Current Issues in Theoretical Philosophy: Volume 3 Prospects for Meaning 9783110216882, 9783110196238

Citation preview

Prospects for Meaning

Current Issues in Theoretical Philosophy Edited by Richard Schantz

Vol. 3

De Gruyter

Prospects for Meaning Edited by Richard Schantz

De Gruyter

ISBN 978-3-11-019623-8 e-ISBN 978-3-11-021688-2 Library of Congress Cataloging-in-Publication Data A CIP catalogue record for this book is available from the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.dnb.de.

© 2012 Walter de Gruyter GmbH & Co. KG, Berlin/Boston Typesetting: Apex CoVantage, LLC Printing: Hubert & Co. GmbH und Co. KG, Göttingen ∞ Printed on acid-free paper Printed in Germany www.degruyter.com

Contents Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

I

1

Meaning and Reference

The Puzzle That Never Was—Referential Mechanics . . . . . . . . . . Joseph Almog

21

Reference and Meaning. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . William P. Alston

35

Still Against Direct Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . Michael Devitt

61

On Meaning, Meaning and Meaning . . . . . . . . . . . . . . . . . . . . . . Ruth Garrett Millikan

85

On Referents and Reference Fixing . . . . . . . . . . . . . . . . . . . . . . . Howard Wettstein

107

II Truth-theoretic Semantics On Some Examples of Chomsky’s . . . . . . . . . . . . . . . . . . . . . . . . Graeme Forbes

121

Truth, Meaning and Contextualism . . . . . . . . . . . . . . . . . . . . . . . Samuel Guttenplan

143

Expression, Truth, Predication, and Context: Two Perspectives . . . James Higginbotham

171

vi

Contents

Prospects for a Truth-conditional Account of Standard Meaning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Guy Longworth

195

A Very Large Fly in the Ointment: Davidsonian Truth Theory Contextualized . . . . . . . . . . . . . . . . . . . . . . . . . . . R.M. Sainsbury

223

From Truth Conditions to Structured Propositions . . . . . . . . . . . Richard Schantz Five Flies in the Ointment: Some Challenges for Traditional Semantic Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . Gabriel M.A. Segal

259

287

III Meaning Skepticism Against Meaning-Skepticism . . . . . . . . . . . . . . . . . . . . . . . . . . . . Paul Horwich

311

Phenomenal Intentionality and Content Determinacy . . . . . . . . . Terry Horgan and George Graham

321

Semantic Realism and the Argument from Motivational Internalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Alexander Miller

345

IV The Metaphysics and Epistemology of Meaning and Content Meaning and Content in Cognitive Science . . . . . . . . . . . . . . . . . Robert Cummins and Martin Roth

365

Knowledge of Meaning and Epistemic Interdependence . . . . . . . . Jennifer Hornsby

383

Contents

vii

Meaning and Ontology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ernest Lepore and Francis Jeffry Pelletier

399

Matter and Meaning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Colin McGinn

435

Meaning, Communication and Knowledge by Testimony . . . . . . . Douglas Patterson

449

Semantics Without Meanings? Sellarsian “Patterned Governed Behavior” and the Space of Meaningfulness . . . . . . . . . . . . . . . . . Jaroslav Peregrin

479

Externalism and Inexistence in Early Content . . . . . . . . . . . . . . . Georges Rey

503

Propositions, What Are They Good For? . . . . . . . . . . . . . . . . . . . Stephen Schiffer

531

Meaning as a Biological and Social Phenomenon . . . . . . . . . . . . . John R. Searle

553

Three Kinds of Meanings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Patrick Suppes

567

V Formal Semantics Requirements on a Theory of Sentence and Word Meanings . . . . . Wilfrid Hodges

583

Quantification and Anaphora in Natural Language . . . . . . . . . . . . Gabriel Sandu and Justine Jacot

609

Equivalence of Semantic Theories . . . . . . . . . . . . . . . . . . . . . . . . Thomas Ede Zimmermann

629

Notes on Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

651

viii

Contents

Index of Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

657

Index of Subjects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

663

Introduction This book is the last of a series of three volumes dedicated to central debates in contemporary theoretical philosophy. The title of volume one is What is Truth?, the title of volume two is The Externalist Challenge. The present volume is a collection of new essays that deal with some of the most basic questions in contemporary philosophy of language and mind, questions concerning both the metaphysics and the epistemology of meaning. Many of the significant issues that must be confronted in developing an adequate theory of meaning are raised, and some of the principal theoretical stances that might be taken with regard to them are discussed. My hope is that this anthology will instigate further fruitful research in the philosophy of language. In its glorious days, during the second half of the twentieth century, philosophy of language was deemed by many authors to be first philosophy, to be the foundation for all the other branches of philosophy. They were confident that it would provide us with the necessary conceptual tools to resolve, in effect, the grand problems in metaphysics, epistemology, and the philosophy of mind. Although those days are gone, philosophy of language still occupies a position of fundamental significance in the contemporary philosophical landscape. Meanwhile, philosophers of language have come to appreciate that investigations of language ought to be an interdisciplinary enterprise. In its ambitious endeavor to come to understand the nature of language and the indispensable role it plays in our lives, philosophy of language has, in recent decades, immensely profited from the impressive progresses that were made in other regions of philosophy— such as, above all, the philosophy of mind—as well as in neighboring disciplines—such as linguistics, logic and psychology. It seems to be relatively uncontroversial that the words, phrases, and sentences of a language mean something—that they have meaning. Consider the word “horse”. It has a distinctive meaning in English in virtue of which it should be applied to horses and only to horses. Moreover, this word apparently has the same meaning as the German word “Pferd” and the French word “cheval”. Surely, “horse” means something different than the English word “cow” and should not be translated into the

2

Introduction

German “Kuh”. Generally, we suppose that speakers of a natural language understand its expressions, or know their meanings. But how are these extraordinary facts possible? In virtue of what properties are expressions meaningful? What distinguishes a speaker’s production of meaningful sounds from his production of meaningless sounds and scratches? How, in a word, do we get from physics to semantics? It is the demanding task of semantic theories, or theories of meaning, to answer such challenging questions, that is, to explain what it is for words and sentences—or, more precisely, for sequences of noises and marks—to possess the particular meanings they possess, and why they possess that meaning rather than some other. They should also be able to explain the remarkable competence of human beings to employ language to express and communicate their thoughts about objects, events and state of affairs in the world. Ultimately, theories of meaning are faced with the difficult and comprehensive job to deliver theoretical descriptions and explanations of all the diverse and complex phenomena of linguistic meaning.

1. Meaning skepticism The title of this volume is Prospects for Meaning. There are, however, philosophers who hold that linguistic expressions do not really, as we commonly and perhaps naively suppose, possess meanings at all. They espouse a form of meaning skepticism, a form of the view that there are no facts of the matter as to what linguistic expressions mean and hence that attributions of meaning to linguistic expressions are neither true nor false. So, according to them, the prospects for meaning are pretty bleak. I think that the most subtle and ingenious of these skeptical attacks on meaning, and on semantic phenomena in general, deserve convincing answers. Unquestionably, W.V.O. Quine is one of the most original and most influential supporters of the doctrine of meaning skepticism. According to his famous thesis of the indeterminacy of translation, manuals for translating one language into another can be constructed in divergent ways, which are all compatible with the totality of speech dispositions, yet incompatible with one another.1 In accord with his behavioristic conception of language,

1 Cp. Quine 1960, 27

Introduction

3

Quine thinks that behavioral evidence is all the evidence that is relevant to theories of translation. From this, Quine concludes that the question whether any one translation is the uniquely correct one makes no sense. Suppose three field linguists construct manuals for translating a foreign language such that the foreigner’s one-word sentence “Gavagai” is translated as “There is a rabbit” by the first linguist, as “There is an undetached rabbit part” by the second, and as “There is a temporal stage of a four-dimensional rabbit” by the third. Quine claims that if all of these mutually incompatible translations are consistent with the verbal dispositions of all foreign speakers, then they are equally justified. The question which of these translations is uniquely correct has no answer and, therefore, no objective sense. If there are no possible facts, which allow us to answer the question “What did the speaker mean by ‘Gavagai’?”, then that question turns out to be senseless. According to Quine, there is no fact of the matter to the question of which of these translations is uniquely correct. It is obvious that the thesis of the indeterminacy of translation is a form of meaning skepticism. For, surely, it is the job of a decent scheme of translation to correlate expressions that intuitively have the same meaning. Hence, if there are no facts about which scheme of translation is correct, it immediately follows that there are no facts about the meanings of the expressions correlated. For Quine, claims about meaning and synonymy do not state objective facts and so cannot be regarded as genuine truths. Quine’s original argument for the indeterminacy of translation relies on the behavioristic premise that people’s overt behavior under publicly observable circumstances furnishes the only facts relevant to determining meaning. Mental and other non-observable facts are entirely ignored. Behaviorism is a highly controversial doctrine, however. Quine himself came to recognize that there are philosophers who will agree with him that facts about behavior are not sufficient to fix the meanings of words, while they will disagree with his view that these are the only meaning-fixing facts. In effect, he confronted these philosophers with the challenge to cite further physical facts that might help to determine meaning.2 Quine is a resolute advocate of physicalism, of the doctrine that all facts are physical facts. And he is firmly convinced that even the totality of all physical facts cannot help to resolve the indeterminacies about translation and meaning. Evidently, an appeal to neurophysiological states, for example, is unpromising.

2 Cp. Quine 1970

4

Introduction

One cannot read off from the neuronal states a speaker is in when he uses the word “gavagai” whether his word means the same as my word “rabbit”. Quine thinks that the same holds for any other physical fact one might adduce. And so the alarming judgment he arrives at is that translation is underdetermined by the totality of all physical facts.

2. Kripke’s Wittgenstein’s skeptical paradox Another important figure in the contemporary debate about meaning skepticism is Kripke’s Wittgenstein. In his extremely influential book Wittgenstein on Rules and Private Language, Kripke presents a skeptical argument, inspired by Wittgenstein’s famous rule-following paradox, for the startling conclusion that there are no facts in virtue of which ascriptions of meaning, such as “Jones means addition by ‘+’ ” are either true or false, and, more generally, that there are no facts about what anyone means by any of the expressions she uses. Faced with this apparently paradoxical and self-refuting conclusion, we tend to suppose that there must be something wrong with the argument supporting it, that it hinges on some controversial assumption which we can discard or that it involves some fallacious step. So the natural response to the skeptical argument consists in the attempt to defend the claim the skeptic called into question, the claim that there are, after all, meaning-constituting facts. But Kripke rejects this “straight solution”; he argues that there indeed are no fundamental facts in which the meanings of our words consist.3 Instead, he embraces a “sceptical solution”, one which accepts the skeptical conclusion but tries to rehabilitate our ordinary practice of ascribing meanings to our utterances. The core of his skeptical solution consists in the proposal to replace the notion of truth conditions, as constitutive of sentence meaning, by that of justification conditions or assertibility conditions. Meaning ascriptions do not need truth conditions or facts making them true or false in order to be perfectly meaningful and to play useful and vital roles in our lives. Kripke’s book is concerned with some of the most basic questions in the philosophy of language and mind. Moreover, the main theoretical options that might be taken with respect to these difficult questions enter the stage: semantic realism and irrealism, naturalism, reductionism, dispositionalism,

3 Kripke 1982, 66–7

Introduction

5

individualism and communitarianism. It is, therefore, worthwhile to take a closer look at how Kripke develops his skeptical argument. He begins by considering a simple arithmetical example. The symbol “+”, so we routinely assume, stands for the addition function, which assigns a unique natural number to infinitely many pairs of natural numbers. Hence, what we mean by this symbol cannot be exhausted by the comparatively small number of cases in which we have actually employed it. Kripke asks what fact about us determines that “+”, as we now employ it, already applies to cases we have not yet considered. Let us suppose that “68 + 57” is a calculation that I have never performed before. Certainly, computing the sum of these numbers now, I shall confidently answer “125”. This answer is correct in the arithmetical sense, since 125 is indeed the sum of 68 and 57. The skeptic, however, questions my confidence in the “metalinguistic” sense. He asks whether it is really the case that the sign “+”, as I used it in the past, stood for the same function as my present use of it, the addition function. It is important to note that in his development of the skeptical paradox Kripke assumes that there is no problem about our present understanding of words. In particular, he grants pro tem that, as we now use it, “+” stands for the addition function. This strategy has the advantage of permitting him to take the meanings of the words employed in stating the argument for granted and so to eschew the objection, which immediately suggests itself, that his argument is self-refuting, that its successful formulation falsifies its conclusion, because it employs certain words with definite meanings while at the same time doubting what they mean. But, of course, if the argument were really cogent, its conclusion could straightforwardly be extended to present and future uses. If there really were a fact about me that determined that I now mean addition by “+”, then I could appeal to this fact to refute the challenges the skeptic will raise tomorrow. Plainly, however, I will not be in a better position tomorrow than I am today. It is the challenge of Kripke’s skeptic to cite some fact about us that constitutes our meaning the addition function, rather than some other function or nothing at all by “+”. He asks: In virtue of what fact are we justified in answering “125” to the question “What is 68 + 57?” Kripke’s skeptic contends that there is no such fact. He proceeds by considering all the facts potentially relevant to determining the meaning of our words— facts about our previous linguistic and nonlinguistic behavior, and facts about our mental history and qualitative mental life—and comes to the conclusion that an expression’s possession of meaning cannot consist in

6

Introduction

facts to be found in these regions. There is no restriction at all upon the kind of fact that might be cited as engendering the meanings of words. So while it exhibits plain similarities to Quine’s famous argument for the indeterminacy of translation, Kripke’s argument seems to many philosophers to be even stronger since it does not exclude access to facts about our mentality, such as mental images, sensations or feelings. Moreover, it is an important feature of Kripke’s argumentative strategy that he makes an idealizing assumption about our epistemic capacities: we are imagined to have perfect epistemic access to all the potentially relevant behavioral, mental, and even physical facts. It becomes evident that Kripke is not seeking to establish a form of traditional epistemological skepticism. Epistemological skepticism about a subject matter is the position that even though there are determinate facts about it, our cognitive powers are incapable of acquiring knowledge or adequately justified beliefs about them. The meaning skepticism Kripke is after, although he challenges us to justify our claim to know what the expressions we use mean and so presents itself in an epistemological guise, is a much more radical view than epistemological skepticism. What he is after is a form of constitutive or metaphysical skepticism, which denies that there are meaning facts at all. For, so the skeptical reasoning goes, if even under the idealizing assumption of perfect access to all relevant areas, we cannot find any meaning-constituting facts within them and thus are still incapable of justifying any assertion about the meanings of our words, the conclusion must surely be drawn that there simply are no such facts.4 Obviously, an appeal to the beliefs, thoughts, intentions or other propositional attitudes connected with our employment of sentences will be of no help. Just as linguistic expressions have meanings, so the propositional attitudes have mental or representational content. At least, this is what we commonly, outside the special context of the skeptical scenario, suppose. There is a close relationship between language and thought. The sentence “Bananas are yellow” means that bananas are yellow, and we use this sentence to express a belief with the content that bananas are yellow. It is well known that philosophers disagree on the question of explanatory priority. Should, in line with the traditional mentalistic order of explanation, the semantic properties of language be derived in virtue of our intentions and conventions from the mental contents of our propositional attitudes, as Paul Grice and Stephen Schiffer think?5 Or should language be given 4 Cp. Kripke 1982, 14–15; Boghossian 1989, 515; Miller 2002, 2 5 Cp. Grice 1989; Schiffer 1972

Introduction

7

priority, as defenders of a linguistic theory of intentionality such as Wilfrid Sellars and Michael Dummett hold, for whom thinking is a kind of internalized saying?6 Or is, after all, Donald Davidson’s influential view correct that neither thought nor language has priority, and that, rather, these realms are interdependent, so that meaning and mental content can only be explicated together?7 Fortunately, we do not have to settle this thorny question in order to see that a persuasive argument for meaning skepticism will straightforwardly bring forth a persuasive argument for content skepticism as well. As Paul Boghossian, aptly and concisely puts it: Whatever the correct answer, however, there would appear to be no plausible way to promote a language-specific meaning skepticism. On the former (Gricean) picture, one cannot threaten linguistic meaning, without threatening thought content, since it is from thought content that linguistic meaning is held to derive; and on the latter (Sellarsian) picture, one cannot threaten linguistic meaning without thereby threatening thought content, since it is from linguistic meaning that thought content is held to derive. Either way, content and meaning must stand or fall together.8

We have to realize that a satisfactory reply to the skeptic will have to discover facts of the matter that fix both mental and linguistic content—content in general. As already indicated, the case for the skeptical conclusion proceeds by examining the potential candidates. First, the meaning of “+” cannot simply consist in the actual uses we have made of it in particular computations, since these are finite in number and so compatible with our having meant some function that coincides with addition over the cases actually encountered but diverges from it over new, so-far unencountered cases. Actual applications of “+” underdetermine which function is meant, since no finite sequence of applications of “+” can determine which of the indefinitely many functions compatible with those applications is the one that was denoted. Deviant interpretations will always be possible. Kripke goes on to consider a natural protest to his reasoning. Surely, our understanding of “+” does not consist in our having given ourselves some finite number of examples from which we are supposed to extrapolate the whole function. Rather, our understanding of “+” consists

6 Cp. Sellars 1963; Dummett 1993 7 Cp. Davidson 2001 8 Boghossian 1989, 144

8

Introduction

in associating with it a linguistic rule or an algorithm that determines its correct application. The rule applies, in principle, to an indefinitely large class of cases. Hence, when confronted with a new computation, we simply apply the general internalized instructions the rule involves and get the correct result. Thus, according to this response, the fact that we mean addition by “+” is constituted by the fact that we have associated a general rule with it that always yields the sum of the numbers to which it is applied. However, as soon as we ask what a rule actually is, we recognize that this response leads us back to where we started and only sets in motion the potential regress of interpretations Wittgenstein accentuated over and over again. In order to give us any general instructions the rule cannot be conceived of as just a collection of uninterpreted symbols. Syntax does not furnish meanings. No symbol intrinsically means anything. Thus the mental symbols of the rule, just as the external symbols used in expressing it, must be interpreted, must possess meanings. But, since this is so, the skeptic can always produce deviant interpretations of the symbols of the rule itself. It is obvious, however, that the process of using symbols to interpret other symbols must stop at some point. Kripke claims here to be expounding Wittgenstein’s well-known remarks in the Philosophical Investigations on “a rule for interpreting a rule”. Ultimately, understanding cannot be based solely upon the association of one symbol with another, upon a kind of translation. As Wittgenstein himself puts it: “. . . any interpretation still hangs in the air along with what it interprets, and cannot give it any support. Interpretations by themselves do not determine meaning.”9 Essentially the same point is expressed in his dictum: “Explanations come to an end somewhere.”10 Sooner or later explanations, justifications, interpretations, rules for applying rules, etc. run out. If we ever mean anything by any word, some words must be understood without explanation and hence without associating any rules with them. It seems then that facts about our actual and previous behavior cannot determine definite meanings. But perhaps the sought-after facts are to be found in our mental life. Can states or events or processes in our consciousness—having some mental image, or some specific mental occurrence having a distinctive qualitative character—supply what we are looking for? Now it was one of Wittgenstein’s central aims to dislodge this

9 Wittgenstein 1958, §198 10 Ibid., §1

Introduction

9

tempting and once very popular mentalistic conception of meaning. Contemporary philosophers of language have absorbed many of Wittgenstein’s negative conclusions in this regard. So Kripke thinks he can be brief here and simply rely on the cogency of the many arguments Wittgenstein developed against the classical empiricist conception which assimilated the concepts of meaning and understanding to concepts of mental phenomena. In the first instance, Wittgenstein emphatically rejects the idea that meaning is a kind of picturing, that understanding the meaning of a word consists in having the experience of a mental picture coming before one’s mind, e.g. a picture of a cube when one uses or understands the word “cube”. Firstly, Wittgenstein argues that it is not a necessary condition for understanding an expression that some particular mental image come before our minds when we hear or use the expression. To be sure, there are typical experiential accompaniments to use and understanding, and such phenomena may suggest a particular meaning, but they do not determine one. The crucial point is that there is no one particular mental item coming before our minds’ whenever we use or understand a given word. The mental episodes associated with understanding may vary from person to person and even for the same person on different occasions. And even if there were a psychological uniformity between a particular word and a particular mental entity we could still conceive of persons who understand a given word in exactly the same way even though they associate quite different images with it. Indeed, we can conceive of a person whose understanding of the word is not accompanied by any such mental representation at all. Secondly, it is not a sufficient condition for meaning a word in a particular way that some picture or some other mental entity come before one’s mind. Wittgenstein argues that this is not sufficient because the picture itself cannot determine the correct use of the word with which it is associated. The picture is in actual fact just another sign, another piece of lifeless syntax, whose meaning has to be fixed, too. Internal mental pictures, no less than external physical pictures, are themselves susceptible of a multitude of alternative interpretations. And so again, the possibility of deviant applications of them cannot be ruled out. There is no logical route from the properties of a mental item to the meaning of an associated word. A picture of a cube may, given a deviant “method of projection”, indeed be employed as a sign of a pyramid.11 Moreover, suppose that a

11 Ibid., §139

10

Introduction

person’s uses of the word “cube” are constantly accompanied by an image of a pyramid, but that nevertheless she applies the word correctly; she applies it to cubes, not to pyramids. It looks as if it does not matter what comes before one’s mind. Mental pictures or images cannot function as criteria of understanding. Consequently, understanding does not consist in having mental pictures.12 Kripke considers a further response to the skeptical challenge, to which a proponent of mentalistic semantics might try to resort. According to this suggestion, the state of meaning something by a sign should not be assimilated to mundane states of consciousness such as having mental images, sensations or other introspectible qualitative states. Rather, the needed mental state might be conceived of as a state sui generis, as an irreducible “primitive” state, a state “of a unique kind of its own”.13 Kripke dismisses this move brusquely because, as he says, it is “desperate” and leaves the nature of postulated primitive state of understanding “completey mysterious”.14 His first objection is that the state is not supposed to be introspectible, and yet we are allegedly aware of it with a fair degree of certainty whenever we enjoy it. This raises the problem, so Kripke argues, that we would have no idea of how we could know the contents of our thoughts, if there really were such primitive non-qualitative states. His second charge is that it is also entirely mysterious how there could be “a finite object contained in our finite minds” that nevertheless bears information about the totality of the correct future uses of an expression. It seems to be a mystery how in understanding the meaning of an expression we could immediately grasp something that already contains in some sense the indefinitely large range of future applications of the expression in advance. Kripke’s skeptic holds that there is no such advance determination in our understanding of a word. After all, to know the meaning of an expression is not a matter of travelling along rails which are already laid down and fix what is to count as its correct application in so far unconsidered cases. We now come to the centerpiece of Kripke’s skeptical treatment of meaning, to his examination of the tempting suggestion that instead of seeking to locate the meaning-determining facts in our actual use of language or in occurent mental states or events, we should turn to dispositional facts about speakers. Our meaning one thing rather than another by

12 Cp. McGinn, 1984, 1–7 13 Kripke, 1982, 51 14 Ibid.

Introduction

11

our use of words resides, according to this view, in our being disposed to apply them in certain ways and not in others. The special appeal of this suggestion is that it can be a determinate fact about us that we were all along disposed to answer “125” to the question “57 + 68 = ?”, but never actually did so, simply because this question was never asked. Whilst the particular additions we have performed have changed over time, our dispositions to answer arithmetical questions in definite ways have not. Hence, so the advocate of the dispositionalist view claims, we denoted the same function by “+” in the past as we do now. The skeptic’s contention that we might have meant some other function seems to be refuted. Kripke argues, however, that the dispositionalist account must be rejected, for two reasons. The first objection, the finitude objection, is that the totality of our dispositions regarding use, like the totality of our actual uses, is finite and, therefore, does not have the required generality. Meanings seem to have an infinitary character. There are indefinitely many questions of the form “x + y = ?”, but it is simply not the case that for every one of them we are disposed to give a certain determinate answer; some pairs of numbers will be too large for us to deal with and some will be even so extremely long that we will die before we can answer them. And so the skeptic, by selecting examples that lie beyond the reach of our finite dispositions, can once again develop deviant interpretations that are nonetheless in accord with all the relevant dispositional facts. The second, and more important objection, the normativity objection, is that the dispositionalist analysis lacks the resources to capture the essential normativity of meaning. Let us first clarify the sense in which meaning is normative. Kripke says: Suppose I do mean addition by “+”. What is the relation of this supposition to the question how I will respond to the problem “68 + 57”? The dispositionalist gives a descriptive account of this relation: if “+” meant addition, then I will answer “125”. But this is not the proper account of the relation, which is normative, not descriptive. The point is not that, that if I meant addition by “+”, I will answer “125”, but rather that, if I intend to accord with my past meaning of “+”, I should answer “125”. Computational error, finiteness of my capacity, and other disturbing factors may lead me not to be disposed to respond as I should, but if so, I have not acted in accordance with my intentions. The relation of meaning and intention to future action is normative, not descriptive.15

15 Ibid., 37

12

Introduction

There is indeed an intimate relation between the meaning of an expression and how it should be used. The fact that the word “tiger” means tiger implies that we should apply it to all and only tigers and not to any other things. Generally, the fact that an expression means something has a normative component: some uses of it in application to certain objects are correct, and some uses of it in application to certain other objects are incorrect. That is why the meaning of a word determines its extension, the set of things to which one ought to aim to apply it. So the decisive point in Kripke’s subtle discussion of the normativity of meaning is the familiar fact that meaningful expressions possess conditions of correct use. Understanding a word involves adopting a standard to which one intends to adhere, and which furnishes the basis for evaluating employments of it as correct or incorrect. Therefore, Kripke argues, it is an adequacy condition on accounts of meaning that any proposed candidate for the property in virtue of which an expression means something must be such as to sustain the normativity of meaning; it must be possible to read off from any alleged meaning-constituting property of a word what is the correct use of that word. And Kripke is convinced that the dispositional theory does not fulfill this adequacy condition because one cannot read off from a disposition to use a word in a particular way what is the correct use of that word. In contrast to meanings, verbal dispositions do not determine extensions. The set of objects to which we are disposed to apply a given term under various conditions is bound to diverge from with its real extension. For, we may well have dispositions to use words incorrectly, to make a few systematic errors. But, if what we mean by “+” is identified with the answers we were disposed to give to questions of the form “x + y = ?”, then there is no room for the needed distinction between the answers we would have given and the answers we should have given. The consequence of equating competence and performance is that all our answers must be correct. But this cannot be right. We have to leave room for the possibility of error. For, surely, a speaker can mean addition by “+” even though she is disposed to make some mistakes in computation in particular cases. Convinced that he has examined and eliminated all the plausible candidates for facts that might fix the meaning of “+”, Kripke draws the radical skeptical conclusion that there are no such facts. And, since the argument does not essentially concern a particular speaker, or the sign “+”, or the past, it seems that the conclusion can be straightforwardly generalized to all speakers, all expressions and all times. It appears to follow that

Introduction

13

no one can ever mean anything by any of the words she uses. The notion of meaning appears to collapse. As Kripke puts it: “It seems that the entire idea of meaning vanishes into thin air”.16 Furthermore, the argument can be rerun for other content-bearing states such as beliefs, intentions, and other propositional attitudes. So we are confronted with a radical skepticism about content in general, with a shocking skepticism about intentionality: no one ever means, or believes, or thinks anything. Language and thought seem to be impossible. This seems to be unadulterated conceptual nihilism.

3. The skeptical solution Such a radical form of skepticism seems to be self-refuting, to lead into a form of pragmatic incoherence. For, if the skeptic really succeeds in stating the argument, then that very achievement seems to refute him. So, apparently, there must be something wrong with the argument. And Kripke himself thinks that we can deny the skeptical thesis that no one ever means or believes anything. As already mentioned, he distinguishes between two ways one might respond to it. A “straight solution” consists of a proof of the thesis the skeptic called in question—the discovery of an appropriate meaning-determining fact would show that the central premise of the skeptical argument is false. Kripke, however, recommends a “sceptical solution”. This alleged solution comprises two essential components. The first component is the concession that the skeptical thesis is correct: there are no meaning facts, and hence no facts corresponding to our ordinary practice of using sentences that ascribe meanings. The second component is an argument that seeks to convince us that we are nevertheless perfectly justified in preserving our ordinary practice of ascribing meanings to the words we use. For this combination of the two parts of the skeptical solution to become consistent, however, we have to provide a radically different account of the significance of statements about meaning, namely that such statements have some other role than that of asserting that certain facts obtain. Since meaning ascriptions do not purport to state facts at all and so do not have truth conditions, but have a quite different, non-factstating function, the discovery that there are no possible facts for them to

16 Ibid., 22

14

Introduction

stand for cannot endanger the legitimacy of their uses and hence their meaningfulness. Thus, the skeptical solution rejects the key presupposition of the radical skeptic’s reasoning, the presupposition that in order for attributions of meaning to be meaningful and fully justified they have to be conceived of as fact-stating and as having truth conditions. Rather, meaning ascriptions are supposed to have a kind of meaning that does not tally with the classical, realist, truth-conditional theory of meaning. Kripke argues that the skeptical solution is made possible by the change in Wittgenstein’s philosophy of language from the Tractatus to the later Philosophical Investigations.17 While the skeptic presupposes the central idea of the Tractatus, the idea that declarative sentences possess their meanings in virtue of their truth conditions, in virtue of their correspondence to facts that must obtain if they are true, the skeptical solution abandons this general picture of meaning and mobilizes instead the different conception espoused in the Philosophical Investigations. Kripke characterizes this alternative, allegedly non-factualist model of sentence meaning as one in which truth conditions are replaced by assertibilty conditions, or justification conditions. If we accept the recommended shift to a global form of non-factualism, we can also apply this position to meaning ascriptions themselves. The welcome result is that we shall no longer be threatened by the skeptical conclusion, for the defender of the skeptical solution can simply argue that the skeptic’s radical claim that no one means anything depends upon a false—even though quite natural—picture of meaning. As Kripke says: Now if we suppose that facts, or truth conditions, are of the essence of meaningful assertion, it will follow from the skeptical conclusion that assertions that anyone ever means anything are meaningless. On the other hand, if we apply to these assertions the tests suggested in Philosophical Investigations, no such conclusion follows. All that is needed to legitimize assertions that someone means something is that there be roughly specifiable circumstances under which they are legitimately assertable, and that the game of asserting them under such conditions has a role in our lives. No supposition that ‘facts correspond’ to those assertions is needed.18

Kripke urges that the emphasis in the Philosophical Investigations is no longer on the truth conditions of declarative sentences but on the description of the conditions under which they may properly and justifiably be 17 Cp. ibid., 71–78 18 Ibid., 77–78

Introduction

15

asserted, and on an explanation of their role and utility in our linguistic and nonlinguistic practices. Accordingly, we are enjoined to look at the conditions under which attributions of meaning are actually made and at their utility or point within our lives. Kripke thinks that it is the key element of Wittgenstein’s skeptical solution that the assertibility conditions and role of meaning-ascriptions essentially involve the verbal dispositions of the linguistic community. Thus, an utterance like “Jones means addition by ‘+’” is justified if we have found that his uses of “+” are in agreement with uses of “+” by most fellow speakers in his community. The role and utility of such utterances is to acknowledge him as a competent member of the community, to express the community’s firm persuasion that he can generally be trusted in his interactions with members of his community in contexts which involve the use of “+”. Kripke’s Wittgenstein is convinced that meaning and rule-following are essentially social affairs. The community is supposed to be needed to underwrite the normativity of meaning. If we consider Jones in isolation from any community, as a speaker of a solitary or private language, all we will come across is his disposition to use the “+” sign in a certain way, but there is nothing for his employments of it to agree or disagree with. For a speaker of a solitary language, the crucial distinction between real and apparent rule-following breaks down. It breaks down because there is— neither for the speaker nor for the hearer—a criterion of correctness, a criterion for telling whether the rules of the language are being followed correctly or not. In Wittgenstein’s own words: “One would like to say: whatever is going to seem right to me is right. And that only means that here we can’t talk about ‘right’.”19 Merely believing that my usage accords with a rule does not suffice for accord with it. Rule-following is an objective phenomenon. So whether one really obeys a rule consistenly is not determined by whether one believes one is obeying it. Therefore, Wittgenstein also remarks in a famous passage: “And to think one is obeying a rule is not to obey a rule. Hence it is not possible to obey a rule ‘privately’: otherwise thinking one was obeying a rule would be the same thing as obeying it.”20 Since the distinction between “is right” and “seems right” is essential to the concept of meaning, an isolated person cannot mean anything by the words he uses.

19 Wittgenstein 1958, §258 20 Ibid., §202

16

Introduction

However, once the social background is taken into account, things present themselves in an altogether different light. Involvement of the community puts us in the position to distinguish between the assertibility conditions of utterances of “Jones means addition by ‘+’” and the assertibility conditions of utterances of “Jones believes that he means addition by ‘+’”. The community, observing his divergent usage, might come to the conviction that Jones does not really mean addition by “+”—although he himself thinks that he does mean addition. Hence, the correlative possibilities of error and correctness can only be explained in terms of the agreement or disagreement between Jones’ use of “+” and that of the rest of the community. In this way, the neutralization of the skeptical paradox at the same time furnishes a demonstration that a solitary language is impossible.

4. Concluding remarks Expectedly, Kripke’s study aroused a great deal of critical discussion. I think it is fair to say that there is widespread consensus among commentators that Kripke misunderstood the strategic purpose of Wittgenstein’s rule-following considerations, and, for this reason, misinterpreted Wittgenstein’s account of rules and rule-following and its putative connection with the private language argument. Apparently, Wittgenstein did not advocate any form of rule skepticism or meaning skepticism. Thus, there was no need for him to propose a skeptical solution essentially involving the linguistic community. Indeed, Wittgenstein even seems to have championed a variant of the dispositional view which is the main target of Kripke’s skeptical argument against the reality of rule-following. It is true that, for Wittgenstein, using a word according to a certain rule does not rest upon justifying reasons. But this is not an occasion for skeptical doubts. The principal positive element of his view is the thesis that to follow a rule, and hence to use a word with a certain meaning, is to participate in a custom or practice. Since customs and practices get established by regularities in behavior, by recurrent actions in appropriate contexts, they are capable of accounting for the essential diachronic character of meaning. Moreover, “practice” does not signify a social practice. The point of the rule-following considerations is not to establish that language necessarily involves a community, a multiplicity of speakers. Their point is, rather, to establish that rule-following is a form of customary conduct, of regular action in accord

Introduction

17

with rules—not, and this is the relevant contrast, a form of thought or private experience. More than a few excellent philosophers have come to believe, however, that the exciting skeptical reasoning Kripke has unfolded and its farreaching consequences are worthy of careful deliberation in their own right. Almost all of his commentators remained unconvinced, however. The majority of his critics think that a straight solution must be possible because the semantic irrealism the skeptical solution brings along is a position of doubtful incoherence in the end. There are critics who, while they share Kripke’s basic naturalistic assumption that meaning facts must be reducible to non-semantic and non-intentional facts, nevertheless hold that some sophisticated form of the dispositional account can be defended after all,21 or that some version of the causal or informational theory of reference and meaning is not vulnerable to the skeptical argument.22 And there are other critics who are convinced that it is impossible to reduce meaning facts to more fundamental non-semantic facts, and, therefore, find fault with the seemingly unwarranted semantic reductionism underlying the skeptical argument.23 According to their estimation, the way meaning and understanding are indeed somehow grounded in physical facts need not comply with the severe requirements of a reductive account. Anti-reductionist views are, however, confronted with the notorious problem of mental causation, the problem of how to reconcile anti-reductionism about meaning and content with a plausible conception of their causal efficacy. If a genuine causal role in the explanation of behavior is indeed attributed to semantic and intentional phenomena, then, given the autonomy of the physical realm, the conclusion cannot be avoided that those behaviors are mysteriously over-determined. Perhaps we can say that the real significance of Kripke’s and, not to forget, Quine’ skeptical lines of thought resides in bringing the profound question of how meaning and content can be objective features of our speech and thought into the focus of our philosophical attention. We do owe the skeptics of various stripes a persuasive reply to the question of what meaning and content really are. The essays comprising this volume are all of them devoted, in one way or another, to this fundamental philosophical question. 21 Cp. Forbes 1983–4; Horwich 1995; Millikan 1990 22 Cp. Fodor 1990, 89–136; Goldfarb 1985 23 Cp. Boghossian 1989; McGinn 1984; McDowell 1984; Wright 1984

18

Introduction

Several people have contributed in various ways to the successful completion of this volume. All of them are warmly thanked. I have to express my gratitude to the contributors, both for their immense patience with the many delays and for the excellence of their papers. Again, thanks are due to Gertrud Grünkorn, and the editorial staff at Walter de Gruyter for their expert assistance in bringing this volume to publication. Finally, I would like to add a word of thanks to my student Michael Helwig, who helped me with the technical implementation of the project.

References Boghossian, Paul: 1989, “The Rule-Following Considerations”, Mind 93, 507–49 Davidson, Donald: 2001, Inquiries into Truth and Interpretation, Clarendon Press, Oxford Dummett, Michael: 1993, The Seas of Language, Oxford University Press, Oxford Fodor, Jerry: 1990, A Theory of Content and Other Essays, MIT Press, Cambridge/MA Forbes, Graeme: 1983–4, “Scepticism and Semantic Knowledge”, Proceedings of the Aristotelian Society 84, 223–37 Goldfarb, Warren: 1985, “Kripke on Wittgenstein on Rules”, Journal of Philosophy 82, 471–88 Grice, Paul: 1989, Studies in the Way of Words, Harvard University Press, Cambridge/ MA Horwich, Paul: 1995, “Meaning, Use and Truth”, Mind 104, 355–68 Kripke, Saul: 1982, Wittgenstein on Rules and Private Language, Blackwell, Oxford McDowell, John: 1984, “Wittgenstein on Following a Rule”, Synthese 58, 325–63 McGinn, Colin: 1984, Wittgenstein on Meaning, Blackwell, Oxford Miller, Alexander: 2002, “Introduction”, in Miller, Alexander and Wright, Crispin (eds.), Rule-Following and Meaning, Acumen, Chesham, 1–15 Millikan, Ruth Garrett: 1990, “Truth Rules, Hoverflies, and the Kripke-Wittgenstein Paradox”, The Philosophical Review 99, 323–53 Quine, Willard Van: 1960, Word and Object, MIT Press, Cambridge/MA Quine, Willard Van: 1970, “On the Reasons for Indeterminacy of Translation”, Journal of Philosophy 67, 178–83 Schiffer, Stephen: 1972, Meaning, Oxford University Press, Oxford Sellars Wilfrid: 1963, “Empiricism and the Philosophy of Mind”, in Science, Perception and Reality, Routledge, London, 127–96 Wittgenstein, Ludwig: 1958, Philosophical Investigations, Blackwell, Oxford Wright, Crispin: 1984, “Kripke’s Account of the Argument Against Private Language”, Journal of Philosophy 81, 759–78

I Meaning and Reference

The Puzzle That Never Was—Referential Mechanics1 Joseph Almog Abstract: The puzzle that never was is the one both Frege-Russell but also their nemesis, Kripke, all see as the source problem of the theory of reference: how—in virtue of what fact—does a name in my mouth, say “Aristotle”, refer to this unique man in history, the ancient Greek Aristotle, and not say, to Plato or Euclid or . . . Kripke tells us two very interesting things about this mother puzzle. First, at the level of answering the darkest puzzle, Frege-Russell have a natural answer to the puzzle, while Mill has none. And secondly and more important, Kripke formulates the question, and thus agrees with Frege-Russell, as if the deepest oracular witchcraft is involved: a name in my mouth reaches out (out from where?), and as if by sheer magic, touches a man long dead and gone now for 2300 years. I submit below that the standard expositions—led by Kripke’s—invert things. First, at the level of theories, it is not Mill but rather Frege/Russell (and both of them) who have a big puzzle of reference-determination on their hands. Secondly and more important, at the level prior to any theory, reference-determination involves no magic, action-at-distance or any other mystery; there is simply no determination-puzzle and stronger yet, there never—say 3000 years ago or all the way back to caveman—was a puzzle of reference-determination.

The puzzle that never was is the one both Frege-Russell but also their nemesis, Kripke, all see as the source problem of the theory of reference: 1 I see this paper as a sequel to my “Semantical Anthropology” of 1984. In relating to Kripke’s work on the “puzzle”, his key text is Naming and Necessity (NN) pp. 27–30, pp. 79–85 and fn. 33–34 therein. As important is Frege’s footnote about the name “Aristotle”, in his touchstone paper “Uber Sinn und Bedeutung” (1892) (Kripke quotes the footnote on p. 30 of NN). I owe many thanks to years of co-teaching NN with David Kaplan (who is not responsible for anything I say except for making to me clear what it is that I think). I also owe many thanks to Keith Donnellan for directing me to the source of the relation of reference. There is further relevant discussion of the matter in Donnellan’s forthcoming volume Essays on Reference (OUP 2012) and my editor’s introduction to it.

22

Joseph Almog

how—in virtue of what fact—does a name in my mouth, say “Aristotle”, refer to this unique man in history, the ancient Greek Aristotle, and not say, to Plato or Euclid or . . . In stage-setting his attack on Frege and Russell, in his NN 27–29, Kripke tells us two very interesting things about this mother puzzle. First, at the level of answering the darkest puzzle, his Frege-Russell (and aside of this fictive philosopher Kripke weaved, also the historical man Frege and the distinct historical man, Russell) have a natural answer to the puzzle, while Mill has none. And secondly and more important, answer the puzzle or not, Kripke formulates the question, as if the deepest oracular witchcraft is involved: a name in my mouth reaches out (out from where?), and as if by sheer magic, touches a man long dead and gone now for 2300 years; Ooga Booga. In contrast to this dramatic casting, I submit below that the standard expositions—led by Kripke’s—invert things. First, at the level of theories, it is not Mill but rather Frege/Russell (and both of them) who have a big puzzle of reference-determination on their hands. Secondly and more important, at the level prior to any theory, reference-determination involves no magic, action-at-distance or any other Ooga Booga; there is simply no determination-puzzle and stronger yet, there never—say 3000 years ago or all the way back to caveman—was a puzzle of reference-determination.

Three hoary puzzles On pp. 27–29 of his game-changing NN, Saul Kripke mentions three puzzles with respect to which he admits the Frege-Russell account of reference to (and thinking about) individuals is doing very well, resolving the puzzles one by one and with “marvelous coherence”. In contrast, Mill’s view—towards which Kripke has much sympathy—viz. that the sole linguistic function of a name is to stand for its bearer and without connoting any (descriptive) content—this Millian view seems absolutely bankrupt when it comes to these three tests-puzzles. The second and third puzzles Kripke classifies as somewhat “subsidiary”, indeed seemingly derivative of the first, source-problem. Let us recall here that the second problem is Frege’s famed question “how could a true identity of the form “Hesperus = Phosphorous” be cognitively significant?” It is ( pre-) supposed—by identifying semantic and cognitive significance for Mill—that for Mill’s view, this identity is as cognitively insignificant as the allegedly trivial “Hesperus = Hesperus” (it is again ( pre-) supposed

The Puzzle That Never Was—Referential Mechanics

23

(for Mill) that this last is cognitively trivial because it semantically is). Mill seems to be missing the further (discerning) object-transcendent “contents” the two names must be carrying, the contents the “head” of the user must be grasping as part of being competent with the names. The third puzzle is in effect two puzzles in one, both of which regard existential sentences. So, first, using a loaded name, a name that does refer, e.g. “Obama”, Kripke asks us, what is the point of asserting the true (but, again, allegedly (semantically? cognitively?) tautological according to Mill) “Obama exists”? And now secondly, using the empty name “Vulcan” (of the purported planet causing Mercury’s perturbations; as is well known, there is no such planet), we say truly “Vulcan does not exist”. How could this last be true, if there is no referent—no astronomical object—making it true? Again having dismissed the hyper-objectual “contents”, Mill is in shambles. So, there; two famous destructive puzzles nabbing the Millian. But the true source of it all is the first problem Kripke describes in pp. 27–8 as the determination of reference (of a proper name). In virtue of what (what fact) does a given historical name, say “Aristotle” (as used for the famed ancient Greek philosopher; recall that there are other people so called) stand for this one worldly object, the man Aristotle, and not any other man, say, Plato or Salvador Dali?

A detour through acquaintance Kripke observes (NN, 28) that for a few closely available items (this must be relativized—to a particular user and at a given time), Russell thinks the relation of acquaintance—with the target object—can provide the answer (for a given user at a time). The impact of this solution is of course relative to the “range” of our acquaintances. But quite apart of the range of this relation (made ever narrower by Russell for epistemological reasons), there is the to my ear much more fundamental question: why would knowledge of this or that kind ground—form the logical basis—of a semantic relation? Russell and many Neo-Russellians—e.g. first my late teacher Gareth Evans, then Tyler Burge, then the recent Kaplan and hordes of others trying to “save” Frege from Kripke by way of a dose of Russell—all write as if they are “ahead” of Frege, when it comes to the problem of determination of reference.2 2 Evans “The causal theory of Names’ (1973); Burge “Belief de re” J. Phil 1977; Kaplan “On an idea of Donnellan” to appear in “Essays on having in mind dedicated to Keith Donnellan” OUP 2012 (edited by Leonardi and Almog).

24

Joseph Almog

Why? Because they do not depend on the assumption I am to have a uniquely identifying piece of purely descriptive knowledge in order to refer to the candidate item, x. Frege boxed himself with an overly strict demand and Kripke has picked on this demand. As against this sense of improvement on Frege (Frege still remains, to my ear, the best Fregean!), let me record two facts. First, Frege himself never insisted (and could not really insist, given his examples on which more below, see e.g. “Aristotle”) that the reference-determiner, the “belief ’ (or “knowledge’, if such there be) in the head of the user is purely descriptive. He gives us, for “Aristotle”, “(best) pupil of Plato” and “teacher of Alexander”, both of which obviously have relations to objects, to Plato and Alexander. And so it goes with other examples glossed in his spirit e.g. “London” glossed as “The capital of England” or “Mont Blanc” as “the tallest mountain of Europe”, all involving relations to individuals. I see no reason—in Frege—why he should commit to eliminating all such relations to individuals (if anybody is committed to such eliminations to a purely “qualitative” language, it is Russell’s epistemology, by 1911). The second point is to me the critical one—making the determination depend on “intuitive” (in the extended Kantian sense of the word) rather than descriptive knowledge is not freeing it from dependence on— internally available knowledge. The Russellian is not saying: sitting in the Chamonix cafe, I can refer to Mt. Blanc because I directly see it. The Russellian is rather saying: I can refer to Mt. Blanc because my seeing it induces in me knowledge of-it (often: knowledge of which thing it is) and this knowledge is the key to reference. In the end, whether by intuitive or descriptive knowledge, the determination of that semantical relation is said—by Frege or by Russell as by their modern apostles—to rest on the epistemic state—the state of knowledge and by this I here mean really justification—of the candidate name user. As is familiar from his work on numbers and extensions, Frege was uneasy about intuitive justifications; Russell may have been more lax (in any thing he did). Either way, both look for justification as the foundation of the relation of reference. And the modern Neo Russellians, still more lax than Russell on when we have the justification, are nonetheless all justification-reductionists: reference rests on the justification affordable by that agent. This justification-key is the the Kant-inspired basis of the FregeRussell approach to the determination of reference: I need some justification to ground my use of “Berthold” in one worldly individual, x and not

The Puzzle That Never Was—Referential Mechanics

25

another individual, y. Barring such justification, there is a breakdown of the norm that I need to be justified in using language to refer to objects.

Back from the acquaintance-tangent: the determination puzzle In any event, whether we allow, like Russell (Kaplan, Burge etc.) acquaintance or, like Frege, bar it, he determination-puzzle sits there in front of us, like the huge Mt. Blanc in front of the alpine climber sitting in the cafe in Chamonix. For Frege and Russell agree—the man Aristotle is no acquaintance of Frege or Russell and the historical name “Aristotle” in their mouths is—indeed, as is often with Kantian epistemologists this soon really means must be—conventionally associated—its a Kantian rule to be followed!-with a definite description, let us say “the best student of Plato”. And thus, a competent user of the name would have the oracle-description in his “head”. And finally, whoever, as a matter of historical fact, satisfies the condition is thereby determined for that thinker-speaker as the referent. This is why this historical man—Aristotle (but you can call him “Zorg”, not to get confused between bearer and name) is the “right” referent (of this use) and no one else is.3 This determination-mechanism, Kripke suggests,is viewed by Frege and Russell as the ur problem of the theory of reference. And on this ur problem, the score is stark: Frege-Russell (keep in mind, these are two philosophers, not one fictional man Kripke has invented) have a powerful answer to it; Mill has none. And answer it or not, the question is presented by Kripke (NN, 27–28) as a confounding puzzle, as if magic or action at distance is taking place: 3 As is familiar, Kripke, Kaplan and others pointed out that Frege-Russell may have further problems in accounting for the reference in modal discourse (or “at” possible worlds) if the oracle description is not modally rigid (assuming, as KripkeKaplan do, the name is). The problem of determination of reference is not about modal discourse. It is about reference that is pre-modal predication. Given the referent, we may go on and ascribe to it the modal predicate (if Frege-Russell still object to such a modal predication of the object, this has to do with their skeptical (Quine-like views) about modal predicates as genuine predicates, not their views about reference determination). The issue in the current paper is the primal, real world bound, issue of reference-determination, not reference in counterfactual situations.

26

Joseph Almog

how on earth (or across the solar system, see Kripke and Donnellan below on “Neptune”) does a name “reach out”? Where, in what place, if it is in physical space at all, does the reaching-out and determination-of-reference occur? What is the mechanism, the referential mechanics, grounding the name’s semantic orbit—e.g. for “Aristotle”, the fixed orbiting is around that Greek man Zorg, aka Aristotle? For a sense of the problem, recall that with planetary orbits, we have celestial mechanics, whereby a genuine mechanism—gravitation—dissolves the puzzle; with atoms (say hydrogen) we have another mechanics— quantum mechanics—to cement the intra-atomic orbits. What mechanics governs the semantical orbits of names? All we are given at the out-set is a name—“Aristotle” (and sometimes, a current user). That’s where Kripke (not just Frege and Russell) makes us start, and where we start in philosophy is often as critical as any further move made later. Given this starting point and the endless plurality of worldly objects, it seems sheer magic that the name would nab this one object, ancient Greek Aristotle, and no others. To dissipate the oracular feel, we need a mechanism determining the reference relation from the name (and user of ) “Aristotle” to the man Aristotle. The italicized prepositions from . . . to . . . hold the key here—Kripke, just like Frege and Russell, poses the problem as involving a directed relation, viz. from the head of a current user to the ancient man who lived 2300 ago. This casting of the issue does strike me as leading to a puzzle.

The puzzle that never was So much for stage setting and puzzle mongering. My own proposal is disappointingly simple. There is no and never was a puzzle of determination of reference. By adding “and never was” I mean to indicate I do not think the puzzle has been dissolved or resolved by some path-breaking recent theoretical development—the kind of theoretical advance Newton’s gravitational mechanics or the later quantum mechanics offered us in order to solve the above cited orbit puzzles, puzzles we did not have the means to solve before. It is not as if the determination of reference puzzle—the explanation of semantic orbits—was ever a truly mysterious puzzle shared by umpteen generations which only a latter day stroke of genius decoded. Rather, I claim, as in the tale of the impertinent kid and the Emperor’s new clothes, there never was a puzzle here to begin with. Any person—including those

The Puzzle That Never Was—Referential Mechanics

27

of past millennia—knew the answer to “the semantic orbit” mystery before breakfast. I don’t say this in a coy or ironic vein. I think the full account of the matter—a complete description as philosophers like to say—is included in the following paragraph I will quote from he who poses it as the darkest mind-boggling puzzle, Kripke. And it is obvious that this paragraph was accessible to any person using names 3000 years ago, way before mankind ascended to the realization there are infinitely many primes. Says Kripke:4 Someone, let’s say, a baby, is born; his parents call him by a certain name. They talk about him to their friends, other people meet him. Through various sorts of talk the name is spread from link to link as if by a chain.

There; the whole determination puzzle dissipated in a short paragraph and please notice no rigid designators, possible worlds, singular propositions and other path-breaking hitherto unavailable to mankind advanced scientific technology was ever mentioned. The only remaining sense of a puzzle is due here to excessive philosophical reflection—at that, not amplifying reflection extending obvious observations (e.g. those the just quoted “baby-is-born” paragraph) but, to the contrary, counter-observational reflection—reflection that embeds us (indeed, drowns us, would be more correct) in a certain complex picture engendering a storm in a tea cup. In this respect, the determination-puzzle blown up here is in my view just like the puzzle of contingent identity or skepticism about the external world: a certain philosophical theory engenders an illusion of depth—lost on ordinary men who live foolishly un-puzzled—and with the “depth” come unresolvable . . . deep puzzles. Of course, such deflationary kid-like observations tend to have a bitter fate among the professionals. When the kid tells his mother: the emperor is naked, and no mother, it is not a sophisticated 22th century hyper transparent gown made of new alloys, its really just the plain bare regal body, as e.g. G.E. Moore tried to tell us with his two hands, both the kid and Moore were dismissed as childish and naïve. Luckily, Kripke could use the authoritative baton of “modal logic” to undo the contingent-identity darkness. In fact, as anybody who knows the fine print appreciates, no distinctly “modal” logic ( past the simple Leibniz law applied to modal features) is really invoked. Well, leaving skepticism and contingent identities to others, I will only assert here about reference determination—there is no and never was 4 NN, op. cit

28

Joseph Almog

a puzzle concerning the determination of this most common relation, reference.

