified by an adjective. It is what I think is properly seen as standing behind misleading ontological talk of truth as not a 'substantive' property - 'substantive' in this context making implicit reference to its availability for a certain sort of explanatory project. And it is this disqualification of truth from playing a substantive explanatory role in accounting for semantic meaningfulness in general that tempts some to expressive deflationism: the view that truth talk adds no significant or indispensable expressive resources to a language. What sort of explanatory undertaking, exactly, is it that global explanatory deflationism about truth rules out? In Fregean terms, what it rules out is theories that seek to put a notion of truth in place in advance of a notion of sense. (Frege's own theory does not have this shape.) That is, it rules out attempts to explain what it is for a sign design to express a thought (that is, the sense of a declarative sentence) by appealing to a prior notion of what it is for the sign design to stand in the right relation to things to be true. The idea would then be to understand the sense, meaning, or content that the sign design expresses in terms of the distinction between ways the world could be that

11 This argument is reminiscent of one Dummett offers against the availability of truth-conditional semantic theories to those endorsing redundancy theories of truth. I think there is something to the analogy, but I think the particular role assigned to the notion of anaphora makes this is a good argument, while I am not convinced that Dummetts is.

118

Deflationism Defended

would make it true, and ways the world could be that would not make it true: its truth conditions. From the point of view of such an explanatory project, a notion of truth (of a sign design) as correspondence (to the world) counts as robust or substantive in case it can itself be explained or otherwise put in place without appeal to a notion of (propositional) sense, meaning, or content. Fregeans, by contrast, see the notions of truth and sense as two sides of one coin - neither as explicable in advance of or without appeal to the other. The anaphoric theory, I think leaves room for the possibility of an account that starts with a notion of sense or content explicated without explicit appeal to a notion of truth (i.e. without the use of truth locutions), to which the expressive power of truth locutions might then be added by introducing suitable anaphoric mechanisms. I'll say a word or two about that converse direction of explanation in closing below. So does the anaphoric account of the expressive role distinctive of 'true' and its cognates show the incoherence or impossibility of an order of semantic explanation that begins with a robust notion of correspondence between linguistic sign designs and features of the world? No. I think it may be possible to mount such arguments, perhaps by arguing first that no such notion of correspondence or propositional representation can do without a notion of facts or states of affairs on the worldly side of the relation, and then second that no story can entitle itself to such a notion unless it appeals to the practices of using expressions as sentences in the making of claims, and finally that such appeals are already tantamount to a theory of sense. But even if that were right, the anaphoric deepening and generalization of disquotational construals of the expressive role of 'true' could serve as nothing more than a preliminary softening up for such an argument. What such accounts can do is to undercut the motivation robust semantic explanatory appeals to notions of truth and truth conditions derive from the practice of saying what some particular expression means by specifying the conditions under which it would be true. By explaining the expressive role of 'true' as they do, such theories challenge the justification for identifying the property sign designs are taken to have in virtue of standing in a specified technical theoretical relation to the world as truth. Thinking that some property could be so identified is a mistake resulting from misunderstanding the grammar of the word 'true' - on a par with taking some object made available by one's theory to be what is referred to by the word 'something', or 'no-one'. It is, according to the anaphoric account of the expressive role 'true', wrong (though tempting) to think that one can explain what propositional contentfulness is in general in terms of possession of truth conditions. So, I have been urging, deflationists ought to acknowledge the possibility of expressing semantic content truth-conditionally, while denying the possibility of explaining semantic content in general truth-conditionally. This result will be unpalatable insofar as one cannot see how else one might begin to think about contentfulness than in terms of truth conditions. 12 Indeed, I take it that

12 Thus Boghossian, for instance, just assumes that content must be understood in terms of truth condi-

Explanatory vs. Expressive Deflationism about Truth

119

one of the major sources of resistance to deflationary approaches to truth is precisely that they rule out what has seemed to many as the only possible form of semantic explanation. If propositional contentfulness is not to be understood in terms of an antecedently explicable notion of truth conditions, how is it to be understood? I think this question should be taken very seriously indeed. Anyone who endorses deflationary accounts of the use of 'true', such as the anaphoric one I have been sketching, or its cruder disquotational cousins, it seems to me, owes an answer to questions like: How do you propose to understand the content of the sentences that serve as anaphoric antecedents for presentences (or result from applying operations of disquotation from quote names)? In addition, anyone subscribing to the specifically anaphoric approach owes a general account of anaphora and anaphoric chains (as the disquotationalist owes an account of disquotation in general). The challenge put by the dominance of truth conditional approaches to semantics is not adequately responded to simply by making the case for anaphoric or disquotational theories that underwrite global explanatory deflationism about truth. According to such theories, semantics - the study of cognitive or conceptual meaningfulness in general - is not best understood as the study of truth and truth conditions. But then, how should it be understood? I think that there are a variety of promising avenues available for exploration in responding to this question. I have developed one of them in detail in Making It Explicit. There I understand propositional contentfulness in terms of inferential relations, specified without use of truth locutions. The contents of subsentential expressions are then explained in terms of their role in specifically substitutional inferences. Anaphora is explained in terms of various sorts of inheritance of substitution-inferential potential. But that is all truly a story for another occasion, a story for which discussion of how to understand the use of 'true' can serve at best as an appetizer.

tions [op. cit., ρ 173]. It should not be surprising that those who start from such a presupposition then find theories that take a deflationary attitude toward the explanatory use of 'true insupportable.

O n Locating Our Interest in Truth DOROTHY GROVER

Deflationists with respect to truth are charged with separating truth from the philosophically interesting issues that are usually associated with truth. Deflationists agree, arguing the separation is a merit. My task will be to show that the claim, while true, is misleading. I shall argue that some of the most challenging of philosophical issues do indeed concern "truth," though not as usually thought the nature of truth, but whatis-true and criteria for determining what-is-true, where 'true' is read in a deflationary way. All of this has implications for what we might classify as theories of truth. I conclude with a brief survey of the possibilities. Deflationists have argued the truth predicate is used by us for expressing generalizations. For many, this logical role is its primary or only role. Deflationists also defend a thesis that is negative relative to the substantive versions of correspondence, coherence, and pragmatic theories. Given the assumption that the truth predicate is used in generalization; and given the failure (in the deflationists' view) of substantive theorists to show that truth has an explanatory role, 1 deflationists have argued there is no need to seek an analysis of the nature of a substantive truth property. This rejection of a truth property, and acceptance of a logical role for 'true', amounts to the separation of truth (as a property, at least) from the more interesting philosophical issues.2 And, anyway, deflationary theories are relatively uninteresting— certainly uninteresting compared with the purported interest of theories that claim an ontological, epistemological, or language-related explanatory role for truth. How can the separation be a merit? Deflationists claim that they have clarified the truth talk employed in discussions of the "deeper" issues; they have also shown that inquiry can proceed without the diversion that has been occasioned by the pursuit of the ever elusive truth property. The merits of the separation lie in this fact that theorizing about a truth property is irrelevant to inquiry. Nevertheless truth is important to us, very important. How, if the truth predicate has only a logical role, can deflationists account for its interest and importance? I address this question from the assumptions of the prosentential theory, the version of deflationism that I favour.

1

2

It has been claimed that truth has an explanatory role in accounting, for example, for realism and the success of science (Putnam 1978, p.21); also, theories of meaning (Devitt 1984, Field 1986). Williams 1986 and Horwich 1990 have argued that the arguments for explanatory role fail to make their point. I have questioned the need for a truth property in Grover 1981, 1990b and 1992. This mention of more interesting topics oversimplifies, given the overlapping nature of all the inquiries in question. I suppose what is at issue is that deflationary theorists have not come up with a theory that provides answers to the kinds of questions many expect from a theory of truth.

O n Locating Our Interest in Truth

121

1. The Prosentential Theory* I initially employed the concept of a presentence (introduced by analogy with pronouns) to provide a reading of anaphoric occurrences of bound prepositional variables. Consider, for example, the formula -Ξρ (Albert believes that ρ and p). 'It is true' provides a reading in English by doing the same kind of logical work that the anaphoric bound prepositional variables do: each provides ways of making the logical connections we need for expressing generalizations. This is demonstrated in the following There is nothing such that Albert believes it is true and it is true or more colloquially,4 Nothing Albert believes is true. The basic claim of the prosentential theory is that 'it is true' and 'that is true' function as presentences in English. Much as pronouns are used for generalizing with respect to nominal positions, so 'it is true' tends to be used for generalization with respect to sentence positions. 'That is true,' on the other hand, tends to be used much as "pronouns of laziness" are used-except that 'that is true' occupies a sentential position. This means that in simple cases it stands in for something else that has been said in the context. For example, in Mary: Snow is white. Tom: If that is true, snow will glare in the sun. 'that is true has the sentence 'snow is white' as its antecedent, just as 'she' in Because Janet likes to ski she will not want to come hiking with us. has the name 'Janet' as its antecedent. And just as 'she' picks up its referent from 'Janet' so 'that is true' will acquire its content from Mary's statement above. So, Tom in effect is saying, 'If snow is white, snow will glare in the sun.' This is partially what it is for 'that-is-true' to function as a presentence. Note that this means 'that is true' is used to say something about snow, and not to say something about a property of a sentence. Presentences can be modified, as in, 'that was true,' 'that might be true,' and 'what is true?' For example:

3

4

Readers unfamiliar with the prosentential theory are referred to my collection of papers in Grover 1992 where the notion of a presentence is explained and where the assumption that 'true' is a presentence forming predicate is explored in much greater detail. There is much work to be done in working out how the syntax works in English for the prosentential reading. Kent Wilson (Kent Wilson 1990) has raised a number of challenges.

122

Deflationism Defended

Tom: That was true, but not in big cities where there is much pollution and the snow gets so dirty. In such simple cases, the modified prosentence can be said to stand in for a modified form of the antecedent which, in this case, would be snow was white'. This prosentential picture serves to explain many pragmatic features of our truth talk, features highlighted in Strawson's 1950 work on truth. For example by using 'that is true' rather than 'snow is white' Tom implicitly acknowledges that snow's being white was mentioned before he contemplated the implications. Anaphoric devices seem essential if we are to establish the connections needed for communication: they facilitate the process whereby different speakers are able to talk about, and know they are talking about, the same things; they also provide logical connections needed for generalization. I mentioned above that 'it is true' tends to be used for generalization, as in, I know that his story sounds far-fetched, but for anything he said, if he said itis-true, it-is-true. Generalizations have instances. In this case, I know his story sounds far-fetched, but if he said that his cat imitated the chirping of birds, his cat imitated the chirping of birds. I have found that in those cases where 'true' is used to express a generalization, the form of the prosentential reading is often best conveyed through a paraphrase that uses bound propositional variables. In the case of the last example, the propositional variable paraphrase might take the form I know his story sounds far-fetched, but Vp(if he said that ρ then p) Other deflationists also claim a generalizing role for 'true'. There is a difference however. While the notion of anaphora is my basic notion, other deflationists tend to take Tsentences (sentences of the form '"Snow is white' is true iff snow is white") as paradigmatic of our use of'true'. In the case of Horwich 1990, propositions are the bearers of truth with, roughly, the proposition χ being true just in case p, where χ is the proposition expressed by 'p'. This means that while the prosentential construal of the truth predicate keeps discourse at the level of the "object language," so to speak, other deflationists describe 'true' as functioning as a metalinguistic predicate.^ But let us now take up the question of deflationism and its relationship to interesting issues.

2. About What-is-True I begin by considering some relatively basic truth talk that arises in ontological and epistemological inquiries. Consider the question, 'What is true?' On a prosentential account of the role of the truth predicate, the question 'What is true?' will have as its

5

I compared these approaches in Grover 1990a.

On Locating Our Interest in Truth

123

form 'For each p, p?' For the moment, the instances of this question provide our focus because they show what is being asked when 'true' is read prosententially. The instances include Are electrons basic units of the physical world? Do people act freely? Does gratuitous violence in entertainment programs nurture violence in society? In asking and answering such questions we hope to learn about the way the world is, i.e., we hope to learn what-is-true. (I use hyphens to remind readers that I intend a prosentential reading of the truth predicate-a reading that keeps 'true' at the level of the object language.) On this prosentential reading the candidate answers to these questions include 'electrons are basic units in the physical world,' 'electrons are not basic units in the physical world,' 'people act freely,' 'people do not act freely', etc. Note that there is no mention in these answers of the property truth or its alleged bearers. Rather, there is talk of worldly things like electrons, people, actions, and violence. Because instances of 'What-is-true?' are used to ask about worldly things, we need experts in these matters (e.g., scientists, perhaps philosophers in the free will case) to provide the answers. By contrast, on the non-deflationary reading, i.e., on an analysis of'true' as property ascribing, the question 'What is true?' does not yield a general question which has specific questions as instances. It seems to ask for a list that might begin: "the sentence 'electrons are basic units of the physical universe'"; or perhaps, 'the belief that people act freely'. Alternatively, the question may be read as asking for a characterization of the bearers that have the property truth. There are also epistemic questions: Under what conditions would we know whether something is true? The instances of the prosentential reading would include: Under what conditions would we know whether electrons are basic units of the physical universe? Under what conditions would we know whether people act freely? Under what conditions would we know whether gratuitous violence in entertainment programs nurtures violence in society? The experts in worldly matters (e.g., physicists) will perhaps have criteria for determining whether we know that electrons are among the basic units of physics, etc.; epistemologists, who presumably are informed by scientific practices, may also have theories as to the criteria that should be used. Whatever this mix is, what is in question are the conditions under which we have knowledge that the world is this way rather than that way (e.g., criteria for determining whether we have knowledge that electrons are basic units of the universe). Criteria for determining whether we have knowledge, that a given linguistic item (as a bearer of truth) has the property truth, does not enter the picture. 'What-is-true?' asks about truth, in the sense that it asks what-is-true (it asks which way the world is). Similarly, instances of the epistemic question, 'Under what condi-

124

Deflationism Defended

tions would we know whether something is true?' are questions about criteria for having knowledge of what-is-true (knowledge of which way the world is). So, I have shown that there are interesting questions that can be asked using 'what-is-true.' They concern truth in the sense that the concern is with what-is-true, and our knowledge of what-is-true. However, so far, this prosentential reading of the truth predicate has led almost only to inquiries found in individual disciplines. How are we to get to the "deep" ontological, epistemological, and communication, issues that have been held by many to be associated with truth?

3. Truth, Inquiry, and Assertion Cheryl Misak 1998 is one philosopher who has criticized deflationary theorists for separating truth from significant philosophical issues. Among deflationists, her primary focus is the Minimalist Theory of Horwich 1990. She represents Horwich as using the schema DS to "define" a logical property truth. (Misak uses 'DS' to refer to the disquotational scheme 'x is true iff p'—'"people act freely' is true iff people act freely" would be one of its instances.) This logical property gives 'true' its role in generalization. While Misak accepts the generalizing role of 'true', she argues that we need more on truth than DS provides: "a kind of pragmatism best captures what is important about truth." (p.407) O f disquotationalists she says that they "will not want to add the pragmatists' thought that truth would forever be assertable" (p.409) and "if we stop with the disquotationalists we fail to give a full account of how truth is linked to our practices of deliberation and experimentation; we fail to live up to the demand of making sense of inquiry" (p.412). Misak has her eye on the importance of connecting truth (seemingly, as a property) to inquiry, assertion, and deliberation. Rather than defending Horwich's theory against Misak's charges, my goal is to defend deflationism more generally by showing that a prosentationalist can articulate connections between truth and inquiry, assertion, and deliberation-should he or she want to do this. Misak points out that while Peirce is opposed to engaging with definition, especially the task of defining truth, he nevertheless attempts an "elucidation" of truth. 6 The elucidation contains various claims about truth that (at least partially) articulate the connections between it and inquiry, assertion, and deliberation. For example, 'a true belief is one on which inquiry could not improve' (Misak 1998, p. 408) and 'we take truth to be our aim when we assert, inquire, and deliberate. ... were we to get a belief which would be as good as it could be, that would be a true belief" 7 (Misak 1998,

6

7

"So a pragmatic elucidation of truth is neither a definition nor a criterion of truth. It is a specification of what one can expect of a true hypothesis. ... The expectation must be pragmatically significant-it must really lead us to expect something of the course of experience," Misak 1991, p.42. See also Misak 1991, p.44. Misak 1998 offers an elaboration of "as good as it could be." Empirical adequacy, coherence, simplicity, and explanatory power are explicitly mentioned on p.410.

O n Locating Our Interest in Truth

125

p.410) Deflationists will have missed the boat if they have eliminated the possibility of connecting truth, in some way, with deliberation, assertion and inquiry. I begin with the pragmatist's claim that truth is our aim when we inquire. I shall assume this means roughly, 'the aim of inquiry is to acquire knowledge of all truths.' 8 Substantive truth theorists (not Peirce, nor deflationists) may likely read the claim as saying that inquiry's aim is to determine for any given bearer of truth whether it has the property truth. In that case, we would need to have knowledge of what the property truth is, as well as criteria for determining when a truth bearer has the property.9 The prosentential reading comes out quite differently, of course. Using propositional variables, we have The goal of inquiry is such that for any p, just in case p, we know that p. This in turn can be paraphrased, using prosentences and a quantifier, as A goal of inquiry is that we acquire knowledge that something is true, just in case it is true. Such prosentential generalizations are best understood through consideration of their instances. Two instances of the thesis are: A goal of inquiry is to acquire knowledge that electrons are basic units of the physical universe, just in case electrons are the basic units of the physical world. A goal of inquiry is to acquire knowledge that gratuitous violence in entertainment nurtures violence, just in case gratuitous violence in entertainment nurtures violence. Because there is nothing remarkable in these instances, I think appreciation of the directness of the prosentential reading is facilitated by drawing the comparison between the above reading and that which is yielded by a substantive truth property ascribing reading of the predicate. In the latter case, 'the goal of inquiry is knowledge of all truths' has the following instances:

8

9

This statement is implausible as it stands. D o we want all truths, or only the interesting ones? D o we care whether true beliefs come at the "cost" of having false beliefs? Does inquiry have other goals? Etc. Misak 1991, e.g., p. 121, allows that such elucidations of truth must be phrased more carefully. Jennifer Faust 1995 draws attention to epistemologists who think they must provide a definition of truth before offering a theory of epistemology. There is Blanshard 1939, for example, who thinks that he needs a coherence theory of truth if he is to show that a coherence theory of the criteria for truth suffices for a theory of justification. Bonjour 1985 argues that he must respond to the sceptic by showing that "an essential part of the task of an adequate epistemological theory, in addition to providing an account of the standards of epistemic justification, is to provide an argument or rationale of some sort for thinking that an inquirer who accepts beliefs which are justified according to that account ... is thereby at least likely to arrive at truth . . . " (p. 157). Bonjour concludes this means that he needs an account of a truth property. Bonjour argues the correspondence theory is needed if he is to provide the meta-justification he thinks he needs for his coherence theory of knowledge. But, as Faust points out, this business of trying to interpose a truth property is a matter of creating pointless work. A task we do have is to find out what-is-true. That is enough to keep all of us busy!

126

Deflationism Defended

A goal of inquiry is that we acquire knowledge that the sentence/proposition 'electrons are the basic units of the physical world' is true just in case electrons are the basic units of the physical world. A goal of inquiry is that we acquire knowledge that the sentence/proposition 'gratuitous violence in entertainment nurtures violence' is true just in case gratuitous violence in entertainment nurtures violence.' The contrast is that on the prosentential reading the goal of inquiry is to know about worldly matters; whereas in the case of the substantive truth reading, the goal of inquiry is to know which candidate bearers of truth have the property truth. Given that substantive truth theorists are likely to use something like DS to (eventually) arrive at knowledge of what-is-true, the difference between the two is that the prosententialist has eliminated the detour through the truth property. Now which rendering would best serve the pragmatist? Is it the property truth that is linked to inquiry, as Misak seems to assume? Consider the prosentential reading. On that reading the aim of inquiry is to learn what-is-true, to get the world figured out, e.g., to learn whether electrons are basic units, whether people act freely, and so on. That is, in saying that truth is the aim of inquiry, we are saying that inquiry is at least partially elucidated in terms of an attitude we have towards the acquisition of knowledge of what-is-true. More simply, the aim of inquiry is knowledge. While this oversimplifies what the pragmatists want to say about inquiry, I suspect something like this reading would serve them well. So, what of the separation of truth from the interesting issues in philosophy? Certainly truth has been separated off (eliminated), in the sense that the truth property doesn't have a role here. But this is of lesser importance than the fact that one can still do important things like, talk about what-is-true, and speculate about coming to know what-is-true, (without having to first figure out what the bearers of truth are and what the truth property is)}Q So truth has not disappeared in every sense: we still can give expression to our concerns with what-is-true. Having made these distinctions with respect to our truth talk, I now find ambiguities in some of the claims of Horwich 1990. Take, for example, this claim on p.8 W h a t I a m c l a i m i n g o n behalf o f the m i n i m a l i s t c o n c e p t i o n o f t r u t h is n o t that it, b y itself w i l l e n g e n d e r realism o r anti-realism; b u t rather t h a t it will m a k e it easier f o r us to see t h a t t h e central aspects o f t h e realism debate have n o t h i n g to d o w i t h t r u t h .

I agree that the realism debate has nothing to do with a truth property-truth does not have an explanatory role in the debate; nor does a deflationary theory engender either realism or anti-realism. But this does not mean the realism debate has nothing to do with truth: the realism debate is about what there is, it concerns what-is-true. A prosentential reading of 'truth is our aim when we assert' is also available to those who want to assert or consider such a claim. 11 The propositional variable form is:

10 11

Horwich 1 9 9 0 has given a good account of the merits of the separation. Again, the claim oversimplifies. For example, sometimes in the interests of efficiency we assert something that is not quite true, but close enough for all concerned; sometimes we may assert something to mislead; we can also assert that something is the case because we want to sound authoritative.

On Locating Our Interest in Truth

127

For any p, our aim is to assert that p, only if p. A prosentential reading is: Our aim is to assert that something is true, only if it-is-true. This has instances like We aim to assert that electrons are basic units in the physical universe only if electrons are basic units in the physical universe. And so we have the possibility of linking our attitudes in making assertions with asserting what-is-true. 'Truth would forever be assertable' might be captured in For any p, if p, then at all times t, it is assertable that ρ and If something is true, then it is forever assertable that it-is-true. The situation of deliberation can be similarly treated. 'Truth is the aim of inquiry,' 'truth is the aim of assertion,' and 'truth is the aim of deliberation belong, if true, to theories of inquiry, assertion, and deliberation, respectively-they are not part of an elucidation of truth as a property. Truth as a property has been removed from inquiry, assertion, and deliberation; but we have seen this does not mean we have left no room for our concern with what-is-true. It is in this sense that 'truth" has not, in fact, been removed from the interesting issues. Given Peirce's reluctance to engage with the task of defining truth, I continue to think he might have happily endorsed a deflationary account of truth, given such an endorsement could be given without cost to his concerns with the nature of inquiry, assertion, and deliberation. 12

4. Theories ofWhat-Is-True I have previously referred to the prosentential theory as a theory of truth, but that is not right: the prosentential theory is a theory of the truth predicate. And now I have shown that a claim like 'truth is the aim of inquiry' belongs, not in a theory of a truth property, but in a theory of inquiry. Theories were classified according to the position taken on the nature of the truth property (e.g., correspondence, pragmatic/coherence, or deflationary). But if there is no truth property, this hardly seems a useful way to describe ongoing work.

12

I gave prosentential readings of both the question 'What is true?', and some pragmatist quotations, in §3.2 of my Introductory Essay in Grover 1992. My position was that pragmatists might have been better served if they had employed a prosentential truth predicate; for prosentential readings of the quotations portrayed pragmatists as offering criteria for determining what-is-true rather than as elucidating the nature of truth.

128

Deflationism Defended

More to the point would be to think in terms of theories of what-is-true. But all inquiries that purport to say how the world is would so qualify, and this includes the sciences, possibly literature, as well as most (perhaps all) of philosophy. The following suggests one way of dividing up the many different ways in which we inquire into what-is-true, and so the many sub-groups of theories of what-is-true that may ensue. Inquiring Into What-is-true I. Disciplines Other than Philosophy (i) The Sciences: the sciences are concerned with what-is-true, i.e., concerned with the way the world is; (ii) Literature, Visual and Performing Arts: though candidates for inclusion, I do not mean to suggest that all they might have to offer is an account of the way the world is; (iii) Religion: also a candidate for inclusion. II. Philosophy (i) Metaphysics (ii) Theories of Knowledge (knowledge of what-is-true) (a) Epistemology (b) Other Disciplines: I include these in so far as they address epistemological issues (iii) Language and Logic (a) I include studies of the expressibility of the truth predicate in English and in formal languages; such investigations have led to theories such as Horwich's Minimalism and the Prosentential Theory; (b) Logic (including logic of the truth predicate): especially theories of inference and Tarski style definitions, since they seek to explain the logical relations that obtain between statements that are used to say what-is-true; (iv) Other Areas of Philosophy These are included in so far as they say something about the way the world is. (v) Our Interest and Concern with What-is-True Candidates for inclusion are theories of inquiry, assertion, deliberation, and rationality, since all seem to connect our attitudes to knowing what-is-true. I would also want to include our explanations as to why, when we are inquiring, asserting, and so on, we care about what-is-true; or perhaps I should say, why we care about inquiry, assertion, and deliberation. This list is not likely to be exhaustive of those inquiries that lead to theories of what-istrue, nor are the divisions mutually exclusive. For example: there's overlap between the expressibility provided by the truth predicate and its logic; also scientists engage in philosophical debates concerning the status of theories as well as debates concerning more worldly things. As there is not space to elaborate on all of the above subdivisions, I will restrict my discussion to some issues that come up under the headings of metaphysics and logic. In metaphysics, the issues raised in debates as to what-is-true include: the status of scientific theories, the unity of science, relations between the scientistic and humanis-

O n Locating Our Interest in Truth

129

tic standpoints, and the status of moral statements. Such debates address philosophically interesting issues, leading us to theories of theories of what-is-true. It is interesting to note the variety of issues raised in these philosophical debates. Take, for example, the debate concerning the relation between the scientistic and humanistic standpoints found in the recent volume of the Library of Living Philosophers, The Philosophy of P. F. Strawson. Several contributors (Susan Haack, E.A. Adams, Payanot Butchvarov, Simon Blackburn, and David Pears) respond to Strawson's (e.g., Strawson 1985) claim that we can, and do, accept both the scientistic standpoint relative to which there are causal explanations and scientifically approved properties; and the humanistic standpoint which we occupy as social beings and according to which there are phenomenal and moral properties. The categorial differences between the two standpoints raise the issue of whether, and if, the standpoints can be unified. Strawson claims there is no contradiction in occupying both standpoints. To abandon one standpoint in favour of the other would be as absurd as supposing that somebody engaged in establishing the validity of a mathematical proof... was incapable of appreciating, at the same time, its elegance ... (Strawson 1998, p.88).

Adams 1998 argues the humanistic standpoint is basic and that the scientific is contained within it, while Blackburn 1998 suggests the humanistic supervenes in some way on the scientistic. Pears 1998 argues that there is no possibility that determinism could nullify responsibility since determinism is an untestable thesis about the general structure of the world. At an earlier time, Sellars 1962, in addressing the compatibility and viability of the scientific and manifest images, considered relevant 'is'-'ought' differences. Prediction, simplicity, comprehensiveness, paradigms and scientific revolutions, are all issues in debates concerning the relative status of scientific theories. Note that the overlap between the issues arising in these debates as to the way the world is (what-is-true) and those found in debates about the property truth is marginal. To the extent that language issues come up in both, there is some overlap with the correspondence theory. In the discussions leading to theories of theories of what-is-true, there is attention paid to categorial differences; also, scientists who are at the cutting edge will be in the business of honing concepts appropriate for use in their proposed accounts of what-is-true. 13 But these concerns are a far cry from those of correspondence theorists. Appeal to prediction, simplicity, and comprehensiveness take us a bit closer to the concerns of the pragmatists. But then is their contribution really best understood in terms of providing an analysis of a truth property, or are they better represented as being concerned with acquiring knowledge of what-is-true? The debates that address the status of our theories of what-is-true all address interesting and important issues; but, yet again, a prosentential truth predicate doesn't deny philosophers the issues and debates-indeed it facilitates them by removing the digression through a truth property. And so, instead of theories of truth, I have talked of theories of what-is-true.

13

See Mark Wilson's writings on these language issues, e.g., Mark Wilson 1982.

130

Deflationism Defended

Are there theories of truth? We could sensibly call all theories of what-is-true, theories of truth. But this would constitute a big break from the tradition as to what gets counted as a theory of truth since there are so many different kinds of theories of whatis-true, including the sciences. Perhaps the closest we could get to the tradition would

be to call the theories of theories of what-is-true, theories of truth.14 Logic also has a special place in the above categorization of the theories of what-istrue because it addresses the logical relations that obtain between statements that can be used to say what-is-true. This way of describing logic seems to make sense of Frege's 1918-19 claim T h e w o r d 'true' indicates the aim o f logic'... All sciences have truth as their goal; but logic is also concerned with it in a quite different w a y f r o m this. ... To discover truths is the task o f the sciences; it falls to logic to discern the laws o f t r u t h . 1 5

I agree with Frege. For, while both science and logic provide theories as to what-istrue, there is a difference: science focuses on what-is-true with respect to the physical universe, while logic focuses on (what-is-true with respect to) the logical relations between claims as to what-is-true. But what of the metalinguistic truth predicate logicians seem to be so fond of? Indeed, some logicians have been wary of the prosentential theory because of its representation of 'trues expressibility being at the level of the "object language." There are so many important results in logic that seem to rely on a metalinguistic truth predicate. Would acceptance of the prosentential account mean that we would be deprived of these results? The short answer is, 'No.' The issue is one I have addressed in a number of places (e.g., §4 of Grover 1990a, §6 Grover 1981b). Briefly, my points have been the following. In formal languages where contexts are constrained in relevantly significant ways, DS (the disquotational schema, 'x is true iff p') can be assumed to hold. This means that one can use the truth predicate for linguistic ascent. 16 (Quine 1970 has pointed out this role of the truth predicate.) Logicians can then treat the logical relations that obtain between things that are said as if they were well defined relations holding between sentences. Since logicians have an interest in facilitating a process whereby certain kinds of thinking processes are reduced to processes involving just the manipulation of symbols, the constraints on formal languages together with DS, make life easy for logicians. (Logicians in their turn try to make rational thinking easy for the rest of us.) Consideration of an analogous exercise may be helpful: Frege 1884 has drawn attention to the work that went into devising a system of numerals that made addition, multiplication, etc., rote exercises—to the extent that most of us do not even know how the algorithm gets the job done. Similarly, logicians who focus on logical issues can develop sophisticated theories of what can or

14

I have found, in responding to questions from non-philosopher friends on my research area, describing my topic as "truth" misleads. "Truth" suggests that I have some insight into the nature of the universe that I do not have! 15 This is from A.M. and M. Quinton's translation of Frege's "The Thought: A Logical Inquiry" reprinted in Essays on Frege, ed. E. D. Klemke. Urbana, Illinois: Univ.of Illinois Press 1968. 16 See my characterization of constructions like '"Snow is white' is true" as a kind of "inheritor," Grover 1977. Presentences are also inheritors.

On Locating Our Interest in Truth

131

c a n n o t be inferred, by reducing such talk to talk o f sentences a n d well-defined ways o f manipulating t h e m . N o t e , however, these symbol-pushing maneuvers w o u l d n o t be o f m u c h interest to us if we did n o t have a truth predicate to return us f r o m talk a b o u t linguistic things to talk o f worldly things-'true' provides the required linguistic descent.

5 . Our Interest in

Truth

Explaining o u r interest in truth c o m e s d o w n to explaining o u r interest in what-is-true. O u r interest in what-is-true takes us to inquiry a n d assertion; to theorizing a b o u t w h a t is-true; legitimizing o u r theories o f what-is-true; a n d theorizing a b o u t the logical relations between statements used to claim w h a t - i s - t r u e . 1 7

Bibliography Adams, E.M. 1998. "On the Possibility of a Unified World View," in The Philosophy of P. F. Strawson, ed. Lewis Edwin Hahn. Chicago and LaSalle, Illinois: Open Court. 1998, pp. 69-85. Blackburn, Simon. 1998. "Relativization and Truth," in The Philosophy ofP.F. Strawson, ed. Lewis Edwin Hahn. Chicago and LaSalle, Illinois: Open Court. 1998, pp. 151-167. Blanshard, Brand. 1939 The Nature of Thought. London: Allen & Unwin. Bonjour, Laurence. 1985. The Structure of Empirical Knowledge. Cambridge, Massachusetts: Harvard Univ. Press. Devitt, Michael. 1984. Realism and Truth. Princeton, NJ: Princeton University Press. Faust, Jennifer 1995. A Deflationary Response to the Ontological Problem: A Defense of the Coherence Theory of Justification. Ph.D. dissertation, University of Illinois at Chicago Field, Hartry 1986. "The Deflationary Conception of Truth," in Fact, Science, and Morality: Essays on A.J.Ayer's Language, Truth and Logic, ed. Graham MacDonald and Crispin Wright. Oxford: Blackwell, 1986, pp. 83-107. Frege, G. 1984. The Foundations of Arithmetic. Trans. J. L. Austin. Chicago: Northwestern Univ. Press. Frege, G. 1918-19. "The Thought: A Logical Inquiry," trans. A. M. and M. Quinton, Mind 65, 1956, pp. 289-311. Reprinted in Essays on Frege, ed. E. D. Klemke. Urbana and Chicago, Illinois: Univ. of Illinois Press 1968, pp. 507-35. Grover, Dorothy. 1977. "Inheritors and Paradox," The Journal of Philosophy, 74, pp. 590-604. Reprinted in Graver 1992. Grover, Dorothy. 1981. "Truth: Do We Need it?" Philosophical Studies 40, pp.69-103. Reprinted in Grover 1992. Grover, Dorothy. 1990a. "On Two Deflationary Truth Theories," in Truth or Consequences, ed. Michael Dunn and Anil Gupta. Dordrecht: Kluwer 1990, pp.1-17. Reprinted in Grover 1992. Grover, Dorothy. 1990b. "Truth and Language-World Connections," Journal of Philosophy 87, pp.671-87. Reprinted in Grover 1992. Grover, Dorothy. 1992. A Prosentential Theory of Truth. Princeton, New Jersey: Univ. of Princeton Press. Horwich, Paul. 1990. Truth. Oxford: Blackwell. Misak, Cheryl. 1991. Truth and the End of Inquiry: A Peircean Account of Truth, Oxford: Clarendon Press. Misak, Cheryl. 1998. "Deflating Truth: Pragmatism vs. Minimalism," The Monist, 81.no.3, pp.407-425. Pears, David. 1998. "Strawson on Freedom and Resentment," in The Philosophy ofP.F. Strawson, ed. Lewis Edwin Hahn. Chicago and LaSalle, Illinois, Open Court 1998, pp. 245-258.

17 I thank my colleagues at UIC for helpful and (as always) supportive comments on a presentation of an earlier draft of this paper.

132

Deflationism Defended

Quine, W.V. 1970. The Philosophy of Logic. New York: Harper and Row. Sellars, Wilfrid. 1963. Science, Perception and Reality. New York: Routledge & Kegan Paul Strawson, P. F. 1950. "Truth," Proceedings of the Artistotelian Society, supp. Vol. 24, pp. 129-56. Strawson, P. F. 1985. Skepticism and Naturalism: Some Varieties. New York and London: Columbia Univ. Press and Methuen. Strawson, P. F. 1998. "Reply to Ε. M. Adams," in The Philosophy of P. F. Strawson, ed. Lewis Edwin Hahn. Chicago and LaSalle, Illinois: Open Court. 1998. pp. 86-90. Williams, Michael. 1986. "Do We (Epistemologists) Need a Theory of Truth?" Philosophical Topics 4, pp.22342. Wilson, Mark. 1982. "Predicate Meets Property," Philosophical Review 91, pp. 549-89 Wilson, Kent. 1990. "Some Reflections on the Prosentential Theory of Truth," in Truth or Consequences: Essays in Honor ofNuel Belnap, ed. J. Michael Dunn & Anil Gupta. Dordrecht, The Netherlands: Kluwer Academic Publishers. 1990.