Frege-Russell vs. Kripke Why has the puzzle spellbound us all? I do understand why Frege and Russell (and all those NN criticized “description theorists” following them, Strawson, Searle etc.) got themselves into a knot about the horrid determination puzzle, though, as we shall see, the originator of the puzzle, Frege, was in one of his own key footnotes, a breath away from bursting the bubble. But what I do not understand is why Kripke plays along and fans the flames of a deep puzzle here. He surely had the resources (and the wonderfully good ear aside of his theories, see e.g. the baby-is-born paragraph) to bust the mystery in one short paragraph (as he did with contingent identities). In other, more technical, realms, Kripke was the one to bring in common sense just where philosophers looked for mystification and dark insolubilia. Thus (following a line of thought also practiced before him by Cantor and Skolem), I heard Kripke give crystalline and full of humor lectures where the likes of Cantor’s paradox, Russell’s paradox and Tarski’s paradox were demystified in short paragraphs—there is no dark paradox here, only a theorem, an impossibility theorem (there is no Russell or universal set; there is no first order definition of a “truth” predicate)5. I believe that the exposure of the hoary paradoxes of the theory of reference—and leading them, our mother paradox of reference determination—requires much less insight than this modern logic/set theory recasting of a technical paradox as an impossibility theorem. In the theory of reference, Kripke’s “theorem” was known to man (in the dark caves) thousands of years ago. And yet, for quite a few of these puzzles of reference, Kripke sustains the reader’s illusion (and often unconscious craving) for a puzzle or a paradox. E.g. he makes the second and third problems sound as extremely hard puzzles (he says he won’t even touch them in NN ) when, in fact, he deals expeditiously with both in other places. And he cunningly brews a huge puzzle over poor Frenchman Pierre—rational on ne peut plus—who is being reported truly by the

5 Skolem’s paradox, another favorite of philosophers, is another such impossibility (first order indefinability) theorem.

The Puzzle That Never Was—Referential Mechanics

29

conjunctive sentence “Pierre believes that London is pretty and Pierre believes that London is not pretty”. Kripke could have said, like Descartes in the end of Meditation VI: I was only joking with this hyperbolic doubt, the crisis was really a reductio of the skeptic.In like vein, Kripke could have said: there is no puzzle with Pierre; there is a true report of an absolutely fine logical thinker. The Pierre situation is not paradoxical, we each live it twice a week and the only puzzle here is not for that-man but for theories that insist “believes that” is some a so called “de dicto” “propositional attitude” towards certain kinds of propositions(“contents”). Kripke could have said this and read the hoary puzzle as an impossibility theorem. But he did not.6

Where is the puzzle and where it is not? So let me say what is going in simple terms, I am afraid somewhat disappointingly for those who look for murky depths where it is all limpid shallows. Quite systematically, the standard philosophical methodology has inverted—in between Frege/Russell and the view of “Mill”—the sense of who is in a crisis over the hoary puzzles and who smoothly sails on. The standard presentation—Kripke’s included—suggests that on all three puzzles and led by the ur-problem of determination of reference—F/R have a marvelous solution and Mill is bankrupt. And what is more, and more important, is that all three problems are deep problems, especially the source problem, the determination of reference. Instead, I would like to claim—on all three fronts there is no pre theoretic puzzlement; furthermore, there is no insoluble position to a Mill like purely referential view of names; and what is more, the positions for which there are three real problems here are the Frege and Russell positions, especially so on the determination of reference front.7

6 Quine did state such an impossibility result very clearly in “Quantifiers and propositional attitudes” in his earlier 1956 sortie on whether we can describe the cognitive life of his point man Ralph by purely de-dicto (or sentential contents involving) relations. For all the new Pierre dramatic paradox pyrotechnics, Pierre is a variation on Ralph. 7 I will not discuss here the two other puzzles by will only claim: it is not the assertion (as in the “description theory of names”) of synonymy between a name and controller description that will suffice for F/R to “solve” the alleged puzzles. See my “Is a unified description of language and thought possible?” J. Phil 2005.

30

Joseph Almog

How does the determination puzzle gets going? The puzzle is posed by presupposing the Frege or Russell picture about historical names. On this F/R view, we start with a historical name, a name already in circulation, say “Aristotle” and as a matter of historical fact loaded (on a given use) with a referent, the man Aristotle. We now indulge in the philosopher’s favorite—rational reconstruction. This is often a euphemism for us philosophers re-doing things cosmic history already did but in a cleaner, more ‘rational’, way. This philosophical rationalizationimprovement on cosmic history often means, ironically for that is the height of ir-rationality, that we end up not being able to do the original thing (e.g. reference to individuals) at all. In the Frege-Russell case, we rationally re-construct the name’s history by abstracting from (that is, subtracting! ) all its circulatory-history, its origination-facts (viz. a baby, Aristotle, was born. . . .) and of course, we subtract the subsequent processing by human communicative history of the name, bringing it all the way to our time, in particular to a user like myself, to my brain/mind. We do the same for the natural history of the name “Plato”; we “logicize” it by subtracting all the historical facts of origination and transmission. Now, it is a huge puzzle how (why?) “Aristotle” stands for this one historical figure, man x, and “Plato” for this other historical figure, man y. Why not “Aristotle” for y and “Plato” for x? We may well pick up the color of my eyes, brown, and the color of Paul Newman’s eyes, blue, and subtracting the sperm and egg we each came from and the subsequent zygotes development into two babies etc., we ponder how (why?) I am brown eyed and he blue eyed? Why not switch the colors? Frege and Russell totally drop the historical relations that fully determined the life and times of (uses of ) “Aristotle”—indeed multiplied over for each use, because many “Aristotle” histories exist, many men so called etc. And having erased this genetic information, they raise a huge puzzle.

The internal determination problem in Frege and Russell So much then is my take—already in pre theoretical times, in caveman’s uses, prior to the debates of Mill and Frege and Russell and Kripke—no magic, action at distance, or mechanical witchcraft was called upon to “explain” the determination of the reference of (the likes of ) “Aristotle”

The Puzzle That Never Was—Referential Mechanics

31

(on a given historical use). Kripke “discovered” in 1970 the bare obvious facts, a repeat of the kid’s discovery of the emperor’s nudity. On the other hand, for Frege/Russell there is—and there must be—a genuine puzzle. Why? The answer is coded on p. 79 of Kripke’s own NN, where Kripke tries to imagine how Frege/Russell may have started to worry about . . . the determination of reference. The ground zero case for Frege/ Russell, their paradigm, is introducing a new immaculate name and not the accounting of the determination of an already loaded historical name. With immaculate names, F/R may well think there is no prior metaphysical/ontological determination of the reference facts. In this (onto-) logical sense—and not in the stipulated epistemic sense favored by Russell— that the de novo un-interpreted names are logically proper names; they are like individual constants of a formal language that are yet to be assigned any reference. Of course, we must be absolutely pure—history cleansed—about this imagined F/R origination “logical big bang” situation. E.g. Kripke is not satisfactorily pure in the “Neptune” introduction case—which he discusses on p. 79 (fn. 33) and that is exactly what Keith Donnellan criticizes him later for 8—it is already metaphysically determined, by historical wiring facts, which planet Leverrier thinks of—he was looking at “footprints left by the burglar”, viz. perturbations of Uranus’ orbit left by . . . Neptune mass. The introduced name is subordinated to that information channel just as a detective introducing the name “Jack’ on the basis of the burglar’s footprints. So “Neptune” is not un-determined at the out-set as Kripke intimates. The (satisfaction of the) description (“the entity causing the perturbations according to Newtonian mechanics”) does not metaphysically determine a referent for Leverrier’s use; for one thing, the description is literally false of Neptune, because Newtonian mechanics gravitation theory mis-describes the planetary orbits (as later was strikingly clear with Leverrier’s positing of . . . Vulcan). A striking example of the non determinative role of the description is offered by Kripke in passing on the very same page 79—the “fixing” of “Hesperus” with Venus as the referent by the “reference-fixing” description ‘The evening star”. Obviously in this case, it could not be satisfaction of the description that determined—logically as it were—Venus as the

8 Donnellan “On the contingent apriori and rigid designators”. See also for a discussion of the “Neptune” debate, my editor’s introduction to the volume of Donnellan’s essays, Essays on Reference forthcoming 2012, OUP.

32

Joseph Almog

referent, for Venus is no star; and assuming that x-328 at the other end of the galaxy is the first star to shine in the evening but that none of us was not in contact with (no light from that area impinging) that direction of the sky, it would be a false prediction to suggest that “Hesperus” referred to x-328. And for a good reason: Venus was ( going to be) “Hesperus”’s referent quite independently of any posterior externalization of that thinking-of-Venus to others by means of the directing “the evening star” which, though false, like a slightly inexact pointing finger, can communicate what I am already thinking of.

Immaculate names? Frege on “Aristotle” We need to drop “Neptune” and “Hesperus” and discuss an absolutely immaculately fresh name, if such there be. In such an originary situation, F/R think that metaphysical and epistemic determination go hand in hand—the originator (the dubbor) is locally in the mantle of a god; he creates the very logical-metaphysical facts by his very action; and in turn, the originator may be said to know apriori what he created, because he knows independently of sense experience his own intentions. This seems like assigning de novo a hitherto un-interpreted formal symbol of a formal language an interpretation. What F/R try to do is take this purely internal immaculate conception of a name as a model and then extend this internal-model to common currency historical names in circulation. This is why in Frege’s often neglected but very important footnote—it is really the Michelson/Morley experiment of semantics—on the historical name “Aristotle”. Says Frege:9 In the case of an actual proper name such as ‘Aristotle’ opinions as to the Sinn may differ. It might, for instance, be taken to be the following: the pupil of Plato and teacher of Alexander the Great. Anybody who does this will attach another Sinn to the sentence ‘Aristotle was born in Stagira’ than will a man who takes as the Sinn of the name: the teacher of Alexander the Great who was born in Stagira. So long as the Bedeutung remains the same, such variations of Sinn may be tolerated,

Frege wonders: what if the allegedly metaphysically creating-determining fact in Joseph’s head (“best pupil of Plato”) and in David’s head (“teacher

9 “Über Sinn und Bedeutung” footnote 4.

The Puzzle That Never Was—Referential Mechanics

33

of Alexander the Great”) are different? Frege says in response—“as long as the bedeutung is the same this may be tolerated”. . . . and thinks he is out of the woods. I don’t think so. What denotation is “the same”? It can’t just be any old same-denotation (say, Euclid or Caligula), for Frege is not indulging in a “model theory” for a purely formal un-interpreted symbol “A” that may land in a given model wherever it may e.g. Euclid or Caligula or. . . . The historical name is damned because it is already bound. “Aristotle” is already. . . . historically fully determined and as such determined to be of this one Greek man, Aristotle. So, Frege, as a user of the historical name “Aristotle”, is bound too. He must co-ordinate Joseph and David beliefs to be not just about the same object but about that man, Aristotle. Indeed all the beliefs Frege cites are referential (in Donnellan’s sense)—they are coming from this one man . . . Aristotle (even if the content happens false of him). In the footnote, Frege is presupposing, as he “determines” reference for “Aristotle”, the more primal notion of belief-about-Aristotle. As Donnellan would say: Frege presupposes that David, Joseph, etc. with all their varying descriptions are having-in-mind . . . Aristotle. Like Kripke, Frege had his “a baby is born . . .” moment, in this confessional footnote. Frege too should have said, at least for historicallydetermined names: I have discovered that there is no determination of reference to be indulged in; reference is already fully determined. Of course, he could not say this—he would have refuted . . . Frege’s basic conception of how we manage the magical feat of thinking of remote objects. But his footnote says it for him.

Ever internally determined? You can now ask yourself, in the vein of the Kripke/Donnellan debate on the introduction of “Neptune” by Leverrier, whether there is ever any case (to simplify: of non mathematical reference; though the mathematical case will in the end converge), where the internal model determination of Frege works—the head of the thinker creates-makes the metaphysical-logical basis of the reference fact; and as the creator of the key fact, the thinker knows apriori that “Aristotle is The F”, “the F” being the logical miracle-fixer? On my reading, Donnellan’s answer is: no, no such case is ever possible for a name making a genuine singular claim of/about Aristotle. The thinker must already have in mind the target object, must have his mind

34

Joseph Almog

already determined by (i) the object and (ii) the history of the connection from the object. As a cosmic act(ion), the name’s introduction is a re-action to the object’s prior action on the introducer’s mind. As for Kripke’s answer to the question, I am not sure. The man who gave us the incisive “a baby is born . . .” passage should have been with Donnellan on this. But I don’t think he is and he seems rather with Frege. Sure enough, unlike Frege, Kripke thinks the name, unlike the referencecreating description, is modally rigid. But like Frege, he lets the description make-create the logical basis of the reference fact; and as Kripke emphasizes proudly, he thus “gains” the apriority of the knowledge for the creator. This magical creationist apriority should remind us that one philosopher’s modus ponens is another philosopher’s modus tollens.

Reference and Meaning William P. Alston Abstract: The author criticizes prevalent confusions among philosophers of language with respect to their employment of the terms ‘reference’ and ‘meaning’. First, he objects to the undiscriminating use of the terms ‘refer’, ‘reference’, and ‘referent’ for all of the semantically significant relations between linguistic units and what they are used to talk about and for the relata of the latter sort. Second, he attacks various projects that involve attempts to let reference take over the role of meaning, thereby shortcircuiting any problems about the relations between meaning and reference.

I This paper is motivated by a conviction that there are widespread, though certainly not universal, confusions, conflations, and just plain sloppiness among philosophers of language with respect to the terms ‘reference’ and ‘meaning’ and their cognates. These give rise to unfortunate results of differing degrees of severity, ranging from crucial theoretical mistakes, on the high end, to infelicitous and incautious formulations that are apt to mislead the unwary on the low end. The paper documents these charges. Rather than plunging straight off into the documentation, it would be useful to lay out what I take to be linguistic meaning and how it is related to ‘reference’, more soberly to the various relations between meaningful linguistic units and what they are used to talk about that are often undiscriminatingly lumped together under ‘reference’. To do this I will also need to specify and distinguish the various relations termed ‘reference’. The defects in handling this material that I will be exposing represent various deviations from the picture that I claim to get it straight. I will not present any significant argument for my picture. It seems so clear to me that it is the right way to think of the matter that I cannot imagine any opposition to it that is not based on the confusions and conflations that I shall be documenting. If that sounds intolerably dogmatic to you, that reaction may be reconsidered when you realize how abstract, how neutral

36

William P. Alston

between a large variety of specific positions, my picture is. I am not presenting any detailed account of linguistic meaning or of any of the various items lumped under ‘reference’, but only an outline of how they fit into a general picture, an outline that can be filled out in a large variety of ways. I will, for conciseness, stick to a consideration of subject-predicate sentences. It should be clear how the picture could be extended to other cases. To cut to the chase, I want to unpack the different relations of linguistic expressions to what is incautiously called “the world”1 that are lumped together under the ‘reference’ label. But since it is an important point that they are related differently to meaning I had better first say just a word about the latter. That won’t take long because in presenting my picture I will be largely concerned with differentiating the various whatevers lumped under ‘reference’, and linguistic meaning will be characterized only in the most neutral functional terms. To put it briefly, the (a) meaning of a linguistic expression is a feature it has in the language, has because of the (semantic) structure of the language. Possessing knowledge of that semantic structure (and hence of the meanings of many expressions in the language), at least a working, know-how sort of knowledge, is a crucial part of what it is to “know the language”, to be a fluent speaker of the language. When one hears (reads) an utterance, knowing what it means, and what its component meaningful linguistic components mean, is what is required in the way of knowing the language, over and above phonology and grammar, to understand what is being said. These are the fundamental constraints on what counts as linguistic meaning. They locate the concept of linguistic meaning. They serve to identify what it is that theories of the nature of linguistic meaning are theories of. And since my picture is designed to be neutral with respect to the differences between such theories, that is all I will say here about how to understand linguistic meaning. To get back to “reference”, where better to begin than with the concept that is most often expressed in ordinary discourse by ‘refer’ and its cognates, where it is language and the use of language that is in question. And that, to put it roughly and with the degree of unspecificity at which I aim here, has to do with what a speaker meant to be referring to with an expression that is designed to pick out what his utterance was “about”, what it was that he was going on to say something about. These are expressions that are suited to be the subjects of declarative sentences, inter alia. I shall call them ‘referring expressions’. They include noun phrases in 1 Incautious because it implies that linguistic expressions and their use do not belong to the world.

Reference and Meaning

37

the singular and plural, pronouns, and proper names. Thus appropriate answers to “To what (whom) were you referring?” include such responses as ‘the car parked across the street’, ‘Jack Robinson’, ‘weasels’, and ‘that dog over there’. Individual pronouns, as in (‘She is late’) are more likely to give rise to such a question than to answer it. There are two crucial things to be noticed about ‘reference’ in this ordinary sense. (1) With an exception to be noted in a moment, it is speakers who refer, rather than linguistic expressions. If you are asked, out of the context of a particular utterance, what does ‘she’ or ‘that dog over there’ or ‘Jack Robinson’ refer to, the only answer is “It all depends”. It depends on who was using it, when, under what conditions, etc. To put it in a slogan, Reference is a phenomenon of speech, not of language. A fact of reference, what is referred to, is a matter of how an expression is used by a speaker, not a matter of the semantic status (meaning) the expression has in the language. To put it in another slogan: Meaning underdetermines reference. Just by virtue of being a fluent speaker of English, you can know the meaning of the phrase ‘that dog over there’ or ‘she’, but that won’t be sufficient to know what was being referred to by one of those expressions on a particular occasion. (2) And that brings us to the second crucial feature of reference in this ordinary sense. What is needed in addition to the meaning of the expression to determine the reference are various features of the context of a particular utterance—the identity of the speaker (for first person pronouns), demonstrative gestures or functional equivalents (for phrases like ‘that dog over there’), the speaker’s intentions (for which of the people named ‘Jack Robinson’ she had in mind), etc. This simple picture needs to be complicated in several ways, even though no particular theory of reference, such as a causal theory, is being spelled out. (a) If, as many philosophers hold, proper names do not have meanings in the language, then in their cases meaning does not even partly determine reference, there being no such meaning to do so. On this view, knowing what a proper name names is no part of knowing the semantic structure of the language. (b) An apparent exception to “Meaning underdetermines reference” is provided by “purely qualitative” definite descriptions (those that contain no referring expressions that are underdetermined by meaning like “Susie’s husband”). ‘The first person to run a 4 minute mile’, assuming that this description is satisfied by one and only one person, would seem to have its reference fully determined by its meaning. But such expressions constitute only a tiny proportion of referring expressions. And, indeed, some philosophers deny them that title altogether. (c) As Keith Donnellan famously pointed out (see Donnellan 1966), speakers

38

William P. Alston

have the capacity to override the (b) kind of reference that is fully determined by meaning. Thus if in introducing a speaker, I refer to her as “the chairperson of the Harvard Philosophy Department”, mistakenly supposing her to be holding that position still, Donnellan argues, convincingly, that I could still be referring to the speaker even though the meaning of the expression points to some other referent. But despite these complications, for the great majority of referring expressions, the crucial points are (1) their reference is a matter of speech, not (solely) of the semantics of the language, and (2) what else is involved has to do with various features of their use on the particular occasion of speech in question. When we turn to predicate terms—nouns, adjectives and adjectival phrases, verbs and verb phrases, which are used to say what one has to say about what is picked out by some referring expression, we get a quite different relationship of the meaning of the term to what it is used to talk about. The “relation to the world” that is most analogous to the referent of a referring expression (though not analogous enough to justify being called “reference”) is extension, the class of things of which the predicate is true, the class of things of which it can be truly asserted. Thus the extension of ‘cat’ is the class of all cats—past, present, and future and wherever they may exist. Another choice from the world of something to which ‘cat’ is related in a semantically salient way is the property of being a cat. But for various reasons, the most important of which will emerge in the course of our discussion, it is the extension of a predicate term that is most often put under the referent rubric. The way a predicate’s extension is related to its meaning is importantly different from the way the referent of a referring expression is related to its meaning, if any. Most crucially, so long as the meaning of a predicate remains constant its extension does not vary in a given possible world. As we will see, this is often put in an unfortunate way—or not to put too fine a point on it, an incorrect way—by saying that the meaning of a predicate determines its extension. But, of course, it would be absurd to suppose that the meaning of ‘cat’ determines just what cats there are, have been, and will be in the world. The meaning could have been just what it is even though the population of cats were as different as you please from what it is. So here, just as with referring expressions, meaning underdetermines reference or, in this case, what masquerades as reference, viz., extension. But still the details of this underdetermination are different, and the first difference is the one noted above. The extension of a predicate remains constant in a given possible world so long as the meaning of the predicate does. In particular the extension does not vary with features of

Reference and Meaning

39

the context of particular utterances, as is the rule for referring expressions. So, as we might say, the extension of a predicate is closer to being determined by the meaning of the predicate than is the case with the referring expression-referent relationship. But just how is the extension of a predicate related to its meaning? How does the partial determination work here? In my view, one not, I fear, universally shared, it goes as follows. The meaning lays down necessary and sufficient conditions for belonging to the extension. The meaning of ‘cat’ embodies such conditions for x’s being a cat, for any x. It is then up to extra-linguistic fact what x satisfy those conditions. That is a simple way of putting it. But to have any chance of being acceptable, various complications need to be recognized. First, the above characterization fully applies as stated only to perfectly precise predicates, and most predicates in natural languages do not come up to that standard; their meaning exhibits various kinds of indeterminacy. To illustrate this with the familiar case of degree vagueness, how many inhabitants does it take for a community to belong to the extension of ‘city’, rather than to the extension of ‘village’ or ‘town’. The meaning does not yield a precise cut off point here. There are clear cases, of course. Chicago is a clear case of a city, and Lower Slaughter in the Cotswolds of southwest England is a clear case of a village. But there is a sizable area of indeterminacy in between. How about 4,00 inhabitants? And for that matter, are there perfectly determinate criteria for someone’s being an inhabitant of a community? Residing there, you say. Yes, but for what proportion of time? And so on. To be realistic we have to take the above formulation of the meaningextension relationship as depicting an ideal case to which most actual cases approximate more or less. Another and more controversial complication concerns the question of whether the meaning of a predicate is purely a matter of the semantics of the language, something that is relatively stable, though of course subject to change over time, and unaffected by contexts of utterance. This has become a major issue in the philosophy of language because of the contention of Putnam and Kripke2 that natural kind terms, and perhaps other predicates as well, have a kind of meaning that involves, inter alia, a set of standard examples. At a first approximation (and more of this later), the view is that what ‘cat’ means is something like this: items that have the

2 See Putnam 1975 and Kripke 1972.

40

William P. Alston

same essential nature as the items in C (a standard set of paradigm cases of members of the extension of ‘cat’). One thing that made this position revolutionary is that it implies that meaning is not wholly a matter of “what’s in the head”, not wholly a matter of a set of conditions, schemata, or patterns that speakers of the language carry around in their minds. Instead an essential element in the meaning is what paradigms of items in the extension are socially recognized in the language community, as well as what essential nature is shared by those paradigms. For terms to which the Putnam-Kripke view applies the contrast with referring expressions is therefore less sharp than it is for terms that satisfy the more traditional view of meaning as a set of internalized conditions for the application of the term. But the Putnam-Kripke view still conforms to the generic account given above of the meaning-extension relationship. The meaning still lays down conditions for membership in the extension. It is just that these conditions themselves involve facts external to the general concepts employed by users of the language. And so the interplay between meaning and extra-linguistic factors becomes more involved. But the original characterization of the meaning-extension relation for predicates given above can remain in force, with the earlier qualification to take care of the pervasive phenomenon of greater or lesser indeterminacy of meaning. That will do as a preliminary statement of what I take to be the bare outlines of the relations between meaning and reference-or-extension for two crucially important groups of meaningful terms. As we go through a discussion of ways in which talk of reference and meaning deviates from this picture, further aspects of the picture itself will emerge. But all in good time.

II The sins of commission and omission that I will survey fall into two main groups. First, there is the undiscriminating use of terms like ‘refer’, ‘reference’, and ‘referent’ for all of the semantically important relations between meaningful linguistic units and what they are used to talk about and for relata of the latter sort. Second, there are various projects that involve attempts to let reference, in the inflated sense just objected to, take over the role of meaning, thereby shortcircuiting any problems about the relation of meaning and reference, i.e., any relation other than identity. The first group is simpler than the second, since it doesn’t involve explicitly bringing

Reference and Meaning

41

meaning into the picture at all. Hence I will begin with the simpler target and only then examine the more complex one. The inflated use of ‘reference’ in which it includes the extension of predicates as a special case is so common in current philosophy of language that it takes a special effort to realize how far removed it is from any ordinary use of the term. If I say “That’s a lovely cat” and you ask me to what I was referring, the normal reply would be to make clear what my demonstrative ‘that’ was intended to pick out. But it would be puzzling at best if I were to reply, “Well, for one thing I was referring to all cats—past, present, and future”. I’m afraid my conversational partner could make no sense of this unless he were a typical philosopher of language. It is interesting to find philosophers using ‘reference’ and its cognates in this inflated way, while, from time to time, evincing awareness of the differences in the way referring expressions are related to referents and predicates are related to extensions. See the citations from Putnam and Devitt in this section. Well, you might say, what’s all the fuss about? Sure, this isn’t ordinary language; it’s technical language. But what’s wrong with that so long as it’s understood. Indeed! But it is not clear that it is always understood. It contains seeds of confusion and misunderstanding. And this can go in both directions. Thinking of the extension of a predicate as its “referent” can encourage distorted ideas of (most) genuine referring expressions, neglecting the way in which the referent can vary enormously over different uses without the meaning of the expression changing, and thereby obscuring the way in which the relation of a referring expression to its referent is a matter of speech rather than of the semantics of the language, and diverting attention from the crucial contribution of context of utterance to the identity of the referent. Or if attention is focused on referring expressions and predicates are thought of in terms appropriate to the former, there is the danger of falling into thinking of predicates as having the function of picking out what an utterance is about rather than providing something to say about it. I will now proceed to present some cases that make it clear that these are no idle possibilities. It is the second confusion, treating predicates like referring expressions that is the more clear and present danger. But before turning to that I will illustrate the less common direction of conflation—treating referring expressions like predicates, ignoring the way in which their referential function is a feature of speech, underdetermined by linguistic meaning. There can be no doubt that this tendency is largely due to the practice of creating “artificial languages” (better termed “artificial fragments of language”) for

42

William P. Alston

purposes of logic. There the usual procedure is to introduce names and predicates in the same way—by assigning individual referents to the names and extensions to the predicates. Since this is what passes as assigning meanings to the terms of the language, as this game is played the referent of a referring expression is fully determined by its “meaning” just as the extension of a predicate is, and that in both cases “reference” is a fact of the semantics of language, rather than, as in ordinary life and with real languages, a fact of speech in the former case. I have no objection to setting up “artificial languages” in this way for logical purposes, provided it is fully realized how different this is from the situation with respect to real languages. If anything is completely clear in this subject matter, it is that speakers of natural languages do not learn the meaning of a predicate by being given an exhaustive enumeration of the members of its extension, something that we are virtually never in a position to do even if we were foolish enough to attempt it. And for that matter, neither do we learn the meaning of pronouns and demonstratives by being given an enumeration of all the items they can be used to refer to, again a quixotic enterprise. The damage is done when philosophers whose thinking about natural languages is over-influenced by their familiarity with artificial languages proceed to treat referring expressions in the former as if they were items in the latter. Thus we have Kripke in a well known essay, Kripke 1979, distinguishing between “speaker’s reference” and “semantic reference” of referring expressions. The fact that the first side of this contrast is recognized shows that some account is taken of reference as a feature of speech, but the presence of the second side of the contrast indicates a tendency to think of the reference of referring expressions as sometimes being a fact about the semantics of the language or at least of an idiolect. To substantiate this diagnosis I would have to go into the complex details of Kripke’s discussion. For example I would have to bring out how Kripke uses his distinction to contest certain claims of Donnellan about definite descriptions and their treatment by Russell. I have no time for that here. I will merely suggest that only one whose thinking about such matters is strongly influenced by dealing with artificial languages would think it plausible to suppose, as Kripke does there, that certain conventions of a speaker’s idiolect determine the referent of a proper name in that idiolect. To turn to the more common conflation, here is a particularly blatant example from Nelson Goodman’s essay, “About”, reprinted in Goodman 1972. There he is looking for a principle on the basis of which it can be determined what a sentence is about. This leads him to the sensible

Reference and Meaning

43

suggestion that a sentence is about anything that any of its components “designates”. But the further extension of this line turns out to be less sensible. . . . a predicate will be considered to designate its extension: the class of those elements it applies to or denotes. A sentence may be said to mention whatever any expression in it designates. Thus Maine prospers Utah is west of the Pine Tree State The Atlantic State farthest from Florida is agricultural all mention Maine, by name or description. The first also mentions the class of things that prosper; the second mentions Utah, the class of things west of Maine, the class of things of which Utah is west, and the relation (class of pairs) designed by “is west of ”; and the third sentence mentions another State and several other classes and relations. (248–249)

Thus by forcing predicates into the function (designating) that is proper to referring expressions, we generate the absurd consequence that in saying “Utah is west of the Pine Tree State” I mention, inter alia, the class of things west of Maine and the class of things of which Utah is west. Here we purchase an inflation of designation at the price of absurdity. For another example consider Putnam’s argument for an indeterminacy of “reference” of terms in Chapter 2 of Putnam 1981. There he speaks of the “reference” of both predicates and referring expressions. He tries to make this procedure respectable by introducing some “technical terms, viz., ‘extension’ and ‘intension’, the latter construed as an ‘extensionfunction’. In these terms he provides a uniform treatment of singular referring expressions and predicates by, in effect, treating them like predicates. Just as the constant meaning of ‘cat’ is a function whose value at every possible world is the set of cats in that world, so the constant meaning of ‘I’ is a function the value of which in each utterance is the speaker. By this maneuver Putnam manages to obscure the differences in the way the referent is related to meaning for referring expressions and the way the extension is related to meaning for predicates. No small achievement! But, again, bought at a price. And it is noteworthy that in order to set up his machinery that gives him a common terminology for referring expressions and predicates, Putnam has to explain ‘extension’ for predicates as the class of things they are true of, and explain ‘extension’ for referring expressions as the referent (in each use). Otherwise we wouldn’t understand

44

William P. Alston

what he was saying. And so the different treatment of the two sorts of expression has to be brought in at the foundation of the project in order to give different accounts of what ‘extension’ means for the two. The differences are only covered up, not abolished. That way lies confusion. Another conflation under the banner of ‘reference’ can be found in Devitt 1981 and 1987. At least he is up front about what is going on. As the title of his 1981 suggests, ‘designation’ is an central term for him. It is initially introduced as a central term for a “semantically important relationship that holds between a name and a certain object”. (7) This doesn’t help us much, because it doesn’t tell us how to locate the object so related to a given name and hence how to identify the relationship. As he points out, he could have introduced the term by saying that it picks out the relationship that holds between a name and what it names, or between a name and that to which it refers or which it picks out. But he has a motivation for not bringing in ‘refer’ here, since he is giving it a wider role, as we shall see in a moment. In order to do a proper job of explaining the way Devitt sets things up, we would have to go into his causal theory of the designational (referential) function of designational terms, since his final account presupposes it. But there is no time for that in the present essay. At least we can point out that his general conception of a ‘designational term’ (more or less what I have been calling the usual sort of a singular referring expression, leaving aside purely qualitative definite descriptions) is a token that refers ( picks out) a particular object by virtue of a causal connection with that object, either relatively directly in case the speaker has perceived the object and “baptized” it (to use Kripke’s phrase) by the use of a token of the type in question, or more indirectly through a chain of transmission reaching back to an initial “baptism”. Designation is then what designational terms do vis-à-vis their objects. He then introduces ‘reference’ as a generic term, ranging over different species of semantically important relations between meaningful units and “the world”, or, as he puts it in the Glossary, “the genus of which all referential relationships are species” (278). I follow Quine in using ‘refer’ as a generic term: proper names, predicates, variables, and so on, may all refer. I use ‘apply’ to express the special relationship between predicates and the world; so ‘raven’ applies to each and every raven and to nothing else. (8–9)

Since Devitt is so explicit in distinguishing the “referential” relation that singular referring expressions and predicates have to what they are used to talk about, why is he so determined to gloss over these differences by

Reference and Meaning

45

subsuming them under a common term ‘refer’? He does suggest a rationale for this. The truth conditions of sentences are determined by the designations of designative expressions and the extensions of predicates + grammatical structure. Thus both what names, etc., designate and the extensions of predicates have a common functional role in contributing to the truth conditions of sentences. But then why not take other components of sentences—prepositions, adverbs, conjunctions, etc., as having a referential role too, as we will see C. I. Lewis doing? In the next section we will see a considerable fallout from Devitt’s inflation of ‘refer’ in his semantic program. Devitt’s reference (in an ordinary sense of the term!) to Quine encourages us to see Quine (along with Frege) as a prime source of the bad terminological habit I have been documenting. At the beginning of “Notes on the Theory of Reference”, in Quine 1953 he writes: When the cleavage between meaning and reference is properly heeded, the problems of what is loosely called semantics become separated into two provinces so fundamentally distinct as not to deserve a joint appellation at all. They may be called the theory of meaning and the theory of reference . . . The main concepts in the theory of meaning, apart from meaning itself, are synonymy (or sameness of meaning), significance (or possession of meaning), and analyticity (or truth by virtue of meaning.) Another is entailment, or analyticity of the conditional. The main concepts in the theory of reference are naming, truth, denotation (or truth-of ), and extension. Another is the notion of values of variables. (130)

Would that Quine and his epigoni had “properly heeded the cleavage” between reference, extension, and truth-of, and thereby realized them to be so “fundamentally distinct as not to deserve a joint appellation”. As for Frege, we have not only the reference of referring expressions and the extension of predicates lumped under reference, but he goes even further to extend the treatment to indicative sentences. In “Sense and Reference”, in Geach and Black, eds, 1970, he puts forward the following line of argument. The fact that we concern ourselves at all about the reference of a part of the sentence [when the sense of the whole sentence depends only on the senses of its components, not their reference] indicates that we generally recognize and expect a reference for the sentence itself . . . We are therefore justified in not being satisfied with the sense of a sentence, and in inquiring also as to its reference. But now why do we want every proper name to have not only

46

William P. Alston

a sense, but also a reference? Because, and to the extent that, we are concerned with its truth value. [I.e., the truth value of the sentence in which the name occurs] . . . It is the striving for truth that drives us always to advance from the sense to the reference . . . We are therefore driven into accepting the truth value of a sentence as constituting its reference. (63)

Lest you think that this is a slip on Frege’s part, he goes on to compound the offense. “Every declarative sentence concerned with the reference of its words is therefore to be regarded as a proper name, and its reference, if it has one, is either the True or the False.” (63). Apart from the absurdity of taking a declarative sentence to be a proper name, the mere idea that a declarative sentence has a reference, i.e., is fitted to perform the referential function of picking out what a statement is about, is absurd enough. Let’s assume that for this strange term, ‘the True’ we can substitute the abstract noun ‘truth’. And suppose I wish to say of truth that it is the basic aim of cognition. Then, following Frege’s lead I pick some true sentence at random, e.g., ‘Tigers are carnivorous’ and say ‘Tigers are carnivorous is the basic aim of cognition’. I fear this would not be an effective way of getting my point across and would greatly puzzle my audience. Finally, C. I. Lewis in his 1946 goes all the above several degrees better (i.e., worse) by ascribing “denotations” (his term for what is often called ‘extension’) to all meaningful linguistic units. First I will note that Lewis marches along with the others discussed above in covering up the heterogeneity of what he lumps under ‘denotation’ by explaining it disjunctively as what a term “names or applies to”. And he follows Frege with respect to sentential expressions, but with some differences. First, he counts as a term not a declarative sentence itself but a participial phrase that he calls a ‘proposition’ and that signifies the content asserted by the sentence.3 Thus “the statement ‘Mary is making pies’ asserts the state of affairs, Mary making pies now [the proposition], as actual” ( p. 49). And Lewis takes all true propositions to have as their denotation the actual world and all false propositions to have a null denotation. This latter is functionally equivalent to the Fregean view we have just exhibited. But where Lewis outdoes all his colleagues in this section is in applying his “modes of meaning” (of which I am here primarily concerned

3 In Lewis’s “modes of meaning”, what a term signifies is “that property in things the presence of which indicates that the term correctly applies”. (p. 39) This is what is more commonly called ‘intension’.

Reference and Meaning

47

with denotation) to all meaningful linguistic units. (To understand the following quotations we need to understand what Lewis calls ‘signification’, which was explained in fn. 3.) All words in fact have meaning, and name or apply to. All words are fundamentally classifiable as substantives or adjectives or verbs; and verbs are a kind of adjectives. More simply, all words are either substantives or predicables . . . (79–80)

Lewis treats the unfamiliar candidates as predicables. Adverbs are predicables which modify other predicables, or a context containing other predicables, verb-predicables being amongst those which may be so modified adverbially. ‘Quickly’, for example, may modify any ‘x’ such that ‘quickly x’ or ‘x quickly’ has meaning . . . Thus ‘quickly’ applies to instances of walking, boiling, being seen, etc., and signifies that property of them which might be called quickliness—the property common to all things that take place quickly—just as ‘red’ applies to all things that are red. (80–81) Prepositions are relation-words, definable by defining phrases such as ‘in x’, ‘of y’, etc. ‘In’ applies to any case of being in something, or of something being in something else; and signifies the relational property being in, or the property of a pair so related. (81)

And even more startlingly: Articles are adjectives which modify substantives or pronominals. ‘An x’ is equivalent to “instance comprehended by ‘x’”; and ‘a’ or ‘an’ signifies that property which is common to all instances of anything; the property of being a thing nameable by a general name or common noun. (81)

If we are to take all this literally, with ‘apply’ being used in its ordinary sense in which to apply a predicate to X is to predicate the predicate of it, then we would be licensed by Lewis to say such things as “Your walking was quickly”, “My present relation to my house is in”, and “You are a”! And these are flagrantly ungrammatical utterances. ‘Quickly’, ‘in’, and ‘a’ just don’t work like adjectives. Procrustes would be thoroughly satisfied with this 20th century American disciple.4

4 For a more up-to-date and much more elaborate treatment with some family resemblance to C. I. Lewis, see Richard Montague’s “English as a Formal Language” in Montague 1974.

48

William P. Alston

III Let us turn now to the second, more serious devastation, an incautious use of ‘reference’ (in the inflated sense exhibited in the previous section) wreaks in the philosophy of language. This has to do with ways in which “reference” is called on to play an unsuitable role in an account of linguistic meaning. The extreme of this tendency would be banishing of meaning altogether from the semantic vocabulary, an attempt notoriously associated with Quine. It would inflate this essay far beyond seemly bounds if I were to delve into the thickets of Quine’s disavowal of meaning.5 I will concentrate on flights into “reference” that march under the banner of a (more or less) “referential” account of meaning. But just to stick my toe into the hotter water, here are a couple of tastes (to deliberately mix the metaphor). First, a glance at Nelson Goodman’s “On Likeness of Meaning” in Goodman 1972. After raising the question “Under what circumstances do two names or predicates in an ordinary language have the same meaning?” (222) and having dismissed a number of attempts involving “non-extensional” concepts, he proceeds as follows. All these difficulties suggest that we might try the very different and radical theory that two predicates have the same meaning if and only if they apply to exactly the same things—or in other words, have the same extension. (224)

He then recognizes that there are many cases of terms with the same extension but different meanings, most obviously where the extension is empty, as with ‘centaur’ and ‘unicorn’, and then proceeds as follows. Despite the obvious inadequacy of the thesis we have been considering, I think that difference in meaning between any two terms can be fully accounted for without introducing anything beyond terms and their extensions . . . [Though] all unicorns are centaurs and all centaurs are unicorns . . . yet not all pictures of centaurs are pictures of unicorns, nor are all pictures of unicorns pictures of centaurs . . . Now the important point here is this: although two words have the same extension, certain predicates composed by making identical additions to these two words may have different extensions. It is then perhaps the case that for every two words that differ in meaning either their extensions or the extensions of some corresponding compounds of them are different . . . if we

5 For a critique of Quine’s abandonment of meaning see my “Quine on Meaning” in Hahn, ed. 1986.

Reference and Meaning

49

call the extension of a predicate by itself its primary extension, and the extension of any of its compounds a secondary extension, the thesis is formulated as follows. Two terms have the same meaning if and only if they have the same primary and secondary extensions. (225–227)

If you are inclined to cavil at the supposition that this account by Goodman is a case of “going all the way” in abandoning meaning for “reference”, on the grounds that it is advertised as an account of “sameness of meaning”, take account of the fact that if the view is successful it provides a basis for a “disappearance” theory of meaning in which all terms that belong to the “intensional” family in which ‘meaning’ is the center can be eliminated in favor of “extensional” terms like ‘extension’. My second taste involves Hilary Putnam’s characterization of the semantics that he takes to be assumed in “metaphysical realism” in his polemic against such realism in “Realism and Reason”, in Putnam 1978. . . . there is a relation between each term in the language and a piece of THE WORLD (or a kind of piece, if the term is a general term). This relation, the relation of reference, is given by the truth-conditional semantics for the language, in the canonical versions of the theory—i.e., understanding a term, say, T1, consists in knowing what piece of THE WORLD it refers to . . . I shall not assume this account of understanding to be part of the picture in what follows, although it certainly was assumed by metaphysical realists in the past. Minimally, however, there has to be a determinate relation of reference between terms in L and pieces (or sets of pieces) of THE WORLD, on the metaphysical realist model, whether understanding L is taken to consist in ‘knowing’ that relation or not. What makes this picture different from internal realism [the position favored by Putnam] . . . is that . . . THE WORLD is supposed to be independent of any particular representation we have of it . . . (124–125)

We have to remember that this is not a semantics espoused by Putnam but one that he attributes to the metaphysical realist. He uses it to argue that metaphysical realism collapses into unintelligibility. I am not concerned here with that argument.6 Here I am only interested in Putnam’s putting this forward as a semantics that a language might have. Indeed, he says that his “internal realism employs a similar picture within a theory”. So this is an extreme case of supposing that one can give a semantics simply by

6 I criticize the argument in Alston 1996, Ch. 5.

50

William P. Alston

specifying referents (in the inflated sense) for every meaningful term of the language. It is thus an exact duplicate of the procedure of logicians who construct “artificial languages” for the purposes of logic. But it is obviously impossible not only as an account of what it is to understand a term, or the language as a whole, but also as an account of how that understanding could be acquired. As I pointed out above, if anything is obvious in the philosophy of language, it is that we do not come to understand predicates by becoming acquainted with their extensions. We all understand ‘elephant’ while remaining either totally ignorant of its extension, or knowing only a tiny fragment thereof. (And, by the way, it is equally obvious that this kind of semantics is not held by most metaphysical realists in the history of philosophy.) I now move to a more extended example of a reference-oriented semantics, albeit one that does not wind up being as purely referentially dominated as the above sketch. This one is found in Devitt 1981 and 1987. In presenting it I can build on Devitt’s terminology we have encountered in the previous section. In Devitt 1981 he sets out a “semantic program”. As the term suggests, this is not fully developed, but there is enough to enable us to see how the land lies. Our problem in semantics is posed by a part of human behavior, the verbal part . . . early in our theorizing about people . . . we see their sounds and inscriptions as items of language: we see the items as having such semantic properties as being meaningful, of referring to parts of the world, of being true or false . . . I take the main problem of semantics to be to explain the semantic notions that appear in the theory. In virtue of what does this sound refer to that object? What is it for an inscription to be meaningful? Why is that sound true? What semantic notions should appear in our theory of people? In my view the central notion is truth. I can see no plausible account of language, of the sounds and inscriptions we produce and react to, that does not see many of them as having a truth value . . . To explain truth we need notions of reference, which must then also be explained. Do we need any other semantic notions? We detect differences among coreferential words that we might well call differences in meaning. These differences can be accounted for, I claim, by differences in the “mechanisms of reference” which we posit to explain reference. Everything that the philosopher of language finds interesting and important about meaning seems well enough captured by truth and reference and what goes into explaining them. The phenomena we seek to explain do not, in my view, require any notion of meaning beyond this. (68–69).

Reference and Meaning

51

Devitt goes on to add the qualification that it is not, as the above would indicate, that all utterances are assertions. But he does not seem unduly concerned about it. Later he will discuss various ways of handling other kinds of sentences and other kinds of utterance. I myself see this as a major problem, but for present purposes I will ignore it.7 Let’s remember too that Devitt is working with the inflated notion of ‘reference’ that was noted earlier. I will fill out a bit the view set out in the last quotation from Devitt. He uses Fregean arguments for the thesis that there are differences in meaning between coreferential terms and, furthermore, that they affect the meaning, including truth conditions, of sentences. Thus even though ‘the morning star’ and ‘the evening star’ refer to the same object, ‘The morning star is the morning star’ is tautologous, while ‘The morning star is the evening star’ is an empirical discovery. And if that is not enough to give them different truth conditions, how about ‘John believes that the morning star is visible from the earth’ and ‘John believes that the evening star is visible from the earth’. Here all the terms (can) have the same referent/ extension, and yet the truth conditions are different. One could be true and the other false. This leads Devitt to follow Frege in attributing a sense as well as a referent to a term. Frege characterizes the sense of a term as the “mode of presentation” of the object. Devitt takes the “mechanism of reference” associated with a term as its sense; it is the way in which the referent “presents itself ” to the user of the term. Fortunately Devitt abstains from following Frege in taking a declarative sentence to be a proper name, the referent of which is its truth value. Devitt’s general view, only partially worked out in detail, is that the “mechanism of reference” for a term is a causal chain leading from its referent/extension to tokens of the term. For proper names he more or less follows Kripke’s well known view, according to which there is an initial baptism of the referent, followed by a transmission of that use of the name (its use to name the object in question) from speaker to speaker. Somewhat different underlying causal chains are involved for other singular referring expressions. For demonstratives, for example, in the simplest case the chain goes from the object perceived by the speaker, plus contextual features that indicate which object is the one in question, to the use of the term by the speaker to refer to that object. The meaning (Frege’s

7 See Alston 2000 for an approach that takes the variety of speech acts into account from the beginning.

52

William P. Alston

“sense”) of the term (type or token) is identified with the associated causal chain. The treatment of predicates, only sketchily presented, is modeled on the treatment of proper names, with some important differences. Here Devitt more or less follows Kripke and Putnam, especially with respect to terms for perceivable natural kinds. The causal chain in question goes from perceptually presented members of the extension (a paradigm set) to the use of the term to apply to all objects with the same internal structure (nature, essence . . .) as members of the paradigm set. The meaning of the term “would be the relevant sort of internal structure together with the causal network by which it is presented”. (195). This is at odds with more traditional accounts of the meaning of predicates, according to which it consists in a “concept”, a “criterion in mind” that embodies a certain set of properties that is a necessary and sufficient condition for membership in the extension. These properties are typically “surface”, readily perceivable properties on the basis of which we typically identify members of the extension. Furthermore, the meaning in this sense is purely intra-mental and is completely independent (in its constitution if not in its acquisition) of the external environment of the language users. The KripkePutnam-Devitt view differs in both the respects mentioned. The meaning involves the non-directly perceivable internal structure; Putnam and Krikpe are fond of claiming that it is not an analytic truth that gold, for example, is yellow, hard, ductile, etc. Anything that has the same chemical constitution as the paradigm cases of gold is gold, whatever its color, weight, malleability, or other surface properties. And, second, the meaning essentially depends on certain aspects of the external environment of the users of the language; it is not something purely intra-mental. Hence this view is called “externalism” about linguistic meaning. In Putnam’s famous illustration, if on some other planet, lacking any communication with ours, there is a liquid with the same readily perceivable properties as water but with a different chemical constitution, the term the inhabitants of that planet use to refer to that liquid has a different meaning and reference from our term ‘water’. Many qualifications to this brief account are needed. Here are a few. (1) Whereas the causal chain underlying a use of a proper name has a single grounding, it is typical for a predicate to have many. And though, by a convenient over-simplification, the advocates of this view speak of a “standard set of paradigm cases”, in practice these may differ from speaker to speaker. (2) Just as a singular referring expression may lack a referent, so the usual set of paradigms in a language community may be diverse in the

Reference and Meaning

53

nature of its members, in which case the description “those objects that are of the same nature as the paradigms” lacks unique reference. (3) Here is an important feature that this externalist view shares with the more sophisticated versions of the intra-mental view. A speaker’s grasp of the meaning can be more or less implicit. All that can be reasonably required for knowing the meaning of the term is a working knowledge, an ability to recognize new members of the extension and use the term intelligibly in other respects. It is not necessary that the speaker be able to articulate the constitution of the meaning. (4) Putnam is to be commended for putting stress on what he calls “division of linguistic labor”. Using terms with certain meanings is a community endeavor, and not everyone participates in the same way. Putnam is famous for saying that he cannot tell a beech tree from an elm tree (I find this hard to believe), but that he uses these terms meaningfully all the same by virtue of deferring to “experts” on whom he can call when needed. Obviously, not all predicates are perceivable natural kind predicates, and others need different treatment. For “theoretical natural kind terms”, Devitt suggests a substitute for direct perception of paradigm cases, viz., “a relation consisting of an instrument ‘perceiving’ the referent and our ‘reading’ of the instrument; we are ‘perceiving the reference through the instrument’.” Predicates other than natural kind predicates are construed as deriving their meaning from definition or explanation in terms of other predicates, and the latter may have their meaning derived from other predicates. . . . . . , the chain so engendered ending in predicates the meaning of which consists in an underlying causal chain stemming from members of the extension. Though most of this program for semantics is, indeed, only programmatic, it contains enough substance to be critically evaluated. It is an impressive attempt to base an account of meaning on reference/extensions without flatly identifying meaning and reference/extension. It avoids the excesses of Frege and C. I. Lewis in ascribing reference or extension where they do not belong. But I do not, in the end, find it satisfactory. If I were doing a full dress criticism, I would take aim at an assumption that Devitt shares with a large majority of contemporary Anglo-American philosophers of language—that truth (or truth conditions) of sentences is a basic semantic notion. I have argued in Alston 1996, Ch. 1, that truth value is, rather, a post-semantic notion, that it is only after a sentence has been semantically interpreted that questions of truth and falsity arise. In fact, I contend, truth values apply primarily to propositions, statements, and beliefs, and only derivatively, if at all, to sentences. But since that criticism is

54

William P. Alston

to one side of my concerns in this essay, I will simply refer the reader to Ch. 1 of the book just cited. My criticism here, restricted to his treatment of natural kind terms, is, rather, that Devitt has gone too far in the externalist direction, has reacted too strongly and undiscriminatingly against the traditional internalism. This results from his neglecting a crucial point about meaning. Knowing the meaning of a term is a matter of having something we can carry around with us and use in various ways—recognizing items the term applies to, raising and answering relevant questions about its object(s), and deploying the term intelligibly in the many contexts in which it enters speech. Hence we are debarred from going externalist so thoroughly as to make the meaning useless for playing these roles. How can we do justice to this desideratum while at the same time recognizing the Kripke-Putnam point about the role of environmental paradigms of the extension? By taking the meaning to include a mental reference to the external paradigms of the extension rather than the paradigms themselves. Thus the meaning is a concept in the mind, but that concept embodies references to items in the external environment, as well as general properties. So the meaning of ‘tiger’ can be formulated as what has the same nature (essence, internal structure . . .) as a, b, c, d, e . . . (the set, or a set, of socially recognized paradigms of tigers). I submit that by putting a (mental) reference, which may be implicit, into the meaning of the term, we have done justice to the Kripke-Putnam insight that what is actually in the environment matters for the meaning and also done justice to the equally crucial point that a meaning is something the language user can have readily available to work with in using the term. So I have toned down Devitt’s externalism by putting part of what he identifies with meaning in the mind. But I haven’t included the chain of transmission from the initial baptism or perception of paradigms, even by way of a mental reference thereto, although that chain is part of the meaning, according to Devitt. That is because it seems to me that such grasp as the typical speaker has of the chain of transmission is, if it exists at all, too implicit to make any contribution to the usability of the meaning to direct and monitor the use of terms in speech. Let us not forget that Kripke, in rejecting a descriptivist account of proper names, rejects the suggestion that there could be a descriptivist meaning that would be in terms of the chain of transmission, and rejects it on the ground that we may have completely forgotten from whom we derived our use of the name. And, he may well have added, we may never have known anything about the earlier stages of the transmission, except for the identity of the referent. A parallel

Reference and Meaning

55

point holds for predicate meaning. Hence I take Devitt to have been ill advised in including the chain of transmission in the meaning.