Norms of Truth and Meaning PAUL HORWICH

It is widely held that the 'normativity' of truth and meaning puts a severe constraint on acceptable theories of these concepts. This constraint is so severe, some would say, as to rule out purely naturalistic' or 'factual' accounts of them. In particular, it is commonly supposed that the deflationary view of truth and the use-regularity conception of meaning, insofar as they are articulated in entirely non-normative terms, must for that reason be defective.1 I want to oppose this point of view. I'm not going to deny that there are correct norms concerning truth and meaning. Amongst the most important of them are that true belief is valuable, and that the meaning of a word determines the things to which one ought, and ought not, to apply it. But I want to argue that these normative commitments can easily be reconciled with fully adequate conceptions of truth and meaning that are wholly wow-normative. Thus my conclusion, in a nutshell, will be that although truth and meaning do indeed have normative import, they are not intrinsically normative.2 So those of us who are attracted to the deflationary view of truth and to naturalistic analyses of meaning (in terms of 'dispositions of use', for example) have nothing to worry about, at least as far as normativity is concerned.3 1

2

3

This essay concerns the meanings of terms in thought, as well as the meanings associated with sounds and marks in public languages. Thus the discussion that follows includes the normativity of mental content. By this I mean, roughly, that although truth and meaning are governed by norms, they are not to be analysed conceptually in terms of explicitly normative notions such as 'ought', 'rational', or 'good'. A more general definition of 'intrinsic normativity' - one allowing that a notion might be unanalysible and yet still be intrinsically (but not explicitly) normative — would have it that the concept expressed by "f" is intrinsically normative if our understanding of "f" in constituted by our underived acceptance of "#f", where "#_" contains some explicitly normative term. Alternatively, and without commitment to any such account of how understanding is constituted, one might say that a concept is intrinsically normative if and only if its possession requires possession of some explicitly normative concept such as 'ought'. The latter formulation is given by Paul Boghossian in his "The Normativity of Content", Minds, Machines, and Meaning, edited by J. Campbell and L. Oliviera, Oxford University Press. None of these attempts to distinguish the normative from the non-normative is free of difficulties. They presuppose, perhaps wrongly, that all normative notions are in some way based on some specified stock of explicitly normative concepts. Moreover, even if that presupposition is correct, it is undesirable for the normativity of the fundamentally normative concepts to remain unexplained - i.e. for there to be nothing more than a list of them. However, we can articulate the central issue here without adverting to the normative/ non-normative distinction. The issue is simply whether or not deflationism about truth and the use theory of meaning can, or cannot, accommodate the value of truth and the relationship between a word's meaning and how it ought to be applied. It might be thought that proponents of non-normative theories of truth or meaning would have nothing to worry about even if these phenomena were intrinsically normative. For it might be thought that the explicitly normative ideas (i.e. 'ought') in terms of which truth and meaning would, in the first instance, be analyzed could then be reduced to non-normative ideas — leaving us, after these two stages of

134

Deflationism Defended

Many philosophers have urged the point of view that I will be opposing in this paper, but let me single out Michael Dummett and Saul Kripke as especially influential examples. Dummett has complained that the redundancy theory of truth is inadequate - too weak - on the grounds that it leaves out the value we attach to truth. And his point applies with equal force to other versions of deflationism.4 In his 1959 paper, "Truth", he compares the making of true statements to the winning of a game such as chess.5 He points out that we could set out the rules for how to move the various pieces and we could specify which positions count as winning the game - but still something vital to the concept of winning would have been left out: namely, that one tries to win. And according to Dummett the same goes for truth: the deflationary theory merely identifies the circumstances in which things are true; it tells us, for example, that the proposition that dogs bark is true if and only if dogs bark; but it leaves out the vital fact that we want our beliefs to be true; this is how they are supposed to be. 6 Another philosopher who has made much of the normative character of language is Saul Kripke. In his Wittgenstein: On Rules and Private Language Kripke emphasizes that, for example, "+" means PLUS —» one ought to apply "+" to the triple and maintains that any account of the underlying nature of this meaning property any reductive theory of the form w means PLUS = P(x)

4

5 6

analysis, with non-normative accounts of truth and meaning. However, there is no plausible conceptual (i.e. definitional) analysis of ought' in terms of'is'; and a weaker (non-conceptual) reduction could not help those, such as deflationists and use theorists, who advocate non-normative accounts of our concepts of truth and meaning. Moreover, it is extremely hard to see how any such two-stage analysis would result in the particular accounts they advocate. The redundancy theory says that "The proposition that ρ is true" means the same as simply "p", whereas certain more recent forms of deflationism about truth — including the 'minimalism' defended in my Truth, 2nd edition, (Oxford University Press, 1998) - propose to implicitly define the truth predicate by means of our commitment to the material biconditional, "The proposition that ρ is true numbers (God, possible worlds) exist" expresses a necessary truth, given the appropriate conception of numbers (God, possible worlds), one should admit that the corresponding inference is not a good one. Horwich objects to theories other than M T on the grounds that they fail to explain the axioms of M T (see, e.g., op. cit., pp. 11-12). But if an intuition of the necessity of X—>Y were sufficient to warrant the claim that X explains Y, then every theory of truth would explain the axioms of M T since they are necessary truths. O f course, meta-deducibility still depends on the specific conception of the nature of propositions remarked on above. Tarski considers a similar notion-applied to numbers rather than propositions-in his "The Concept of Logical Consequence", in Tarski, op. cit., p. 411.

174

Deflationism Attacked

explainable/deducible, in the old sense, on the basis of M T & X. It is hard to see how this goes beyond the bare assertion that the general fact is explainable because the particular facts are explainable—the very claim that was at issue to begin with. Moreover, this new notion of explanation is rather puzzling. Ordinarily, one thinks that facts about objects of kind Κ are explained by other facts about objects of kind K, L, and M. The claim that certain facts about Ks are explained by the fact that we can explain other facts about Ks suggests that the facts in question are not regarded as real facts. A minimalist may want to respond to all this: "Why be so conservative? Why not accept a non-standard notion of deducibility*?" Let us grant, for the sake of argument, that some appropriate non-standard notion(s) can be made sense of. The result must be that the Adequacy Thesis is far weaker than originally advertised. I remarked earlier that the thesis plays a crucial role in the minimalist best-explanation argument for the exclusivity of MT, where it is argued that other theories of truth ought to be rejected because they fail to provide adequate explanations of the facts about truth. For such an argument to work, there has to be a shared standard of adequate explainability. Deducibility from the proposed truth theory (plus background theories) could serve as such a standard. Given this standard, one half of the minimalist best-explanation argument seems indeed correct: other truth theories do not enable us to deduce all the facts about truth. However, neither does MT. To save the Adequacy Thesis, the minimalist trades deducibility for some extended notion of deducibility*. But now the common standard of adequate explanation is discarded and the best-explanation argument loses all force. Other truth theories will be able to provide adequate explanations of the facts about truth, provided they can select some notions of "deducibility" appropriately extended to serve their needs. I also remarked that the Adequacy Thesis, construed in terms of deducibility, would help ensure that the facts about truth are completely reducible to M T (and truth-free background theories). This would lend some support to the minimalist claim that the facts about truth do not involve a truth property more substantial than the one covered by deflationary MT. But once deducibility is traded in for some extended notion of deducibility*, the Adequacy Thesis lends no support to this claim. Why not claim, on the contrary, that truth must be a substantial property because universal generalizations about truth are not deducible from MT? Similar considerations apply with respect to the issue of admissible background theories. The Adequacy Thesis can be defended only relative to a very specific background theory about the nature of propositions; moreover, background theories that are somehow enmeshed with truth have to be reconstrued as infinite theories. Such specific and contentious commitments help save the Adequacy Thesis only by weakening it. I want to return briefly to a curious feature of Horwich's account of universal facts about truth that I have set aside earlier. At the outset, Horwich promises an explanation of (Imp), but his account ends with (7.1):

Minimalism and the Facts about Truth

175

(Imp) For all propositions x, y: if χ is true, and χ implies y, then y is true; (7.1)

For all proposition x: if χ is of the form, ([(p) is true & (p) implies (q)] —» [(q) is true]), then χ is true. 27

On the face of it, these generalizations differ in content as well as in form. This shift from one generalization to another will be a general trait of all explanations modeled on Horwich's account. When we ask the minimalist to explain, say, the generalization that a proposition is known only if it is true, the last line of his argument will offer us instead the generalization that every proposition of the form 'If (p) is known then (p) is true' is true. This raises the objection that Minimalism is unable to explain ordinary general facts about truth; instead, it offers us general form-facts about truth. It seems the minimalist will have to respond that (Imp) and (7.1) express the same fact. To put it more generally and in the material mode, he will have to maintain that ordinary general facts about truth are really form-facts about truth. This response further weakens the Adequacy Thesis. For it turns out that the thesis holds only modulo a contentious claim about fact identity. The claim would have to be backed-up with some theory of forms that applies to facts and propositions-such a theory remains to be spelled out (and it is sure to be contentious too). 28

27

It seems that (7.1) is misstated. Since '(...)' abbreviates 'the proposition that...', (7.1) comes out as "Every proposition of the form that the proposition that [(p) is true & (p) implies (q)] —> [(q) is true] is true," which does not make any sense. T h e outermost'(...)' in (7.1) should be replaced by quotes. This fits in with what Horwich says about form-talk elsewhere. Note that (dogs fly —> dogs fly) is not of the form '(p —> p)'; rather, it has the form 'ρ —> ρ'. The former is not a form of a proposition at all because it is not the form of a sentence expressing a proposition; it is the form of a noun-phrase referring to a proposition; cf. Horwich, op. cit., p. 123. 28 At times Horwich construes form-talk linguistically, so that "Every proposition of the form '...p...' is true" comes out as "Every proposition expressed by a sentence of the form '...p...' is true"; cf. Horwich, op. cit., p. 123. On this interpretation, form-facts like (7.1) would come out as disguised linguistic facts and the thesis that all ordinary general facts are form-facts would be quite untenable. Horwich may ultimately prefer a non-linguistic interpretation of form-talk, on which the thesis might be somewhat less implausible. But there is little by way of a non-linguistic theory of forms. O n pp. 17-20 he tries to construe propositional forms/structures as functions-unsuccessfully by my lights. He holds that the propositional form/structure (E*) «p> is true iff p) is a function from propositions to propositions. If so, it would have to make sense to say that, given a proposition as argument, there is a proposition which is the value of the function (E*). That is, it would have to make sense to say that, for every proposition y, there is a proposition x, such that χ = {(y) is true iff y). But this does not make sense. Take y = the proposition that snow is white; we then have χ = the proposition that the proposition that the proposition that snow is white is true iff the proposition that snow is white.

Disquotationalist Conceptions of Truth WOLFGANG K Ü N N E

'Truth is disquotation', Quine says,1 and this slogan, like all the others he has coined, bears spelling out: (QUINE 1) The truth predicate is a reminder that, despite a technical ascent to talk of sentences, our eye is on the world. This cancellatory force of the truth predicate is explicit in Tarskis paradigm: 'Snow is white' is true if and only if snow is white. Quotation marks make all the difference between talking about words and talking about snow. The quotation is a name of a sentence that contains a name, namely 'snow', of snow. By calling the sentence true, we call snow white. The truth predicate is a device of disquotation. [../..] So long as we are speaking only of the truth of singly given sentences, the perfect theory of truth is [...] the disappearance theory of truth. {PL 12; 11.)

Attaching the predicate 'is true' to the quotational designator of a (declarative) sentence has the same effect, or so we are told, as would be obtained by simply erasing the quotation marks: what is said by such a truth ascription could just as well be said by uttering the quoted sentence itself. This Redundancy Thesis is the first tenet of disquotationalism. The 'So long as' clause at the end of (Q-l) hints at the limitations of this claim. As it stands, it can at best hold of revealing truth-ascriptions, i.e. of those which display a truth candidate between quotation-marks: No disquotation without quotation (if I may venture to offer a slogan myself). Since ever so many truth-ascriptions are unrevealing, it is to be hoped that the redundancy claim isn't the whole message of disquotationalism, and indeed it isn't. But let us start with a discussion of this claim.

1

Redundancy

The disquotationalist Redundancy Thesis should be strictly distinguished from another one. Frege and Ramsey are concerned with instances of the Denominalization Schema (Den)

It is true that p, iff p,

whereas disquotationalists focus on homophonic T-equivalences, i.e. on instances of the Disquotation Schema

1

Quine (1992), 80.

Disquotationalist Conceptions of Truth

(Dis)

177

'p' is true iff p.

It is one thing to maintain that what is said remains unaffected whether we attach '(it) is true' to a that-clause or whether we simply erase the complementizer 'that': (IDENTITY 1)

[It is true that p] = [p].

(You can read the square brackets as 'The proposition that' or as 'What is said by " "'. The Denominalization Schema is weaker than the identity schema.) It is another thing to contend that what is said remains unaffected whether we append the truth-predicate to the quotational designator of a (declarative) sentence or whether we simply delete the quotation marks: (IDENTITY 2)

['p' is true] = [p\.

The Disquotation Schema, too, is weaker than the corresponding identity schema: if you accept an instance of (IDENTITY 2), you are committed to endorsing the corresponding instance of (Dis), but there is no such obligation in the other direction. I am ready to dispute the former as well but in this paper I am only concerned with the latter. (Most arguments which can be used here would be entirely inappropriate there.) Not only disquotationalists subscribe to the second identity thesis. Wittgenstein, Carnap, and McDowell, for example, would not like to be classified as partisans of disquotationalism. Nor would Quine (as we shall soon see), although disquotationalism owes more than its name to him. Nevertheless, occasionally at least they all assent to (IDENTITY 2), even if most of them might not like to put it in terms of propositions:

2

(WITTGENSTEIN 1)

'/>' ist wahr, sagt nichts Anderes aus als p! 'p' is true, says nothing else but p! (Notebooks, 9.)

(WITTGENSTEIN 2)

Was heißt denn, ein Satz 'ist wahr? 'p ist wahr = p. (Dies ist die Antwort.) What does it mean, a sentence 'is true'? 'p' is true = p. (This is the answer.) {Bemerkungen über die Grundlagen der Mathematik, 117. 2 )

Wittgensteins use of'=' in (W-2) is ungrammatical, and his use of quotation-marks tends to be sloppy. In spite of the two passages from his early and late work (to which Philosophical Grammar, 123 and Philosophical Investigations, § 136 could be added), it is by no means clear that he really favours (IDENTITY 2) rather than (IDENTITY 1). In one of his unpublished manuscripts and in his Philosophical Grammar he seems to find fault with (IDENTITY 2): 'The proposition (Satz) 'It is raining' surely says something about the weather but nothing about the words I am speaking. But how can 'It is raining' say the same as: 'The proposition (Satz) "It is raining" is true', since it does say something about the words?' (ms. quoted after Hallett (1977), 237). Cp. Philosophical Grammar, 123-124.

178

Deflationism Attacked

(CARNAP)

To assert that a sentence is true means the same as to assert the sentence itself; e-g· [···] "The moon is round' is true" and 'The moon is round' are merely two different formulations of the same assertion. (IS 26. 3 )

(McDOWELL) Appending a truth predicate to a designation of a sentence produces a sentence apt [...] for saying [...] the very thing [...] that could have been said by using the original sentence. ('Truth-Conditions, Bivalence and Verificationism', 7.) Is this identity claim plausible? (In what follows I shall pretend that all truth-candidates are free of ambiguous and indexical elements. 4 ) Perhaps the most famous objection against the first tenet of disquotationalism is Dummetts Argument From TruthConditional Semantics: (DUMMETT) [I]f the whole explanation of the sense of the word 'true', as applied, e. g., to the sentence 'Frege died in 1925', consisted in saying that '"Frege died in 1925" is true' is equivalent to 'Frege died in 1925', then my understanding of the sentence 'Frege died in 1925' could not in turn consist in my knowing what has to be the case for the sentence to be true. Given that I knew what it meant to apply the predicate 'true' to that sentence, such knowledge would reduce to knowledge of a mere tautology in the most literal sense: if all that it means to say that 'Frege died in 1925' is true is that Frege died in 1925, then the knowledge that 'Frege died in 1925' is true just in case Frege died in 1925 is simply the 'knowledge' that Frege died in 1925 just in case Frege died in 1925. {FPL 458. 5 ) The Argument From Truth-Conditional Semantics is a conditional refutation of the first tenet of disquotationalism: if bur understanding of the (declarative) sentences of a language consists in our knowledge of their truth-conditions, then what these conditions are conditions of is not identical with disquotational truth. Dummett asserts the antecedent and applies modus ponens. Disquotationalists deny the consequent and apply modus tollens. Hartry Field who has become the most resourceful and vociferous advocate of disquotationalism, writes: (FIELD 1)

Accepting [disquotationalism] requires dethroning truth conditions from the central place in the theory of meaning [...] that Frege [...] and many others have given them. My current view is that this is probably a good thing. ('Disquotational Truth' [...] , 408 f.)

It is noteworthy that Quine does not share this view: 'First and last, in learning language, we are learning how to distribute truth values. I am with Davidson here; we are

3

4 5

Carnaps first remark is confusing. First he takes what is asserted to be a proposition (albeit one about a sentence), then he takes it to be a sentence. But the second claim makes only dubious sense (as was pointed out by Moore in 1942: see his (1962), 229): By uttering certain words one can assert something, but in so doing one does not assert those words. Disquotationalist strategies for handling ambiguity and indexicality are sketched in Field (1994b), 278281, (1994a), 405 n„ 428-430. Cp. also Dummett (1959), 7.

Disquotationalist Conceptions of Truth

179

learning truth conditions'. So in spite of Quine's midwifery at the birth of disquotationalism, charity forbids to count him among its supporters.6 Is the first tenet of disquotationalism compatible with the contention that one can use a sentence like (A) 'Snow is white' is true iff snow is white at least as a partial explanation of the meaning of 'true' as applied to the quoted sentence? The Argument From Explanatory Loss seems to show that the redundancy claim, if true, nips all such explanatory aspirations in the bud. According to disquotationalism, the left branch of (A) has the same meaning as the right branch, hence (A) has the same (conventional linguistic) meaning as the tautology (B) Snow is white iff snow is white. (Within an environment like (A), supplanting an embedded sentence by another sentence with the same meaning leaves the meaning of the whole intact.) But then telling somebody that (A) can no more amount to giving a partial explanation of 'true' (or of anything else) than would telling him that (B). This worry is just another illustration of a general problem that has surfaced in the debate about the Paradox of Analysis.7 Compare the following pair of sentences: (a) Donald is a drake iff Donald is a male duck (b) Donald is a male duck iff Donald is a male duck. Since 'drake' and 'male duck' have the same meaning, (b) has the same meaning as (a). Nevertheless, (a) can, whereas (b) cannot, be used to explain the meaning of 'drake'. The fact that the predicates in (a) are related as analysandum and analysans is quite accidental to the problem. What really matters is that only one of the biconditionals can play an explanatory role, as the next example shows: (a*) Kaa is a serpent iff Kaa is a snake (b*) Kaa is a snake iff Kaa is a snake. As the two predicates have the same meaning, so do (b*) and (a*). But once again, (a*) can, whereas (b*) cannot, be used to explain the meaning of 'serpent'.8 So the disquotationalist can disarm the objection along the same lines: Why should (A) not be used to explain the meaning of 'true' as applied to the quoted sentence even though (B) is entirely useless for this explanatory purpose? That two sentences have the same meaning does not exclude that only one of them has an explanatory potential. A familiar objection against the first tenet of disquotationalism is the Argument From Modal Difference. Let us embed the sentences

6

The quotation is from Quine ( 1 9 7 3 ) , 65; cp. also his ( 1 9 8 1 ) , 3 8 . In my reading of Quine I follow Davidson ( 1 9 9 4 ) .

7 8

Cp. Carnap (1947), 63. Pairs like (a*/b*) show that Carnap's attempt to solve the problem with the help o f his notion of intensional isomorphism doesn't work (loc. cit.). Cp. my ( 1 9 8 3 ) , 2 1 9 f.

180

Deflationism Attacked

(S) Snow is white (T) 'Snow is white' is true in a modal context, as consequents in a subjunctive conditional: (MS)

If we all came to use the sentence 'snow is white' for saying that snow is black, it would not be the case that snow is white.

(MT) If we all came to use the sentence 'snow is white' for saying that snow is black, it would not be the case that'snow is white is true. Obviously (MT) is true, whereas (MS) isn't. The colour of snow does not depend on the way the word 'white' is used. After all, one cannot change the colour of snow simply by using the word 'white' for the colour of coal. Now the application of the modal operator 'it would not be the case that' to (S) and (T) could not enforce the assignment of different truth-values to (MS) and (MT), if an ascription of truth to 'Snow is white' were really nothing but an ascription of whiteness to snow. Hence, the objector concludes, the first tenet of disquotationalism is false. A disquotationalist can shield off this attack by formulating his homophonic Tequivalences in a more circumspect way:9 (Dis*) 'p' is true as used by me now iff p. This reformulation is independently motivated, as the following observation by Davidson (who does not intend to help disquotationalists) shows: (DAVIDSON) T h e instances of the disquotational schema are guaranteed to be true, in fact, only in the very special case where the quoted sentences are guaranteed to have the same truth-values as those same sentences shorn o f quotation marks on the right o f the biconditional. This guarantee is lacking, for example, when I surmise that your sentence 'Snow is white' is true if and only if snow is white. ('What is Quines View of Truth?', 4 3 9 . )

If we apply (Dis*) to the consequent of (MT) we obtain: (?)

If we all came to use the sentence 'snow is white' for saying that snow is black, it would not be the case that 'snow is white' is true as used by me now.

This is clearly false. So the intended reading of (MT) must be rather this: (MT*) If we all came to use the sentence 'snow is white' for saying that snow is black, it would not be the case that 'snow is white' is true as used by me then.

9

The following attempt at answering the Modal Objection is due to McGee (1993), 92-93. Cp. also David (1994), 130-135; Field (1994a), 408, (1994b), 277 f.

Disquotationalist Conceptions of Truth

181

Why shouldn't disquotationalists assent to (MT*), and hence to (MT), without any reservation? Now consider (MS). According to advocates of (Dis*) this amounts to the same as (?), which is plainly false, hence disquotationalists can deny (MS) as firmly as anyone else. Perhaps the following Argument From Entailment is a more powerful objection against the disquotationalist Redundancy Thesis. 10 It runs as follows: There is a reading of entails' under which (T*)

'Snow is white' as used by me now is true

entails both (M) and (N): (M) (N)

'Snow is white' as used by me now is meaningful Something is sometimes used by someone.

But (S), as used by me now, entails neither (M) nor (N). Therefore (S), as used by me now, does not have the same content as my utterance of (T*). 11 The conception of entailment appealed to in this objection is such that two sentences which are necessarily equivalent may nevertheless differ as to what they entail, for in our reply to the Argument From Modal Difference we have assumed that the necessitation of'(T*) iff (S)' is true. Now I do not doubt that we have a conception of entailment which allows us to say, for example, that although '2 is prime' and 'All drakes are ducks' are necessarily equivalent, only the latter sentence entails 'If Donald is a drake then Donald is a duck'. But since it is none too clear that our pre-formal conception of entailment can be explained in such a way that all counter-intuitive consequences are avoided, it would be unwise to rest one's criticism of the disquotationalist redundancy claim entirely on this Argument From Entailment. The most powerful, indeed lethal challenge to the first tenet of disquotationalism is the Argument From Doxastic Difference. In spite of its somewhat pompous name it is exceedingly simple. If 'Snow is white' (S) and "'Snow is white" is true' (T) were to express the same proposition, then nobody could believe what either of these sentences expresses without eo ipso believing what the other expresses. But somebody who does not know what 'Snow is white' means may believe that snow is white without believing that 'Snow is white' is true, or he may believe that 'Snow is white' is true without believing that snow is white. Hence (S) and (T) do not express the same proposition. 12 Nor does (S) express the same proposition as (T*)

'Snow is white' is true as used by me now.

Somebody who does not understand 'Snow is white' at all may believe that snow is white without believing of any person X and time t that 'Snow is white' is true as used by X at t.

10 11

It is sketched in passing, and with very little sympathy, in Quine (1953b), 163-164. It is noteworthy that the Fregean Redundancy Thesis cannot reasonably be attacked along similar lines. It would be a petitio against advocates of that position to argue: 'That snow is white is true' entails, whereas (S) does not entail, that something is true.

182

Deflationism Attacked

The following argument drives the same point home. Consider the disquotationalist Redundancy Thesis in the light of the Conceptual Balance Requirement: Two utterances express the same proposition only if there is no concept whose mastery has to be exercised only in understanding one of them. Surely one cannot understand an utterance of the English sentence (S) without exercising one's mastery of the concept of snow. But imagine a Bedouin whose rudimentary English comprises only vocabulary which is useful in the desert: he might very well understand the truth ascription (T) without even having that concept. In order to understand a quotational designator one does not have to understand the quoted expression. Otherwise the following verdict would be comprehensible only if it is false: 'The slithy toves did gyre' is incomprehensible. So (S) and (T) do not express the same proposition. This time an explicit relativization of the truth predicate will make the identity claim even less defensible, for it will cause trouble in both directions: Mastery of the concept of use is required for understanding (T*), but not for understanding (S). Like its predecessor this Argument From Conceptual Overloading shows that the disquotationalist Redundancy Thesis cannot be upheld. Let me briefly go into the question whether Tarski would have considered a T-equivalence like (A) 'Snow is white' is true iff snow is white to be 'the whole explanation' of the sense of the word 'true', as applied to the quoted sentence. According to Quine and Dummett, the answer is No. 'It is sometimes overlooked', Quine says, 'that there is no need to claim, and that Tarski has not claimed, that [homophonic T-equivalences] are analytic'. 13 Dummett concurs: 'Tarski was concerned to claim no more than material equivalence, i.e. identity of truth-value' for the two sides of a T-equivalence.14 Criterion Τ lends no support to the disquotationalists' Redundancy Thesis, for it demands only that 'p' translates S, not that it translates 'S is true'. But at other places Tarski is concerned to claim far more than material equivalence. He calls T-equivalences 'partial definitions (Teildefinitionen)' •} ^ they are only partial because they do not tell us what being true comes to in all cases (i. e. for all sentences of the given language), and they are definitions because they tell us completely what being true comes to in the case at hand (i. e. for the quoted sentence). Tarski explicitly declares the right branch of a T-equivalence to be an explanation of the meaning of its left branch: 16

12 Notice that neither disjunct would be convincing if we were to replace (T) by 'It is true that snow is white'. The first disjunct would be a petitio against advocates of the Fregean Redundancy Thesis, the second disjunct would be plainly wrong. 13 Quine (1953a), 137 n. 9 (cp. also his (1953b), 164). 14 Dummett, (1978), p. xx. 15 Tarski (1936a), 8 - 1 0 (155-7) on schema (2) and sentences (3) and (4); 98 (236), 1 1 6 (253), 129 (264); (1936b), 264 (404); (1944), § 4; (1969), 413. 16 The statement (T-l) is repeated almost verbatim in Tarski (1936b), 264 (404).

Disquotationalist Conceptions of Truth

183

(TARSKI 1) [T-equivalences] explain in a precise way, in accordance with linguistic usage (Sprachgebrauch), the meaning (Bedeutung) of the phrases χ is a true sentence' which occur in them. [...] ( W B 45 (187).)

So even if there were no need for Tarski to claim more than identity of truth-value for the two branches of a T-equivalence, as matter of fact he did claim more than that, and what he claimed at this point is refuted by the Argument From Doxastic Difference.17 But is there really no need for Tarski to require anything stronger than material equivalence? As we saw, Quine absolves him from any obligation to maintain even the necessitation of (A), (Nec) Necessarily, 'snow is white' is true iff snow is white. As can be seen from his classic paper 'Two Dogmas of Empiricism',18 Quine took 'Necessarily p' to come to the same (obscure) thing as '"p" is analytic'. 19 Now the truth of (Nec) is a necessary (but not a sufficient) condition for the truth of the corresponding identity claim. (Although, necessarily, ABC is an equilateral Euclidean triangle iff ABC is an equiangular Euclidean triangle, it is not the case that anyone who was to say the former / the latter would thereby be saying the latter / the former.) So, when speaking on Tarskis behalf, Quine denies any need to embrace (Nec), but, when speaking in propria persona he upholds a thesis that is stronger than (Nec).20 (There is nothing contradictory about this in a philosopher who declares talk of analyticity / necessity to be at bottom incomprehensible.) But let us return to our question: is Tarski really not committed to accept (Nec)? I think he is. Criterion Τ demands that the pertinent Tequivalences follow from the definition of 'true' for L (plus some non-contingent syntactical truths like '"snow is white" is not identical with "blood is red'"). The definition itself is not a contingent truth, for it is constitutive of L that its sentences mean what they do mean. Now a conceptual truth A cannot entail a contingent truth B; for otherwise, by contraposition, the negation of B, which is as contingent as B, would entail the negation of A, which is conceptually false. And this is absurd, since whatever entails a conceptual falsehood is itself conceptually false.21 Hence Tarski ought to be ready to accept (Nec). Disquotationalists should weaken their first tenet, but they can make a stronger claim than that of necessary equivalence.22 They can plausibly maintain that an ascription of truth to a sentence in a certain context is cognitively equivalent with the sentence itself as used in that context. Take (T*) (S)

17 18 19

20 21 22

'Snow is white' as used by me now is true Snow is white

This was pointed out by Soames (1999), 240. Written at about the same time as the two papers referred to above and likewise included in his ( 1 9 6 1 ) . Maybe equally obscure, but certainly not the same thing (as Kaplan has taught us): every utterance of Ί am here now' is true solely in virtue of the meaning of this sentence (as friends of analyticity would put it), but no utterance of'Necessarily, I am here now' is true. See his (1960), 24; (1970), 12; 11. A variant of his argument is used, in a different context, by Lewy (1976), 49. As we shall soon see, Field is a disquotationalist who does weaken it in a similar way.

184

Deflationism Attacked

with respect to a certain speaker at a certain time. Nobody who understands both sentences (who grasps the linguistic meanings they have in her mouth at that time) can take one of them to express a truth (falsehood) without immediately being ready to take the other to express a truth (falsehood) as well. That's why for most, if not all, communicative purposes of those who understand the sentence to which truth is ascribed, the plain sentence itself will do just as well as the truth ascription. Even after this weakening of the first tenet one might have the impression that disquotationalists cannot see much point in using a truth predicate. But this impression is deceptive.

2 Truth for the sake of brevity When it comes to stressing the importance of the truth predicate, advocates of disquotationalism again take a leaf out of Quine's book: (QUINE 2) The truth predicate proves invaluable when we want to generalize along a dimension that cannot be swept out by a general term. The easy sort of generalization is illustrated by generalization on the term 'Socrates' in 'Socrates is mortal'; the sentence generalizes to 'All men are mortal'. The general term 'man' has served to sweep out the desired dimension of generality. The harder sort of generalization is illustrated by generalization on the clause 'time flies' in 'If time flies then time flies'. W e want to say that this compound continues true when the clause is supplanted by any other; and we can do no better than to say just that in so many words, including the word 'true'. We say "All sentences of the form 'If ρ then p' are true." W e could not generalize as in 'All men are mortal', because 'time flies' is not, like 'Socrates', a name of one of a range of objects (men) over which to generalize. W e cleared this obstacle by semantic ascent: by ascending to a level where there were indeed objects over which to generalize, namely linguistic objects, sentences. ( P T 80 f.)

Disquotationalists are prone to add here a further observation. Sometimes we want to voice our acceptance or rejection of what somebody said when the speaker's words are unavailable for quotation. In such situations, too, the truth predicate proves invaluable. Here are two examples: 'We shall never know how she answered his question, but her answer was certainly true, for she knew the answer, and she would never have lied to him.' 'What he said to our pursuers cannot have been true, for otherwise they would have found us.' (One may have doubts whether such statements are really about the speakers words, but let that pass.) Disquotationalists are prone to say that the raison d'etre of a truth predicate resides entirely in its utility for the kinds of generalization and of 'blind' acceptance or rejection just described. This is an exaggeration even if disquotationalists are right in pushing truth-conditions from their throne in the theory of meaning. In his description of the performative potential of 'true' Strawson pointed out various other purposes which are served by this word. How do disquotationalists account for unrevealing truth ascriptions of the compendious, and the indirect, kind whose availability they esteem so highly? They maintain that an unrevealing general truth ascription like

Disquotationalist Conceptions of Truth

185

Every English sentence of the form 'If ρ then p' is true abbreviates (or encodes') an infinite If time flies then time flies, if duty calls then duty calls, if lions roar then lions roar,

conjunction and and and ...,

which contains all English sentences of the form 'if p, then p'as its conjuncts. (As Quine said, 'if we want to affirm some infinite lot of sentences [...], then the truth predicate has its use.' 2 3 ) Similarly, disquotationalists contend, an unrevealing singular truth ascription like 'The last sentence of Goethe's Faust is true' or (U)

Alfred's favourite English sentence is true

abbreviates an infinite disjunction. Here is a tiny fragment of this disjunction, using 'a' as short for 'Alfred's favourite English sentence': (V)

(a is 'blood is red', (a is 'coal is black', (a is 'snow is white',

and blood is red), and coal is black), and snow is white),

or or or ...

From here it is only a short step to the second tenet of disquotationalism: 's is true in L' abbreviates an infinite disjunction of conjunctions. 24 Let us try to capture this by (Df. DisT)

Vs (s is true in L if and only if (s is 'pj', and pj), or (s is 'p 2 ', and p 2 ), or (s is 'p 3 ', and p 3 ), or ...,

where 'pj', 'p 2 ' , 'p 3 ', etc. are (almost 2 5 ) all and only declarative sentences of English if L is English. As a handy notational variant of (DisT), we could use (Of. DisT)

Vs (s is true in L iff {Ξ p}{s is 'p', and p}),

where the substitution-class of 'p' is taken to comprise (almost) all and only declarative sentences of L, and the 'existential' substitutional quantification '{3 p}{...p...}' is understood as tantamount to the disjunction of all the substitution-instances of the open sentence after the quantifier. 26 Tarski endorses something like (DisT) for codes, i.e. for languages that contain only a finite number of sentences that can be enumerated: truth in a code is defined by

23 24

25

26

Quine (1970), 12. The idea, hinted at by Quine, was first put into relief in Leeds (1978), 1 2 1 , and a few years later it was developed in Soames (1984), 4 1 2 - 4 1 4 , Because of the Paradox of the Liar, exceptions must be made for certain sentences (or utterances) containing 'true' (if classical logic and its semantical principles are to remain unmodified). Hence, like all other conceptions of truth, disquotationalism must be supplemented by an answer to the question how the exceptions are to be circumscribed (or how classical logic is to be modified). The question which reaction to the semantic paradoxes is most attractive to disquotationalism is canvassed in Field, 'Postscript', § 2. Field (1986), 58 f.; David (1994), 9 8 - 1 0 0 . Assuming, plausibly enough, that for every sentence there is exactly one quotational designator, the right-hand side of (DisT) is logically equivalent with '{Vp} {(s =

V)

p}'·

186

Deflationist!! Attacked

means of a disjunction of conjunctions. For languages that are not codes Tarski mobilizes the technique of recursion (and in this respect Quine always remained faithful to Tarski). When the language for which a truth predicate is to be defined is sufficiently complex to demand the detour through satisfaction, Tarski's method still turns on disquotation (provided that the object-language is part of the metalanguage), but what gets disquoted are singular terms, open sentences and logical operators which, though finite in number, suffice to form all sentences of the language. But the recursive machinery works only if the language obeys certain tight syntactical and semantical constraints. Unlike Tarski, disquotationalists do not presuppose that the language for which 'true' is to be defined has a specific kind of syntactic structure, and they do not have to worry about the apparent non-extensionality of many constructions in natural lan27

guages. When it comes to determining their relation to Tarski's work, disquotationalists are likely to remind us that Tarski says even of a definition of a truth-predicate for a language with infinitely many sentences that it is, in a sense, tantamount to a conjunction of infinitely many T-equivalences. The continuation of our last extract is pertinent here: (TARSKI 2 ) [...] Not much more in principle is to be demanded of a general definition of true sentence [sc. for the language under investigation] than that it should satisfy the usual conditions of methodological correctness and include all partial definitions of this type [sc. T-equivalences] as special cases; that it should be, so to speak, their logical product. (WB 45 (187).) Given a suitable infinitary logic allowing conjunctions and disjunctions to be of infinite length, 28 we can derive T-equivalences from (DisT): If we replace's' by 'Snow is white', there will be just one true disjunct on the right-hand side. After eliminating '('snow is white' is 'blood is red', and blood is red)' and all the other false disjuncts, we are left with this: 'Snow is white' is true in English iff ('snow is white' is 'snow is white', and snow is white). We then drop the 'tautologous' conjunct and obtain our snow-bound triviality. So (DisT) complies with Criterion T. If L is not English, friends of (DisT) must be ready to put up with language-mixtures like's is "Schnee ist weiß', and Schnee ist weiß ?ΐ> And there will be much more of this they have to face. Suppose we replace (U) by 'Alfred's favourite sentence is true (i. e. 'There is a language L such that Alfred's favourite sentence is a true sentence of L'). Then the disjunction has to contain odd-looking, and odd-sounding, clauses like

27 28

29

Cp. Field (1994b), 2 6 9 . In the late fifties Tarski himself published pioneering work in this field. For a more precise rendering of the derivation of T-equivalences from (DisT) with the help of an infinitary logic cp. David (1994), 1 0 0 104. In presentations of disquotationalism this need to mix languages is often swept under the carpet: thus Blackburn & Simmons (1999), 1 2 f. assume as a matter of course that their speaker Claire speaks English.