IV It will be useful to contrast Devitt’s program with a somewhat similar one that gets it right in ways that Devitt does not, viz., the one in Lewis 1972. The details are complicated, and I will only seek to give the general idea. Though Lewis says: “My proposals are in the tradition of referential . . . semantics descended from Frege, Tarski, Carnap (in his later works) and recent work of Kripke and others on semantic foundations of intensional logic . . .”(170), and we can see how this is not incorrect, still his theory gives an appropriate place to a robust concept of meaning in a way that Devitt’s does not, and it does not seek to replace meaning with reference and/or extension in the style of Goodman. The “referential” character comes from the key role that ‘extension’ plays in the theory. “Intension”, which with some alternative refinements plays the role of meaning, at least for what Lewis calls “basic categories” of linguistic expressions, is a function from various relevant factors to an extension. (We get the same kind of function for some derived categories (see below) as well, such as noun phrases.) On one alternative systematization there are three basic categories, sentence, name, and common noun. Lewis, unfortunately, follows Frege in taking the extension of a sentence to be a truth value. (He takes all sentences to be equivalent to assertion making indicative sentences in a way I will leave aside here.)8 The “extension” of a name is what it names. The extension of a common noun is the set of things it applies to. Thus Lewis is subject to the strictures leveled at Frege, Putnam, and Devitt in section ii for conflating very different word-world relationships under a common term, while explaining the term differently for different sorts of expressions. But perhaps Lewis is less guilty than his partners in conflation in view of the fact that he presents his semantics as applying to “possible languages . . . as abstract semantic systems whereby symbols are associated with aspects of the world”, leaving as a further task “the description of the psychological and sociological facts whereby a particular one of these abstract semantic systems is the one used by a person or population”. (170) This may justify him in using a term like

8 For criticism of this see Alston 2000, Ch. 9.

56

William P. Alston

‘extension’ as a technical term for this limited purpose. In any event, I am not concerned here with the issues discussed in section ii, but only with showing ways in which Lewis’s “General Semantics” steers clear of the shoals that wreck Devitt’s account of meaning. So the intension of a term in a basic category is a function from various relevant factors, which together he terms an index, to an extension. What factors are relevant is different for expressions tokens of which differ in extension from expressions of which that is not true. For a name, and most other singular referring expressions, the relevant factors are features of the context of utterance for the token—the speaker, time, place, audience, indicated object, previous discourse, etc. For common nouns and many noun phrases and verb phrases, the relevant factors are possible worlds, for they have different extensions in different possible worlds. For sentences with both the above kinds of components both contextual and possible world factors are included in the index; for sentences with only the second kind of component only possible world factors are relevant. Derived categories are generated by a concatenation of simple and derived components. Thus an adjective and a common noun make a noun phrase, a verb and an adverb make a (simple) verb phrase, and by adding a noun phrase as object we get a more complicated verb phrase. Instead of treating the intensions of all derived categories as functions from indices to extensions, Lewis treats many of them as functions from indices to further functions, the chain eventually ending in a function from an index to an extension. Thus since it is inappropriate to think of ‘alleged’ as having an extension that consists of things it applies to (alleged “things” are too heterogeneous for this purpose), we will think of the intension of ‘alleged’, in each use, as a function from an index to the function of a complex expression in which it figures, e.g., ‘alleged spy’. That noun phrase, in turn, has an intension that may be construed as a function from an index to an extension, the class of alleged spies, i.e., of people who are alleged to be spies. Or, alternatively the intension of ‘alleged spy’ can, in turn, be construed as a function from an index to the intension of some more complex expression, e.g., ‘falsely alleged spy’. Any such chain of intensions as functions from indices to intensions, which are functions from indices to intensions. . . . . . will end in a function from an index to an extension when the process gives us a sentence intension. In this way Lewis is able to treat all meanings of sub-sentential units as consisting in contributions they make, ultimately, to the meanings of sentences, without following his namesake, C. I. Lewis, in the extravagance, or rather absurdity, of assigning extensions to all meaningful linguistic expressions.

Reference and Meaning

57

Lewis points out that for sentences a function from possible worlds to truth values is too coarse grained to qualify as meanings. (182) For since all necessarily true sentences are true in all possible worlds, they all have the same intension, and so do all inconsistent sentences, which are false in all possible worlds. What we need here is the equivalent in his system for C. I. Lewis’s “holophrastic meaning”, viz., a “semantically interpreted phrase marker” minus the final sentential node, which exhibits the way in which the intensions of the various meaningful components contribute to the final sentential function. This will give us something that differs for different necessarily true or necessarily false sentences, something that conforms to our intuitive judgments of sameness and difference of meaning. Is David Lewis superior to Devitt in making his basic “mechanism of reference” something that a user of the language can “carry around with him” to be readily available (cognitively) when needed? Are the Lewisian functions something that are internalized by users of the language in a way that makes them usable? In answering this we must keep in mind the point that it is only practical, know-how cognitive internalization that is in question. It is out of the question to think of ordinary language users as able to explicitly articulate the Lewisian account of the intension or semantically interpreted phrase markers of each meaningful linguistic unit they employ. It is enough if their internalization of the relevant functions gives them the capacity to recognize the components of the appropriate index when there is need to do so and to spot what it would take for that expression to yield as value one function or one extension rather than another, when a question arises as to that. (Again, they need not be doing the recognition or spotting in precisely those terms.) And it does seem that Lewis has the edge here. Whereas Devitt’s causal chains stretching from the foundational baptisms, etc. are clearly not the sort of thing speakers can “carry around with them mentally”, there is, at least, a good chance that Lewisian functions are that sort of thing, provided we restrict the requirement sufficiently to a practical, know-how kind of availability.

V I hope and trust that the above discussion gives some idea of the unfortunate consequences of prevalent ways of using ‘reference’ and its cognates in recent English speaking philosophy. These consequences could be documented much more thoroughly, but sufficient unto the day . . . My aims in this essay have been purely negative, confined to exposing misleading

58

William P. Alston

confusions and conflations. No doubt I would have contributed more to philosophy by laying out and defending adequate accounts of linguistic meaning and of the various word-world relationships lumped under the heading of “reference”. But such was not my intention here, and, in any event, carrying out even part of such an enterprise would have burst the bonds of a single essay. And although cleaning up philosophical messes is a relatively humble task, compared with erecting a magnificent philosophical edifice, it still has a honorable place in the philosophical enterprise. I am reminded in this connection of a story told in my youth about the first-half-of-the-twentieth-century American philosopher, Morris Raphael Cohen. In one of his classes he had subjected some arguments of F. H. Bradley to devastating criticism. After the class one of the students came up to him and asked him what positive alternative he had to offer to Bradley’s absolute idealism. To which Cohen responded, “When Hercules had finished cleaning the Augean stables, was any further task required of him?”.

References Alston, William P. 1986. “Quine on Meaning”. In The Philosophy of W. V. O. Quine, ed. Lewis E. Hahn. La Salle, IL: Open Court. Alston, William P. 1996. A Realist Conception of Truth. Ithaca, NY: Cornell University Press. Alston, William P. 2000. Illocutionary Acts and Sentence Meaning. Ithaca, NY: Cornell University Press. Davidson, Donald & Gilbert Harman, eds. 1972. Semantics of Natural Language. Dordrecht: D. Reidel. Devitt, Michael. 1981. Designation. New York: Columbia University Press. Devitt, Michael & Kim Sterelny. 1987. Language and Reality. Cambridge, MA: MIT Press. Donnellan, Keith. 1966. “Reference and Definite Descriptions”. Philosophical Review, 75. Frege, Gottlob. 1970. “On Sense and Reference”, tr. Max Black. In Philosophical Writings of Gottlob Frege, ed. Peter Geach and Max Black. Oxford: Basil Blackwell. Goodman, Nelson. 1972. Problems and Projects. Indianapolis, IN: Bobbs-Merril. Kripke, Saul A. 1972. “Naming and Necessity”, in Davidson and Harman 1972. Kripke, Saul A. 1979. “Speaker’s Reference and Semantic Reference”. In Contemporary Perspectives in the Philosophy of Language, ed., Peter A. French, Theodore E. Uehling, Jr., and Howard K. Wettstein. Minneapolis, MN: University of Minnesota Press. Lewis, C. I. 1946. An Analysis of Knowledge and Valuation. La Salle, IL: Open Court. Lewis, David. 1972. “General Semantics”. In Davidson & Harman 1972.

Reference and Meaning

59

Montague, Richard. 1974. Formal Philosophy. New Haven, CT: Yale University Press. Putnam, Hilary. 1975. “The Meaning of ‘Meaning’”. In Language, Mind, and Knowledge, ed. Keith Gunderson. Minneapolis, MN: University of Minnesota Press. Putnam, Hilary. 1978. Meaning and the Moral Sciences. London: Routledge & Kegan Paul. Putnam, Hilary. 1981. Reason, Truth, and History. Cambridge: Cambridge University Press. Quine, Willard Van Orman. 1953. From a Logical Point of View. Cambridge, MA: Harvard University Press.

Still Against Direct Reference Michael Devitt Abstract: According to direct reference (“DR”), the meaning or “semantic content” of a name is simply its referent. I have argued against this in “Against Direct Reference” and many times elsewhere. In this paper I do so again from the methodological perspective of Coming to Our Senses. This leads to the view that a name’s causal mode of reference is (at least one of ) its meanings. This view, ignored by Soames, is the core of my argument against DR. I look critically at Braun’s explanationist defense, and Soames’ pragmatic defense, of DR. Finally, I argue, again, that DR faces an insurmountable difficulty: the problem of identity statements. The magnitude of this problem is demonstrated by Soames’ own discussion.

According to the doctrine of direct reference “the utterance of a simple sentence containing names or demonstratives normally expresses a “singular proposition” – a proposition that contains as constituents the individuals referred to, and not any descriptions of or conditions on them” (Crimmins and Perry 1989: 686). I have criticized this doctrine many times, including in a paper called “Against Direct Reference” (1989). In brief, I have argued that the doctrine is theoretically unmotivated and faces an insurmountable difficulty, the well-known problem of identity statements. My criticisms have had sadly little effect. But I am still against direct reference and so I shall try again. I shall consider some of the latest from directreference theorists, particularly from Scott Soames. My case against direct reference is in section 3. It rests on considerations I have presented in arguing for a naturalistic truth-referential theory in Coming to Our Senses (1996).1 I shall summarize that argument in sections 1 and 2.

1. The methodology of naturalistic semantics Three important questions get insufficient attention in semantics. What are the semantic tasks? Why are they worthwhile? How should we 1 See also the following responses to critics: 1997a,b,c. I defend naturalism at greater length elsewhere (2010a: 253-91).

62

Michael Devitt

accomplish them? The central purpose of Coming is to answer these “methodological” questions and to see what semantic program follows from the answers. I summarize. It is troubling that much semantic theorizing proceeds with inexplicit reliance on apparently ad hoc views of the semantic tasks. Thus it is common to take for granted that semantics is concerned with truth and reference. I think that this view is right, but why is it right? What can we say to someone like Paul Horwich (1998a,b, 2005) who disagrees, claiming that semantics should be concerned with use?2 Furthermore, it is troubling that, in attempting to accomplish the semantic task, we all go in for “intuition mongering”. Broadly, it is troubling that we seem to lack a scientifically appealing method for settling the disputes that bedevil semantics. In Coming (ch. 2, and in Devitt 1994) I propose a view of the semantic tasks by looking at the purposes we attempt to serve in ascribing meanings. And I propose a way of accomplishing the tasks. This methodology has a place for intuitions, but it is the same limited place that they have elsewhere in science.

1.1 Semantic tasks So, what are the semantic tasks? What should semantics be trying to do? There seems to be a simple answer: The “basic” semantic task is to say what meanings are, to explain their natures. It is thus analogous to such tasks as saying what genes, atoms, acids, echidnas, or pains are but not, we should note, to such tasks as saying what genes and so forth do, stating the laws that advert to them. However, we start the semantic task in rather worse shape than we do its analogues. With them, the subject matter of investigation is already identified relatively uncontroversially. This reflects the fact that we have clear and familiar theoretical or practical purposes for which we identify the subject matter. Semantics does not start out like that. It is far from clear what counts as a meaning that needs explaining. Indeed, the intractable nature of semantic disputes largely stems from differing opinions about what counts.3 (And we should note that the problem is not

2 Devitt 2002 and 2010b are in fact responses to Horwich. 3 “The chief problem about semantics comes at the beginning. What is the theory of meaning a theory of ?” (Higginbotham 1991: 271). “Meaning is notoriously vague” (Block 1986: 615). Lycan has brought out the problem wittily with his “Double Indexical Theory of Meaning”: “MEANING = def Whatever aspect of linguistic activity happens to interest me now” (Lycan 1984: 272).

Still Against Direct Reference

63

simply with the expression ‘meaning’. Those who prefer the expressions like ‘semantic content’, ‘semantic value’, or ‘propositional content’ for what has to be explained have an analogous problem: What counts as a content or value?) We start semantics in the unusual position of having to specify a subject matter. We should not insist on great precision about this in advance of theory, but we do need some explication of our vague talk of “meanings” (“semantic content”, “semantic value”, etc.).4 Furthermore, we must specify a subject matter worthy of investigation; we need an explication that is not ad hoc. Finally, the semantic task of interest to philosophers should be not only worthwhile but fundamental. I seek a solution to this problem by focusing on the purposes for which we ascribe meanings using ‘that’ clauses (“t-clauses”) in attitude ascriptions: in particular, the purposes of explaining intentional behavior and of using thoughts and utterances as guides to reality. I call these purposes “semantic”. I say further that a property plays a “semantic” role if and only if it is a property of the sort specified by t-clauses, and, if it were the case that a token thought had the property, it would be in virtue of this fact that the token can explain the behavior of the thinker or be used as a guide to reality. We are then in the position to add the following explication to the statement of the basic task: A property is a meaning if and only if it plays a semantic role in that sense. And the basic task is to explain the nature of meanings in that sense (secs. 2.3–2.6). Consider another task, that of explaining the nature of the properties that we ought to ascribe for semantic purposes. It is easy to see that this “nor mative” task is closely related to the basic one. Indeed, my first methodological proposal is that we should tackle the basic task by tackling the normative one. There is an obvious contrast between the normative task and what I call “the descriptive task”. This is the task of explaining the natures of the properties we do ascribe in attitude ascriptions for semantic purposes; it is the task of explaining the semantic status quo. It is what most people working in semantics—philosophers, linguists, and psychologists—are in effect doing. Yet it is very different from the normative and basic tasks. Note particularly that the “putative meanings” it investigates really are meanings only if ascribing them really does serve our semantic purposes, only if they

4 Hereinafter I shall mostly omit such parenthetical additions but they should always be taken as read.

64

Michael Devitt

really play semantic roles. Once the descriptive task is sharply distinguished from the others, the question of its bearing on them arises. We shall consider this question in a moment. Semantic tasks are often defined using terms like ‘meaning’ with either no attempt to explicate this talk or with an explication using terms like ‘proposition’ and ‘information’ that are equally unclear. Given the prima facie unclarity and vagueness of all this talk, this practice is surely unacceptable: It leaves us both with no firm idea of what the issue is and with the likelihood of being at cross purposes. Once an explication is offered, there is little point in a verbal dispute about the appropriateness of using ‘meaning’ and hence ‘semantics’ in the way proposed. However, there is a lot of point in asking why a task defined in terms of ‘meaning’ thus explicated is worthwhile. We are, of course, free to study anything. But if semantics is a genuine science, as the naturalist thinks it must be, we should be able to say why it is interesting. A definition that does not say this will be ad hoc. My discussion of the semantic tasks is an attempt to meet these demands for explication and worth. What about other views of the tasks? Coming takes a brief and largely critical look at these (2.7). But my point is not that the properties that I have called “meanings” are the only real meanings nor that the tasks I have called “semantic” are the only proper tasks for semantics. I doubt that there is any interesting matter of fact about such claims. I do claim that those properties are worth investigating and that those tasks are worth performing. I claim further that those tasks are appropriately fundamental. Perhaps all this is true of other tasks, but that always needs to be demonstrated.5 I have four comments on this discussion, some that go a little beyond what is in Coming.6 (a) The focus of this discussion is on the meanings of thoughts rather than of utterances. I argue that it is because of our interest in the meanings of thoughts that we are interested in the meanings of utterances. It is by ascribing meanings to thoughts that we directly explain behavior and are guided to reality. But a person’s utterances are our main route to her thoughts. If we take a speaker’s utterances as sincere expressions of her thoughts then we

5 So I am much more open to other views of the semantics tasks than David Braun allows in his review (2000). My criticism of direct-reference theorists like him is that they do not attempt to motivate any view of the tasks that would support direct reference; see section 3 below. 6 I draw on discussions in my Ignorance of Language (2006: chs. 8–10).

Still Against Direct Reference

65

will ascribe the same meanings to her thoughts as to her utterances. So ascribing meanings to utterances is a way to ascribe meanings to thoughts and hence indirectly to explain behavior and be guided to reality. (b) Paul Grice (1989) has drawn our attention to the distinction between the conventional meaning of an utterance and its speaker meaning. A speaker may mean something by the sentence he utters that is different from what that sentence conventionally means on the occasion of the utterance. In doing semantics, we are primarily interested in conventional meanings on occasions of utterance because those are the meanings that are the main routes to speaker meanings and hence thought meanings. It is because sentence tokens have those meanings, largely by convention but partly as a result of the “pragmatic” fixing of indexical reference and removal of ambiguities, that uttering the sentences is so worthwhile.7 For, on hearing an utterance, a person who participates in the conventions for the type of sentence it involves can, with the help of “pragmatic” abilities at detecting indexical references and removing ambiguities, immediately grasp the thought that the utterance expresses. So the basic task for utterances is to explain the nature of properties that utterances have largely by convention and that play semantic roles. (c) This task concerns the meanings of tokens and the earlier explication should be understood accordingly: a property of a (thought or expression) token is a meaning if and only if it plays a semantic role. Yet many focus their semantics on the meanings of types. An interest in types is already implicit. Although our theoretical interest in meanings starts with tokens, that immediately leads to an interest in types. For, as we have just noted, an utterance token has a meaning we are interested in partly in virtue of a conventional meaning of tokens of that type in the speaker’s language. So our basic task for tokens leads to an interest in those conventional properties of types which, along with pragmatic features on an occasion of utterance, determine the properties of tokens that play semantic roles. And the basic task for expression types is to explain the nature of those conventional properties. (d) The discussion opens up the possibility that a token has more than one meaning. For, it opens up the possibility that a token might have a variety of properties playing semantic roles: One meaning might explain one bit of behavior; another, another; and a third might serve as a guide to

7 It should be noted that if Chomsky is right, as I think he is, some of the syntactic properties of a sentence token are not conventional but innate.

66

Michael Devitt

external reality. If so, there would be no theoretically interesting reason to insist that one rather than another is the meaning (2.8). Coming argues that this is in fact the case. My first methodological proposal was that we should tackle the basic task by tackling the normative one. Turn now to the methodology of the latter task. 1.2 Methodology for the normative task What should be the basis for our choosing to ascribe one rather than another of the many properties of a token that are candidates to be meanings? The answer is implicit in the task. We should ascribe the ones that play semantic roles and hence that it serves our semantic purposes to ascribe: the ones that play the roles outlined in explaining behavior and informing us about reality. It is clearly not easy to tell which properties to ascribe nor to say how to tell. Coming makes a modest proposal about the bearing of the descriptive semantic task on this normative task. Our ordinary attitude ascriptions attribute certain properties for semantic purposes. Given the apparent success of the ascriptions it is likely that these putative meanings are real meanings. So I make my second methodological proposal: We should look to the descriptive task for evidence for the normative/basic one. Further, I suggest, although I do not argue, that the main, if not the only, justification for the usual focus on the descriptive task is the bearing of that task on the normative one (2.9). 1.3 Methodology for the descriptive task Turn now to the methodology of the descriptive task. Coming approaches this by considering descriptive tasks in general. How do we tell what is the nature of some property, being an F, that we ascribe? Sometimes we already have a well-established theory; for example, thanks to molecular genetics, we had one for being a gene. But suppose that we do not have a theory and are starting pretty much from scratch. How then do we tell what is common and peculiar to F’s “in all possible worlds”? What is the “ultimate” method? The answer breaks into two stages. First, we must identify some apparently uncontroversial examples of F ’s and non-F ’s. Second, we must examine the examples to determine the nature of being an F. The second stage is a straightforwardly scientific one. The preliminary first stage may not be.

Still Against Direct Reference

67

It involves using “identification experts” who may be scientists but may be just plain folk (2.10). Coming goes on to apply this general discussion to the descriptive task in semantics. In semantics we surely start from scratch: Any theories we have are as much in contention as anything is. So, I make my third methodological proposal: We should use the “ultimate” method on putative meanings to accomplish the descriptive task. To apply the “ultimate” method to a putative meaning, we must first identify some apparently uncontroversial examples of tokens with that property and of tokens without it; and second we must examine the examples to determine the nature of the property. Who are the experts to be consulted in the first-stage identifications? The folk, in their frequent use of t-clauses in attitude ascriptions, are as expert as anyone. And the folk are in an advantageous position to become experts because they themselves are competent to produce token thoughts and utterances to which the properties are ascribed. Finally, theorists can count themselves among the expert folk. So, in the first stage, semantic theorists have two advantages: no need to consult others and ready access to data. Intuitions and thought experiments of the sort that dominate semantics are important in the first stage. However, they are empirical responses to the phenomena and are open to revision at the second stage (2.11).8 The next step is to use this methodology to argue for a truth-referential view of meanings.

2. A truth-referential theory Coming argues for the view that the meanings of a token are entirely constituted by “representational properties”. These properties include any property that plays a role in determining what the token represents. So the meaning of a sentence token is exhausted by the properties that determine its truth condition.9 And the meaning of a word token is exhausted by properties that determine its reference.10 8 For more on intuitions and their role, see Devitt 2010a: 292–302. 9 Not quite. A sentence token represents some situation that would make it true, complied with, or whatever, but it also asserts that the situation obtains, requests that the situation be brought about, or whatever; a sentence has “a force”. I shall overlook these forces. 10 Any truth-referential theory of meaning gives an explanatory role to truth and reference and so is at odds with deflationist views. For a discussion, see Devitt 2010a: 155–81.

68

Michael Devitt

2.1 The descriptive task The second methodological proposal instructs us to look to the descriptive task for evidence about the nature of meanings. The third proposal instructs us to use the “ultimate” method on the descriptive task. Applying this “ultimate” method, and drawing on the classic discussion of transparent and opaque attitude ascriptions generated by Quine,11 we find support for the descriptive version of our representationalism. The focus is on definite singular term tokens and their mental correlates. I shall talk of these mental correlates as if the controversial “language-of-thought hypothesis” were true. Thus, I shall call the part of a thought that is expressed by a proper name “a proper name”. I do think that the language-of-thought hypothesis is true (4.4), but even if it is not I can see no harm in this convenient way of talking. As Quine pointed out, the most common attitude ascriptions are opaque. In these, substituting a co-referential term for a singular term in a t-clause cannot be guaranteed to save truth; the rule of “substitutivity of identity” does not hold. (The proponents of direct reference disagree; see sec. 3.) Thus, consider Quine’s famous example. Ralph observes a man in a brown hat behaving suspiciously and as a result it is true to say, (1) Ralph believes that the man in the brown hat is a spy. But, as typically understood, it is not true to say, (2) Ralph believes that Ortcutt is a spy, even though Ortcutt is that very man in the hat. For, Ralph does not know this identity. In light of this, what property do singular terms in the t-clauses of opaque ascriptions ascribe to thoughts? I argue that they ascribe the property of ( purportedly) referring to a specified object under a specified mode. All token beliefs that would make (1) true have in common, so far as their singular terms are concerned, only the property of ( purportedly) referring to Ortcutt under the mode exemplified by ‘the man in the brown hat’ in (1); similarly, (2), and the mode exemplified by ‘Ortcutt’ in (2). (1) is true because Ralph believes spyhood of Ortcutt 11 See particularly, Quine 1953: 139–59; 1960: 141–51, 166–9; 1966: 183–94; Kaplan 1968.

Still Against Direct Reference

69

under the former mode; (2) is false because he does not believe it of him under the latter mode. Since these modes are ascribed to explain behavior and guide us to reality they are putative meanings at least (4.2). 2.2 The normative/basic task But are they really meanings? If my argument is right then we have learnt what folk ascribe to explain behavior and guide us to reality and hence what they are, in effect, treating as a meaning. It describes the semantic status quo. We have solved the descriptive task. But we need to solve the normative/basic task. Perhaps the folk are wrong. Perhaps the property of a mental word that really does play a role in explaining behavior, and hence is its meaning, involves a stereotype or a non-reference-determining functional role. Or perhaps the eliminativist is right and there are no thoughts: something else altogether explains behavior. Of course, if the status quo is as we have described it, such positions are all revisionist. But that does not show that they are wrong. Showing this is, of course, a lengthy business involving criticism of revisionism and its arguments. I attempt to tackle this in Coming (ch. 5). Now I just want to emphasize one simple yet powerful argument for the status quo: briefly, it works. Day in and day out the folk use ordinary thought ascriptions to explain behavior and guide us to reality. For example, they say “Oscar believed that Mary was thirsty” to explain his giving water to Mary and inform us of Mary’s thirst. And the folk are not alone in this habit: social scientists do it all the time too. Furthermore, these ascriptions appear to be, by and large, successful; the ascription to Oscar really does seem to explain his behavior and inform us of Mary’s thirst. This is evidence that thoughts really do have whatever properties the folk and social scientists ascribe to them, and that those properties really do serve semantic purposes and so are meanings. And, I have argued, those properties are modes of reference. Given the explanatory success of this status quo, overthrowing it needs both a powerful argument and a plausible alternative semantics (4.3).12 12 So Braun is wrong to say that I assume that t-clauses “ascribe only semantic properties to tokens” (2000: 491). The properties they ascribe are semantic at all, on my explication, only if they really do explain behavior or guide us to reality. I argue that these properties do have these “semantic” roles. A central point of my methodology is to avoid tendentious assumptions about semantics.

70

Michael Devitt

2.3 Other meanings Now, in this paper, I shall be focusing on these common opaquely ascribed meanings. But I think that there is a very good case that there are others. First, using an example due to Stephen Schiffer (1979: 67), I argue sentences like (1) and (2) also have a “simply-transparent” reading according to which they ascribe the property of simply referring to a specified object. And the literature makes it plausible that sentences like (3) Ortcutt is such that Ralph believes him to be a spy are “rapport-transparent” ascribing the property of referring to an object that the speaker has particularly in mind (4.2). Furthermore, I argue that these other ascriptions also serve our semantic purposes, particularly the purpose of guiding us to reality. So, these putative meanings are also really meanings (4.3). And that is not the end of it. Attention to various ingenious “puzzles” (Richard 1983; Perry 1993; Castaneda 1966, 1967; Kripke 1979) complicate the story. The explanation of behavior sometimes demands that a demonstrative has a “finer-grained” meaning than may be ascribed by a normal opaque ascription (4.14). First-person pronouns have a similar finer-grained meaning ascribed by a special form of attitude ascription (4.15). The meaning ascribed by a proper name in a t-clause is normally “coarser-grained” than my discussion of (2) suggests: thus ‘London’ in a t-clause normally ascribes a “disjunctive” mode that includes the mode of ‘Londres’ and all other translations of ‘London’. Yet, the explanation of behavior sometimes demands that a name has a finer-grained meaning than may be ascribed by a normal opaque ascription (4.17). If this is all right, the earlier-mentioned possibility that a token belief or utterance has more than one meaning property turns out to be actual.

2.4 Explaining meanings We need to explain the nature of these various meanings, of course. What we most need for that are theories of reference. Coming argues that three sorts of theory of reference are possible (4.5). According to “description” theories, the reference of a word is fixed by certain of the descriptions that speakers associate with the word; it refers to whatever those descriptions, or a weighted most of them, apply to. The received view for decades was that proper names are to be explained by a description theory. But then

Still Against Direct Reference

71

came the revolution. Such theories were seen to have serious problems, particularly the problem of “ignorance and error”: speakers who seem perfectly competent with a word are too ignorant to provide the appropriate descriptions of its referent; worse, speakers are often so wrong about the referent that the descriptions they would provide apply not to the referent but to other entities or to nothing at all (Kripke 1980; Donnellan 1972). Aside from that, description theories in general are essentially incomplete, explaining the reference of some words in terms of the reference of others, a reference which then needs to be explained. If we are to be naturalistic, some singular terms at least must be covered by “causal” or “descriptive-causal” theories, explaining reference fully or partly in terms of direct noninferential relations to reality. These theories must appeal to one or more of three types of causal relations: historical, reliablist, or teleological. Although much progress has been made with theories of these ultimate referential links, I don’t think that we yet have any completely satisfactory theory. Theories of reference of these three sorts will explain the referential meanings we have posited. Consider the meanings that are our focus, opaquely ascribed properties of referring to an object in a certain way. Some of these—for example, that for ‘the man in the brown hat’—might be descriptive modes of reference but it simply follows from the above discussion that some must be nondescriptive causal modes of reference.13 Coming illustrates the latter idea with “IT”, a historical-causal theory of names and other singular terms (4.6).14 On this view, the opaquely ascribed meaning of a name is a property of referring by a certain sort of causal chain. For example, the mode for ‘Mark Twain’ is the property of referring by means of causal chains grounded in Mark Twain and involving the sounds, inscriptions, and so on, that constitute the history of the name’s use to designate Mark Twain; and the mode for ‘Samuel Clemens’ is similar but involves the sounds, inscriptions, and so on, of this different name. If another sort of causal theory is right for a word, its mode will be its property of referring by some other sort of causal relation to external

13 Note that the view is not that such a meaning is a property of being associated with a description of a certain causal way of referring to the object: that would be what is known as “causal descriptivism”, a description theory that is parasitic on a causal theory; see Devitt and Sterelny 1999, p. 61, for a discussion. The view is that the meaning is the property of having a certain causal way of referring to the object. So the name is rigid ( pp. 79–81); c.f. Braun 2000. 14 I proposed this theory in earlier works: 1974, 1981a, 1981b, 1989.

72

Michael Devitt

reality. The commitment to meanings as causal modes does not rest on IT or any particular theory of these modes.

3. Direct reference 3.1 Ignoring causal modes Having outlined its program, Coming turns to the rejection of its rivals. The first of these is the direct-reference view of proper names, “DR” (4.8, 4.18).15 DR has its roots in the just-mentioned refutation of description theories for proper names. DR responds to this refutation with the surprising claim that the meaning—DR’s preferred term is “semantic content”—of a name is simply its referent, and the claim that assertions containing the name semantically express only “singular Russellian propositions”, propositions containing that very referent.16 Haven’t we known for more than a century that such a Millian view faces overwhelming problems with identity statements, empty names, and the role of names in the content sentences of attitude ascriptions? Proponents of DR think not. They acknowledge the problems, of course, but seek to explain them away because they find DR so prima facie plausible. Thus, in his muchdiscussed book, Beyond Rigidity (2002), Scott Soames claims that “if names don’t have descriptive semantic contents, then it would seem that their only semantic contents are their referents” (2002: 5); “the semantic theses [of DR] I have adopted are highly motivated; it is not easy to see how they could be wrong” ( p. 142). What is the source of this complacency? The answer is clear. In support of the DR view that ‘Carl Hempel lived on Lake Lane in Princeton’ semantically expresses the proposition made up 15 I am indebted to David Braun for a very helpful exchange about DR, lasting over many years and including lengthy comments on a draft of the present paper. He has tried to answer my criticisms of DR with great patience but not, it seems to me, much success. He has also been an invaluable guide to the literature. 16 Given my scruples about treating meanings as objects, particularly as propositions, and preference for treating them as properties, I would express the DR view as that a name’s meaning is simply its property of referring to its referent. But this preference is beside the point of my disagreement with DR and so I shall (mostly) go along with DR’s expression. DR applies also to indexicals and perhaps some other terms but I shall discuss only proper names. It should be noted that “direct reference” is used quite loosely and does not always refer to what I am here calling “DR”. I have elsewhere summarized the various “direct reference” theories and their histories (1989: secs. 1–2).

Still Against Direct Reference

73

of Carl Hempel and the property of living on Lake Lane in Princeton, Soames has this to say: “Since there seems to be no better candidate, it is reasonable to identify the proposition semantically expressed” with this singular Russellian proposition ( p. 66). In brief, he thinks that, once description theories are abandoned, DR is “the only theory in town”. This is curious. DR is obviously not the only theory in town. I have briefly described another in the last section. It is the theory that the meaning of a name is its property of ( purportedly) referring to a specified object in a certain causal way; the meaning is a causal mode of reference. Neither this theory nor any of the many works in which I have argued for the theory and against DR (inter alia, 1974, 1980, 1981a, 1989, 1996, 2001) is even mentioned by Soames. This is disappointing, even a bit depressing.17 What are we to make of it? Soames cannot be unaware of the theory; indeed he gave me helpful comments on a draft of “Against Direct Reference” (1989). The charitable thing to suppose is that he finds the view too preposterous to be worthy of discussion. This is pretty much the view of Nathan Salmon, another “hard man” of DR. Citing my Designation (1981a), Salmon calls the view “ill conceived if not downright desperate . . . wildly bizarre . . . a confusion, on the order of a category mistake” (1986: 70-1).18 Indeed, the received

17 But I don’t take it personally for I am far from alone in being thus neglected. For example, even though Soames rests his defense of DR on appeals to pragmatics, the only work in the extensive pragmatics literature that he mentions is Grice 1989. And I must acknowledge that Soames devotes a whole chapter (2002: ch. 7) to criticizing the views of Richard Larson and Peter Ludlow (1993) and Mark Richard (1990). These views, which take attitude ascriptions to ascribe “linguistically enhanced propositions”, have some similarities to mine, as I note (1996: 166n, 202, 237n). But the differences are important and none of Soames’ criticisms seem to me to undermine my view. Thus, (i), my argument that the meaning ascribed by a proper name like ‘London’ in a t-clause is normally a “disjunctive” mode that includes the mode of ‘Londres’ and all other translations of ‘London’ (1996: 232–4) avoids some criticisms Soames makes of both views (2002: 156–7, 165–7, 174). (ii) I do not embrace the Davidsonianism of Larson and Ludlow that leaves them open to one of Soames’ criticisms ( pp. 150–1). (iii) Most of Soames’ criticism of Richard is of views that arise from Richard’s “hidden-indexical” theory of attitude ascriptions. I also reject hidden-indexical theories. Indeed I have criticized them extensively (1996: 4.11; 1997b), sometimes along lines rather like Soames’. 18 Note that my view is not the genuinely preposterous view that the meaning of a name is a particular token causal link (c.f. Yagisawa 1993: 144) and so is not open to Salmon’s “argument from subjectivity” ( p. 70).

74

Michael Devitt

view, even among those who do not subscribe to DR, seems to be that DR is simply a consequence of accepting a nondescriptive causal theory of names.19 Why would anyone think that the idea that the meaning of a name is its causal mode of reference should not be taken seriously? The idea that the meaning is a mode of reference (“presentation”) of some sort has, of course, been familiar since Frege. It was proposed precisely because of the now-familiar problems for a Millian theory, particularly the problem of identity statements. According to the Fregean tradition this mode is expressed by the descriptions competent speakers associate with the name. But then the revolution showed that the reference of a name is not determined in this descriptive way. So where do we go from there? Well, clearly, the name’s reference must be determined somehow, presumably causally in something like the way that Kripke sketched. So the name must have some sort of causal mode of reference. So perhaps that mode is the meaning. This surely should be a candidate for being the meaning, perhaps not the right candidate, but still a candidate. Yet the idea that this causal mode is a meaning is clearly alien to the semantic tradition; it is, as I have said in the title of a paper, “A Shocking Idea about Meaning” (2001). Why is it so shocking? I think the main cause is a ubiquitous Cartesianism about meaning. It is a truism that competent speakers of a language “know the language”. The Cartesian assumption is that this involves (tacitly) knowing facts about meanings: if an expression has a certain meaning in the language then speakers know that it does.20 Then, since the typical speaker knows nothing about causal modes of reference, those modes cannot be meanings. Yet this popular Cartesianism is almost entirely unsupported and is, I have argued, undermined by the revolution. We should embrace the much more modest view that linguistic competence is an ability or skill, a piece of knowledgehow not knowledge-that (1996: 51–4, 172–6; 2001: 479–81).21 In any case,

19 Some examples: Loar 1976; Ackerman 1979a: 58; 1979b: 6; McGinn 1982: 244; Baker 1982: 227; Lycan 1985; Block 1986: 660, 665; Lepore and Loewer 1986: 60; Wagner 1986: 452; Wettstein 1986: 187; Katz 1990: 31–4. 20 An analogous Cartesianism is rife in linguistics. My Ignorance of Language (2006) is an extended critique. 21 Jason Stanley and Timothy Williamson (2001) have argued ingeniously for the surprising thesis that knowledge-how is really a species of knowledge-that. I have responded (2011).

Still Against Direct Reference

75

this Cartesian explanation should not appeal to the proponents of DR. We would expect them to be skeptical of semantic Cartesianism, and Soames, for one, seems to be (2002: 70–1). Without Cartesianism, the cursory dismissal of a causal mode as a candidate meaning is unwarranted: the dismissal needs an argument. And Salmon and Soames need more than that. For, they have another candidate—the meaning of a name is its referent— and so they need an argument that their candidate is better. To my knowledge, no argument of either sort has ever been attempted by Salmon, Soames, or any other proponent of DR. Instead of argument, we get special pleading: DR is “reasonable” (Soames 2002: 65), “highly motivated” ( p. 142), “highly plausible” (2004: 102), “appealingly simple” (Braun 2001: 256), and so on. 3.2 The argument for causal modes as meanings Of course, the view that a name’s meaning is a causal mode of reference also needs to be argued for and shown to be better than DR. This I have attempted in Coming and the other works cited. I have outlined the argument in the last section. Let me now give a snappy summary to sharpen the issue with DR: 1. A name in a t-clause of an apparently opaque attitude ascription conveys information about a mode of referring to the name’s bearer. 2. A name’s mode of referring to its bearer is causal not descriptive.22 3. Apparently opaque attitude ascriptions explain behavior in virtue of what they convey. 4. So, the causal mode is the name’s meaning. Given DR’s revolutionary ancestry, it surely must accept (2). And Salmon (1986: 70–1) and Soames (2002: 19–20) do seem to accept it (whilst remaining noncommittal about the precise details of these causal modes).

22 Perhaps I should be a bit more cautious: the mode is largely if not entirely causal rather than descriptive. This allows for the fact that the revolution did not show that no descriptive element plays a role in determining the referent of a name. And, as Soames points out nicely, a case can be made that names like ‘Princeton University’ are “partially descriptive” (2002: 51–2).

76

Michael Devitt

What about (1)? I start with a preliminary point about its talk of “apparently” opaque ascriptions. This is to avoid begging the question against DR. Salmon and Soames claim that attitude ascriptions are not really opaque—substitutivity of identity does not really fail—but they, like just about everyone else, agree that some ascriptions seem to be opaque. (1) goes back at least to Frege, is much discussed by Quine, and is very hard to deny. Salmon does seem to accept it (1986: 116). But, whereas a common view is that conveying the mode of reference is a semantic matter, for Salmon it is merely a “pragmatic function” of the ascription ( p. 117). We will get to this difference in a moment (3.4). What about Soames? So far as I can see, he does not explicitly discuss (1) in Beyond Rigidity. He thinks that apparently opaque ascriptions pragmatically convey more than singular Russellian propositions but what he principally has in mind is information captured by non-reference-determining descriptions that the speaker and her audience associate with the name (2002: 210–14). He does not seem to have in mind that these ascriptions also convey modes of reference. However, he earlier accepted that they do (1987: 68; 1988: 117–25) and he offers no reason for going against this near-irresistible view. Now consider (3). The view that apparently opaque attitude ascriptions explain behavior in virtue of what they convey is a pillar of folk psychology, accepted one would have thought by everyone bar a few eliminativists. Yet it is not as obvious as one would like that either Salmon or Soames accepts it. Salmon thinks that it serves “our normal purpose in attributing belief . . . to convey how the believer stands with respect to a proposition” (1986: 116) but he says nothing to suggest that an important “normal purpose” is to explain behavior. So far as I can see there is no mention of explaining behavior in Soames’ book. Still, in a response to Maite Ezcurdia, he does seem prepared to go along, “for the sake of argument”, with her emphasis on the importance of attitude ascriptions in explaining behavior (2004: 103; see also 1987: 68–9). In sum, (2) must be accepted by any proponent of DR. (1) and (3) are widely accepted and Salmon and Soames do not attempt to throw any doubt on them. (4) follows easily. From (1) and (2) it follows that an apparently opaque attitude ascription conveys information about a name’s mode of referring and that mode happens to be causal. Then from (3) it follows that it is partly in virtue of conveying this mode that the ascription explains behavior. So that mode plays a semantic role and is a meaning: it is in virtue of having the property of referring to its bearer in a certain causal way that a name contributes to causing behavior.

Still Against Direct Reference

77

Now a proponent of DR is likely to respond that semantics, properly conceived, is not concerned with this property of causing behavior. Soames does in effect respond in this way to Ezcurdia (2004: 103–4) and Salmon surely would, given his resistance to connections between semantics and psychology (1986: 174, n. 2). This sort of response raises two questions. First, if the response is not to be a merely verbal point about what to call “semantic” or a “meaning” (or “semantic content”) then it needs to be accompanied by an argument that the property of a name that causes behavior is not theoretically interesting, that it is not adverted to in a theory that explains some phenomena. Second, the response needs to be accompanied by an argument that a name’s property of simply referring to its bearer—what DR attends to—is theoretically interesting, that it is adverted to in a theory that explains some phenomena. And it is not sufficient to argue that this property of a name is interesting because it is what a name contributes to the truth conditions of sentences containing it. For, we need an argument that a sentence’s truth condition, its property of representing a certain situation, is theoretically interesting. Perhaps, rather, its property of representing that situation in a certain way (or, indeed, its property of having a certain use) is what is theoretically interesting. We need these arguments if the DR’s preference for a name’s property of referring to its bearer over its property of referring to its bearer in a certain way is not to be theoretically arbitrary, a mere ad hoc stipulation. I made this point in Coming (4.8). I had made it before (1989) and have made it since (2001). Yet, so far as I know, no attempt has been made to supply the needed arguments. This is not to say that the property of referring to the name’s bearer is not theoretically interesting nor to say that an argument for its interest cannot be given. Indeed, I claim to have given something close to this in Coming, attending particularly to the role of names in guiding us to reality (4.2–4.3). But I also claim to give an argument for the theoretical interest of modes of reference in explaining behavior and in guiding us to reality. So, contra DR, these properties are the most theoretically interesting and certainly deserve to be called “meanings” (1996: 150–4). The most objectionable feature of DR is not what it includes in semantics but what it excludes. So my conclusion is that there are good reasons for the view that a name’s causal mode of reference is its meaning and no good reason has been presented against this view. I shall now go on to say more against DR, but the above argument is the core of my case.

78

Michael Devitt

3.3 Braun’s defense of DR I start with a defense of DR offered by David Braun (2001). He takes critics like me to argue that if DR were true, “then attitude ascriptions could not explain (certain sorts) of behavior” ( p. 254). Yet these ascriptions obviously do explain that behavior and so DR must be false. He responds by arguing ingeniously that ascribing a singular Russellian proposition to a person does contribute to explaining a person’s behavior: it “provides a substantial portion of the information contained in some ideal explanation” ( p. 267). But Braun has missed the main point of my argument. That point is not that if DR were true the ascriptions could not explain behavior. The point is that ascribing non-Russellian propositions involving causal modes of reference does, as a matter of fact, explain behavior. This point is quite compatible with the claim that ascribing a singular Russellian proposition can contribute to the explanation in the way Braun illustrates. Indeed, in effect, I explicitly endorse that claim (1996: 153). But ascribing modes of reference provides a more complete explanation. As Braun states nicely: “Clearly, the way in which an agent believes or desires a [singular] proposition can make a difference to that agent’s behavior” ( p. 257). So that way, involving in the case in question a name’s causal mode of reference, is the meaning. There is no principled basis for Braun’s view that the referent is theoretically interesting in a way that the mode of reference is not. Braun is taking as obvious precisely what is in question.

3.4 The pragmatic defense of DR I turn next to “the pragmatic defense” of DR. At the beginning of his book, Soames says: “A central feature of this account is the explanation it provides of how sentences containing names or indexicals may be used to convey, and even assert, propositions the contents of which exceed the semantic contents of the sentences uttered” (2002: v). Now, everyone should accept something along these lines; in particular, in a situation Soames describes ( p. 63), we should probably all accept that ‘Carl Hempel lived on Lake Lane in Princeton’ conveys the information that the wellknown philosopher of science Carl Hempel lived on Lake Lane in Princeton even though the information that Hempel is a well-known philosopher of science is not part of the semantic content of the sentence. And we should all accept that it is often hard to know precisely where to draw the line between what is semantically and what is pragmatically conveyed by a

Still Against Direct Reference

79

sentence.23 Still any claim that an utterance semantically expresses that p but pragmatically conveys that q needs to be supported by an argument. And the argument needs to show not only how an utterance that semantically expresses that p could, in the circumstances, pragmatically convey that q; it needs show that it is more plausible to think the utterance does pragmatically convey that q whilst semantically expressing only that p rather than semantically express that q. I don’t think that DR comes close to providing the arguments we need here. The key issue is whether a name conveys information about its mode simply in virtue of a semantic convention. If it does then that information is semantically not pragmatically conveyed. Here is evidence that it does. Let us attend to simple predicative utterances of the form ‘a is F ’. First, we note that it is highly relevant to the explanation of the behavior of a person who utters ‘a is F’ to know that the thought she expresses is under the mode of reference of ‘a’. Thus, whether Mary’s belief of Twain/Clemens that he is sitting at a nearby table is under the mode of ‘Twain’ or ‘Clemens’ might make a big difference in her behavior; for example, given her other beliefs, if the belief is under the former mode she might rush up to Twain/ Clemens for an autograph but if it is under the latter mode she certainly will not. So we need people regularly to convey information about modes of thought and we would expect there to be a semantic convention to facilitate their doing so. Second, the fact that people regularly—indeed always?—convey this information in uttering ‘a is F’ is evidence that doing so is a semantic convention. When a person has a thought under a certain mode she standardly uses an expression under that very mode to express the thought. Soames states a nice requirement for semantic content: “the semantic content of s . . . should consist of information that a competent speaker who assertively utters s asserts and intends to convey in any context in which s is used nonmetaphorically” (2002: 57). Information about modes fits this requirement perfectly. Finally, as we have already noted, our standard way of ascribing meanings to utterances to explain behavior— opaque attitude ascriptions—conveys information about modes. This case is persuasive that information about a name’s mode of reference is semantically conveyed. Even if the contrary view that this information is pragmatically conveyed faced no insurmountable difficulty, it is hard to see any reason for preferring that view. And the view does seem to face an insurmountable difficulty: the problem of identity statements. 23 Coming throws sadly little light on how to draw this line in general. I think I have done better in some later works (1997b, 2004, 2008 and, particularly, forthcoming).

80

Michael Devitt

3.5 Identity statements That the true identities ‘a = a’ and ‘a = b’ differ in meaning is about as powerful a semantic intuition as we have. Still, we should not rest with an intuition: we need an argument. I do argue for this difference, but not in the usual way that appeals to the difference in informativeness of the identities. For, what have epistemic issues about informativeness got to do with semantic issues about meaning? People think that the epistemic issues are relevant because they take something like the following for granted: ‘S1’ and ‘S2’ mean the same only if all competent speakers know that they do. This is a dramatic example of the sort of Cartesianism that Coming rejects. I argue for the difference in meaning by applying the methodology: the folk normally distinguish the meanings of ‘a = a’ and ‘a = b’ in using t-clauses to serve their semantic purposes (descriptive), and they are right to do so (normative); the t-clauses are opaque and ought to be so to serve the purposes, particularly those of explaining behavior. So the two identities differ in meaning. And, of course, the semantic view I am urging accounts for this difference easily: ‘a’ and ‘b’ have different modes of reference (4.7; see also 1997c, pp. 382–6 for a correction). DR has to start its treatment of identity statements with the highly implausible claim that ‘a = a’ and ‘a = b’ do not differ in meaning. So the differences between these statements have to be explained away pragmatically. And so too do differences in ascriptions of the attitudes that a = a and that a = b. The dimension of this problem for DR has been demonstrated countless times but Soames demonstrates it as well as anyone has (2002: 230–2). Suppose that Tom says “a = b” to Mary and that as a result Dick says “Mary has just learned that a = b”. Set aside any propositions Dick thereby pragmatically conveys that Mary has just learned and consider the proposition that, according to DR, he semantically expresses that Mary has just learned. That proposition is a triviality. For, Dick semantically expresses just what he would have had he said “Mary has just learnt that a = a”. Now had he said that he would have expressed something false: Mary has known the triviality that a = a ever since she first heard of a. Yet nobody (except of course proponents of DR) would accept that Dick’s actual utterance falsely asserts that this triviality is what Mary has just learnt. Soames unflinchingly bites the bullet: we are all wrong. This is not plausible (to put it delicately). In sum, I have argued that DR is theoretically unmotivated. And it has no convincing answer to the well-known problem of identity statements.

Still Against Direct Reference

81

4. Conclusion I started by summarizing my argument in Coming to Our Senses (1996) for a certain semantic methodology and truth-referential theory of meaning. This explanation of meaning is far from complete, of course. What it mostly lacks are completely satisfactory theories of reference. Central to this truth-referential theory is the view that a name’s causal mode of reference is at least one of its meanings. This is the core of my argument against direct reference, DR. I went on to dismiss Braun’s defense of DR based on the correct claim that ascribing a singular Russellian proposition can contribute to the explanation of behavior. Finally, I considered the pragmatic defense of DR. Even if the DR view that information about a name’s mode of reference is pragmatically conveyed faced no insurmountable difficulty, there seems to be no reason for preferring it to the view that this information is semantically conveyed. And the DR view does face an insurmountable difficulty: the problem of identity statements.24

References Ackerman, Felicia (Diana). 1979a. “Proper Names, Propositional Attitudes and NonDescriptive Connotations”. Philosophical Studies 35: 55–69. Ackerman, Felicia (Diana). 1979b. “Proper Names, Essences and Intuitive Beliefs”. Theory and Decision 11: 5–26. Baker, Lynne Rudder. 1982. “Underprivileged Access”. Nous 16: 227–42. Block, Ned. 1986. “Advertisement for a Semantics for Psychology”. In French, Uehling, and Wettstein 1986: 615–78. Braun, David. 2000. Review of Devitt 1996. Philosophy and Phenomenological Research 60: 489–92. Braun, David. 2001. “Russellianism and Explanation”. In Philosophical Perspectives, 15, Metaphysics, 2001. James Tomberlin, ed. Atascadero, CA: Ridgeview Publishing Company (2001). 253–89. Castaneda, Hector-Neri. 1966. “He : A Study in the Logic of Self-Consciousness”. Ratio 8: 130–57. Castaneda, Hector-Neri. 1967. “Indicators and Quasi-Indicators”. American Philosophical Quarterly 4: 1–16. 24 Parts of this paper were delivered at the Australasian Association of Philosophers at Armidale, July 2007, and at a conference on John Perry’s Work, XVIII InterUniversity Workshop on Philosophy and Cognitive Science in Madrid, April 2008.

82

Michael Devitt

Crimmins, Mark, and John Perry. 1989. “The Prince and the Phone Booth: Reporting Puzzling Beliefs.” Journal of Philosophy 86: 685–711. Reprinted in Perry 1993. Devitt, Michael. 1974. “Singular Terms”. Journal of Philosophy 71: 183–205. Devitt, Michael. 1980. “Brian Loar on Singular Terms”. Philosophical Studies 37: 271–80. Devitt, Michael. 1981a. Designation. New York: Columbia University Press. Devitt, Michael. 1981b. “Donnellan’s Distinction”. In French, Uehling, and Wettstein 1981: 511–24. Devitt, Michael. 1989. “Against Direct Reference”. In French, Uehling, and Wettstein 1989: 206–40. Devitt, Michael. 1994. “The Methodology of Naturalistic Semantics”. Journal of Philosophy 91: 545–72. Devitt, Michael. 1996. Coming to Our Senses. Cambridge: Cambridge University Press. Devitt, Michael. 1997a. “A Priori Convictions about Psychology: A Response to Sosa and Taylor”. In Truth: Philosophical Issues, 8, 1997, Enrique Villanueva, ed. Atascadero: Ridgeview Publishing Company (1997): 371–85. Devitt, Michael. 1997b. “Meanings and Psychology: A Response to Mark Richard”. Nous, 31: 115–31. Devitt, Michael. 1997c. “Responses to the Maribor Papers”. In The Maribor Papers in Naturalized Semantics. Dunja Jutronic, ed. Maribor: Pedagoska fakulteta (1997): 353–411. Devitt, Michael. 2001. “A Shocking Idea about Meaning”. Revue Internationale de Philosophie 208: 449–72. Devitt, Michael. 2002. “Meaning and Use”. Philosophy and Phenomenological Research 65: 106–21. Devitt, Michael. 2004. “The Case for Referential Descriptions”. In Descriptions and Beyond, Marga Reimer and Anne Bezuidenhout, eds. Oxford: Clarendon Press (2004): 280–305. Devitt, Michael. 2006. Ignorance of Language. Oxford: Clarendon Press. Devitt, Michael. 2008. “Referential Descriptions and Conversational Implicatures”. European Journal of Analytic Philosophy 3: 7–32. Devitt, Michael. 2010a. Putting Metaphysics First: Essays on Metaphysics and Epistemology. Oxford: Oxford University Press. Devitt, Michael. 2010b. “Deference and the Use Theory”. ProtoSociology 27: 196–211. Devitt, Michael. 2011. “Methodology and the Nature of Knowing How”. Journal of Philosophy CVIII: 205–18. Devitt, Michael (forthcoming). “What Makes a Property ‘Semantic’?”. In Perspectives on Pragmatics and Philosophy A. Capone, F. Lo Piparo, and M. Carapezza, eds. Springer (forthcoming). Devitt, Michael, and Kim Sterelny. 1999. Language and Reality: An Introduction to the Philosophy of Language, 2nd edn (1st edn 1987). Oxford: Blackwell Publishers. Donnellan, Keith S. 1972. “Proper Names and Identifying Descriptions”. In Semantics of Natural Language, Donald Davidson and Gilbert Harman, eds. Dordrecht: D. Reidel (1972): 356–79. French, Peter A., Theodore E. Uehling, Jr., and Howard K. Wettstein, eds. 1979. Contemporary Perspectives in the Philosophy of Language. Minneapolis: University of Minnesota Press.