Disquotationalist Conceptions of Truth

187

'Alfred's favourite sentence is "Sniegjest bialy", and snieg jest bialy, or it is "Schnee ist weißand Schnee ist weiß, or it is "La neve έ bianco", and la neve e bianca, or ...'. Such motley sentences would have to be taken as true in a hybrid language which results from pooling English, Polish, German, Italian,... 30 and which is spoken at best by a very few polyglots. But I cannot see that registering this linguistic oddity by itself amounts to making a principled objection against (DisT). 31 But as it stands, (DisT) falls victim to another Argument From Conceptual Overbading. Consider (U) again. If (U) abbreviates something which contains (V), then the sense of (V) is a component of the sense of (U), and necessary conditions for understanding (V) are also necessary conditions for understanding (U). In order to understand the disjunction (V) you must understand 'Blood is red', 'Coal is black', and 'Snow is white', for these sentences appear in (V) not only between quotation-marks. So you cannot understand (V) without exercising your mastery of the concepts of blood, of coal, etc.. But you do not have to understand any of the sentences mentioned and used in (V) in order to understand the truth-ascription (U, hence grasping the concepts of blood etc. is not involved in grasping the content of (U). And this is just a tiny portion of the conceptual load which is to be carried when it comes to understanding the whole disjunction which (U) is supposed to abbreviate. Surely understanding that is no small feat. Is there anyone who understands all English sentences? And even if there were such a person, her spectacular conceptual competence is certainly not required for understanding a humble truth ascription like (U). So application of our Conceptual Balance Requirement shows that 'true' is not a device for abbreviating infinite disjunctions in the sense of (DisT). 32 Once again a weakening of the disquotationlists' claim may seem commendable: instead of maintaining that (U) abbreviates an infinite disjunction of which (V) is a fragment, they should contend at most that (U) and its infinitely long disjunctive counterpart are cognitively equivalent. But is this weaker contention really correct? Consider the following existential quantification and (what could with some patience be turned into) the alphabetically ordered disjunction of all its instances: (P) At least one Oxford College is a graduate institution (Q) Either All Souls is an Oxford College that is a graduate institution, or Balliol is, ..., or Wolfson is, or Worcester is. Somebody who understands both sentences might take (P) to express a truth without being ready to take (Q) to express a truth as well: He might suspect that (P) is true because an Oxford College which is not mentioned in (Q) is a graduate institution. Now the same applies, mutatis mutandis, to (U) and its infinitely long disjunctive coun-

30 31

Cp. Quine (1953a), 135. As to the actual use of hybrid languages in certain circles, let me mention in passing that in philosophical seminars in Germany you will nowadays occasionally hear sentences which can be assigned a truth-value only in the hybrid language German-English (often referred to as 'Denglisch'). 32 The final move in my application of my (Bolzano-inspired) argument from conceptual overloading is identical with Anil Guptas 'objection from ideology': see his (1993), 69-71.

188

Deflationism Attacked

terpart: Even if there were somebody who understood that disjunction she might not realize that it comprises all English sentences, and then she might very well suspect that (U) owes its truth to a sentence omitted in the disjunction. 33

3 Truth in my present idiolect Let us now scrutinize the version of disquotationalism which is favoured by Hartry Field. 34 (It was foreshadowed in the reply to the Argument From Modal Difference given above.) Field replaces (DisT) by two concepts of disquotational truth, a primary notion and a derivative one, both relativized to a speaker and a time. (The relativization to time is not officially taken into account, but it is clearly implied.) Here is his exposition of what he takes to be the primary notion of truth: (FIELD 2)

[I] η its primary ('purely disquotational') use, (1) 'true' as understood by a given person applies only to utterances that that person understands, and (2) for any utterance u that a person X understands, the claim that u is true is cognitively equivalent for X to u itself. [...] The intelligibility of such a disquotational notion of truth should not be in doubt; you could think of it as an indexed concept, meaning in effect 'true on my understanding of the terms involved' [...]. The deflationist allows that there may be certain extensions of the purely disquotational truth predicate that don't have features (1) and (2); but he requires that any other truth predicate be explainable in terms of the purely disquotational one, using fairly limited additional resources. ('Disquotational Truth [...]', 405.)

By clause (2) Field does not commit himself to identity claims like 'What (S) expresses in X's present idiolect is the same as what is expressed by (T) in X's present idiolect'. He takes 'cognitive equivalence' to be a matter of inferential role: 'for one sentence to be cognitively equivalent to another for a given person is for that person's inferential rules to license [...] fairly directly the inference from either one to the other'. 35 Presumably my inferential rules license a very direct inference from 'Snow is white, and two is larger than one' to (S), et vice versa, but since the conjunction is conceptually more demanding than either of its conjuncts, it is reasonable to deny that they have the same propositional content either in English or in my present idiolect. So cognitive equivalence ä la Field no more guarantees identity of content than does our (Fregean) condition of the same name. In any case, the arguments from Doxastic Difference and from Conceptual Overloading do not apply to Field's view, for a thinker who does not know what (S) means (e. g. because she has not mastered the concept of snow) cannot believe that (S) is true, if the only notion of truth currently available to

33 34 35

Cp. the analogous reflections on infinite conjunctions and universal generalizations in Gupta, Ά Critique of Deflationism', 63, 77 and 80, notes 21 and 31. Cp. Field (1986), 58, (1994b), 250, 265-267; David (1994), 135-148. Michael Resnik calls this 'the immanent approach to truth': cp. his (1990), 414. Field, η. 1 to (F-2). Cp. his (1994b), n. 2.

Disquotationalist Conceptions of Truth

189

her is the one she could now express by 'true on my understanding of that sentence'. I shall call this notion of truth, which Field takes to be primary, Idiolectic Disquotational Truth. Using the substitutional quantification format we can codify it as follows: (Of. IdDisT)

Vs (s is true in my present idiolect iff {3 p } m p i {s is identical with 'p\ and p}).

If this is said by myself, the substitution-class of 'p' is to comprise (almost) all and only those declarative sentences which belong to m[y] ρ [resent] i[diolect]. (As before, the substitutional quantification '{Ξ p}{...p...}' is supposed to abbreviate the disjunction of all substitution-instances o f ' . . . p . . . ' . ) The right-hand side of the biconditional is guaranteed to mention only sentences I now understand. This multitude happens to contain elements from various 'national' languages, most of them from German, many of them from English, and a few from some other European languages: so there will again be plenty of motley clauses in the disjunction. Never mind! What we should mind is something else. The concept of idiolectic disquotational truth, or rather each of the numerous concepts of idiolectic disquotational truth, differs widely from our everyday concept of truth. Here are two respects in which they are unlike. (i) It is extremely improbable that any of my readers would express by 'true in my present idiolect' a concept that has the same extension as the concept I now express by this locution. Furthermore, the latter concept also differs from the one I expressed by it, say, three years ago (when I had no idea what 'Snieg jest bialy' means). Such concepts are not expressed in ordinary truth talk. Here are three pieces of evidence. Firstly, suppose Sandra comments on a newspaper report by saying, 'That's true', and Alec retorts, 'No, it isn't true, I am afraid'. Obviously Alec takes himself to be contradicting Sandra, but then the concept which he refuses to apply to the report in question must be the same as the one Sandra applied to it. (The information that she speaks English and Russian fluently whereas he is a monolingual Englishman doesn't disconfirm this sameness claim in the least.) Secondly, suppose he says, 'If that report is true, the government is in trouble', then she hastens to assure him, 'It is true', and finally he concludes, 'So the government is in trouble'. This very much looks like a modusponens argument (delivered by two speakers), but the argument is formally valid only if the truth predicate in his utterance of the first premiss expresses the same concept as the truth predicate in her utterance of the second premiss. (The above information about the difference between our speakers' linguistic abilities does not provide us with a reason for condemning the argument as exemplifying the fallacy of equivocation.) Thirdly, suppose several years later she confesses, 'At that time I thought that the report was true, but now I no longer think so'. According to her confession, the very concept she once took to apply to that report does not really apply to it. (This identity claim is not refuted by the information that she has learnt a third language in the meantime.) (ii) In the following episode from Alessandro Manzoni's novel Ipromessi sposi another important aspect of our everyday use of 'true' becomes conspicuous. The simpleminded old sacristan was shocked to see Father Cristofero late at night with two women in the church, but a few words set him at ease again: although he did not understand a word of Latin he took Father Cristofero to have spoken the truth when he said to

190

Deflationism Attacked

him, 'Omnia munda mundis, since he had uttered these 'solemn, mysterious words with great determination (By taking these words to express a truth the old sacristan did not acquire the belief that to the pure all things are pure, although this is what those words mean.) It is a constitutive feature of our concept of truth that we can suppose, and even believe, that something true is being said in an utterance which we do not understand. As long as children can think of what is said as true only when they understand the utterance, they have not yet fully grasped our concept of truth. Field tries to make sense of the practice described in (ii) by extending the notion of disquotational truth: 3 7 (FIELD 3)

[A disquotationalist] can say that what we are doing [when we conjecture whether some utterance we don't understand is true] is conjecturing whether a good translation of the utterance will map it into a disquotationally true sentence we do understand. ('Disquotational Truth' [...], 408.)

Whether this explanation of a non-idiolectic truth predicate in terms of the idiolectic one uses only 'fairly limited additional resources', as was required in (F-2), depends on the notion of a good translation which is invoked. So Field adds that it 'should be taken to be a highly context-sensitive and interest-relative notion' (loc. cit.). One may wonder whether it is possible to explain without recourse to the notion of truth-value preservation what the goodness of a good translation consists in. But let us subdue all nagging doubts. Field's derivative concept of disquotational truth, I shall call it Translational Disquotational Truth, can be captured by the following universally quantified biconditional: (TrDisT)

V L Vs (s is true in L iff

3 χ (x is a good translation of s into my present idiolect) & Vx (x is a good translation of s into m.p.i —» χ true in m.p.i.). As this is conceptually dependent on the notion of idiolectic disquotational truth, it does not reduce the distance between disquotational truth and truth which was described under (i) above (and it is not meant to do so). Let us return briefly into the Lombardian church. Suppose the old sacristan's anti-clerically minded wife, who doesn't understand a word of Latin either, was also witness of the midnight event, but she took the priest to have swindled. When the trustful sacristan comments on Father Cristofero's utterance, 'That's true', his distrustful wife retorts, 'No, it is not true'. Now this very much looks as if she contradicted her husband, but Translational Disquotational Truth does not save the appearance, for the matrix on the right-hand side of (TrDisT) expresses different concepts in our quarrelling speakers' mouths. There is a further respect in which the derivative notion of disquotational truth fails to match the real thing. You could very well speculate that the hypothesis for which

36 37

Manzoni, The Betrothed, chpt. 8 (quoting from St. Paul, Titus 1,15). Cp. Field (1986), 6 1 , (1994b), 273; David (1994), 1 7 7 - 1 8 6 ; McGee (1993), 99.

Disquotationalist Conceptions o f T r u t h

191

a scientist will receive the first Nobel Prize for physics in the 25th century is true, even if you are convinced that this hypothesis cannot be translated from that physicists language into any sentence you now understand. Our everyday concept of truth allows you to reckon with the possibility that there is at least one true sentence which cannot be translated into a sentence you now understand. 38 Let us consider the following re-

marks by Field in the light of this Argument From Untranslatability Into My Present Idiolect: (FIELD 4)

Doesn't this show that the average person is clearly not using the word 'true' in its [...] disquotational sense? And doesn't that in turn show that a version of deflationism that puts [...] disquotational truth at the centre of things [...] is gratuitously departing from common sense? I don't think so [...]. [Perhaps] ordinary speakers are committed to a notion of truth that goes beyond the [...] disquotational. But if we can lessen those commitments in a way that is adequate to all practical and theoretical purposes [...], then the charge that we are 'gratuitously departing from common sense' is quite unfounded. ('Deflationist Views [...]', 2 7 7 f.)

This reply will not satisfy any philosopher who aims at elucidating our workaday concept of truth. He does not want to lessen the commitments of those who use this concept, since his theoretical purpose is to get clear about these very commitments, and he will take the Argument From Untranslatability to show that the derived notion of disquotational truth, or rather each of the numerous concepts of translational disquotational truth, is by no means 'adequate to all practical and theoretical purposes'. It is only a poor surrogate for the concept of truth which we all have. Adherents of idiolectic disquotationalism could deflect the objection from untranslatability, as presented above, if they were to replace 'translation into my present idiolect' in (TrDisT) by 'translation into a potential expansion of m.p.i'. 3 9 But this proposal falls victim to a variant of the original objection (which I promise to refer to only once as the Argument From Untranslatability Into Any Potential Expansion of My Present Idiolect). Not every possible language which contains my present idiolect, I take it, is a potential expansion of this idiolect, but only a language which humans can come to be able to understand. Now there could be intelligent beings, Alpha-Centaurians, say, endowed with modes of sensory awareness and conceptual abilities that we and our descendants constitutionally lack. We may have the resources to understand the linguistic expression of some of their thoughts and thus have very good reasons for taking certain other noises they produce to be assertoric utterances in 'Alpha-Centaurian' as well, although they are such that humans are constitutionally incapable of ever understanding them. W h y should the fact that some of their utterances are forever incomprehensible to members of our species prevent them from being true? 40 Suppose

38

39 40

Once again I whole-heartedly agree with Gupta: 'The problem of explaining how one goes from "truein-my-present-idiolect" to "true" seems to me much harder than that of explaining 'true' using a limited ideology [i.e. limited conceptual resources]' (Gupta (1993), 8 1 , n. 48). As recently suggested in Field, 'Postscript', 148. Aping Peter Singer one might call the position I am about to attack alethic speciesism'.

192

Deflationism Attacked

that the system of signs some chimpanzees have been taught to use deserves to be called a language. From the fact that many of our sentences, e. g. 'This is a worthless 19th century copy of a painting most art-historians nowadays attribute to Giorgione', cannot be translated into a possible expansion of 'Chimpanzee', a language which could be learned by chimps, we would certainly not be inclined to conclude that these sentences are incapable of being true. So why should it be legitimate to conclude from the untranslatability of some sentences used by Alpha-Centaurians into a possible expansion of my idiolect that they cannot be true? Our concept of truth allows us to reckon with the possibility that there is at least one true sentence which cannot be translated into a sentence humans are capable of understanding. 41 Though richer than its predecessors, the extended derivative notion of disquotational truth, or rather each of the numerous notions of this type, is again only a substitute of the concept we esteem, chicory rather than coffee.

Bibliography Blackburn, S., Simmons, K. (eds.), 1999, Truth, Oxford. Carnap, R., 1942, [IS] Introduction to Semantics, Cambridge/Mass. —, 1947, Meaning and Necessity, Chicago. David, M., 1994, Correspondence and Disquotation, Oxford. Davidson, D., 1994, "What is Quine's View ofTruth?', Inquiry 37, 437-440. Dummett, M„ 1959, 'Truth', repr. in his 1978, 1-24 —> 1973, [FPL] Frege - Philosophy of Language, London. —, 1978, Truth and Other Enigmas, London. Field, H., 1972, 'Tarski's Theory ofTruth', as repr. in Platts, Μ. (ed.), 1980, Reference, Truth, and Reality, London, 83-110. —, 1986, 'The Deflationary Conception ofTruth', in: Macdonald, Wright (eds.), 55-117. —, 1994a, 'Disquotational Truth and Factually Defective Discourse', Philosophical Review 103, 405-452. —, 1994b, 'Deflationist Views of Meaning and Content', Mind 103, 249-284. —, 2001, (forthcoming), 'Postscript [to his 1994b]'. Gupta, Α., 1993, Ά Critique of Deflationism', Philosophical Topics 21, 57-81. Hallett, 1977, A Companion to Wittgensteins 'Philosophical Investigations', Ithaca. Kiinne, W., 1983, Abstrakte Gegenstände, Frankfurt/M. Leeds, S., 1978, 'Theories of Reference and Truth', Erkenntnis 13, 111-129. Lewy, C., 1976, Meaning and Modality, Cambridge. Manzoni, The Betrothed, chpt. 8 (quoting from St. Paul, Titus 1,15). McDowell, J., 1976, 'Truth-Conditions, Bivalence and Verificationism', repr. in his 1998, 3-28. —, 1998, Meaning, Knowledge and Reality, Cambridge/Mass. McGee, V., 1993, Ά Semantic Conception ofTruth?', Philosophical Topics 21, 83-111. Moore, G. E„ 1962, Common Place Book 1919-1953, London. Quine, W. V. O., 1953a, 'Notes on the Theory of Reference', as repr. in his 1961, 130-138. —, 1953b, 'Meaning and Existential Inference', as repr. in his 1961, 160-167. —, I960, Word an Object, Cambridge/Mass.

41 Notice that an attack on alethic speciesism is not yet a plea for alethic realism as a claim about truths which are comprehensible to human beings. On the other hand, the attack presupposes that not every truth is in principle knowable or warrantedly assemble by us, for knowability and warranted assertibility imply, of course, comprehensibility. In the last chapter of my forthcoming book 'Conceptions of Truth' I shall try to earn the right to neglect this option here by refuting the weaker claim of alethic anti-realism.

Disquotationalist Conceptions of Truth

193

—. 1961, From a Logical Point of View, 2nd edn., New York. —, 1970, [PL] Philosophy of Logic, Englewood Cliffs. —, 1973, The Roots of Reference, I.a Salic. —, 1981, Theories and Things, Cambridge/Mass. —, 1992, [PT] Pursuit of Truth, 2nd edn., Cambridge/Mass. Resnik, M„ 1990, 'Immanent Truth', Mind 99, 405-424. Soames, S„ 1984, 'What Is a Theory of Truth VJourn. Phil. 81, 411-429. —, 1999, Understanding Truth, Oxford. Tarski, Α., 1936a, [WB] 'Der Wahrheitsbegriff in den formalisierten Sprachen', enlarged transl. (by Leopold Blaustein) of his 1933, repr. in his 1986, 2: 51-198, transl. as 'The Concept ofTruth in Formalized Languages', in his 1983, 2: 152-278. —, 1936b, 'Grundlegung der wissenschaftlichen Semantik', repr. in his 1986, 2: 259-268; transl. as 'The Establishment of Scientific Semantics', in his 1983, 401-408. —, 1944, 'The Semantic Conception ofTruth and the Foundations of Semantics', repr. in his 1986, 2: 661699. —, 1969, 'Truth and Proof', repr. in his 1986, 4: 399-423. —, 1983, Logic, Semantics, Metamathematics, Papers from 1923 to 1938, transl. by Joseph H. Woodger, ed. by J. Corcoran, 2nd edn., Indianapolis. —, 1986, Collected Papers, Vols. 1-4, Basel-Boston-Stuttgart. Wittgenstein, L., 1956, Philosophical Investigations, Oxford. —, 1961, Notebooks 1914-1916, Oxford. —, 1969, Philosophical Grammar, Oxford. —, 1978, Remarks on the Foundations of Mathematics, 2nd edn., Oxford.

The Truth about Truth COLIN M C G I N N

Disquotation is the essence of truth. This much is widely accepted. It is less clear quite what this tells us about the concept of truth. My aim in this paper is to articulate exactly what the disquotational insight implies, and does not imply, about what it is for a proposition to be true — what follows from it and what it reveals about the inner nature of truth. I think the disquotational property of truth has been widely misinterpreted, and that the correct interpretation of it shows truth to be a far more interesting concept than has been recognised. Specifically, I am concerned with the question of whether disquotationalism entails something deserving the label 'deflationism'. The view I shall defend, which I call thick disquotationalism, holds that there is something importantly wrong about the deflationary interpretation. Truth is a more robust property than deflationism allows, despite its disquotational essence. Truth can be predicated of both linguistic items (sentences) and of what they express (propositions). This difference will not concern me in what follows. For convenience and purity I will take truth to apply to intensional items like propositions or beliefs, so that the question of their semantic interpretation is irrelevant. Given this, 'disquotational' isn't quite the right word, since only words can be quoted or disquoted: 'dis-intensional' or 'dis-representational' might be more accurate. But I will stick with 'disquotational', warning that it must not be taken too literally. Using italic 'p' as the name of a proposition, then, we can express the disquotational property of truth in the following familiar way: 'p is true iff p\ i.e. 'the proposition that ρ is true iff p'. Thus the truth predicate takes us from something referring to a proposition to something referring to what that proposition is about. Analogous general principles can be formulated for satisfaction and reference, as in: 'F is true of χ iff Fx' and 'a refers to b iff a=b', where again the italicised letters refer to concepts. My question, then, is exactly what the significance is of the truth - indeed the analytic truth - of these biconditionals: what do they tell us about truth, satisfaction and reference? Focusing on truth, what is the concept of truth such that the disquotational principle holds of it? What is it about truth that makes disquotationalism true? But before I get to this, I want to make a few remarks about competing theories of truth and how they run afoul of the disquotational property of truth. It can be quickly seen that the classic coherence and pragmatist theories of truth are committed to idealism of some sort by the disquotational character of truth, and idealism is not something one wants to be commited to just by virtue of one's analysis of the concept of truth. This is because it is facts about beliefs or desires and actions that constitute truth according to these theories. Thus the coherence theory says that ρ is true iff the belief that ρ coheres with other beliefs, and the pragmatist theory says that ρ is true iff the belief that ρ leads to the satisfaction of desires or to successful

T h e Truth about Truth

195

action. But now if we substitute in these biconditional the disquoted sentence licensed by the disquotation principle we get the result that facts about the world are constituted by beliefs and desires. For example, we start by saying: 'the belief that snow falls from the sky is true iff the belief that snow falls from the sky coheres with one's other beliefs'. Then, by disquotation, we can substitute to derive the following: 'snow falls from the sky iff the belief that snow falls from the sky coheres with one's other beliefs'. But this makes snows falling from the sky consist in something about one's beliefs which is a form of idealism. Snow could surely fall from the sky even if there were no beliefs in the world to cohere with each other. So the biconditional cannot possibly express an analytic truth, and it needs to if it is to claim to be a definition of truth. Similarly for the pragmatist theory: it makes snow's falling from the sky consist in a fact about human desires and actions - which it does not. To say that snow falls from the sky is not, failing idealism, to say that the corresponding belief coheres with other beliefs or that it leads to desire satisfaction. There is simply no analytic connexion here. That was the simple way to make the point, but it is open to a natural reply: why not make the two theories conditional on the existence of beliefs and desires? Thus we might say: 'Given that there is the belief that snow falls from the sky, that belief is true iff it coheres with other beliefs', and similarly for the pragmatist theory. Then we only get the result that, given the existence of an appropriate belief, snow falls from the sky iff the belief that it does coheres with other beliefs — and this does not commit us to idealism. No longer are we making the existence of the fact depend upon the existence of the corresponding belief. But this only postpones the problem, because we are still saying that snow's falling from the sky consists in the coherence of beliefs, or the satisfaction of desires - and these are still mental facts. We are still committed to supposing that snow cannot fall from the sky unless there is coherence among beliefs or satisfaction of desires - even though we have made the disquotational biconditionals conditional on the existence of beliefs and desires. Of course, proponents of these theories often had an idealist agenda; what I am saying is that this is built right into their theory of truth, once we take notice of disquotation. And we surely don't want idealism to follow directly from our definition of truth alone. The correspondence theory has a different kind of problem. Suppose we say '/> is true iff ρ corresponds to the fact that p'. Then the question is why it is only the fact that ρ that p corresponds to, and not say that fact that not-p. For example, we naturally say 'the proposition that snow falls from the sky corresponds to the fact that snow falls from the sky', taking the correspondence relation to relate that proposition to that fact and to that fact alone. But why not say that there is another correspondence relation that relates the proposition that snow does not fall from the sky to the fact that snow does fall from the sky? There surely is such a relation: it is the relation, not of truthmaker, but of falsity-maker. For any truth-making correspondence relation there is a falsity-making relation - the one that maps facts onto the negations of the propositions that they make true. So we can say that the proposition that snow does not fall from the sky corresponds to the fact that snow does fall from the sky - in the sense that there is a mapping from fact to proposition according to the making-false rule. Just as there is a function from facts to true propositions, so there is a function from

196

Deflationism Attacked

facts to false propositions. But clearly that is not the correspondence relation we need when we affirm the correspondence theory of truth: we need the correspondence relation that takes us from propositions to their irwrA-makers, not their falsity-makers. So let us amend the theory as follows: 'p is true iff ρ corresponds to a fact that makes p true (not false)'. But of course now this is blatantly circular, since it uses the concept of truth on the right-hand-side. The trouble is that the neutral notion of correspondence lets propositions and facts stand in the wrong kinds of relations to serve the purposes of the correspondence theory; but if we tie the correspondence relation down in the right way we have to stipulate that it is the relation that holds when a fact makes a proposition true - which uses the concept of truth again. Correspondence has to be understood as truth-making correspondence or else it won't work. This explains the air of triviality that surrounds the correspondence theory, and hence its apparent undeniability: it implicitly builds the idea of truth into the notion of correspondence. O f course, if we read the correspondence theory as simply a windy way of asserting disquotationalism, then there is no problem; but if the notion of correspondence is to do real work then we cannot avoid the question which sort of correspondence - and it will have to be defined as the truth-making kind. But there is no interest in a theory that says lp is true iff ρ corresponds to the fact that ρ in such a way that that fact makes p true. To put the point more generally: if we say that ρ is true iff />Rx, for some relation R and entity x, then we need to know that R maps ρ onto the right x, since there are many relations that map propositions onto entities (e.g. the proposition that snow falls from the sky maps onto the fact that I am sitting at my desk, since there is the relation of being typed by a person sitting at a desk that relates that proposition to that fact). But what is it that selects the relation that intuitively constitutes the truth of the proposition in question? If we say it is the truth-making relation, we have a circle; but it is hard to see what else we can say to avoid this circle. Certainly the correspondence theorist owes us an answer to this question or else his theory is quite unexplanatory. In the light of these problems, and many others I have not mentioned, but which are only too familiar, we do well to explore the prospects of the disquotational theory. Maybe registering the disquotationality of truth is all we need to say about truth to have a satisfactory account of its nature. This is a view that has been widely held and I am very sympathetic to it; but I think that the real import of the view has never been properly articulated - the task to which I now turn. There has been a standing tendency to suppose that the right and left sides of'ρ is true iff p' express the same proposition, that they say the same thing. This has fuelled a number of conceptions of truth that go by such names as 'the redundancy theory', 'the re-assertion theory', 'the pseudo-property/predicate theory', 'the disappearance theory'. The thought behind these slogans is obvious: if '/> is true' and 'p' express the same proposition, then the former expresses no more than the latter, and the latter contains no predicate at all applicable to propositions. The concept of truth gets swallowed up by the proposition to which we apply it. Thus we get the idea that to apply this concept to a proposition is simply to affirm the proposition, that 'true' is redundant, that it does not express a genuine property, that it disappears under analysis. To use the concept of truth is just an indirect way to say something you can say without it - some-

The Truth about Truth

197

thing not about propositions but about what propositions are about. When I say that the proposition that snow falls from the sky is true I am simply saying something about snow, namely that it falls from the sky; the reference to a proposition cancels out, as does the impression of a special kind of property expressed by 'true'. Truth is simply a way to go from outside a proposition to inside it (so to speak) - a device of semantic descent. Let us state this thesis as follows: disquotation entails disappearance. Then my argument will be that this thesis is mistaken: disquotation does not entail disappearance. If we call the disappearance thesis 'thin disquotationalism', then my own position can be called 'thick disquotationalism': disquotation with a robust truth property. What I primarily want to claim is that the left side expresses something logically stronger than the right side, so that they cannot be synonymous. The first reason for saying this is straightforward and I think uncontroversial, though its implications are sizable (as I shall argue): the logical form and ontological commitments of the left are different from those of the right. The left side has the logical form 'Fa', a one-place predicate 'true' attached to a singular term for a proposition (or other truth-bearer); whereas the right - though it may have this logical form - typically does not. The left refers to a proposition and thus is ontologically committed to such, while the right makes no such reference and has no such commitment: it refers to snow and whiteness and suchlike things. Therefore they cannot express the same proposition: for if propositions ρ and q have different logical forms and different ontological commitments, then they cannot express the same proposition. The plain fact is that the left side ascribes a property to something that the right does not; as I would put it, the left contains a predicate that denotes a property that the right does not contain, even implicitly. Thus the left has entailments that the right does not have; it is logically stronger in that it entails the right while entailing other propositions that the right does not entail. And this already shows that adding the truth predicate to a language expands the expressive power of that language; it increases the entailments of the language. As Tarski would put it, the metalanguage is always logically stronger than the (truth-free) objectlanguage. Secondly, there seem to be cases in which the right side does not entail the left. Take any example of a proposition that suffers from truth-value gaps, according to your philosophical predilections - fictional propositions, counterfactuals, ethical propositions, vague propositions. Now it is not that I am myself particularly wedded to truthvalue gaps for these types of propositions; my point is that the very concept of a proposition does not seem to imply that substituting it into the truth schema yields a truth. Consider the sentence: 'the proposition that Sherlock Holmes is a detective is true iff Sherlock Holmes is a detective'. Can't we quite comfortably affirm that Holmes is a detective without having to be committed to the claim that this proposition is truei For it to be true, 'Holmes' would have to refer to something, but it does not, so the proposition cannot be true. There is nothing to make this proposition true, but it still seems to be a proposition; even if it cannot be properly asserted, it can at least occur as the antecedent of a conditional, so it can function propositionally. Or consider the view that 'the king of France is bald' is neither true nor false and yet expresses a proposition; if so, what propositions bearing the properties of truth or falsity does this proposition imply? It seems to imply nothing about the truth or falsity of itself; certainly it does

198

Deflationism Attacked

not imply that it is true - if it did it would imply its own falsity, since it is not true (and not false either)! Propositions suffering from truth-value gaps cannot entail that they are true propositions, so they cannot entail the left side of a truth schema in which they occupy the right side. Most propositions are true or false, so this disparity does not show up in their case - we can generally say 'if p, thenp is true'. But this does not seem to be guaranteed by the very idea of a proposition, so it is not analytically the case that, for all propositions p, if p, then p is true. The notion of truth is more demanding than the mere notion of proposition - of what can be intelligibly said. We want to leave conceptual room for the idea of propositional speech acts that fail of truth and falsity. Calling a proposition true elevates above merely uttering it. So there is no entailment from the right to the left of an arbitrary instance of the truth schema. Let us agree, then, that the left side is richer expressively than the right; and let us agree that 'true' really does express a property, just as much as any other meaningful predicate expresses a property. The truth predicate of a language is a genuine semantic predicate, ascribing the property of truth to what instantiates it. Now the point I want to insist upon is that this is consistent with the thesis that truth operates disquotationally. I want to put together these two claims - that truth is a robust property, and that it is disquotationally definable — and ask what conception of truth emerges. Here, then, without further ado, is the essence of the concept of truth: truth is a property whose application conditions can be stated without making reference to that property — moreover, it is the only property of which this can be said. Let us accordingly say that truth is a self-effacing property in the foregoing sense. The first part of the claim that truth is self-effacing is easy to grasp: the right side gives a necessary and sufficient condition for truth to apply to a proposition but it makes no reference whatever to the property of truth - the application conditions of 'true' are given without alluding to the property this predicate denotes in any way. We must be careful to understand this claim correctly: the claim is not the trivial one that the application conditions of 'true' can be given in other words or in other concepts. Obviously, we can give the application conditions of 'bachelor' in other words, like 'unmarried male', and obviously too we can give the application conditions of 'water' by using the concept H20. But in these cases we are still referring to the same property we started out with, though using other words to refer to that property. My point is not that 'true' can be noncircularly defined - for that is true of many concepts. It is that truth can be defined without even making reference to the property 'true' denotes - or using any predicate equivalent to it. That is what is special about the disquotation principle: it explains truth without referring to it in any way, under any description. As we might put it, a shade paradoxically, truth applies to a proposition in virtue of something other than itself. I also claim, and will soon argue, that only truth is self-effacing in this sense, so that we can define truth as 'the self-effacing property' and pick it out uniquely. More exactly, I will argue that no other property sustains disquotation: only truth allows the kind of semantic descent it warrants. I am now in a position to spell out the essence of the disquotational nature of truth as I see it. There are two parts to this: First, truth is a property of a proposition from which one can deduce the fact stated by the proposition. Second, it is the only such property. Together: truth is that (unique) property of a proposition from which one

The Truth about Truth

199

can deduce the fact stated by the proposition. In other words, truth is the only property of a proposition which entails the fact that makes the proposition true. Propositions can have many properties - they can be believed, justified, denied; they can entail other propositions; they have constituent structure - but none of these properties entails the very fact stated by the proposition. They entail other facts, to be sure, but not the fact stated. This point is really quite self-evident: If you know that ρ is true you can thereby deduce that p. But if you merely know that p is believed or justified you cannot deduce that ρ - though you may be able to deduce some other proposition q. It is as if the property of truth enables you to look through the proposition right to the fact it states. In saying that truth is disquotational we are saying that it is reality-implying in this sense. By knowing that truth applies to a proposition you come to know, not just facts about propositions, but facts about the world. And this is a remarkable thing once one takes the measure of it: who would have thought that there is a property propositions have that points beyond them to the extra-propositional facts? All the other properties of propositions stay at the level of propositions - whether they are believed or justified or entail other propositions or what elements make them up. But there is this one property that takes us outside of propositions and down into the world beyond them. And this is directly related to the selfeffacing character of truth: because its conditions of satisfaction make no reference to it, but only to objects and properties in the world, via the nonsemantic right side of the truth schema, we can infer worldly facts from the application of 'true' to propositions. If the satisfaction conditions of this property were correctly stated by referring to it, then we would still be at the level of propositions; but because it is self-effacing in the way it is we can move from its application to a proposition to a fact about the world. It is not that 'true' expresses no property, so that '/> is true' means 'p'; rather, it expresses a genuine property that has the characteristic of being self-effacingly fact-implying. This is not then a trivial result of a synonomy, as it would be if we thought both sides of the schema said the same thing, but a substantive fact about a real property. How does it work with falsity? Falsity is not, strictly speaking, disquotational: we have the schema 'ρ is false iff not p', and the right side is not a disquotation of the left, since it contains 'not' and ρ lacks this word. But we can easily modify the rule to accommodate this: if a proposition has the property of falsity, then you can deduce its negation - if you know that ρ is false, then you thereby know that not p. Falsity is simply disquotation plus negation. When you know that a proposition has this property you can deduce the opposite fact to the one it states - so we have the same semantic descent and world directedness for falsity too. And we can similarly assert that falsity is the only property of a proposition that licenses the inference to the negation of the fact it states (or purports to state). Let us now consider some alleged counterexamples to the uniqueness claim. We should note that this claim is crucial if we are to have defined the notion of truth, since if truth is not the only disquotational property then this property does not individuate truth. (No one ever seems to concern themselves with establishing the uniqueness claim, being content to show that truth is disquotational - but this is clearly not enough if we are to have succeeded in fixing upon what distinguishes truth from all

200

Deflationism Attacked

other concepts.) Consider then the concept of knowledge: we have it that if χ knows that p, then p. So knowledge disquotes: you can infer the fact stated by a proposition from the property a proposition has of being known. Similarly for the property of following from something true, or the property of being believed by an infallible God. These are all distinct properties from truth, it may be said, but they all license the inference I am saying is distinctive of truth. I take it the answer to this question is obvious: each of these properties includes or embeds the notion of truth, and it is this embedded truth element that is doing all the disquotational work. Knowledge, omniscience, and entailment-by-a truth all imply that the proposition in question is true. So these are no more counterexamples to the uniqueness claim than the property of being both true and written in red ink is. Yet it does not seem self-evident that no other property could license disquotation. Take the property of being intelligible as applied to propositions. Certainly we cannot infer that grass is green from the fact that the proposition that grass is green is intelligible. But what about a proposition that says of itself that it is intelligible - the proposition expressed by the self-referential sentence 'the proposition expressed by this sentence is intelligible'? This proposition does have the property of intelligibility (just about!), and from its having that property we can infer that it is intelligible - but isn't that exactly what it says? So we can deduce that that proposition is intelligible, which is what it says, just from the fact that it has the property of intelligibility - we can therefore deduce the fact stated from the property ascribed to the proposition that states this fact. The proposition states that it has the property of intelligibility, and we can infer this fact from the knowledge that it has the property of intelligibility. Or consider the sentence: 'this sentence is written in English'. That sentence has the property of being written in English, so we can trivially infer that it is written in English: but that is what the sentence says; so we can infer the fact it states from the knowledge that it has the property of being written in English. Or suppose I have a belief that it realised in my brain by state S, and suppose this belief has that very content - that it is realised in my brain by state S. Then the fact stated by the proposition I believe can be inferred from the property my belief has of being realised by state S. So here we have three cases in which there is a convergence between the property possessed by a proposition and the fact it states. But I hope it is obvious that these contrived cases are extremely special and do not really threaten the claim I am making. My claim is that for any proposition truth licenses the inference in question; no matter which proposition you choose you can always make this move. But the properties just contrived are not such that for any proposition they permit the move — the move works only in these strange self-referential cases. It is not in general true that intelligibility, being-written-in-English and being realised by a particular brain state allow one to infer the fact stated by the propositions or sentences with these properties. In the vast majority of cases such an inference would be lamentably of the mark. So these are not counterexamples to the general claim I am making - that truth always permits the inference and nothing else has this power. Still, the putative counterexamples may serve to indicate why it is that my claim may not seem totally self-evident, since there is a kind of local violation of it in special cases of self-reference. The uniqueness claim is not trivial, though it is I think virtually unas-

The Truth about Truth

201

sailable. In fact, I think it is sufficiently obvious that no one has ever thought to defend it before — or even formulate it explicitly. But it does need to be made explicit and evaluated, as I have done here, if we are to assure ourselves that we have captured the essence of what truth distinctively is. It might help to bring out the role of the concept of truth if we try to imagine doing without it. Imagine being a member of a community of propositional beings who think and communicate but have no concept of truth. Someone says something to you and you register it, but you cannot apply the concept of truth to what is said. It seems, in these circumstances, as if you are in no position to form beliefs about the world as a result of what people say to you: all you know is that the speaker said that p, not that p is true. You cannot disquote on p and hence form beliefs about the world as a result of testimony, since you lack the device of disquotation that is the essence of truth. But now suppose you suddenly acquire the concept of truth, perhaps with the help of a friendly alien. All at once you can apply this concept to what people say and hence infer facts about the world. If you take what someone says to be true, then you can infer that p, for some ρ - you can acquire knowledge of facts. Of course, you can also come to know facts about the world directly without deploying the concept of truth, as when you simply see (say) that grass is green; in such cases there is no detour through other peoples speech acts. But if you are to acquire knowledge of the world on the basis of testimony, then you need the truth concept. Truth thus comes into its own when we start using other people s beliefs to acquire knowledge of the world. This is the pragmatic side of the disquotational property of truth; it explains why we care about this property, what it does for us. Without the concept of truth we could not learn from others; no truth, no education. Education is one long exercise in disquotation, using the teacher's beliefs to acquire knowledge of the world, this being mediated by the concept of truth. Without truth we would be condemned to be complete autodidacts. From this point of view, truth is essentially a device of inference. When I learn that a bird is yellow by seeing it I learn something about that bird, and what I might infer from this is incidental to acquiring that knowledge. But when I learn by testimony that a proposition is true the interest of this lies in what I can infer from this knowledge, namely the fact stated by the proposition. When I learn by testimony that the proposition that a certain bird is yellow is true what I learn is something I infer from this, namely that the said bird is yellow. I make a transition from proposition to world, and this transition is the whole point of the notion of truth. In saying that truth is a device of disquotation, then, we are also saying that it is a device of inference - inference to the disquoted form. It is not the mere fact that a proposition is true that is interesting; it what can be inferred from this about how the world is that is important. Truth is essentially a method for deducing facts from propositions. To bring out how special the inferential powers of truth are, let us formulate these powers abstractly. Suppose you are told that there is a certain property Ρ such that when Ρ applies to an object χ it is logically implied that some other object y has some distinct property Q; and further that this property Ρ is a nonrelational monadic property. This would be like being told that the property of being yellow is such that when it holds of an object χ it logically follows that some other object y is (say) square. That

202

Deflationism Attacked

would seem like a very remarkable claim: how can this object's being monadically Ρ possibly imply that that object is monadically Q? Surely that is impossible: what has the condition of the one object got to do with the condition of the other, logically speaking? But this is precisely how truth works. From the fact that one object, say the proposition that a certain bird is yellow, has the property of truth we can deduce that another object, a particular bird, has the property of being yellow: we can jump between entities and properties using truth as our springboard. Moreover, we are moving from properties of abstract or linguistic entities to properties of concrete things: and that sounds like saying that the number 2's being even logically implies that a particular bird is yellow! But that is precisely what truth allows: from the fact that an abstract proposition has a (nonempirical) property, viz. truth, we can deduce that the world of concrete objects instantiates certain empirical properties. This should strike us as more remarkable than it does; we are so familiar with this property of truth that we fail to appreciate how anomalous it is. Truth is not just a device of disquotation; it is a device of ontological leapfrog - or rather, that is what disquotation really amounts to, properly understood. Truth forms a logical bridge between the world of propositions and the world of objects and properties; it enables us to travel from propositions to the objects and properties they are about. No other concept can cross this ontological and deductive gap; truth is the only disquotational concept. In the light of all this we can now state a 'definition of the concept of truth, i.e. a condition that truth and only truth satisfies. Truth is to be defined as that property of a proposition that entails the fact stated by the proposition. This definition focuses on the disquotational aspect of truth and is intended to be a reformulation of that idea. But we can also define truth by employing the related notion of self-effacement introduced earlier: truth is to be defined as the self-effacing property, i.e. that unique property whose application conditions can be stated without reference to that property. Thus we can say 'p is true iff ρ has the self-effacing property' and offer this as a definition. These definitions do not compete with other definitions, say Tarski's; rather, they home in on certain features of truth, features that distinguish truth from other concepts. But the definitions themselves are far less important than grasping the unique (and remarkable) way that truth operates: it performs the miraculous feat of taking us from language and thought, on the one hand, to the world of objects and properties, on the other. No other concept has this power: truth is the adhesive that binds mind and world, to put it metaphorically and portentously. More soberly, when a belief has the property of truth (as opposed to any other property it might have), then the world is guaranteed to be a certain way. If this remystifies the concept of truth, then so be it. I shall now discuss the metaphysics of truth more directly. Again, I will be brief and dogmatic. The concept of truth seems simple in the following sense: it has no conceptual decomposition, and no empirical essence or nature. We cannot analyse it into conceptual constituents, and we cannot expect to discover a hidden underlying empirical structure for it. Truth is primitive, in this sense (which is not to say that nothing illuminating can be said about the concept). It has no analysis. But, despite the unanalysability of truth, it is possible to give a noncircular definition of the concept. This is peculiar: one would have thought that if a concept was simple and unanalysable then no account could be given of it in other terms. One would think, that is, that all we

The Truth about Truth

203

could say about 'true' is that it applies to a proposition iff that proposition is true. That is surely what one would say about 'blue' if one took this concept to be analogously primitive: 'blue' applies to an object iff that object is blue. But the peculiar thing about truth is that we can define it by means of the disquotation principle, even though it is primitive. Truth is a simple unanalysable property that can be defined: that is to say, noncircular necessary and sufficient conditions, of an analytically warranted kind, can be given for the instantiation of this property by a proposition. This is because truth is self-effacing: its application conditions can be given without referring to it under any description, and so its primitiveness does not stand in the way of providing these conditions. Truth is thus both definable and primitive (in the sense of having no conceptual decomposition and no underlying empirical nature). It is as if blueness could be the simple property it is and yet have application conditions given by reference to something quite other than blue objects. Put simply, the primitive property of truth applies to the proposition that snow falls from the sky in virtue of the fact that snow falls from the sky - and not in virtue of the proposition meeting some condition that analyses (or simply re-uses) the concept of truth. The predicate 'is true' holds in virtue of a condition not specified by the use of some (intensionally or extensionally) equivalent predicate. Yet it is still itself a genuine predicate standing for a real property - as much as 'blue' is. Therein lies the essential character of the truth concept, making truth an oddity in our conceptual scheme. Truth is really a very strange property indeed, when viewed in the right light. And it is precisely the disquotational aspect of truth that lies behind this oddity. To call the disquotational view 'deflationary' therefore strikes me as wide of the mark: truth turns out to be very interesting in its workings, not the banality some people suppose. Of course, it would be fair enough to use the term 'deflationary' if one held that 'p is true' says nothing different from - is synonymous with - p\ but we have seen that that is the wrong way to interpret the disquotational character of truth. Truth becomes interesting precisely when one accepts this principle and yet recognises that '/> is true' says something stronger than merely 'p' - in particular, that the former refers to a property the latter does not refer to. This is why I call the view I am defending thick disquotationalism, unlike the thin disquotationalism implied by the synonymy thesis. Truth is a substantial, robust property, as thick as any property, not the disappearing pseudo-property it is sometimes said to be. Or, putting the point in the formal mode, 'true' is as genuine a predicate as 'blue' or 'square'. What makes it unique is that it is a predicate that applies in virtue of something other than a predicate of what it applies to. Does any version of supervenience hold for truth? I think it does: if you fix all the nonsemantic facts, and you fix all the propositions, then you fix the application of the truth concept, analytically so. That is, if two possible worlds are exactly alike in the facts obtaining in them, and they contain the same propositions (which on reasonable assumptions they will), then exactly the same propositions must be true and false in the two worlds. This is indeed a simple consequence of the disquotational biconditional, read from right to left. If snow falls from the sky in a world, and there exists the proposition that snow falls from the sky in that world, then that proposition must have the property of being true in that world. In short, the truths are supervenient on the facts. Here, supervenience is trivially assured. Yet it would be a mistake to say that truth

204

Deflationism Attacked

is nothing over and above the facts on which it supervenes. Truth does not collapse into facts and propositions, since it is an irreducible property — though one whose instantiation is fixed by conditions that make no reference to it. Such supervenience may remind us of a familiar position with respect to moral goodness. G.E. Moore took goodness to be simple, unanalysable and non-natural, but he also took it to supervene on the descriptive and natural. I would say the same of the concept of truth, and I would adopt the same kind of realism about truth that Moore adopted for goodness. The truth property is a constituent of reality as much as blueness or electric charge or goodness is (though it is what we have been calling a logical property). It is a primitive constituent that nevertheless supervenes on facts that do not involve the notion of truth. But there is a significant disanalogy with goodness, namely that there is no counterpart to the naturalistic fallacy for truth. We cannot deduce that something is good simply from information about its nonmoral properties - there is always a logical gap here. There is always an 'open question' as to whether something is good, given that it has such-and-such descriptive properties. But nothing like this holds of truth with respect to its supervenience base: you can deduce that p is true given the information that ρ and the existence of p. There is no logical gap whatsoever here, thanks to the disquotational biconditional. So the irreducibility of truth does not result from a nonsequitur analogous to the naturalistic fallacy - there is no fallacy involved in inferring truth from fact. And yet the property of truth is not reducible to its supervenience base. Where there is still an analogy with goodness (and the other concepts discussed in this book) is on the 'nonnatural' status of truth. Truth is not a property that has causal powers or can be perceived by means of the senses; it is an object of intellectual cognition. It flouts naturalistic epistemology. It is 'queer'. But, as I remarked earlier, sometimes we just have to learn to live with the 'queer': denial and denigration are not sensible responses. What does seem clear, in the light of this nonnaturalism, is that 'deflationism' is not the right word for the kind of thick disquotationalism I have defended, if this is taken to imply that this view of truth is philosophically unproblematic or somehow 'tame'. If anything, my conception of truth deserves to be labelled inflationary. As I am conceiving it, truth raises many ontological and epistemological perplexities - but I do not regard this as an objection to the view I am defending. It is just the way things are. Often the right view in philosophy is the one that identifies accurately just where the problems lie; evading real problems can never be the route to philosophical understanding.

Generalizations of Homophonie Truth-sentences1 PETER VAN INWAGEN

I Homophonie truth-sentences are of three types. The following three sentences illustrate these types. 'Snow is white' is true if and only if snow is white The proposition that snow is white is true if and only if snow is white It is true that snow is white if and only if snow is white. Sentences of the first two types are predicate (homophonic truth-) sentences. Sentences of the first type are sentential predicate sentences. Sentences of the second type are prepositional predicate sentences. Sentences of the third type are operator (homophonic truth-) sentences. Homophonic truth-sentences have been of interest to philosophers for various reasons. One of the most important of these reasons is that they have seemed to many philosophers to suggest that the adjective 'true' is in some sense redundant, and that it may be possible to eliminate this adjective from our discourse without loss of content - and without introducing some new adjective or adjectival phrase (such as 'in correspondence with reality') to take its place. Homophonic truthsentences suggest that it may be possible to eliminate 'true' because 'true' occurs in the left-hand constituent of each homophonic truth-sentence and not in the right-hand constituent 2 ; nevertheless, there seems to be an obvious sense in which the right-hand constituent of a homophonic truth-sentence contains no less information than the lefthand constituent. And why would one wish to be able to eliminate 'true' from our discourse? Many philosophers believe - I do myself — that the perfect definition is the eliminative definition. Suppose, for example, that our vocabulary comprises four items, A, B, C, and D, and that we regard D as somehow philosophically problematical. If we can find a systematic method of translating all our sentences into sentences that convey exactly the same information but contain only the vocabulary items A, B, and C, then the theory that the perfect definition is the eliminative definition tells us that we have thereby achieved a perfect understanding of D. (The theory does not demand that we

1 2

This paper is dedicated to the memory of Herbert Heidelberger. More exactly: 'true' occurs one fewer times in the right-hand constituent of a homophonic truth-sentence than in its left-hand constituent.

206

Deflationism Attacked

find some phrase containing only A, B, and C with which to replace D at all its occurrences; it demands only that we find some systematic way to replace each sentence containing D with a sentence that contains only the other vocabulary items.) And almost all philosophers — from the pre-Socratics to Tarski and Heidegger — have regarded truth as a philosophically problematical concept. Analytical philosophers, at least, generally believe that the way to give an account of a philosophically problematical concept is to provide a non-trivial definition of the word that expresses that concept, and 'true' is the word that expresses the concept of truth. 3 To show how systematically to eliminate 'true' from our discourse, therefore, would be - at least so many philosophers would agree — to give an account of the philosophically problematical concept of truth. Homophonie truth-sentences suggest that it may be possible to eliminate 'true' from our discourse, but they do no more than suggest this. This is because 'true' occurs in sentences that are not of the forms exhibited by the left-hand constituents of homophonic truth sentences. There are for example, such sentences as 'The Axiom of Choice is true'; and there are generalizations like 'Only true propositions follow logically from true propositions' or 'Some of the things Alice said were true and some were not' or 'An analytic sentence is one that is true in virtue of its grammatical structure and the meanings of the words it contains'. Nevertheless, the suggestion is a powerful and attractive one, since it is hard to avoid the impression that each homophonic truthsentence is a particular instance of a general thesis. For example, '"Snow is white" is true if and only if snow is white' and '"Cows are purple" is true if and only if cows are purple' seem to be in some sense two instances of some one general thesis. And it is very tempting to believe that if we could find these general theses (there would be three, corresponding to the three types of homophonic truth-sentences), they would show us how to eliminate every occurrence of 'true' from our discourse. Can such generalizations be found, and (if they can be found) will they indeed show us how to eliminate 'true' from our discourse? Let us see. Let us first consider the predicate 'is true'. To eliminate a predicate systematically from our discourse, it is necessary to find some other predicate with which it may be replaced at any of its occurrences, and which in some sense has the same content as the original. We may, for example, replace the predicate 'is a sphere' with the predicate 'is a surface such that, for some point, it comprises all and only those points equidistant from that point'. It may be hard to spell out the sense in which 'is a surface such that, for some point, it comprises all and only those points equidistant from that point' "has the same content as" 'is a sphere', but it does seem that there is some important sense in which these predicates have the same content. Can something comparable be done for the predicate 'is true'? Let us first consider the case in which this predicate is a pred-

3

'True' is, to be sure, only the English word that expresses this concept. But no one, I suppose, doubts that if a non-trivial definition of 'true' could be given in English, then this definition could be "translated" into German, and its translation would be a non-trivial definition of 'wahr' in German - and so for French and 'vrai', and so for every language in which there is a word that means 'true'.

Generalizations o f Homophonie Truth-sentences

207

icate of propositions (non-linguistic objects of affirmation, denial, and acceptance or belief). Can 'is (a) true (proposition)' be eliminated from our discourse about propositions - and do prepositional predicate sentences provide some sort of clue as to how this might be done? The idea that suggests itself is this: find a biconditional whose lefthand constituent is 'χ is true' and from which all prepositional predicate homophonic truth-sentences (and no false sentences) follow by universal instantiation. In such a biconditional, it would seem, the right-hand constituent will contain the variable V free and will not contain 'true'. What would such a biconditional be? Certainly not this sentence, which is not even well formed: χ is true if and only if x. But if not this sentence, what? The problem of finding a biconditional from which all prepositional predicate homophonic truth-sentences (and no false sentences) follow by universal instantiation seems insoluble. If we say this, however, we shall no doubt be told that the problem is insoluble only because we have made it so: we have made it insoluble by treating V as a nominal variable; and (we shall be told) there are other kinds of variables than nominal variables. There are also predicate variables, and among these there are 0-place predicate variables, or sentential variables. (Nominal variables have the syntax of names or terms; sentential variables have the syntax of sentences.) If we have sentential variables ('/>', 'q' and so on) at our disposal - and "sentential quantifiers" to bind them - , we can find the general sentence we want. It will not, of course, be 'ρ is true if and only if p\ for this sentence, too, is not well formed - and its lefthand constituent is not 'x is true'. In order to get the general sentence we want, we need some construction that connects terms and sentences (or sentential variables). The obvious candidate for this construction is the operator 'the proposition that', which takes a sentence (or sentential variable) and makes a term. We can use 'the proposition that' and sentential variables to construct various sentences that are in some sense generalizations of homophonic truth sentences. For example: V/>: The proposition that p is true . if and only if p. But this sentence will not help us to find an equivalence for 'x is true', for it contains no nominal variables. The best solution to our problem, I think, is this χ is true if and only if 3\p . ρ and χ = the proposition that p. Admittedly it is not true that all prepositional predicate sentences follow from this sentence by universal instantiation. In fact, none do. What follow from this sentence by universal instantiation (if the nominal variable ranges only over propositions) are sentences like these: The Axiom of Choice is true if and only if 3\p . ρ and the Axiom of Choice = the proposition that p. The proposition that snow is white is true if and only if 3ρ . p and the proposition that snow is white = the proposition that p.

208

Deflationism Attacked

Still, both these sentences are true (provided, at any rate, that sentential variables and quantifiers make sense). And the propositional predicate sentence The proposition that snow is white is true if and only if snow is white follows logically from the second. 4 More generally, the sentence a: is true if and only if 3\p . ρ and a; = the proposition that ρ has no false instances,5 and every propositional predicate sentence can be deduced from it. This sentence may, therefore, be regarded as, in a loose sense, a generalization whose instances are propositional predicate sentences: it is a sort of compendium of the information they collectively provide. And it has these two important features: its left-hand constituent is 'x is true' and its right-hand constituent does not contain 'true'. We can, therefore, use this sentence to remove all mention of truth from our discourse - provided, of course, that every mention of truth that occurs in our discourse can be understood in terms of a truth-predicate that applies to propositions. Consider for example Some of the things Alice believes are true only if the Axiom of Choice is true. Our general sentence provides us with the resources to rewrite this sentence, preserving its content but removing all occurrences of the truth-predicate: 3x (Alice believes χ and :3p . ρ and χ = the proposition that p . only if 3p . ρ and the Axiom of Choice = the proposition that p). The case is less clear with sentential predicate homophonic truth-sentences, however. Can 'is true' (understood as a predicate of sentences) be eliminated from our discourse by the application of some generalization of 'Snow is white' is true if and only if snow is white? How should this generalization be stated? These questions will not be of any great interest to philosophers who accept the existence of propositions and who regard propositions as the "primary" bearers of truth-value. Such philosophers, having at their disposal a definition of 'is true' that applies to propositions can simply say that a sentence

4

5

Proof [we use obvious abbreviations and omit quotes]: LEFT-TO RIGHT. Assume TP(V is w). Our premise is TP(i is w) 3p . ρ & Ρ(ί is w) = Ρ (p). Hence, we have 3\p . ρ St Ρ (5 is w) = Ρ (ρ). By existential instantiation - or perhaps it would be better to say "particular instantiation" - we then have q & P(i is w) = P(q). We now introduce a premise that would obviously be a theorem of any logic whose vocabulary contained sentential variables and the operator 'the proposition that': V/> Vq : Ρ (ρ) = Ρ (q) —ρ q. By universal instantiation, we have P(i is w) = Ρ(q) —s is w q. From this and q &C P(f is w) = Ρ(q) we deduce s is w. RIGHT-TO-LEFT. Assume s is w. We have P(i is w) = P(i is w) and hence s is w & P(j is w) = P(i is w). And then by "particular generalization" we have 3\p . ρ &C P(i is w) = V(p). From this and our premise we have TP(i is w), and the proof is complete. Or at least this is true if we ignore the possibility that there are propositions inexpressible by any sentence — which I propose to do.

Generalizations of Homophonie Truth-sentences

209

is true just in the case that it expresses a true proposition - or is true on an occasion of utterance just in the case that the proposition it expresses on that occasion of utterance is true. But not all philosophers accept the existence of propositions. Those philosophers who see sentences as the only bearers of truth-value will wish to define 'is true' as a predicate of sentences without appealing to propositions. They will therefore be interested in the question whether this can be done by "generalizing" sentential predicate sentences - by finding a biconditional whose left-hand constituent is 'x is true' (the range of 'x' being understood to comprise sentences), which has no false instances, and from which all sentential predicate sentences can be deduced - whether by a single application of universal instantiation or by some more complicated deductive route. What might such a sentence be? Our generalization of propositional predicate sentences does not provide us with much guidance in answering this question, for that generalization depended on the fact that the left-hand constituent of a propositional predicate sentence contains a phrase ('the proposition that') that when prefixed to a sentence yields a name of the proposition that sentence expresses. T h e left-hand constituent of a sentential predicate sentence, however, contains a name of a sentence, and there is no phrase that, when concatenated with a sentence, produces a name of a sentence. There is no such operator as 'the sentence that', and, even if there were, the expression 'the sentence that snow is white is true' would be a different expression from '"Snow is white" is true'. (The operator 'the quotation name o f is not such an operator, since it must be prefixed to names of sentences - as in 'the quotation-name of the first sentence of this paper'.) A pair of quotation-marks itself cannot be thought of as an "operator" in any useful sense. It is true that applying a pair of quotation marks to a sentence produces the quotation-name of that sentence, but quotation marks interact with sentential variables in a way that renders them useless for our purposes, for a variable does not occur (and hence does not occur free) in its own quotation-name. The sentence χ is true if and only if 3p . ρ and χ = the proposition that p has, perhaps, no false instances. (The phrase 'the proposition that p' is an open term, a term containing a free sentential variable.) But the sentence χ is true if and only if 3 ρ . ρ and χ = '/>' has false instances whenever the value of V is a true sentence, for it is extensionally equivalent to χ is true if and only if 3p . ρ and χ = the sixteenth letter of the roman alphabet (italicized). And 'Snow is white' is true if and only if 3\p • ρ and 'Snow is white' = the sixteenth letter of the roman alphabet (italicized) is false. If we are to find a way of generalizing sentential predicate sentences, we must find

210

Deflationism Attacked

some phrase to do (mutatis mutandis) the work that 'x = the proposition that p' does in our generalization of prepositional predicate sentences. Actually, such a phrase is not hard to find. If one is willing to accept the existence of propositions, one may write such a phrase this way: χ expresses the proposition that p. If one does not accept the existence of propositions, one will have to find a sentence that, intuitively, seems to say the same thing as this (a sentence, of course, in which 'x' and 'p' and no other variables are free), but which does not involve an open term such that, when the free sentential variable in that term is replaced by a sentence, the result represents itself as naming a proposition. I would suggest this: 'x says that p\ Thus: 'Snow is white' says that snow is white The first sentence of this paper says that homophonic truth-sentences are of three types. (Both examples are intended to be true sentences. It would, of course, be no challenge to find false instances of 'x says that p\) Those who accept the existence of propositions will say that 'x says that p could be defined as meaning 'x expresses the proposition that p'\ those who deny the existence of propositions will have to be content to take 'x says that p as indefinable. We may then offer the following as the generalization of sentential predicate sentences we are seeking: χ is true if and only if Ξ\p . ρ and χ says that p. (The nominal variable, of course, ranges over sentences.) This sentence obviously has no false instances: the first sentence of this paper is true if and only if 3ρ . ρ and it says that ρ - and so on. 6 Can we deduce all sentential predicate sentences from it? Can we, for example, deduce 'Snow is white' if and only if snow is white from it? The answer to this question is a qualified Yes. The simplest deduction I have been able to construct uses two premises that are worthy of some comment. One is

6

I leave to one side the problem o f "self-referential" or "liar" sentences. (See Tarski's " T h e C o n c e p t o f Truth in Formalized Languages," in Alfred Tarski, Logic, Semantics, Metamathematics, translated by J . H . Woodger, second edn. edited and with an introduction by J o h n C o r c o r a n (Indianapolis: Hackett Publishing C o . , 1 9 8 3 ) , pp. 152-287. See particularly p. 162.) I justify this omission simply by pointing out that it is not m y project to defend the project o f defining truth by generalizing h o m o p h o n i c truthsentences. I mention "liar" difficulties at this point, rather than in connection with propositional predicate sentences, because I am convinced that these difficulties can be overcome, or evaded, in the case o f propositional predicate sentences. I will not defend this conviction in the present paper. If I a m w r o n g on this point, m y mistake will not weaken the argument o f the paper: again, it is not m y project to defend the project o f defining truth by generalizing h o m o p h o n i c truth-sentences.

Generalizations of Homophonie Truth-sentences

211

Vx V/> \/q : χ says that ρ &c χ says that q . —» . /*-> The other is 'Snow is white' says that snow is white. The first premise might be justified by saying that it obviously ought to be a theorem of any 'says that' logic: any logic whose vocabulary includes nominal and sentential variables and the 'says that' connective. (This premise plays roughly the same role in the proof as the role played by 'V/> V^ : Ρ (ρ) = Ρ {q) —>.p q in the proof in note 4.) The second premise - which is needed both to deduce 'Snow is white' from '"Snow is white" is true' and to deduce '"Snow is white" is true' from 'Snow is white' - might be justified by the contention that the following is a reasonable rule of inference for any 'says that' logic: Any expression formed by writing the quotation name of a sentence and then 'says that' and then that sentence may occur as a line in a proof. To generalize predicate homophonic truth-sentences, as we have seen, presents technical problems. These problems are due to the fact that the generalizations must contain both nominal and sentential variables. To generalize operator homophonic truth-sentences, however, presents no technical problems because the required generalization contains only sentential variables: V/> : it is true that />./>. Every sentence that follows from this general sentence by universal instantiation is an operator sentence and all operator sentences follow from this general sentence by universal instantiation. The simplicity of this case suggests that those philosophers who are interested in an eliminative definition of truth ignore predicate sentences and their generalizations and concentrate their attentions on operator sentences. Whether this policy would be workable depends on just one question: Can everything we want to say by using the predicate 'is true' be equally well expressed by using the operator 'it is true that' — given, of course, that we have sentential variables at our disposal? Or put the question this way: Can the truth-predicate be eliminated in favor of the truth-operator? It is evident that this can sometimes be done. For example, one need not take the sentence 'Certain things that Monica affirmed and Bill denied are true' to have the logical structure 3 x . Monica affirmed χ and Bill denied χ and χ is true if one has sentential variables at one's disposal. One can instead write 3ρ . Monica affirmed that ρ and Bill denied that p and it is true that p. But the case is more difficult with sentences that contain names of propositions that do not contain sentences that express the propositions they name. I have in mind expressions like 'the Axiom of Choice', 'the special theory of relativity' and 'Erdös's first important theorem'. (Not all philosophers will be willing to call these expressions "names of propositions"; let those who reject the description characterize these expressions as they will.) Consider the sentence

212

Deflationist!! Attacked

The Axiom of Choice is true. Can the content of this sentence be expressed using only the truth operator and not the truth predicate? How? As 'it is true that for every non-empty set of pairwise disjoint sets χ there exists a set that contains exactly one member of each member of χ and contains nothing else'? No, for no part of our original sentence expresses the content of the Axiom of Choice. (Or so I should suppose. Anyone who doubts this may substitute 'the principle that is the topic of Chapter X of Quine's Set Theory and Its Logic for 'the Axiom of Choice' in the example.) With 'it is true that the Axiom of Choice holds'? No, for 'holds' is just another way of saying 'is true'. (It should be evident, by the way, that there is no problem "in the other direction." Given the truth predicate, it is easy to eliminate the truth operator: 'it is true that p may be replaced by 'the proposition that p is true' or 'for some χ, χ says that p and χ is true'.) I doubt whether there is any solution to this problem. I conjecture that anything we can say using the adjective 'true' can be said in a language in which this adjective occurs only in the predicate 'is true'; I conjecture that some things we can say using the adjective 'true' cannot be said in a language in which this adjective occurs only in the operator 'it is true that'. (I take it for granted that anything we can say using the noun 'truth' can be said using only the adjective 'true'.) Still, we have seen that it is at least plausible to suppose that 'is true' (whether this predicate applies to propositions or to sentences) can be eliminated from our discourse by means of what may be loosely described as "generalizations of homophonic truth-sentences" - given that we have at our disposal sentential variables. We have been assuming that sentential variables make sense. It is time to examine this assumption.

II What are sentential variables and what are the quantifiers that bind them? How are these devices to be understood? One possibility is that they be understood substitutionally. But if we attempt to understand them this way, we face grave problems. Most writers explain the substitutional quantifier by providing a systematic statement of the truth-conditions for the sentences in which it occurs. These truth-conditions are standard and well known. If, for example, 'Σ' is the existential (or "particular") substitutional quantifier, the truth-condition for 'Σχ χ is a dog' is: 'Σχ χ is a dog' is true if and only if for some χ, χ is a name [not necessarily a name that has a referent] and "x is a dog" is true. 7

7

I use bold-face double quotes to do the work of "Quine corners" or "quasi-quotation-marks." Thus For some χ, χ is a name and "x is a dog" is true is read For some χ, χ is a name and the sentence that results from writing χ and then writing 'is a dog' is true.

Generalizations of Homophonie Truth-sentences

213

(The variable in this example is nominal. In cases in which 'Σ' binds a sentential variable, the right hand side will be a generalization on sentences rather than names.) But this sentence does not tell us what 'Σχ χ is a dog' means - it tells us only that, whatever it means, it is true just in this case: prefixing some name to 'is a dog' yields a truth. That is to say, it is true just in the case that some "substitution-instance" of 'x is a dog' is true. What, then, does 'Σχ χ is a dog' mean? One possibility, of course, is that it means is an abbreviation of - 'For some χ, χ is a name and "x is a dog" is true'. Most advocates of substitutional quantification deny that this is how substitutional quantification should be understood, but the question that should interest us is whether this interpretation enables us to understand the sentential variables that, as we saw in Part I, are essential to an eliminative definition of truth. And it would seem that the answer to this question must be No. Consider our definition of the truth-predicate (as a predicate of propositions): χ is true if and only if 3p . p and χ = the proposition that p. Suppose '3' is understood as the substitutional particular quantifier, 'Σ', and that the substitutional particular quantifier is understood in the way suggested. Then 'Σ/> . ρ and χ = the proposition that p' abbreviates For some y, y is a sentence and "y and χ = the proposition that y" is true. Not only does this expression contain 'true', the word the definition is supposed to eliminate, but the nominal variable 'x' does not occur free in this sentence; in fact, strictly speaking, V does not occur in this expression at all - which is an abbreviation for For some y, y is a sentence and the sentence that results from writing y and then writing 'and χ = the proposition that' and then once more writing y is true. This expression is a closed sentence - and a false one, since its instantiation to any sentence is false. For example, 'Snow is white' is a sentence and the sentence that results from writing 'Snow is white' and then writing 'and χ = the proposition that' and then once more writing 'Snow is white' is true is false, since 'Snow is white' is a sentence and 'Snow is white and χ = the proposition that snow is white' is true is false - open sentences being neither true nor false. It seems, then, that if we wish to eliminate 'true' from our discourse, and if the way we propose to do this requires sentential variables and quantifiers, and if we propose to understand sentential quantifiers as substitutional quantifiers, we had better not understand sentential substitutional quantifiers "metalinguistically": we had better not contend that a sentence of the form 'Σρ . . . . / > . . . ' is no more than an abbreviation for the corresponding sentence of the form

214

Deflationism Attacked

For some χ, χ is a sentence and " . . . χ . . . " is true. But if we do not understand substitutional quantifiers metalinguistically, how shall we understand them? In "Why I Don't Understand Substitutional Quantification", 8 I defended the thesis that there that there is no way to understand substitutional quantifiers, that substitutional quantification is simply meaningless. (I explicitly considered only substitutional quantification into nominal positions, but my argument is equally applicable to substitutional quantification into sentential positions.) I argued that this was the case because the advocates of substitutional quantification had done nothing to explain the meaning of substitutional quantifiers other than to give truth-conditions for the sentences in which they occur and to say that a sentence containing substitutional quantifiers does not have the same meaning as the statement of its truth-conditions. The argument was essentially this. Suppose I introduce some new vocabulary item, X, into our language. I set out the syntactical features of X, so that it is clear which sentences containing X are well-formed or grammatical. Let A be the (infinite) set of well-formed sentences containing X. I go on to offer a mechanical procedure that pairs each sentence a belonging to A with a sentence b that does not contain X (and which we understand), and I say two things: first, that a is true if and only if b is true; secondly, that a means something other than b. If I say only this much (and "this much" is all that stating systematic truth-conditions for sentences containing substitutional quantifiers comes to), I have not told you what X means. Consider this simple case. I introduce a unary sentence-operator ' — I tell you these two things and tell you nothing more: the result of prefixing ' — t o a sentence is true if and only if the negation of that sentence is true; the result of prefixing '—i' to a sentence means something different from what the negation of that sentence means. Do you now know what '—i' means? No, you do not. There are lots of sentences that have the same truth-value as the negation of 'Snow is green'. There are even lots of sentences that can be known a priori to have the same truth-value as the negation of 'Snow is green': the negation of 'Snow is green, the conjunction of the negation of 'Snow is green' with '2 + 2 = 4', the conjunction of the negation of 'Snow is green' with the alternation of 'Snow is green' and its negation . . . . If I tell you only that '—ι Snow is green' doesn't have the same meaning as the first of these, I don't tell you that it doesn't have the same meaning as the second, and I don't tell you that it doesn't have the same meaning as the third. And those two sentences have different meanings. Providing a systematic statement of truth-conditions for all sentences containing substitutional quantifiers does not, therefore, enable us to understand these sentences. Since I wrote "Why I Don't Understand Substitutional Quantification," another way of understanding substitutional quantification has been proposed. The advocates of this second approach propose to regard particular substitutional quantifications on a predicate as disjunctions - disjunctions that include as disjuncts every substitution-instance of that predicate.9 That is, where α is any variable and F any predicate, they propose to re-

8 9

Philosophical Studies 39 (1981) pp. 281-285. As far as I know, this approach to substitutional quantification was first mentioned in print by Hartry Field, in his review of Dale Gottliebs Ontological Economy (Noüs 18, 1984, pp. 160-165). I. L. Hum-

Generalizations of Homophonie Truth-sentences

215

gard "Σ α F a " as an abbreviation for the disjunction of all substitution-instances of " F a " . (Universal substitutional quantifications on a predicate are to be understood in the parallel way as conjunctions.) But this proposal has odd results if there are as many names as there are natural numbers - and there certainly are that many names, for the standard, arithmetical names for the natural numbers are as numerous as the natural numbers. For the sake of the example, let us suppose that the standard, arithmetical names of the natural numbers are the only names there are. Then Σχ χ

is odd

is equivalent, on this proposal, to 0 is odd ν 1 is odd ν 2 is odd ν . . .

k

is odd ν . . . .

This proposal, or so one might argue, does provide a meaning for 'Σχ· χ is odd'; it does tell one how to translate this sentence into language one already understands. So one might argue. But if one does so argue, one seems to be employing the premise that there are infinitely long sentences, sentences that have no end. And this is very hard to believe. And matters seem, if possible, to become even worse when we consider open sentences whose variables are bound by more than one quantifier. Suppose the dual of 'Σ' is A'. Consider the sentence Αχ Zj/ y

>x.

This sentence will be an abbreviation for an unending conjunction of unending disjunctions: 0 >0ν 1>0v2>0v...£>0v... & 0 > 2 ν 1 > 2 v 2 > 2 v . . . k > 2 v . . .

& 0 > l v l > l v 2 > l v . . i > l v . . . & . . . 0 >

ν 1 > j v 2 > j v . . . k > j

ν ...&.. . . It really is very hard to believe that there are such sentences as these. Even if one does not share my scruples about the existence of such sentences, one must admit that the advocates of the proposal we are considering must solve some difficult technical problems. Consider only nominal quantification. It is not enough to say that Ax "Ly y >x abbreviates an infinite conjunction of infinite disjunctions; one must also devise an algorithm that - in the present case - takes 'Ax Xy y >x' and yields a finite display (no doubt containing a lot of 'fs and 'k's and lots o f ' . . .s) that uniquely represents the of course unwritable "doubly infinite" sentence that lAx Zjy y >x supposedly abbreviates. One would then have to devise explicit rules of inference (corresponding to UI, EG, and so on) for manipulating the finite displays the translation algorithm yields. One would, finally, have to present a proof that the display the algorithm correlates with a sentence Q in the language of quantification is a theorem of the "new" system if and

berstone suggested this approach to me in 1981, in a letter. A recent, very sophisticated treatment of quantification, partly based on the idea that existential or particular quantifications are disjunctions of their substitution-instances, can be found in Mark Lance, "Quantification, Substitution, and Conceptual Content," Nous 30, 1996, pp. 481-507.

216

Deflationism Attacked

only if Q is a theorem o f substitutional quantifier logic. Devising such a translation algorithm, such a set o f rules, and such a proof, will present a considerable technical challenge. But let us suppose that this challenge can be met. (I am no logician, but my guess is that it can be.) Even if it can be, and even i f there are "multiply infinite sentences" - and, o f course, it would be no profound problem to find set-theoretical constructs with the properties necessary to play the role o f such sentences - I do not understand any o f them, for I can understand only finite sentences. And if the only thing that can be said about the meaning of'Ax Eyy >x is that it abbreviates a certain (specifiable) "doubly infinite" sentence, I shall therefore not understand 'Ax y >x\ This point does not essentially involve the infinite length o f the unabbreviated sentences. Essentially the same point could be made about very long finite sentences. Suppose someone proposes to use the sentence 'The natural numbers less than one thousand are even-odd' as an abbreviation for Ό is even and 1 is odd and . . . 9 9 9 is odd'. Although I believe that the unabbreviated sentence exists, and although I believe that it is true, I cannot understand it (for more or less the same reason I cannot visualize a chiliagon: I can't "get it into my mind"), and I therefore have no understanding o f the sentence that abbreviates it. M y contention that I cannot understand Ό is even and 1 is odd and . . . 9 9 9 is odd' should not be confused with the contention that I cannot understand Ό is even and any number less than 1 0 0 0 is odd if it is the successor o f an even number and even if it is the successor o f an odd number'. T h a t sentence I can understand. But then that sentence contains only thirty-one words: the sentence I cannot understand contains 3 , 9 9 9 words. I recognize, moreover, that the sentence Ό is even and 1 is odd and . . . 9 9 9 is odd' is a perfectly meaningful English sentence and that it is true. But to say that is not to say that I understand it.