Still Against Direct Reference

83

French, Peter A., Theodore E. Uehling, Jr., and Howard K. Wettstein, eds. 1981. Midwest Studies in Philosophy Vol. 6, The Foundations of Analytic Philosophy. Minneapolis: University of Minnesota Press. French, Peter A., Theodore E. Uehling, Jr., and Howard K. Wettstein, eds. 1986. Midwest Studies in Philosophy Vol. 10, Studies in the Philosophy of Mind. Minneapolis: University of Minnesota Press. French, Peter A., Theodore E. Uehling, Jr., and Howard K. Wettstein, eds. 1989. Midwest Studies in Philosophy Vol. 13, Contemporary Perspectives in the Philosophy of Language II. Notre Dame: University of Notre Dame Press. Grice, H. P. 1989. Studies in the Way of Words. Cambridge, MA: Harvard University Press. Higginbotham, James. 1991. “Truth and Understanding”. Iyyun, the Jerusalem Quarterly 40: 271–88. Horwich, Paul. 1998a. Truth, 2nd edn (1st edn 1990). Oxford: Clarendon Press. Horwich, Paul. 1998b. Meaning. Oxford: Clarendon Press. Horwich, Paul. 2005. Reflections on Meaning. Oxford: Clarendon Press. Kaplan, David. 1968. “Quantifying In”. Synthese 19: 178–214. Katz, Jerrold. 1990. “Has the Description Theory of Names Been Refuted?” In Meaning and Method: Essays in Honor of Hilary Putnam, George Boolos, ed. Cambridge: Cambridge University Press (1990): 31–61. Kripke, Saul A. 1979. “A Puzzle about Belief ”. In Meaning and Use, A. Margalit, ed. Dordrecht: D. Reidel (1979): 239–83. Kripke, Saul A. 1980. Naming and Necessity. Cambridge, MA: Harvard University Press. Larson, Richard K., and Peter Ludlow. 1993. “Interpreted Logical Forms”. Synthese 95: 305–57. Lepore, Ernest, and Barry Loewer. 1986. “Solipsistic Semantics”. In French, Uehling, and Wettstein 1986: 595–614. Loar, Brian. 1976. “The Semantics of Singular Terms”. Philosophical Studies 30: 353–77. Lycan, William G. 1984. Logical Form in Natural Language. Cambridge, MA: MIT Press. Lycan, William G. 1985. “The Paradox of Naming”. In Analytical Philosophy in Comparative Perspective, B. K. Matilal and J. L. Shaw, eds. Dordrecht: D. Reidel (1985): 81–102. McGinn, Colin. 1982. “The Structure of Content”. In Thought and Object, A. Woodfield, ed. Oxford: Clarendon Press (1982): 207–58. Perry, John. 1993. The Problem of the Essential Indexical and Other Essays. New York: Oxford University Press Quine, W. V. 1953. From a Logical Point of View. Cambridge, MA: Harvard University Press. Quine, W. V. 1960. Word and Object. Cambridge, MA: MIT Press. Quine, W. V. 1966. Ways of Paradox and Other Essays. New York: Random House. Richard, Mark. 1983. “Direct Reference and Ascriptions of Belief ”. Journal of Philosophical Logic 12: 425–52. Richard, Mark. 1990. Propositional Attitudes: An Essay on Thoughts and How We Ascribe Them. Cambridge: Cambridge University Press.

84

Michael Devitt

Salmon, Nathan. 1986. Frege’s Puzzle. Cambridge, MA: MIT Press. Schiffer, Stephen. 1979. “Naming and Knowing”. In French, Uehling, and Wettstein 1979: 61–74. Soames, Scott. 1987. “Direct Reference, Propositional Attitudes, and Semantic Content”. Philosophical Topics 15: 47–87. Soames, Scott. 1988. “Substitutivity”. In On Being and Saying: Essays for Richard Cartwright, J. J. Thomson, ed. Cambridge: MIT Press (1988): 99–132. Soames, Scott. 2002, Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity. New York: Oxford University Press. Soames, Scott. 2004. “Reply to Ezcurdia and Gomez-Torrente”. Critica, Revista Hispanoamericana de Filosofia 36: 83–114. Stanley, Jason, and Timothy Williamson. 2001. “Knowing How.” Journal of Philosophy 98: 411–44. Wagner, Steven J. 1986. “California Semantics Meets the Great Fact”. Notre Dame Journal of Formal Logic 27: 430–55. Wettstein, Howard. 1986. “Has Semantics Rested on a Mistake?” Journal of Philosophy 83: 185–209. Yagisawa, Takashi. 1993. “A Semantic Solution to Frege’s Puzzle”. In Philosophical Perspectives, 7: Language and Logic, James E. Tomberlin, ed. Atascadero, CA: Ridgeview Publishing Company: 135–54.

On Meaning, Meaning and Meaning Ruth Garrett Millikan Abstract: To understand how language works, we look, first, to the cooperative functions that various language forms perform, understanding these on a biological model as what they accomplish for both speaker and hearer that keeps them in circulation. Next we look at language mechanics, at how language forms perform their functions. For many (but not all) language forms there are conditions in the world that would be necessary to support their functions and that vary systematically with certain variations in the forms themselves. These are truth or satisfaction conditions. Last we describe the psychological mechanisms that are involved in implementing the functions of various language forms. For extensional terms these involve methods of identification of incoming information about the extension of the term, methods that are usually multiple and various for each person and that may also vary widely from person to person.

I. Introduction Various cris-crossing distinctions have been drawn in the philosophical tradition between kinds or dimensions of linguistic meaning or between meaning and other dimensions of linguistic function. Here I’ll try to collect together from various books and papers the results of my own investigations on different aspects of meaning. The underlying idea is that to understand how language works, one must look, first, to the functions that various language forms perform, understanding these on a biological model as what these forms accomplish that keeps them in circulation. To explain the function of a language form is to explain its survival value, the source of its proliferation, what it does that accounts at the same time for the fact that speakers continue to use it and that hearers continue to react to it often enough in standard ways. Next we should look at language mechanics, at how language forms perform their functions. For some language forms there are conditions in the world that are necessary to support their functions and that vary systematically with certain variations in the

86

Ruth Garrett Millikan

forms themselves. These are truth conditions, determined as such by the kind of “meanings” that are semantic mapping functions—“functions”, this time, in the mathematical sense. (Conversely, truth conditions delimit but do not determine semantic mapping functions. I will get to this.) Last we need to describe the psychological mechanisms that are involved in implementing the functions of various language forms, the ways that speakers and hearers manage to produce and understand these forms so as to promote performance of their linguistic functions. These vary from person to person. There is also an important distinction, of course, between speaker meaning and linguistic or conventional meaning. This is the difference between the function, with its associated truth condition and so forth, of a public language form and functions that individual speakers may use or try to use the form to serve. I will discuss this distinction, but only to set speaker meaning aside. My basic proposal is that there are these three basic kinds of linguistic meaning: (1) Conventional linguistic function, to be called “stabilizing function”. (2) Conventional semantic mapping functions (“functions” in the mathematical sense) which determine truth and other kinds of satisfaction conditions. (3) Methods of identifying—to be called “conceptions” and “conceptual components”—that govern individual speakers’ grasps of referents and of truth or satisfaction conditions, hence help to determine their dispositions to use and understand various conventional language forms. Conceptions governing individual speakers’ grasps of referents and satisfaction conditions may exhibit little or no overlap among competent members of a language community. The upshot of the whole is that none of the aspects of meaning corresponds at all well either to any traditional notion of intension or to any Frege-related notion of sense.1

II. Stabilizing Functions When speakers are conforming to the conventions of a language, speaker meanings coincide with conventional meanings. But speakers often use 1 This despite the embarrassing fact that in my (1984), “(Fregean) sense” was the name I gave to what I now call “semantic mapping” and “intension” my name for what I now call “conception.” I had my reasons, but they were not good.

On Meaning, Meaning and Meaning

87

language forms for purposes that diverge from conventional meanings. To distinguish speaker meanings from conventional linguistic meanings, we need to know what a public language convention is. Probably the bestknown theory of convention and its application to language is that of David Lewis (1969, 1975). I will clarify my position by comparing it with his. Lewis describes a convention as a regularity in the behavior of a population such that within the population there is mutual knowledge (1) that everyone conforms to the regularity, (2) that everyone prefers to conform given that the others do and (3) that everyone expects everyone else to conform for the same reason they do. The reason each prefers to conform is that conforming solves a coordination problem. A coordination problem arises when people have a purpose in common which must be achieved by joint action, where the contribution that each must make will vary depending on what each of the others contributes, and where there is more than one acceptable way of combining contributions to produce a successful outcome. Then coordination is necessary. It is best for everyone if everyone makes his contribution according to the same solution plan. To each it doesn’t matter as much which plan is chosen as it matters that the same plan is chosen by all. In many cases, Lewis says, the plan that is chosen will be the one for which there is a precedent. It has been used before, which makes it a salient plan, one that comes to mind and that each participant assumes will come to the mind of the other participants. Each participant thus steps into his role in this plan on the assumption that the others will adopt their roles according to the same plan. When a precedent for solving a coordination problem spreads in this way a convention is born. Thus Lewis claimed that social conventions of all kinds, including linguistic conventions, are supported by rational beliefs and intentions concerning one another’s thoughts. It is true that children and idiots may conform to the conventions of language without having reasons of this sort but, Lewis claims, “they are not parties to the convention and their linguistic competence is incomplete” (1969 p. 51). Now I agree that the conventions of language arise and spread because they solve certain kinds of coordination problems. Not all conventions solve coordination problems, however. And for those that spread because they do, the “because” is almost never a reasoned because but some more mundane kind of causal because. The rest of us conform to linguistic conventions in exactly the same unreasoned way that the idiot and the child do. Further, despite apparent consensus among philosophers that conventions always involve regularities of behavior within a group (Searle 1969; Lewis 1969, 1975; Schiffer 1972; Bach & Harnish 1979; Gilbert

88

Ruth Garrett Millikan

1983, 1989/1992; Recanati 19872), my claim is that conventions do not generally require regularities of behavior, either de facto or de jure. In particular, conventional coordinations, including linguistic coordinations, do not, in general, require regularities of behavior. These claims are defended in (Millikan 1998). Here I will just try to make them clear. A convention, in the sense that a natural language contains conventions, is merely a pattern of behavior that is (1) handed down from one person, pair, or group of persons to others—the pattern is reproduced— and (2) is such that, if the pattern has a function, then it is not the only pattern that might have served that function about as well. Thus if a different precedent had been set instead, a different pattern of behavior would probably have been handed down instead. Putting a wreath on the door at Christmas time, dyeing eggs for Easter and drinking green beer on St Patrick’s day are conventions in this sense. In Japan the convention is to eat with chopsticks, in America, with a knife and fork. Against Lewis, that these are conventions (1) does not necessarily mean that they solve coordination problems. Also (2) it does not necessarily mean that they are universally followed. Indeed, there are many conventions for which conformity is neither prescribed nor mandatory in any sense. Of course, some conventions, such as driving on the right in the United States, do solve coordination problems, are universally followed, and are mandatory. But that is not what makes them be conventions. Also, linguistic conventions do solve coordination problems but, as I will explain, they are neither universally followed nor mandatory. When a conventional pattern of behavior is handed down because it is solving a coordination problem, the mechanism for this is usually quite simple. No matter how the precedent for the convention was originally set, if the coordination it effects is an obvious and important one, it will tend to proliferate without anyone’s thinking about anyone else’s thoughts. Like other higher animals, people repeat behaviors that have been successful in achieving wanted results in the past. Unlike other animals, they tend also to copy behaviors of others that have been successful in producing wanted results. Behaviors that constitute solutions to coordination problems achieve results desired by all parties to the coordination, hence these

2 Recanati (1987) takes it that conventional language devices “indicate” or, using the linguists’ term, “mark” uses of language, that is, conventions mandate that these devices shall be used only for those purposes. This is not explicitly stated, but see, for example, §22.

On Meaning, Meaning and Meaning

89

behaviors will tend to be reproduced when similar results are desired. There is no need for the various parties in the coordination even to recognize the problem as a coordination problem let alone to think about one another’s thoughts in order for the convention to proliferate. If other people are driving on the right, then I will drive safely only if I drive on the right. Thus I might learn to drive on the right without ever quite realizing that it is only a convention to do so. Exactly so, not only children but very smart primitives typically are unaware that the languages they speak are merely conventional. Specific language forms continue to be reproduced by speakers within a language community merely because, often enough, they prompt hearer responses that contribute to the fulfillment of speaker purposes in speaking. Similarly, hearers continue to respond in conventional ways, for example, by believing or by doing what they are told, because, often enough, the result is rewarding for them. Often enough, believing or doing what one is told leads to believing or doing what is useful or what will keep one out of trouble. Speakers within a language community are, simply, adapted to an environment in which hearers are responding, sufficiently often, to the forms speakers produce in ways that reinforce these speaker productions. Correlatively, hearers in the community are, simply, adapted to conditions under which speakers, sufficiently often, produce these language forms in circumstances such that making conventional responses to them aids those hearers. Consider, for example, a speaker whose purposes in using the word “dog” will be achieved only through calling attention to dogs or to facts that concern dogs or through changing behaviors toward dogs. Such a speaker will eventually stop trying to use the word “dog” for these purposes if they are never achieved. Similarly, a hearer whose languageunderstanding faculties turn his mind to dogs whenever speakers use the word “dog” will soon unlearn this response if speakers never use the word “dog” such that it carries information or expresses intentions that concern dogs. Consider those syntactic forms that get labeled “indicative” in various languages. These forms usually have a number of alternative functions, but no form will be so labeled unless one of its functions is to effect production of true beliefs having the propositional contents carried by the other aspects of these sentences. These conventional forms are surviving in part because, often enough, this particular effect is of interest both to speakers and to hearers. Production of false hearer beliefs may occasionally interest speakers, but rarely serves the interests of hearers. A hearer unable to interpret the indicative sentences he hears so as sometimes to extract

90

Ruth Garrett Millikan

genuine information from them would soon cease to form beliefs on their basis. And if hearers ceased ever using indicative sentences as guides in forming beliefs, speakers would stop trying to use them for purposes that required imparting beliefs. Similarly, if it were not sometimes in the interest of hearers to comply with imperatives—advice, instructions, directions, friendly requests, sanctioned directives, and so forth—they would soon cease ever to comply. And if hearers never complied with imperatives, speakers would soon cease to issue them. Imperative syntactic forms would disappear from the language.3 A corollary is that the functions of public language forms are not on the same level as either speaker purposes or hearer purposes taken alone. The conventional functions of language forms are not, for examples, merely standard speaker purposes. Conventional language forms are selected for performing services satisfactory at once to both partners in communication. Their functions must balance speaker with hearer interests. Because the conventional function of a linguistic form will remain stable only if it continues to serve the interests of both speakers and hearers often enough, I call it a “stabilizing function.” Linguistic “meaning” in the sense of stabilizing function is on a entirely different level from, for example, average speaker meaning. Similarly, on this analysis a linguistic convention consists in a pattern that includes both a conventional contribution by the speaker and a conventional contribution by the hearer. The hearer’s contribution is as much a part of the convention as is the speaker’s. Thus the linguistic convention includes important aspects of what Austin called the perlocutionary act. As such it effects a genuine coordination between speaker and hearer, each of whom must play his part if the coordination is to be successful. Contrary to this, Lewis claimed that “[a] member of the audience, as such, is not constrained by convention . . . Only when he takes his turn as communicator does he himself act in conformity to the convention of truthfulness in L” (1969, p. 179–80). We can see why Lewis took this position, however. According to his analysis, a convention was a “regularity of behavior in a population” such that “everyone conforms to the regularity.” As the book Convention proceeds, he does modify this to allow that almost everyone conform almost all of the time (1969 p. 78), but still, it seems

3 Critics often ask at this point about Chomskian linguistic universals. Chomskian universals, if such should exist, would merely be filters restricting the aspects of what one hears that one will reproduce. There is no incompatibility.

On Meaning, Meaning and Meaning

91

plainly false that almost every hearer of a directive complies with it and plainly false that almost every hearer of a description believes it. So on Lewis’s account the hearer’s response could not be part of a linguistic convention. Lewis also says, “forming a belief . . . is normally not a voluntary action and hence not an action in conformity with convention,” and of directives, “[e]ven if the audience should act, the action may not answer to an interest common to the communicator and the audience” ( p.180). But if, as I have claimed, a convention is merely a reproduced pattern whose form, should it have a function, is arbitrary with respect to that function, then there is no requirement on how voluntarily or how regularly the pattern is reproduced or on how often the pattern is broken, with either speaker or hearer failing to contribute his proper part (Millikan 1984 chapter 4; 1998). Sometimes the speaker is not interested in genuine cooperation. Sometimes the hearer is not. Sometimes mistakes are made. Conventional coordination patterns need to succeed only often enough to avoid extinction. Notice as well that alternative coordination conventions that serve the same purpose and from which one can choose often happily exist side by side in a community. Besides linguistic conventions, Lewis talked of “signaling conventions” which he did describe as involving the receiver’s responses as well as the signaler’s gestures. He illustrated with signals used by a man standing outside to help a truck driver back into a tight space. The helper and the driver both want to maneuver the truck safely into the space. How the signals are composed by the helper and how they are read by the driver have to be coordinated if this common end is to be achieved. Lewis apparently overlooked that both helper and driver might easily be acquainted with more than one signaling system commonly used for this purpose. So long as these different systems didn’t happen to contain identical signals that meant different things, it wouldn’t matter which system the helper chose to use, the driver would recognize the signal and follow it. What is necessary for success is only that the same precedent should be followed by both helper and driver on each individual occasion. What they do on other occasions doesn’t matter. An initiating move by one party will immediately be recognized by the other as coming from a particular lineage of precedent with which both are familiar. Thus it is that linguistic conventions are neither universally followed nor mandatory. Many alternative conventions can possess the same stabilizing function. (For a fuller discussion of this issue, see (Millikan 1998).) Language conventions can be considered as lineages of precedent. A public language is a huge web of crisscrossing lineages of reproduced

92

Ruth Garrett Millikan

patterns consisting of tokens of linguistic forms and responses to them. People listen to one another, then repeat words and idioms they have heard, syntactic arrangements they have heard and tonal inflections they have heard, arranging these into new combinations. Words, idioms, syntactic forms, tonal inflections and so forth are handed down from one person to others because these elements are helping to serve coordinating functions. These stabilizing functions are, in one of that term’s various senses, their “meanings,” the first of the three “meanings” listed in this essay’s title. One thing to investigate then is exactly what kinds of stabilizing functions compose these meanings. What various jobs do linguistic forms do to keep themselves in circulation? The thesis that linguistic conventions correspond to reproducing lineages of cooperatively-used tokens-with-responses has direct implications for the individuation of linguistic forms. For the purposes of semantics, what makes two tokens be tokens of the same linguistic type is not their sound or shape, or the phonemes or letters of which they are composed, or their surface syntactic arrangements. They are tokens of the same type only if they have been copied from the same pool of tokens reproducing in the same language community. They must be members of the same historical lineage or lineages. Genuine words cannot be accidentally formed by the wind. Further, any genuine linguistic token is automatically a token from one particular language or another. When discussing linguistic forms, reference to the form as being “in L1” or “in L2,” etc., may help the hearer to identify the form intended, but whether identified or not, if the form is a genuine natural language form, it already is essentially either in L1 or L2 or some other language without that. Otherwise it is not a linguistic form, but merely a shape or sound. Unlike the lineages that make up animal species, linguistic lineages frequently acquire new functions without changing their physical forms. Novel uses of conventional linguistic forms introduced by speakers through metaphor and other figures of speech or through Gricean implicature are similar to mutations in biological evolution. If the hearer understands the metaphor or the implicature, the novel use will serve a new coordinating function. It may then be copied by other speakers and may in time be understood directly by hearers without having to go through the process of unpacking a metaphor or an implicature. Then a new lineage of tokens with a different stabilizing function has branched off from the original lineage but without any change in physical form. Suppose, for example, that a new metaphor is copied again and again. For a very long time, those who use it and those who understand it may continue to read

On Meaning, Meaning and Meaning

93

it as derived from its original source. For most speakers and hearers, acquaintance with the old lineage and with the new lineage may both together and equally be responsible for its continued use and for its easy comprehension. Later, however, the new use may become as familiar as the old and may start to be proliferated quite independently. Then the metaphor becomes “dead.” An entirely independent branch of the family has been formed from tokens with exactly the same physical form. The result is “polysemy”—one sound, many meanings. Families of linguistic forms quite typically form wide-spreading bushes, many different branches having slowly formed over time, and more branches from those branches. Since branches often take a long time to separate off completely, the places where true branchings begin are not at all sharp. At a given time, exactly how many branches there are is not definite. Still, it is helpful to give a name to the branches of a given form that are currently fairly independent, each being well enough established that it would survive even if all the others should die out. These independent branches I call “least types.” Least types correspond to the various relatively independent stabilizing functions (different “senses”) of a polysemous language form (Millikan 1984 Chapter 4). However, typically the number of such “meanings” and the divisions between them are very far from precise. Surface syntactic forms may also branch into independent least types. They too may be polysemous, having a number of branches that propagate more or less independently. For example, definite descriptions are sometimes used by speakers merely to identify a particular referent for a hearer, the description itself being of no interest at all. Other times definite descriptions are of interest in their own right, indeed, the speaker may not care whether the hearer identifies the referent or not so long as the description is remembered. Donnellan’s distinction (Donnellan 1966), when understood this way, is a distinction between two stabilizing functions of definite descriptions that tend to divide these descriptions into two least types. Definite descriptions tend to be polysemous in stabilizing function.4

4 It doesn’t follow that they are polysemous in semantic mapping function or that they correspond to more than one kind of truth condition. I will discuss this below. Donnellan’s claim that there is a use of definite description for which it doesn’t matter whether the description is empty so long as the speaker’s intended referent is understood by the hearer is best interpreted as a claim about speaker meaning rather than linguistic meaning. Certainly it is not a stabilizing function of definite descriptions to bring things not correctly described by them to hearer’s minds.

94

Ruth Garrett Millikan

Similarly, consider the various grammatical moods in a language. Sometimes you impart beliefs with the indicative mood but in the army it is used to give orders. Sometimes you ask questions with the interrogative mood but other times you use it to make requests. A number of relatively separate conventions are helping to propagate the same surface syntactic forms. Differences in what Austin called “illocutionary” and also “perlocutionary” force may often be carried conventionally by syntactic forms that are polysemous in this way (Millikan 1984 chapter 4; 2001). Then understanding the force of a particular utterance using the form will require disambiguation from context, just as understanding the meanings of polysemous individual words often does.

III. Semantic mapping functions Many linguistic forms work in part by mapping or, as Wittgenstein put it, “picturing.” When functioning in the ways that have accounted for their proliferation, they correspond to states of affairs in accordance with semantic mapping functions that have been determined by convention. Directive communicative forms have as their stabilizing functions to yield states of affairs—completed actions—that vary with variations in the composition of the sentences exemplifying these forms. For example, directive least types used in giving orders have as stabilizing functions to produce compliance, where what constitutes compliance is determined, along the lines Tarski proposed, by the composition of the rest of the sentence. The state of affairs that would result from compliance is called the “satisfaction condition” of the directive sentence. Descriptive communicative forms have stabilizing functions that can be performed through normal mechanisms only if the forms correspond to states of affairs existing prior to or independently of performance of their stabilizing functions. For example, conventional fact-stating least types function to produce true beliefs in hearers, but a true belief will be formed by normal mechanisms only if the sentence corresponds to a world affair in accordance with its conventional semantic mapping function. False sentences do not cause true beliefs in hearers, at least not through normal mechanisms. For brevity I will also call the truth conditions of a descriptive sentence “satisfaction conditions.” The semantic mapping function for a sentence determines the sentence’s satisfaction condition, but the satisfaction condition does not determine the semantic mapping function. The semantic mapping function

On Meaning, Meaning and Meaning

95

is given by rules according to which significant transformations of the sentence—typically substitution transformations—that conserve its syntactic form yield different truth or satisfaction conditions. Compare the sentence “It’s raining” with the sentence “Rain is falling here now. “It’s raining” contrasts with “It’s snowing,” “Its hailing,” “It’s sleeting,” and so forth. All display the same syntactic form, the transformations that substitute “snow,” “hail,” and “sleet” for “rain” determining different satisfaction conditions in a systematic way. “Rain is falling here now” also contrasts with “Snow is falling here now,” “Hail is falling here now” and so forth. But it contrasts, further, with “Mist is rising here now,” and with “Rain was falling in Rome yesterday.” The truth conditions of “It’s raining” and of “Rain is falling here now” are the same, but the semantic mapping is different. Similarly, “Many drops of water are presently precipitating from the atmosphere and landing close to this place” is articulated by yet another semantic mapping function. Or compare the semantic mapping function for a bee dance with that of an English sentence having the same truth condition. There are no transformations of the bee dance that would tell about nectar location relative to objects other than the hive and the sun, or about the location of anything other than nectar. Only reference to the angle between the nectar and the line from the hive to the sun can be varied in the bee dance. Further, the bee dance is not subject to a negation transformation. No English sentence with the same truth conditions approaches this degree of inarticulateness. The semantic mapping of a sentence articulates it, placing it in a logical space of contrasting possibilities. Its truth condition is not, as such, articulated. Stabilizing functions can vary while semantic mapping remains the same. Compare “Jane will close the door” to “Will Jane close the door?” And there are more interesting examples. Wilfrid Sellars claimed that the function of the sentence form “‘X’ means Y,” as in “‘Rot’ means red” and “‘und’ means and,” is to produce in the hearer a disposition to use “X” in the same way he already knows to use “Y.” The “Y” in this rubric, Sellars said, is neither mentioned nor used in the usual way (Sellars 1956). It is used in a special way, held up, as it were, as a model (Millikan 2004 Chapter 7). Compare the function of the form “‘X’ and ‘Y’ have the same meaning.” Here “Y” is mentioned rather than used. This sentence has the same truth condition as “‘X’ means Y” but its function is different. Its function is to produce a belief about words whereas the function of “‘X’ means Y” can be performed even if the hearer lacks a concept of words (as very young children apparently do—Susan Carey, private correspondence).

96

Ruth Garrett Millikan

Peter Strawson claimed that the function of the identity form “A is B,” as in “Cicero is Tully,” is to induce the hearer to merge all of the information he has accumulated under the concept he associates with the word “A” with the information he has accumulated under the concept he associates with “B,” so that he no longer harbors this information under two separate concepts (Strawson 1974). More accurately, the stabilizing function of “A is B” must be to induce the hearer to do this appropriately, such that the resulting concept is not confused or equivocal (Millikan 2000). If this is the function of the form “A is B,” then its truth condition is the same as that of the form “‘A’ has the same referent as ‘B’,” in which the “A” and the “B” are mentioned rather than used. But these two sentence forms do not have the same function. The hearer of “‘A’ has the same referent as ‘B’” is to form a belief about words, hence needs concepts of words and also the concept of reference, whereas the hearer of “A is B” needs neither. I have claimed (Millikan 1984, chapter 12) that the stabilizing function of the form “A does not exist” is to (correctly) induce the hearer to disengage his concept associated with “A” from ordinary referential uses, relegating it, for example, to pretend uses, or eliminating it entirely from his conceptual repertoire. Correlatively, the function of “A exists” is to (correctly) begin to engage a previously disengaged concept associated with “A”. But if this is true, the sentence forms “A does not exist” and “A exists” have the same truth conditions as do “‘A’ has no referent” and “‘A’ has a referent” though, again, the functions of these sentences are not the same. The latter two have as stabilizing functions to cause beliefs about words. Adding a different kind of example, the two uses of definite descriptions mentioned above in connection with Donnellan’s distinction may correspond to two independent stabilizing functions of these, but these two uses require exactly the same conditions for truth. The world affair needed to make one least type of definite description serve its stabilizing function through normal mechanisms is the same as that needed for the other. In both cases the truth conditions are Russellian.5 The study of semantic mapping functions should include a study of the peculiarities of indexicals and demonstratives. As I understand it, there is more than one way of describing their semantic mapping functions and I have suggested somewhat different though, I believe, compatible ways of

5 See footnote 4 above.

On Meaning, Meaning and Meaning

97

thinking about the matter in (1984 chapter 10) and (2004 chapter 12). I omit discussion of these forms here, my purpose being only to make clear why we need to distinguish among the three broad aspects of meaning I have mentioned.

IV. Conceptions I think that Frege made a mistake in positing something common beyond Bedeutung that is grasped by the mind of every competent speaker using the same unambiguous linguistic form. A related mistake suffuses the tradition of conceptual analysis in seeking shared “criteria” for the correct application of various terms, criteria taken to be learned, in some mysterious way, when one learns one’s language. On the contrary, the public meaning of a simple referential term typically includes only its stabilizing function and its reference, and since the stabilizing function depends largely on the sentential context the term is in, the public meaning is essentially just reference.6 I intend this sweeping claim to include terms for properties, kinds, stuffs, and so forth, which I will treat here as also being, in a broad sense, referential. The claim will need qualifications, but first I’ll just try to explain it. The idea to be scouted is that for different users to understand the same referential term as having the same meaning requires that their psychological processing be similar in certain ways. Briefly, “Meaning is not in the head.” But I want to launch an attack on Frege’s idea and on conceptual analysis that is more radical and exhaustive than the familiar offenses once launched by Putnam, Burge and Evans. My argument grows out of a view about thought structure, a view about what it is to have a concept of a property, an individual, a kind or a stuff and so forth. It is fully articulated in (Millikan 1984, 2000, 2010). Here I can make only a small sketch. Consider what is involved in being able to recognize, for example, shapes. Think of the variety of proximal visual stimulations to which a given shape may give rise when viewed from various angles, from different distances, under different lighting conditions, through various media such as water or fog, when colored different ways, when partially occluded and

6 See (Millikan 2010 and 1984 chapter 4) on the most general stabilizing function of all referential terms.

98

Ruth Garrett Millikan

so forth. How shape constancy is achieved by the visual system, the capacity to recognize the same shape as the same under a range of conditions, is a problem of enormous complexity that is still being investigated. Similarly, how color constancy, texture constancy, size constancy and distance constancy are achieved are enormously complicated problems. We are also adept at perceiving sounds, especially speech sounds, as the same sound at origin whether near or far, through air or through water, muffled or distorted and so forth. What does seem clear in each of these cases is that no single rule is applied. Different clues are used by the perceptual systems in different circumstances, separately or together. For example, distance is perceived with the help at least of ocular disparity, tension in the focusing muscles, occlusion of one object by another, knowledge of the size of objects viewed, and atmospheric haze. We also recognize distances by touch and stretch using many different parts of the body, and we recognize distances of things that make noises fairly well by ear. And, of course, there is measuring with a ruler or a tape measure or just a string, or measuring as a surveyor does by triangulation, or measuring with an odometer or a micrometer or by the time of the return of light . . . None of these ways of telling distances is infallible nor is any definitional of our concepts of distances. On the other hand, each adds something to our concepts of distances, nor could we have distance concepts at all were we not in command at least of some of these methods of recognition. The situation is similar, if not always so extreme, with our grasp of other perceptual constancies. The perceptual systems do their work in flagrant violation of the ideal set by champions of operational definitions. The more ways the better when it comes to methods of perceiving a property. After all, the ways in which empirical properties affect the various senses through intervening media is a thoroughly empirical matter, a question of natural law, not a matter of logic or definition. That is why neither phenomenalism nor verificationism could survive. Now it is conceivable that all normal persons perceive some constancies, for example depth, the same way, conceivable even that they are genetically programmed rather than perceptually tuned to perceive some of these constancies in standard ways. The issue is under debate. But surely whether one’s perceptual capacities are entirely normal in this regard does not affect what one means by the English words one uses in designating depths or shapes or textures. Being blind in one eye so that one could not perceive depth using ocular disparity would not change what one meant by “near” and “far” nor, indeed, is it sensible to suppose that Helen Keller meant something different by “near” and “far” than you do. I don’t want

On Meaning, Meaning and Meaning

99

to debate about whether there are secondary qualities7, but surely we recognize in perception enough properties and relations that obviously are primary to make the point. There are many different ways of recognizing each of these properties, but none defines either the property or the words that stand for it. Turning to the opposite extreme, consider proper names. Besides having a referent does your name have a definition? What is involved in someone’s understanding who’s meant by your name—say, a child in your family, your child’s teacher, a student of yours, the student’s wife, a reader of your essays, the pharmacist who fills your prescriptions. Do these people all understand who’s meant by your name in the same way? The reasonable answer is that there is no special thing common in the minds of all people who understand your name except, I have argued (Millikan 1984 chapters 4 and 9, 2000 chapter 6, 2010), some practical capacity to reidentify as such, in actual context, the least type that is your name (rather than the name of someone else with “the same name”), so as to recognize when information is being offered about the same person, you, again. Speaking more generally, what it is to have a concept of an individual is, in part, to have an ability to recognize, in one way or another, under at least some circumstances, when one is encountering information concerning that individual, and one recognizable way that one encounters information about a thing, besides through direct perception, is by encountering sentences that contain its name. To defend this position properly, the right characterization of information is needed (Millikan 2004, chapters 3–4; forthcoming) and the right story about perception through language (Millikan 1984 chapter 9, 2000 chapter 6, 2004 chapter 9) as well as a story about abilities that allows for their fallibility (Millikan 2000 chapter 4). But that names of individuals need not be associated either with independent publicly agreed on ways of recognizing these individuals or with agreed on descriptions associated with these individuals in order to do their work is generally accepted, I think. Call the sum of the various ways that you have of recognizing a thing or, what amounts to the same, of recognizing when you are receiving information about a thing, your “conception” of that thing.8 Your conceptions of most common things have many components, for you have many ways 7 The very fact that psychologist’s can make a study of how color constancy is perceived seems to cast doubt on the idea that colors are secondary qualities, anyway in the Lockean sense of that phrase. 8 In (Millikan 1984) I misleadingly called these conceptions “intensions”.

100

Ruth Garrett Millikan

of recognizing these things—no infallible ways, of course, but many fairly reliable ways. Whatever you know about a thing is part of your conception of it too, for whatever you know might help you to identify it, or help prevent you from misidentifying it, under some circumstances. Some components of conceptions are explicit, involving the use of descriptions hence of prior concepts in their application. Other conceptual components are implicit, moving one directly from perceptual experience to an identification of what is perceived. Neither the names of perceivable properties nor the names of individuals are associated, however, with conceptions or conceptual components, either explicit or implicit, that all users must possess in order correctly to understand these names. No specific way of identifying their referents is required. True, in some cases there does exist considerable overlap in the conceptions that most people use in understanding the referent of such a name, for example, since “Mark Twain” was Samuel Clemens’s pen name, a large proportion of people know that Mark Twain was a writer, perhaps even that he wrote Huckleberry Finn. You can usually count on someone having that knowledge if the name “Mark Twain” is in their vocabulary. And the implicit conceptions, the ways, by which we recognize many common properties may be shared among a large majority of adults. But if someone were born with bats ears and could only hear shapes, this would not prevent him from understanding the English words “round” and “square”. I take it that this much is not highly controversial. But the principle can be extended in less obvious ways. Many terms for kinds name kinds that are objective natural units, discovered rather than created by thought and language (Millikan 1984 chapters 16–17, 2000 chapter 2, 2010). These “real” kinds are important subjects for knowledge because there is a reason why the various members of the kind mostly resemble one another in a good number of ways, hence there is a reason why one can learn from observation of one or a few examples of the kind much that is likely to be true of most others. Most single terms designating kinds designate real kinds of this sort (Millikan 2000 chapter 3, 2010). Typically, these kinds not only have many properties, there are also many ways to recognize them. Think how many ways there are of telling that something is copper, or that a dog is present. Do you have to look to tell it’s a lemon? Or that it’s raining? How much of what portion of The First Noel or The Lord’s Prayer do you have to hear to recognize it? To have a valid concept of a real kind one needn’t know the reason for the resemblance of its members, what natural principles hold the kind together. One only needs some fairly reliable ways of reidentifying the kind, the more the better, of course, since most ways one can use only on some occasions. Like

On Meaning, Meaning and Meaning

101

concepts of individuals, concepts of real kinds can be supported by alternative conceptions, alternative methods of recognition, and there are no conceptual components that all users of a real kind’s name must possess in order to understand it (Millikan 1984 chapter 9, 2000 chapters 3 and 5, 2010). The third aspect of meaning, conception, is not then essentially public. It attaches in the first instance to idiolects rather than to public languages. However, there usually is considerable overlap among people’s conceptions corresponding to names of common real kinds. Also, sometimes conceptual components are passed on explicitly from generation to generation, for example, the definitions of certain geometrical figures. One could, after all, “define” a circle, instead, as a closed plane figure with but one side of uniform curvature. And in the case of fictional names, and empty names like “phlogiston” and “witch” that are mistakenly thought to have referents, there is no public meaning beyond certain traditional explicit conceptual components, traditional descriptions, passed down from person to person. There is no more to public meaning in these cases than public conception, indeed, public conception that is highly subject to drift. Santa Clause acquired red and white attire and reindeer rather late in his career while phlogiston and witches took on different diagnostic properties over time in the eyes of different investigators. Traditional descriptions associated with empty terms fail to reach anything real, hence do not correspond to real abilities to identify. But having empirical concepts, having thoughts of objects, properties and so forth, essentially involves abilities to identify, from which it seems to follow that empty terms do not express real concepts. This brings us to the externalist core of this chapter on meaning. The claim is that the meaning of an empirical term is, in the first instance, its referring to something, and only in the second instance, ways one has of identifying this thing through various of its manifestations. Wittgenstein was right, after all, that the only check on whether we mean the same by our words is agreement in judgments, but agreement in judgments proves nothing about agreement in the methods of identifying used in making those judgments. Now nothing inside the head or mind determines, in and of itself, whether one’s dispositions to react to sensory stimulations with would-be thoughts manifest real abilities to identify anything, or whether one’s inference dispositions whose job is to connect prior concepts with would-be thoughts (ones explicit conceptions) are actually doing this job. If the would-be thought is governed by explicit conceptions, and if the prior concepts in the descriptions used in these

102

Ruth Garrett Millikan

conceptions are not themselves empty, then there is a legitimate, though secondary, sense in which even a name connected with an empty thought can have a meaning.—because components of its conception have meanings. Indeed, if the term is public, it will have conceptual components that are both explicit and traditional, having been handed down from speaker to speaker, hence it will have a sort of public conception/meaning. But suppose there were an empty concept that had only implicit conceptual components, that was not anchored by any prior nonempty concepts. Such a term would have no more claim to membership in the realm of the intentional or the semantic than a sneeze. It would merely be a quirkish regular response to certain sensory stimulations, resulting, presumably, from the faulty operation of systems designed to design genuine concepts, genuine thoughts, through experience but that had failed in that task. As outlined above, my proposal on meaning has a huge hole in it however. A crucial task incumbent on any advocate of meaning externalism is to explain how we acquire evidence through experience that our concepts are not empty, that they are anchored externally in what is objectively real. The externalist is obliged to accompany her claims about the ontology of meaning with a plausible epistemology of adequacy in empirical concepts. She must construct an epistemology of meaning to support her claims in the philosophy of mind. I consider this an urgent matter, though one sorely neglected in the literature on externalism. The epistemology of concepts, or of meaning, is the subject of (Millikan 1984 chapters 18–19, 2000 Chapter 7, 2004 chapter 19).

V. Replacing intensions and Fregean senses As said at the start, none of my trio of meanings corresponds at all well either to any traditional notion of intension or to any Frege-like notion of sense. Both these notions mistakenly assume that a grasp of certain ways of identifying or certain properties by which a thing may be identified must be shared by users of any public term that refers to it, whereas I claim that any such grasp is, in the first instance, a private matter. But this claim needs to be defended with an alternative explanation of the phenomena that lead to the postulation of intensions and Fregean senses. I take it that there are three central classical arguments for something like intensions or senses, from the informativeness of sentences asserting identity, from the need to analyze statements asserting the existence of things, and from the

On Meaning, Meaning and Meaning

103

behavior of referential terms in intensional contexts. I will discuss each in turn.9 First, identity statements. I have agreed with Strawson that the stabilizing function of an identity sentence, “A is B” is to encourage the hearer to merge under a single concept all of the information she has accumulated under the concept she associates with the word “A” with the information she has accumulated under the concept she associates with “B.” Stating this a bit more accurately now, an identity statement that serves its stabilizing function joins the conception the hearer has associated with “A” to the conception she has associated with “B” so that these now correctly govern the same concept. Rather than inducing beliefs—analogizing beliefs to mental sentences—it alters conceptions, ways of identifying (Millikan 2000 chapter 12). Thus for any hearer who associates a different conception with “A” than with “B,” the effect of a true identity statement “A is B” obviously is different from that of “A is A.” This can be true and important even if no two hearers who react in the stabilizing way to “A is B” happen to share their conceptions associated with “A” or with “B.” Second, existence statements. Statements of the form “A doesn’t exist,” I have claimed, induce a hearer to disengage his concept associated with “A” from ordinary referential uses, eliminating it entirely or reserving it only for pretend uses. Statements asserting the existence of A reverse this effect. To engage or disengage a concept is the same as to engage or disengage the conception that governs that concept. The forms “A exists” and “A doesn’t exist” can serve these functions regardless of how diverse listeners are in the conceptions they associate with the name “A.” But as noted earlier, there is usually a good deal of overlap in conceptions for names that are very common, and names that have no referents, unless passed on by sheer repetition, can only be passed on by description, so they are especially likely to have conceptions attached that are largely public (though perhaps shifting). Third, intensional contexts. A well known way of extensionalizing intensional contexts was suggested by Davidson in “On Saying That” (1968–9). His idea was that a sentence such as “Galileo said that the earth moves” is true just in case uttering the words inside the that-clause of this sentence makes the speaker and Galileo into “same sayers,” people who have

9 A fourth argument might appeal to a need for reference to intensions or “narrow meanings” in givng psychological explanations. On this, see (Millikan 2006).

104

Ruth Garrett Millikan

uttered words with the same import. I have adopted a similar view but generalized it, claiming that when one representation is held up or put on display in order to show what another representation is like, the kind of similarity intended may concern any aspect of meaning, or may even concern some aspect of the vehicle of the displayed expression (Millikan 1884 chapter 13, 2004 chapter 7). Giving an example of the latter, consider the sentence, “John kept insisting that he didn’t see a woodchuck but instead a groundhog.” It is clear here that the similarity intended must concern the very words “groundhog” and “woodchuck” since these two are names for the same. Intentional attitude contexts yield to a similar analysis. In “John firmly believes that it was not a woodchuck that he saw but a groundhog,” again something about the words “woodchuck” and “groundhog” is surely at stake. One way to understand this is to assume that an embedded sentence displayed in an intentional attitude context refers to an intentional attitude that is relevantly like one it is its own stabilizing function to produce. The sentence “It was not a woodchuck but a groundhog that John saw,” if it were to serve its stabilizing function with John as a hearer, would produce just the mental state John is in, right down to the last conceptual component, his full conception. That the message concerns not merely some proposition associated with John’s mental state but also his conceptions, including the very words through which he would try to recognize information coming in about the subjects of his thought, is clear on the ( pragmatic) assumption that John does not think a thought that shows, from the inside, that it is contradictory. This reading also nicely accommodates the fact that definite descriptions appearing inside intentional attitude contexts are sometimes read as attributing the description to the thinker as part of his conception but sometimes as attributing to him only a thought of the description’s referent. Thus “Ralph thought that our venerable dean was a spy” might or might not imply that Ralph knew that the man he thought was a spy was our dean. This is entirely natural if the intentional attitude description works by displaying a sentence whose function is to produce an attitude like the one being attributed, and if definite descriptions have alternative stabilizing functions corresponding to Donnellan’s distinction, as discussed above.10

10 To accord with this treatment, modal contexts need to be understood as contexts in which representations, rather than possible situations or worlds, are the basic subject matter.

On Meaning, Meaning and Meaning

105

Thus the three aspects of meaning that I have discussed are sufficient, I believe, to account for those properties of natural language traditionally associated with meaning.

References Boër, S.E. and W.G. Lycan 1986 Knowing Who (Cambridge MA: MIT Press) Bach, K. and R. M. Harnish 1979 Linguistic Communication and Speech Acts (Cambridge MA: The MIT Press). Davidson D 1968–9 “On Saying That” Sythese 19:130–146. Reprinted in A.P. Martinich ed. The Philosophy of Language, fourth edition (Oxford: Oxford University Press) Donnellan K 1966 “Reference and Definite Description” Philosophical Review 75: 281–304 Gilbert, M. 1983 “Agreements, Conventions, and Language,” Synthese 54:375–407. Gilbert, M. 1989/1992 OnSsocial Facts (Princeton: Princeton University Press 1992, reprinted from Routledge 1989 edition). Grice, H. P. 1968, “Utterer’s Meaning, Sentence-Meaning and Word-Meaning” In Studies in the Way of Words (Cambridge MA: Harvard University press). Lewis, D. 1969 Convention; A Philosophical Study (Oxford: Basil Blackwell). Lewis, D. 1975 “Languages and Language”. In K. Gunderson, ed., Language, Mind and Knowledge (Minneapolis: University of Minnesota Press). Reprinted in H. Geirsson and M. Losonsky eds., Readings in Language and Mind (Oxford: Blackwell) 1996, 134–155. Millikan, R.G. 1984 Language, Thought and Other Biological Categories (Cambridge MA: The MIT Press). Millikan, R.G. 1998 “Proper Function and Convention in Speech Acts,” in ed. L.E. Hahn, The Philosophy of Peter F. Strawson, The Library of Living Philosophers (LaSalle IL: Open Court) 25–43. Millkina, R.G. 2000 On Clear and Confused Ideas (Cambridge: Cambridge University Press). Millikan, R.G. 2001 “Purposes and Cross-purposes: On the Evolution of Language and Languages,” The Monist 84.3 (Special issue The Epidemiology of Ideas, ed. Dan Sperber) 392–416. Millikan, R.G. 2006 “Reply to Lalumera” SWIF Philosophy of Mind Review Vol.5 No.2, http://www.swif.uniba.it/lei/mind/swifpmr.htm Millikan, R.G. 2010 “On Knowing the Meaning,” Mind 119, No. 473. 43–81. Millikan, R,G. forthcoming, “Natural Information, Intentional Signs and Animal Communication” In Animal Communication Theory: Information and Influence, Ulric Stegmann ed. (Cambridge: Cambridge University Press). Recanati, F. 1987 Meaning and force; the pragmatics of performative utterances (Cambridge: Cambridge University Press). Searle, J. 1969. Speech acts; an essay in the philosophy of language. Cambridge: Cambridge University Press.

106

Ruth Garrett Millikan

Sellars, W. 1956. “Empiricism and the philosophy of mind.” In Minnesota studies in the philosophy of science, vol. I, ed. K. Gunderson, 253–329. Minneapolis: University of Minnesota Press. Reprinted in (Sellars 1963). Sellars, W. 1963 Science, Perception and Reality (New York: The Humanities Press). Schiffer, S. 1972. Meaning. Oxford: Clarendon. Strawson. P.F. 1974 Subject and Predicate in Logic and Grammar. London: Methuen and Co.

On Referents and Reference Fixing1 Howard Wettstein Abstract: The paper consists of two notes, one in which I challenge the relational conception of reference. The conception is problematic if what it requires is the existence of a referent. For referential expressions, I argue, sometimes lack such a relatum; and in other cases while the term refers to existent aspects of the world, it’s difficult to know what we are talking about. “Pegasus” and “Zeus” are examples of the former; “water” is an example of the latter. The second note concerns the idea of reference-fixing, one of the exciting ideas in Kripke’s ground breaking Naming and Necessity. I discuss Frege’s anticipation of the idea in his pre-sense/reference Begriffsschrift, and explore different ways of characterizing the idea, one in Kripke’s work, the other in Kaplan’s. The paper draws attention to what I argue is the Fregean lineage of reference-fixing. Not only because Frege makes early use of it. The Fregean inspiration is evident in Kaplan’s approach to indexicals: Kaplan associates descriptive reference-fixing “characters” with modes of presentation, and then uses the characters to explain the Frege puzzle in Frege’s way. Frege’s influence also seems at work in Kripke’s view of name introduction. While Kripke denies a role for descriptions in the use of proper names, he maintains that the paradigm for name introduction is introduction by description. Even Kripke, a strong voice for contemporary Millianism, makes room for a descriptional connection between a name and the thing named.

On the idea of a referent Wittgenstein speaks of the fog that surrounds the workings of language. Our ways of thinking about reference contribute generously to the fog. While a full discussion would constitute a book-length project, my aims here are quite limited. I want to have a look at the idea that reference is a relation between a piece of language and a piece of reality. The idea might seem unexceptional and unexceptionable; names, for example, name 1 I am grateful to Richard Mendelsohn and Andrea Bianchi for extremely helpful comments on previous drafts.

108

Howard Wettstein

things, and “relation” seems just right. But there is casual relation-talk, and then more serious talk of a genuine relation, one that requires the existence, in every case, of a referent. Referential expressions, I’ll argue, sometimes lack such a relatum, and other times—when in some straightforward way there is such a thing—it’s quite difficult to know what we are talking about. Reference as a relation becomes especially interesting in the context of the widely accepted idea, that reference constitutes a master key to the understanding of the link between language (or even representations generally) and reality. It would be interesting to trace the history of attempts to take one sort of expression or another as the model for “reference as the master key.” Fregeans have found it tempting to focus on the definite description; contemporary anti-Fregeans (with whom I have much sympathy) have gone with the proper name, some with the indexical expression. And some have resisted a single model, still however seeing reference as involving a relation between a piece of language and a piece of reality. Proper names both make the relational picture intuitive—here’s the name, “David Kaplan,” and here’s the referent, DK himself—and dramatize the difficulties. “Pegasus,” certainly to non-philosophers a perfectly good name, one that people use to say things, lacks a referent.2 What then of the genuine relation? I raise this issue not at all in the spirit of any sort of anti-realism about reference; “David Kaplan” still refers to DK. But it does reveal that things do not go as smoothly as we might have supposed for the relational picture, even for proper names. Let’s move on to natural kind terms like “water.” The term seems to function like a name. But what of its referent? Unlike the case of “David Kaplan,” the matter of its referent is certainly not straightforward. Quine says that the referent of “water” is a scattered object. That’s a weird bird. Quine, I’m thinking, wants a referent for the term ‘water’ and wants

2 Such “empty names” raise a number of well-known problems, as do names in fiction. Prominent for Millians like me is the problem of the significance, the meaningfulness of reference-less names. This is a problem, or at least certainly appears to be, since Millians are fond of saying things like “the sole semantic significance of a name is its reference.” At the same time, as I discuss in Chapter 7 of The Magic Prism: An Essay in the Philosophy of Language (Oxford University Press, 2004), empty and fictional names had better be meaningful. We certainly use them in significant sentences. (See my criticism there of “pragmatic approaches” that deny the semantic significance of empty and fictional names.)

On Referents and Reference Fixing

109

something more concrete, less abstract, than a stuff. Notice that in practical terms there is no puzzle about the functioning of terms like ‘water’. It’s not as if we have trouble applying the term. But when one asks focused questions about the identity of the referent, something goes haywire. ‘Horse’ is another problematic sort of “natural kind term.” This one doesn’t look like a name at all. One might suppose, and I’ve heard it said, that the term names a species; we certainly can speak of “the species, horse.” But the subject term there is not “horse” but rather “the species, horse.” In such contexts the term “horse” identifies a species without naming it. Indeed one cannot begin a sentence about the species with simply “horse,” as in the incoherent, perhaps ungrammatical, “Horse is a species. . . .” It seems important that “horse,” unlike proper names, takes plurals. Indeed, “horses” seems more like a device of reference than “horse,” although not singular reference. And again while “reference” seems like it ought to be an appropriate idea here, the referent is not some chunk of the world. One begins to wonder about whether “natural kind term” identifies a natural kind of expression, and about what exactly a natural kind is supposed to be. I raise these questions as a way of generating interest in another picture. As Wittgenstein emphasized, the connections between language and the world are much more multifarious than we are apt to suppose. To look—at least to look through a philosophical lens—for the referent of a referential expression, as simple, as trivial, as this sounds, often leads to a dead end. Not with proper names of course—at least the happy ones that stand for real things in the real world. But such expressions are, I want to suggest, a special case; hardly paradigmatic. I’m not sure there is a paradigm; perhaps illumination comes, as Wittgenstein supposed, from seeing a variety of cases and developing a sense of the family they constitute. Consider an analogy concerning the concept of truth. Here’s our picture: “The cat is on the mat” is true just in case the cat is on the mat. Everything, again, is out in the open. A propositional representation is true just in case it matches, corresponds to, what it purports to represent. But then we come to contexts that don’t quite fit, like negative existentials and fictional contexts. The usual way has been to make the paradigm fit, using more or less force as needed. If the student speaks truly when she says that Hamlet is a prince, then we need a referent for “Hamlet”; Kripke, et al., thus posit an abstract entity engendered by the act of writing the fiction. And so on. I prefer a less orthodox option: truth only sometimes conforms to the correspondence paradigm, which is in fact a special case of truth. In fictional

110

Howard Wettstein

contexts, the explanation of truth does not require positing a fictional abstract entity. A remark about Hamlet can be true, correct, just because it gets the story right. Perhaps there is a vague and general idea associated with our use of the concept of truth—crudely put, it’s a matter of getting things right. This is not to say that one could not further articulate and illuminate the more general idea. But it won’t be simple; and the result will surely not be as simple as the ideas with which we have been working.3 Let me return from this digression about truth to the topic of reference. In what follows, I’ll sketch an alternative model. I don’t know the limits of my model; it’s not a new paradigm that’s supposed to cover everything. Consider our talk of pain. Pain talk is surely referential; we are speaking of things that go on with us. But the referent of “pain” is another matter. First, there is the question of whether the term is something like a family resemblance term. Is there something like pain-in-general, the sort of thing that might turn out to be identified with C-fibers firing? Or are there pains and pains? Perhaps pain resists a uniform sort of physical treatment. Leaving that question aside and focusing on a single pain—this pain in my leg—isn’t there something a little strange about the question of where and indeed what it is? In one sense, “where and what” is apparent to all pain-sufferers; we know where it hurts, and what it is. But that’s not the “where and what” of interest to philosophy. What seems strange is the idea that it—my leg hurting—might turn out to be an event in my brain, or maybe in some immaterial substance associated with my body.4 The putative referent is puzzling in other ways: Does it have boundaries? What if it’s sort of there but elusive? Is it really a something? One is (almost) tempted to resort to Wittgenstein’s non-answer to a parallel question about universals: it’s not a something, which is not to say it’s nothing. Wittgenstein sympathizers sometimes emphasize the role of paintalk as expressive of pain—as opposed to referential about pain. But it’s worth pointing out that these functions—expressive, referential—do not preclude one another. I find it helpful to think developmentally here,

3 “Truth” also seems subtle and complicated with respect to reports of speech, belief, and the like, as I suggest in Chapters 8 and 9 of The Magic Prism. 4 I surely don’t mean to deny the utility of “identifying” the pain with some CNS event, process or whatever. “Identifications” of that sort can be of use to us even when they involve no claims about real identity, my interest here. Cf. Paul Benacceraf, “What Numbers Could Not Be,” (Philosophical Review, 1965, Vol 74, pp. 47–73) for the distinction between questions of real identity and identifications that have utility.