Ill How is sentential quantification to be understood if not as a species o f substitutional quantification? In this section, I will examine the most important non-substitutionalist attempt to explain sentential quantification: that o f Dorothy Grover. (What I have called sentential quantification, Grover calls propositional quantification. I will stick to my own term in the following discussion o f her views.) Recognizing that the "variables" o f nominal quantification are essentially pronouns, and that a proper understanding o f nominal quantification must proceed from this fact, 1 0 Grover proposes that the variables o f sentential quantification — the 'p's and 'q's that occupy sentential positions and are bound by sentential quantifiers - be understood in an exactly parallel way: as formal versions o f words or phrases o f a kind she calls "presentences," items that stand to sentences as pronouns stand to nouns. It is,

10

I say "recognizing that" because I agree with this view of nominal quantification - or, as I prefer to say, quantification tout court, quantification full stop, quantification period. I defend this understanding of quantification in Chapter 1 of Being: A Study in Ontology, forthcoming from Oxford University Press, and, more briefly, in "Meta-ontology," Erkenntnis 4 8 (1998), pp. 233-250.

Generalizations of Homophonie Truth-sentences

217

however, o f at most heuristic value to say, "Presentences are items that stand to sentences as pronouns stand to nouns." Some sort o f definition is required, and Grover has provided one. She has defined presentence' by providing a list o f three defining properties: a prosentence is any word or phrase that (1) Is not a sentence but has the syntactical properties o f a sentence (2) Can be used to make an assertion, serve as the antecedent o f a conditional assertion, and can, in general play all the linguistic roles a sentence can play. (Or can with the help o f context. This is parallel to the case o f pronouns and nouns: a third-person-singular pronoun - 'it' - can be used to refer to an object, but it must pick up its referent from the context in which it is used.) (3) Can be used anaphorically - or, near enough, different occurrences o f a presentence in the same sentence can have the same antecedent. 11 But we may ask: W h a t reason have we to believe that anything has all the properties in the list or even that those properties are mutually consistent? We need some reason to believe that presentences exist, or at least that the defining properties o f a prosentence are consistent. (If the defining properties o f a prosentence are consistent, then, even if prosentences do not in fact occur in English or other natural languages, they could easily be added to the vocabulary o f a language by stipulation: one would simply pick some class o f words or phrases - preferably ones that do not already occur in the language - and stipulate that its members have the defining properties o f prosentences.) I, at any rate, am sufficiently sceptical about the linguistic category "prosentence" that I should want to see an argument for the conclusion that the defining properties o f prosentences were consistent before I was willing to accept any theory that made essential use o f this category. I know o f no other way to show that the list o f defining properties o f prosentences is consistent than to provide an example o f a prosentence. Can this be done? Grover has endorsed Joseph Camp's suggestion that 'it is true' (or 'that is true') has the defining properties o f a prosentence, and is therefore a prosentence o f English. 1 2 Will this example do? Does it prove that there are English prosentences? I doubt whether it does. I will try to explain my doubts. I begin with an observation: even if we have a prosentence, we shall have to have some way o f connecting its occurrences in a sentence with their intended antecedents if we are to be able to understand the sentences that contain arbitrarily many sentential quantifiers and arbitrarily many sentential-position variables. In Being and "Meta-ontology" (see note 10), I solve the analogous problem for 'it' by supplying 'it' with an indefinite stock o f subscripts. (That is, 'it x ', 'it ', and so on. Thus, the second occurrence o f ' i t ' in the quantifier-phrase 'it is true o f

11

12

See " Prepositional Quantifiers" in Dorothy Grover, A Prosentential Theory of Truth (Princeton: Princeton University Press, 1992), pp. 46-69. (This essay was originally published in the Journal ofPhilosophical Logic, in 1972.) The defining properties of a prosentence are set out on pp. 52-53. See the "Introductory Essay" in A Prosentential Theory of Truth (pp. 3-45), p. 12.

Deflationism Attacked

218

everything that it x is such that' is an antecedent of appropriately placed occurrences of 'it x \ but not of occurrences of 'it y ' or 'it z '.) But the "prosententialist" who tries to use subscripts in a like manner will face a problem that I did not face: where the subscripts are to be placed. This problem arises because 'it is true' (unlike 'it') has a grammatical structure, and thus affords more than one "site" at which a subscript might be placed. There seem to be two serious candidates for the placement-site for the subscript, which we may illustrate as follows: it is true Ρ and (it is true)p. That is, we might attach the subscript to the (apparent) grammatical subject of the prosentence-candidate, or we might attach it to the prosentence-candidate "as a whole." (To attach it to the copula would seem to be simply a variant on the second option: to attach it to the copula would be a way of attaching it to 'it is true' "as a whole"; after all, the copula in a subject-predicate sentence is a mere grammatical convenience, one that many languages manage very nicely without. And I can't see what could be intended by attaching a subscript to 'true'.) Let us suppose that we have chosen the first option: we will attach our subscripts to 'it'. Are we now in a position to translate (say) 'V/> (ρ ν -ρ)' into what Grover has called "philosopher's English"? (If we cannot succeed in this simple "one-variable" case, there will be no point in proceeding to more complex examples.) Well, we know how to translate '/> ν -/>': it is true ν -it is true. Ρ Ρ But how are we to translate 'V/>'? How are we to translate the whole sentence 'Vp (ρ ν "/>)'? Someone might offer: For all p, it p is true ν -it p is true. But, to my mind, at least, this will not do. This leaves us with an untranslated variable in the quantifier-phrase. How are we to translate 'For all pi How, that is, are we to write the quantifier-phrase so that it contains not 'p' but 'it p is true'. (In my account of "nominal" quantification in Being and "Meta-ontology," it is 'it x ' and 'it ' that occur in quantifier-phrases - not V and j/\) I do not see any way to do this. It is, of course, easy enough to get 'it ' into the quantifier-phrase. We could read 'for all ρ as it is true of everything that it p is such that. But in this phrase, the subscripted pronoun is not followed by the predicate 'is true': the supposed prosentence 'it p is true' does not occur in this phrase. This reading of'for all p' would seem, therefore, to leave us with good, old-fashioned nominal quantification. If we so read the quantifier-phrase, then the sentence 'V/> (ρ ν -p)' reads as follows in "philosopher's English": It is true of everything / b that p it is such that it ρ is true ν - itρ is true.

Generalizations of Homophonie Truth-sentences

219

The meaning of a sentence containing subscripted pronouns can hardly be supposed to depend on the particular symbols that are employed as pronoun-subscripts. 13 This sentence, therefore, is equivalent to It is true of everything that itx is such that itx is true ν - itx is true. That is to say: Vx (x is true ν - χ is true). So interpreted, the language of prosentential quantification seems to be no more than a rewriting of the language of nominal quantification over sentences or propositions for only sentences or propositions satisfy 'x is true'. And a rewriting of only a part of this language, for there is nothing in the language of sentential quantification that corresponds to the identity-sign. For this reason, there are things that can be said by those who are willing to combine the identity-sign and quantification over propositions that cannot be said by those who restrict themselves to that part of the language of quantification over propositions that can be translated into the language of sentential quantification. For example: 'For any contingent proposition, there is some distinct contingent proposition that it entails'.14 Perhaps, however, these untoward results are a consequence of the fact that we chose to attach the differentiating subscript to 'it' in 'it is true', thus (it might be argued) forcing 'it' to behave as a pronoun, contrary to the intention of the proponents of sentential quantification that 'it' should be an inseparable part of the presentence 'it is true', a part that has no syntactical properties (as the two characters that make up 'it' have no syntactical properties). Let us now turn to the other option we mentioned. Let us suppose that the differentiating subscripts are to be attached to the prosentencecandidate "as a whole," that is, in the following fashion: (it is true) p (it is true)^. (In the following discussion of this suggestion, I shall drop the round brackets; but it must be understood that throughout this discussion the 'p'- and 'q'-subscripts apply to 'it is true' and not simply to the adjective 'true'.) Then the above objections have no force against the idea of understanding sentential-position variables as prosentences. But we are still faced with the problem of how we are to formulate quantifier-phrases. Let us pose the problem this way. Consider the formula Vpipv-p).

13 Cf. the principle of the equivalence of alphabetic variants ('Vx χ = χ and 'Vz ζ = z\ for example) in formal logic. 14 I mention this only as a problem. It might be possible to solve this problem by introducing some "sentential counterpart" to '=' — something that does the work in respect of sentential quantification that . . and . . . are propositions & . . . = . . . ' does in respect of nominal quantification and has the syntax of a sentential connective. Such a connective might be introduced as equivalent to ' • (. . . < - » . . . ) ' or to 'necessarily, to assert that . . . would be to assert that . . . and vice versa'.

220

Deflationism Attacked

How is the whole of this formula to be turned into English — or into a supplemented English that contains phrases like 'it is true p ' and 'it is true^'? That is to say, with what expression of generality may we prefix it is true or it is not the case that it is true Ρ Ρ so as to produce something that (a) is a sentence of (supplemented) English, and (b) seems intuitively to express what is supposed to be expressed by 'V/> (ρ ν -/>)'? Grover generally uses the following as a universal quantifier phrase: 'for any proposition. (Thus, she might offer 'For any proposition, it is true or it is not the case that it is true' as an English reading of 'V/> (ρ ν -/»)'. Let us leave aside the charge that this looks a lot more like an English reading of ' V x (x is a proposition —» (x is true ν - χ is true))' than an English reading of 'V/> (ρ ν -/>)'. How something looks is, after all, a matter of subjective judgment.) This quantifier-phrase will do only in the simplest cases, the one-variable cases. Since it offers no way of distinguishing one universal quantifier phrase from another or one existential quantifier phrase from another, it will not do in cases involving multiple generality. We have to be able to distinguish, say, Vp3q(p->

q)

from Vf

3p {p -» q),

and 'For any proposition there is a proposition' does not enable us to distinguish 'V/> 3 q from 'Vq 3\p\ We need some way to "connect" each quantifier-phrase with a particular occurrence of a prosentence. There is, of course, always the "brute force" method: For any proposition p , there is a proposition^ such that if it is truep then it is true^. But, really, what does this meant We might try to make sense of it by supplementing English with lots of exact synonyms for 'proposition', perhaps by the ever-useful "subscript" method: 'proposition p ', 'proposition^', and so on. Would this serve to establish an intelligible connection between quantifier-phrases and the sentential-position variables they are supposed to bind? I think not. By way of illustration, let us suppose that 'proposition' and 'thesis' are exact synonyms. And let us suppose that 'it is true propo ar| sition' d 'it is true thesis ' are distinct presentences. Having made these suppositions, we might try to distinguish our two sentences by providing the following readings for them: For every proposition, there is a thesis such that if it is tme

p r0 p 0S i t i 0n then it is

true

prci p 0S i t i 0n

thesis

For every thesis, there is a proposition such that if it is true

true

then it is

thesis-

But I confess that these two sentences make little sense to me. (Well, to be frank, no sense at all.) 'Proposition' and 'thesis' both have meanings (even if their meanings are the same): they are not mere indices. It seems to me to be absurd to suppose that these

Generalizations of Homophonie Truth-sentences

221

two sentences are sentences of a transformed, regimented English between which there is a clear difference of sense (or even an inchoate difference of sense that one might have some hope of making explicit and precise). Let us return to the question we have posed and to which we have so far failed to find an answer: with what expression of generality may we prefix it is true or it is not the case that it is true Ρ Ρ so as to produce something that (a) is a sentence of (supplemented) English, and (b) seems intuitively to express what is supposed to be expressed by 'V/> {ρ ν -/>)'? It should be evident from the foregoing that any prefix that satisfies these conditions will have to contain 'it is truep'. But let us make things as easy as possible for ourselves. Let us take a step backward and neglect the subscripts; let us ask this question with respect to 'it is true or it is not the case that it is true'. (If we can find no answer in this case, there will be no point in going on to consider multiple generality.) I will examine two possible answers. (In each case, I will write out the whole sentence and italicize the suggested prefix.) 1. Under whatever circumstances it is true, it is true or it is not the case that it is true. But this cannot be right. For one thing, it doesn't make sense, or doesn't make sense unless 'it' (which stubbornly insists on behaving as if it were a pronoun and the subject of the verb 'is') can pick up a referent from its context. "Under whatever circumstances what is true?", one wants to ask. For another, if this suggestion confers truth on 'V/> (j> ν ~p)\ it seems to be just as happy to confer truth on 'V/> f - for if one does manage to assert something by saying, "Under whatever circumstances it is true, it is true," what one manages to assert will certainly be true. This second problem is avoided by the following answer to our "prefix" question: 2. Whether it is true or not, it is true or it is not the case that it is true. Someone who managed to assert something by saying this would no doubt say something true, and someone who said, "Whether it is true or not, it is true," would, or so I should think, say something false if'it', in the context of utterance, referred to a false statement. But the first problem is intractable by any variant on this suggestion. The word 'it' insists on behaving like what it is: an ordinary pronoun, and, if the sentence is uttered in a context that fails to supply it with a referent, the person to whom it is addressed is going to want to ask, "Whether what is true or not?" In the end, it seems that we do not have to go on to determine whether any suggestion along the present lines can be elaborated to solve the problem of multiple generality. Any such suggestion must fail even in the simplest case. The English replacement for the sentential quantifier phrase ' Vp' must contain 'it is true' if it is to interact in the appropriate way with the sentences to which it is prefixed (sentences in which 'it is true' occurs and allegedly functions as a prosentence). And in any such replacement, the phrase 'it is true' will stubbornly behave not like a syntactically structureless unit (as the second 'it' does in 'it is true of everything that it is such that') but like what it is: a phrase having a subject-copula-predicate structure. And any utterance of a

222

Deflationism Attacked

sentence that starts with this replacement will therefore constitute an assertion only if the 'it' occurring in the replacement manages somehow to pick up a referent from the context of the utterance. 'It is true' is therefore not a presentence - or if it is, then presentences cannot per se be used to make sense of sentential quantification. (Maybe there are presentences, hiding somewhere in the jungle of natural language, and perhaps they can be used to make sense of sentential quantification. But if there are natural-language presentences, I have no idea what they might be.) If 'it is true' is not a presentence, we are left without an example of a presentence and are thus left without any reason to suppose that Graver's list of properties is consistent — without any reason to suppose that "presentence" is a possible grammatical category. I therefore judge her attempt to explain sentential quantification to be an extremely interesting failure. I conclude, tentatively, that it is at least extremely doubtful whether the idea of sentential quantification can be made sense of. But, as we have seen, the project discussed in Section I (finding the general theses of which homophonic truth-sentences are instances, and using these general theses to provide an eliminative definition of 'is true') depends essentially on the use of sentential quantification. It is therefore extremely doubtful whether this project can succeed.

IV Tarski Challenged

An Argument Against Tarski s Convention Τ ANIL GUPTA

1 .Introduction Alfred Tarski proposed his celebrated Convention Τ as a contribution to the solution of an important philosophical problem: that of giving a satisfactory definition of truth. 1 As Tarski explains, a satisfactory definition of a notion needs to meet two sets of requirements. The definition has to be, in Tarski's terminology, formally correct and materially adequate. A definition is formally correct iff 2 (i) it conforms to the logical rules governing definitions and (ii) only acceptable vocabulary occurs in its definiens. For example, the definitions χ is true iff χ is true and χ is true iff χ corresponds to reality, Tarski would say, fail the condition of formal correctness. The first fails because the occurrence of 'true' in the definiens makes the definition circular and, thus, results in a violation of the logical rule that bars circularity in definitions. The second definition fails because unacceptable terms occur in its definiens. 'Corresponds' and 'reality' are much too vague and unclear to be acceptable in a definition of truth. A definition can meet all the requirements of formal correctness and yet be unsatisfactory. The definition χ is true i f f χ = χ is undoubtedly formally correct. But it is unsatisfactory because it fails to be materially adequate. The definition implies, for instance, that everything is true and thus plainly fails to capture our ordinary concept of truth. A materially adequate definition of truth is required to accord with the ordinary uses of'true'; it should, in Tarski's words, "catch hold of the actual meaning" of'true' ( S C T p . 13). Tarski constructs in his paper "The Concept of Truth in Formalized Languages" a definition of truth for a particular language — the language of the calculus of classes — and offers a mathematical proof that it is satisfactory. Tarski's definition is surprising -

1

2

See Tarski 1935. See also Tarski 1944. T h e page references for the first paper, cited as CTFL, will be given for the Logic, Semantics, Metamathematics volume. T h e page references for the second paper, cited as SCT, will be given for the Linsky reprinting. T h e expression 'iff' abbreviates 'if and only if'.

226

Tarski Challenged

it is quite unlike any earlier proposal —, but his proof that it is satisfactory is, if anything, even more surprising. Formal correctness of a definition, it may be granted, can sometimes be rigorously established. But the idea that the same is possible for material adequacy is highly surprising. Material adequacy of the definitions of most concepts (e.g., "plant" and "chair") can only be established in an unsystematic and experimental way. One compares piecemeal the consequences of a definition with our uses of the concept; no a priori condition of adequacy can be laid down in advance. Tarski showed, however, that a plausible and relatively precise material adequacy condition can, surprisingly, be laid down for the definition of truth. This adequacy condition is Convention T. Convention Τ enables Tarski to reduce a philosophical problem - and a seemingly intractable one at that - to a mathematical one. And it enables him to construct a proof that his definition is satisfactory. Convention Τ contains Tarski's principal philosophical claim. If it is granted, there can be little doubt about the philosophical merits of Tarski's definition. Convention Τ is plausible and has been highly influential in philosophy. There is, however, a fairly straightforward argument that refutes it - an argument that I have been unable to prove unsound. My aim here is to present this argument. Towards the end of the paper I shall state the insights in Convention Τ that are left untouched by the refuting argument.

2. Statement of Convention Τ Convention Τ makes, roughly, the following claim: Let D be a proposal for the definition of truth for a language L ("the object language"). Let us call a sentence a T-biconditional for L iff it is of the form (T), (Τ) X is true in L i f f p , where X is a perspicuous name of a sentence of L and p is a translation of that sentence into English. 3 Then, D is a materially adequate definition of truth if, and only

3

Tarski gives different accounts of the names that may replace 'X in the form (T). In CTFL, he requires that they be structural-descriptive names (p. 188). In SCT, he allows the T-biconditionals to be formed using all names, simple or compound (p. 16). The proposal of SCT can be shown to be unacceptable. Let φ be an arbitrary true sentence of English. Let a be a name of a tautology of the object language L and b a name of its negation. Let ψ be a translation of this tautology into English, and not-1// a translation of the negation. Finally, let t be the following complex term: the unique object χ such that [(χ = a & φ) or (x = b Sc not -φ)]. The SCT proposal counts the following as T-biconditionals of L. t is true iff ψ a is true iff ψ. b is true iff not- ψ. These equivalences logically imply that t and a are true and that b is not true. But the logic of definite descriptions yields that

An Argument Against Tarski's Convention Τ

227

if, D implies (i) all the T-biconditionals for L and (ii) only sentences of L are true in

r.4

Convention T, as stated above, is not general; it should be applied only in a restricted range of situations. Most obviously, it should be applied only to definitions of 'true in Ü that are formulated in English. Furthermore, it should be applied only when the object language L satisfies the following conditions: (a) L contains none of the complexities of indexicals, ambiguity, self-referential truth, etc.; (b) L contains no sentences that are untranslatable into English; and (c) L contains no sentences that lack perspicuous names in English. For languages that fall outside these restrictions, Convention T, as formulated above, does not provide a proper adequacy condition for the definition of truth. A significant effort has been devoted in the philosophy of language to formulating more widely applicable versions of Convention T, but the effort has yielded only limited success. Typically the newer formulations gain generality at the expense of precision. Convention T, as stated above, appears to be subject to a simple counterexample. Consider the following proposal for the definition of 'true in L\ χ is true in Ζ iff χ is not true in L. T h e definition is not materially adequate, but it appears to be inconsistent and, hence, to satisfy trivially the demands imposed by Convention T. For our purposes, the simplest way of dealing with this problem is to restrict Convention Τ to non-circular definitions.^ The argument to be presented remains unaffected by the restriction. t = a or t = b, and t = a iff 0.

4

5

It follows that the three T-biconditionals jointly imply φ. Hence, any definition that implies all the Tbiconditionals will have to imply all the expressible truths. T h e account of the T-biconditionals in SCT is therefore much too liberal. A doubt is possible about the proposal of CTFL also. Why should structural-descriptive names occupy a special position in the material adequacy condition on a definition of truth? I have avoided this doubt by requiring that the T-biconditionals contain "perspicuous" names. Intuitively, these are names whose sense can be identified with their reference (e.g., quotation names). Only for such names can one say, asTarski says, that the logical relation between 'Λ'is true' and '/>' is one of equivalence (SCT, p. 16). In Tarski's own explicit formulation, Convention Τ states only a sufficient condition on a materially adequate definition of truth, not a necessary condition. There are reasons, however, for working with the stronger reading given in the text, (i) Tarski himself says in both the cited essays that a definition of truth must imply the T-biconditionals, and that the definition "has to be, in a certain sense, a logical conjunction o f " the biconditionals {CTFL, p. 187; SCT, pp. 15-16). (ii) Tarski needs, in order to achieve his goals, a necessary condition on a definition of truth. Tarski wants to show not only that certain definitions of truth are materially adequate, but also that under certain conditions materially adequate definitions of truth cannot be given, (iii) In the literature on the subject, Convention Τ is often taken to state both a necessary and a sufficient condition. See, for instance, Rudolf Carnap, Introduction to Semantics (Carnap 1942, pp. 26-28); Donald Davidson, Inquiries into Truth and Interpretation (Davidson 1984, pp. 23 & 66); Richard L. Kirkham, Theories of Truth·. A Critical Introduction (Kirkham 1992, p. 144); and R. J. Nelson, "Proxy functions, Truth and Reference" (Nelson 1997). I believe that this restriction is consistent with Tarski's intentions. Let me add that in my view the above objection is unsound. I do not think that the proposed definition satisfies the requirements of Convention T. But, as I do not want to argue this here, I have let the objection stand and have followed a simple way out. T h e interested reader can find an exposition of the theory of definitions that justifies my view in The Revision Theory of Truth (Gupta and Belnap 1993).

228

Tarski Challenged

Convention Τ, as stated above, is not fully precise; it does not delineate the notion of implication that it invokes. The notion plainly cannot be that of logical implication. For that makes Convention Τ much too strict: even Tarski's own definition (in CTFL) fails to be materially adequate. The relevant notion of implication is, therefore, one that allows the use of some nonlogical resources. But what resources are to be allowed - analytic truths, mathematical truths, or something else? Tarski himself does not provide, as far as I know, an explicit answer to this question. In practice he allows himself the use of limited syntactic information about the object language and some simple principles of higher-order logic (equivalently, set theory). The argument below is built on some minimal assumptions about implication. It assumes that at least one true sentence - say 'It rained in Bloomington, Indiana, on 22 July 1997', or some such - may not be used in deriving consequences from a definition of truth. More precisely, let Γ ζ be the set of sentences that may be used in deriving consequences from a definition of 'true in L'. Then Γ^ will be assumed to be regular in the following sense. Γ^ is regular (for L) iff there is a true sentence Ρ such that Γ^ does not logically imply Ρ and, furthermore, 'true in L' occurs neither in Γ^ nor in P. One part of the argument below will rest on another assumption about Γ^ (its nonvacuity). This will be introduced later. Let us say that a definition D satisfies the Tarski condition for L i f f D logically implies, in conjunction with Γ^, (i) all the T-biconditionals for L and (ii) 'only sentences of L are true in Z'. 6 We can now state a more precise version of Convention T. This is the version that the argument below will attempt to refute. Let £ be a language that meets the three conditions, (a) - (c), stated above. Then, Tarski's Convention Τ makes the following claim. (*) For all non-circular definitions D of'true in L\ D is a materially adequate definition of 'true in L' iff D satisfies the Tarski condition for L?

6

7

Some imprecision remains in this definition. The notion of logical consequence is not fully clear and precise in its application to English. However, the fact that we are allowing the character of Γ^ to be so open provides a way out of the difficulty. We can read 'logical consequence' in a precise way - say as, 'classical first-order consequence' - and we can require that all other statements that are needed for deriving consequences from D be included in Γ^. Note that we shall work with a "tidied up" fragment of English, one free from ambiguity, amphiboly, etc. Note that, in the Tarski condition, the set Γ^ of sentences that may be used in deriving consequences from a definition D depends on the language L but not on the definition D. If Γ^ were allowed to depend on D, Convention Τ would reduce to a triviality, one incapable of serving as a touchstone for definitions of truth. For consider an arbitrary non-circular definition D and let its definiens be ψ(ζ). Either D is materially adequate or it is not. If D is materially adequate, we can set Γ^ so that it contains the sentence For all z, if ψ ( ζ ) then ζ is a sentence of L and all the biconditional of the form, ψΡΟ iff/. where X is a perspicuous name of a sentence of the object language and ρ is one of its translations into English. The condition imposed by Convention Τ holds. (Note that the resulting Γ^ is bound to be

An Argument Against Tarski's Convention Τ

The argument below will show that, for some languages L, (*) is false if tain minimal conditions.

229

meets cer-

3. The Argument A materially adequate definition aims, Tarski says, to "catch hold of the actual meaning" of the term being defined (i.e., the άφηίεηάηηΐ). This characterization contains an ambiguity — an ambiguity that it inherits from the notion of meaning. The aim of the definition may be to capture the extension or the intension or the sense of the definiendum. Let us call a definition of a one-place predicate G extensionally adequate (respectively, intensionally adequate and sense-adequate) iff it succeeds in capturing the extension (respectively, the intension and the sense) of G. Extensional adequacy requires that the extension of the definiendum G be the same as that of the definiens, that all and only things that are G satisfy the definiens. Intensional adequacy is a stricter requirement. The definition has to be extensionally adequate, not just contingently but necessarily; in other words, the extension of the definiendum has to coincide with that of the definiens in all possible situations. The definition For all objects z, [z is a planet of the Sun iff ζ = Mercury or ζ = Venus or ... or ζ = Pluto], for example, is extensionally adequate. But it is not intensionally adequate, for in a possible situation in which Pluto is not a planet of the Sun, the definiendum and the definiens do not have the same extension. The requirement of sense-adequacy is even stricter than that of intensional adequacy. Sense of an expression is that which one grasps when one fully understands the expression. Sameness of sense implies sameness of intension,8 but expressions with the same intension — for example, 'prime number between 6 and 10' and 'identical to 7' — may differ in sense. The concepts needed to fully grasp 'identical to Τ are not the same as those needed to fully grasp 'prime number between 6 and 10'; hence the two expressions express different senses. The notion of sense, it must be said, is not as clear as that of extension and intension. I shall make explicit the claims about it that my argument uses. I can now present the argument that Convention Τ fails to provide a satisfactory criterion with respect to all three standards: extensional adequacy, intensional adequacy, and sense-adequacy. Let L be a first-order language with identity that meets the three conditions, (a)-(c), stated in section 2. It will prove convenient if we assume that L is normal in the sense that L satisfies not only (a)-(c) but also two further conditions:

8

regular, if the translation of L into English is effective.) On the other hand, if D is materially inadequate, we can let Γ be the null set. Now the condition imposed by Convention Τ fails. In either case, Convention Τ holds trivially. (In these remarks, I am equating the notion of material adequacy with the notion of extensional adequacy explained below.) I am ignoring indexical expressions.

230

Tarski Challenged

(d) L has proper names for the numbers 0 and 1 and (e) the nonlogicaJ constants of L are translatable into constants of English. (These latter assumptions will come into play only in the third part of the argument. And, if the length of argument were not a limiting factor, these assumptions could have been weakened.) Set D to be the following non-circular definition of 'true in L\ {D) For all objects ζ, ζ is true in L iff ψ(ζ) , and let us consider the three readings of 'material adequacy' one by one. Case 1. Extensional adequacy. Suppose D is an extensionally adequate definition of truth for L? That is, all and only the true sentences of L satisfy the definiens, ψ (ζ), of D. Now, since Γ^ is regular, there is a true sentence, say P, such that Γ^ does not logically imply Ρ and, furthermore, 'true in L' occurs neither in Y L nor in P. Since Ρ is true, the formula, If Ρ then

ψ(ζ),10

is also true precisely of the true sentences of L. It follows that the definition D\ (D') For all objects ζ , [ ζ is true in L iff (if Ρ then ψ (ζ))], is also an extensionally adequate non-circular definition of 'true in L\ But, as we now show, Γ^ and D' do not logically imply all the T-biconditionals for L. Suppose, for reductio, that they do. It follows that YL and D' logically imply that not everything is true in L, for the T-biconditionals logically imply that a contradiction is untrue. But in virtue of the character of D\ we have also that Γ ζ , D' and not-P logically imply that everything is true in L. That is, TL, D and not-P are logically inconsistent. Hence, Γ^ and D' logically imply P. But, given our hypotheses, this is impossible. For we know that Γ^ does not imply P. So there is an interpretation on which all members of Γ^ are true and Ρ is false. If we assign to the predicate 'true in V the set of all objects, we obtain an interpretation on which D' and all members of T L are true but Ρ is false. This completes the reductio. We have established that a definition that is extensionally adequate may fail to imply the T-biconditionals. Convention Τ is thus not a satisfactory standard for evaluating definitions for extensional adequacy. Case 2. Intensional adequacy. The argument here is parallel to the previous one. Consider the "modalization," Actually-P, of the sentence P. 1 1 Since Ρ is true in the ac-

9 10 11

Tarski has provided a method for constructing extensionally adequate definitions of truth for first-order languages. The conditional employed here is the material conditional. The semantics of the operator 'Actually is as follows. Actually-φ is true at a possible world w iff φ is true in the actual world. So, if φ is true in the actual world, then Actually-φ is true in all possible worlds, i.e., Actually-φ is necessarily true.

An Argument Against Tarski's Convention Τ

231

tuaJ world, Actually-P is true in all possible worlds. Hence, if D is intensionally adequate - i.e., if in each possible world all and only the true sentences of L satisfy ψ(ζ) then the same must hold for if Actually-P then ψ(ζ). So, the non-circular definition (.D") For all objects ζ, [ζ is true in L iff (if Actually-P then ψ(ζ ))] is also intensionally adequate. But an argument parallel to the one given above shows that D" does not imply ail the T-biconditionals. 12 Convention Τ is thus not a satisfactory standard for evaluating definitions for intensional adequacy. Case 3. Sense-adequacy. Convention Τ is commonly supposed to be most plausible when it is read as laying down a criterion for sense-adequacy of a definition of truth. And it is on this reading that Convention Τ has had its greatest philosophical influence. It has lent support to deflationism, the view that truth cannot serve a substantial philosophical function. The role of truth, according to deflationism, is purely logical; it is to provide a device for expressing certain sorts of generalizations that would otherwise be difficult or impossible to express. 13 The thought underlying this reading of Convention Τ is that sentences such as "snow is white' is true' and 'snow is white' have the same content - they say the same thing — , and that they do so solely in virtue of the sense of 'true'. Hence, the thought continues, any satisfactory explanation of the sense of 'true' must imply the equivalence of "snow is white' is true' and 'snow is white' - and other similar equivalences (i.e., the T-biconditionals). This reasoning appears compelling, but I suspect that its force issues not from its intrinsic merits but from the obscurity and ambiguity of the notion of sense. Let us make explicit the "sense-adequacy" reading of (*): (**) For all non-circular definitions D of'true in L\ D is a sense-adequate definition of 'true in Ü iff D satisfies the Tarski condition for L. And let us begin by noting that (**) cannot be true if "sense" is understood, as it sometimes is, in a fine-grained way. In particular, (**) fails if sense is sensitive to logical structure. Let φ and φ' be logically equivalent formulas that differ in sense and that do not contain any occurrences of'true in ΖΛ Suppose that D satisfies the Tarski condition for L. And let D' and D" be definitions like D but with the definientia, respectively, ψ(ζ)δί(φ

iff φ),

and ψ(ζ) & (φ iff φ').

12 13

Note that the equivalence of Ρ and 'Actually-P' holds in virtue of the logic of 'Actually. I provide an exposition, and some criticisms, of deflationism in my "A Critique of Deflationism" (Gupta 1993).

232

Tarski Challenged

The definitions D, D\ and D" are logically equivalent. So D' and D" also satisfy the Tarski condition for L. Hence, according to (**), both D' and D" are sense-adequate. But, on the conception of sense under discussion, this is impossible: the definientia of the two definitions must differ in sense. Note that in this argument - as well as in the argument below - a new constraint on Γ^ comes into play. The argument assumes that Γ^ is not vacuous in the sense that Γ^ is strong enough to allow at least one definition to satisfy the Tarski condition for L. Let us now turn to a different notion of sense, one that is intermediate between the fine-grained notion just considered and the notion of intension. How might we individuate senses of this intermediate grade? We must allow arbitrarily complex expressions to have the same sense, so we cannot individuate using logical structure. And we must distinguish between the senses of intensionally equivalent expressions, so we cannot individuate solely on the basis of the intensions of the constituents. The only way left open is to appeal to concepts that are expressed or brought into play by the constituents. But this raises the difficulty that the notions of "concept," "expression," etc. are as obscure as the notion of sense. Fortunately, the argument against Convention Τ relies not on a particular detailed account of these notions but only on four minimal constraints, which I state below. Let us take as primitive the notion "the ideology of an expression X." Intuitively and roughly, the ideology of an expression may be thought of as a set that contains the primitive concepts that the expression brings into play. We shall assume nothing about this notion other than the constraints given below. These constraints leave open, for example, whether logically equivalent expressions have the same ideology, whether the ideology of a nonlogical constant is a set with one or with many members, and whether distinct nonlogical constants have distinct ideologies. Let us define the ideology of a language to be the sum of the ideologies of its nonlogical constants, and the ideology attributed by a definition to its definiendum to be the ideology of the definiens. We can now state the first constraint. Constraint A. All sense-adequate definitions of a definiendum G attribute the same ideology to G. Constraint A imposes a minimal condition on sense-adequacy. It allows two sense-adequate definitions of G to differ in logical complexity. It also allows them to contain entirely different nonlogical constants. All that Constraint A requires is that the ideologies of the definientia of sense-adequate definitions be the same. Let us say that a nonlogical constant Κ occurs essentially in a formula φ iff AT occurs in every formula that is logically equivalent to φ, and let us say that Κ occurs essentially in a set of formulas iff AT occurs essentially in one of the members of the set. The next two constraints are as follows. Constraint Β. If a nonlogical constant A'occurs essentially in φ then the ideology of Κ is included in the ideology of 0. 14 14

A significantly weaker version of Constraint Β suffices for an argument against Convention T: we can restrict φ in Constraint Β to range over formulas that contain no occurrences of defined expressions. Since this weakening complicates the argument, I work with the version given in the text.

An Argument Against Tarski s Convention Τ

233

Constraint C. Nonlogical constants that are translations of each other have the same ideology. Note that if the ideology of a nonlogical constant Κ is identified with the set containing the concept expressed by K, then Constraint C reduces to the familiar idea that nonlogical constants that are translations of each other express the same concept. We can now turn to the argument that (**) is false. Set ( t ) be the right-to-left part of (**). (t) For all non-circular definitions D of 'true in L\ if D satisfies the Tarski condition for L, then D is a sense-adequate definition of'true in L\ Three little lemmas will help reduce ( f ) to an absurdity. The First Lemma. Suppose ( f ) holds and consider a definition D that satisfies the Tarski condition for L. Now the ideology of every nonlogical constant that occurs essentially in Γ^ is included in the ideology of the definiens, ψ(ζ), of D. Proof. Let φ be an arbitrary member of Γ^ and let A" be a nonlogical constant that occurs essentially in φ. Observe that /if must occur essentially in either (1) or (2).

(1) (φ&ψ(ζ)). (2) If φ then

ψ(ζ)).