On Referents and Reference Fixing

111

even in the armchair way for which we are notorious.5 I like the Wittgensteinian idea that expressive pain talk begins as a socialized replacement for more primitive pain behavior; thus “ow!” in place of writhing or an even less articulate groan. At a later stage, “Ow!,” in a matter of short geological time, begets words like “owie”6 that work their way into sentence frames and become referential. Our pain vocabulary is referential and so it’s natural to wonder about the referent, the item picked out by “this pain in my leg.” But my developmental sketch aims to make us pause over the idea of the item picked out. At one moment—or epoch—we are crying out. Later this behavior gives way to words that express our hurting. Still later these same expressions or their progeny work themselves into sentence frames; now we can talk about our hurts. But where exactly do discrete pain-entities—subject to theoretical identification—enter the story? One might, contrary to what I’m suggesting, suppose that the crying out was responsive to such discrete pain-entities. And further that with the birth of referential pain vocabulary, speakers wake up, as it were, to the fact of such entities. I am not providing anything like a refutation of such an approach idea. Just an alternative, one that I hope the reader will find natural. Let’s return to natural kind terms, “water” and “horse.” The direction I’m pursuing is easier for “horse” than for “water.” My suggestion is that we not move too quickly from talk of language being referential to questions about the identity of the thing picked out. In the case of “water,” to which I’ll return, the term at least appears to have a referent, namely water. But with “horse,” things seem different. Here we really need not think of it as picking out an item to think of it as referential, as applying to things in the real world. “Water,” as I say, is more difficult for my picture, since the term does, as we say, refer to water. But to stay for a moment with linguistic practice, it will be very difficult and involve serious strain to make the argument that linguistic practice—the ways we talk of water—allows us to discriminate between the stuff view and the scattered object view. Probably these philosophical views each pick up on aspects of the way we speak of water. Still, there may be some sort of consideration—philosophical, scientific, or whatever—that reveals the superiority of some way of thinking about the 5 The following remarks on the development of pain vocabulary are adapted from my paper, “Terra Firma” (The Monist, October 1995, Vol. 78, Number 4, pp. 425–446). 6 I hope this children’s term for pain is recognizable.

112

Howard Wettstein

referent. But perhaps not. The suggestion I’ve been making about reference leaves the referential character of “water” intact, the vicissitudes of these philosophical views aside. Wittgenstein, in his Brown Book, discusses what he refers to as names of directions, “North,” etc. He suggests, characteristically, that we should attend to the analogies with proper names without neglecting the differences. I don’t know that such names of directions should be thought of as referential; nor am I confident that they should not be. The category of referential is no doubt not sharply bounded.7 My point here was that attention even to a small sample of our variegated referential vocabulary suggests that the proper name picture of reference—“David Kaplan” and DK—does not represent anything like the general case. This is not in the service of some sort of linguistic idealism or anti-realism about reference. Nor is my proposal in any serious way anti-theory, except in the sense that we owe to Aristotle that the degree of precision in a theoretical proposal needs to be sensitive to the subject matter.

Fixing the reference Frege, Kaplan, and Kripke: Three treatments of the idea Naming and Necessity broke new ground in a number of philosophic domains. Reference-fixing was just one of Kripke’s exciting ideas: it is one thing for a definite description to be synonymous with a name, to formulate its meaning; quite a different thing is a description’s merely determining or fixing the name’s reference. To be clear about the distinction, one wants to hear more about synonymy; and one wants to know the cash value, as it were, of the difference between giving the meaning and fixing the reference. Two avenues have been discussed for the understanding of the distinction, one modal, the other concerning propositions. Kripke emphasizes the first of these; Kaplan, the second. Kripke’s modal characterization: if we are thinking in terms of synonymy, then the description formulates the conceptual content of the name. Accordingly, “Aristotle taught Alexander” would be a tautology 7 Another interesting case is that of the indexical expression,“now.” Kaplan’s widely accepted suggestion that “now” refers to a time raises all sorts of issues. One wants to know more about the referent, for example.

On Referents and Reference Fixing

113

and so a necessary truth. By contrast, if the description merely fixes the reference of the name, the idea is that the description, as it were, searches the world, finds its satisfier and attaches the name to it. But the referent itself, the person who really did teach Alexander, might well not have gone into education; someone else might have taught Alexander. So “Aristotle taught Alexander” is a contingent truth. The other way to get at the distinction is by way of the notion of proposition. Proposition-talk is not Kripke’s preferred idiom; nevertheless, for better or worse, it’s become more or less standard. Using propositiontalk, one can say that if the description is synonymous with the name, then the proposition expressed by “Aristotle was wise” contains as its subject constituent the conceptual content of the description, “the teacher of Alexander.” However, if the description merely fixes the reference of the name, the proposition expressed is—Kaplan and others argue—the singular proposition that contains the man, Aristotle, in the subject position. The idea that a description might merely fix the reference of an expression and not formulate its meaning has the appearance of a radical departure from Frege. My idea, on the contrary, is that the idea of reference fixing represents a lingering Fregean element in the thought of Kripke, Kaplan, and those of us who have been influenced by them. Interestingly, Frege himself flirts with the distinction. In the Begriffsschrift, Frege identifies (in a way that anticipates Russell and Kaplan) the notion of content with reference: the content of a name is the thing named. Nevertheless Frege supposes that names apply to things in virtue of descriptively given “ways”8 in which the things are presented. Put these two ideas together—descriptive ways of determining reference and content as reference—and you have arrived virtually at Kaplan’s approach: Reference is fixed by the “ways of determining reference” but the proposition contains the referent, the object. This is of course very different than Frege’s mature sense-reference view, a view that breaks with the Begriffsschrift’s content-as-referent doctrine. I want to dwell for a moment on Kaplan’s use of reference-fixing (specifically for indexicals) since it is interesting in its own right and importantly different than Kripke’s. In “Demonstratives,” Kaplan says that indexicals, unlike proper names, have descriptive meanings, formulated by Kaplan’s “character rules.” “I,” for example, means “the agent of the 8 As with Frege’s later notion of sense, thinking of senses, and ways of determining reference, as descriptive seems to capture what Frege has in mind and yet does not fit with everything he says.

114

Howard Wettstein

context,” roughly, the speaker. So far, Kaplan’s characters sound just like Frege’s senses. But there is a difference: the descriptions that formulate Kaplan’s characters are improper, they fail to denote. “The speaker,” fails to apply uniquely to anything—unless of course one is speaking of a specific context of utterance.9 And relativize them to context is exactly what Kaplan does; characters determine reference relative to context. This is Frege updated for the special case of indexicals.10 Kripke’s use of reference fixing might seem less Fregean. A Kripkean reference determining description should not be thought of as formulating the meaning of a name. One of Kripke’s examples of reference-fixing by description is the name “Neptune,” introduced by the description, “the cause of perturbations in the orbit of Uranus.” Kripke’s point is not that when members of the linguistic community use the name ”Neptune,” they somehow fix its reference in this way. If that were Kripke’s idea, it would be parallel to Kaplan’s. The relevant description would then be like a character rule for the name. Instead, for Kripke the description is germane only to the question of how the name was originally introduced; “Neptune” was introduced into the language by way of the description in question. When the name is later passed along, this reference-fixing information is not necessarily or typically passed along. Others down the chain pick up the name—that is, they achieve competence with it—without necessarily learning the reference-fixing information. So reference-fixing descriptions come in only at the name-introduction phase. That Kaplan, by contrast, is really thinking of the reference fixing descriptions as meanings is also evident in Kaplan’s use of characters to resolve the Frege puzzle for the special case of indexicals. Kaplan explains the informativeness of, for example, “I am he,” in terms of the different characters of “I” and “he”; these characters capture the respective modes of presentation under which speakers conceptualize the referent. Characters clearly have nothing special to do with the introduction of expressions, but with the continuing practice. Indeed in the case of indexicals, introduction of the expression is not what’s at issue. And similarly for Frege’s early view. It is concerned with names, not indexicals. But the idea of way of determining reference is not restricted to

9 And even then there are issues. Relativize “the speaker” to a conversational context and there may be more than one speaker. Such issues are beyond my scope here. 10 This is Kaplan’s Fregean side. His anti-Fregean side is represented by his Russellinspired view of propositions as containing objects.

On Referents and Reference Fixing

115

name introduction. These precursors of Frege’s modes of presentation are meanings, even if only in the attenuated sense of continuing reference-fixers,11 even if they are not elements of the propositional content.

The Fregean heritage Kripke suggests in Lecture 2 that name introduction by reference-fixing description is the paradigm. Even when a name is introduced by some ostensive gesture, a reference-fixing description is in fact operative. This view of Kripke’s has always seemed to me strange, out of character for Kripke. Part of what I took to be the big lesson of the work of people like Kaplan, Kripke, Donnellan, Marcus, and Putnam was a certain skepticism about traditional ideas about the role of descriptions: reference by names is direct, conceptually unmediated, and the contrary Frege-inspired way overintellectualizes language. The spirit of many of Kripke’s wonderful examples also seemed radically non-descriptivist. Take Madagascar. Kripke’s powerful response to Evans and other critics was the idea that the (current) reference “Madagascar” to an island was born in a mistake, a mere speaker’s reference. People were misusing this name that really was a name of a part of the mainland. But the mistake became socialized and generalized—that’s how the name became a name of the island. At least on the face of it, there are no reference-fixing descriptions launching this new use of the name. Perhaps one could find a way to push a descriptivist analysis here as well; but it does not seem the most natural way to proceed. Indeed I would have thought that for Kripke, it’s not in general semantically important how a name gets going. There might indeed be a variety of ways. One need not deny that there are cases in which the name begins with an intellectual act (as if in the privacy of one’s study—to paraphrase Kripke in another context) of reference fixing by description. But what’s important is that the name-user just uses the name as Mill suggested, without descriptive meaning. Descriptions are irrelevant, one might say, to the semantics of names. So it seemed strange that Kripke would insist that name-introduction be by description. Let’s think for a moment about baby-naming and the role or lack of it for descriptions. At one point in his discussion, Kripke mentions a 11 By “continuing reference-fixers” I mean that competence with a name involves associating the name with such a reference-fixer.

116

Howard Wettstein

baptismal ceremony. In my own ( Jewish) religious tradition—Kripke’s is the same—a name is not used until a formal act of naming occurs. But this is quite a distinctive practice and hardly the way it goes typically for people. Parents may muse for quite a while—at the expense of their friends and relatives—about names, and then the baby is born and they just start using the name; no baptism; no act of giving the name, a fortiori no intellectual act of giving a name by means of a description. When we name a baby, we may already know a great deal about the child. But unless one has descriptivist commitments, one will not naturally see the name giving as involving a reference-fixing description. The name is applied to the baby directly. I think the examples by means of which Kripke generates intuitions about reference fixing are worth attention. Kripke’s Neptune example is quite an oddity. Isn’t it unusual for us name givers to first have a theoretical idea—like “cause of the perturbations in the orbits of Uranus”—and then coin a name to stand for the thing? That’s about as typical as definition by stipulation. Another of Kripke’s examples is the meter. Notice that “one meter” is hardly a name. Any more than “two meters.” Words for lengths— perhaps this generalizes to words for measures—like “meter” are not proper names. And concerning the standard meter bar, who would suppose that the assignment of the bar to the measurement in question involves reference-fixing, indeed by description? It seems important that Kripke’s two real-world examples of reference fixing by description— Neptune and the meter—are atypical. I’ve been trying to draw attention to what I see as the Fregean lineage of the idea of reference-fixing. Not only because Frege make early use of it. I sense the Fregean inspiration in Kaplan’s approach: associating his reference-fixing characters with modes of presentation, and then using the characters to explain the Frege puzzle in Frege’s way. I also see it in Kripke’s intuition that somehow descriptions have to be involved, if not in typical name use then in name introduction. The idea is that at some level the connection between a name and the thing named is descriptional, conceptual.

Afterthought: Kripke’s chains of communication Kripke says, in Lecture 2, that when one hears another using a name and starts using it himself, the current speaker intends to be using it with the

On Referents and Reference Fixing

117

reference of the prior user. Kripke’s remark is subject to interpretation. It might mean—if one pushed the text in a descriptional direction—that one fixed the reference of say, “Jones,” by the description “the person who was spoken of by the person from whom I learned the name, ‘Jones’.” I don’t believe that this is Kripke’s intent. Much of the spirit of the first two lectures suggests otherwise. Indeed on page 91 he strongly suggests quite the opposite, in the paragraph that begins, “But that’s not what most of us do.” What we don’t do, Kripke is saying, is to assign names by intellectual acts. What then of the referential intention of which Kripke speaks? I take Kripke’s remark to be of a piece with his comment in “Speaker Reference and Semantic Reference” in which he distinguishes general from specific referential intentions. The most fruitful and plausible reading of that latter passage I owe to David Kaplan. He suggested that what Kripke means by general intention is simply the intention to use the name conventionally. That’s not to say that the speaker performs some mental ceremony in first using the name, but just that in general the speaker means to be using language conventionally. To return to Naming and Necessity Kripke means, I’m supposing, that when I pick up a name, of course I mean to use it as it’s been used. Period. My reading of Kripke—both generally and with respect to this remark about intention—admittedly emphasizes passages in Naming and Necessity and deemphasizes others. The work, after all, is ground-breaking and highly suggestive. And like other seminal works that are prior to developed theory, one sometimes struggles to see how, and indeed if, the work’s insights all cohere. The core idea of the work is clearly Millian, as Kripke emphasizes: names are non-connotative. Kripke’s descriptivist remarks about name-introduction seem (to me at least) to come from another place. But Kripke still goes a long way with Mill. For the name user down the chain, the name is fully Millian—no descriptions or modes of presentation, not even in a reference-fixing capacity. Indeed, the later name user does not fix reference; she becomes competent in the use of a name that already has a referent.12 12 This brings me to another point of some importance about what I take to be, and hope is, Kripke’s picture. The question is one of the character of the chain of communication. Here I’m making use of some thoughts of David Kaplan and Joseph Almog. One might suppose that the chain of communication story of Kripke’s—he denies in conversation that he meant to emphasize “causal”—is his externalist alternative to the internalism of Frege; an externalist alternative to

118

Howard Wettstein

I said above that my picture of Kripke’s view is based upon emphasizing some passages and deemphasizing others. Here I need to mention a passage that is least amenable to my sort of treatment, “least” meaning “not at all.” I tend to joke that Lecture 3 is apocryphal. I hope Kripke finds this amusing; it’s meant to express reverence for the first two lectures and my sense that something important changed in the third. Here’s what I mean: Kripke’s treatment of natural kind words suggests that they are name-like but that their references are fixed by description. And indeed the role he gives to these reference-fixing descriptions is not limited to anything like name-introduction. Perhaps I am wrong here, but it seems that our concepts of the relevant kinds somehow involve the reference-fixing conceptual material. From where I sit, that makes such expressions very much unlike names; they are rather some sort of name-description hybrids. In the course of this discussion, Kripke remarks on an important difference between proper names and such natural kind words: when someone learns a name, we don’t care much how he fixes it’s reference, whereas with natural kind expressions, it really matters. This is trouble for my picture. For it suggests that indeed name users do indeed fix the reference of names, and in different ways, presumably by different descriptions. Frege’s internalist story of what determines reference. The semantics of the name then crucially involved the chain; the links are links, as it were, in the semantics. And one might suppose—there may be alternatives—that there is a question about what determines the reference at each link. It’s at least thinkable that at each link, the new user fixes reference, etc. But there is an alternative I prefer. When one learns the word “table,” for example, it is implausible to suppose that the transmission of this word from one to another constitutes something internal to the semantics of the name. One is rather passing along something that is semantically whole and complete. This is of course not to deny that one can change the meaning of the word, and then something really new may happen. But barring that, the “chain” is not semantically relevant. The idea is that names are like that. At some point the name “Aristotle” entered our practice and then its semantics was finished. Passing it from one to another is like passing the salt; like passing “table” from one to another. The chain is of interest in various ways, but it’s not an externalist link of name to referent. That link was whole and complete; if not there was nothing to pass along. If this picture has merit, then again we see that it’s a mistake to see each user as fixing reference. This would be like reinventing the wheel. It’s there already. Question: if Kripke’s is not an externalist response—that is an explanation of the name-referent link—how does one respond to the Fregean challenge for an explanation of the link between name and reference? I’m inclined to suppose that the question is ill-framed and that the challenge withers upon analysis. I argue this in Chapter 5 of The Magic Prism.

II Truth-theoretic Semantics

On Some Examples of Chomsky’s Graeme Forbes Abstract: Chomsky (1975) gave some examples which he took to be problematic for compositional, truth-conditional semantics. One example is the contrasting pair (i) Poems are written by fools like Smith, and (ii) Mountains are climbed by fools like Jones. (i), he notes, makes a claim about all poems, while (ii) does not make a claim about all mountains. Chomsky doubts that this difference can be traced to structure. In my paper, I argue that we can account for this contrast (and some others Chomsky notes) within compositional truth-conditional semantics, provided, first, that the semantics is a neoDavidsonian event-semantics, and second, that we combine it with a theory of focus. The natural reading of (i) is then “every poem has a fool like Smith as agent of its writing” and of (ii), “every climbing of a mountain has a fool like Jones as agent”. My thesis is that event semantics provides the materials for analyzing the contrast between the likely meanings of the two statements, and the theory of focus explains why we read one sentence one way and the other another.

I take the title of this volume, Prospects for Meaning, to subsume the topic of prospects for the theory of meaning. Specifically, I will be concerned with whether compositional, truth-conditional semantics embodies the central part of the explanation of how language functions as a vehicle of communication. My own view is that it does, provided the semantics in question is, in the usual jargon, neo-Davidsonian.1 That is to say, typical assertions are taken to characterize states, processes, or events (henceforth just ‘events’), with the various parts of speech identifying specific features or constituents of events in a fairly constrained way. In this paper I will consider some examples of Chomsky’s which prima facie present difficulties for the compositional, truth-conditional approach, but I will argue that instead, the examples bring out some advantages of the neo-Davidsonian framework. I begin with an overview of how this framework may be implemented within a type-logical semantics. 1 After (Davidson 1967). The canonical source for the ‘neo’ variation is (Parsons 1990).

122

Graeme Forbes

1. Type-logical event-semantics One particularly appealing model of the composition of meanings is the function-argument model: when two meanings compose, one meaning is a function which takes the other as input, and the composed meaning is the result, or output, of this function-application. If two meanings cannot be composed, that may be because neither is a function whose domain of application includes the other. The standard realization of this idea is in the simple theory of types, in which basic types of meaning, i (individual) and b (truth-value, ‘boolean’), are posited; all other types of meaning are built up recursively by the rule that if t1 and t2 are types of meaning, there is a type of meaning denoted t1 ѻ t2, or t1t2 for short, namely, the type of functions from meanings of type t1 to meanings of type t2. In categorial grammar, expressions are assigned syntactic categories that determine their potential for type-theoretic semantic composition. For example, (1a) below may have the semantic representation (1b), reflecting the syntactic and semantic classifications of (1a)’s constituent words and phrases given in (1c): (1) a. Tom chased Jerry b. (chased( jerry))(tom) c. ‘Tom’: NP, i; ‘chased’: (NP\S)/(NP), i(ib); ‘Jerry’: NP, i; ‘chased Jerry’: NP\S, ib; ‘Tom chased Jerry’: S, b (1b) is in a type-theoretic language in which the order of application of functions to arguments is unambiguously displayed. The function chased applies to the individual jerry and produces the function chased( jerry), which applies to the individual tom to produce a truth-value. On the syntactic side, ‘chased’ merges with an appropriate sister expression on its right (indicated by ‘/’), that is, an NP, and forms the VP ‘chased Jerry’; but ‘VP’ is not a basic syntactic category, rather, it abbreviates ‘NP\S’, an expression which merges with an appropriate sister expression on its left (indicated by ‘\’), again an NP, to form an S. Note that without directionality, ‘cat the’ should be a meaningful phrase, since ‘the cat’ is. But if ‘the’ is (only) of category NP/N ( producing a noun phrase from a noun, or more generally a nominal, on its right) it cannot merge with what is on its left. This is a second way in which the meanings of two phrases may not be able to compose.

On Some Examples of Chomsky’s

123

The semantic representations do not indicate the directionality of a word or phrase, so we cannot read syntax off them. But in the other direction, those representations can be recovered from syntax by the rules that NP correlates with i, N with ib, S with b, and any syntactic category of the form A/B or B\A with the functional type t1t2, where t1 is the type of expressions of category B and t2 is the type of expressions of category A. So ‘chased’ is of category (NP\S)/(NP) because it looks right to merge with an NP and produces an expression such as ‘chased Jerry’. ‘Chased Jerry’ is of category NP\S because it looks left to merge with an NP, say ‘Tom’, to produce a sentence. Correspondingly, the meaning of ‘chased’, which we variously write ‘[[chased]]’ or ‘chased’, is of type i(ib), because it takes an input of type i, such as [[ Jerry]], and produces a function of type ib, [[chased Jerry]], that takes an input of type i, such as [[Tom]], and produces an output of type b. Notoriously, natural-language semantics based on the simple theory of types is hugely oversimplified, since only two sentence meanings, true and false, are available. However, one can modify the simple theory into a hyperintensional theory, in which prop, an unanalyzed type of sentence-meanings, replaces b (Thomason 1980; further elaboration in Muskens 2005). For instance, ‘chased’ in the hyperintensional theory is of type i ѻ (i ѻ prop), since it maps an individual such as jerry to a propositional function such as chased( jerry), which maps the individual Tom to the proposition that Tom chased Jerry. This theory makes enough sentence meanings available, and it trades not offering an analysis of ‘proposition’ for not running up against the problems of grain that bedevil, for instance, the analysis of propositions as sets of possible worlds. But because the simple and the hyperintensional theory are close to isomorphic under interchange of b and prop, it is often adequate to formulate proposals in the simple theory, with the assurance that any extensionality objections can be met by moving to the hyperintensional theory without having to restructure the proposal. The input-output arrow has the logic of a substructural conditional logic. For given a meaning of type t1 ѻ t2 and a meaning of type t1, we obtain or ‘infer’ a meaning of type t2 by functional application (ѻE(limination)); and if a meaning of type t2 can be inferred from the hypothesis of a meaning of type t1, then a meaning of type t1 ѻ t2 may be inferred by functional abstraction (ѻI(ntroduction)). There are special restrictions on ѻ I; for a full account, see (van Benthem 1995, Chs. 2–4).

124

Graeme Forbes

ѻE by itself suffices for the orthodox proof that (1b) is the semantics of (1a): (2)

Wherever it is convenient to do so, we apply K-reduction, so chased rather than Ȝx.Ȝy.chased(x)(y) appears in (2). The lexicon provides both a syntactic category for the word, and, for its semantics, a constant of the typetheoretic language along with a type-label (which must be the one that can be read off the category label). The proof demonstrates both that ‘Tom chased Jerry’ is a sentence, since we derive S on the last row, and also that it has the semantics (1b), since we derive (1b) on the last row. The rule ѻE that we are using here applies to semantic types; but we could as well have used rules of \E and /E, applying to syntactic categories. The neo-Davidsonian semantics for (1a) is more involved than (1b), since it represents (1a) as meaning ‘in the past, some event was a chase whose agent was Tom and whose theme was Jerry’. In other words, ignoring tense, we have (3) a. Tom chased Jerry. b. (some)Ȝe.chase(e) and agent(e)(tom) and theme(e)( jerry). The most obvious question raised by (3) is how (3b) is to be recovered from (3a). One possibility is that the procedure is indirect: first (1b) is derived—the ‘atomic’ semantics, in Parsons’ terminology (Parsons 1990:8–9)—then a further analysis produces (3b), the ‘subatomic’ semantics. This route preserves a role for the familiar categorial syntax in deriving (1b). Alternatively, the subatomic semantics may be derived directly from (3a), without reliance on (1b). A direct-derivation account must be premised on a verb’s lexical entry assigning it a property of events. This may be a complex property, as proposed in (Parsons 1995), with conjuncts for the thematic relations to events whose relata are obligatory NP-arguments to the verb. But this raises problems of distinguishing obligatory from non-obligatory arguments, which we can avoid if we take lexical entries to be as simple as possible. Thus for ‘chase’ we would just have Ȝe.chase(e). However, we may take it that an English-speaker who understands ‘chase’ knows implicitly that in any event of chasing there is a chaser and a chased, or an agent and a theme, and also knows implicitly that the subject and object NP’s specify them. So we will assume that the input to interpretation is not the likes of the bare (3a), but rather a theta-labeled version:

On Some Examples of Chomsky’s

125

(4) |Tom|agent chased |Jerry|theme. The labelling allows us to generate the likes of ‘Ȝe.agent(e)(tom)’ from ‘|Tom|agent’, as the following derivation illustrates: (5)

A number of comments about (5) are in order: (i) We have introduced a type, e, which we will think of as a new primitive type, which generates new complex types in the standard way. (ii) The step from agent(e)(tom) to Ȝe.agent(e)(tom) illustrates the use of ѻI: assuming an item of type e, we obtain an item of type e ѻ b from an item of type e ѻ (i ѻ b) and an item of type i; cp. p ѧ (q ѧ r), q ՗ p ѧ r. The assumption is bracketed when it is discharged, and is co-indexed with the discharging use of ѧI. (iii) We also need some extra-logical machinery. The presence of |NP|T, where T is a label for a thematic relation, permits the introduction of the constant for T at a node on a path ascending from the node n where [[NP]] is input to the ib function. Exactly how one arranges this is partly a matter of aesthetics, but in derivations here, it will always be a twonode path, with a branch point at the lower node, as illustrated on the extreme top left of (5). When NP has semantic structure, the right branch from n will generally have more than two nodes (see below for the treatment of ‘some mouse’). Note that the T-label is erased when the ib-function derived from the T-relation is applied to [[NP]]. At the end of the derivation, I leave the leaf node with the constant for T undischarged, though no doubt some other convention is possible. (iv) The step labelled ‘Ȝ&’, for ‘lambda conjunction’, is another piece of extra-logical apparatus, based on Parsons’ non-logical rule of the same name (1995:651–2). A fixed verb, with its arguments and adjuncts, gives rise to an array of event-properties. As shown in (5), terms for these properties are applied to the same event-term and the result is abstracted. E-reduction (lambda conversion) and D-reduction (change of variable) then produce the penultimate line of (5). Note that in (5),

126

Graeme Forbes

and is of type b(bb) (i.e., it is the familiar sentential connective), but we write p and q rather than the official (and(q))(p). (v) The last step is by Parsons’ non-logical rule of default existential quantification (1995:652). This rule is only a default, as we shall see. (vi) At the second last line of the derivation, we have three conjuncts in an order that reflects the word-order in (3a). But clearly, nothing turns on which order the conjuncts are presented in, so in the hyperintensional semantics, the same element of the prop domain should result no matter what the order of the conjuncts. Our final illustration simultaneously shows the treatments of nonsentential co-ordination, type-raising, and quantifier-raising, first at the atomic level, then at the sub-atomic: (6) a. |Tom|agent chased and caught |a mouse|theme. b. ȜP.P(tom)Ȝx.a(mouse)Ȝy.chased(y)(x) and caught(y)(x). c. a(mouse)Ȝy.(some)Ȝe.agent(e)(tom) and chase(e) and theme(e) (y) and (some)Ȝe.agent(e)(tom) and catch(e) and theme(e)(y). (6b) is the reading of (6a) on which there is a mouse that Tom both chased and caught. In (6c) the chasing and the catching are different events (the very same chase could have ended with a fumble) and if ‘chased’ is intensional, then it is the relational reading that we aim to capture here. The primary difference between (6a) and (6b) is that in (6b), and coordinates two expressions of sentential type (i.e., b), so and has its basic type b(bb).2 But in (6a), ‘and’ appears to co-ordinate two transitive verbs, expressions of type i(ib); this ‘and’ has type (i(ib))[(i(ib))(i(ib))]. When and coordinates two expressions of type t, we may write it as andt. So (6a) has andi(ib) while (6b) has andb. To get from one to the other we use the polymorphic combinator Ю&, which is defined for certain types t as input, and produces andt as output.3 So in particular, Ю&(i(ib)) = andi(ib). Ю&(t) can be defined so that ultimately, for any appropriate t, the behavior of andt is fixed by the behavior of andb. The rule is 2 An expression with free variables has the type that results when entities from the domains for the types of the variables are assigned to those variables. chased(y) (x) and caught(y)(x) each have type b because the output is a truth-value when entities from the domain of individuals are assigned to x and y. 3 Here I follow (Carpenter 1997:180–1). The input types have to be boolean, that is, ones which produce the type b after all arguments have been consumed.

On Some Examples of Chomsky’s

127

(7)

where (7b) is the base-case stipulation. So, by (7a) we have (Ю&(ib)(snored)) (slept) = (andib(snored))(slept) = Ȝxi.(Ю&(b)(snored(x)))(slept(x)). By (7b) this in turn is Ȝxi.(andb(snored(x)))(slept(x)); thus andib is explained in terms of andb. We refer to (7) as Simp(lification), and we continue to use postfix positioning of conjunction; for example, slept andib snored rather than (andib(snored))(slept). The other special features of (6b) are (i) the quantifier-type semantics of ‘Tom’ and (ii) the raising of the terms for ‘Tom’ and ‘some mouse’ to have scope over the sentence chased(y)(x) and caught(y)(x); in other words, we have type raising and quantifier raising. The type-raising we are employing takes an individual and produces a function from properties to truthvalues, so we go from i to (i ѻ b) ѻ b (this is known as the Montague Rule) and again there is a parallel with logical “if . . . then”, since p ՗ ( p ѧ q) ѧ q. The derivation below of ȜP.P(tom) follows the natural proof of p ՗ ( p ѧ q) ѧ q: we assume something of type i ѧ b, use i and i ѧ b to obtain b, then discharge the assumption i ѻ b by lambda abstraction. Quantifier raising is accomplished in two steps: after a QNP has been derived, it may be substituted by an individual variable v, and an expression of type b derived using v; a rule μ (for ‘Moortgat’) then allows abstraction on v and introduction of the QNP that was substituted (see Carpenter 1997:220–27). The derivation of (6b) now follows. Here and in subsequent proofs, we use a double line to abbreviate a transition for which a sequence of steps is illustrated in previous examples or elsewhere in the current proof. The type of determiners, (ib)((ib)b), is abbreviated as det, and the type of quantifiers, (ib)b, as q. (8)

128

Graeme Forbes

Two applications of ȕ-reduction to the last line of this derivation would distribute ‘tom’ through, resulting in a(mouse) Ȝy.chased(y)(tom) andb caught(y)(tom). We derive the subatomic semantics (6c) as follows, using ag for agent and tm for theme: (9)

At the first step in (9), we obtain Ȝe.agent(e)(x) from |x|ag in the same way as we obtained Ȝe.agent(e)(tom) from |tom|ag in (5); mutatis mutandis for Ȝe.theme(e)(z) and |z|tm. The next step uses Ȝ& and default existential quantification to obtain the proposition that x is an agent of a chasing of z. The assumptions |x|ag and |z|tm are then discharged by successive lambda abstractions, producing Ȝz.Ȝx.(some)Ȝe.agent(e)(x) and chase(e) and theme(e)(z) as the interpretation of (6a)’s ‘chased’. On the right of the proof we have suppressed the corresponding derivation of Ȝz. Ȝx.(some)Ȝe.agent(e)(x) and catch(e) and theme(e)(z) as the interpretation of (6a)’s ‘caught’, which is achieved at the same line. Both these terms are of type i(ib), that of a transitive verb (fortunately, ‘chased’ and ‘caught’ are transitive verbs) and so the meaning we need for (6a)’s ‘and’ is andi(ib). This is obtained by extracting andb from the lexicon and (not shown) applying C&(i(ib)). Successive uses of Simp produce the desired i(ib) term with separate event quantifiers but commonly bound agent variables and commonly bound theme variables. The proof continues with this i(ib) term consuming y then tom, and finally Moortgat’s rule for quantifier-raising is applied. Perhaps the reader will have noticed that while most semantic type information has been omitted from (9) (and some occasionally abbreviated ‫ )׌‬simply to allow the proof to fit on the page, the syntactic category information is missing from (9)—and also (5)—for a different reason. It is

On Some Examples of Chomsky’s

129

not that the standard categories make no sense: we can still say, if we like, that ‘chased’ is of category (NP\S)/NP. But by going directly to subatomic semantics, we have dissolved the close relations between syntactic and semantic classification, so that a hand-in-hand derivation like (2) is no longer possible. One option here is just to accept that the proof that a given string is of category S proceeds entirely independently of the derivation of the string’s meaning. Another is to drop the categorial syntax entirely and rely on phrase structure plus thematic labelling to screen out nonsense strings that threaten to acquire meanings in the framework. It is an interesting question how much service of this kind can be performed by the thematic requirements of, and options for, verbs. We already noted that though semantics can explain why some phrases lack meaning (there is a type mismatch both ways round), syntax is needed to prevent others being interpreted: either there is no type mismatch but the order is wrong, or, though the types and order are correct, there is a category mismatch (cf. ‘the sings’, where the problem is that ‘the’ needs N to the left, not NP\S). However, I will not pursue this issue any further in this paper.

2. Chomsky’s examples Already in (Chomsky 1975) we find Chomsky expressing scepticism about truth-conditional semantics, on the basis of a range of examples which, he says, tend to show that ‘even a principle of compositionality is suspect’ (1975:31). Some of these examples are beyond the scope of this paper, but two pairs that are particularly interesting from the neo-Davidsonian perspective are (10) and (11) below: (10)

a. Poems are written by fools like Smith. b. Mountains are climbed by fools like Jones

(11)

a. Beavers build dams. b. Dams are built by beavers.

It is hard to see why there would be any structural differences in the semantics of (10a) and (10b), but (10a) requires all poems to be written by fools like Smith, whereas (10b) may be true even though there are unclimbed mountains. Chomsky doubts that these differences can be ‘traced to structure’ (1975:39); we might, for instance, restore a parallel with (10a) by substituting a creation verb in (10b) (‘mountains are formed by cataclysms like this’). Apparently, the lexical meaning of the verb is playing a

130

Graeme Forbes

role that cannot be cashed out in terms of structural effects.4 In the case of (11), though passivization is meaning-preserving in simple cases (‘Tom chased Jerry’, ‘Jerry was chased byTom’), apparently (11a) can be true while (11b) is false.5 But once we discern quantification over events, we can interpret the examples in (10) and (11) in a way that agrees with Chomsky’s intuitions. For (10), (12)

a. Every poem has a fool like Smith as agent of its writing. b. (every(poem))[Ȝx.a((like(smith))(fool)) [Ȝy.(the(Ȝe.writing(e) and theme(e)(x)))[Ȝe.agent(e)(y)]]]. c. Every climbing of a mountain has a fool like Jones as agent. d. every(Ȝe.climbing(e) and some(mountain)[Ȝx.theme(e) (x)])[Ȝe.a((like( jones)) (fool))Ȝy.agent(e)(y)].

In both (12b) and (12d), we use the syntax of restricted quantifiers: a determiner, in these cases, every, applies to a range-restricting condition (poem in (12b)) to produce a restricted quantifier of the general form det(rr-cond), which in turn is applied to a scope [. . .] of the required type. In (12b), every is of type (ib)((ib)b), while in (12d), it is of type (eb)((eb)b), as is the in (12b). In (12b), we say that every poem is such that the writing of which it is theme is such that a fool like Smith is its agent. like(smith) is the meaning of a complex extensional attributive adjective (type (ib)(ib)) which applies to nouns (type ib) to interpret first-order predicates such as ‘fool like Smith’. The contrast between (10a) and (10b) emerges as a contrast in the restricting condition to which every is subject, simply poem in (12b) but ‘thing such that it is a climbing and some mountain is its theme’ in (12d). Of course, it is all very well to trot out interpretations that capture the readings Chomsky proposes; the challenge is to describe a compositional, truth-conditional, and non-ad hoc way of arriving at them. How and why do we end up with the contrasting (12b) and (12d), beginning from the similar (10a) and (10b)? We address this question in the next section. But before turning to it, we note that we can formulate the contrasting meanings of the examples in (11) in essentially the same way: 4 A similar theme is expanded on in ( Johnson 2004). 5 For further elaboration of Chomskyan scepticism about truth-conditional semantics, see (Pietroski 2003, 2005).

On Some Examples of Chomsky’s

(13)

131

a. Significantly many beavers are agents of dam-buildings. b. (s-many(beaver))[Ȝx.(some(Ȝe.building(e) and a(dam) [Ȝy.theme(e)(y)])[Ȝx.agent(e)(x)]].6 c. Significantly many dams have beavers as agents of their building. d. (s-many(dam))[Ȝx.(the(Ȝe.building(e) and theme(e)(x))) [Ȝe.a(beaver)[Ȝy.agent(e)(y)]].

(13)’s use of ‘significantly many’ (with proportional ‘many’) embodies a stop-gap proposal about how to provide a bare plural with a determiner in a use of the plural as a generic, though to keep things manageable, the interpretations revert to the singular (unrealistically assuming distributive readings). One difficulty with generics is to decide exactly what quantification, if any, they are making: neither ‘all’ nor ‘most’ nor ‘many’ seems right, as examples like ‘Dutchmen are good sailors’ and ‘mosquitoes spread malaria’ show (only 5% of mosquitoes carry the virus).7 The idea behind ‘significantly’ is that standards of significance—how many is significantly many?— can vary from context to context; the same effect would be obtained by using ‘enough’. “Mosquitoes don’t spread malaria” would then come out false because “significantly many/enough don’t” means ‘practically none do’ to the epidemiologist, and 5% is more than practically none. Perhaps ‘significantly’ can even be made to capture whatever modal force there is in these examples (which are from Carlson and Pelletier 1995:44, 81).8

6 One complication here is that (11a) and (13a) would (I think) be true even if beavers have poor follow-through and are inclined to abandon their dam-building projects before any dam exists. The intensionality of creation verbs in the progressive is often noted, but characterizing sentences seem to have the same feature. However, I will not burden the reader with the apparatus of (Forbes 2006:130–38) that would accommodate this complication. 7 See (Pelletier and Asher 1997) for criticism of quantificational accounts of generics. 8 Modal force is imputed to ‘mosquitoes spread malaria’ by a paraphrase like ‘significantly many mosquitoes spread malaria and it is in the nature of those mosquitoes to do so’. Thus an accidental universal truth does not support a generic attribution, but at the same time, ‘mosquitoes spread malaria’ would be false if it was in the nature of all to do so but none do, because of a vaccination program. Conceivably the modal force itself comes from the ‘significantly’, in which case the ‘in the nature of ’ conjunct need not be stated separately. However, I am sceptical that there is modal force in such cases. ‘Mosquitoes spread malaria’ is true because of the percentage of cases of the disease caused by mosquitoes, and would remain true no matter how accidental the role of the insect in propagating malaria, if the percentage was unchanged. As for the children of Rainbow Lake (Pelletier and Asher 1997:1132), ‘children from Rainbow Lake are unusually tall’ could be an

132

Graeme Forbes

3. Focus effects We turn now to the main problem, that of formulating a compositional, truth-conditional, and non-ad hoc way of arriving at Chomsky’s readings of his examples. The proposal we will try to implement is that the difference between the (a)-sentences and the ( b)-sentences in (10) and (11) has to do with different ways of applying the background/focus (topic/comment, subject-matter/statement) distinction to them.9 One way of imposing a background/focus distinction on a statement is to take it to be the answer to a question—the focussed material is the part that provides the answer to the question. For example, there are at least four different ways of distinguishing background from focus in (10a), as the following question/answer pairs reveal (the fourth is rather marginal): (14) a. How do poems come to exist? Poems are written by fools like Smith. b. Who writes poems? Poems are written by fools like Smith.10 c. What do fools like Smith write? Poems are written by fools like Smith. d. How do fools like Smith spend their time? Poems are written by fools like Smith. The same focussing possibilities are available for (10b): (15)

a. What happens to mountains? Mountains get climbed by fools like Jones. b. Who climbs mountains? Mountains are climbed by fools like Jones. c. What do fools like Jones climb? Mountains are climbed by fools like Jones. d. How do fools like Jones spend their time? Mountains are climbed by fools like Jones.

If we now suppose that the focus differences, especially the ones between the (a) and (b) sentences in (14) and (15), manifest themselves at the subatomic accidentally true characterizing statement offered to a photographer in search of unusually tall children. See (Krifka et al. 1995) for a broad-ranging discussion of generics, and (Koslicki 1999) for the prospects of a unified treatment of the generic uses of various categories of expression. 9 Identifying these three distinctions is an oversimplification; see (Partee 1991: 163–5). 10 The questions in (14) are to be read without any special focus themselves.

On Some Examples of Chomsky’s

133

level as structural differences, then we have made a start on explaining how the difference between (10a) and (10b) can, after all, be traced to structure. For (10a) and (10b) do not give any clue about how the focus/background distinction applies to them. Chomsky’s intuition that only (10a) is correctly paraphrased by prefixing ‘all’ may simply indicate that in the absence of other cues (in the “null context”), the focus displayed in (14a) is preferred for (10a), while the focus displayed in (15b) is preferred for (10b). If so, there is no semantic difference between (10a) and (10b), for they have the same range of readings discriminating background from focus; the difference would only have to do with which is preferred in the null context, and so would be pragmatic. But why is there such a pragmatic difference, exactly? The use of the passive promotes either poems or poem-writings as the topic of (10a), and it is quite hard to distinguish the two cases: (16)

a. Every poem has a fool like Smith as agent of its writing. b. Every poem-writing has a fool like Smith as agent.

(16a) would imply (16b) but for (i) the possibility of the writing of a poem being wisely abandoned before any poem exists. And given at most one writing per poem, (16b) would imply (16a) but for (ii) the possibility of a poem being entirely orally composed (or in some other way brought into existence). But (16a) and (16b) may be understood in such a way that (i) and (ii) are not the basis of potential counterexamples to equivalence. So defaulting to (16a) in the null context may just be a consequence of (16a)’s greater accessibility. By contrast, with no creation verb, (12c) and ‘every mountain has a fool like Jones as agent of its climbing’ are easily distinguished, and it requires some effort to get the topic to be mountains ((15a) is rather lame).11 Another factor contributing to 11 Chomsky (p. 39) considers the hypothesis that his passive examples all have an implicit ‘only’, as in ‘mountains are climbed only by fools like Jones’, and that ‘we are saying something about all poems, mountains, dams’. [We would have the atomic semantics every(mountain)Ȝx.only(a fool like jones)Ȝy.y climbs x. This holds iff whenever a mountain is assigned to x, y climbs x is true only if a fool like Jones is assigned to y—only has type ((eb)b)((eb)b) in this case—so the statement neither entails nor presupposes that every mountain is climbed.] But when the verb is creative, Chomsky proposes, ‘we understand further that the entities are formed in no other way’. However, this ‘understanding’ seems to be a reading. Event-quantification, as in ‘every mountain-climbing/poem-writing’ gets the effect of ‘only’ without postulating a covert occurrence of the word. (See n.12 for more on only.)

134

Graeme Forbes

the preference for (16a) is also pragmatic, namely, that if we think of ‘Poems are written by fools like Smith’ as an answer to ‘How do poems come to exist?’, Gricean mechanisms will create the implicature that there is no other way they come to exist, justifying ‘all poems’. But even if nothing much else happens on mountains, there remains the option of being a mountain on which nothing at all happens, so ‘all mountains’ cannot be used. (There is no option of being a poem which doesn’t exist.) (10a), ‘poems are written by . . . ’, and (11b), ‘dams are built by beavers’, are also comparable; indeed, Chomsky says ( p. 39) that the most natural reading of (11b) takes ‘dams’ as ‘all dams’, as he interpreted ‘poems’ in (10a). Perhaps this is right, but I think there is also a generic reading, which states a not-necessarily-exceptionless characteristic of dams, and which is perhaps more useful for bringing out the core difference between (11a) and (11b). The availability of the generic reading may have to do with the existence of an option of being a dam that wasn’t built, as Chomsky notes ( p. 39) when he contrasts dams that are built with ones formed naturally by falling trees.

4. Focus in type-logical event semantics If (10a) prefers the focus of (14a), and (10b) the one in (15b), how do we get from those readings to the different interpretations in (12)? Here we will employ the core idea of (Herburger 2000), where it is argued (p. 43) that there is an operation of focal mapping, which assigns background material to the restriction of the event quantifier and focussed material to its scope. In other words, and roughly speaking, the topic becomes the quantifier restriction and the comment its scope. We need to specify exactly how we get this outcome in the derivational model of interpretation, beginning with a simple case, such as our initial example, (1a), ‘Tom chased Jerry’. Our neo-Davidsonian semantics for (1a), given in (3b), (some)Ȝe.chase(e) and agent(e)(tom) and theme(e)( jerry), is, with its unrestricted some(eb)b, suitable only for a use of (1a) with no focus/background contrast (Q: ‘What happened here last night?’; A: ‘Tom chased Jerry.’). But if (1a) were the answer to ‘Who chased Jerry?’, then ‘Tom’ would be focussed, as in (17a) below, requiring, on Herburger’s theory, the interpretation ‘some chasing of Jerry was by Tom’, as given in (17b):

On Some Examples of Chomsky’s

(17)

135

a. Tom chased Jerry b. (some(Ȝe.chase(e) and theme(e)( jerry)))[Ȝe.agent(e) (tom)].12

To use this idea to meet the challenge of Chomsky’s examples, we first have to show how we can derive (17b), starting with the same English sentence ‘Tom chased Jerry’ as in the original derivation (5). Fortunately, (5) is easily adjusted to produce (17b). On (5)’s fourth row, the constituents of the interpretation are available for combination into a truth-condition. Ȝ-conjunction of all constituents is the only combination method we have used so far, but the phenomenon of focus calls for more flexibility: we should be able to apply separate Ȝ-conjunctions to the focussed constituents and the background ones. There is no mystery about the successful interpreter’s knowledge of which constituents are focussed and which are background, since the 12. In this example, the focussing makes no difference to truth-conditions (compare (17b) and (3b)). That focus does have truth-conditional impact is illustrated by a well-known example due to Jackendoff, ‘John only introduced Bill to Sue’. When ‘only’ immediately precedes a verb and its complements, there are usually multiple ambiguities over what the argument of only is. ‘Only’ and its argument can be discontinuous, and separated by expressions which may themselves be intelligibly understood as only’s argument. So in Jackendoff ’s example, either ‘Bill’ or ‘to Sue’ may be the argument. Such ambiguities are resolved by focus, normally marked by an intonation specifying the argument, as in (i) ‘John only introduced Bill to Sue’ versus (ii) ‘John only introduced Bill to Sue’. And these have different truth-conditions. In fact, (iii) ‘Bill’ and ‘Sue’ may both be focussed (the only introduction John performed), (iv) ‘introduced’ may be the focus (he didn’t arrange their marriage), (v) ‘introduced Bill to Sue’ may be the focus (the only thing he did), and no doubt there are other possibilities. For (i) we have the atomic semantics (only(bill))Ȝx.introduce(to sue)(x)( john), while (iv) asks for a sub-atomic semantics, perhaps (only(introduce))ȜP.some(Ȝe.agent(e)( j) and theme(e)(b) and goal(e)(s))[Ȝe.P(e)]. ‘Only’ can apply to adjectives, determiners, quantifiers, even adverbs (“I’ve only heard informally, but I got the job”). An evaluation rule for (i) and (ii) is that if ˙ is a constant of type i and v a variable of that type, then (only(˙))Ȝv.ij is true iff: [[ij(v/x)]] is true (if and?) only if x = [[˙]]. For (iii) we can use product types, or perhaps, as in (Krifka 1991), lists. But this is insufficiently general. For if John only introduced a rich man to Sue, and the man he introduced was sick, didn’t he only introduce a sick man to Sue? But in that case a generalized version of our rule for (i) and (ii) would make both statements false, since both rich and sick satisfy the relevant open sentence. However, I think that a contextually-supplied domain restriction is all that is needed to fix this, not a more radical recasting of the semantics such as that in (Rooth 1992). Whether the audience has to know what the individual alternatives are for full understanding to occur is debatable.

136

Graeme Forbes

question-answer contexts, or intonation, provide exactly that information. So expressions can be marked as focus or background without circularly assuming the interpretation of the statement they are part of. We can then apply one Ȝ-conjunction to items marked F and another to items marked B, with the Ȝ-conjunction of the latter group becoming, by default, the restricting condition of an existential event-quantifier. In the following variant of (5), which produces (17b), we mark focus with italics and background with roman text: (18)

As (18) shows, we are not employing an additional non-logical rule, focal mapping, but rather a liberalized version of Ȝ-conjunction. Previously, Ȝ-conjunction consolidated the event properties corresponding to a verb and all its arguments and adjuncts. But we now permit separate Ȝ-conjunctions for focus properties and background properties. In (18), because the focus is a single property, only one Ȝ-conjunction is required. But the one that is used, which produces Ȝe.chase(e) and theme(e) ( jerry) on the third row, is restricted to background properties. (However, focal mapping is present as a constraint, since we cannot apply default quantification to a conjunction of focussed properties.) The bare-plurals examples show how a specific focus can produce a meaning specific to that focus. We consider (10b) in detail below, assuming the “natural” focus ‘mountains are climbed by fools like Jones’. As already indicated, we follow Chomsky in ascribing universal force to these examples (see note 11) and in taking the agent specification to be exhaustive ( perhaps as a consequence of the focus). The universal force is explained by (10b)’s being a “characterizing sentence” (Krifka et al. 1995:3), while the focus-background split determines what is characterized,mountain-climbings in the case of (10b) and poems in the case of (10a). ‘Mountains are climbed by fools like Jones’ is synonymous with ‘fools like Jones climb mountains’, where the focus is preserved, and ( granted universal force) with ‘every mountain-climbing is by a fool like Jones’—all three answer ‘Who climbs mountains?’ So neither passivization nor the use of plurals is essential to the meaning. In that case, a simplified derivation that produces (12d) as the meaning of (10b) is possible:

On Some Examples of Chomsky’s

137

(19)

At the fourth line, default existential quantification gives way to a universal, because we are generalizing over events of a certain sort. Also, ‘fools like Jones’ is not labeled ‘agent’, because the preposition ‘by’, in its sense here, expresses this concept (similarly, ‘of ’ has a sense which expresses theme and ‘with’ has a sense which expresses instrument). Last, ‘are climbed’ does not really get its interpretation from the lexicon; some processing for ‘be’ + past participle is at work. But the details are not relevant here.13 There is a collective-distributive ambiguity in (10b), since a team of climbers may be said to climb a mountain without all the team’s members climbing it, but it is only the distributive reading we aim to capture. In effect, the singular NP readings we give to the bare plurals in (19) presume distributivity. The equivalent of (12d) in which plurals are treated as such, would read: mountains are such that for every one of them, any climbing of it is such that fools like Jones are such that at least one of them is agent of that climbing.14 Simply for illustrating the treatment of focus, it seems better to capitalize on the distributivity.15 13 They may be as simple as that participles get the same semantics as the main verb form, while auxiliary ‘be’ has no semantic effect (Parsons 1990:91–2). Alternatively, a participle may be of adjective type, with a meaning fixed by the verb (something is a climbed F iff it is an F which someone climbed). ‘Be’ then produces an appropriate predicate from the adjective; cf. Partee’s account of ‘be’ in (Partee 1986), reprinted in (Partee 2004), esp. pp. 213–5, 223–4. 14 In these terms, (10a) reads ‘poems are such that for each of them, fools like Smith are such that one of them is agent of the writing of that poem’. 15 Carpenter (1997:297–307) suggests that bare plurals are accompanied by a covert determiner, which for ‘mountains’ in our examples is presumably all (this allows for uniform treatment with cases where the determiner is explicit). Let plu be an operator such that given mountain of type ib, plu(mountain) is the subset of the powerset of mountains in which each set is of size t2. Define the plural determiner all so that all(plu(mountain)) takes (ib)b into b. Its complement is of the form ‘every one of them is F ’, which Carpenter provides with an (ib)b semantics by defining a distributor I write ‘devery’ that consumes, not all(plu(mountain)), but rather a term Pib that would be substituted for it in a derivation. So we get devery(Pib), for which we substitute a term xi, and construct a sentence ij(xi). An application of μ produces devery(P)Ȝx.ij(x) and

138

Graeme Forbes

We turn now to the examples in (13), repeated here, (13)

a. Significantly many beavers are agents of dam-buildings. b. (sig-many(beaver))[Ȝx.(some(Ȝe.building(e) and a(dam))) [Ȝy.theme(e)(y)]) [Ȝx.agent(e)(x)]]. c. Significantly many dams have beavers as agents of their building. d. (sig-many(dam))[Ȝx.(the(Ȝe.building(e) and theme(e) (x)))[Ȝe.(a(beaver))[Ȝy.agent(e)(y)]]].

Chomsky’s intuition that it could be that (11a), ‘beavers build dams’, is true while (11b), ‘dams are built by beavers’, is false, appears to depend on letting focus shift rather than on reading ‘dams’ as a strict universal, ‘all dams’. (11a) might be a correct response to the question ‘what do beavers do that is distinctive of them?’, which puts the focus on ‘build dams’. (11b) might be an incorrect response to the question ‘where do dams come from?’16 But it sounds to my ear that if we treat (11b) as an awkwardly formulated response to ‘what do beavers do that is distinctive of them?’, that is, if we keep the focus on ‘build dams’, (11b) is true. With focus having truth-conditional impact, allowing it to shift in passivization cannot be expected to be truth-preserving. And it is an immediate pay-off of Herburger’s account of focus in the neo-Davidsonian framework that it pins down exactly what the semantic effect of the shift is and makes it easy to understand why equivalence is lost. For focus on ‘build dams’ puts ‘beavers’ into the restriction of the main quantifier and ‘build dams’ into its scope, while focus on ‘built by beavers’ makes ‘dams’ the quantifier restriction and puts ‘built by beavers’ into the quantifier scope. The contrast between (13b) and (13d) is therefore not altogether unlike that between a conditional and its converse. As for deriving (13b) and (13d), no new issues arise, other than the adequacy, or otherwise, of ‘significantly many’. So given focus on ‘build dams’ in ‘beavers build dams’, the semantics has ‘Ȝx.there is a dam-building

another use of μ results in all(plu(mountain))ȜP.devery(P)Ȝx.ij(x). The semantics requires that for this to be true, all sets in plu(mountain) make devery(P) Ȝx.ij(x) true when assigned to P. The same derivation process within the Ȝx.ij(x) will accommodate ‘for every climbing of it, the/some fools like Jones are such that at least one of them is agent of that climbing’. The enthusiastic reader may wish to provide the unsimplified version of (19) to confirm that this nice treatment suffices. 16 Incorrect because the proportion of dams not built by beavers is too high.