For suppose otherwise. Then there are formulas logically equivalent to (1) and (2) say, and θ 2 respectively - that contain no occurrences of K. It follows that the formula or not -θ2, which is logically equivalent to φ, also contains no occurrences of K. This contradicts the hypothesis that Κ occurs essentially in φ . Now consider definitions Dj and D2 that are just like D except that their definientia are, respectively, (1) and (2). These two definitions are, in the presence of Γ^, equivalent to D. Hence, they also satisfy the Tarski condition for L. Since D is non-circular and since 'true in L' does not occur in Γ^, the definitions Dj and D2 are non-circular also. Our hypothesis (t) yields that D^ and D 2 are sense-adequate. By Constraint A, the definitions D, Dv and D2 all attribute the same ideology to 'true in L\ That is, the ideology of ψ(ζ) = the ideology of (1) = the ideology of (2). But Constraint Β yields that the ideology of A'is included in either the ideology of (1) or of (2). So the ideology of Κ must be included in the ideology of ψ{ζ). Q.E.D. The Second Lemma. If a definition D satisfies the Tarski condition for a normal language L then every translation of every nonlogical constant of L occurs essentially either in Γ^ or in the definiens, of D. Proof. Let us establish the lemma for H, a translation of a one-place predicate G of L. The argument is easily generalized to function symbols, names, and other predicates of

234

Tarski Challenged

Ζ. 15 Let Q be a sentence of L that translates Ό is H' and let c be a perspicuous name of Q. (By the normality of L such a Q does exist.) So the following is a T-biconditional for L: (3) c is true in L iflf 0 is H. Now suppose, for reductio, that Η occurs essentially neither in Γ^ nor in ψ{ζ). Consider an interpretation Μ that makes D and all the sentences in Γ^ true. (Since Γ^ is consistent and does not contain any occurrences of 'true in L\ there is bound to be such an interpretation.) Suppose Q falls in the extension of 'true in U in M. Consider a variant interpretation M' in which Η is assigned an empty extension. Since //occurs essentially neither in Γ^ nor in D, M' preserves the truth of D and of all members of T L . But the biconditional (3) is false in M', thus violating our hypothesis that Γ^ and D logically imply all the T-biconditionals. A parallel argument yields a reductio for the other case, namely, that Q does not fall in the extension of'true in L' in M. Q.E.D. The Third Lemma. Suppose (t) holds and suppose that the definition D satisfies the Tarski condition for a normal language L. Then the ideology D attributes to 'true in L' includes the ideology of L. Proof. Suppose ( t ) holds and that D satisfies the Tarski condition for L. Consider an arbitrary constant K* of L and let Κ be its translation into English. The Second Lemma implies that Κ occurs essentially in either a member of Γ^ or in the definiens ψ (ζ) of D. If Κ occurs essentially in a member of Γ^ then, by the First Lemma, the ideology of /Tis included in the ideology of ψ ( ζ ) . On the other hand, if Κ occurs essentially in ψ(ζ) then, by Constraint B, again the ideology of /Tis included in the ideology of ψ(ζ). So, in either case, the ideology of Κ is included in that of ψ(ζ). But, by Constraint C, the ideology of K* is identical to the ideology of K. So the ideology of K* must be included in the ideology of ψ(ζ). Q.E.D. We have arrived at the final step of the reductio. Convention Τ implies that a definition that satisfies the Tarski condition for L is bound to attribute to truth an ideology at least as large as that of L. But for some languages L this ideology is much too large. A grasp of the sense of 'true in IS does not require a grasp of all of the concepts in the ideology of L. (Example: One can grasp the concept "true sentence in the language of physics" without grasping all the primitive concepts of physics.) It follows that the satisfaction of the demands of Convention Τ is no proof that the definition is sense adequate. We are appealing here to another constraint on sense. Let us make it explicit. Constraint D. For some normal languages L, the ideology of L is not included in the ideology that a sense-adequate definition attributes to 'true in L\ The argument above shows that Convention Τ does not provide a correct sufficient condition for sense-adequacy. Might it provide a correct necessary condition? That is, might the left-to-right part of (**) - i.e., (φ) - be true?

15

For instance, for a proper name t distinct from Ό', set Q in the argument below to be a translation of V=0'; otherwise set Q to be a translation of 't= 1'.

An Argument Against Tarski s Convention Τ

235

(φ) For all non-circular definitions D of'true in L\ if D is a sense-adequate definition of 'true in U then D satisfies the Tarski condition for L. Statement (φ) might be true — and if it isn't true, it can be made true through a suitable choice of Γ^. If there are no sense-adequate definitions of truth, then (φ) holds vacuously. Furthermore, if there are sense-adequate definitions, we can ensure the truth of (φ) by making Γ^ suitably large.16 But in both these cases, (φ) is a triviality, not a substantial constraint on a definition of truth. On a natural construal of Γ^, there are reasons for thinking that (φ) does not provide a satisfactory constraint on a definition of truth. For, on a natural construal, the ideology of Γ^ - i.e., the sum of the ideologies of the members of Γ^ - is limited. (Note that the IT^s employed by Tarski have a limited ideology.) By the Second Lemma, a definition that satisfies the Tarski condition must attribute to 'true in Z' at least that part of the ideology of L that is disjoint from the ideology of Γ^. So, if the ideology of L is large, the definition attributes a large ideology to 'true in Π and hence cannot be sense-adequate. Consequently, the Tarski condition is not a satisfactory requirement on sense-adequate definitions of truth. Indeed, satisfaction of the Tarski condition is sometimes good evidence that the definition is not sense-adequate. The argument above sets, at the very least, a challenge to a defender of Convention T: articulate a plausible notion of sense for which Convention Τ holds. The notion will have to violate, the argument shows, at least one of the constraints A - D. Since the constraints are weak, the prospects of articulating a suitable notion of sense are not bright. What makes Convention Τ seem plausible for sense-adequacy, we noted above, is the thought that the two sides of the T-biconditionals have the same content, and that they do so in virtue of the sense of'true'. The first part of this motivating idea can be accepted: there is a notion of content on which the two sides of the biconditionals do have the same content. What should be rejected is the idea that the sameness of content issues from the sense of'true'.

4. Concluding Remarks Our argument shows that Convention Τ fails to provide a necessary condition for the intensional adequacy of definitions of truth (and hence also for their extensional adequacy). The argument shows, furthermore, that Convention Τ fails to provide a sufficient condition (and also a satisfactory necessary condition) for the sense adequacy of definitions of truth. There is, however, an important insight in Convention Τ that is left intact by the argument: the T-biconditionals of a language L (in conjunction with 'only sentences of L are true in L) fix the extension and the intension of'true in Z'. 17

16 We can construct Γ^ using the method given in note 7. 17 This claim does not hold unrestrictedly; it requires some important qualifications. See Chapters 1 & 4 of Gupta and Belnap 1993.

236

Tarski Challenged

Hence Convention Τ formulates a correct sufficient condition for extensional and for intensional adequacy. But Convention Τ provides no basis for claims about the sense of 'true'. Tarski's work on the concept of truth has inspired two conflicting developments in logic and philosophy. First, his recursive method of constructing definitions of truth has inspired truth-conditional semantics. Tarski's method rests upon a construction in which semantic properties of compounds (e.g., truth-conditions) are explained systematically in terms of the semantic properties of the components. Tarski worked out this construction only for relatively simple formal languages. But he inspired others to try to extend it to richer formal languages — and even to natural languages. The semantic programs of Richard Montague and Donald Davidson have their roots in this work of Tarski's. Note that neither the truth-conditional semantics nor Tarski's contribution to it depends upon Convention T. Even if Convention Τ is shown to be false on all of its readings, truth-conditional semantics and Tarski's contribution remain undisturbed.18 In particular, the validity of Tarski's recursive method is entirely independent of Convention T. The second development inspired by Tarski's work is deflationism. And Convention Τ - more particularly, its sense-adequacy reading - is the bridge that connects deflationism with Tarski's work. Convention Τ (on its sense-adequacy reading) implies that Tarski's definition succeeds in capturing the sense of 'true'. And this supports the deflationary idea that truth is a lightweight concept, one incapble of serving a substantial semantic and philosophical function. 19 Our argument against Convention Τ undermines this support for deflationism. 20 It leaves untouched, however, Tarski's contribution to truth-conditional semantics, which, in my opinion, is the greater and the more permanent legacy of Tarski's work. 21

18

19 20

21

Plainly Convention Τ is irrelevant to Montague's model-theoretic semantics. The latter is based on the relativized notion "truth in a model," while the former concerns an absolute notion of truth. Davidsons semantics, though based on an absolute notion, is also independent of Convention T. Davidson appeals much to Convention Τ in his work. His semantic program, however, rests not on Convention Τ but on a radical modification of it — a modification that is far in substance and spirit from the original. See Davidson 1 9 8 4 , especially p. xiv. The weaker readings of Convention Τ (namely the extensional-adequacy and the intensional-adequacy readings) are insufficient to establish deflationary ideas. See Gupta 1993. The argument does not refute deflationism, however, since Tarskian truth-definitions are not the only route to deflationary conclusions. Christopher Hill's "Truth in the Realm of Thoughts" (forthcoming) provides one example of a deflationary theory that is independent of Tarskian truth definitions and of Convention T. I wish to thank A n d r i Chapuis, Christopher Hill, Ray Jackendoff, Byeong Deok Lee, and Michael Lynch for their comments on this paper.

An Argument Against Tarski's Convention Τ

237

References Carnap, Rudolf 1942. Introduction to Semantics. Harvard University Press, Cambridge, MA. Davidson, Donald 1984. Inquiries into Truth and Interpretation. Oxford University Press, New York. Gupta, Anil 1993. A Critique of Deflationism. Philosophical Topics, 21:57-81. Gupta, Anil and Belnap, Nuel 1993. The Revision Theory of Truth. The M I T Press, Cambridge, MA. Kirkham, Richard 1992. Theories of Truth: A Critical Introduction. The M I T Press, Cambridge, MA. Nelson, R. J. 1997. Proxy Functions, Truth and Reference. Synthese, 111:73-95. Tarski, Alfred 1935. Der Wahrheitsbegriff in den formalisierten Sprachen. Studia Philosophica, 1:261-405. Translated by J. H. Woodger, "The Concept of Truth in Formalized Languages," in Logic, Semantics, Metamathematics·. Papers from 1923 to 1938, Oxford: Clarendon Press, 1956; second edition, Indianapolis: Hackett Publishing Company, 1983, ed. J. Corcoran, pp. 152-278. Tarski, Alfred 1944. The Semantic Conception of Truth. Philosophy and Phenomenological Research, 4:341376. Reprinted in Linsky (ed.), Semantics and the Philosophy of Language, Urbana, IL: University of Illinois Press, 1952, pp. 13-47.

What is Truth? Stay for an Answer JAAKKO H I N T I K K A

"What is truth?" said jesting Pilate and would not stay for an answer. (Francis Bacon, O n Truth") What is truth? How should you respond to the tide question of this volume? If your name is not Pilate or Richard Rorty and you are not jesting, the obvious response presumably it to try to define truth. But such a definition is impossible, you are likely to object, in any philosophically important sense. Isn't this precisely what Tarski showed by his famous impossibility theorem? (Tarski 1956, pp. 246-266.) Tarski showed how truth-definitions can be given for first-order languages, but he also showed that such definitions can only be formulated in a richer metalanguage. Now a philosophically interesting truth-definition should certainly be applicable to our actual working language, or at least to representative approximations to it. But there is no richer metalanguage that we could resort to over and above the language we actually use. Hence truth-definitions are impossible to all actual language — "the only language that I understand", as Wittgenstein might have put it. Moreover, Tarski showed - didn't he? - that ordinary ("colloquial" to use Tarski's word) language is so irregular that no coherent concept of truth can be formulated for it and used in it. (Tarski 1956, pp. 154-166.) Hence there is for this reason too, no hope of defining truth for our actual working language. Fortunately or unfortunately, these widespread views are radical misrepresentations of what Tarski's results and arguments actually amount to. (Cf. Hintikka 1996(a), ch. 5; Hintikka and Sandu, 1997.) Tarski proved the impossibility of a self-applied truthdefinition only for such languages as satisfy certain further conditions. These conditions are not trivial. Indeed, they apply to ordinary first-order languages only because their logic, the Frege-Russell first-order logic, is seriously flawed. Here we come to the insight that for all its basic simplicity is revolutionizing all discussion of truth and truth-definitions. (Cf. here Hintikka 1998.) Its starting-point is one of the most basic requirements that one has to impose on any fully adequate representational language. Whatever else a fact-stating language does, it ought to be capable of expressing any possible complex of dependencies and independencies between different variables. Now in a language like our usual logical languages, dependencies between variables are expressed by means of dependencies between the quantifiers to which the variables are bound. For one simple example, the fact that a sentence of the form (1) (V*)(3y)S[*,y] expresses that the truth-making choice of the value o i y depends on the choice of the value of χ is shown by the fact that the quantifier (3j) depends on (Vx) or is "in its scope" as the usual (but dangerous) term goes.

What is Truth? Stay for an Answer

239

Hence an adequate logical language should be capable of expressing all possible configurations of dependence and independence between quantifiers. But the received (Frege-Russell) first-order logic does not do so. There are forms of independence and forms of dependence that are inexpressible by means of ordinary first-order logic. (For examples, see below.) In the interests of historical accuracy, perhaps I should speak here of the defects of the Frege-Russell logic of quantifiers, instead of first-order logic, which was separated from the general logic of quantifiers of any logical type only by Hilbert and Ackermann in 1928. But this is distinction without difference. The flaws I am pointing out are independent of the type of quantifiers involved, and in any case the logics of Frege and Russell are best viewed as many-sorted first-order logics. In any case, once this important defect of received first-order logic is perceived, it can also be seen how it can be eliminated by making the logical language in question more flexible (and hence more expressive). One way of doing so is to introduce on additional notation to exempt a quantifier, say (Q^y), from its dependence on another, say (Qjx), by writing it (Q^y/Qjx). Then we can express types of independence that were not expressible before, as e.g. (2) (Vx) (Vz) (3y/Vz) (3 uNx)S[x,y,z, u] We can likewise express new types of dependence, as in (3) (Vi)(Vx)(Vy)(BzfVx)(3uNy) ((x = z) & (y = u) & 5[i,x,_y]) Here the variables χ and y depend on each other symmetrically. Such dependence obviously cannot be expressed by any transitive ordering of quantifiers. The resulting firstorder logic has been called independence-friendly (IF) logic (cf. here Hintikka 1996(a), chapters 3-4; Hintikka and Sandu, 1996) in contradistinction to "ordinary" first-order logic. This terminology is biased, however, for IF first-order logic is simply the unrestricted first-order logic which does not deserve any qualifying epithets, unlike the "ordinary" first-order logic which hence ought to be referred to as dependence-handicapped logic. But what does this liberation of first-order logic have to do with truth-definitions? In order to answer this question, we can ask a further one: What prevented us from defining truth for an ordinary ("dependence-handicapped") first-order language in the same language? In order to formulate such a definition, we must of course first formulate the syntax of the language in itself. Let us assume, for the sake of an example, that this is done in an arithmetical language by means of the usual Gödel numbering technique. That technique amounts to speaking of numbers in two different ways: either as numbers simpliciter or as codifications of certain expressions of the same language. You can have such double-entendre language without any paradoxes. But one restraint is obvious. When you quantify over numbers in their two different roles, clearly the two quantifiers must be independent of each other. For the two quantifiers talk about altogether different things, they are to be interpreted differently. Now it is easily seen that you need such independent quantifiers for your self-applied truth-definitions. But they are not available in ordinary ("dependence-handicapped") first-order logic. This is the reason for Tarski s impossibility theorem.

240

Tarski Challenged

The other side of the coin is that as soon as you have at your disposal the requisite independent quantifiers, a truth-definition becomes feasible. Small wonder, therefore that an IF first-order language allows for the formulation of its own truth predicate. All that is needed is that the language is expressive enough to enable it to speak of its own syntax. This result deserves (and needs) a number of further comments and explanations. (i) But is there any reason to think that this kind of truth-definition can be extended to other languages, let alone all of them? (Actually, an extension to our actual working language should be enough.) Do I have a proof that such a truth-definition is possible even in all explicitly defined ("formal") languages? No, I do not, but such a proof would not be the most conclusive way of settling the issue anyway. In fact, if such a proof were presented to me, I would be highly suspicious. Such a proof would have to cover so many different and unlike cases that the implementation of a promised definition might in some cases end up so unnatural as to be unacceptable for collateral reasons. The situation might in other words be as in the well-known story about the man whose wishes began to become true, but always in such an unexpected way that they had disastrous consequences he had not foreseen. In fact, in another department of logico-semantical studies there is a theorem that says that any recursively definable language has a compositional semantics. (See Zazarozny 1994, Janssen 1997.) Yet Cameron and Hodges (forthcoming) have proved that there is not any compositional semantics for independence-friendly first-order languages which preserves the normal interpretation of quantifiers. (There is an interesting contrast here to Hodges (1997(a)-(b)).) What is much more convincing here is a look at the reasons why attempted internal truth-definitions are impossible for Frege-Russell ("dependence-handicapped") languages. The folk wisdom tells us that such definitions fail because these languages are too strong, so strong that paradoxes of the liar-type arise in them. What was seen above is that in a perfectly natural sense the opposite is true. Truth-definitions fail in "ordinary" first-order languages because they are too weak in their expressive power. They do not allow arbitrary patterns of dependence and independence. But as soon as this defect is corrected, there is every reason to expect that truth predicates can be formed in any language that is strong enough to allow arbitrary patterns of dependencies and independencies — which we must allow anyway in an adequate language. There is every reason to think that our actual working language - or at least the relevant descriptive fragments of it — is flexible enough in this respect. In fact, independent quantifiers are known to occur in natural languages like English. (Cf. e.g. Barwise 1979.) (ii) But didn't Tarski (1956, pp. 154-166) show that natural language is too messy structurally to allow for a genuine truth predicate? No, he did not prove it, he asserted it. There is no stronger a priori reason to believe Tarski here than when he asserted that the syntax of natural language is too messy ever to be spelled out. (Go tell that to Chomsky!) It can be shown (see Hintikka and Sandu 1997) that Tarski, like an overly puritanical moralist, is focusing too much on one particular alleged sin. For Tarski, the sin committed by the colloquial language is violation of compositionality, that is to say, violation of semantical context-independence. Tarski was motivated by the same meth-

What is Truth? Stay for an Answer

241

odological vision as categorial grammars which are little more than an apotheosis of compositionality. IF languages clearly violate compositionality. In them, the interpretation of a quantified formula can depend on its context, more specifically, depend on which outside quantifiers its own quantifiers depend or do not depend on. For this reason, IF firstorder logic is a resounding refutation of the very thesis Tarski was relying on, viz. the thesis that one cannot formulate an explicit truth predicate for a noncompositional language. Tarski apparently realized, correctly, that ordinary ("colloquial") language is not compositional. He was wrong in concluding from this valid insight that one cannot consistently use the concept of truth in natural languages (iii) But why should one believe that the truth predicates that can be formulated for IF first-order languages are truth predicates, that is, that they capture our pretheoretical notion of truth? Tarski s truth definitions have in fact been criticized in this way, that is to say, by alleging that all he does is to correlate certain sentences with certain facts, without showing that this correlation captures our "real" pretheoretical idea of truth. But what is this pretheoretical idea, anyway? Let us go back to (1). This sentence is true according to our commonsense conception of truth if and only if for any given value of χ there is a way of finding a value of y which satisfies S|xy]. Now for a mathematician or logician such a "way of finding" is nothing but an euphemism for a function / t h a t maps any value of χ to a suitable value of y = fix). In other words, according to our pretheoretical conception, (1) is equivalent with (4)

(3fi(Vx)S[x,fix)}

Here the function f can be described as a "witness function" in the same sense as an individual b such that S[£] can be said to be a "witness individual" attesting to the truth of the existential sentence (Ξχ)5[χ]. Likewise, the truth-conditions of (2)-(3) can be formulated as asserting the existence of their respective witness functions. In other words, the truth-conditions of (2)(3) are, respectively. (5) (3fi(3g)(Vx)(Vz)S[x,fix),

ζφ)]

and

(6) (By5(3^)(Vf)(Vx)(Vj) ((* = fiy,t)) & (y = gix.t)) & S[t,x,y\) A moments reflection will show you that (5)-(6) express what is meant by the truth of (2)-(3). Logicians have a name for such witness functions for a given sentence S. They are the Skolem functions of 5. What I am saying is that according to our commonsense (or common logic) conception of truth, the truth of a sentence S amounts to the existence of its Skolem functions. Maybe this is not obvious to you, but if so, I will ask you to re-enact the old story: Rush to another room to contemplate the matter, and I am confident that after fifteen minutes you are ready to return and triumphantly claim, "Yes, it was obvious!". Such truth-conditions as (4)-(6) are second-order formulas and are therefore not directly suitable for the construction of a truth predicate for a first-order language, IF

242

Tarski Challenged

or otherwise. But they belong to that fragment of second-order logic, known as the Σ} (sigma one-one) fragment, that is translatable into IF first-order logic. (See Walkoe 1970.) This translatability enables us to express truth-conditions like (4)-(6) in the very same IF first-order language for whose sentences they are truth-conditions of. And the truth predicate itself is nothing but a result of integrating these truth-conditions into a single predicate of the Gödel numbers of the sentence of that language. Accordingly, the truth predicate for a suitable IF first-order language in that language itself is little more than a way of spelling out the idea of the truth of a quantificational sentence as the existence of its Skolem functions. And it was shown above that the existence of such functions is what our pretheoretical ("intuitive") conception of truth amounts to in the first place. Hence a truth predicate that asserts the existence of the Skolem functions for a sentence is not subject to the same criticisms (no matter whether they are fair or unfair) as Tarski's definition. On the contrary, this truth predicate can be thought of as an explication and systematization of our very own substantive idea of truth. This is illustrated indirectly by what happens when someone tries to modify the notion of truth that is explicated by my truth predicate. (Cf. Hintikka 1996(a), ch. 10.) For instance, a constructivist might require that the witness functions (Skolem functions) must be constructive in some suitable sense. In other words, in the truthconditions like (4)-(6) the function quantifiers must be restricted to constructive functions as their values. But if so, the imaginary constructivist must have changed his or her interpretation of quantifiers already in the sentences whose truth conditions we are looking for. And the re-interpreted truth-conditions will then serve to spell out what this new interpretation of his or her first-order quantifiers is. In other words, if a constructivist consistently changes the interpretation of both his first-order and secondorder quantifiers, as he or she ought to do, the same truth-conditions remain applicable. (iv) But what happens to the different approaches to truth and theories of truth in the light of the new developments? This question is too large to be dealt with adequately here. I can make only a few brief remarks. First, a truth predicate defined along the lines outlined here is in some ways in keeping with the way of thinking of the minimalist and disquotational approaches to truth. In a way, I have turned merely the T-schema into a truth predicate. The joker in the pack of disquotationalists is that if they had tried to carry out their program explicitly, they would have run into the same problems as the users of Tarski-type truthdefinitions. They, too, would have needed IF logic to carry out their project. Once IF logic is available, I can graciously admit that they have been right in a sufficiently abstract (albeit underdeveloped) sense. Second, and most importantly, all theorizing (or, rather, philosophizing) that has relied on - or has been aided and abetted by - the alleged inexpressibility of truth has to be re-evaluated and often rejected. This puts into the same dock such otherwise unlike defendants as Wittgenstein, Heidegger and Quine. (Cf. here Hintikka 1990.) Third, much - maybe most - of the traditional discussion of truth has been marred by a failure to distinguish from each other questions about truth and questions about

What is Truth? Stay for an Answer

243

our knowledge of truth. These latter questions nevertheless turn out to be much subtler than what has ever been realized before. Hence a mere distinction between epistemic and nonepistemic frameworks is not enough. Even this particular confusion is too pervasive and widespread to be analyzed (deconstructed?) here. What I can do is to point to the subtler sources of this confusion in the very logic of truth that I have been expounding. Besides serving as steps toward truth-conditions, such second-order translations of first-order sentences as (4)-(6) bring out another aspect of the theoretical situation. In asserting a sentence like say (1), what am I doing? In order for (1) to be true, its Skolem function or functions must exist. Hence, in asserting (1), i.e. asserting that it is true, I assert the existence of its Skolem functions. (In this case, there is only one function.) But that is strictly (semantically) speaking all I do. I do not have to divulge to my audience any particular Skolem function. I need not myself know any such Skolem functions. It is not even necessary for anyone in the world to know any specific witness functions in order for my assertion to be understood. In asserting a sentence like one of (l)-(3), I am making a purely existential statement. I am merely asserting the existence of certain kinds of mathematical functions in whatever Platonic heaven such abstract entities might enjoy or suffer their existence. This result shows a remarkable fact about our fact-stating discourse in general, at least insofar as it uses quantifiers. Such discourse in effect relies on statements about the existence of Skolem functions rather than about any knowledge of such functions themselves. As I might put it, our fact-stating discourse conducts its business by means of promissory notes rather than cash transactions. But this cannot be the whole story either in business or in logical semantics. Often my communication partner is not only interested in the existence of the verificatory Skolem functions. He or she wants my promissory note to be cashed in, in other words, wants to know what the relevant witness (Skolem) functions are. In still other words, he or she might want to know how to actually play a verification game (semantical game) so as to be sure of winning. Such information cannot be conveyed by the literal meaning of quantificational sentences. But since it can be important for the recipient of my message, in ordinary discourse there are ways of indicating what the Skolem functions of the sentence might be that I am asserting. How such information is actually coded in colloquial discourse is not the crucial question here. What concerns the logical form of this extra information which I once called strategic information as distinguished from literal or abstract information, in other words, between strategic meaning and literal meaning. This does not transform into a distinction between two kinds of truth, however, except in a dangerous secondary sense. Strictly speaking, the distinction is between the truth simpliciter of the same sentence under two different interpretations. With this proviso, we can nevertheless speak of two uses of truth, strategic and abstract. Using such a terminology, one can say that very often what is meant by the truth of a sentence S of ordinary discourse is not its literal (ordinary) truth (e.g. the existence of the Skolem functions of 5) but its strategic truth, that is to say, the truth of the implicit claim that the speaker or writer knows that its strategic meaning is true, e.g. knows what the Skolem functions of S are.

244

Tarski Challenged

What is the logic of such "strategic" discourse? How is strategic meaning expressed explicitly? One striking thing is that merely introducing an epistemic element into our analysis is not enough. It is not enough to be able to express that someone knows the truth of a sentence like one of the (l)-(3). For this knowledge is still merely knowledge merely to the effect that there exist Skolem functions of a certain kind. For the purpose of capturing strategic meaning, it is not enough to be able to express merely knowledge that, i.e. propositional knowledge. We must also be able to express knowledge what (or who, where, when, ...), that is, identificatory knowledge. But surely such knowledge cannot be sui generis, you might object. Surely it must be analyzable in terms of knowing that. (Cf. here Hintikka 1996(b).) The answer is that identificatory knowledge is indeed analyzable in terms of propositional knowledge, but only if we also have the notion of independence applied to the epistemic operator a knows that. Then we can express the fact a knows who the person b is by (7) K ( 3 x / K ; (b = x) Likewise, I can express that a knows which function the function f \s (as in (4)) by (8) Κ Λ (νχ)(3 7 /Κ Λ )(/χ)= 7 ) which is equivalent with (9) a ^ V a - X / x ) = g{x)) Thus the logic of identificatory knowledge is IF epistemic logic. My two-part working hypothesis for future research in this area is thus twofold: First, the logic of strategic truth attributions is the logic of identificatory knowledge; second, this logic is construed as IF epistemic logic. This working hypothesis can be expected to serve as a basis of a theory of those uses of the concept of truth in our actual discourse that go beyond the literal truth exhausted in quantificational languages by the existence of Skolem functions. As an indication of what turns such a theory might take, I can recall that I have independently argued that IF epistemic logic is the true intuitionistic logic, that is, a realization (shall I say) of the correctly interpreted intuitions of the intuitionists. (See Hintikka 1996(a), ch. 11.) If so, intuitionistic logic (thus understood) will turn out to be the logic of strategic truth, but not of literary (abstract) truth. Generally speaking, from the epistemic viewpoint we can expect to understand the so-called "theories of truth" other than the correspondence and the minimalist theories. For instance, whereas coherence with other propositions will undoubtedly play a role in any full account of how we can actually come to know the truth of a proposition, albeit not necessarily in the form prescribed by some particular coherence "theory of truth." But to consider such accounts of our knowledge of truth constitutive of truth simpliciter is at best to commit the naturalistic fallacy. At worst such accounts are circular, for surely the notion of truth is more fundamental than the notion of a propositions possibly being true on the same occasion as another one, a.k.a. coherence.

What is Truth? Stay for an Answer

245

References Barwise, Jon, 1979, "On branching quantifiers in English", Journal of Philosophical Logic vol. 8, pp. 47-80. Cameron, Peter, and Wilfrid Hodges, forthcoming, "Some combinatories of imperfect information" Hintikka, Jaakko, 1998, "Truth-definitions, Skolem functions, and axiomatic set theory", Bulletin of Symbolic Logic, vol. 4, pp. 303-337. Hintikka, Jaakko, 1996(a), The Principles of Mathematics Revisited, Cambridge U.P., Cambridge. Hintikka Jaakko, 1996(b), "Knowledge acknowledged: Knowledge of propositions vs. knowledge of objects", Philosophy and Phenomenological Research vol. 56, pp. 251-275. Hintikka, Jaakko, 1990, "Contemporary philosophy and the problem of truth", in L. Haaparanta et al., editor, Language Knowledge, Intensionality: Perspectives on the Philosophy of Jaakko Hintikka, (Acta Philosophica Fennica vol. 49, pp. 23-39) Societas Philosophica Fennica, Helsinki. Hintikka, Jaakko, and Gabriel Sandu, 1997, "Tarski's guilty secret: compositionality", in Wolenski and Köhler, pp. 217-230. Hintikka, Jaakko, and Gabriel Sandu, 1996, "A revolution in logic?", Nordic Journal of Philosophical Logic vol. 1, pp. 169-183. Hodges, Wilfrid, 1997(a), "Compositional semantics for a language of imperfect information", Lope Journal of the IGPL, vol. 5, pp. 539-563. Hodges, Wilfrid, 1997(b), "Some strange quantifiers", in Jan Mycielski et al, Lecture Notes in Computer Science, Springer, Berlin, pp. 51-65. Janssen, Μ.V Theo, 1997, "Compositionality", in van Benthem and ter Meulen, pp. 417-473. Tarski, Alfred, 1956, Logic, Semantics, Metamathematics, Clarendon Press, Oxford. Van Benthem, Johan, and Alice ter Meulen, editors, 1997, Handbook of Logic and Language, Elsevier, Amsterdam. Walkoe, W. Jr., 1970, "Finite partially ordered quantification", Journal of Symbolic Logic vol. 35, pp. 535555. Wolenski, Jan, and Eckehart Köhler, editors, 1999, Alfred Tarski and the Vienna Circle (Vienna Circle Institute Yearbook vol. 6), Kluwer Academic, Dordrecht. Zazarozny, W., 1994, "From compositional to systematic semantics", Linguistics and Philosophy vol. 17, pp. 329-342.

ν Alternative Approaches

The Two Faces of the Concept of Truth MICHAEL DUMMETT

A great many philosophical and logical treatments of the concept of truth are concerned with solving the problem of diverse semantic levels: let us call this for short 'the problem of levels'. By this I mean that they seek a satisfactory analysis of one or other of the following two evidently valid patterns of inference: (A) Henderson made the statement "Venus has no satellites". Venus has no satellites.Henderson made a true statement. (B) Wilkinson believes that Venus has no satellites. Venus has no satellites. So Wilkinson has at least one true belief. The inference (A) can be readily analysed if it is permissible to add as an additional premiss: (Al) The statement "Venus has no satellites" is true if and only if Venus has no satellites. Let us call (Al) a 'Ts-principle' (T for 'true', s for 'statement'). Such a Ts-principle obviously involves a shift of semantic level: on one side of the 'if and only if' is a sentence ascribing truth to a certain statement, on the other the sentence that expresses that very statement, and is not about any statement, but about the planet Venus. If a way can be found of justifying (Al), and all similar Ts-principles, then the inference (A), and all similar inferences, will cease to be problematic. Not only that, but a host of other sentences involving the word 'true' will become readily explicable. Let us call those who believe that an analysis of the concept of truth wholly consists in a solution to the problem of levels 'level theorists': they are often called 'deflationists', 'minimalists' or 'redundancy theorists', but we shall use the term 'level theorists' to emphasise their defining contention. Level theorists come in two varieties. One comprises those who take the problem of levels to amount to that of justifying Ts-principles. They are those level theorists who take the predicate 'is true' to apply primarily to linguistic items, sentences or statements. For present purposes, we need not be precise about what exactly a statement is to be taken to be, as long as it is clear that it is a piece of language: a declarative sentence, considered as uttered with assertoric force on its own, not as part of some more complex sentence. Level theorists of the type I am considering may be called 'sentential level theorists': 's-level theorists' for short. When they are concerned with statements made by means of sentences containing indexical or demonstrative components, they will need to relativise the Ts-principles to possible occasions of utterance. It is useless to invoke a Ts-principle of the form:

250

Alternative Approaches

The statement 'It is very cold today is true if and only if it is very cold today. This would serve to validate such an inference as: Brown made the statement, 'It is very cold today'. It is very cold today. So Brown made a true statement. The inference is obviously invalid if the first premiss is understood to be rendered true by Brown's having said yesterday, 'It is very cold today': yesterday may have been fairly warm. Exactly how the s-level theorist goes about the necessary task of relativisation need not detain us. In a similar way, the inference (B) is rendered easily analysable if it can be shown to be justifiable to insert the additional premiss: (Bl)

The proposition that Venus has no satellites is true if and only if Venus has no satellites.

If it be allowed that what are believed are propositions, so that in the sentence 'Wilkinson believes that Venus has no satellites', the clause 'that Venus has no satellites' denotes a proposition, and if, trivially, to have a true belief is equated with believing a true proposition, then the addition of (Bl) as a premiss renders the validity of the inference (B) very easily accounted for. (Β 1) may be called a Tp-principle. It can be more colloquially expressed as: (ΒΓ)

It is true that Venus has no satellites if and only if Venus has no satellites.

If (ΒΓ) is to be regarded as equivalent to (Bl), the phrase 'it is true that' cannot be understood as a sentential operator that carries the sentence to which it is prefixed into one with the same truth-value. Rather, the clause 'that Venus has no satellites' must again be construed as denoting a proposition and 'It is true' as a means of applying the predicate 'is true' to that proposition. There is again a shift of semantic level: on one side of the 'if and only if a proposition is referred to, and truth predicated of it; on the other side the same proposition is expressed, but not referred to. We may call (Bl) a 'Tp-principle'. All Tp-principles involve a shift of semantic level, although a different one from Ts-principles. On the most common understanding of propositions, Tp-principles require no relativisation: propositions do not themselves have any indexical or demonstrative components, even if the identity of a proposition expressed by uttering a sentence that does have such components depends in part upon their presence in that sentence. The second category of level theorists comprises those who believe that truth is primarily predicated of propositions, and who therefore think that the problem of levels is to be resolved by showing how Tp-principles are to be justified. We may call these 'p-level theorists'. Now s-level theories can be further classified as global or piecemeal. The crudest form of global s-level theory simply stipulates outright that every instance of the schema:

The Two Faces of the Concept of Truth

(S A )

251

The statement "A" is true if and only if A

is to hold, where an instance of S A is obtained by replacing both occurrences of the letter Ά by a declarative sentence. More exactly, whenever Ά' is to be replaced by a declarative sentence having one or more indexical or demonstrative components, what is stipulated to hold good will be the appropriate relativisation of the resulting instance of S A . By this means, all Ts-principles are simultaneously decreed to hold. More elegant global s-level theories claim to derive the instances, or relativised instances, of S A from some general thesis such as: A statement is true if and only if things are as it states them to be. The objection to global s-level theories is that they do not really solve the problem of levels, but simply decree that it is to be treated as solved. They do not explain what the semantic connection is between saying that Venus has no satellites and ascribing truth to the statement "Venus has no satellites": they merely announce such a connection. The symptom of the failure of a global theory to solve the problem of levels is the fact that the quotation marks in S A do not function as quotation marks ordinarily do. Ordinarily understood, the term '"A"' denotes the letter A', and is thus complete in itself and hence immune to any replacement of the letter A' which the quotation marks enclose by any other expression, just as it would be in the sentence: "A" is the first letter of the alphabet. But in S A the quotation marks are not to be so understood: they are to be understood as due to function in the normal way only when they come to enclose a sentence by which the letter A' which they enclose in S A has been replaced. It is by means of this suspension of the usual effect of the quotation marks that the shift of levels is brought about. It is brought about by this device: it is not explained. This defect of global s-level theories is overcome by piecemeal s-level theories, the prototype of which is Tarskis celebrated truth-definition. A piecemeal s-level theory provides a semantic theory for a language for statements made in which the predicate 'is true' is to be specified. Such a semantic theory assigns semantic values to the primitive expressions of the language and lays down how sentences of the language are determined as true or otherwise in accordance with their composition out of those primitive expressions. It effects the shift of levels because it is so framed as to determine that every Ts-principle shall hold; but it is not open to the criticism that applies to the global theories, namely that they fail to explain how a shift can be made from one level to the other. It escapes this criticism because it does not treat explicitly of the shift of levels at all: what it does not speak of it cannot be blamed for failing to explain. It justifies each particular Ts-principle, but it takes no cognisance of the fact that it is a Tsprinciple: it cannot characterise Ts-principles as a class. Notoriously, Tarskis construction does not allow the predicate 'is true' to be applied to sentences containing that predicate: the truth-predicate must therefore belong to a language distinct from that to whose sentences it is applied. Devices for avoiding this re-

252

Alternative Approaches

suit and constructing a language containing its own truth-predicate have been proposed by Kripke, Hintikka and others; but my present purpose is only to give a very general conspectus of level theories of truth, and I shall therefore abstain from examining these proposals. In principle p-level theories might similarly be divided into global and piecemeal ones; but piecemeal p-level theories cannot be framed. We have a sufficiently good understanding of the structure of propositions: what we lack is a means of circumscribing the basic components out of which all propositions are composed, analogous to our means of circumscribing the primitive expressions out of which all the sentences of a given language are constructed. So p-level theories of truth are necessarily subject to the charge that they fail to offer any semantic analysis or explanation of the shift of levels, but simply impose it by force majeure. There is a far deeper reason for rejecting p-level theories. In agreement with Frege, plevel theories take truth as primarily ascribed to propositions. It may plausibly be argued that this is in conformity with our habitual practice. Someone says something, and I respond, Ί think that that is true'. A little later someone else expresses the same proposition, using different words, and perhaps substituting 'he' for the previous speaker's T . If I reply, Ί already said that I thought that that was true', what can the demonstrative pronoun 'that' which I employed have referred to? Not, surely, to either of the sentences uttered, since they are different: it can refer only to the proposition that they both expressed. Since p-level theories form one species of level theory, they represent their method of handling the shift of levels by determining that all Tp-principles shall hold good as constituting the whole core of a philosophical account of the concept of truth. In doing so, they take the propositions to which truth is to be ascribed, or of which it is to be denied, as given: we must know what a proposition is before we can understand an account of what is meant by ascribing truth to it. I am not here claiming that a p-level theory presupposes a philosophical account or defence of the general notion of propositions: it presupposes only that we can know what proposition is expressed by a given utterance, in which proposition someone manifests his belief, and so on. What is needed in order to know what proposition is expressed by the utterance of a given sentence in given circumstances? Obviously we must know what the sentence means and apprehend the relevant circumstances. More exactly, we must grasp one central ingredient of the sentence's meaning — that which determines what proposition it expresses according to the circumstances in which it is uttered. The meaning of the sentence, as a type sentence or as uttered on a specific occasion, may have other ingredients, for instance, whether it carries assertoric or interrogative force: but that central ingredient must be grasped if we are to know what proposition its utterance expressed. What, then, determines that central ingredient of a sentence's meaning? An answer favoured by many philosophers, from Frege onwards, is: the condition for the sentence, when uttered in given circumstances, to express a true proposition. But the p-level theorist cannot adopt this answer without being caught in a vicious circle. He

The Two Faces of the Concept of Truth

253

has offered an account o f the concept o f truth which takes propositions as given when the notion o f truth is to be grasped in accordance with that account: his theory treats the notion o f a proposition as prior to that o f truth. He must therefore eschew any explanation o f the meanings o f sentences which makes use o f the concept o f truth. He must explain what it is for anyone to grasp what proposition is expressed by uttering a sentence in a language he knows without attributing to him an understanding o f what it is for that proposition to be true. And how is the theorist to do that? It may be replied that there are theories o f meaning which do not represent the meaning o f a statement as consisting in or determined by the condition for it to be true. For example, the intuitionist theory o f the meaning o f a mathematical statement holds that it is determined by the recognisable condition for a construction to constitute a proof o f it. An intuitionist may or may not propose to stipulate what it is for a mathematical statement to true; but this is not an essential constituent o f his theory o f meaning for such statements. T h e primary role o f proof in mathematical discourse is that it serves to justify our treating the theorem proved as true and asserting it as true. I f we attempt to generalise the intuitionist theory o f meaning to non-mathematical statements, we must replace the notion o f a proof o f a statement by that o f whatever serves to justify our treating it as true and asserting it as true. Such a theory o f meaning explains knowing what proposition a statement expresses in terms o f knowing what would justify treating it as true: and to understand what it is, in general, to treat a statement as true plainly involves a grasp o f the concept o f truth, or at least o f one aspect o f it. Hence the p-level theorist could no more avoid being entrapped in a vicious circle if he appealed to such a justificationist theory o f meaning than if he appealed to a straightforward truth-conditional one. T h e same holds good o f pragmatic theories o f meaning. Such theories characterise the meaning o f a statement as determined by how acceptance o f it as true is manifested: that is, in the difference accepting it makes to the subject's actions and linguistic utterances - the consequences o f accepting it as true, or what constitutes acting on its truth. Once again, a grasp o f the general notion o f acting on the truth o f a statement involves apprehending the concept o f truth, or at least one aspect o f it: once again, the p-level theorist cannot have recourse to a pragmatic account o f meaning without being caught in a vicious circle. We have here three general types o f theory o f meaning: the truth-conditional type, which takes as its central notion that o f a statements being true, o f its possessing the quality o f being true, whether we know it or not; the justificationist type, which takes as its central notion that o f what justifies our treating the statement as true; and the pragmatic type, whose central notion is that o f what is involved in accepting the statement as true. All three connect meaning with a salient aspect o f the concept o f truth: in doing so, they are responsive to the powerful philosophical intuition that the concepts o f meaning and o f truth are inextricably entwined with one another. According to Frege, anyone who makes a judgement knows implicitly what truth and falsehood are: how could anyone grasp what proposition a given statement expressed without

254

Alternative Approaches

knowing, at least implicitly, what condition must hold for that proposition to be true, or what would justify treating it as true, or what would constitute acting on its truth? And yet the p-level theorist must maintain that this is possible, because he offers to explain what it means to ascribe truth to a proposition on the assumption that the proposition can be identified in advance of understanding such an ascription. This leaves him owing us an explanation of how, on his view, we do identify the proposition that an utterance expresses; and this he has no way of doing. The presence in p-level theories of this fatal defect may suggest that s-level theories are to be preferred. Since they take truth as primarily ascribed to linguistic items, sentences or statements, they do not tacitly invoke the notion of meaning by deploying the concept of propositions. It may be retorted that global s-level theories tacitly invoke the notion of meaning, in that Ts-principles must be so understood that the sentence enclosed in quotation marks on one side of the connective 'if and only if has the very same meaning as the same sentence occurring without quotation marks on the other side of the connective. But piecemeal s-level theories certainly do not invoke the notion of meaning in any way. A piecemeal s-level theorist is nevertheless left as incapable of explaining what determines the meanings of sentences of the language as the p-level theorist was. The slevel theory purports to explain the predicate 'is true' as applied to such sentences. The predicate will therefore not be available to the theorist to frame a theory of meaning for the language. As Davidson has many times observed, if we treat the predicate 'is true' as already given, we can turn a Tarskian truth-definition upside down and, while of course no longer viewing it as the predicate 'is true', convert it into a truth-conditional theory of meaning for the language. But, so long as it is claimed as a definition or explanation of the predicate 'is true', we cannot appeal to a truth-conditional theory of meaning: it would obviously be frivolous first to give the truth-definition to explain what 'is true' means, and then, with this predicate now available to us, to repeat the whole truth-definition a second time as a way of specifying the meanings of the sentences. It may be objected that this argument will not hold good if we favour a justificationist or pragmatist account of meaning instead of a truth-conditional one, but this is wrong. A justificationist will not understand the connective 'if and only if' as explained by the classical two-valued truth-table: rather, he will take a biconditional Ά if and only if B' to hold good just in case whatever would justify treating A as true would likewise justify treating Β as true, and conversely. I do not know exactly what a pragmatist logic would be like, but I presume that a pragmatist would regard the biconditional as holding if whatever counted as acting on the truth of A counted equally as acting on the truth of B, and conversely. For both, therefore, the Ts-principles will give the meanings of the quoted sentences in terms of the notion each theorist takes to be central: what justifies treating it as true, or what constitutes acting on its truth. For both, turning the Tarskian truth-definition on its head converts it into the kernel of a theory of meaning for the language, provided that the logical constants used in the definition are uniformly understood in the senses appropriate to the theory of meaning in question. In no case, therefore, can the s-level theorist, who presents the Tarskian truth-definition, or whatever variant of it he favours, as defining or explaining the

The Two Faces of the Concept of Truth

255

predicate 'is true', appeal to any theory of meaning of one of the three types here considered. Like the p-level theorist, he cannot invoke the notion of truth to explain meaning, and is left unable to explain it at all. A level theory of truth - one that seeks to characterise the concept of truth by laying down or deriving all Ts-principles or all Tp-principles - undoubtedly succeeds in specifying a notion of truth. It is arguably a notion a grasp of which suffices to explain all uses of the word 'true' that occur in everyday discourse. By 'everyday discourse' I mean remarks like 'She would not have said it if it were not true' and 'Very little of his testimony was true': I exclude statements such as 'To know what a statement means you must know what is required for it to be true' or 'An inference is deductively valid if it is guaranteed to preserve truth from premisses to conclusion'. We do not, of course, maintain any formal boundary marking off the object-language which we speak from a metalanguage in which we formulate principles governing how that object-language functions. But we do reflect, sometimes very informally, sometimes with a high degree of formality, on how our language works; and, when we engage in such theorising, we must necessarily regard the theoretical terms we employ in our inchoate or precisely delineated theories as standing outside the language the principles governing whose working we are seeking to frame: we maintain a notional distinction between objectlanguage and metalanguage. The concept of truth has two faces. On the one side, the word 'true' belongs to our everyday language: it is part of the object-language whose operation a theory of meaning seeks to describe. It is the role of the word 'true' within the language which a level theory seeks to explain. For this purpose we need a level theory that is neither circular nor lame: it is lame if it fails to provide an adequate solution to the problem of the shift in levels. From what has gone before, it is clear that our best hope must lie in a piecemeal s-level theory that allows the predicate 'is true' to belong to the language to whose sentences or statements it applies. It is not to my purpose here to review the candidate theories. But the concept of truth has another, different face: it is an indispensable theoretical term in any conceivable theory of meaning aiming to describe how a language functions and how its sentences and other expressions are thereby determined as having the meanings that they have. The notion of truth explained by a level theory cannot fulfil this role. For someone to grasp the explanation that the level theory supplies he must already understand a large part of the language concerned. It is not enough for him to acquire the knowledge that the form of words expressing a Ts- or Tp-principle is to be treated as holding good: he must already understand that form of words, save for the predicate 'is true' which is being explained. It is not enough, for example, that he should learn that the sentence: The statement 'Venus has no satellites' is true if and only if Venus has no satellites, is to be treated as saying something true, without his knowing what 'satellite' means or what the name 'Venus' denotes: to grasp the contribution that this Ts-principle makes to explaining what 'is true' means, he must understand the Ts-principle as thus formulated. This involves that he must already know what the sentence 'Venus has no satellites' means, since this sentence figures as the clause following the connective 'if

256

Alternative Approaches

and only i f in the statement of the Ts-principle. It follows that the notion of truth explained by appeal to Ts-principles cannot serve to explain the meanings possessed by sentences of the language such as 'Venus has no satellites': the explanation presupposes a prior grasp of their meanings. Level theories are impotent to characterise a notion of truth capable of serving as a theoretical notion within a theory of meaning. I claimed that the concept of truth is central to any conceivable theory of meaning; this claim needs to be made more precise. The kernel or nucleus of a theory of meaning for a language is a semantic theory for that language, in the sense in which mathematical logicians speak of a 'semantic theory'. When truth is classically conceived, as it usually is, a semantic theory operates to lay down how any sentence of the language is determined as true or otherwise in accordance with its composition. It does this by assigning a semantic role to every primitive expression of the language, and laying down how the semantic role of a complex expression is determined by the semantic roles of its constituents. The notion of truth may be said to be classically conceived if it is regarded as subject to the principle of bivalence, namely that every meaningful statement is determinately either true or not true, independently of our knowledge. The semantic role of an expression is that feature of it which constitutes its contribution to determining the truth or otherwise of any sentence in which it occurs. Almost all semantic theories specify the semantic role of each expression by associating with it something extra-linguistic, which may be called the 'semantic value' of the expression: thus the semantic value of a proper name or individual constant will be an object or element of the domain of quantification, that of a functional expression a function defined over the domain, and so on. The semantic value of an expression is essentially what Frege called its Bedeutung (reference): a semantic theory of the usual type is in Fregean terminology a theory of reference. In a two-valued semantics, the semantic value of a sentence will of course simply be its being or not being true: but this will not be so in all semantic theories, even when truth is classically conceived. An obvious case is a possible-worlds semantics, designed to handle modal operators like 'Necessarily' and 'Possibly'. In such a theory, a subsentence of a complex sentence does not contribute to determining whether or not the complex sentence is true absolutely, that is, true in the actual world, by its being itself true or false absolutely: it contributes, rather, in virtue of the possible worlds at which it is true. In this semantics, therefore, the semantic value of a sentence consists in the set of possible worlds at which it is true. So to claim truth to be central to a semantic theory, or to the theory of meaning of which it is the nucleus, is not to claim that the semantic value of a sentence must consist in its being or not being true. It is to claim only that the semantic value of a sentence must be something in terms of which its being or not being true simpliciter can be specified, as, in the possible-worlds semantics, a sentence's being true simpliciter can be specified as its being true in the actual world, so that the semantic theory can still be characterised as analysing how any sentence is determined as being or not being true, i.e. true simpliciter or true absolutely, in accordance with its composition. A semantic theory is only the nucleus of a theory of meaning. It does not in itself constitute a theory of meaning, or even that component of a whole theory of meaning

T h e Two Faces of the Concept of Truth

257

for a language which specifies the meanings of particular sentences and expressions. It does not do so because it is inadequate to explain in what the understanding of such expressions and sentences consists which is possessed by the speakers of the language. It is because they mutually understand those expressions that the speakers can communicate with one another by means of the language: if a theory of meaning for the language is to explain how the language functions, it must be able to explain what it is for a speaker to understand an expression, that is, to know what it means. A bare semantic theory cannot explain this, because it can never be a complete account of what a speaker knows concerning an expression that he knows its semantic value. We can never think or conceive of an object, or a function, or of anything that can be the semantic value of an expression, just as that object, or that function, etc.: the object, function or whatever must be given to us in some particular way. An object, for instance, may be given as the object presently perceived, or as the one previously perceived, or as that which plays a certain role in events, or as that which stands in a certain relation to some other object, or in any of a multitude of other particular ways. The aptitude of the language for communication would be undermined if every speaker took the semantic value of any specific expression as given in a different way. This of course does happen to a limited extent, especially with terms such as placenames; but for successful communication, speakers need to be able to be sure that each is taking any one expression as having the same semantic value as is the other. For this reason the way in which the semantic value of an expression is to be taken as given the way in which the speakers are to think of the object, function, etc. in question is to a large extent a feature of the language as such: part of what must be known by a speaker if he is to understand the expression in accordance with its meaning in the common language. O n this account, then, the sense of an expression - what a speaker must grasp concerning it in order to understand it - is the particular way in which its semantic value must be taken as being given when the expression is used, by that speaker or another, as part of an utterance in the language concerned. In giving this account, I am following Frege: for him, the sense of an expression is the way in which its reference — that is, its semantic value - is given to one who knows the language: die Art des Gegebenseins, in the famous phrase. Thus, on this account, a full-fledged theory of meaning for a language will specify the senses of all its expressions as the particular ways in which a speaker is to think of their semantic values if he is to assign the right contents to sentences in which they are used. This theory of sense thus rests on the underlying theory of reference - the underlying semantic theory — as a base. How, after all, could there be any other relation between the theory of sense and the correct semantic theory? Sense must determine semantic value. The sense which the language requires a speaker to attach to a name or singular term must determine which object a speaker is using it to refer to: how else could it be determinate which object that was? The same goes for all other categories of expression: the sense which the language requires a speaker to attach to a functional expression must determine which function he is invoking; and similarly with predicates and all other categories of expression. If sense did not determine semantic value — if the theory of sense did not rest on semantic theory - where would the semantic values of the

258

Alternative Approaches

expressions come from? Now, since the concept of truth is central to the semantic theory, it is likewise central to the theory of sense as well. Truth is not always classically conceived. It is not so conceived by a proponent of a justificationist theory of meaning, and I see no reason why it should be so conceived by a proponent of a pragmatist theory, either. What form, then, does a semantic theory assume when truth is not classically conceived? Does it remain the case that truth is central to such a semantic theory? Semantic theories are used to provide a criterion - the semantic or model-theoretic criterion - for the validity of deductive inferences: usually of just those inferences whose validity depends on the patterns according to which premisses and conclusion are constructed out of primitive expressions by means of the logical constants. When truth is classically conceived, the criterion for deductive validity is taken to be that the hypothesis that the premisses are determined as true in accordance with their composition guarantees that the conclusion will likewise be determined as true: truth is the property required to be transmitted from premisses to conclusion. The ability of the semantic theory to provide a criterion for deductive validity corroborates the contention that a semantic theory forms the nucleus of a theory of meaning. A mastery of a language certainly requires a general ability - though not a comprehensive or infallible one - to recognise the validity of deductive arguments framed in that language, since that is a prominent part of linguistic practice: without it there would be no such thing as giving grounds for an assertion that has been challenged or whose justification has been asked for. So what is the criterion for deductive validity to be when truth is not classically conceived? Should it be truth, or some different property, that is required to be transmitted from premisses to conclusion? When a deductive argument is seriously advanced, the premisses must, mistakenly or justifiably, have been recognised as true. It cannot be required of a valid deductive argument that it transmit the property of being recognisably true from premisses to conclusion. If every valid deductive argument had this characteristic, then it could only lead from premisses already recognised as true to a conclusion already recognisable as true; and if this were so, deductive reasoning could never bring about an extension of our knowledge. But it can unquestionably do so: we have only to consider how in mathematics deductive reasoning can establish the truth of statements whose truth we were previously unable to recognise in order to acknowledge the power of deductive reasoning to extend our knowledge. So the property which a valid deductive inference is guaranteed to transmit from premisses to conclusion cannot be that of being recognisably true, or any other property whose possession by a statement can always be effectively recognised. It must be a property such that, once we have recognised a statement as having it, we are thereby justified in asserting that statement, but such that we cannot know, simply from understanding a statement, whether or not it has it. There are different conceptions of truth. Within a justificationist or a pragmatist theory of meaning, the conception of truth will not be classical. But what entitles us to call a given notion of a property that statements can possess a conception of truth? If we view a semantic theory merely as explaining how the semantic value of a sentence is

The Two Faces of the Concept of Truth

259

determined from the semantic values of its constituents, then, as we have seen, the theory need not always operate with any notion of truth, or of being true absolutely, at all, since the semantic value of a sentence may not consist in its being or not being true (absolutely). In a possible-worlds semantics, the semantic value of a sentence consists in the set of possible worlds at which, i.e. relatively to which, it is true. In the intuitionistic semantics, the semantic value of a mathematical statement is constituted by the decidable relation between it and a construction which proves it. It is only when the semantic theory is put to use to characterise the validity of deductive inferences that a further notion is needed, that of the property which the inference must transmit from premisses to conclusion in order to be valid. We may reasonably call this property 'truth': whatever notion is used for this purpose by the semantic theory in question will constitute the conception of truth proper to that semantic theory. There is another condition which a notion of a property that statements can possess must satisfy if we are to be entitled to call it a conception of truth. There are two aspects to the meaning of a declarative sentence. On the one hand, we must, in order to understand it, know what is conveyed by a speaker who uses it on its own to make an assertion: how we must expect things to be if he spoke correctly. We may call this its assertoric content. But, in order fully to understand the meaning of the sentence, we must grasp the contribution it makes to determining the assertoric content of any more complex sentence of which it is a subsentence: we may call this its ingredient sense. An example is provided by a sentence in the present tense containing a locative adverb or adverbial phrase. The sentences "It is raining here" and "It is raining where I am" have the same assertoric content: they provide just the same information to a hearer. But, subjected to the temporal quantifier 'always', they yield sentences with different contents: "It is always raining here" and "It is always raining where I am" do not say the same at all. This is because the adverb 'here' is temporally rigid, while 'where I am' is temporally flexible. Clearly it is possible to apprehend the assertoric content of a sentence without mastering its ingredient sense. The ingredient sense is what a semantic theory seeks to characterise by assigning to the sentence its semantic value; but, to grasp its assertoric content, we do not need to know its semantic value: we need only know the condition for a speaker to have made a correct assertion by uttering the sentence on its own, that is, as a complete sentence. We may say that the assertoric content of a sentence is given by the condition for an utterance of it on its own to be true (to constitute a true statement). The notion of a property that statements may have that is used within some given theory of meaning may qualify as the conception of truth proper to that theory of meaning if it holds good, within that theory of meaning, that the assertoric content of a statement is determined by the condition for it to have that property - and hence for it to be true. Many writers take a theory of meaning for a language to be one concerned solely to specify the particular meanings attached to the words of the language and the principles governing their combination, and by this means to determine the meanings of all

260

Alternative Approaches

the sentences of the language. A good example is the earlier Davidson, who thought that a theory of meaning could be obtained by turning a Tarskian truth-definition on its head. But the problem for a philosopher of language is to explain the very concept of linguistic meaning: to say what it is for the words and expressions of a language to have meanings. To explain this is to explain what a language is: and this can be done by describing ab initio the practice of speaking a language. Such a description will explain the significance of saying something in the language: that is, the potential difference that is made by its being said to what subsequently happens. The practice of speaking a language is of course exceedingly complex, and a description of it will be correspondingly complex. Central to it will be an account of the use of language to make assertions, including the responses they evoke in the hearers. The description must also cover other uses of language, to ask questions, make requests, issue demands (of which commands are a special case) and so on. But assertion is primary: questions call for assertions to be made, and it is essential to the activities of making requests and demands that it be possible to say whether a request was granted or declined, a demand complied with or flouted. And it is from the practice of assertion that the notion of truth first stems. It is essential to a mastery of that practice that assertions be classifiable as correct or incorrect. To distil the notion of a true statement from that of a correct assertion it is necessary to distinguish a justifiable assertion from one that the speaker was in fact justified in making; but this may well not be enough to yield the conception of truth that is aimed at. It will not yield the classical conception of truth, for example, since according to it there will be many true statements no justification for making which exists. But, provided that it be possible to explain truth as classically conceived (something of which I personally am extremely sceptical), the notion can then be used in describing the practice of assertion. We have a strong intuition that the concepts of truth and meaning are inextricably linked. Why? A salient ground for deeming two sentences, considered as uttered in given circumstances, to express different propositions is that it is possible for someone who understands both sentences to believe the proposition expressed by one but not that expressed by the other. (It might be that the hearer took expressions occurring in the two, with the same reference but different senses, to have different references.) Now to believe a proposition is to take it to be true: so the propositions expressed by the two sentences are different if one can be taken to say something true while the other is not so taken. The concepts of truth and meaning must thus be explained together, as part of a comprehensive description of the practice of speaking a language. We cannot take the meanings of statements as given before stipulating what it is for them, or the propositions they express, to be true: nor can we take the notion of truth as given and use it to explain what it is for the words and sentences of a language to have the meanings that they have. The concept of truth, under the second of its two aspects, is a theoretical notion belonging to a comprehensive theory of meaning, and gains its content from the role that it plays in that theory. To be comprehensive, a theory of meaning for a language must not only give a general description of the practice of speaking the language, but must specify the particular meanings of its words and sentences. It will do this by the means we have already

The Two Faces of the Concept of Truth

261

reviewed: a theory o f sense, resting upon a semantic theory as base. T h e semantic theory will employ a notion o f truth to characterise the validity o f deductive inferences. I f the theory o f meaning is to be coherent, this notion o f truth must coincide with that which was used in describing the practice o f making assertions. According to the character o f the theory o f meaning in question, it will embody one or another conception o f truth. There is much room for philosophical dispute over what this conception o f truth ought to be, and hence what general form the theory o f meaning ought to take. I am not concerned here to engage in such disputes: I have wanted only to locate the concept o f truth in its correct place in the web o f concepts. T h e semantic theory which forms the nucleus o f the theory o f meaning will closely resemble a piecemeal s-level theory. How closely? Will it yield all Ts-principles as consequences? It will not necessarily do so. Since the assertoric content o f any statement Ρ is determined by the condition for Ρ to be true, it follows that the assertoric content o f 'S is true', where S is a term denoting P, must coincide with the assertoric content o f P. It does not, however, follow that the biconditional 'S is true if and only i f P' holds good. (Strictly speaking, we need to use Quine's quasi-quotes in this connection, but ordinary quotation marks are unambiguous in practice.) I f the biconditional did hold good, then any two statements differing only in that in one 'S is true' occurred where in the other Ρ occurred would likewise be equivalent: in other words, 'S is true' and Ρ would have the same ingredient sense; in particular, 'S is not true' and 'Not P' would be equivalent. But there is no requirement on a predicate which within a given theory o f meaning expresses truth as conceived in that theory that a statement ascribing truth in this sense to another statement Ρ should have the same ingredient sense as P. A simple counter-example is a conception o f truth according to which some statements acquire the property o f being true at a certain time, before which they lacked it. According to one conception, a statement that a state o f affairs obtains, or some event occurs, at a certain time may become true at that time and remain true thereafter, but was not true, nor, o f course, false, before that time. Likewise, one conception o f truth for mathematical statements regards them as becoming true only when they are first proved. These conceptions may well be controverted: but there is no ground to deny that they are conceptions o f truth. Now, when truth is conceived in such a way, a statement ascribing truth to another statement Ρ will obviously not in general be completely equivalent to P, and the biconditional connecting them may well not hold. For one thing, the statement ascribing truth to P, say 'S is true', is genuinely present-tensed, rather than in what Frege called the tense o f timelessness, and hence is subject to the operator carrying it into the past tense, while the statement Ρ may not be subject to any modification o f tense. Further, the negation o f ' S is true' may hold good because Ρ has not yet become true, and therefore is not equivalent to the negation o f P. 'S is true' and Ρ therefore certainly do not, on such a conception o f truth, have the same ingredient sense, and the Ts-principle 'S is true if and only if P' does not hold good. Completely to grasp the sense o f a sentence involves understanding its ingredient sense - the contribution it makes to the sense o f a complex sentence o f which it is a constituent. According to the sense o f the term 'true' required within a meaning-theory, a statement must have the same assertoric content, but need not have the same sense, as an ascription o f truth to that statement. No complex sentence containing both

262

Alternative Approaches

can express the relation between them, because as soon as either becomes a subsentence of a more complex sentence, its ingredient sense is called into play: it is only when it stands on its own that its assertoric content suffices to determine 'what it says'. Taken independently of the classical conception of truth, the truth-conditional account of meaning is right in holding that the condition for a statement to be true is the condition for an assertion of it to be correct: it is wrong in holding that this condition must determine the entire sense of the sentence used to make the assertion. Does this make clear the relation between a conception of truth that may play an essential role within a comprehensive theory of meaning - truth in its meaning-theoretic aspect - and the workaday notion of truth that is explained by a level theory? Or is the notion of truth specified by a level theory a spurious result of failing to distinguish between assertoric content and ingredient sense? I do not think it is spurious. It represents the only notion of truth we need to have in order to understand the use of the adjective 'true' within the language, that is, before we reflect on language and how it works. But, as philosophers, we are committed to engaging in such reflection.

Truth: A Prolegomenon to a General Theory LORENZ Β . PUNTEL

Arguably the explication of no other central notion in philosophy raises so many problems as the concept of truth. Not surprisingly, therefore, one encounters in the philosophical literature a great deal of conceptions or theories of truth, which are the result of extremely divergent intuitions and assumptions in different areas, especially in semantics, logic, epistemology, ontology. Very generally speaking, three questions are considered extremely controversial. Perhaps the most fundamental dissent concerns the question whether the notion of truth is "substantive" or "deflationary". 1 Another point of contest in the debate ranges over the kind of notion truth is: Is it a semantic, an epistemic, a linguistic, an ontological, a pragmatic etc. notion? Finally, there is the question whether truth is expressible or inexpressible, whether it is explicitly definable or undefinable or only implicitly definable, and whether it is a primitive or a derivative notion, and the like. An eloquent illustration of the deep dissent pervading the present discussions about the theory of truth are the extremely divergent interpretations, endorsements, modifications, adaptations and rejections of the theory of truth developed by A. Tarski, the author classicus with respect to this topic. 2 This essay presents a new approach to the question of truth. It attempts to disentangle several central threads of the highly controversial and complicated discussion about truth in contemporary philosophy. The general perspective turns upon a central insight about what might be called the linguistic-semantic Urfaktum: the internal indeterminacy of "simple" language, i.e., language without semantic vocabulary (abbreviated as LQ). Section 1 is devoted to working out and explaining this feature of "simple" language. To achieve this, the "phenomenon" of determinacy and indeterminacy of language will be scrutinized in detail. One important result will be the elucidation of the function of semantic vocabulary in language. Special attention will be given to the distinction between semantics and pragmatics. Section 2 shows that fundamental assumptions and presuppositions most contemporary theories of truth rely upon with-

1

2

Simple quotation marks ( " ) are employed: (i) to mark a quotation within a quotation; (ii) to indicate that the expression put into quotation marks is mentioned, i.e., is taken as a linguistic expression. Double quotation marks ( " " ) are used: (i) to mark a normal quotation for which a reference is given; (ii) to indicate that the expression in quotation marks belongs to a passage that has been already quoted or otherwise referred to; (iii) to highlight a special, unusual or very important meaning attributed to the expression in quotation marks. To cite only some literature which appeared in the last years: Hintikka 1996 (especially ch. 6), Heck 1997, Harrison 1998, Boisvert 1999, Soames 1999, Hintikka/Sandu 1999, Hintikka (unpublished).

264

Alternative Approaches

out further ado are mistaken because they do not take into account the internal indeterminacy of "simple" language and the very idea of the function of semantic vocabulary. Section 3 contains a sketchy explanation of the root idea of truth (for elementary sentences and propositions) that is shown to be a naturally ensuing consequence of the propounded approach. Finally, section 4 contains concluding remarks in view of developing a general theory of truth.

1. The internal indeterminacy of "simple" language and the function of semantic vocabulary 1.1 The "phenomenon" of (in)determinacy of language It is claimed that "simple" language (Z^) - in the semantic (not in the syntactic) perspective - is internally indeterminate. 3 To explain the concept of internal indeterminacy, the phenomenon of (in)determinacy of language must first be considered in a global perspective. One general characterization of what is (more exactly: should be) understood by language can be formulated thus: Language is a semiotic system endowed with a syntactic structure and interpretation capable of being used (and/)or applied in different areas like communication, exposition or presentation of theories etc. It should be clear from the outset that this understanding of language is situated in the tradition of "interpretation semantics". The expression 'semantical' has not been used in the characterization because "interpretation" can be taken as "existing" without the explicit occurrence of any semantic vocabulary. This may be seen - and has generally been seen as a very familiar and uncontroversial "minor detail". But upon closer examination this "minor detail" emerges as very surprising and in need of careful explanation. It will turn out to be the pivotal point in the approach to be presented in this paper. The "use" (and/)or "application" of language can be dubbed "pragmatics". But it should be pointed out that "pragmatics" in the sense intended should not be identified with communication in the usual sense, according to which languages are understood

3

The internal indeterminacy of language meant in this paper should be carefully kept distinguished from Quine's famous indeterminacy thesis (more exactly: theses). In the following passage Quine characterizes the exact meaning of his theses: Three theses of indeterminacy have figured conspicuously in my writings: indeterminacy of translation, inscrutability of reference and underdetermination of scientific theory. Each of the three presupposes a distinctive further topic with concerns of its own. The topic presupposed by a translation is stimulation. The topic presupposed by the inscrutability of reference is reification. The topic presupposed by underdetermination of science is empirical content. A further topic still, pursuant on the underdetermination of science, is truth of empirically equivalent theories... (Quine 1990a, p. 15. Cp. also: Quine 1990b, pp. 47-52, 1 0 1 - 0 2 ) Using Quine's terminology, the "indeterminacy", as understood in this paper, is to be characterized thus: the "topic presupposed" by indeterminacy of "simple" language (according to the above established understanding of this expression) is language-internal indeterminacy.

Truth: A Prolegomenon to a General Theory

265

simply as systems of communication. Indeed, semiotic systems can be conceived of or constructed as set-theoretic objects and models can be brought in. To be sure, one can question the legitimacy of calling 'languages' such semiotic systems constructed as set-theoretical objects. Hugly and Sayward, to give an example, claim that it is "plain that formal systems and pairs comprising formal systems and models are languages only by courtesy of definition". 4 But this is a matter of terminology. In this paper the expression 'language' is not restricted to designating only systems of communication. If language is understood in the sense indicated, the notion of internal (semantic) (in)determinacy can be clarified. "Internal (semantic) (in)determinacy" is short for "internal (semantic) (in)determinate interpretation (of language)". To work out the specific notion of internal (semantic) (in)determinacy of language, the most salient forms of (in)determinacy of language will first be briefly described. "Pure" formal systems are often called "languages". This claim can be justified if the formulation is understood to mean: "pure" formal systems are "actually" uninterpreted or semantically indeterminate, but "potentially" interpreted or semantically determinate languages, for short: interpretable or semantically determinable languages. This is performed when a model is effectively presented. An interesting form of the phenomenon of (in)determinacy of language is language endowed only with "literal (or lexical) meaning". Such a language is "understood" in the sense that its expressions are taken to mean only what a dictionary of the language indicates. Sentences like Ί am writing a paper' or 'Snow is white' having only literal (or lexical) meaning (i.e., taken as examples of learning the language in question) must be said to be "interpreted" sentences, since they are understandable/understood. Thus, they are - to a certain extent - determinate or they have a determinate status. But it is clear that they are not fully determinate or that they do not have a fully determinate status since no determinate reference of the occurring expressions, for instance of T , is given, although every speaker of English knows what the expression Τ means (in the literal/lexical sense) and the status of both sentences is not established (it could be a statement, an example of a sentence of English etc.). Such a language which possesses only literal/lexical meaning can, therefore, be considered as being merely "semi-determinate" or "underdeterminate", i.e., as not being "fully determinate". What about the most common kind of sentences of natural languages, i.e., sentences uttered in every-day contexts, in contexts of communication? There can be no doubt that most sentences of natural languages do have a fully determinate status in the sense to be explained below. For instance, every speaker of English understands the sentence 'Snow is white' uttered in relevant communication contexts and takes it as being fully determinate within such contexts. This fact seems so obvious that it would appear pointless to deny it or to take it not seriously. But the real question to be asked with respect to this obvious fact is not whether it makes sense to deny it or to dismiss it in some way or another; rather, the real question concerns the philosophical significance of this fact. The determinateness of the sentence 'Snow is white' uttered in com-

4

Hugly and Sayward 1983, p. 81.

266

Alternative Approaches

munication contexts does not derive from within language itself and, consequently, not from within the sentence itself, i.e., from the sentence taken as a string of words, even presupposing that they are syntactically structured, i.e., that they constitute a linguistic expression called 'sentence' . Rather, the determinateness of the sentence derives from the context of the utterance. From this it follows that one has to be very careful when one simply quotes sentences like 'Snow is white' as examples for illustrating semantic features. Without specifying the character or the status of the determinatedness of the quoted sentence, the quotation of the sentence cannot serve as an example - let alone a proof - of any semantical feature or thesis. A very different form of simple language or L0 is language endowed with a pragmatic vocabulary, i.e., a vocabulary by means of which what is being performed is made linguistically explicit at the same time. An example of pragmatic vocabulary, thus understood, is the sentence 'Peter asserts that snow is white.' As will be shown in the next subsection, pragmatic vocabulary must be clearly distinguished from semantic vocabulary. As the example shows, Peters assertion is a factor which renders the sentence 'Snow is white' fully determinate. If we scrutinize more closely the forms of the (fully) determinate languages just described, we find that their determinatedness is afforded by either purely external or internal-external factors. Let us explain. So far we have found two "dimensions" or "factors" which afford determination of languages described. The first is a dimension purely external to language, "from outside language": the dimension of "external" contexts in which the speaker of a language is situated and in which he utters sentences of the language without using any kind of vocabulary (like the pragmatic and the semantic vocabulary) that makes explicit how language is to be interpreted. Simply speaking natural language does not in general explicitly qualify the way it should be fully interpreted; in other words, it doesn't explicitly fix the full interpretation and, thus, the full determination of language. (Languages endowed only with "literal/lexical meaning" represent a special case of languages determined by a purely external factor. This factor is a codified form of the language, a dictionary.) The second dimension is a mixed one: Sentences are uttered in a certain context and just this circumstance is also made linguistically explicit. Here the context is something non-linguistic, for instance an action and the like; but making this external circumstance linguistically explicit is an internal factor of language. We have, thus, an external-internal dimension, a mixed one. This is the locus originarius of speech acts. Pragmatic vocabulary is that segment of language which verbally expresses the fact that language is being made determinate "from outside" language.

1.2 The function of semantic vocabulary and the root idea of truth [1] There is a third dimension affording the determination of language, a purely internal dimension, a kind of "self-determination" of language. This dimension is constituted by the semantic vocabulary in the strong sense, whose main expressions are 'meaning', 'reference', and above all 'truth (true)'. It is of utmost importance to avoid at the

Truth: A Prolegomenon to a General Theory

267

outset two kinds of misunderstandings concerning this dimension. The first is the opinion that talk of "self-determination" of language necessarily induces a semantics exclusive of any language-world-connections. "Internal" determination or "selfdetermination" of language as understood in this paper means "only" that the determinate status of language is given by an explicitly occurring semantic vocabulary or, equivalently for that matter, by a presupposed semantic vocabulary. If no semantic vocabulary occurs, as will be shown, the language can still be fully determinate, but this fully determinate status relies on a presupposed semantic vocabulary. What semantic vocabulary affords, namely the fully determinate status of language from within language (in the sense hinted at above and to be explained in more detail in a moment), is in the first place the relationship of language precisely to the ontological dimension. The second misunderstanding results from confusing semantic vocabulary with any other kind of vocabulary. This can be made clear by comparing semantic and pragmatic vocabularies. Pragmatic locutions like 'to utter, 'to assert', 'to state', 'to claim' etc. appear to be meaningful only if we are prepared to answer a question presupposed by these locutions: Who utters, asserts, states, claims etc.? The relation to a factor external to language, to a speaker or user of the language, is an essential feature of pragmatic locutions. By contrast, there is no relation of the purely semantic locutions to a factor external to language. To say 'It is true that snow is white' doesn't contain or imply any relationship or reference to a speaker or subject or user of language. Language itself, as it were, is saying that snow is white. Indeed, language contains semantic formulations in the purely internal sense, for instance: 'The sentence S expresses ... or says that ...'; 'It is true that S', and the like. Thus, the fiinction of semantic vocabulary is to make language fully determinate from inside language itself independently of any (external) factor whatsoever. Semantic expressions are, therefore, the global interpretive determiners of language from inside language itself. In the present context only a first hint at the concept of internal semantic (in)determinacy can be given; this concept will turn out to be pivotal for the explication of the concept of truth to be delineated below. For a linguistic item to be fully determinate means possessing a fixed, definite and definitive position in what might be called the semantico-ontological space or framework presupposed. As is well known, the idea of "linguistic frameworks" has been developed by R. Carnap. There are many unclarities and problems with this idea, especially concerning the concept of ontology presupposed or implied by it. But it is argued that this idea can be improved by working out the exact meaning and place of ontology.5 It is apparent that the linguistic frameworks structuring "natural", i.e., spoken or written, languages are complex and to a large extent vague. Natural languages are pervaded by a fundamental ambiguity deriving from the double-face character of the semantico-ontological framework which features those languages: on the one hand, a world completely independent of any language, mind, theory and the like is presupposed as the ontological framework within which every semantico-ontological item

5

See Puntel 1997 and 1999a.