On Some Examples of Chomsky’s

139

of which x is agent’ as comment and beavers as topic. The semantics of the comment uses default existential quantification again.

5. Undergeneration and overgeneration Semantic apparatus of the kind employed in this paper raises completeness and soundness questions: can we derive all the interpretations there are, and, can we derive some interpretations there aren’t? We want the answers to be yes and no respectively. On the completeness side, the main challenge comes from cases where what is focused does not correspond to a Ȝ-term to which Ȝ-conjunction applies. For instance, perhaps an adjective is focussed. Intersective adjectives are not too difficult, but we might instead focus an extensional attributive, as in (21)

A large beaver built this dam.

A large beaver is something that (i) is a beaver and (ii) is large for a beaver (we need (i) since (ii) by itself could be an argument against the thing in question being a beaver, as in ‘too large for a beaver’). The way we handle this also has to work for cases such as ‘no large beaver built this dam’ and ‘large beavers build big dams’. And there is the additional complication that any proposal also has to work for intensional adjectives, since presumably the effect of adjective focus in a nominal is to be explained in a uniform way: focus is doing the same thing in (21) and ‘an alleged thief removed the picture’ (you made the allegation but I suspect he had permission). These various constraints rule out many different proposals in which the adjective and the noun get separated. ‘Large’ and ‘alleged’ cannot meaningfully apply to an expression of type i, such as an individual variable, and in the intensional case any quantifier formed from a nominal involving ‘thief ’ but not ‘alleged’ is likely to be extensionally incorrect. It seems, then, that the first step is to apply Ȝ-conjunction so that the entire Ȝ-term that contains the adjective, a(large(beaver))[Ȝx. agent(e)(x)] or an(alleged(thief ))[Ȝx.agent(e)(x)], becomes the scope of the event quantifier.17 But this is insufficient, since it does not distinguish between

17 I am assuming that an(alleged(thief )) is an expression of hyperintensional typetheory, in which alleged is of type (i ѻ prop)(i ѻ prop).

140

Graeme Forbes

focus on ‘large’ versus focus on ‘beaver’ versus focus on ‘large beaver’; mutatis mutandis for ‘alleged thief ’. One solution to this problem is to suppose that within the topic, or within the comment, a subsidiary background/focus distinction can apply.18 So taking ‘a large beaver’ in (21) as the argument to ‘built this dam’, we should be able to configure its interpretation in different ways, plausibly reflecting the discourse felicity of different prosodies.19 The three intonations relevant to the interpretation are ‘a large beaver’,‘a large beaver’ and ‘a large beaver’. These suggest three different groupings of the words in the NP: ‘a (large beaver)’, ‘a (large) beaver’, and ‘(a large) beaver’ respectively, where in the second we are thinking of ‘a beaver’ as constituting the background for the focussed ‘large’. These groupings in turn suggest how we might use type shifting to realize three distinct interpretations of the phrase, all of which work out to type (ib)b, and so can apply to [Ȝx. agent(e)(x)]. Uniqueness is another question, but one trio which does the job is: (22)

a. a(large(beaver))[Ȝx.agent(e)(x)] b. a(beaver(large))[Ȝx.agent(e)(x)] c. (a(large))(beaver)[Ȝx.agent(e)(x)].

In deriving (22c), the interpretation of ‘a large beaver’ still has to be of type (ib)b, and we can accomplish this if a is of the type of functions from (ib) (ib) to (ib)((ib)b). And the familiar determiner type, that is, the type of functions from properties to properties of properties—(ib)((ib)b)—shifts to exactly this type via the principle (23) p ѧ q ՗ (r ѧ p) ѧ (r ѧ q) also known as the Geach Rule (see van Benthem 1995:26). For (22b), where large is the argument of beaver, we use the Montague Rule p ՗ ( p ѧ q) ѧ q already encountered in the derivation of ȜP.P(tom) in (8). In this application, we shift beaver from the standard nominal type ib to a type which consumes adjectives, (ib ѻ ib) ѻ ib. beaver(large) is therefore 18 Krifka (1991:153) gives the following example of focus within the topic: “What did Bill’s sisters do?––Bill’s youngest sister kissed John”. See (Glanzberg 2005:90–105) for discussion of this phenomenon. 19 Here and in the rest of this paragraph I am borrowing ideas from (Steedman 2000:89–95).

On Some Examples of Chomsky’s

141

of type ib, so if a is of its usual type (ib)((ib)b), ‘a large beaver’ ends up being of type (ib)b again. These examples illustrate the flexibility of the framework, but may also raise the suspicion that it is too flexible. Perhaps we will end up with interpretations for the uninterpretable, like an example of Mark Liberman’s reported in (Steedman 2000:92), ‘Harry likes the nuts and bolts approach’. In this case the comment is not split across the Ȝ-terms to which Ȝ-conjunction applies (unlike Steedman’s own case, loc. cit., ‘Three mathematicians in ten prefer margarine’), since ‘the nuts and bolts approach’ expresses the theme of the state of liking. Hence an attempted derivation will not break down at focal mapping. But for Liberman’s example, there does not seem to be a way of shifting types in the logic so that an intelligible topic/comment differentiation emerges within the theme and reflects the bizarre intonation. This is far from a soundness proof, but provides some reassurance that not just anything goes. I conclude that the original examples Chomsky gave, or at least the four we have investigated, do not threaten the central role that compositional, truth-conditional semantics is commonly thought to have in the explanation of how language functions as a vehicle of communication. For the event-based, derivational semantics presented in this paper shows some promise of handling them acceptably, and this situation can only improve with the increasingly sophisticated understanding of such topics as generics, focus, and type-logical derivation, that, if the last decades are a guide, we can reasonably look forward to.20

References Carlson, Gregory, and Francis Jeffry Pelletier, eds., 1995. The Generic Book. University of Chicago Press. Carpenter, Bob. 1997. Type-Logical Semantics. The MIT Press. Chomsky, Noam. 1975. Questions of Form and Interpretation. In Essays on Form and Interpretation, edited by N. Chomsky. North Holland. Davidson, Donald. 1967. The Logical Form of Action Sentences. In The Logic of Decision and Action, edited by N. Rescher. University of Pittsburgh Press. Forbes, Graeme. 2006. Attitude Problems. Oxford University Press. Glanzberg, Michael. 2005. Focus: A Case Study on the Semantics-Pragmatics Boundary. In Semantics versus Pragmatics, edited by Z. Szabó. Oxford University Press.

20 I thank Kathrin Koslicki and Paul Pietroski for helpful comments on earlier versions of this paper.

142

Graeme Forbes

Herburger, Elena. 2000. What Counts: Focus and Quantification. The MIT Press. Johnson, Kent. 2004. Systematicity. The Journal of Philosophy 101:111–139. Koslicki, Kathrin. 1999. Genericity and Logical Form. Mind and Language 14:441–467. Krifka, Manfred. 1991. A Compositional Semantics for Multiple Focus Constructions. In Proceedings from Semantics and Linguistic Theory I, edited by S. Moore and A. Z. Weiner. Cornell University. Krifka, Manfred, Francis Jeffry Pelletier, Gregory Carlson, Alice ter Meulen, Godehard Link, and Gennaro Chierchia. 1995. Genericity: an Introduction. In The Generic Book, edited by G. Carlson and F. J. Pelletier. University of Chicago Press. Muskens, R. 2005. Sense and the Computation of Reference. Linguistics and Philosophy 28:473–504. Parsons, Terence. 1990. Events in the Semantics of English. The MIT Press. Parsons, Terence. 1995. Thematic Relations and Arguments. Linguistic Inquiry 55:663–679. Partee, Barbara. 1986. Noun-Phrase Interpretation and Type-Shifting Principles. In Studies on Discourse Representation Theory and the Theory of Generalized Quantifiers, edited by J. Groenendijk, D. d. Jongh and M. Stokhof: Foris. Page references are to (Partee 2004). Partee, Barbara. 1991. Topic, Focus and Quantification. In Proceedings from Semantics and Linguistic Theory I, edited by S. Moore and A. Z. Weiner. Cornell University. Partee, Barbara, ed. 2004. Compositionality in Formal Semantics: Selected Papers by Barbara Partee. Basil Blackwell. Pelletier, Francis Jeffry, and Nicholas Asher. 1997. Generics and Defaults. In Handbook of Logic and Language, edited by J. v. Benthem and A. T. Meulen. Elsevier and The MIT Press. Pietroski, Paul. 2003. The Character of Natural Language Semantics. In Epistemology of Language, edited by A. Barber. Oxford University Press. Pietroski, Paul. 2005. Meaning before Truth. In Contextualism in Philosophy, edited by G. Preyer and G. Peter. Oxford University Press. Rooth, Mats. 1992. A Theory of Focus Interpretation. Natural Language Semantics 1:75–116. Steedman, Mark. 2000. The Syntactic Process. The MIT Press. Thomason, Richmond. 1980. A Model Theory for Propositional Attitudes. Linguistics and Philosophy 4:47–70. van Benthem, Johan. 1995. Language in Action. The MIT Press.

Truth, Meaning and Contextualism Samuel Guttenplan Abstract: The idea that meaning is best understood by showing how to deploy a systematic theory of the truth conditions for sentences in a language has come under sustained attack in recent years by adherents of what is often called ‘contextualism’. In its most radical form, contextualism casts doubt on the idea that sentences can so much as have truth conditions outside of the context in which they are used. This is supposedly because ‘context’ includes not only the usual reference-fixing resources, but also interpretations of relevant sub-sentential items—interpretations we need before we can so much as assign truth conditions. Others, especially Cappelen and Lepore (2005), have defended the truth conditional account by arguing that the standard clauses of truth conditional theories already give us all the interpretative materials we need for assigning truth, and hence, meanings (or content) to sentences. In my paper I set out to show that both sides in this dispute implicitly rely on an assumption about the business of theories of meaning which is not in any sense mandatory. When this assumption is dropped, as I suggest it should be, the basis for the dispute simply dissolves.

1. Michael: ‘I have looked everywhere for the keys and cannot find them.’ Alison: ‘Those keys are in the sideboard where you put them yesterday.’ Michael ( goes to the sideboard, retrieves the keys): ‘Sorry. You were right.’ This example of an apparently straightforward conversational interchange can trigger bewilderingly complex (and even bad-tempered) philosophical interchange. What contributes to such complexity in the philosophical interchange is that various parties seem to be operating against a background of different ground-rules for central notions such as truth, truthcondition, context, what-is-said, what-is-implicated and, not least, semantics, communication and meaning. Moreover, the many subtle, and not so subtle, differences in the background governing these notions are not merely careless. Contributors argue for the rightness of their understanding of the background and the ground-rules that shape it, insisting that only by seeing

144

Samuel Guttenplan

matters in their way do we have any hope of a proper theoretical grasp of what Michael and Alison managed with so little effort. The theoretical debate now has too many twists and turns to be untangled by any single paper. And, though tempting, the Gordian stratagem is not even of metaphorical help here, since there is likely to be as much debate about what would count as cutting the knot as about anything else in this area. Still, the aim of this paper is untangling, rather than tightening, and while the length of unknotted string it will produce is not long, I hope you will come to think that I am pulling on the right thread. The particular thread I have in mind is commonly called ‘contextualism’, a label that covers both lively debates in epistemology and in the philosophy of language. My concern will be with the latter, and the little dialogue that opened the paper can be used to sketch the shape of the debate. 2. A first step in an attempt to explain the interchange between Michael and Alison might go as follows: (i) Michael and Alison uttered sentences in English, and these have the meanings they do in virtue of the meanings (or senses) of their words, and the meaning-relevant effects of their sub-sentential principles of construction. Both Michael and Alison, as speakers of English, are able to take advantage of those meanings (and effects) in their communicative interchange. Moreover, though only implicit in the description of that interchange, it takes place against a background that extends beyond the fact that the words used have the meanings they do. Included here are facts about each of the speakers, and facts about what each of them knows or believes, and knows or believes the other knows or believes, and more general facts about the world which together constitute the context, or separately (depending on one’s view) contexts of the interchange. In sum, the communicative success of the interchange is brought about by what the participants know about the meanings and context(s) of the relevant sentences. Of course, the above description barely scratches the surface (though I will find a use for it later). What is wanted is a more detailed account, both of the meaning/sense properties of the sentences Michael and Alison used, and the contextual factors that, together with these meanings, provide a genuine explanation of communicative success. Here is one kind of continuation:

Truth, Meaning and Contextualism

145

(ii) The way to understand the idea that the sentences Alison and Michael use have meanings is to see them has having truth conditions which are, in turn, the result of the truth-conditional contributions of their words and constructions. Consider, for example, Alison’s sentence: Those keys are in the sideboard where you put them yesterday. Some of the words in this contribute to truth conditions in a way not dependent on context. Thus, given its lexical content, ‘sideboard’ can be counted on to have a satisfaction condition which is intelligible without reference to the context in which Alison used it. However, the plural demonstrative ‘those keys’, the pronoun ‘you’ and the noun ‘yesterday’ are all context-dependent expressions, so we cannot understand their contribution to truth conditions without somehow calling on context, and there is an ample choice of familiar ways of accounting for this context-dependence. Of course, context’s contribution to our understanding of the communicative interchange doesn’t stop here. Even in the simple scenario above, it is not difficult to recognise a kind of undertone to the interchange which is context-dependent and which contributes, not to the truth conditions of the sentences used, but to something we might think of as its social intelligibility. Michael’s initial sentence has the form usually found in an assertion, but of course Alison clearly takes it to be a question. Moreover, the tenor of her reply suggests her exasperation in her having been asked it in the first place. (One has to imagine the context here, but then again one usually has to do this.) In sum, the way to understand the contribution that context makes to the interchange is to see it contributing, on the one hand, to the truth conditional content of sentences used in that exchange and, on the other hand, to the (broadly) social/psychological point or points of using sentences with those truth conditions there and then.

One way to put the message of the (ii)—one I will go along with for the moment—is to say that truth conditional content is the joint product of the semantic features of the words and constructions in the relevant sentences, and the semantic contribution of context. The further contribution of context is to be counted as pragmatic: it lays bare features arising from (broadly) the further purposes of participants in using the sentences they do in the communicative interchange. One recent expression of what is roughly this view comes from King and Stanley (2005). Their view has it that the sentence used by Alison has the semantic properties it does, partly in virtue of the semantic properties of its words and their modes of combination, but partly also in virtue of a certain

146

Samuel Guttenplan

restricted kind of contextual input which they call ‘weakly pragmatic’. The weakly pragmatic contribution of context is that which can be directly linked to words or constructions (or possibly hidden syntactic markers) in the sentence, whereas strong pragmatic contextual effects—the proper subject matter of pragmatics—consists in the contribution that context makes when it cannot be sheeted home to any overt or hidden item in the relevant sentence. Thus, fixing the truth conditional content of ‘those keys’ or ‘yesterday’ in Alison’s sentence is a weakly pragmatic contribution of context, whereas the fact that Michael’s initial sentence was taken as a question is a strongly pragmatic effect. As noted, account (ii) of the Michael-Alison interchange is controversial partly because of the way it draws the semantics/pragmatics distinction, and partly because of the way it uses the idea of truth conditions to characterise the semantic side of this distinction, but mostly because of the way it suggests using the semantics/pragmatics distinction in explaining communicative interchanges. I will have more to say about truth conditions, and, by implication, something to say about the semantics/pragmatics distinction, but understanding this third area of controversy will take us directly to the debate about contextualism which is central to my paper. 3. Attempt (i) to explain the interchange between Michael and Alison appealed only to the notions of meaning and context. The further story (ii)—the one involving talk of semantics, pragmatics and their respective contributions—was intended to elaborate, underpin or explain the relatively common-or-garden notions of meaning and context. Roughly, semantics (where this includes so-called ‘weakly’ pragmatic input to semantics) is supposed to account for the meaning-properties of the sentences that Michael and Alison used; it is supposed to generate truth conditional contents for those sentences which are available to Michael and Alison, and which therefore contribute to the success of the interchange. As already noted, context plays two roles in this: when required by certain overt, or possibly hidden words or constructions in the relevant sentences, it contributes directly to their truth conditional content; when sentences do not contain these context-apt ‘hooks’, the effect of context is more diffuse, contributing not to the understanding of the sentential content but to the understanding of the social and psychological point of the interchange.

Truth, Meaning and Contextualism

147

Against this background, the contextualist has this basic complaint: this picture fails to accord the proper role to context, at least partly because it gives semantics a role it cannot support, but mostly because it doesn’t give context its due. This complaint can take one of two forms. Borrowing a term from King and Stanley, though not using it in exactly the way that they do, one form of the complaint could be called ‘contextual intrusion’, and for reasons which will be obvious, the other will be called ‘contextual invasion’. Citing specific examples, intrusionists maintain that there are features of context which operate, not on specific words and constructions in sentences, but, instead, globally on the sentences as used in relevant contexts. In effect, they maintain that, for many specific kinds of sentence, semantics, as described in the picture above, does not by itself result in appropriate contents for those sentences—contents that account for what actually goes on in communicative interchanges. King and Stanley speak of ‘pragmatic’ rather than ‘contextual’ intrusion because they take as read their division of contextual effects into those which are weakly or strongly pragmatic. For them, pragmatic intrusion is essentially the intrusion of strong pragmatic effects, and while there is no harm in this terminology, it is too closely tied to their conception of the semantics/pragmatics distinction, and it makes it difficult to understand why the issue in the literature concerns contextualism, and not something that might be called ‘pragmaticism’. I shall not have a lot to say about contexutual intrusion, but it is probably a good idea to have an example of it. King and Stanley discuss a number, including this one: (1) I am not happy, I’m ecstatic. It has been claimed that the negation in (1) is difficult to interpret, unless one sees it as operating, not on the semantic content of ‘I am happy’, but on the contextually salient suggestion (implicature, as it is better known) that ‘I am not happy’ brings in train. For in usual cases, someone who asserts ‘I am not happy’ implicates something like: I am miserable, depressed, etc., and in (1) it is this implication that is negated. Now (1) is only a very gentle example of contextual intrusion, and King and Stanley are quick to argue out that it can be resisted by telling a more complicated story about the semantics of the negation in (1). If this story is found acceptable, and they suggest that there are independent reasons for this, (1) turns out to be just like any other case in which context tamely makes what is in the end a semantic contribution by hooking onto

148

Samuel Guttenplan

a syntactically appropriate item, in this case ‘not’. Of course there are more difficult cases than this one, and King and Stanley spend some time discussing them. But the result they believe is the same: in every case where context threatens to intrude globally on the assignment of semantic content to sentences, one finds that what we really have is the perfectly ordinary semantic contribution that context makes to relevant words and constructions within such sentences. As the label implies, contextual intrusion arguments turn on one’s finding specific kinds of sentence that, when used in certain contexts, seem to have a content that cannot be traced back to the semantic assignments made to their words and constructions. The intrusionist thus exerts pressure on the idea that the semantics of words and constructions (where this includes the semantic contribution of context linked to those words and constructions) generates truth conditional content for sentences in which these words and constructions figure. However, as is obvious, this pressure is exerted on a case-by-case basis, and it thus contrasts with what I have called contextual invasion. This is not to say that the invasionist eschews argument by cases, it is just that those used are meant to show something more general about our understanding of linguistic interchanges. Here is a clear statement of intent: Truth conditional content depends on an indefinite number of unstated background assumptions, not all of which can be made explicit. A change in background assumptions can change truth-conditions, even after disambiguation and reference assignment. That is, even after disambiguating any ambiguous words in a sentence and assigning semantic values to any indexical expressions in the sentence, truth-conditions may vary with variations in the background.1

And she adds a little later on: The contextualist argues for the radical context-dependence of what is said. Yet we clearly have no trouble in understanding what other people say, despite this radical context-dependence. If we assume that this understanding is not magical, there must be a systematic account to give of what is understood (the truth conditional content of utterances) and how it is we are able to know this (what our semantic knowledge consists in and how this knowledge is used in context to understand what is said).

1 Bezuidenhout 2002, p. 105.

Truth, Meaning and Contextualism

149

Together these imply that the picture given by King and Stanley (and by many others before them) is fundamentally mistaken. Semantics, even when this includes any contextual contribution to the semantics of this or that word or construction, falls short of delivering truth-conditional content. And Bezuidenhout, along with many other writers, sees the mastery by speakers and hearers of such content as necessary to their being able to participate intelligibly—not magically—in linguistic interchanges. As noted, the basic argument for contextual invasion tends to proceed by way of an examination of cases. Though these cases are often discussed in enormous and hotly contested detail, what matters to this paper is the shape they impose on the argument, not their details. Still, as with contextual intrusion, it will be good to have a headline case, and without suggesting that it is the most powerful example, let me borrow one from Bezuidenhout: My son comes in from the backyard and when I ask him what he has been doing he replies: ‘I’ve been playing baseball’. Is what he says true? Well, the game he was playing resembles standard league baseball games only rather remotely. There certainly is no baseball diamond in our backyard. In the game my son plays with his father and our dog in our backyard, bases are marked by three trees that stand in a very rough diamond shape with respect to ‘home plate’ which is itself a rather poorly defined place somewhere at the fourth point of the rough diamond. The game is played with only a pitcher and a batter. When the batter makes it to a base, he leaves an ‘invisible man’ on base and returns to bat. The dog plays in the outfield. Sometimes he returns the ball to the batter and sometimes he chases the runner around the bases with the ball in his mouth, but not in any predictable way. Yet this joint activity counts as playing baseball, as playing baseball is understood in this context. So if my son was in fact playing baseball on this understanding, then what my son says is true.2

The point of this example, and others like it, is that of convincing us that the semantics of expressions cannot be depended upon to yield appropriate truth conditions. In the above setting, the sentence ‘I’ve been playing baseball’ has a truth condition—one that, in the circumstances makes it true—but a semantic account of its key predicate—‘playing baseball’— does not get right the truth-conditional contribution one would expect of this expression. And this is so even when semantics is understood in a liberal enough sense to allow it to appeal to context. This is partly because,

2 Ibid., p. 106.

150

Samuel Guttenplan

as the example suggests, there is an indefinitely large range of things that can count as satisfying ‘playing baseball’. It is partly because, even aside from the fact that there is no obvious extra syntactic ‘hook’ onto which one can hang some particular interpretation of this predicate expression, the range of things which count as ‘playing baseball’ cannot be summarised in a way amenable to syntactic hooks. But it is mostly because, once one takes on board the difficulties revealed in this example, one can see that they could well show up in the use of virtually any predicate expression. Because the upshot of these considerations will prove crucial to my central contention, it is worth spelling them out a bit more patiently. 4. If we take the quantifier in the sentence: (2) Every book is mouldy, as unrestricted, we get an absurdly false claim. Of course, when this sentence is used in a typical context, speakers and hearers will understand the quantifier as restricted to a particular domain of books—perhaps those in the attic—and whether actually true or false, it will be plausible. Now while some believe this kind of example supports contextual intrusion, Stanley has argued (in many places3) that it does no such thing. Ignoring details, what he argues is that there is a covert marker of domain in the quantified expression ‘All books’, and that this marker is targeted by context. Stanley’s point is that the presence of this covert marker allows context here to make the same kind of contribution to truth conditions as it does in respect of expressions such as ‘she’ or ‘that’. Using their terminology, King and Stanley would say that context here makes a ‘weakly pragmatic’ contribution, one that we would be well within our rights to count as semantic. As Stanley also claims, there are cases in which context makes what some might regard as a contribution to truth conditions, but in which it would be absurd to think there are covert markers. For example, in (3) there is no reason to search for a marker of ‘manner’, nor for the ‘source of the food’. (3) John ate his breakfast. Thus, if he ate his breakfast with his hands, or was fed intravenously, or if John’s breakfast happens to come from Mars, these contextual features do not have a syntactic marker on which to batten, and they therefore cannot 3 For example, Stanley 2000, Stanley and Szabó 2000, and Stanley 2002.

Truth, Meaning and Contextualism

151

be semantically house-trained. This means that, if we have the intuition that one or the other such circumstance does affect the truth conditions of (3), this sentence becomes a problem for those who think that speakers and hearers use their semantic understanding to access the truth conditions of sentences. But, as has been argued by Stanley, among others, it is far from obvious that judgements about the truth of (3) vary with the manner in which John ate, and the planet from which his food came.4 Matters are not so straightforward with the baseball-playing example. The first of the worries cited above is that there is an indefinitely wide range of activities which, in certain circumstances, could be judged as satisfying ‘playing baseball’, but which in other circumstances would not. Given that these judgements of truth are all over the place, it is not obvious how one could count ‘playing baseball’ as contributing something invariable to the truth conditions of the son’s sentence: (4) I’ve been playing baseball. By itself, however, this is not a definitive problem. If one could find some way to harness the contextual variations to the variations in truth value, one could still count the contribution of ‘playing baseball’ as semantic, at least in the liberal sense. But Stanley’s model of how this might be done in the case of quantified phrases seems hopeless here. On the one hand, there seems no obvious reason to think that there is some covert syntactic marker in ‘playing baseball’. In the quantifier case, there are arguments that suggest we might need, or have independent reason for, such markers—arguments involving issues of quantifier binding—but these do not transfer to the present case. But—and this is the second of my earlier worries—the presence of such a marker is only of use if we can specify the kind of contribution that context makes, a kind of contribution which brings order to the otherwise unruly behaviour of the predicate. And there seems no obvious candidate here. In the case of quantifier phrases, the marker holds a place for ‘domain of quantification’, and we understand how specifying this, case-by-case, might allow us to make sense of truth-value judgements that would otherwise be puzzling. But how could we capture the kind of contribution that context might make to the ‘playing baseball’? Should it be ‘manner’, ‘style’, ‘duration’ . . . —the list is open-ended.5 4 See Stanley 2005, pp. 224–5. 5 The idea there might be a marker for ‘respect’ in which one is said to be playing baseball might be tempting, but is in the end no less hopeless. See Bezuidenhout 2006, p. 4, esp n2.

152

Samuel Guttenplan

In view of all this, why not just adopt the fall-back strategy that Stanley recommends in respect of (3)? What is required by this strategy is, basically, that we ignore any contextual variations that might crop up. As we saw in a previous case, whether he ate with his hands, was fed intravenously, or, whether his food came from Mars, these just don’t figure in the truth conditions of (3). Stanley writes: . . . when someone tells us that John ate this morning, we assume that he did so in the normal way. But no one would deem an utterance of (3) false if, contrary to the default assumptions, they discovered that John ingested his breakfast in some non-standard way, such as being spoon-fed. So the manner of eating is no part of the intuitive truth conditions of (3).6

However, unlike (3), the ‘intuitive’ truth conditions of (4) cannot be so easily ignored, for, as Bezuidenhout insists, there will be contexts in which what is recognisably the same activity will be judged to satisfy ‘playing baseball’ and others in which this would be denied. I will come back to this important issue below, but the point now is that there is prima facie reason to think that we cannot simply look the other way—as we can with ‘ate’—in respect of the variation one finds in the use of the predicate in (4). Moreover, and this is the final worry raised earlier, there is nothing special or restricted about the baseball example, or about the other similar cases one finds throughout the literature on contextualism. Though they can often seem recondite or even forced, once one gets the hang of them, one can come to think that any predicate can suffer the same fate. Even the simple dialogue between Alison and Michael might lead down the same path, since it would take only a little imagination, for example, to think of ways in which ‘in the sideboard’ could be made to show relevant contextual variation. It is precisely because virtually any predicate expression can succumb that what is in question is best described as contextual invasion. But invasions, even philosophical ones, can be resisted. And the logic of the examples intended to spear-head the invasion, has been rejected by various writers, perhaps most robustly by Cappelen and Lepore 2005 (hereinafter ‘C&L’). 5. Given the complications that have piled up in the literature, detailing the anti-invasionist side of the case could be a full-time occupation. But presenting what is, I think, the central focus of disagreement will 6 Stanley 2005, p. 224.

Truth, Meaning and Contextualism

153

suffice.7 The invasionist presents examples meant to show that the truth conditions of virtually any sentence are not made available to speakers and hearers by the semantics of its contained expressions, even when we allow contextual contribution to that semantics. The anti-invasionist strategy of C&L consists basically of two elements, one defensive and one offensive. First, and defensively, they ask us to focus on the distinction between what a sentence means and which speech acts it effects. For example, Alison’s sentence: (5) The keys are in the sideboard where you put them yesterday, can figure in as wide a variety of speech acts as there are correct ways of reporting what went on. Thus, someone might say of Alison’s use of the original sentence that she: told Michael where the keys are, reminded him where the keys are, chided him, showed her exasperation with him, responded to what she took to be his question, told him to search in the sideboard, told him to look where he had himself put the keys, asked him to think what he had done with the keys yesterday.

C&L note that there is no reason to take any one of these as giving the truth conditions of the original sentence. Indeed they regard it as a fundamental mistake made by many (sometimes, one gets the feeling, most) theorists to think that we can move easily (in one or both directions) between whatever speech act is performed by the utterance of a sentence and the truth-conditions of the sentence itself. The latter, they maintain, is exactly what semantics is responsible for, and only confusion—and error— can come of thinking that speech act theory can make a direct contribution to this task. Cleaving to this distinction, they regard the examples offered by the contextual invasionist as model exhibits of such confusion and error. 7 Enormous amounts of energy have been expended in arguing about conditions for context-dependency, but, as will be clear, this is not something central to this paper. Nor, for that matter, is it clearly crucial for the debate, since neither side thinks that there can be necessary and sufficient conditions for context-dependency tout court.

154

Samuel Guttenplan

Returning to the Bezuidenhout example, C&L would treat (4) as having the following pretty straightforward truth conditions (leaving aside the ‘I’ which I have dealt with only cursorily below): (6) ‘I’ve been playing baseball’ is true iff the speaker has been playing baseball. And they regard all the stuff about what specific type of game makes (4) true or false, in this or that context, as relevant to our assessment of the reports of speech act or acts that (4) figures in, rather than as relevant to its truth conditions. Thus, the mother in the example can correctly believe and report that her son said he was playing baseball; she can also report him as having said that he was playing a game with his father, filing in time before lunch, etc. In some contexts the content of the reported speech act will be counted as expressing a truth (e.g. in the context where the mother tells a friend what her son said about what he was doing), in others a falsehood (e.g. in the context where a major league scout who lives locally, and having asked if anyone in the his neighbourhood can play baseball, is told that so-and-so’s son plays baseball). But these differences are neither here nor there in the face of the claim that (6) suffices to give the truth conditions of (4). Having shunted the problematic cases into the siding of speech act theory, and insisted that semantics yields a perfectly good truth condition for (4), C&L are not content to rest with this merely defensive move. The second element of their strategy sees them taking the fight to the enemy. Again, the details of this can be daunting, but the underlying point is simple enough: C&L argue that the invasionist makes it difficult, if not impossible, to understand how we ever manage to communicate with one another. This is because, as they see it, if the truth conditions of virtually any sentence depend on the particular context that figures in its use, then our understanding of that sentence would indeed be magical. In a recent confrontation with Travis, they write: We claim that radical contextualists [i.e. my invasionists] are stuck with a kind of ‘contextual content solipsism’ where contents cannot be carried from one context to another. We claim radical contextualism implies that the same proposition can’t be expressed by people differently situated. If Travis were right, it shouldn’t, for example, be possible for us to say what Travis said, and Travis shouldn’t be able to say what we said.8

8 C&L 2006, p. 57.

Truth, Meaning and Contextualism

155

In his own commentary on C&L, Travis anticipates this criticism, and writes: What communication requires, C&L suppose, is that from the fact that Max said, on some occasion, ‘Sid Grunts, of Sid (at time t), those words meanings in Max’s mouth what they do mean, one can extract what it is that Max thus said to be so. R[adical] C[ontextualism] of course denies that one can do that. But why in the world should one need to? Perhaps their idea is this. If Max’s words are to be any use to me, I must be able to identify precisely the proposition they expressed. To do that would be to identify precisely which understanding they bore . . . But (their idea would continue), to do that you would really need to be there (at Max’s speaking). On the RC view, though, communication just isn’t like that. True, Max’s words are worth little to me if I know nothing of the circumstances in which he spoke. But knowing something of them may be enough to know something of what is to be expected if things are as he said.9

This rejoinder is in turn greeted with a kind of derision by C&L: Accordingly, we can never grasp exactly what another says; we can only muster something similar; something that will do for whatever purposes we are up to . . . In [C&L], we advance a battery of arguments for why we abhor this way of construing communication . . . [W]e find it shocking that Travis is willing to reject that component of our self-understanding as linguistic creatures that requires that we are capable of understanding each other across contexts.10

Nor is the issue of how invasionists might explain linguistic communication local to the interchange between C&L and Travis. Bezuidenhout, noting that she has not herself presented ‘anything like a full-blown contextualist theory of meaning’, has this to say of what would be required of such a theory: How is it that these two characteristics [the reference here is to something Travis claims, but could as well refer to the views of the other contextualists she cites] allow users to understand what speakers say, given that what speakers say goes beyond sentence meaning and relieves crucially on background assumptions not all of which can be made explicit? It is not surprising that no contextualist has yet spelled this out in any detail, for it calls for

9 Travis 2006, p. 46. 10 C&L 2006., pp. 58–9. C&L note that they are currently developing these arguments into a book, Shared Content and Semantic Spin.

156

Samuel Guttenplan

something beyond an account of language. It calls for a general account of how our inferential abilities, of how our memories are organised and retrieved, of how we are able to integrate action and belief and of much more besides.11

To which, it is quite predictable that C&L would respond: it calls for no such thing. The truth conditional biconditionals made available, as in, say, (6), give us what we need (so far as ‘playing baseball’ is concerned) to explain how hearers come to understand sentences such as (4). These citations chart one front in the war, and it is clear from the current literature that that war continues unabated. Certainly, it is not yet possible to say whether the contextual invasionists will come to hold sway over the domain of linguistic meaning, or whether blockade and counteroffensive will prevail. I have given these rather long citations, not merely to give one the shape and flavour of the debate, but because they contain claims that will be crucial to my central contention. Perhaps surprisingly, I think that one can find in them something rather basic that is shared by both sides but which is nonetheless misguided. Since nothing in this area is simple, I think it best to preface my account of this shared mistake with some observations, first about truth conditions, and second about context. 6. Central to both sides of the contextual invasionist debate is appeal to the notion of a sentence’s truth condition, or the proposition it expresses. Invasionists deny that there is any such thing, as witness: RC’s [i.e. invasionists] deny that (e.g. English) sentences are in the business of being true or false. They are not (RCs hold) because there is, systematically, no such thing as ‘that which a sentences says to be so’.12

And opponents such as C&L insist that there is, as witness13: Any utterance of: Rudolf has had breakfast, is true just in case Rudolf has had breakfast, and expresses the proposition that Rudolf has had breakfast.

The ‘any’ that C&L use is important. For while the above citations might make it seem as if the issue turns on the notorious distinction between 11 Bezuidenhout 2002, p. 127. 12 Travis 2006, p. 40. 13 C&L 2005, p. 3.

Truth, Meaning and Contextualism

157

sentences as abstract objects and the linguistic acts of utterance in which they figure, this would be a mistake. For though it C&L do often speak of the truth condition of this or that sentence, they make it quite clear that it is the sentence uttered that has this truth condition. And when Travis denies this, he is not merely denying that an abstract object has a truth condition; he is also denying that it makes sense to speak of the truth condition of the utterance of a sentence. His view, along with other invasionists, certainly allows that the utterance of a given sentence in its contextual setting will have a truth condition—indeed that is the beginning of the truth conditional pragmatics programme that Bezuidenhout imagines. But the invasionist insists that there is no reason at all to think that this truth condition is the one that any utterance of that sentence will have. One way to capture what separates the two sides is their attitude towards the biconditionals that figure so prominently in the literature on meaning. In the above citation, C&L claim that the relevant biconditional gives the content of a sentence—what they often call the minimal proposition expressed by that sentence—and they argue that only a case in which the antecedent of such a conditional is true and its consequent false would be a counterexample to their view. Thus, in response to any one of the invasionists cases (e.g. the one involving the predicate ‘playing baseball’), they would say: To refute the view that playing baseball is that way an activity is iff it is playing baseball, you need to convince us that the biconditional is false. How would you do that? You need to find a case in which some activity satisfies ‘playing baseball’ but isn’t playing baseball (or the other way around).14

In response, invasionists could with some justice say: Providing such a case is certainly impossible, but that is beside the point. This is because we deny that this biconditional, wrenched out of its various contexts, provides anything like a content for ‘playing baseball’, or for sentences in which this predicate figures. We can only make sense of such biconditionals fulfilling the task that Davidson and others have set them if they lead speakers and hearers to make a consistent pattern of true/false judgements across a range of linguistic interchanges. But when one looks at this range, the variability in such true/false judgements shows this to be impossible.

14 This is pretty much a direct quote from C&L 2006, p. 55 that has been adapted to the baseball example.

158

Samuel Guttenplan

The issue of who is right in this particular interchange—fundamentally right, not merely right about the burden of argument—is the basic thing dividing the two sides. However, the observation I want to make concerns an understanding of ‘truth condition’, or ‘proposition expressed’ that seems to be shared by both parties. It is this: both sides assume that if a speaker uses a sentence which has a truth condition or expresses a proposition, then the grasp of this by the hearer will at least in principle allow the hearer to determine the truth value of the sentence that the speaker uses.15 That is, they both presume that a speaker’s mastery of the truth condition of a sentence prepares her to judge the sentence true or false. (I shall ignore the complications that would be needed to transpose the discussion for the case of ‘proposition expressed’, but the cases are parallel.) The evidence for this is on the part of invasionists is everywhere, and lies behind every one of the complex examples that support their case. Whatever the details, each such case is intended to show that mastery of biconditional-style truth condition for the relevant sentence will not enable a hearer to form an adequate view of the actual truth or falsity of that sentence when used on some occasion, even though it purports to do this. This is apparent in Travis’s claim (above) that English sentences are not ‘in the business of being true or false . . . because there is, systematically, no such thing as “that which the sentence says to be so”.’ And it comes out in other ways, even if these are not always put with sufficient caution. For example, Recanati writes: The central idea of truth-conditional semantics . . . is the idea that, via truth, we connect words with the world. If we know the truth conditions of a sentence, we know which state of affairs must hold for the sentence to be true.16

And I am sure that we are meant to take the italicised phrase ‘which state of affairs’ as directly referring (via what is intended as a relative pronoun) 15 Here and in what follows, I leave out the obvious further fact that mastery of a truth condition isn’t by itself enough to decide truth value: one has also to be appropriately placed epistemically. This can sometimes cause confusion: it can seem as if the assumption I am describing requires that a truth condition already includes something like a method for verifying the truth or falsity of the sentence. But talk in the text of a truth condition as ‘enabling’ or ‘equiping’ someone to decide truth value should be understood as saying no more than that it would so enable or equip someone who was appropriately placed epistemically. 16 Recanati 2005, p. 185.

Truth, Meaning and Contextualism

159

to some bit of extra-linguistic reality, as witness something Recanati says on this topic (though in another place): . . . there obviously is another way of eliciting truth-conditional intuitions. One has simply to provide subjects with scenarios describing situations, or, even better, with—possibly animated—pictures of situations, and ask them to evaluate the target sentence as true or false with respect to the situation in question.17

The assumption about the relationship between truth conditions and truth value tends also to be made by anti-invasionists. For example, C&L imagine the following question put to them: How can speakers go around making assertions by uttering sentences that semantically express propositions [i.e. have truth conditions] they don’t even believe (since they certainly can’t believe the semantic content unless they have beliefs about the actual truth values)?

And they refer the reader to the following chapter (‘Semantics and Psychology’) where, as part of their answer to the above question, they write: We think . . . that there is a minimal semantic content or proposition expressed that is semantically expressed by (almost) every utterance of a wellformed English sentence. This proposition is not a ‘skeleton’; it is not fragmentary; it is a full-blooded proposition with truth conditions and a truth value . . . We think . . . that this minimal semantic content is an essential

17 Recanati 2004, p. 92. as cited in C&L 2005, p. 197. C&L find the talk of ‘animated pictures’ bizarre, and the garbled talk of ‘scenarios describing situations’ certainly doesn’t help. In any case, it is clear enough what Recanati is getting at: he wants us to understand talk of truth conditions as somehow bringing worldly things directly into discussion. Unkindly, perhaps, but accurately, one could see Recanati’s aim as an updated version of what happened in Swift’s Academy of Lagado in Gulliver’s Travels. Members of that Academy carried around objects so as to avoid the difficulties of words; Recanati recommends scenarios and pictures. The observation I make in the text is not concerned with this aspect of Recanati’s proposal, but rather with his assumption that mastery of a truth condition of a sentence enables one with this mastery to judge whether the sentence is actually true or false. Clearly there is a connection between a ‘Swiftian’ conception of truth conditions and this latter point, but the two are separate. For example, I don’t think that Travis would share the idea of scenarios and pictures with Recanati, but he does insist on the relationship between truth conditions and actual truth or falsity.

160

Samuel Guttenplan

part of all communicative interactions. The minimal semantic content has a function in the cognitive life of communicators that no other content can serve.18

Alright, so both sides seem to think that when someone grasps the truth condition of a sentence, this equips her with a crucial part of what is needed to judge the sentence true or false.19 Aside from the fact that both sides disagree about whether semantics do in fact yield truth conditions, is this a problem? Well, though I don’t have any direct argument (yet) to show that it is, my aim just now is more observation than argument. And my observation here is that this assumption is not one we have to make, even if it is encouraged by talk of truth conditions. There are ways of understanding truth-conditional accounts of meaning which do not see them as furnishing speakers and hearers with specific resources enabling them to make judgements about the actual truth or falsity of sentences. More on that after an observation about the notion of context. 7. Everyone who takes sides in the contextualist wars, uses the notion of ‘context’ as if understanding it was pretty straightforward. Now it is obviously possible to question whether we really do understand what kinds of thing go to make up the contexts that figure so prominently in the literature. But, even though this pedantic stance has something to be said for it, it is not one that I will take up here. Indeed, I admit to having the perfectly ordinary reaction that I suspect we all do to examples of this or that effect of context. When words like ‘I’ or ‘yesterday’ are in question, I go along with the assumption that the context includes the speaker or the time of utterance; when a demonstrative is used, that the context includes a speaker with some relevant object-directed intention which is known or at least knowable by any hearer. When Michael tells Alison that he has looked for the keys everywhere without success, I don’t have any trouble imagining why his saying this is counted as a question for Alison: the context has it that he believes she put the keys some place where he couldn’t find them, and that she will understand his hyperbolic claim as a plea to reveal where they are. When she produces her reply, it is also easy enough to imagine what makes what she says an expression of only barely concealed contempt: the context is such that she is exasperated with 18 C&L 2006, p. 181. 19 See note 15 for an explanation of why I say ‘a part’ here.

Truth, Meaning and Contextualism

161

Michael who she regards as having the childish tendency to blame others for his own failings. In sum, while it is surprisingly difficult to tell a single comprehensive story about what a context is, getting a grip on this notion in—well, here I have to say ‘specific contexts’—is not difficult. In this way, context resembles another notion in this area, viz. meaning. As that concept figured in the attempt (i) above to explain the interchange between Michael and Alison, it is perfectly familiar: each of the sentences that Michael and Alison used were said to have meanings—meanings that themselves result from the meanings of the sentence’s words and constructions—and it was claimed that the participants’ grasp of these is what, at least partly, makes their linguistic interchange work. But, as we know, though meaning is easy enough to employ in this informal kind of way, proper philosophical accounts of it are anything but straightforward. That the notion of context is both commonsensically familiar and philosophically difficult, and its sharing this with the concept of meaning, suggests something further. In its ordinary employment, we find meaning something it is natural to attribute to, or project onto, items in the linguistic interchanges they participate in. Some have, I think rightly, insisted that we do this in the direct, unmediated, way in which we perceive the world when our eyes are open. But this is controversial, and the point about attribution is not meant to be. For whether or not it has the directness of perception, what talk of attribution or projection suggests is that the notion of meaning is something we employ to make sense of linguistic interchanges, rather than something that makes them possible in the first place. Put another way, meaning is the upshot of various communicative interchanges, not their determinant. My suggestion—and this is second of my tentative observations—is that something like this might well be true of the notion of context. In the hands of theoreticians, context can seem to be one of the things which, even if only implicitly, speakers and hearers exploit to make sure that their communicative interchanges work; context, on this understanding, counts as a determinant of those interchanges. But, following the suggestion above, mightn’t it be better to think of context instead as the upshot of communicative interchange? This would fit better with the perfectly ordinary employment of this notion. It certainly seems to be something we find it natural to attribute to successful (or unsuccessful) interchanges, rather than something which we take ourselves to have access to independently of any such interchange.

162

Samuel Guttenplan

8. Underlying my observations about truth conditions and context is a certain conception of the semantic project shared by many writers on both sides in the contextualist debate. My observations in the previous two sections were meant to register doubts that might be traced back to that conception, and in this section I should like to confront it more directly. Central to that project is the demand that the materials with which speakers and hearers are credited suffice for explaining the success of their linguistic interchanges. And, even without worrying too much about the niceties of this contested notion, it is clear enough that the idea of explanation here is broadly causal/explanatory: the materials the theorist credits to speakers are thought of as bringing about, or being antecedents to, any such success. Using our earlier example, the idea is that those who have mastered the predicate expression ‘playing baseball’ can deploy it so as to achieve communicative success amongst those similarly endowed. (Of course, there must also be mastery of other words and constructions to provide a setting for ‘playing baseball, but for the present point we can assume that this is in place.) When the mother in Bezuidenhout’s example hears her son say: ‘I’ve been playing baseball’, then, assuming that the son has mastered the materials—the package—that goes with ‘playing baseball’, his use of this predicate will result in (or at least constrain) his mother’s understanding. Of course, it is Bezuidenhout’s view that traditional semantic accounts of this package will not work. As she sees it, such accounts understand the package that goes with ‘playing baseball’ as supposedly captured by its contribution to the truth conditions of sentences in which it is used: in the familiar idiom, some activity satisfies ‘playing baseball’ iff that activity is playing baseball. And given that this contribution is a truth conditional one, any hearer also in possession of the relevant package, will be enabled to say whether the son’s claim is true or false. However, as we saw, Bezuidenhout’s worry is that the standard truth conditional characterisation of the package will do no such thing. For, in a different context, one might well perfectly well possess the package characterised by the biconditional satisfaction condition, while being unable to judge, or judge correctly, the claim’s truth or falsity. What this failure shows, or so say the invasionists, is that the traditional picture of truth conditional semantics cannot be depended upon to give us an adequate characterisation of the package that explains successful communication. For that, we need to tell a much more complex story

Truth, Meaning and Contextualism

163

that takes context into account, hence Bezuidenout’s view of her project as ‘truth conditional pragmatics’. Defending the traditional conception from contextual invasion, C&L argue first of all that the biconditional characterisation does suffice to determine the truth values of the relevant sentence, and, secondly, that trying to extend the project in the way suggested by talk of truth conditional pragmatics, will actually make impossible any explanation of communicative success. It should be clear how my earlier observations link to the underlying conception that figures in both sides of the debate. First, this conception makes it possible to understand why one would think of a truth condition for a sentence as something which would lead those who use or grasp the sentence to be in a position to judge it actually true or false. Treated as a way of characterising the package that enables speakers and hearers to deploy the sentence (and its parts), it is a natural assumption to make. Thus, the thought is that mastery of the package governing ‘playing baseball’ must surely equip someone with what it takes to frame or judge sentences in which it occurs in this or that context as true or false. Second, having argued that traditional, semantically generated, truth conditions cannot cope with contextual variability, one can understand why truth conditional pragmatists draft context itself in to help with the job. In so doing, context becomes part of the apparatus that speakers and hearers have at their disposal; it becomes a determining element in communicative success, rather than—as was suggested by my observation—something which is the upshot of it. There is, however, an alternative conception of the overall project, one which does not have these consequences. Instead of trying to attribute theoretical resources—packages—which enable speakers and hearers to achieve communicative success, one begins by assuming, what is in fact the case, that we do successfully communicate. Then the aim of the project becomes, not providing a (roughly) causal explanation of that success, so much as a description of what constitutes it. Let me explain. Think first about how we ordinarily understand some stretch of human activity when language is not in question. Though the details can be controversial, the basic outline of the enterprise is not: we attribute attitudes— desires, beliefs, knowledge, intentions, needs, interests, etc.—to the agents involved in the relevant activity, and these attributions are governed (even if only implicitly) by a scheme or framework of what it is rational to

164

Samuel Guttenplan

believe, know, intend, desire, need or have an interest in, etc.20 Moreover, while the framework of rationality—the network of concepts and their inter-relationships—is something that, on specific occasions, we bring to bear on human activity, it is not something we manage from outside that framework. The very idea of our own agency is bound up with the ability to deploy the framework, and deploying it presupposes agency.21 With language, a new range of actions enters the picture, one that fits into, but also requires some expansion of, the framework of rationality. And partly because linguistic actions have a kind of systematicity not present in stretches of non-linguistic activity, the expansion of that framework will have to reflect that systematicity. Here is McDowell on this very topic: The adequacy of the total theory [of sense or meaning] would turn on its acceptably imposing descriptions, reporting behaviour as performance of speech acts of specified kinds with specified contents, on a range of potential actions—those that would constitute speech in the language—describable, antecedently, only as so much patterned emission of noise. For that systematic imposing of descriptions to be acceptable, it would have to be the case that speakers’ performances of the actions thus ascribed to them were, for the most part, intelligible under those descriptions, in the light of propositional attitudes; their possession of which, in turn, would have to be intelligible, in the light of their behaviour—including, of course, their linguistic behaviour—and their environment. The point of the notion of sense—what the content-specifying component of a total theory of that sort would be a theory of—is thus tied to our interest in understanding—fathoming— people. We have not properly made sense of forms of words in a language if we have not, thereby, got some way towards making sense of its speakers. If there is a pun here, it is an illuminating one.22

20 Many writers seem to misunderstand this familiar claim. Most importantly, ‘rational’ here is not some narrowly calculative notion. More in a Kantian (or Sellarsian) way, the framework of rationality is the structure which contains and links those notions (intention, belief, knowledge, desire, . . .) and which thereby defines what it is to be a rational subject /agent. Any such subject /agent can do and think quite ‘irrational’ things, while still being rational in the relevant sense. 21 The whole of the scheme of rationality thus sketched is often described as ‘folk psychology’, and this label suggests, what many think anyway, that it is somehow optional. Though I have no room to argue it here, and though it is not needed for the point I shall go on to make, the claim in the text about the interdependency between the framework and the very possibility of agency is intended to make that kind of optionality seem, to put it plainly, bizarre. 22 McDowell 1998, p. 172.

Truth, Meaning and Contextualism

165

How does this systematic imposing of descriptions work? Well, instead of trying to match contents directly to sentences, one tries to do this by generating the conditions of application of some predicate. That is, instead simply of looking for a systematic way to match each sentence s with its content p, we look for a systematic way to generate instances of: s is ) if and only if p,

where ‘)’ is a predicate of sentences of the language, and p is the contentspecification for the relevant s. If s and p are as described, then there is nothing to prevent our understanding ‘)’ as ‘true’, and thus understanding the systematic generation of contents as a generation of truth conditions. Finally—and this is part of the point of casting the account in this way— the actual business of generating these contents goes via further conditions that are put on the words and constructions are found in the sentences of the language. I trust that the above is familiar enough, but, without distorting it, let me put what might be a less familiar gloss on it. The project of generating truth conditions for sentences in the above way can be thought of expanding the conceptual resources of the framework of rationality. Added to the usual list of such concepts as intention, belief, desire, etc., and its supplementation with general notions relevant to speech act types such as asserting, asking, ordering, etc., the project provides a whole panoply of concepts—those needed at both the sub-sentential and sentential levels— which we can then use to fit relevant linguistic actions into the larger framework of rationality.23 In effect, it is the very point of a truthconditional account that it makes systematically available a large repertory of concepts linked in appropriate ways to words, constructions and sentences in a language. One can think of these as like the dots in a pointillist painting: each one is not anything much on its own, but together they constitute the picture itself.

23 This way of putting it makes it sound as if the resources of the framework might just happen to include these systematically generated further elements. This is, however, merely an artefact of my exposition, for I regard language as in fact integral to that framework, and hence integral to the kind of agency that makes us, in McDowell’s word, ‘minded’. (See McDowell 1998, p. 105, where he insists on the centrality of language to the framework, and attributes the use of ‘minded’ to Jonathan Lear.)

166

Samuel Guttenplan

Here is one rough account of how we might use this repertory in dealing with a stretch of linguistic activity. In response to his mother’s utterance: ‘What have you been doing all morning?’, a boy utters ‘I’ve been playing baseball’. Taking these two bits of behaviour to count as successful linguistic interchange—as a case in which there has been what I like to think of as an ‘attunement’ between the two agents—we can, from our perspective, capture that attunement by, among other things, counting the son has having said or asserted that he is playing baseball.24 In doing this— and again this is only a small part of the larger story—we are using a concept (that of ‘playing baseball’) from the range provided by the truthconditional account described above. In effect, in taking ourselves to understand the interchange, we are taking ourselves also to be in attunement with the mother and son. And, as it happens, we mark that attunement by ourselves using the expression ‘playing baseball’. Nothing in this story involves attributing to the participants ‘packages’ of extension- or truth-value-determining information. Thus, the idea of one or other of the participants deploying such a package to achieve, perhaps impose, communicative success drops away. To be sure, we mark the attunement by using the predicate expression ‘playing baseball’, and one could imagine using this expression on another quite different occasion to mark what we take to be required for attunement in that circumstance. But this in no way undermines the fact that we can correctly characterise, by our lights, the linguistic interchange between the mother and son using this expression. It would undermine it, if we insisted that any use of this predicate expression had itself to be associated with a concept that applied determinately to this or that kind of baseball-related activity; that it is either used of the informal kind of game played in the backyard, or the kind that requires a stadium, a diamond, nine players and the rest, or something in between. But that requirement is one born of the package idea, and it ought to have no place in our understanding of the truth conditional account. Properly understood the truth conditional account is a systematic way of specifying what it is that constitutes the success (or, with suitable modifications, the failure) of linguistic interchange. It is not a psychological or cognitive scientific account of the capacities speakers and hearers deploy to achieve that success (or fail to deploy in cases of failure).

24 I first used ‘attunement’ in Guttenplan 2005 (see pp. 122–28), and though the context was different—I was defending my account of metaphor—the basic aim of that discussion overlaps with the central theme of this paper.

Truth, Meaning and Contextualism

167

This way of looking at the matter shows how there can be what I think is a salutary decoupling of truth conditions and assessments of actual truth value. The truth-conditional account supplies us with the materials we need to describe what goes on in a linguistic interchange, but it does not do this by putting some bit of reality in front of us, whether in the shape of ordered n-tuples of particulars-properties-relations, scenarios, or animated pictures, so the idea that we can directly move from a truth condition to a judgement about actual truth value is inappropriate. Instead, a truth-conditional account uses the words, constructions and sentences of a language to make available what I characterised above as a repertory of concepts. To be sure, in its most familiar guise, this is brought about by the systematic generation of conditions for the application of a predicate which, as it happens, turns out to be ‘true’. However, what this shows is simply that the materials yielded by the account ‘bear on reality’, and this is just what would one expect, given that the truth-conditional account stands in for the ordinary notion of meaning, a concept naturally understood to relate the relationship between words or actions and reality.25 My suggested way of understanding the truth-conditional account also allows us to continue to think of context as we ordinarily do, as a feature we reconstruct in the attempt to describe and rationalise linguistic interchange, rather than as an input to the psychological component required for the production of such interchange. As noted, the truthconditional account aims to describe communicative success, not only in general, but in specific cases. For this, in at least the homophonic case, it calls on the words, constructions and sentences of some language to serve as markers of that success, as markers of the concepts and truths shared in such communication. Of course, as also noted, this does not require us to think of each predicate marker (for example) as somehow permanently linked by the account to some extension-determining concept. Indeed, the very idea that there might be some such concept is problematic.26 But 25 The useful and non-committing phrase ‘bear on reality’ comes from McDowell 1998, p.128, and it figures in his debate with Dummett about modesty in theories of meaning. However, the relevance of the Dummett/McDowell dispute to my concerns is only tangential, and a proper discussion of it would have made this paper much too long. 26 I have in mind here the arguments that make up Wittgenstein’s so-called ‘rulefollowing considerations’. Properly understood, I believe that these fit rather nicely with the picture of the truth-conditional account in this paper. And this is especially interesting, given that Wittgenstein’s arguments are often cited as the inspiration for the invasionist position.

168

Samuel Guttenplan

this does not mean we have to think of each particular use of the predicate expression as merely schematic, as not really standing for a determinate concept. When we describe the mother as understanding what her son said—as accepting that he was ‘playing baseball’—there is nothing merely schematic about it: we are, in that case, counting the mother and son—and us—as sharing a quite specific concept, one marked by ‘playing baseball’. The phrase ‘in that case’ indicates what is I think the proper place for context. As suggested in §7, context is attributed to a successful communicative interchange, and therefore works in much the same way as the other items we use in fitting behaviour into the framework of rationality. 9. So where does this leave us in respect of the invasionist war? Both sides assume it is reasonable to demand the broadly psychologistic account of communication that I have called the ‘package’ view, and then argue about whether the truth-conditional account does or does not satisfy that demand. I have presented a picture of the truth-conditional account which shows such a demand inappropriate: the account is simply not in the business of telling us what equipment is required to underwrite success in linguistic communication. If I am right, this war, like many others, is being fought for no good reason. Still, I can imagine someone responding as follows: Alright, so there is a conception of meaning and its truth-conditional embodiment which doesn’t take itself to be characterising speaker-hearer resources. But speakers and hearers do manage successful communication, and surely there is something they must possess which makes this possible; it is not simply magic. Our concern is with this possession, and our war is being fought over how it is best understood.

A proper counter-response to this ecumenical suggestion is a task for another paper. But, rather satisfyingly, the very arguments that each side deploys suggest a shorter way of dealing with it. The point of those arguments was to show that the materials the ‘other side’ uses are not up to the task of describing the communication-enabling resources of speakers and hearers. And while I have argued in this paper that these arguments are not in fact required—that the basic premise of the invasionist war is misguided—I have not offered any argument against their cogency. However, if the invasionist is right in thinking that a truth-conditional, minimalist account of speaker-hearer competence is not viable, and if the invasionist’s opponent is right in thinking that the contextualist programme is doomed to inadequacy, or, even worse, incoherence, then they will have done my work for me. For together these sets of arguments show the project of giving an

Truth, Meaning and Contextualism

169

account of speaker-hearer competence using resources like that of truth condition and context is hopeless.

References Bezuidenhout, A. 2002: ‘Truth Conditional Pragmatics’. In Language and Mind, Philosophical Perspectives, 16, edited by James E. Toberlin. Bezuidenhout, A. 2006: ‘The Coherence of Contextualism’, Mind & Language, 21, 1–10. Cappelen, H. and Lepore, E. 2005: Insensitive Semantics. Oxford: Blackwell Publishing. Cappelen, H. and Lepore, E. 2006: ‘Response’, Mind & Language, 21, 50–73. Guttenplan, S. 2005: Objects of Metaphor. Oxford University Press. King, J.C. and Stanley, J. 2005: ‘Semantics, Pragmatics and the Role of Semantic Content’. In Semantics vs Pragmatics, edited by Zoltán Gendler Szabó. Oxford University Press. McDowell, J. 1998: Meaning, Knowledge and Reality. Cambridge, MA: Harvard University Press. Recanati, F. 2004: Literal Meaning. Cambridge University Press. Recanati, F. 2005: ‘Literalism and Contextualism: Some Varieties’. In Contextualism in Philosophy: Knowledge, Meaning and Truth, edited by Gerhard Preyer and Georg Peter. Oxford University Press. Stanley, J. and Szabó, Z. 2000: ‘On Quantifier Domain Restriction’, Mind & Language, 15, 219–61. Stanley, J. 2000: ‘Context and Logical Form’, Linguistics and Philosophy, 23, 391–434. Stanley, J. 2002: ‘Making it Articulated’, Mind & Language, 17, 149–68. Stanley, J. 2005 ‘Semantics in Context’. In Contextualism in Philosophy: Knowledge, Meaning and Truth, edited by Gerhard Preyer and Georg Peter. Oxford University Press. Travis, C. 2006: ‘Insensitive Semantics’, Mind & Language, 21, 39–49.

Expression, Truth, Predication, and Context: Two Perspectives James Higginbotham Abstract: In this article I endeavor to untangle several questions about the concept of truth (as applied to potential utterances of sentences in human first languages), and especially the notion of “what is expressed” by an utterance, through a critical examination of some of writings, both early and late, of Donald Davidson and others, concentrating especially on the explanation of predication and the question of predicate reference. I also sketch the Sententialist view of these matters, according to which we allow a redundancy theory of “what is expressed,” but not of truth itself. I compare this proposal to others in the literature, derived from Intensional Logic or refinements thereof, by David Lewis and Jeffrey King.

Semantic theory as commonly practiced recognizes two central notions, namely that of the expression of a proposition by a linguistic form (or its utterance), and that of truth. Because truth applies both to forms or utterances, on the one hand, and to propositions expressed on the other, the data to be accounted for in a theory of truth split in two; but the theory of expression of a proposition may reunite them, reducing the account of the truth of linguistic forms to that of the propositions that they express. Let L be an actual or potential human first language, perhaps our own. Setting aside for the moment all contextual matters, and taking the canonical structural descriptions of the sentences of L for granted, the project is to clarify the status of unexceptionable examples from the usual semantic trio, as in (I)-(III): (I) s expresses in L (the proposition) that p (II) s is true in L l p (III) (The proposition) that p is true l p How do these elements fit together? By a redundancy theory of expression as in (I), or of truth with respect to either (II) or (III), I shall mean any account according to which the

172

James Higginbotham

standard examples need only to be clarified to be seen as in one way or another trivial, or anyway not demanding of a separate, presumably theoretical, explanation. Of course, my basis for calling something a redundancy theory is vague; but then there is a variety of redundancy theories. Redundancy theories of (III) have been advanced on the ground that it is the unexceptionable instances of (III) in our own speech that sum up the point of the predicate ‘true’ in the first place. But complex theories of truth, for which certain instances of (II) are consequences, may also be advanced as redundancy theories, particularly for the case where the metalanguage is merely an extension of L containing the metalanguage predicate ‘true in L’, and enough apparatus to formulate at least an inductive definition of this predicate along Tarskian lines. Again, suppose that (I) is solved for expression of a proposition by a sentence in L. As we are assuming that a sentence is true just in case it expresses a true proposition, any instance of (II) is a consequence of the corresponding instances of (I) and (III), there is no need for a separate account of it. Likewise, the instances of (III) could be viewed as consequences of (I) and (II). There are also redundancy theories, as I intend this notion, of (I). One of these may be illustrated through Richard Montague’s language IL of Intensional Logic (or, more strictly speaking, by a language L for which IL constitutes the logical apparatus, L being further enriched with ordinary names, predicates, and so forth). In that setting, the semantic values of all expressions are relativized to the parameter i of possible worlds, and we may propose a substantial account of the notion (II’) corresponding to (II) above: (II’) is true in L at i l p(i) However, once (II’) is taken care of, the notion of expression follows right along, as we shall have, for every s and i, (I’): (I’) s expresses in L at i (the proposition) ^p where Montague’s hat ‘^’ represents O-abstraction over the possible worlds. (Alternatively, one could start with the expression relation (for all categories) and derive the instances of (II’).) Another redundancy theory of (I) is that of sententialism, as I and others have used this term, following Schiffer (1987). On the sententialist view, clauses ‘that p’ refer to themselves, understood as if uttered. With

Expression, Truth, Predication, and Context: Two Perspectives

173

suitable auxiliary premisses, the instances of (I) and (III) will follow (I give these premisses in Higginbotham (2006)). In this case, the account of expression presupposes the account of truth. Similar remarks will apply on an extension to all occurrences of complement clauses of Donald Davidson’s “paratactic” theory of indirect discourse (Davidson (1968), though I think this theory incorrect, for reasons I will not go into here). There are, then, redundancy theories of (I), and of (II), each of which advances substantive theories of the other. Is there a prospect of taking all of (I)-(III) as redundancy theories? Not if we are serious. If in Oxford I glance at a newspaper in Romanian, I can certainly wonder what the headlines say, and whether whatever it is they say is true. My (extremely partial) knowledge of Romanian gives me some information, but likely not enough. What do the Romanians know that I don’t know? Whatever it is, it is sufficient for them to determine what is said, and the conditions under which it is true, even if I don’t know what they are. In short, the concepts of truth and of expression of something, or of what is said, by a sentence or potential utterance, apply beyond the borders of our individual competences. Assuming, then, that redundancy theories of truth or of expression may be offered, I will turn first to some very general issues for redundancy theories of truth based upon accounts of expression in terms of structured propositions as in King (2007), adverting for this purpose to critical discussion of some proposals early and late by Donald Davidson. I will conclude that Davidson’s considerations (or similar points made by Cartwright (2005) and Schiffer (2003)) are effective only against particular versions of that account; at the same time, these considerations cannot be waved aside through casual, or even sophisticated, talk of properties and relations I shall assume for present purposes that a structured proposition of the sort expressed by an elementary subject-predicate sentence is composed in some way out of the interpretations of the subject and the predicate, and perhaps other things, as expounded and defended recently in King (2007). The problem is to explain how those things, together with the mode of composition, are to be seen as composing something with a truth value. It is straightforward to say, or anyway to construct a theory that implies, that these objects have truth values, and even to say how these values are to be determined; the problem lies in doing so without importing other information. In his posthumous book Truth and Predication (Davidson 2005), Donald Davidson rehearsed much of the long history of a notorious question: how

174

James Higginbotham

should we understand the different roles of subject and predicate (or, in more contemporary terms, of argument expressions and predicate) and their contributions to the sentence? There is a cluster of questions here. What makes for the different roles of argument expression and predicate? Why does each require completion (by the other, putting quantification aside) in order to form an expression that can have a truth value, and give the content of a speech act? Is there a principled distinction between the kinds of things argument expressions and predicates are to be associated with in an overall account of the workings of language, and if there is does that distinction explain the difference in their roles? And if we suppose as usual that the role of an argument expression is to refer to an object, do predicates also refer, and if so to what? Davidson’s views on the above questions comprise three theses, which I believe may be arranged in order of increasing strength. The first is that the reference of predicates must be of a fundamentally different kind from the reference of singular terms; or, to put it metaphysically, the reference of predicates is not to objects. The second is that, on the assumption that revealing the full workings of a language involves giving an inductive definition of truth for that language (and, perhaps we should add, to the extent that it involves only that), we need only say that predicates are distinguished as those expressions that figure in the inductive definition as being true of objects, or sequences of objects, under such-and-such conditions, as spelled out predicate by predicate, by stipulation for the primitives, and by recursion for complex predicates. The different roles of argument expressions and predicates are then explained through the difference between refers to an object, and is true of an object. The third thesis is that even if, as in Frege, the reference of predicates is not to objects in the universe of discourse, but to other things of some special kind altogether, the nature of such reference cannot explain predication. (To explain predication does not mean: to explain it to someone who did not know what it was; for I would have to use predication in order to do that. Rather, it would mean to explain why argument expressions and predicates “fit together,” and why neither argument expressions nor predicates fit together with each other.) I consider these in turn. Davidson’s argument for the first thesis is well-worn, but I am going to rehearse it anyway in a particular form, using one- and two-place predicates, and supposing for simplicity that they refer to classes. The thesis to be disputed then would be: a one-place predicate refers to a class of things, and a two-place predicate refers to a class of ordered pairs. On the view in question we would have, e.g.:

Expression, Truth, Predication, and Context: Two Perspectives

175

(1) ‘Theaetetus sits’ is true l Theaetetus  the class of sitters (2) ‘John loves Mary’ is true l  the class of lovers and beloveds, in that order. Now, the argument goes, we will, having the relation expressed by ‘’ to hand, give the role of ‘’ on the right hand side of (1) by (3), and similarly (4) for the right hand side of (2): (3) ‘Theaetetus  the class of sitters’ is true l  the class of ordered pairs of things of which the first is a member of the second. (4) ‘  the class of lovers and beloveds, taken in that order’ is true l  the class of ordered pairs of which the first is a member of the second. In short, the explanation of predication as relating things, or sequences of them, to classes cannot apply to ‘’ itself, which must be used as a relational predicate in giving the explanation. I take it that such is the “infinite regress” that Davidson remarked as early as “Truth and Meaning” (Davidson (1967)). I have illustrated the issues through the use of membership ‘’, and classes. But they arise equally for any other purported relation R in place of that expressed by ‘’, and for purported objects other than classes. The conclusion then is: the semantic role of predicates can’t be to refer to special objects in the universe of discourse, on pain of being unable to unravel the role of R. The result may be unsatisfying; but that doesn’t make it a regress, at least not yet. Before offering a summary of what the argument does show, I will consider an attempt to defang it. The defender of structured propositions, relating individuals or pairs of them to classes, or to properties and relations, may respond, as in King (2007), that the relation of membership, or analogous relation, that is exploited in securing a truth value for (1) is not given by any piece of vocabulary hidden somehow in that sentence, but rather by the syntax of the sentence. Thus King (2007: 34) writes: . . . we can think of this bit of syntax [concatenation, or the breakdown of S into NP+VP] as giving the instruction to map an object o (the semantic value of an expression at its left terminal node) and a property P (the

176

James Higginbotham

semantic value of an expression at its right terminal node) to true (at a world) iff o instantiates P (at that world).

And in defense of the view that the syntax “provides instructions,” he adds on the same page: Semantic approaches differ only on what they claim is the instruction that a given piece of syntax provides. They are all stuck with the idea of syntax providing instructions.

and gives as an example IL as invoking function application.1 If we take up this last statement as a thesis about linguistic competence, it seems to be false, and false not only for the toy examples of instantiation and membership, but also for accounts such as IL. Taking King’s example first, what do you need to know to know that ‘Theaetetus sits’ (or more precisely the string ‘Theaetetus sits’ together with its syntactic structure) is true (at the world i) just in case Theaetetus instantiates at i the property of sitting? According to the account envisaged, you need to know (i)–(iii): (i) The NP ‘Theaetetus’ refers in English at i to Theaetetus (for every i). (ii) The VP ‘sits’ refers in English to the property of sitting. (iii) A sentence S = NP+VP is true at i iff the reference of NP at i instantiates at i the property referred to by the VP. None of these statements “gives instructions” to anybody: they are just statements of fact, known to speakers of English. In IL the statement corresponding to (iii) would be (iii’): 1 King (2007: Chapter 1) finds structured propositions as n-tuples (or, I assume, labelled trees) inadequate, for at least two reasons. The first is that there are several ways of constructing these objects, and the second, if I have it right, is that one must independently say what endows them with truth values; whereas he wants it to be evident in some way that they do have truth values. The first worry seems to me insignificant, reflecting only the ontological relativity that is characteristic of mathematics generally. The second worry does not arise for IL (where propositions are just sets of possible worlds, and so lack internal structure); but I don’t see that it cannot be answered, so far as an answer is wanted, by a construction that would yield trees analogous to compositional intensions, as in Lewis (1970) (where the label of the root of a syntactic structure is the intension got by combining as function and argument the intensions of its daughters). To put it another way: there is no possibility of doing away with statements such as (iii) governing the syntactic structure T for ‘Theaetetus sits’, or of replacing them with additional structural elements added on to T; for the combinatorial rule that applies to a structure cannot itself be determined by the structure to which it applies.

Expression, Truth, Predication, and Context: Two Perspectives

177

(iii’) A sentence S = NP+VP is true at i iff the reference of VP at i(the reference of NP at i ). The statement (iii’) of course uses functional application, but it seems a far fetch to say that the syntax “provides instructions” to use it. To grant the legitimacy of (iii’) is effectively to concede Davidson’s point, as I will argue more fully below. I agree with King that appeal to (iii), or to the corresponding (iii’) of IL, do not of themselves involve any sort of regress or incoherence. True, (iii) uses the relational predicate ‘instantiates’, which does not itself appear in the sentence S; and (iii’) applies function to argument without having displayed any expression in S that expresses that notion. In each case, there is something that, according to the theory in question, we must grasp in order to understand it that is not represented by any vocabulary in the sentence; and in sentences where that is explicitly represented, as in ‘Theaetetus instantiates the property of sitting’, we still must grasp it in the background, as yielding truth just in case Theaetetus and the property of sitting, in that order, instantiate the relation of instantiation. But that isn’t yet a regress: that’s just the theory. But now, for King’s account, or for the toy version above in terms of classes, there is another issue. How do expressions such as ‘the class of sitters’, ‘the property of sitting’, ‘the property of loving’, or ‘the class of all lovers and beloveds, taken in that order’, get their reference? These expressions are the nominalizations of predicates: the class of sitters is {x: x sits}, and the class of all lovers and beloveds, taken in that order, is {: x loves y}. Likewise, the property of sitting is the property of being an x such that x sits, and the relation of loving is the relation of x to y in which x loves y. Moreover, the account of predication illustrated by (1) and (2), or King’s examples, must apply to all 1-place and 2-place predicates in the language, with each of which must be associated a canonical singular term referring to a class, property, or relation. But then, in characterizing the things to which a predicate corresponds we will necessarily be using, inside the abstracts, the very predicates that we were supposed to be explaining. In Truth and Predication ( pp. 145–146), Davidson cites Frank Ramsey’s partial dismissal of the question of predication in Ramsey’s statement that ‘Socrates is wise’ and ‘Wisdom is characteristic of Socrates’ invert subject and predicate but “say the same thing.” For Davidson, this thesis is an instance of falling into his, Davidson’s, regress, through the use in this case of the predicate ‘is characteristic of ’ (an inverted epsilon, as it were),

178

James Higginbotham

expressing the relation converse to instantiation. But it should also be noted that there is no systematic way of forming, for an arbitrary Adjective Phrase A, a subject comparable to ‘wisdom’, except by recognizing nominalizations such as ‘to be/being A’, which are manifestly predicative in nature. The first objection to Ramsey, then, should be that in transforming subject and predicate we must recognize the predicative character of the new subject ‘wisdom’, which refers to the property of being an x such that: x is wise. Moreover, just to get off the ground, Ramsey’s consideration must appeal to some form of comprehension principle. What makes the class of things x such that x is wise the class that it is, or the property of being a thing x such that x is wise the property that it is? These expressions, ‘wisdom’, ‘the class of wise things’, ‘the property of being wise’, must themselves have a compositional semantics that justifies the assertion that ‘Wisdom is characteristic of Socrates’ does in fact have the same truth value as ‘Socrates is wise’. I have been skeptical of the thesis that there is a regress in a semantics that uses (i)-(iii) in giving the truth conditions of ‘Theaetetus sits’. Whether there is or not, however, we can advance the view that the whole route was redundant, brought nothing at all to the semantic project, and, if applied even mildly to membership or instantiation itself, either paradoxical or mysterious. It was redundant, because to understand (1), taken up now as (5), we must already understand ‘x sits’, and we could just write (6): (5) ‘Theaetetus sits’ is true l Theaetetus  {x: x sits} (6) ‘Theaetetus sits’ is true l Theaetetus sits It is no advance, because predication inside the class abstract is not eliminable. And if the schema is applied to ‘’ itself then, because the truth of such as (6) has to be acknowledged in any case, and the analogues of (5) and (6) together will imply all instances of the comprehension axiom, and Russell’s paradox will be swiftly derivable (alternatively, one might take the right hand side of (6) as an abbreviation for the right hand side of (5), thus leaving it utterly mysterious what classes are referred to by the abstracts, and how come). Russell’s paradox looms anyway, to be sure, as the naive comprehension principle (7) (cited by Rudolf Carnap in Carnap (1956: xyz)) cannot be accepted: (7) x has the property of being an F l F(x)

Expression, Truth, Predication, and Context: Two Perspectives

179

(I supply some details in Higginbotham (1989)). But that it should arise at this level is problematic. (IL, incorporating as it does the simple theory of types, evades the paradox, by syntactic fiat. In the respect we have been discussing, the difficulty with King’s and similar accounts of predication and properties is not shared by IL, for in that setting predicate and argument fit together naturally, as we might say: the clause (iii’), repeated here, employs functional application, but invokes nothing expressing that operation: (iii’) A sentence S = NP+VP is true at i iff the reference of VP at i(the reference of NP at i ). Moreover, (III) above, repeated here, falls out trivially. (III) The proposition that p is true if and only if p. It is possible that Davidson’s neglect of IL led him to overlook a simple response to his chief argument against deflationism about truth in Davidson (1996), that it was simply incoherent to advance (III) as saying, of something denoted by the subject ‘the proposition that p’ that it was true if and only if p. That argument I understand as follows:‘that p’ (or ‘the proposition that p’) must be a singular term, and what replaces ‘p’ must be the same sentence, with the same semantic features in both occurrences; but how is this to be done? The answer, chez IL, is that both sides of the biconditional contain a tacit reference to the actual world @; ‘p’ in both occurrences expresses a proposition (a function from possible worlds to truth values); and ‘that’ is a nominalizer, referring (for each possible world i) to the proposition that ‘p’ expresses. IL provides an explicit account of propositional nominalizations ‘that p’, producing their interpretations in terms of the interpretations of their parts, ‘that’ and ‘p’; any semantical account that did the same could be used to respond to the difficulty that Davidson raises. There is a worry about the justification of models for IL in the usual set-theoretic terms, inasmuch as within set theory functions are just other objects. And there is a further related worry about stating the theory itself: if the reference of ‘sits’ (in i) has to be given as something like ‘the function that maps x into Truth if x sits (in i), and Falsehood otherwise’, it appears that this reference is to an object. If t is a singular term, then we can put the words ‘the reference of t‘ salva grammaticate for t anywhere. Should not the same be true for any predicate P and whatever corresponds to its

180

James Higginbotham

reference? Here we are up against questions very close if not identical to those that Frege faced in “On Concept and Object” (Frege (1892)). I will consider some of these more closely below. Davidson cites Plato as having observed that you cannot make a sentence out of two names, or two predicates. But there is a way within a fragment of English taken up within IL to make ‘Theaetetus Theaetetus’ a sentence: just let the latter occurrence of the name express ^Ox(x = Theaetetus). This is “type-lifting,” much beloved of Montague grammarians. The fact that ‘Theaetetus Theaetetus’ isn’t a sentence, and couldn’t be, would be an adventitious lexical or morphosyntactic matter. However, I think that we should reject, not Plato’s observation, but type lifting, and likely the theory of types altogether, at least in application to the semantics of human first languages. There is an issue in the other direction as well; that is, in the free admission of NP-VP sentences, inasmuch as the free combination of many predicates with expressions of different types will, according to the theory, force those predicates to be arbitrarily ambiguous (for many details, see Chierchia (1984)). We have been considering Davidson’s first thesis, that the reference of predicates cannot be to objects. It is possible to invert this thesis; that is, to hold that no (apparent) reference to objects can give the reference of a predicate. I take this to be the view of Cartwright (2005). There, considering what he perceives to be the inadequacy of expressions like ‘the property of being so-and-so’ to reveal what is going on in predication, he mentions Frege’s view that the inadequacy merely reveals what Frege called an “awkwardness of language.” He then writes: I have nothing better to say. Cartwright (2005:918).

I assume that this confession is not merely an autobiographical statement, but rather an expression of despair at not finding Frege’s gesture satisfactory. The definite article, said Frege, “points to an object.” That statement is not quite true for English anyway, as we have statements like (8): (8) John is the very thing we expected him to become: a philosopher (or: honest to a fault, etc.) where what follows the copula is pretty clearly a second-order definite description (more than a hint of the “awkwardness” is provided by the observation that we had to use the dummy noun ‘thing’ in the description).

Expression, Truth, Predication, and Context: Two Perspectives

181

But such examples are rather puny, and they yield a far less than robust interpretation of quantification into predicate positions. Others have been highly skeptical of property talk, or anything similar, as explaining predication. Thus Schiffer (2003) (discussed in King (2007: 103 ff.) argues that accounts of structured propositions along King’s or similar lines must make what is expressed by ‘Fido barks’ a proposition made up of two singular terms, so that the predicate position in the nominalization ‘that Fido barks’ should be a quantifiable place, and sentences such as ‘John believes that Fido barks’ should allow the implication ‘For some x, John believes that Fido x’, which is, in Schiffer’s words, “incoherent” (Schiffer (2003: 30)). Here Schiffer is proposing, or perhaps assuming, that in virtue of the use of such words as ‘the property of barking’ as giving the reference of ‘barks’ we have, in Frege’s terms, “pointed to an object.” But this view need not be accepted (King disputes it, at least in the sense that he denies that his view is committed to holding that predicates are ever what he calls “referring expressions”). In any case, I review quickly, and without presenting the textual evidence from Frege’s writings here, one way of defending Frege’s views.2 In a language such as Begriffschrift, with function symbols as primitive, and similarly in second-order logic, there is no difficulty in writing definite descriptions for functions, or definite descriptions that would give the reference of predicates. The function “add one” is just (9), and the reference of the predicate ‘sits’ is given as in (10) (in an extensional setting): (9) (the f )(x) f(x) = x+1 (10) (the F )(x) (F(x)lx sits) Moreover, (9) is a function expression, just like ‘f ’ itself, and (10), like ‘sits’, is a predicate. Hence we may write (11) and (12): (11) [(the f )(x) f(x) = x+1](2) = 3 (12) Theaetetus (the F )(x) (F(x)lx sits) In High School Mathematese we pronounce (9) as something like, “the function f such that for all x f-of-x is x+1.” But then we had to use a Noun, namely ‘function’. And how do we say (10)? We can try something like, “What (or: that which) a thing is if and only if it sits,” roughly as 2 The ensuing discussion of Frege is in part what is called in Los Angeles a “remake” of Higginbotham (1990).

182

James Higginbotham

suggested in Dummett (1973: 213–218). We should then declare that ‘Theaetetus is what a thing is if and only if it sits’ shall be a sentence meaning that Theaetetus sits. By no means, however, can we always use the definite or indefinite articles of English—they would give the wrong idea, as producing singular terms, or indefinite descriptions of objects. If, moreover, as is overwhelmingly the case in natural languages, the arguments of predicates are in Frege’s sense saturated, then neither (9) nor (10) can itself appear as a subject of predication. Now, nothing that I can see prevents endorsing Frege’s view, absorbing it into possible-worlds semantics, and thereby recruiting it into one’s conception of structured propositions. For ‘Theaetetus sits’ we would have a tree or phrase marker as in (13), where D is the person Theaetetus, and E is the intension of ‘sits’: (13) [[D] [E]] The parent node, marked by the exterior square brackets, is that sentential intension that is Truth in a world i just in case [E(i )](D(i )), and Falsehood otherwise. The point of the structured-propositions account would thereby at least in part be realized, in that it would prevent the collapse of the proposition expressed by a sentence into its intension. Should scruples about what is and what is not sayable given the design of natural languages be a barrier to adopting Frege’s thesis, that what we face is just an “awkwardness of [natural] language,” or a precaution against launching ourselves as it were into the unsaturated sea? That is a question beyond my scope here. In any case, it is not to be decided in virtue of the properties of language that figure in those parts of semantics we have been considering. We have been examining Davidson’s first thesis, that the reference of predicates cannot be to objects. The second thesis I mentioned above is that if we adopt the view that the interpretation of a predicate is given by saying what it is for that expression to be true or false of objects, then nothing further that is peculiar to predication need be stated. The question of the reference of predicates then appears to become otiose. Thus in his useful review of Truth and Predication, Jeff Speaks (2006) observes that we might read the book back to front, as it were, taking the solution that Davidson offers to constitute an objection to assigning universals, or concepts in the sense of Frege (i.e., functions from objects (with or without possible worlds) to truth values), or properties, or anything of the sort to predicate expressions. In a similar vein, Orenstein (2006) remarks that if

Expression, Truth, Predication, and Context: Two Perspectives

183

we can use the notion of truth to characterize predication (in the sense of saying that predicates are those expressions that are satisfied or not (in a possible world) by objects), then the question what predicates stand for, refer to, or mean falls away. This conclusion is too swift for philosophy, however, as it presupposes that the expressive limitations of natural languages must carry over into the philosophical exposition of science and mathematics. The questions here are beyond the scope of this paper; anyway, they aren’t to be decided offhand within the semantics of simple sentences of natural language. I conclude that Davidson’s second thesis, although it may be correct, depends upon matters outside the immediate task of theories of truth and propositional expression. Davidson’s third and strongest thesis, remarked above, is that even if predicates have a reference, their reference cannot explain predication. This thesis, I believe, does not survive. If higher-order logic is admitted, then predicate reference, and with it the clauses of a theory of truth that characterize predicates in terms of their reference, come naturally. For, say, ‘sits’ we shall have (14): (14) ‘sits’ refers to an F such that (x) (F(x)lx sits) (depending upon other matters, there may or may not be a unique such F ). Or, if we aren’t too squeamish, just ‘the Verb ‘sits’ refers to sits’, because ‘refers to’, as we are now using it, has already crossed the barrier, having been recruited as a predicate of arguments of the second level; compound predicates will have their own recursive clauses; and so on. If I understand him correctly, Burge (2005: 18 ff.) argues, contrary to Davidson, that predicate reference can profitably be grounded in the notion of functional reference, and he therefore finds what I have called Davidson’s strongest thesis simply mistaken.3 The grounding of which he speaks, however, must amount to more than just an appeal to talk of functions as entrenched in mathematics, unless indeed it could be shown that, say, the functions spoken of in real analysis are not objects. The proposed reduction of predication to functional application is indeed, as one might

3 Burge appeals here to Church’s calculi of O-conversion. Burge’s own view of the relation between ‘Ox F(x)’ and ‘F(x)’ may differ from that of Church (Burge (2005: 21, fn. 13)). IL itself is a O-calculus; but it is typed, thereby granting a sense to the metaphysical distinction between concept and object.

184

James Higginbotham

say, a reduction to the previous problem. But in this case the previous problem is the same problem.4 To some extent, functional reference can be grounded in the notion of the reference or denotation of a function symbol, something undertaken on behalf of predicate reference in Furth (1968). For a function symbol to denote (relative, perhaps, to a restricted universe of discourse) would be for it to figure syntactically in an expression referring to a single object whenever a singular term that refers to a single object is put in its open place: thus ‘_2’, over the natural numbers, denotes, since if t refers uniquely to some number n, then ‘t2’ will do so likewise; more generally, in settings where not all objects have or could have singular terms referring uniquely to them, the function symbol ij, when completed by an object assigned to its empty place, will, on that assignment, denote an object. The attempt would be to redeem talk of functional denotation in grammatical, rather than metaphysical, terms, relieving it of the metaphorical baggage of Fregean “unsaturatedness” and the like. However, the construction is evidently too weak to allow quantification over all functions, say from natural numbers to natural numbers, something that is wanted classically. As Parsons (1971) notes, it would not allow impredicative definition, which nevertheless may be wanted, and Frege himself did require, for Cantor’s conception of cardinal number.5 In IL, or full second-order logic, which accepts the two-sorted universe of objects and functions, the predicative barrier is breached; but then we have gone metaphysical again. The same point applies within the refinement of IL that brings in structured propositions. I have tacitly taken for granted (as King and many others do) that IL, despite its elegance, is an unsatisfactory theory of propositional identity, insofar as propositions are objects of speech, thought, desire, and knowledge. The approach in terms of structured propositions, specifically as in King, but extending to other proposals, is an attempt to have a share of the elegance, without suffering the collapse of necessary equivalents. That

4 Davidson, both early and late, thought of expressions like ‘the father of ___’ as function expressions (which is of course wrong: they are definite descriptions with relational head Nouns), and concerning their interpretation contented himself with the remark that they can be given a semantics without saying that the words ‘the father of ’ refer to anything. Presumably he would have said the same for expressions like ‘__ squared’. So far as I know, he nowhere considered quantification over functions. 5 See Parsons (1983: 68, fn. 11).

Expression, Truth, Predication, and Context: Two Perspectives

185

account, however, presupposes an account of predicate reference, for which there is scant evidence, at least within the most basic parts of our language. Davidson’s discussion is valuable here, even if his repudiation of predicate reference amounts in the end to a rejection, not a refutation. With the above to hand I turn now to the proposal that a substantive theory of truth might be combined with a redundancy theory of the expression of a proposition by (an utterance of ) a sentence, by allowing the sententialist proposal, that what is expressed is simply the sentence itself, not as quoted, but understood as if uttered. On sententialism, the objects of thought reported through closed English sentences without indexicals are just those English sentences, understood as if uttered. Where we understand free variables or indexical reference, we allow syntactic structures where things other than expressions may be found at points where terms of the appropriate sort could go. Quantification into objects of thought, however, is by no means linguistically restricted. There are no doubt many objects of our thought that we do not express in English, and perhaps could not express, as they may involve non-linguistic representations of objects, for example. Again, suppose I am in Bucharest, and I see a man say something to another man, only to be slapped in return. I conclude, correctly we may suppose, that the first man said something insulting to the second man. I know nothing much of Romanian, and so do not know what was said, or whether it has an English equivalent; but nothing in sententialism requires that I do. On the view under consideration, the relation of the sentences we use to the objects of our thought is something like the relation of the numerals to the real numbers. We can refer to numbers in all sorts of ways; but we really get hold of them through numerals, and expressions constructed with the aid of numerals (to the extent that these expressions are surveyable; and on the linguistic side to the extent that the sentences in question can be taken in). There intimate relation between the numbers and the numerals, fractions, exponents, and other features of our notation through which we canonically refer to them. But that relation need not be viewed as limiting our capacity to talk of real numbers in general. Similarly, the intimate relation between the sentence and the thought doesn’t rule out thoughts for which we don’t have the sentences. Elsewhere (Higginbotham (2006)) I have tried to show how at least certain kinds of context-dependency, such as the use of demonstratives, or gestures toward confinement of quantificational domains, are consistent with sententialism. There are other types than these, however, and the remainder of my discussion will explore some issues in that domain. The

186

James Higginbotham

specific view I will examine will take a strong stand on some matters under current discussion, and (in the case of the contextual feature of Tense) about which considerably more has now been examined cross-linguistically than was the case some time back. In many expositions, the fundamental object of study is the interpretation of a sentence in a context. The notion context here is to be understood in a technical sense, as comprising a bundle of features that interact in specific ways with formatives of the language; thus in the development in Kaplan (1977) and elsewhere, including the speaker, the time and place, the possible world, and so on, cashing out demonstratives and other indexical expressions, and redeeming talk of possibility and necessity. An important element of Kaplan’s account is the thesis that demonstrative and indexical reference are settled by the context, independently of depth of embedding. The general theory permits expressions to behave otherwise, and it is therefore an empirical discovery, about English, or about some or all human first languages in general if indeed that thesis holds. But it may appear not to hold. For instance, Schlenker (2003), relying upon the descriptive literature, reports that in Amharic the word that in root clauses amounts to the first-person pronoun can take on reference to the higher subject in a subordinate clause. Representing this word by Ï, we would have (15) as spoken by Mary, meaning that she herself is happy, but (16), as spoken by Mary, would assert that John said that he, John, was happy: (15) Ï am happy. (16) John said that Ï am happy. There has been considerable further discussion of the above and similar phenomena.6 The contrast with English is stark: obviously, Mary’s English statement ‘John said that I am happy’ only means that John said of Mary that she was happy. It appears, then, that we must allow that the context “shifts” as between superordinate and subordinate clause in the Amharic (16), as only then can we have reference to “the speaker,” of the subordinate clause, John, rather than the speaker of the superordinate clause, Mary.

6 See especially the survey and conjectures in Anand (2006). Of course, it is much less than theoretically satisfactory to declare that Amharic ‘Ï’ is just a homonymic form, as that would lose the generalization that it looks back to the (main or attributed) speaker. (Thanks to Barry Schein for remarks on this point.)

Expression, Truth, Predication, and Context: Two Perspectives

187

But there is no need to accept this point of view. It is, I shall argue, an artifact of the decision to treat sentence in a context as the object of inquiry, rather than potential utterance. I elaborate briefly, drawing upon views I have expressed elsewhere (Higginbotham (2002a)). First of all, we extend the theory of truth for natural languages so as to allow conditional truth conditions, whose antecedents involve the referential intentions of speakers, and whose consequents are like the familiar biconditionals of theories of truth, with the peculiarity that they contain variables that, although bound in the larger conditional, are free in the consequent. A canonical example is an utterance of (17): (17) This is red. The semantics of English, we suppose, stipulates some satisfaction condition for the predicate ‘red’. For the subject, the semantics gives a rule of use, approximately as follows: (18) ‘this’ as a full NP is to be used to refer to a single proximate, salient object. For the sentence (17) as a whole we will have (19): (19) (s)(u)(x) If the speaker s uses ‘this’ in accordance with its rule of use (16), and thus refers to a single object x, then s‘s utterance u of (15) is true l x is red. In any particular case in which the antecedent of (19) comes off, we shall have (20): (20) s‘s utterance u of (15) is true l x is red. for some definite trio of values for s, u, and x. Now, the account of (17) applies also when that sentence is embedded, as in (21): (21) Mary said that this is red. That is to say, in reporting what Mary said it is the speaker who uses the word ‘this’ to refer to a thing: however Mary may have referred to it is not revealed. What is to be understood as if uttered is just (17).

188

James Higginbotham

We have seen that the complement clause in (17) is to be understood as if the speaker (not the person whose speech is being reported) had said it. Moreover, if the general account of (17) and (21) fully generalizes, it cannot be otherwise; that is to say, you could not introduce a “shifted” demonstrative into the language, because the speaker’s referential intentions always govern her speech as a whole. On this view, Kaplan’s thesis, which was originally applied to English, would be universal. But now what of Schlenker’s and similar examples? To answer this point, I shall first advert to Tense; but the moral will be general. Tense is an indexical feature of English and other languages. According to the simple sententialist interpretation of complement clauses, Tense should be understood in a complement just as it is understood in a root clause. But now, are there not examples where the subject’s perspective on time, not the speaker’s, is at stake? Indeed there are, and a simple case is provided by the rules governing Sequence of Tense in English. Consider (22): (22) Mary said that she was once happy. Let u be the speaker’s true utterance of (22), and let its actual time be W(u). Let e be an utterance by Mary that makes (22) true, as reported by the speaker. The use of ‘once’ in the complement makes the only reasonable interpretation of that clause one in which what Mary said at some time t prior to W(e) (not W(u)) was to the effect that she was happy at t. Therefore, the speaker’s utterance of ‘She (Mary) was once happy’ is not interpreted as if uttered by the speaker of u; for any such utterance requires for its truth only that Mary was happy at some point prior to W(u). Rather, it is interpreted as if uttered by the speaker of u at W(e), the time of Mary’s speech. The phenomenon I have just described extends, with further consequences, downward clause by clause; I won’t give further examples here. The question is what the phenomenon may show about the interpretation of complements and other embeddings on the sententialist hypothesis. The answer is that it shows very much indeed if the formula “as if uttered” must be adjusted with respect to time or other features. But it is easily taken on board, consistently with the view that indexical and demonstrative reference is fixed once for all in the antecedent of conditional truth conditions, if it consists in anaphoric relations between lower and higher clauses. For an obvious example of anaphora in English, consider (23):

Expression, Truth, Predication, and Context: Two Perspectives

189

(23) Mary thought that pictures of herself would please Susan. Examples such as (23) are generally ambiguous, in that the antecedent of ‘herself ’ can be either the higher subject, or the lower object of the psychological Verb, either ‘Mary’ or ‘Susan’ in this case. But the lower clause, ‘pictures of herself would please Susan’ is of course unambiguous. By hypothesis, the complement clause in (23) refers to the syntactic structure 6 for ‘pictures of __ would please Susan’, with Mary herself (the person, not the expression) in the position marked by the underscore. On this formulation, there is no shift of context as between the higher and lower clauses. Indeed, insofar as the complement expresses just a de re thought about Mary, it is as if the speaker had said (24): (22) Mary thought that picture of HER [indicating Mary] would please Susan. We may propose, therefore, that it is anaphoric relations between the Tenses in (22) that are responsible for the shift from W(u) to W(e) in determining the reference of ‘Mary was once happy’ in that sentence. No shifting of context is involved.7 But now Schlenker’s examples, and others, may be understood along the same lines. My ersatz Amharic Ï is grounded by default in the speaker, just as the Tenses are grounded by default in the actual time of speech. But it is also anaphoric, like the Tense, and for that matter the third-person pronouns of English; so it is anaphoric when embedded, though indexical in root clauses. Some examples from Korean may illustrate the point further.8 Korean has a word, usually transliterated as ‘cikum’, which in a root clause is properly translated by English ‘now’. So (the Korean equivalent of ) ‘I cikum stand+Present’ means that I am standing now.9 Korean (unlike English) allows cross-reference in Tense between an embedded Present and a superordinate Past, so that the Korean sentence whose structure and vocabulary are indicated as in (25) does not mean that John said that Mary is in

7 For elaboration, see Higginbotham (2002b) and references cited there. 8 I am indebted here to the judgements, scholarship, and written work of Ms Hyuna Byun Kim; see Kim (2008). 9 Korean is Verb-final, so I give examples in ersatz Korean.

190

James Higginbotham

Seattle at a time that continues up to the present, but only that John said that Mary was in Seattle at the time he, John, spoke: (25) John [that Mary in Seattle be+Present] say+Past But now, if we retain the past-time reference in the higher clause, and add ‘cikum’ to the lower clause, as in the make-believe Korean (26), the meaning is just that of English (27), not English (28): (26) John last week [that Mary in Seattle cikum be+Present] say+Past (27) John said last week [that Mary was then in Seattle] (28) John said last week that Mary is now in Seattle. Korean ‘cikum’, unlike English ‘now’, is therefore sensitive to embedding. (Similar remarks, I have learned, hold for the Korean ‘ecey’, which means yesterday in root clauses, but may mean the day before when embedded in complement clauses.) But here again we may have just an anaphoric phenomenon, or one that is sensitive to the results of anaphora. I assume as in other work (e.g., Higginbotham (2002b), and as proposed in various places cited there, that Tenses are not sentential operators of any sort, but rather express temporal relations between events or situations as arguments These arguments can further be bound by adverbial quantifiers, or by existential closure in the syntax. Thus the English present in an utterance of ‘Mary is in Seattle’ expresses the temporal relation ~ of (the actual time of ) one situation e surrounding (the actual time of ) another, which latter will in a root clause be the time W(u) of that utterance itself; similarly, the Past Tense of ‘Mary was in Seattle’ expresses the relation W(e) iff: on the understanding that it is enough to be painted green, x grunts.

The anomalous understanding is one that would be needed in an account of “green”, so it will be a member of the domain of quantification over understandings. But this instance is either nonsense, or it delivers the wrong result. (For the second alternative: an understanding of what it is to be a grunter on which it is enough to be painted green would ensure that suitably painted benches are grunters.) Hence understandings must be restricted to ones appropriate to “grunts”. Using the theory now presupposes that the user knows in advance which these understandings are, for the theory does not say. But to know which understandings of “grunts” are appropriate entails knowing what “grunts” means. The knowledge the theory was supposed to state has not been stated but has been presupposed. I conclude that a Davidsonian cannot respond to the kind of global argument we have considered by quantifying over contexts or understandings. How, then, should Travis’s point be met? For examples like “grunt” I think the main thing is to distinguish the common phenomenon of unspecific meaning from semantic context-dependence. I’ll illustrate with an example that will be uncontroversial for many Contextualists (though I fear may not be so for Radical Contextualists of Travis’s kind). There are many ways to run, east or west, to work or to the gym, in the morning or in the evening. An utterance merely of “John runs” does not provide any of these details, though if the utterance is true, it will be made true by an event which resolves every such issue. Context may make some more specific way of running salient, but in doing so, the semantics are not touched. The test is that one can coherently deny that John runs in a

A Very Large Fly in the Ointment

245

salient way, without this being either a retraction or a contradiction. Hence the salient way of running is not part of the semantics. For example, it would be natural to interpret the utterance of “John runs”, as it occurs in the following context Jill walks to work. John runs.

as committing the utterer to the claim that John runs to work. The question is whether this commitment (supposing it to be genuine) emerges from the semantics of “John runs”. A negative answer is suggested by the following possible variant: Jill walks to work. John runs. Indeed, he runs 20 miles a week. But never to work, on account of the traffic.

The coherence, and absence of retraction, suggests that in this utterance the semantics of “John runs” does not assign it the content “John runs to work”. This suggests that the same is true of the shorter utterance, for the longer one has the shorter one as a proper part. By the time the interpreter had reached the second full stop in the longer utterance, he should presumably have reached just the state he would have reached when interpreting the shorter one, and so, on the rival view, would have believed that John had been said to run to work. Such an interpreter would have to regard the remainder of the longer utterance as either containing a contradiction or a retraction of the earlier part. Intuitively, however, that is not the case. The same point can be reached by a slightly different route. Consider “Bob walks to work. Jill doesn’t run. But she runs a quarter marathon every Sunday.”

On a Contextualist view, it should be easy to hear this as consistent, for the second sentence will be equivalent to “Jill doesn’t run to work”. In fact it is hard to hear the whole as consistent, suggesting that the second sentence tells us that Jill doesn’t run anywhere (or in any way). Unspecific meaning is the category to which the Davidsonian should assign Travis’s grunter. True, there are many ways of grunting, as there are many ways of doing anything. If an attribution of grunting is true, it is made so by some specific form of grunting. None of this entails that the semantic content of “grunts” varies from context to context (nor that the pragmatic content varies). A test is this: if we can add something equivalent

246

R. M. Sainsbury

to “in some way or other” without making a significant difference, the verb is semantically neutral concerning the way it is to be satisfied. The default reading of “Sid grunts” is that he grunts in some way or other: the truth conditions are unspecific relative to various modes of grunting. The default reading of “Sid doesn’t grunt” is that he doesn’t grunt in any way. The same goes for color terms: “red” applies to the things that are red in any one of possibly indefinitely many ways (on the inside, on the outside, naturally, through being painted, etc.). If one of these ways is highly salient, we may criticize a speaker for applying “red” to something not red in the salient way; we may voice this criticism by saying that what the speaker said is not true. But we normally do not care about the distinction between semantic content and what a speaker meant, and so we would not discriminate between these different targets of our criticism. Suppose external redness is salient, and that someone says, of something which is red inside but not outside, that it is red. It would be natural to respond like this: You’re wrong: it’s not red in the relevant way. It may be red inside, but it’s not red outside.

The whole exchange does not require any more specific semantics for “red” than that it is satisfied by something which is red in some way or another. This permits a sensible story about “red inside”, which is hard to tell if the salience of external redness made the contained occurrence of “red” apply only to things externally red. These cases contrast sharply, I believe, with other Contextualist examples. “Jill is ready” is not equivalent to the near-trivial “Jill is ready for something or other”, and “This girder is strong enough” is not equivalent to the trivial “This girder is strong enough for something or other”. We should not let the fact that there are specific ways to grunt undermine our confidence in the full correctness (barring considerations related to tense) of the claim that “Sid grunts” is true iff Sid grunts. By contrast, “ready” and “enough” may well demand a treatment which reveals their content as context-sensitive. We cannot offer a global response to the global argument offered by Radical Contextualists. We have to look at their examples case by case. One avenue of response is to say, as with “grunts”, that they confuse many ways in which a sentence can be made true with many distinct contents, when really we have a single rather unspecific content. This will not do for

A Very Large Fly in the Ointment

247

all cases. In what follows, I consider further weapons which a Davidsonian should stock in his armory.

4. Cases in which the content-sensitivity is not associated with a specific expression Two examples: Little Johnny cuts his hand and his mother says: (a) It’s OK. You won’t die. According to a Contextualist, the mother does not say, falsely, that Johnny is immortal, but that he won’t die from that cut. So the semantic content of the utterance is not that Johnny is immortal. Davidsonian truth conditional semantics cannot associate this utterance with a content that refers to a cut, since (among other reasons) there is no expression which would justify invoking a semantic axiom introducing the notion of being a cut. “Bridging inferences.” ( b) He took a book from the shelf and sat down to read. In many contexts we will take it that the speaker is claiming that the protagonist read the book he took down. So in these cases, this is the semantic content of the utterance. This content cannot be delivered by a truth theoretic semantic theory, since there is no expression in the utterance which could merit invoking an axiom involving a relation between a book taken down and a book read. I have emphasized ( by italicizing the occurrences of “so”) the Contextualist moves that a Davidsonian should say are non sequiturs. A speaker may say, communicate or claim less (as in case (a)) or more (as in case ( b)) than the semantic content of the utterance she uses, and these reduced or expanded contents are plainly determined by context in all sorts of complex ways, with no upper bound on what collateral knowledge may need to be brought to bear to extract it. The Davidsonian should say that the semantic content of (a) is not reduced, but is that Johnny won’t die (ever); and that the semantic content of ( b) is not expanded, but is simply that he took a book from the shelf and sat down to read. The reduction or expansion is something in some way obtained from semantic content along with other information; we can label it pragmatic content. A test for this being

248

R. M. Sainsbury

the right thing to say is as follows. If we suppose that (a) does not have as its semantic content that Johnny is immortal (or anything entailing this), we should be able to add its negation explicitly, without contradiction or retraction. That is, the following should be fine: You won’t die. You’re not immortal.14 But it is jarring: one’s first reaction is that one has misheard. By contrast, a standard form of qualification or retraction is entirely in order: You won’t die. I don’t mean you’re immortal, only that this cut won’t kill you.

Typically, you need to explain “what you mean” when you have not said what you mean, which seems precisely the right description of how the speaker is related to “You won’t die”: she said it but did not mean it. This is reinforced by supposing that the mother had said: You won’t die, though you will die sometime.

If in the original scenario the semantic content of “You won’t die” had been “You won’t die from this cut”, one would expect the utterance of these words in the revised scenario to have the same content (the revision affects only what words the mother said subsequent to these). In that case, the total utterance would be perfectly normal, but in fact it is odd, perhaps (and certainly to my ears) contradictory. A similar point applies to the additional content associated with (b). A Davidsonian can allow that this content will typically be communicated, but can deny that it belongs to semantic content. A sufficient condition for this verdict is that the speaker can go on to deny the additional content without oddity, contradiction or retraction, which would not be possible if it belonged to semantic content. So suppose the original utterance was followed by a subsequent one: He took a book from the shelf and sat down to read. He chose The Paintings of Michelangelo, because it was large enough to conceal what he was reading, a dog-eared copy of Penthouse.

If the semantic content of the utterance of the first sentence included that he read the book he took down, the utterance of the second sentence 14 Julie Hunter suggested a nice variant: If the child responds to its mother’s “You won’t die” by something like “Come off it Mom: you know we’re all going to die sometime”, we cannot doubt that the child has contradicted its mother.

A Very Large Fly in the Ointment

249

should be odd or contradictory or constitute a retraction; but none of these things is the case. Hence we can conclude that his reading the book he took down is not part of the semantic content of the first utterance. From this portion of the discussion, we collect a fairly obvious resource for the Davidsonian: not all context sensitivity need affect semantic content, so not all such sensitivity need be recorded in a Davidsonian semantics. Everyone should agree that this resource is available, though there will be room for disagreement about how exactly it should be applied. I take it to be unpromising in the case of the examples which follow.