268

Alternative Approaches

(sentences, propositions, facts...) is attributed a fixed and definitive position, i.e., a fully determinate status; on the other hand, what such semantico-ontological frameworks, based on natural languages, really afford, i.e. express or articulate, is something like a "life-world (Lebenswelt)", i.e., a world not just independent of language or mind, but a world which is the product as well as the articulation of everyday linguistic behavior of speakers, a world understood as the totality of middle-sized objects, artifacts, and the like. In other words, the ontological status of the "world" articulated by the linguistic framework of natural languages is profoundly ambiguous and problematic. In our natural linguistic behavior we take it as something unproblematic and indeed very "natural" that the sentences we utter and the propositions these sentences express do have a fixed and definitive status. But since the semantico-ontological framework presupposed by natural languages is to a large extent vague and, thus, indeterminate, the very meaning of the alleged "determinacy" of natural language is only a surface phenomenon. In contrast to natural languages, artificial - especially formal - linguistic frameworks are well defined and, therefore, devoid of any indeterminacy. And philosophical and scientific linguistic frameworks should be required to be as much free from indeterminacy as possible. This difference between natural language, on the one hand, and artificial, philosophical and scientific language, on the other, is of great importance for working out the idea of truth for philosophical purposes. (In)determinacy of language is therefore always relative to a framework which, in the present case, is a semanticoontological framework. As will be shown below (see section 4. [2]), the idea of semantico-ontological framework(s) is central for the approach propounded in this paper. There is no "absolute" (in)determinacy, unless one admits something like "absolute" (atomic) items of whatever kind; but this appears to be extremely implausible. Being internally indeterminate, "simple" language LQ is in need of (further) qualification. It follows from this that "simple" language Z,Q is indeterminate precisely because it is in need of (further) qualification. This is a feature of LQ whose importance has been so far overlooked. [2] The central or most fundamental semantic expression is 'truth (true)' . As applied to sentences and propositions (and utterances) 6 , 'true' (or more exactly: 'it is true that...' 7 ) must be considered the supreme internal interpretive determiner of language. This presupposes that the arguments of the operator 'it is true that..', sentences

6

7

The assumption of propositions as truth-bearers is not essential for working out the root idea of truth in the present context. Later on (in section 3) this question will be addressed explicitly. In other writings the author has shown that propositions are the fundamental truth-bearers and that sentences and utterances are "derivative" truth bearers (see especially Puntel 1990, section 4.3). But note that the author defends a very peculiar conception of "propositions" or "states of affairs" (those expressions are taken synonymously in a purely terminological respect) that must be strongly distinguished from what most philosophers take "propositions" and/or "states of affairs" to be . An extensive treatment of this topic is to be found in Puntel 1990 (section 3.3), Puntel 1993, Puntel 1999b and 2001. As will be shown below in section 3, it is best to interpret 'true not as a predicate but as an operator taking sentences (and propositions) as arguments.

Truth: A Prolegomenon to a General Theory

269

and propositions, have an indeterminate status, i.e., the status of "determinables", of something that is in need of being qualified and can be qualified. This should be considered an elementary insight that can hardly be denied. Indeed, the price of rejecting this insight amounts to making the occurrence of 'true' (at least in most cases) completely redundant and, therefore, irrelevant. But this contradicts the real usage of this expression in language. To put it very succinctly: 'True' expresses an operator that takes indeterminate sentences/propositions as arguments and yields fully determinate sentences/propositions. The argument and the value of the function (operator) 'it is true that...' are in the perspective of "content" the same sentence/proposition; but in the perspective of "status" argument and value are profoundly different: as argument the sentence/proposition is indeterminate, as value it is fully determinate. This root idea of truth is in one respect a very simple one: generally understood, it is the idea of the internal "transformation" or "change" of the "status" of language: the "transition"from internal indeterminacy to full determinacy from inside language itself, i.e., afforded or made explicit by explicitly given or at least presupposed semantic vocabulary:8 But in another respect this idea is extremely complex: specifically understood, it expresses or articulates the pluridimensional and multi-faceted structure of language: its logical, ontological (and even metaphysical), epistemological, communicative etc. dimensionality. To adequately characterize the fully determinate status of language amounts to making explicit this pluridimensionality. To accomplish the formidable task mentioned would require much more space than is at the author's disposal in this paper. The purpose of the present paper is, therefore, a very modest one. It attempts to clarify only a prolegomenon to a general theory of truth. Mainly two restrictions are important and should be explicitly mentioned. First, it is intended to merely articulate what might be called the root or fundamental idea of truth. Second, this idea will be worked out only insofar as it applies to singular or particular sentences/propositions, in which the expression 'true' does not occur.9 Spe-

8

Interestingly, this "root idea of truth" — putting aside questions of terminology and some fundamental statements - is to be found in Frege. In his "On Sinn and Bedeutung (1892) he writes: We can never be concerned only with the Bedeutung of a sentence; but again the mere thought alone yields no knowledge, but only the thought together with its Bedeutung, i.e. its truth-value.

Judgements can be regarded as advances [Fortschreiten] from a thought to a truth-value. (Beaney 1997, p. 159; emphasis of the last sentence added) 9

In his Truth and Proof (1969) Tarski considers the formulations '(1) 'Snow is white' is true if and only if snow is white' and '(Γ) 'Snow is white' is false if and only if snow is not white' and claims that "... (1) and (1') provide satisfactory explanations of the meaning of the terms 'true' and 'false' when these terms are referred to the sentence 'snow is white'. We can regard (1) and (1') as partial definitions of the terms 'true' and 'false', in fact, as definitions of these terms with respect to a particular sentence" (p. 64; emphasis added). In the same vein, this paper aims at working out the explication of the term 'true' only with respect to particular sentences/propositions. The question whether an adequate explication of'true' should be understood as a definition in the strong sense or as an axiomatic theory will be briefly addressed in section 4.

270

Alternative Approaches

cial attention will be devoted to making explicit the specific character of the semantic vocabulary in opposition to pragmatic vocabulary and to other contextual factors. But beyond this point it will not be possible to do more than sketching the kind of semantics and ontology suitable to making intelligible the propounded explanation of the concept of truth. 10

1.3 The relationship between semantic vocabulary, pragmatic vocabulary and "communicative" contexts Roughly, the mainstream of contemporary semantics could be characterized by a maxim R. Brandom has formulated with admirable precision and clarity: "Semantics must answer to pragmatics." The sketched conception runs directly against this mainstream. [1] According to many contemporary semanticists pragmatic factors are essential to explaining the semantic features of language. Among such factors speech acts, like "assertion", play the decisive role with respect to what might be called the theoretical segment of language. But this is a mistake, since the specific character of semantic vocabulary is ignored. Using Brandom's terminology, the maxim should be: Pragmatics must answer to semantics. Indeed, as has been already remarked above, semantic vocabulary does not contain any reference to speakers, language users, subjects, speech acts or communicative contexts. A sentence/proposition need not be asserted to count as true (or false); at best, it could (perhaps must) be said that the truth of a sentence/ proposition entails that it must be assertable. But the pragmatic entity called "assertion" can hardly be understood without assuming its being directed towards truth. C. Wright correctly speaks of the "platitude" "that to assert is to present as true". 11 On the basis of these considerations it comes as a surprise that many truth theorists seem to countenance a link (an equivalence or even identity) between truth and assertion; but this is the result of confusing or at least not sufficiently distinguishing two different dimensions. [2] In subsections 1.1 and 1.2 we introduced the concept of "simple" language or LQ, i. e., language without semantic vocabulary, and we characterized it as being internally indeterminate. But we stressed the fact that simple language can be a "concrete" language (in the sense of language endowed with a fully determinate status), when it is determined either by an external-internal factor, i.e., by pragmatic vocabulary, or by a purely externalfactor, i.e., by the context of pure communicative utterance or discourse. The question now arises as to how those three factors affording full determination of

10 O n this reason the reader is asked to consult other writings of the author, especially Puntel 1990 (ch. 3 and 4), 1993, 1999a, 1999b, 2000, and 2001. 11 Wright 1992, p. 34. This insight has been already formulated by Frege: In order to put something forward as true [Um etwas als wahr hinzustellen], we do not need a special predicate: we only need the assertoric force with which the sentence is uttered. (Frege, "Logic in Mathematics," in Frege 1979, p. 233).

Truth: A Prolegomenon to a General Theory

271

language are to be seen more precisely. Do they afford "full determination" in exactly the same sense? This question constitutes a central issue underlying in many ways most discussions about the truth topic in contemporary philosophy. The thesis to be propounded in the remainder of this subsection is this: Purely external determination of language, given by pure communicative contexts, presupposes (or relies upon) the external-internal determination afforded by pragmatic vocabulary; in its turn, external-internal determination presupposes (or relies upon) purely internal determination brought about by semantic vocabulary. It follows from this thesis that the fully determinate character of language is ultimately given by semantic vocabulary; every form of language not explicitly containing semantic vocabulary can be considered fully determinate only by - explicitly or at least implicitly - presupposing the determination of language afforded or made explicit by semantic vocabulary. Let us explain. Simple language LQ (in his "declarative" segment) given in purely communicative discourse of everyday life, i.e., language containing no pragmatic vocabulary like 'to assert' and no semantic vocabulary like 'true', can be considered fully determinate only by assuming that utterances of (declarative) sentences are intended as assertions. For instance, if the sentence 'Snow is white', uttered in the context of a purely communicative discourse, is intended to having assertoric character, it will be attributed fully determinate status. Indeed the sentence is intended as saying: Real snow is really white. But from this it follows that it is just pointless to simply "present" (or "utter", "write down"...) the sentence 'Snow is white' — without making explicit in what sense and under what conditions it should be taken. What should be made explicit is the fundamental circumstance that the sentence can be attributed a fully determinate status only if it is presupposed or made explicit that it is meant as an assertion. But if so, then to point to the sentence (for instance, in the context of the theory of truth) does make sense only if one makes explicit what "assertion" means. It is easy to see that Tarski's famous biconditionals according to his T-schema "'p' is true if and only if p" are highly ambiguous, since they do not make explicit how the sentence represented by 'p' on the right-hand side is to be exactly understood. Clearly, so long as the status of 'p' on the right-hand side is not made explicit, such biconditionals are useless. Now, if it is presupposed that p' on the right-hand side is taken as (elliptically) meaning "it is asserted that p", then two problems arise for those philosophers who rely on such biconditionals as being basic for a theory of truth. The first problem is induced by the fact that, if it is presupposed that 'p' on the right-hand side is meant as being an assertion, then just this character of 'p' should be worked out or made explicit; otherwise, the only relevant point would be missed. The second problem derives from the result one would reach in making explicit the assertoric character of 'p' (on the right-hand side). One would obtain: "'p' is true if and only if it is asserted that p". But this would be circular, since "to assert that p" is tantamount to "to present ρ as true". So, externally determined language as well as externally-internally determined language presupposes the internally given determination or, as it were, the "self-determination" of language afforded by semantic vocabulary. From this it follows first that a sentence qualified as 'true' cannot simply be equated to the "sentence itself", i.e., the

272

Alternative Approaches

sentence without 'true' ('p') or to the sentence endowed only with a pragmatic vocabulary ('it is asserted that p'); second that 'true' is by no means "redundant" or "dispensable" in the sense ail "redundancy theories of truth" and all "deflationary theories of truth" claim it is. Those theories assert that 'true' is dispensable when attributed to "singly given sentences" 12 , or "to sentences that are explicitly before us". 13 But this is a mistake, as will be shown in more detail in the next section.

2. A fundamental

mistake: the ignorance of the proper function of semantic

vocabulary

[1] The function of semantic vocabulary is to make language determinate internally, from within language itself, independently of any other purely external as well as external-internal factors. Indeed, meaningful occurrence of semantic vocabulary does presuppose a kind of indeterminacy of language and at the same time does surmount this indeterminacy. To have ignored this function of semantic vocabulary is a fundamental mistake of most theories of truth. This can be best illustrated by considering the widespread usage most truth theorists make of Tarski's biconditionals. What they don't account for is the fact that the sentence occurring on the left-hand side of the biconditionals has an indeterminate status whereas the (same) sentence occurring on the right-hand side is endowed with a fully determinate status. Since 'true' is an internal qualifier or determiner, the linguistic as well as the non-linguistic item (i.e., the sentence and the proposition expressed by the sentence) to which it is attributed must be considered qualifyable or determinable, i.e., must be taken as being indeterminate. This is easily shown both in the case that 'true' is understood as a predicate of a nominalized sentence (and of the proposition expressed by such a nominalized sentence) and in the case that it is interpreted as an operator ('it is true that ...') taking genuine (non-nominalized) sentences (and propositions expressed by such sentences) as arguments. In the first instance the determinable sentence is put in quotation marks or is indicated by using a "structural-descriptive name" of the sentence1^; since the biconditional brings about the "disquotation" of the sentence or the "translation" of the structural-descriptive name of the sentence into "the sentence itself", it is manifest that the nominalized sentence changes its status: it becomes (fully) determinate. If 'true' is taken as an operator, then the biconditionals, as they are usually formulated and interpreted, clearly seem to contain exactly the same (genuine, non-nominalized) sentence both on the left-hand side and on the right-hand side of the equivalence. The consequence most analytic philosophers draw from this form of the biconditionals is the claim that 'true', at least as attributed to singly given sentences, is redundant and, therefore, dispensable. But this, as has been already hinted at in the preceding subsection, is a mistake on two accounts. First, speaking in general terms, it seems highly implausible and even absurd to assume that the semantic vocabulary oc-

12 13 14

Quine 1970, p. 11. Quine 1987, p. 2 1 4 . See Tarski 1933/1983, p. 1 5 6

Truth: A Prolegomenon to a General Theory

273

curring in language is redundant (with respect to particular sentences). Moreover, such a claim is refuted by the understanding speakers of language associate with their usage of 'true'. Second, language devoid of any explicitly given or implicitly presupposed semantic vocabulary is simply unintelligible, since such a language is neither interpreted nor interpretable. Indeed, if the singly given sentence on the right-hand side of the biconditional is attributed any determinate "meaning" at all, then it must be taken as being (at least) an assertoric sentence. But if so, then the sentence can be understood only if one knows what an assertion is. But if "to assert" is "to present as true", then the sentence on the right-hand side presupposes the understanding of truth. And this shows that every "form" or "shape" of determinate language is based on the previously understood concept of truth. But it should be considered a truism that the concept of truth is expressed by a semantic expression. [2] To what extent the described mistake pervades the literature on theories of truth might be illustrated by an example. In his Philosophy of Logic Quine presents a highly significant reading of Tarski's biconditionals: No sentence is true but reality makes it so. The sentence 'Snow is white' is true, as Tarski taught us, if and only if real snow is really white. The same can be said of the sentence 'Der Schnee ist weiss'; language is not the point. In speaking of the truth of a given sentence there is only indirection; we do better simply to say the sentence and so speak not about language but about the world. So long as we are speaking only of the truth of singly given sentences, the perfect theory of truth is what Wilfrid Sellars has called the disappearance theory of truth.15 The reason why this example is so significant is that it clearly confirms the manifest unclarity and incoherence pervading contemporary theories of truth developed in the wake of Tarski's Truth Schema. On the one hand, Quine wants us to read the biconditional '"Snow is white' is true if and only if snow is white" as saying '"Snow is white is true' if and only if real snow is really white", the "reality" of "snow" and of "white" being, thus, what is induced by the occurrence of 'is true' on the left-hand side of the biconditional; on the other hand, he interprets 'is true' occurring in singly given sentences as perfectly explained by the "disappearance theory of truth", i.e., according to Quine, 'is true' is to be taken as a simple device of disquotation 16 with the result that the explanans is "the sentence itself". But what is "the sentence itself"? How to interpret it? According to Quine it is obvious that the "sentence itself" is perfectly meaningful in the sense of being perfectly well determined in such a way that it constitutes the ultimate basis for all statements. When he explains "we do better simply say the sentence and so speak not about language but about the world" he takes the second sentence of the conjunction to be the explication or paraphrasing of the first sentence: "To simply say the sentence" is tantamount to "speak not about the language but about the world". But such an assumption is by no means the expression of a matter of fact: there is no matter of fact of an "unqualified" sentence (a "sentence itself") being a sen-

15

Quine 1970, pp. 1 0 - 1 1 (emphasis added).

16 Cp. ibid.., p. 12.

274

Alternative Approaches

tence not about the language, but about the world. In other words, Quine's claim is based on some implicit presuppositions with respect to "the sentence itself". The most fundamental assumption he implicitly presupposes is that "the sentence itself" is an assertion about the world. But then he should make explicit what an assertion is. If, as shown above, to assert is to present as true, then to say that the sentence speaks about the world is to say that the sentence itself is taken as being assertoric and, thus, it is presented as true. But then "truth" is presupposed by the sentence itself taken as an assertion. Consequently, a sentence qualified as true cannot be reduced to "the sentence itself". It is apparent that in Quine's treatment of the truth topic some fundamental things go awry. His view is the result of an oversimplification; no room is left for a detailed and differentiated analysis. [3] No other theoretical piece has been more important and influential in contemporary theories of truth than Tarskis famous "Schema (or Convention) T": (TS) χ is a true sentence if and only ifp^ (with 'p' replaced by any sentence of the language to which 'true' is applied, and 'x' replaced by a name or a structural description of this sentence). Many philosophers dealing with the question of truth have accepted and understood this Schema as being clear and unambiguous. But in recent years discussions about the exact meaning and the theoretical relevance of the Schema have intensified. Surprisingly, to the author's knowledge no one among Tarski scholars, critics and followers seem to have noticed what might be called the most fundamental problem at the heart ofTarski's semantic theory of truth: the problem posed by his Truth Schema. But no exhaustive analysis of Tarski's own understanding of the Schema will be presented here, since this is not required for attaining the aim of this paper. Tarski's own original Truth Schema (TS) and its different forms have been interpreted as expressing a deflationary conception of truth. As will be shown below, this conception contains an interesting and valuable insight. But this insight cannot be articulated unless a deep mistake underlying Tarski's Schema and its forms is corrected beforehand. The mistake is to be found in Tarski himself in surprisingly clear terms. Tarski started articulating what he called a semantical definition of truth for the sentences of colloquial language which he expressed in the following words: (T)

A true sentence is one which says that the state of affairs is so and so, and the state of affairs indeed is so and so. 18

Tarski claims that the intuitive meaning and the intention of this formulation is quite clear and intelligible, regardless of the circumstance that it leaves much to be desired concerning formal correctness and clarity. In order to find out a starting-point for the attempt to make the intention of (T) more definite and to give it a correct form, he considers "certain sentences of a special kind" which could be taken as partial defini-

17 Cp. Tarski 1933/1983, p. 155. 18 Tarski 1933/1983, p. 155 (emphasis added).

Truth: A Prolegomenon to a General Theory

275

tions of the truth of a sentence; those sentences are sentences of the form "x is a true sentence". He, then, introduces his famous "general schema" (TS) quoted above. Clearly, (TS) is intended to correctly and formally articulate "the intuitive meaning and the general intention" of (T). But now it appears that (TS) does not explicitly articulate what (T) explicitly, though not formally, says, i.e., what according to Tarski has to be taken as "the intuitive meaning and the general intention" of (T). This is easily shown. The relative clause in (T) reads "[A true sentence is one] which says that the state of affairs is so and so"; the second main clause of (T) contains exactly the same words as the relative clause but one, namely the small word 'indeed': "... and the state of affairs indeed is so and so" (emphasis added). 'Indeed' is added, so that the second main clause is not simply a repetition of the relative clause. The decisive point of (T) is articulated in this small word 'indeed'. But where does the weight of this decisive word appear explicitly in (TS)? Nowhere. Tarski simply presupposes or assumes that the "standing sentence" 'p' on the right-hand side of the equivalence schema is to be understood as "meaning" "... indeed p". But if so, why not make this fundamental circumstance explicit? The sentence 'p', by itself, i.e., taken as an unqualified standing sentence, does not say or contain something like the qualification expressed by 'indeed'. But the biconditionals would not make sense unless one takes the sentence on the right-hand side as being qualified or determined by 'indeed'. Quine once realized this, namely in the above quoted passage from his Philosophy of Logic. To understand the sentence 'Snow is white' in the sense of "real snow is really white" is to understand the sentence according to its fully determinate status, both semantically and ontologically. The function - or, if one likes, the "meaning" - of 'is true' is precisely to express the transformation or the transition of the indeterminate status of a sentence ('Snow is white') into its fully determinate status ('Real snow is really white' or 'Snow is indeed white'). It is precisely this status or character of the sentence on the right-hand side of the biconditional that should be made explicit as what is afforded by the occurrence of 'true' on the left-hand side. But this is not explained by something like a "disappearance theory of truth". The negative upshot of these reflections is a negative judgement about the most fundamental tenet permanently repeated and defended by those theories of truth that are called "deflationary" in contemporary philosophy. From this critique of the "deflationary" conceptions of truth it does not follow that the only alternative are the socalled "substantive theories of truth". As will be shown below, the distinction between "deflationary" and "substantive" theories is not an adequate one, since it is highly ambiguous. The explanation of'is true' to be delineated in the next section can be considered in one important respect a deflationary one, in another respect it can be seen as the result of a rejection of truth deflationism, as this position is understood in contemporary philosophy.

276 3. 'It is true that...'

Alternative Approaches

as a composite function: a "cataphoric" conception of truth19

In this section a brief explication of 'true' will be sketched on the basis of the considerations presented in sections 1 and 2. The explication is restricted to the concept of truth attributed to particular or individual sentences (and propositions) in which the term true' does not occur (see footnote 9). In this sense the explication is restricted to the elementary concept of truth or to the elementary case of truth. The explication is intended to work out only a prolegomenon to a general theory of truth. An important further restriction must be pointed out: the explication of 'true' to be propounded aims at explaining the concept of truth not for what is called "natural language", but only for philosophical (and scientific) language. A philosophical (and scientific) language can be globally characterized as a language suitable for pursuing theoretical aims. Therefore, it should at least be required to be clear, unambiguous, devoid of "natural" (in the sense of "non-examined") assumptions in all areas of philosophical (and scientific) interest. In the space of this essay it will not be possible to properly justify all assumptions and claims argumentatively. The exposition will be confined of necessity to sketching the outlines and motivations for the root idea of truth resulting from the approach presented in sections 1 and 2. But first a preliminary question must be broached. From the syntactical point of view, there are three occurrence forms of the term 'true' on the elementary level20 : "p' is true'; 'That ρ is true'; 'It is true that p'. According to the first occurrence 'true' is a predicate of a sentence, according to the second form it is a property of a proposition, according to the third form it is an operator taking sentences (and propositions) as arguments. Most philosophers consider 'true' a predicate; many among them claim that this predicate expresses a genuine or substantive property, others deny this claim. 21 Some philosophers reject the thesis that '... is true' is a predicate proper at the level of the deep structure of language. Thus, according to the original prosentence theory of truth 2 2 '... is true' is a syncategorematic fragment of a semantically atomic generic prosentence in the sense that it functions anaphorically and not descriptively; in other words, it marks the use of a prosentence which by analogy to pronouns has an anaphoric antecedent. Contrary to this original version of the theory, R. Brandom sees '... is true' as a prosentence-forming operator, it applies to a term that is a sentence nominalization or that refers to or picks out a sentence tokening. It yields a prosentence which has that tokening as its anaphoric antecedent. 23 Brandom's version of the prosentence theory was highly inspirational in developing the explanation presented below. In particular, the denial that '... is true' is a predicate

19 20 21 22 23

Parts of this section draw heavily on the author's paper "What Does '... Is True' ('It Is True That...') Express?" 2000. That is, on the level of particular (atomic) non-quantified sentences, in which the term true' does not occur. Cp. Horwich 1990/1998. See Grover/Camp/Belnap 1975. Cp. Brandom 1994, p. 305.

Truth: A Prolegomenon to a General Theory

277

expressing a descriptive property is accepted. There are many reasons that support this position, as Brandom has shown. But the main reason (not to be found in Brandom) lies in the function o f semantic vocabulary just worked out in section 2. T h e function o f making language (fully) determinate is not adequately captured by saying that it is a descriptive property o f expressions o f the language. Thus, 'true' as the central semantic expression should not be considered a predicate (expressing a substantive property). Another reason for not taking 'true' as a predicate derives from ontological considerations. I f the Principle o f Semantic Compositionality is rejected in favor o f the Principle o f Contextuality (or Context Principle), then the language-world connection, which constitutes a central feature o f the concept o f truth defended in this paper, is best explained in terms o f what sentences interpreted non-compositionally express (propositions or states o f affairs and facts), instead o f what the subsentential constituents refer to (singular terms - objects) or designate (predicates - attributes). Consequently, the "ontological dimension" o f 'true' is adequately accounted for only by considering 'true' not as a predicate, but as an operator. This fundamental topic, whose impact on an adequate explanation o f the concept o f true can hardly be exaggerated, has been extensively dealt with in the author's paper "Truth, Non-Compositionality, and Ontolo» 24

gy ·

Now, contrary to Brandom, it is claimed that 'it is true that' is not an operator that applies to a sentence tokening and yields a presentence, such that this presentence has that tokening as its anaphoric antecedent. Just the opposite view emerges from the basic idea that the function o f semantic vocabulary is to make language fully determinate; i.e., 'it is true that' is an operator that applies to sentences (and propositions) with a merely indeterminate status and yields not anaphorically understood PROsentences, but another kind o f sentences (and propositions) which can be dubbed "PERsentences (PERpropositions)". (The expressions 'PERsentence/PERproposition' are formed by analogy to Brandom's 'PROsentence', the prefix ' P E R ' being taken from 'PERfect' or 'PERform'.) Such PERsentences (and PERpropositions) are explained cataphorically in the following sense: they have as their consequents (not: antecedents) the "same" sentences (propositions), to which the truth operator applies, except for the significant fact that those sentences (propositions) adopt a new status as consequents, to wit, a fully determinate status. Thus, 'true' is not a PROsentence-forming operator, as Brandom asserts; rather it is a PERsentence-(PERproposition-)forming operator in the sense indicated. I f the PERsentence-(PERproposition-)forming operator 'it is true that...' is accepted and if the PERsentences (PERpropositons) are interpreted along the lines just described, then it seems more adequate to explain the notion o f truth as the composition o f two functions, that is, as a composite function. To be sure, one could simplify matters and say: 'It-is-true-that' is a function taking indeterminate sentences/propositions as arguments and yielding the "same" sentences/propositions endowed with a fully determinate status. But in so doing, the internal working or the fine-grained structure

24

Puntel 2 0 0 1 .

Alternative Approaches

278

of the operator 'it-is-true-that...' would not be made explicit in an adequate way. At least for philosophical purposes, one has to explicitly take into account two steps in the "working" of the operator 'It-is-true-that...' The first is the (result of the) application of this operator to an (indeterminate) sentence/proposition: this result is a PERsentence/ PERproposition ('It is true that p'), i.e., a sentence/proposition having a cataphoric status, i.e., the status of being "(in)directed" towards or of referring to a consequent.2^ The second step consists in making explicit what the consequent is: it is the "same" sentence/proposition with the difference that now it is endowed with a fully determinate status. The two steps are two functions which, combined, constitute a composite function. Let those two functions be ' T ' ('it-is-true-that' as a PERsentence/ PERproposition-forming operator) and 'T + ' ('it is true that...' as a "cataphorically" working operator), respectively. Taken together, Τ and T + express T, the "meaning" or the "notion" of 'true'. Accordingly, the composite function can be formulated thus: (Τ)

Τ = T+

0

T*

In particular, the two functions T* and T + can be represented as follows: (Τ*)

Τ* : X

>

Y

(where 'X' is the set of indeterminate sentences (propositions), Ύ the set of the resulting PERsentences (PERpropositions)) For particular members of X, i.e., sentences ρ (and the propositions expressed by them), and particular members of Y, i.e., PERsentences p P E R (and PERpropositions expressed by them): Ρ (T+)

T

~ +

:

PPER

Y

> Ζ

(where ' Y is the set of PERsentences (PERpropostions) p P E R , 'Z' the set of the sentences 'p' (and the propositions expressed by them), but now endowed with a fully determinate status) For particular PERsentences p P E R and fully determinate sentences p: PpER

^

Ρ

It is clear that the function of'truth (true)' consists in bringing about or affording the fully determinate status of sentences (propositions). To say of a sentence (proposition) that it is true is to fully determine this sentence (proposition) or to take it as being fully determinate. This theory of truth could, therefore, be dubbed determination theory of truth. There is no place for a relation like the correspondence relation, since there is absolutely no need to admit of two different kinds of entities of which such a rela-

25

α ν α φ ο ρ ά and κ α τ α φ ο ρ ά have opposite (correlative) meanings: carrying back (coming up, rising, ascent) and carrying down (rushing down, having a downward tendency), respectively. The English derivatives of both expressions have also a grammatical sense: 'anaphoric' has the meaning of referring back to or substituting for a preceding word or group of words; 'cataphoric' means the use of a word to refer to a following word or group of words, as in the phrase 'as follows.'

Truth: A Prolegomenon to a General Theory

279

tion could or should be established. That is, there is no entity on the side of language or the mind or the thinking (like propositions or thoughts or states of affairs understood as some sort of mental or linguistic entities) and neither is there on the side of the world some ontological entity corresponding to the proposition taken as a different entity in the indicated sense. Rather, if we call "proposition" what is expressed by a declarative sentence, then a true sentence expresses a true proposition, and a true proposition is simply a worldly fact, an ingredient of the world. This last feature has been dubbed in recent years "identity theory of truth". 2 6 There is no correspondence relation, there is simply identity between the proposition (or states of affairs) expressed by a true (that is, a fully determinate) sentence and a fact in the world. 27

4. Concluding remarks [1] Tarkis T-biconditionals, duly and profoundly reinterpreted, can be seen as an excellent articulation of the notion of truth just worked out. The reinterpretation can be formulated using the following notation: It-is-true-that ρ if and only if ρ where 'p' is an indeterminate sentence (expressing an indeterminate proposition), 'p' (in bold character) is the "same" sentence as 'p', except that this sentence now has a new status, a fully determinate status (and such that 'p' expresses the "same" proposition as 'p', except that the proposition expressed by 'p' is endowed with a fully determinate status). [2] But now a fundamental question arises for the propounded theory of truth: How is the concept of the fully determinate status of sentences (propositions) to be worked out exactly? Within the limited space of this paper only some hints at an answer could be given. What in the outlined theory of truth is called the fully determinate status of sentences (propositions) can be generally characterized as the exact, definite and ultimate place the sentence (proposition) occupies in the chosen or presupposed framework. Carnaps conception of "linguistic frameworks" should be considered of fundamental importance in this respect. 28 Lack of space does not permit to explain this matter in more detail. But it is worth stressing the following central point: The concept of the fully determinate status of sentences/propositions is an extremely complex concept and, consequently, it induces a formidable task which has to be situated at the intersection of several philosophical disciplines, especially semantics, logic, ontology, epistemology, pragmatics. To illustrate the task that should be tackled just consider this: In some way or other most theories of truth speak of "the world" or of an ontological dimension in connection with

26 See Puntel 1999b. 27 See also Brandom 1994, pp. 327-333. 28 See Carnap 1950/1956; Puntel 1997 and 1999.

280

Alternative Approaches

"truth". But they do not in the least - not even in very general terms - make explicit in what sense they are referring to the ontological dimension ("the world"). A case in point is Quine's famous statement: In speaking of the truth of a given sentence there is only indirection; we do better simply to say the sentence and so speak not about language but about the world?**

This topic has been extensively dealt with by the author in his "Truth, Sentential Non-Compositionality and Ontology". 30 [3] The sketched explanation of 'true' with respect to particular sentences (and propositions) cannot pretend to be taken as an adequate explanation of the meaning or the notion of truth simpliciter. In order to develop a general theory of truth a great deal of work remains to be done. Central problems have not even been mentioned, for instance, the problem of falsity, the truth paradox etc. But to accomplish this aim, the clarification of what has been called "the elementary truth case" should not be underestimated. If the explanation of the notion of truth is presented as a definition (in the strong sense) of the term 'true', then the definiens is nothing more than the briefly described composite function: Τ = T+ ° Τ . In this respect the explanation could be considered "deflationary" (according to the sense commonly attributed to this expression in contemporary philosophy), since no explicit articulation of something like the languageworld connection traditionally associated with the concept of truth is contained in such a definition. But to qualify the sketched view as "deflationary" is not correct, since the presented view in no way does countenance one of the central tenets of every brand of deflationism: acceptance of the "original" (uninterpreted) T-biconditionals either in the disquotational form ("'p' is true if and only if p") or in the propositional form ("It is true that ρ if and only if p"). On the other hand, it would be equally inadequate to characterize the propounded explanation as a "substantive" one, at least insofar the expression 'substantive explanation is associated with something like a correspondence theory in the commonly accepted sense. This "definitional" way or procedure cannot be taken as unproblematic for philosophical purposes, since to present the explanation of'true' only as a definition does not make explicit or articulate the dimensions of the concept of "fully determinate status" (of sentences/propositions qualified as true). To be sure, this philosophical deficiency would be overcome if those "dimensions", especially the semantic and the ontological dimension, are worked out independently of the task of explaining the concept of truth. This is a question of how to methodologically develop a systematic philosophical conception. As a matter of fact, contemporary philosophical theories of truth, which in one way or another intend to give a definition (or explication or characterization...) of truth, do not work out the semantic and the ontological "dimension" of truth.

29 30

Quine 1970, p. 11 (emphasis added). Puntel 2 0 0 1 .

Truth: A Prolegomenon to a General Theory

281

An alternative way of dealing with the truth topic which avoids the problems just alluded to does suggest itself: instead of simply presenting a strong and explicit definition of'true', it seems more suitable for philosophical purposes to develop an axiomatic theory of truth.31 Such a theory should be conceived of as containing axioms in which at least the most important logical, semantical and ontological aspects of the concept of a fully determinate status of sentences/propositions are explicitly formulated.

Bibliography Beaney, M „ 1997, The

FREGE READER.

Oxford: Blackwell

Boisvert, D. R., 1999, "The Trouble With Harrison's 'The Trouble With Tarski'", in The Philosophical Quarterly 49, pp. 376-383. Brandom, R. B., 1994, Making It Explicit. Reasoning, Representing, and Discursive Commitment. Cambridge, Mass. - London: Harvard University Press. Carnap, R., 1950/1956, "Empiricism, Semantics and Ontology", in Revue Internationale de Philosophie 4 (1950), pp. 20-40. Page references are to the reprint in R. Carnap, Meaning and Necessity. 2nd ed. Chicago: The University of Chicago Press, 1956, pp. 205-221. Frege, G., 1979, Posthumous Writings, ed. by H. Hermes, F. Kambartel, and F. Kaulbach, transl. by P. Long and R. White. Oxford: Blackwell. Grover, D./Camp, J. L./Belnap, N. D., Jr., 1975, "A Prosentential Theory of Truth", Philosophical Studies 27, pp. 73-125. Halbach, V., 1996, Axiomatische Wahrheitstheorien. Berlin: Akademie Verlag. Harrison, J., 1998, "The Trouble With Tarski", in The Philosophical Quarterly 48, pp. 1-22. Heck, Jr., R. G., 1997, "Tarski, Truth, and Semantics", in The Philosophical Review 106, pp. 533-544.. Hintikka, J., 1996, The Principles of Mathematics Revisited. Cambridge: Cambridge University Press. Hintikka, J., unpublished, "Post-Tarskian Truth". Ms., pp 1-29. Hintikka, J./Sandu, G., 1999, "Tarski's Guilty Secret: Compositionality", in J. Wolenski/E. Köhler (eds.), Alfred Tarski and the Vienna Circle. Austro-Polish Connections in Logical Positivism. Vienna Circle Institute Yearbook (1998). Dordrecht/Boston/London: Kluwer, pp. 217-230. Horwich, R, 1990/1998, Truth. 2nd edition (1st edition 1990). Oxford: Clarendon Press. Hugly, Ph./Sayward, Ch., 1983, "Can a Language Have Indenumerably Many Expressions?", in History and Philosophy of Logic 4, pp. 73-82. Puntel, L. B., 1990, Grundlagen einer Theorie der Wahrheit. Berlin-New York: de Gruyter. Puntel, L. B., 1993, "The Context Principle, Universals and Primary States of Affairs", in American Philosophical Quarterly 30, pp. 123-135. Puntel, L. Β., 1997, "Metaphysikkritik bei Carnap und Heidegger: Analyse, Vergleich, Kritik", in LOGOS 4, pp. 294-332. Puntel, L. B., 1999a, "On The Logical Positivists' Theory of Truth: The Fundamental Problem and a New Perspective", in Journal for General Philosophy ofScience 30, pp. 101-130. Puntel, L. B., 1999b, "The 'Identity Theory of Truth': Semantic and Ontological (Metaphysical) Aspects". To appear in Analyomen III. Proceedings of the 3r