5. Allegedly problematic context-sensitive expressions Four examples: Rain

(a) It’s raining. Imagine this utterance having been made in Austin on 06/06/06. The semantic content of the utterance (the contextualist may claim) is that it is raining in Austin on 06/06/06. Even supposing that the Davidsonian approach can handle the tense (a member of the Basic Set), it cannot account for the location, since there is no lexical element in the utterance which could be associated with a location-introducing axiom. Let’s accept without discussion that the semantic content of the utterance is as specified in this argument.15 It does not follow that a Davidsonian cannot weave it into a conditional truth condition, for example: if in an utterance u of “it’s raining” the speaker referred to place or range of places p, then u is true iff it’s raining somewhere there [within p].16 15 Location fails the envisaged test for not belonging to semantic content. “It’s raining. But not anywhere.” sounds odd, perhaps contradictory. 16 The use of “there” rather than just “at p” conforms to the demands of the scenecontent approach. One who uses such a theorem en route to a report of the speech of one who utters “It’s raining” may refer to the relevant place without so much as hinting that this corresponds to the way the utterer thought about it:

250

R. M. Sainsbury

I have remained neutral on whether location is part of semantic content, but have shown that if it is, this poses no special difficulty for a Davidsonian. Possessives Uses of possessives (as in “John’s car”) invoke different relations in different contexts. The Contextualist will say that it is up to a semantic theorist to deliver a semantic content which involves the contextually appropriate relation, but that a Davidsonian cannot do this. Why not? One answer might be that the range of relevant relations is huge, and the relations need not be, indeed cannot be, known on a one-by-one basis to speakers and hearers merely in virtue of mastery of possessive locutions. Yet, in context, specific relations enter semantic content. For example, ( b) John’s car may refer to the car John owns, the one he has borrowed, the one he has painted, the one he covets, and so on. All being well, context settles the relevant relation, which in turn, a contextualist will say, helps fix the semantic content of an utterance containing (b). A Davidsonian need not deny that there is no knowing the full range of possible relations in advance. But he will not see this as an obstacle: perhaps a semantic clause could simply quantify over the relations, and so provide a conditional reference condition. for all possession relations R, all objects z, all referring expressions X, all predicates Y: if in an utterance u of “X’s Y” the speaker refers to relation R, then u thereby refers to z iff z = the satisfier of Y which stands in relation R to the referent of X.

Speaking (in Utah) of some remote corner of Alabama where my sister lives, Bill said it was raining there. Likewise, the content of the claim in the text ( given that the speaker may not know he is in Austin or that it is 06/06/06) is better reported: speaking in Austin on 06/06/06, he said that it was raining there then. Davidson himself, without comment, inserts location into the truth condition: “‘Es regnet’ is true-in-German when spoken by x at time t if and only if it is raining near x at t” (1973: 135). This will not always give the right result, e.g. not for the example displayed in this note.

A Very Large Fly in the Ointment

251

Setting aside detailed inadequacies (for example, “John’s car” can with entire propriety be used to refer to a car John owns even if he owns more than one), we encounter again the difficulties which led to the scene/ content approach: we would not wish an interpreter’s way of referring to z to replace the speaker’s way. Given the background information about what relation a speaker referred to, we can determine to which object the possessive phrase refers, but we risk leaving behind information about how that object is to be referred to. For example, if our instantiation of variable z is VIN number 1079856291, we will end up saying that an utterance of “John’s car is rusty” is true iff 1079856291 is rusty, and even if this is true, it is not interpretive. This problem is not much to do with possessives, for it springs from the fact that even for context free languages, we need to treat with care an atomic composition axiom like this: for all subject expressions, S, and predicate expressions, P, S+P is true iff the referent of S satisfies P.

Hesperus is the referent of “Phosphorus”. If this fact is available to the theory, one will be able to derive the uninterpretive T-theorem “Phosphorus is visible” is true iff Hesperus is visible.

In the Davidsonian tradition, the idea has been to ensure that such “extrasemantic” facts as that “Phosphorus” refers to Hesperus are not expressed by theorems of the theory, and hence cannot interact in this unsatisfactory way with the atomic composition axiom. This approach has to be abandoned once conditional truth conditions are supposed to lead to interpretive truth conditions via modus ponens, for here the input to the inference, the antecedent of the conditional, includes material that is not strictly part of the theory, and so cannot be constrained in that way to semantic propriety. This was the problem addressed by the scene-content approach, which in effect isolated, for indexicals, the way in which an interpreter referred from the way in which the speaker referred. But for complex referring expressions, we cannot accept that loss: something is clearly missing if we report an utterer of “John’s car is rusty” as having said, concerning vehicle 1079856291, that it is rusty. (Contrast with an utterance of “That is rusty”.) The concepts of being John and being a car need to get into the content reported. I think a Davidsonian would do better to take a different tack, and claim that possession is a case of very unspecific semantic content: the

252

R. M. Sainsbury

apostrophe indicates that some relation holds, but not which. If you are lucky, some relation rather than another will be salient, and will be the one the speaker hoped you would be moved by. Here the sign that the specific detail is not part of semantic content is that a speaker may have no specific relation in mind: one who uses “John’s car” will typically have a view about which car is in question, and will expect his hearer to, but may not have a developed view of the specific relation in which John stands to it, involving as this does nuanced distinctions between, for example, ownership and possession. Specificity is an optional extra, and not a semantically demanded norm. The approach would be illustrated by a kind of homophonic axiom: “X’s Y” refers to the referent of “X” ’s satisfier of “Y”.

In short, this case belongs with “grunts”. Tall Contextualists say that utterances of (c) Sally is tall can differ in truth value (even holding the reference of “Sally” unchanged) because different comparison classes may be salient in different contexts. In one context, the speaker may be saying truly that Sally is tall for a 6-year-old. In another, the speaker may be saying falsely that Sally is tall compared to her classmates. Hence the semantic contents must be different on the two occasions; and the Davidsonian has no way to deal with this. Setting aside the problematic character of the emphasized inference, it must be admitted that the semantics of adjectives like “tall” are genuinely puzzling, from any point of view known to me, especially because of their relation to comparatives and to nominal qualification (“tall F”), and I don’t think there is a widely accepted approach. The question is whether there is a special problem here for a Davidsonian. A starting point is to take as fundamental the role of adjectives of this kind, the “noun-hungry” ones, as noun qualifiers. The axiom for “tall” would be based on the idea that something satisfies “tall”+X iff it is a tall satisfier of X (Evans 1976). Having this as the basic axiom allows a Davidsonian to describe context sensitivity in terms of an elided qualified noun phrase. One of the two utterances of (c) just envisaged can be regarded as introducing (“at the level of logical form”) the noun phrase “6-year-old”,

A Very Large Fly in the Ointment

253

the other as introducing “member of class II”. This explains the difference in truth conditions. It is a contextual matter, in that context will determine which noun phrase has been elided; but once that is settled, the semantics carries the load. A bare occurrence of a noun-hungry adjective, like “tall”, must be seen as elliptical for an occurrence in which the adjective has an appropriate noun to qualify. If this is not provided by syntax, it must be provided by context. Although this illustrates a further resource for a Davidsonian, it would be naive to suppose that it unlocks the key to the semantics of these adjectives. Some will complain that no explanation has been given of what it is to be a tall satisfier of X. It is in keeping with Davidsonian minimalism to say that there is no explanation; alternatively explanations could be offered in terms, for example, of being taller than most satisfiers of X. However successful a Davidsonian may be in response to this objection, there are other difficulties. As C&L say, if Sally is a tall 6-year-old and Jack is a tall (adult) basketball player, then both Sally and Jack are tall. The supposed ellipsis must now take a more complex form: they are both tall, Sally for a 6-year-old, Jack for a basketball player. The original direct mode of composition, “tall”+X, needs to be replaced by a more complex form “tall for an X”. A Davidsonian, possibly in common with all of us, also needs to reject apparently natural principles like: if x is tall and y is taller then y is tall.

Counterexample: assume Harry is 5 ft 2 inches and that context supplies the italicized material: Sally is tall for a 6-year-old and Harry is taller, so Harry is a tall for an American adult. Another problem is that we may not have come to the end of the relevant context dependence. The envisaged elided material is itself sensitive to context. Maybe Sally is not tall for a Namibian 6-year-old, so perhaps just inserting “6-year-old” does not fully describe whatever context delivered. But what does? And is it plausible, as the noun phrases become more complex (6-year-old caucasian 21st century child raised in America . . .), that such linguistic material was present to the mind of the speaker or hearer? I don’t deny that these are problems, but either they have some kind of systematic solution, which comports with the appearance that utterances about tallness pose no special interpretive difficulties for actual speakers, or they do not. If they do, I am not aware of a reason for thinking that that the solution cannot be brought within a Davidsonian framework.

254

R. M. Sainsbury

Ready A Contextualist will say that the context of an utterance of (d) Jill’s ready will normally resolve what it is that Jill is being said to be ready for. If this is not resolved, there are no truth conditions. Its resolution does not relate to any syntactic element of the sentence, so, according to the Contextualist, there is no way it can be reflected in compositional truth theoretic semantics. There are a number of words which can occur both relationally (“phrasally”) and also non-relationally: married to, ready for, enough for, willing to, resting on, fed up with, moved by. The relation between the forms is different in different cases: “Jill is married” can be thought of as equivalent to “Jill is married to someone”, but “Jill is ready” seems not to be equivalent to “Jill is ready for something”, and “Jill is resting” is certainly not equivalent to “Jill is resting on something”. Any semantic theorist will sense a tension in treating these expressions: on the one hand, to preserve something of the common meaning shared by relational and nonrelational forms, on the other hand to do justice to the differences (and the different differences). One option for the Davidsonian has been adopted with bravura by C&L. They claim that there is nothing problematic about a theorem like: “Jill is ready” is true iff Jill is ready.

In effect, they treat “ready” as unspecific, the different things one can be ready for relating to “ready” rather as the different ways one can run relate to “run”. If they are right, then of course these examples pose no special problem for Davidsonian approaches. They offer two grounds for their opinion that there is such a proposition (i.e. entity possessed of truth conditions) as that Jill is ready. (1) If I hear someone utter “Jill is ready” without knowing what she is said to be ready for, I can still report the speaker as having said that Jill is ready, and this is not equivalent to reporting the speaker as having said that Jill is ready for something. (2) Given the truth of “Jill is ready” and “Jack is ready” I can infer that Jack and Jill are both ready, even if they are ready for different things, and I don’t know what either is ready for. The second point suggests that my conclusion cannot be treated as contextually enriched by an implicit single answer to the question what they are both ready for.

A Very Large Fly in the Ointment

255

Much as a Davidsonian would wish success to C&L, it is not clear that these considerations constitute any justification of their claim. As we saw, the capacity to re-use words without infelicity in a report does not guarantee that the reporter has identified truth conditions (e.g. the envisaged report that he told her he was leaving right away). As we saw with “tall”, two things can both be F without being F in the same way or respect, and without there being such a thing as being F simplicitur. C&L do not appear to address the Contextualist riposte that “Jack and Jill are both ready” involves a double contextual enrichment: “ready” is enriched one way for Jack and another for Jill. They are both ready, Jack for the climb and Jill for her exam. A Davidsonian might consider incorporating this kind of context dependence into the semantics, along the following lines: if in uttering “ready” the speaker refers to X, and this occurrence of “ready” is not part of a phrase “ready to” or “ready for”, then for all z, z satisfies “ready” iff z is ready to/for X. for all z, z satisfies “ready to/for” + F iff z is ready to satisfy F (or to be a satisfier of F, or for being a satisfier of F).17

This does not do justice to the use of “ready” with a plural subject (“Jack and Jill are both ready”), but although the whole question of plurals should be on the list of technical difficulties for Davidsonians, the use of “ready” in plural constructions justifies no special qualms. In the case of “ready”, a Davidsonian has two promising options: to say, with C&L, that it is unspecific, and to adopt a conditional truthcondition approach, as for expressions in the Basic Set. I have given reasons to prefer the second option.

6. Summing up Context is clearly a crucial concern for all those interested in language, no matter how its effects are partitioned between semantics and pragmatics. There are too many varieties of contextual effect for it to be likely that there is a single way of dealing with them, and the apparent panacea

17 I am assuming, probably unrealistically, that if x is ready to F, and something is F iff it is G, then x is ready to G. Non-extensionality is a major problem for the Davidsonian approach, but is not within the scope of this paper.

256

R. M. Sainsbury

considered in §3 turned out not to deliver what was needed. It has been a premise of this paper that the Davidsonian approach to the members of the Basic Set is adequate, using conditional truth conditions perhaps with the additional nuance of scene-content reporting. Some kinds of contextual dependence, it was suggested, are to be dealt with by extending the underlying idea to them (e.g. “It’s raining”). Other kinds need to be treated on the lines illustrated by “tall satisfier”, showing that not all words can receive an adequate semantic description except as they occur in a larger setting (they are “syncategorematic”). Some cases of alleged context dependence are shown to be, from the semantic point of view, best described as lack of semantic specificity (e.g. “runs”, possessives), with context often making salient more specific ways of making utterances containing such words true, ways not semantically determined. Finally, some context dependence seems to be independent of the words in the utterance (“You won’t die”, bridging inferences) and in these cases the relevant content is properly relegated to pragmatics, as involving non-language-specific processing. I have certainly not considered all the interesting kinds of contextual effect; but in suggesting that the ones I have considered do not threaten a Davidsonian, I hope that at the very least I will cause Contextualists to rely on different kinds of example, if such there be, ones which cannot be given a Davidsonian treatment in any of the ways considered here.

References Borg, E. (2004). Minimal Semantics. Oxford, Clarendon Press. Cappelen, H. and E. Lepore (2005). Insensitive Semantics: a defense of semantic minimalism and speech act pluralism. Oxford, Blackwell. Cohen, L. J. (1985). “A problem about ambiguity in truth-theoretical semantics.” Analysis 45: 129–34. Davidson, D. (1967). “Truth and meaning.” Synthese 7. Reprinted in his Inquiries into Truth and Interpretation Oxford, Clarendon Press, 1984: 17–36. Davidson, D. (1970). “The method of truth in metaphysics.” In his Inquiries into Truth and Interpretation Oxford, Clarendon Press, 1984: 199–214. Davidson, D. (1973). “Radical interpretation.” Dialectica 27. Reprinted in his Inquiries into Truth and Interpretation Oxford, Clarendon Press, 1984: 125–39. Evans, G. (1976). “Semantic structure and logical form.” Reprinted in his Collected Papers, Oxford, Clarendon Press, 1985. Frege, G. (1892). “On sense and meaning.” In his Collected Papers on Mathematics, Logic and Philosophy, B. McGuinness (ed), Oxford, Basil Blackwell, 1984: 157–77. Frege, G. (1923). “Compound thoughts.” In his Collected Papers on Mathematics, Logic and Philosophy, B. McGuinness (ed), Oxford, Basil Blackwell, 1984: 390–406.

A Very Large Fly in the Ointment

257

Frege, G. (1897). “Logic.” In his Posthumous Writings H. Hermes, F. Kambartel, F. Kaulbach (eds), P. Long and R. White (trs). Chicago, University of Chicago Press: 126–51. Frege, G. (1918). “Logical investigations: thoughts.” In his Collected Papers on Mathematics, Logic and Philosophy, B. McGuinness (ed), Oxford, Basil Blackwell, 1984: 351–72. Higginbotham, J. (1994). “Priorities in the philosophy of thought.” Supplementary Proceedings of the Aristotelian Society 68: 85–106. Künne, W. (1992). “Hybrid proper names.” Mind 101(4): 721–31. Rumfitt, I. (1993). “Content and context: the paratactic theory revisited and revised.” Mind 102: 429–54. Sainsbury, R. M. (1998). “Indexicals and reported speech.” Proceedings of the British Academy 95: 45–69. Reprinted in his Departing From Frege, London, Routledge, 2002: 137–158. Sainsbury, R.M. (2005). Reference Without Referents. Oxford, Oxford University Press. Travis, C. (2006). “Insensitive semantics.” Mind and Language 21(1): 39–49. Weinstein, S. (1974). “Truth and demonstratives.” Noûs 8(2): 179–84.

From Truth Conditions to Structured Propositions Richard Schantz Abstract: The author scrutinizes the classic truth-theoretic approach to semantics. The focus is on Davidson’s influential contention that a theory of truth can be employed in pursuit of a theory of meaning. It is argued that the main problem for truth-theoretic semantics is that there is no conception of truth conditions that is sufficiently strong to capture meaning. In line with several other philosophers, the author proposes an alternative approach to semantics, according to which structured propositions function as the meanings or contents of sentences.

I It is relatively uncontroversial that the words, phrases, and sentences of a language mean something—that they have meaning. Consider the word “horse”. It has a distinctive meaning in English in virtue of which it should be applied to horses and only to horses. Moreover, this word apparently has the same meaning as the German word “Pferd” and the French word “cheval”. Surely, “horse” means something different than the English word “cow” and should not be translated into the German “Kuh”. Generally, we suppose that speakers of a natural language understand its expressions, or know their meanings. It is the demanding task of semantic theories, or theories of meaning, to supply theoretical descriptions and explanations of all the diverse and complex phenomena of linguistic meaning. Such theories attempt to develop theoretical frameworks which enable us to explain in a systematic way what it is for words and sentences to possess the particular meanings they possess, and why they possess that meaning rather than some other. But what are these “meanings”? Although sentences are used to communicate information about the objects and events in the world, one cannot simply identify the messages that speakers communicate and the meanings of the sentences that they employ because one and the same sentence can be used to convey different messages in different contexts.

260

Richard Schantz

What a speaker means in uttering a typical sentence is generally underdetermined by the linguistic meaning of the sentence. Therefore, we must distinguish between literal meaning and speaker meaning, between what is said and what is implicated by a speaker. Very roughly, semantics is the study of literal linguistic meaning, while pragmatics is the study of the use of language to communicate information, information that often goes far beyond what the expressions used mean.

II We are chiefly concerned here with the metaphysics of meaning, with the nature of literal linguistic meaning. Our foundational question is: what are meanings? Are they entities, and if so, what sorts of entities are they? In our everyday practice we use sentences to express our thoughts, and so the idea quite naturally suggests itself that the meaning of a sentence is something like the content of the thought, or the proposition, it is typically used to express. This idea is reflected in a classic approach to semantics, according to which theories of meaning are theories of meanings or propositions, of things expressed by the use of declarative sentences. These propositions are thought of as the primary bearers of truth and falsity. A sentence is true or false derivatively, by expressing, in the context in which it is uttered, a true or false proposition. Moreover, advocates of this approach hold that propositions perform several other philosophically relevant functions as well. They are taken to be the objects of the so-called propositional attitudes like belief, doubt, and wish, the objects of illocutionary acts such as assertion and denial, and the things that are necessary, possible or contingent. Of course, this approach has led, in turn, to questions about the nature of propositions themselves. What are propositions, and in what kind of relation do we stand to them? Traditional answers to these questions conceived of propositions as denizens of a “third realm”, which we somehow grasp or apprehend by a mysterious faculty of intellectual perception. In light of these rather puzzling answers, many philosophers preferred to dispense with proposition altogether, and to develop an alternative approach to semantics, one, which attempts to avoid including meanings as entities in their ontology. The most important of these alternative approaches is the view that a theory of meaning is a theory of truth conditions—not a theory of meanings or propositions. Most of the work within the truth-conditional paradigm owes its inspiration, either directly or indirectly, to Alfred Tarski’s groundbreaking

From Truth Conditions to Structured Propositions

261

work on truth.1 Tarski had shown how a formal language can be provided with an interpretation by specifying a model and by using a definition of truth in a model to assign truth conditions to every sentence of the language. He demonstrated how it is possible to derive the truth conditions of all sentences of certain artificial languages from clauses specifying the referents of their subsentential constituents. Rudolf Carnap was then the first philosopher who applied Alfred Tarski’s methods of defining truth predicates for formal languages to the philosophical and scientific study of meaning.2 What Carnap did, was to employ Tarski’s technical methods of introducing interpreted formal languages in order to describe the meanings of sentences of already meaningful formal languages, such as the languages spoken by logicians. Later on, Donald Davidson made the philosophically important suggestion that theories of truth can be used as theories of meaning for natural languages, too. He came to believe that Tarski’s own despair of the possibility of applying his methods to natural languages was unwarranted. Davidson’s optimism rested upon the great theoretical advances that had been made by philosophers of language, logicians, and linguists who had been engaged in employing and enriching Tarski’s techniques to bring larger and larger areas of natural languages under logical and semantic control. It is worthwhile to examine carefully the development of Davidson’s enormously influential approach to the semantics for natural languages.

III There have been many attempts to approach the nature of meaning in a direct and informal way. Davidson, in contrast, does not aim to analyze the ordinary concept of meaning in other terms. Rather, he approaches the nature of meaning indirectly, by way of constructing and confirming formal, axiomatic truth theories for natural languages, modeled on axiomatic truth definitions for formal languages in the style of Tarski. In his foundational early paper Theories of Meaning and Learnable Languages, Davidson lays down a requirement that any adequate theory of meaning for a natural language must satisfy: In contrast to shaky hunches about how we learn language, I propose what seems to me clearly to be a necessary feature of any learnable language: it 1 Tarski 1958 2 Cp. Carnap 1942

262

Richard Schantz

must be possible to give a constructive account of the meaning of the sentences in the language. Such an account I call a theory of meaning for the language, and I suggest that a theory of meaning that conflicts with this condition, whether put forward by a philosopher, linguist, or psychologist, cannot be a theory of a natural language; and if it ignores this condition, it fails to deal with something central to the concept of a language.3

What Davidson calls a “constructive account of the meaning of the sentences in the language”, is usually called a “compositional theory of meaning”. A theory of meaning is compositional if it shows how the meaning of any sentence or complex expression depends on the meaning of its component words and the ways these words have been combined with one another. Such a theory must, on the basis of only finitely many axioms, governing the semantic properties of the primitive expressions and of the ways of putting them together, generate a meaning-specifying theorem for each complete sentence of the language. The derivation of a meaning-specifying theorem for a sentence displays how the meaning of the sentence is determined by the meanings of its significant parts and their modes of combination. By placing individual sentences in the context of a complete theory for the language, axiomatic semantic theories help us to reveal the structure that is present in natural languages. So compositional semantic theories seem to appreciate that in order to understand a natural language, we have to know both the meanings of its primitive expressions, and to know the entailment relations that hold among its sentences on the basis of their structures. The insight that natural languages have a compositional structure is the foundation of Davidson’s program in the theory of meaning. Certainly, the credit of having given prominence to the significance of the requirement that an adequate theory of meaning has to reveal the compositional structure of natural languages belongs to Davidson. The reason he offers for the compositionality requirement, in his famous learnability argument, is that only if natural languages are compositional, can we understand how they can be learned, how “an infinite aptitude can be encompassed by finite accomplishments”.4 Speakers of natural languages are finite beings, with finite capacities. They can learn languages, in a finite amount of time, languages that contain an infinite number of nonsynonymous sentences. And speakers do not possess magical

3 Davidson 2001a, 3 4 Ibid. 8

From Truth Conditions to Structured Propositions

263

abilities. Here we can rely on certain empirical assumptions: “for example, that we do not at some point suddenly acquire an ability to intuit the meanings of sentences on no rule at all.”5 The conclusion seems to be that there are only finitely many primitive expressions on the basis of which we are able to understand a potential infinity of sentences because their meanings are systematically dependent upon the meanings of their constituent words and rules for their combination. Moreover, by assuming a finite vocabulary of primitive expressions, we can also explain our ability to understand an indefinitely large number of novel sentences; we understand them by understanding the familiar words of which they are composed and the rules for their composition. A compositional meaning theory is a formal theory with a finite base, which should enable any person who knows it to understand every sentence of the language. Hence such an axiomatic theory should generate, for each sentence of the language under study, a theorem that specifies the meaning of that sentence. What form should such a theory take? The first proposal Davidson considers is that the theory should yield, for each sentence of the object language L, a theorem of the form: (1) s means m in L, where “s” is replaced by a structural description of a sentence of the object language, and “m” by a singular term of the metalanguage which refers to the meaning of the sentence. Davidson rejects the traditional view of meanings or senses as entities emphatically.6 His rejection of the view that meaning is some sort of thing, which certainly was prevalent in the long history of philosophy, is not based on a general rejection of Platonism and a fondness for nominalism. Rather, he is deeply convinced that reifying meanings is neither necessary nor sufficient for constructing a semantic theory for a natural language. As he puts it: Paradoxically, the one thing meanings do not seem to do is oil the wheels of a theory of meaning—at least as long as we require of such a theory that it non-trivially give the meaning of every sentence of the language.7

The fundamental difficulty confronting theories that assign reified meanings to expressions is to explain how to derive the meanings assigned to 5 Ibid. 9 6 Davidson 2001b, 20–1 7 Ibid.

264

Richard Schantz

the complexes from the meanings assigned to their components. Davidson illustrates the problem with reference to the simple sentence “Theaetetus flies”.8 Assigning a meaning to the singular term and to the predicate of this sentence does not exhibit how their concatenation generates a new meaning. If we suppose that each term refers to, or expresses, a meaning, then their concatenation produces merely a list. If we further suppose that concatenation is a significant part of the sentence, we might assign to it the relation of instantiation. But this does not help. We just get one more meaning, and still do not understand how to combine these meanings to generate the meaning of the whole. Davidson reminds us that Frege tried to avoid the threatening regress by treating the entities associated with predicates as “unsaturated” or “incomplete”, and critically remarks that “this doctrine seems to label a difficulty rather than solve it”.9 So Davidson is persuaded that the assignment of meanings to expressions is not sufficient to furnish a compositional meaning theory. What we entirely lack so far are rules governing the ways complex expressions are composed of their primitive parts. And, once we are in possession of such rules, we see that the assignment of meanings to expressions does no work at all in explaining how we are able to grasp the meaning of complexes on the basis of our grasp of their elements. Hence, according to Davidson’s reasoning, postulating an ontology of meanings seems to be superfluous. We shall later come back to the fundamental question of whether the view that meanings are entities is really explanatorily worthless. Alternatively, the theorems of a meaning theory might have the form: (2) s means in L that p, where “s” is replaced by a description of a sentence of the object language, and “m” is replaced by a sentence of the metalanguage that gives its meaning. But Davidson also jettisons this format. In his eyes, the basic difficulty facing this proposal, with its direct use of the notion of meaning, is that the “. . . means that . . .” construction creates intensional contexts, contexts in which the substitutability salva veritate of coreferential or coextensional terms, or sentences with the same truth value, fails. As he tersely puts it, “it is reasonable to expect that in wrestling with the logic of the apparently

8 Ibid. 17 9 Ibid.

From Truth Conditions to Structured Propositions

265

non-extensional ‘means that’ we will encounter problems as hard as, or perhaps identical with, the problems our theory is out to solve.”10 Davidson’s problem with this suggestion seems to be that in “. . . means that . . .” locutions we can substitute expressions only if they are synonymous, that is, have the same meaning. So the option of developing a logic that seeks to represent intensional features presupposes that we are already capable of specifying the meanings of every expression of the language. And indeed the only systematic explanation of intensional contexts we possess has to make use of the concept of meaning. Thus, it becomes obvious that a meaning theory that generated theorems of form (2) would presuppose the very concept that it is the task of a meaning theory to scrutinize. Davidson’s contrasting aim is to deliver a meaning theory that is couched in purely extensional terms. His own positive proposal is to use an axiomatic truth theory in the style of Tarski’s pioneering work on truth theories for formal languages to do the work of a compositional meaning theory. His famous proposal is expressed in the following passage: The theory will have done its work if it provides, for every sentence s in the language under study, a matching sentence (to replace ‘p’) that, in some way yet to be made clear, ‘gives the meaning’ of s. One obvious candidate for matching sentence is just s itself, if the object language is contained in the metalanguage; otherwise a translation of s in the metalanguage. As a final bold step, let us try treating the position occupied by ‘p’ extensionally: to implement this, sweep away the obscure ‘means that’, provide the sentence that replaces ‘p’ with a proper sentential connective, and supply the description that replaces s with its own predicate. The plausible result is (T)

s is T if and only if p.

What we require of a theory of meaning for a language L is that without appeal to any (further) semantical notions it place enough restrictions on the predicate ‘is T ’ to entail all sentences got from schema T when ‘s’ is replaced by a structural description of a sentence of L and ‘p’ by that sentence.11

What Davidson requires of a satisfactory meaning theory is that it yields theorems that match mentioned object language sentence with used object language sentences that have the same meaning, in a way that uncovers

10 Ibid. 22 11 Ibid. 23

266

Richard Schantz

structure. And he suggests that a recursive characterization of a predicate “is T ” of the object language sentences that yields all instances of (T ) will do the trick. It is clear that the predicate “is T ” will apply to all and only the true sentences of the object language. So it turns out that the requirement a theory of meaning has to meet is in essence Tarski’s Convention T on a materially adequate formal definition of a truth predicate for a language. Convention T requires that an adequate theory of truth for a formal language should have as consequences all instances of the (T ) schema (T)

s is true if and only if p,

where “s” is replaced by a structural description of an object language sentence, and “p” is replaced by that sentence or a translation of it, depending on whether or not the object language is contained in the metalanguage. Hence, it is Davidson’s positive proposal that a truth theory meeting Tarski’s Convention T, appropriately modified to accommodate the context sensitive features of natural languages, can do the work a compositional theory of meaning has to do. Initially, Davidson was convinced that a theory of truth for a language can count as a theory of meaning just because it gives the truth conditions of every sentence of the language. He said, for example, . . . the definition works by giving necessary and sufficient conditions for the truth of every sentence, and to give truth conditions is a way of giving the meaning of a sentence. To know the semantic concept of truth for a language is to know what it is for a sentence—any sentence—to be true, and this amounts, in one good sense we can give to the phrase, to understand the language.12

Davidson cherished the hope that a theory of truth itself could serve as a theory of meaning. However, a basic question arises right at the start of this project: how can a theory of truth deliver meanings? It only seems to deliver truth conditions. And statements of truth conditions are weaker than statements of meaning. From instances of schema M (M ) s means in L that p

we can infer instances of schema T (T ) s is true in L if and only if p.

12 Ibid. 24

From Truth Conditions to Structured Propositions

267

But there is no acceptable converse rule that permits us to infer instances of schema (M ) from instances of schema (T ). Nonetheless Davidson makes the surprising claim that a merely extensionally adequate truth theory for a natural language would suffice for understanding it. So he must have thought that a theory satisfying the requirement of extensional adequacy would ipso facto satisfy Tarski’s Convention T, that is, would ipso facto be interpretive. There is no explicit requirement that the sentence used on the right-hand side of each T-sentence be a translation of the sentence mentioned on the left. Rather, the hope was that if a truth theory for a language meets certain other constraints, the right-hand side would at least be an approximation to a paraphrase of the sentence mentioned on the left. Note that the connective “if and only if ” in the T-sentences is the material biconditional. A material biconditional is true just in case the sentences flanking the connective have the same truth value. Hence a standard T-sentence such as (ST )

“Snow is white” is true if and only if snow is white

is true. But so is the nontranslational or noninterpretive T-sentence (S )

“Snow is white” is true if and only if grass is green

since the sentence on the right-hand side of the connective has the same truth value as the sentence mentioned on the left. Obviously, however, we are not tempted to conclude that the sentence “Grass is green” can be used to give the meaning of the sentence “Snow is white”. There is no conceptual connection between the greenness of grass and the truth of the sentence “Snow is white”. This points out how weak the sense is in which Davidson’s theories of truth state the truth conditions of sentences. In the usual, strong sense, two sentences are regarded as having the same truth conditions just in case they have the same truth value in all possible state of affairs or circumstances. But a T-sentence does not state what would and must be the case if the sentence to be interpreted were true. Rather, it merely states, that, as things actually are, the relevant sentence is true if and only if some state of affairs obtains. Astonishingly, for Davidson, two sentences have the same truth conditions if and only if they have the same truth value. Since T-sentences furnish only weak, extensionally adequate truth conditions, Davidson himself was well aware that he has to deal with the problem of whether a correct truth theory for a language could have as consequences true but nontranslational or noninterpretive T-sentences like

268

Richard Schantz

S. He responded to this serious problem by, heroically, telling us that we should not think that a theory that entails (ST ) is any more correct than one that entails the deviant (S ) instead, . . . provided, of course, we are as sure of the truth of (S ) as we are of the truth of its more celebrated predecessor. Yet (S ) may not encourage the same confidence that a theory that entails it deserves to be called a theory of meaning. The threatened failure of nerve may be counteracted as follows. The grotesqueness of (S ) is in itself nothing against a theory of which it is a consequence, provided the theory gives the correct results for every sentence (on the basis of structure, there being no other way). It is not easy to see how (S ) could be party to such an enterprise, but if it were—if, that is, (S ) followed from a characterization of the predicate ‘is true’ that led to an invariable pairing of truths with truths and falsehoods with falsehoods—then there would not, I think, be anything essential to the idea of meaning that remained to be captured. What appears to the right of the biconditional in sentences of the form ‘s is true if and only if p’ when such sentences are consequences of a theory of truth plays its role in determining the meaning of s not by pretending synonymy but by adding one more brush-stroke to the picture which, taken as a whole, tells us what there is to know of the meaning of s; this stroke is added by virtue of the fact that the sentence that replaces ‘p’ is true if and only if s is.13

We see that Davidson attempts to rule out deviant T-sentences like (S ) by invoking a holistic constraint on meaning. A truth theory allows us to derive a T-sentence for each sentence of the object language on the basis of its semantically significant parts and compositional structure. But it is not suggested that the meaning of each sentence is given entirely by that individual T-sentence which states its truth conditions. Rather, the idea underlying this picture is that the meaning of each sentence is revealed only by locating its position in the language as a whole. The meaning of each sentence does indeed depend on the meanings of its components. But the meanings of the components, in turn, consist only in their systematic contribution to the truth conditions of all the sentences in which they might occur. So the job of interpreting each sentence of a language is distributed over the whole theory. What a competent speaker knows about meaning is not a set of unrelated T-sentences but, rather, a complicated

13 Ibid. 26

From Truth Conditions to Structured Propositions

269

system of relations that encompasses every expression, simple or complex, of the language. Moreover, unlike the formal languages Tarski was interested in, natural languages are context sensitive languages. Many sentences of natural languages contain expressions, in particular, indexicals and demonstratives, whose contributions to the semantic content of an utterance are determined only relative to the context of the utterance. Davidson assumed that the context sensitive elements of natural languages would place sufficiently strong constraints on the theory to prevent grotesque theories that issue in true but noninterpretive T-sentences like (S ). He said: Sentences with demonstratives obviously yield a very sensitive test of the correctness of a theory of meaning, and constitute the most direct link between language and the recurrent macroscopic objects of human interest.14

Davidson urges that a theory that allows us to derive (S ) would also have to generate true theorems for “That is snow”, “That is grass”, “That is white”, and “That is green”. In order to derive (S ) the theory would have to include axioms stating that the predicate “is white” is true of green things, and that the predicate “is snow” is true of grass. With these axioms, and an additional axiom for the demonstrative, specifying that the referent of “that” as used by a speaker S at time t is the object demonstrated by S at t, however, one could derive not only accidentally true T-sentences like (S ), but also false T-sentences for demonstrative utterances like “‘That is snow’” is true if and only if it is grass”, and “‘That is white’” is true if and only if it is green”. And, of course, theories that imply false T-sentences must be dismissed. Thus, Davidson held the hope that truth theories that are not only true, but also holistic and capable of accommodating the context sensitivity of many sentences of natural languages, would deliver only translational theorems. However, all the constraints added so far do not suffice. John Foster was the first to note the crucial difficulty.15 He observed that the constraints proposed still cannot block the derivation of nontranslational T-sentences. Suppose we have an extensionally adequate truth theory T for an elementary quantificational language. If we now replace any predicate axiom, giving the application conditions of an objectlanguage predicate, with another that employs a metalanguage predicate that is 14 Ibid. 35 15 Foster 1976; cp. also Loar 1976

270

Richard Schantz

extensionally equivalent to the original but unlike in meaning, we will produce an alternative truth theory T’ that is also extensionally adequate. If the theorems of the original theory T were true, then the theorems of the modified theory T’ will also be true. But at most one of the theories can be interpretive. For example, the original truth theory T for German might include the following T-sentence (1) “Zitronen sind sauer” is true in German if and only if lemons are sour, which is plainly meaning-specifying. It is easy, however, to construct another truth theory T’ that satisfies all of Davidson’s constraints, but generates a different T-sentence for the German sentence “Zitronen sind sauer”. Modifying the axiom for the predicate “ist sauer” appropriately, so that it states that “is sour” applies to something if and only if it is sour and the earth moves, we can prove the T-sentence (2) instead of (1) (2) “Zitronen sind sauer” is true if and only if lemons are sour and the earth moves. Since it is true that the earth moves, (2) will be true if (1) is true. Evidently, however, the right-hand side of (2) does not give the meaning of the German sentence “Zitronen sind sauer”. Therefore, merely extensionally adequate truth theories cannot be regarded as proper theories of meaning.

IV Davidson himself quickly realized that his initial hope had to be abandoned. But if extensional adequacy and the constraints recommended as yet are still not sufficient, the obvious question arises of what additional constraints one needs to impose on a truth theory for a natural language to guarantee that the theory yields only meaning-preserving T-sentences like (1). Davidson dealt with this question in his essay Radical Interpretation,16 about which he remarked in his Reply to Foster that it criticizes his earlier attempts to say what the relation is between a theory of truth and a theory of meaning, and that he “tried to do better”.17 16 Davidson 2001c 17 Davidson 2001e, 171

From Truth Conditions to Structured Propositions

271

Radical Interpretation starts by asking two principal questions: (1) What could we know that would enable us to interpret the utterances of other speakers? (2) How could we come to know it? Davidson remarks that the first question does not ask what we do know that suffices to interpret all utterances of a speaker. And he adds that the second question does not concern the actual process through which speakers acquire knowledge of natural languages, or the ability to interpret one another. Rather, these questions seek out conceptual elucidation, and the project of radical interpretation is supposed to provide this elucidation by relating the central concepts of the theory of interpretation—such as meaning, truth, reference, rationality, belief, action, preference, etc.—to more primitive evidence that does not presuppose anything about their employment. Ultimately, the evidence available to the radical interpreter is confined to knowledge of a speaker’s actual and potential behavior with respect to his environment. The crucial point is that the radical interpreter is neither allowed to invoke knowledge of the meanings of any expression of the object language nor to invoke any detailed knowledge of the speaker’s propositional attitudes. Davidson claims that an appeal to knowledge of the finely discriminated contents of speakers’ propositional attitudes must be excluded, because it is ultimately based on the same kind of evidence as our attribution of meanings to their sentences. We cannot find out what speakers’ utterances mean independently of finding out what they believe, and indeed independently of what they want, intend etc., and we cannot find out what they believe or want or intend independently of finding out what they mean by what they say. So a theory of interpretation must produce the meanings of sentences and the contents of beliefs and the other propositional attitudes jointly. And it must accomplish this complex task on the basis of purely behavioral evidence. Davidson bases his expectation that the radical interpreter can break into the circle of meaning and belief on the assumption that he is able to identity the attitude of holding sentences true or accepting them as true.18 He maintains that arriving at knowledge of which sentences speakers hold true presupposes neither knowledge of the detailed contents of their beliefs, nor knowledge of the meanings of the sentences they hold true. Even the identification of speakers’ attitudes toward the truth of sentences ultimately depends on behavioral evidence. So behavioral facts, facts accessible to a person engaged in radical interpretation, are the basic facts, the facts that in the end must determine all

18 Davidson 2001c, 135

272

Richard Schantz

the meaning facts. It is the fundamental aim of a theory of interpretation to identify and systematize patterns in the behavior of people and their interaction with the environment. From the perspective of the radical interpreter, the key concepts of the theory of interpretation are theoretical concepts, instruments whose role it is to bring order in our understanding of behavior.19 It is Davidson’s proclaimed ambition to describe what we could know that would enable us to interpret a native speaker without appealing to any semantic concepts. And he is well aware that the constraint of extensional adequacy is not enough. So the pressing question arises what stronger constraints are really added in Radical Interpretation? The following passage, which puts emphasis on the empirical character of theories of meaning for natural languages and on the ensuing empirical restrictions on such theories as wholes, contains the core of his answer to that question. If we knew that a T-sentence satisfied Tarski’ Convention T, we would know that it was true, and we could use it to interpret a sentence because we would know that the right branch of the biconditional translated the sentence to be interpreted. Our present trouble springs from the fact that in radical interpretation we cannot assume that a T-sentence satisfies the translation criterion. What we have been overlooking, however, is that we have supplied an alternative criterion: this criterion is that the totality of T-sentences should (in the sense described above) optimally fit evidence about sentences held true by native speakers. The present idea is that what Tarski assumed outright for each T-sentence be indirectly elicited by a holistic constraint. If that constraint is adequate, each T-sentence will in fact yield an acceptable interpretation.20

The additional constraint that is supposed to be strong enough to guarantee that a theory of truth can be used for interpretation is that it should optimally fit evidence available to the radical interpreter. In other words, the theory should have been empirically verified or confirmed from the perspective of the radical interpreter. It is not immediately obvious, however, how the empirical confirmability of a theory of truth could ensure its interpretiveness. If the mere truth of the theory is not enough, why should one assume that confirmability by the procedures of the radical interpreter could signify a remarkable improvement? The fact that a theory is empirically confirmed, that it 19 Davidson 2001d, 161 20 Davidson 2001c, 139

From Truth Conditions to Structured Propositions

273

optimally fits evidence about sentences held true by speakers, epistemically justifies us to believe that the theory is true. But if the truth of a theory is not sufficient for interpretation, it is hard to see how the process of confirming that it is true could help much. There does not seem to be any additional property, which might be revealed by the process of verifying a theory, which is such that possession of it could convert a noninterpretive theory into an interpretive one. In later texts,21 Davidson clarified his intentions. He now declared that the crucial point of passing the empirical tests was to draw attention to the fact that the theorems of an empirically verified theory are laws of nature, appropriately supported by their instances, and so capable of supporting subjunctive and counterfactual claims. The confident hope was that laws about speakers’ behavioral dispositions, formulated as universally quantified biconditionals, might succeed in ruling out theories issuing in anomalous T-sentences. And indeed Davidson’s new suggestion allows him to reject contrived theories from which T-sentences of the following kind can be derived: (1) “Snow is white” is true if and only if snow is white and Aristotle was the teacher of Alexander the Great. We have seen above, how easy it is to construct such theories. All one has to do to get from one extensionally adequate theory to another is to replace a predicate axiom of the original theory with another that is extensionally equivalent but nonsynonymous, and a simple way to achieve this is to form a conjunction of the original axiom and a contingently true statement. The new truth theory will be true if the original theory is true. But though (1) is true, and a theorem of an extensionally adequate theory, it is not interpretive. It is an advantage of Davidson’s new suggestion to treat the theorems and axioms of a truth theory as laws of nature that it is capable of eschewing such bizarre consequences. For, plainly, it is not a matter of nomological necessity that snow is white and that Aristotle was the teacher of Alexander. There is no nomological connection between these two states of affairs. We can easily imagine a nomologically possible world in which snow is still white, but in which Aristotle has never been the teacher of Alexander. But even this new suggestion does not fill the bill. While it allows us to distinguish between predicates, which are only accidentally coextensive, 21 Davidson 2001e, 172–4; 2001b, 26 n.11; 1999, 688

274

Richard Schantz

it does not allow us to distinguish between predicates, which are nomologically coextensive, such as “is water” and “is H2O”, or “is silver” and “has atomic number 47”. What is more, there are even logically or conceptually necessarily coextensive predicates, which are nevertheless nonsynonymous, such as the three predicates “is a triangle”, “is a closed figure with three sides”, and “is a closed figure with three angles”. It is logically impossible for something to satisfy one of these predicates without satisfying the others. Nevertheless, they do not have the same meaning. In light of this problem, the familiar question immediately suggested itself anew of what further constraints have to be placed on acceptable truth theories to ensure that they really satisfy Convention T. But there is a difficulty lurking beneath the surface, a difficulty that goes much deeper. For even if we suppose that this obstinate difficulty had somehow been solved, there would still remain a more principal problem for regarding a theory of truth as a theory of meaning. Even if we knew an interpretive truth theory T for a language L, knowledge of the facts T states would not be sufficient for understanding L. The crucial point is that someone might know all the facts which are stated by T without knowing that T is interpretive. Such a person might hold true beliefs about the truth conditions of a sentence but false beliefs about its meaning, and hence would not understand the sentence. At first glance, there seems to be a natural response to this problem, the response namely, that in order to interpret speakers of L, one has to know a truth theory for L and, in addition, to know that it is a truth theory for L. Foster argues that Davidson came to realize that he had to revise his original position accordingly.22 The new position Foster attributes to Davidson is that knowledge that T is a truth theory for L, plus knowledge of the facts T states, is sufficient for understanding L. Foster attacks this thesis because, so formulated, there is no direct connection between the content of the truth theory and the claim of its T-theoreticity. The trouble, according to Foster, is that someone might know that T is a truth theory for L, and know all the facts T states, but nevertheless not know that T states those facts. The conclusion he draws is that what we have to know in order to use the theory to interpret speakers is what the axioms of the theory are, and what they mean or state. To remedy this defect, Foster makes the proposal, which he regards as “perhaps what Davidson is trying to convey”, that the knowledge that is

22 Foster 1976

From Truth Conditions to Structured Propositions

275

supposed to be sufficient for understanding L is knowledge expressed by the single sentence: “Some truth theory for L states that . . .”, where the blank is filled with the conjunction of all axioms of T.23 If we had the knowledge expressed by this claim, so Foster argues, we could deduce the meaning of any sentence of L. Consider an interpretive T-sentence like (1) “Zitronen sind sauer” is true in German if and only if lemons are sour. According to Foster’s proposal, we would know the proposition this T-sentence expresses, and we would also know that this proposition is expressed by a biconditional the right branch of which is a translation of the German sentence quoted on the left. Hence, it seems, we are in a position to infer (2): (2) “Zitronen sind sauer” means in German that lemons are sour. So, apparently, by knowing the facts which T states, we can work out what any sentence of L means. Foster claims that this revision does indeed secure interpretiveness. Nonetheless, he proclaims that the present proposal must be discarded. The reason for his negative estimation lies in his belief that the use of an intensional idiom like “states” in “Some truth theory for L states that . . .” in specifying an interpreter’s knowledge conflicts with Davidson’s fundamental philosophical aims. Therefore, he announces, Davidson’s grand plan is in ruins. In his reply to Foster, Davidson cleared up his position. First of all, he denied that he ever held the view Foster attributes to him, the view that does not appropriately connect the knowledge of a truth theory and the knowledge of its T-theoreticity. Moreover, he points out that the view he really advocates is precisely the view Foster thought might better capture his real intentions, the view according to which knowledge that some truth theory for L states that . . . suffices to understand L. Davidson goes on to argue that Foster’s criticism of his standpoint rests upon a serious misunderstanding. He insists that it is not his goal to characterize in purely extensional terms what a person has to know to be capable of interpreting each sentence of a language. A truth theory, the recursive core of a theory of meaning or interpretation, cannot allow an intensional logic. Davidson thinks that to satisfy Convention T, an extensional logic 23 Ibid. 20

276

Richard Schantz

must be used within the truth theory. But what we would have to know about the truth theory to be capable of using it for interpretation does not have to be formulated in the vocabulary of a purely extensional first-order language. As we have seen, it is because what the truth theory itself states is not strong enough to guarantee that it generates interpretive T-sentences that Davidson seeks to say something about it that is sufficient for interpretation. Foster seems to have been mislead in this respect by Davidson’s different and independent thesis that the semantics of natural languages can be given by using only the resources of an extensional truth-theoretic semantics. Davidson is committed to showing that the seemingly intensional constructions in natural languages are in reality extensional. So his famous paratactic semantic analysis of indirect discourse, which may also function as a model for an analysis of other intensional or opaque sentential contexts, treats an utterance of “Galileo said that the earth moves” as semantically equivalent to the utterance of two sentences, “Galileo said that” and “The earth moves”. The “that” is a demonstrative referring to the second utterance. Only the first sentence is asserted, the second has the function of giving the content of the subject’s saying.24 Davidson contends, and rightly so, that two problems have to be kept apart. One is whether the paratactic account of indirect discourse is correct and can properly be applied to “states” or, as he remarks, to the “slightly more appropriate” word “entails”.25 The other problem is whether the knowledge the radical interpreter has to possess to employ a truth theory for interpretation can be specified in a non-question-begging way, that is, without making use of unanalyzed linguistic concepts. These problems are largely independent of each other. So even if Foster were right that the paratactic analysis founders, that would not be directly relevant to Davidson’s project to explicitly state something knowledge of which would suffice for interpretation. The following passage is a succinct statement of Davidson’s proposal: Indeed it still seems to me right, as far as it goes, to hold that someone is in a position to interpret the utterances of speakers of a language L if he has a certain body of knowledge entailed by a theory of truth for L—a theory that meets specified empirical and formal constraints—and he knows that this knowledge is entailed by such a theory.26 24 Cp. Davidson 2001h 25 Davidson 2001e, 178 26 Ibid. 172

From Truth Conditions to Structured Propositions

277

Although Davidson succeeded in refuting Foster’s objection, his position still seems to be faced with another intractable problem. I think the problem to be considered now is the hardest problem for the truth-theoretical approach to semantics. The problem arises because the interpretiveness of the truth theory guarantees only that among its theorems will be one interpretive T-sentence. But it does not enable us to determine which T-sentence that is.27 In general, truth theories will also generate a lot of T-sentences which are not interpretive. For example, suppose we have a proof of some T-sentence like “s is true if and only if p”, and that q is any logical truth. Then the alternative T-sentence “s is true if and only if p and q” will also be a theorem of the theory. It cannot be the case that both of these T-sentences are interpretive. The problem, then, is to identify which among the infinitely many theorems an interpretive truth theory provides actually are its interpretive theorems. Here Davidson’s grand semantic approach seems to come to grief.

V In an excellent exegetical and critical investigation of Davidson’s philosophy, Ernie Lepore and Kirk Ludwig offered a defense of the truth-theoretic format for semantics, which, however, differs in some important respects from its development in Davidson’s hands.28 In particular, they do not accept Davidson’s central constraint that one should not employ any semantic concepts beyond those needed in the theory of reference in describing what is required for a truth theory to provide interpretations of utterances of sentences of the object language. In this context, they reject what they call “Replacement Theory”, according to which it is Davidson’s aim to abandon the pursuit of a theory of meaning, as it was traditionally conceived, in favor of a more austere truth theory. Against the commentators who espouse the Replacement the theory, such as Stephen Stich,29 Lepore and Ludwig object, rightly in my view, that it was not Davidson’s intention to replace the theory of meaning with theories of truth and reference, but, rather, to use a truth theory to do the work of a compositional meaning theory.

27 Cp. Soames 2003, 306–9; 2009a; 2009b 28 Lepore & Ludwig 2005; cp. also Lepore & Ludwig 2007 29 Cp. Stich 1976

278

Richard Schantz

Lepore and Ludwig argue that their account of the relation between a theory of truth and a theory of meaning depends on distinguishing between two different, though related, projects: Davidson’s “initial project” and his “extended project”. The initial project is the one propagated in Truth and Meaning, the project of explaining the meanings of complex expressions on the basis of their structure and the meanings of their significant components. By contrast, the extended project seeks not only to explain how the meanings of complexes depend on the meanings of their components, but is also motivated by the more ambitious aspiration to provide insight into what is for any linguistic expression, primitive or complex, to have the meaning it does. Lepore and Ludwig claim that the initial project gradually became subordinated in the development of Davidson’s philosophy of language to the extended project, the project of illuminating the concept of meaning in general. And they argue that it is Davidson’s desire to pursue the extended project that motivates his requirement that the theory of meaning should not make use of any semantic notions other than truth and reference. We know that what Davidson wants to exclude is any appeal to notions such as “meaning”, “proposition”, “synonymy”, and “translation”. It is this requirement that Lepore and Ludwig object to. They maintain that to meet the goals of the initial project, the goals of a genuinely compositional semantics, we can presuppose knowledge of the meanings of the primitive expressions of the language. As Davidson himself once said, “it was not in the bargain also to give the meanings of the atomic parts”.30 Thus, the constraint the authors place on truth theories in order to serve as theories of meaning is that they must have “interpretive” axioms, axioms giving interpretations of expressions of the object language by using translations of them in the metalanguage to provide reference and satisfaction conditions for them.31 It is an advantage of the requirement that the truth theory is interpretive that it automatically solves the extensionality problem, the problem of limiting the class of extensionally adequate theories to only those the logical consequences of which will always include interpretive T-sentences. However, even if it is assumed that a truth theory is interpretive, another problem still remains, the “determination problem”, as Lepore and Ludwig fittingly call it. This is the problem formulated above of determining

30 Davidson 2001b, 18 31 Cp. Lepore & Ludwig 2005, 71–74

From Truth Conditions to Structured Propositions

279

which among the infinitely many different theorems of an interpretive truth theory actually are its interpretive theorems. Lepore and Ludwig claim that the determination problem can be solved by way of appeal to an appropriately restrictive characterization of the concept of a canonical proof.32 Their strategy consists in describing a formal proof procedure for identifying, for each object language sentence, a single theorem that specifies its meaning. The interpretive T-sentences are supposed to be all and only the canonical theorems of the theory, the theorems resulting from a canonical proof. The guiding idea is to portray the canonical theorems of the theory as those theorems which are derived solely on the basis of the content of the axioms, and whose proofs rely on no additional resources of the logic of the truth theory. Such a proof of a T-sentence will reveal how the meanings of the parts and their mode of combination combine to determine the meaning of the object language sentence. The authors give a relatively informal illustration of how they intend to restrict the logic of the metalanguage so that only interpretive T-sentences can be derived from its axioms. The main device to achieve this aim is to characterize the permissible rules of inference in such a way that the gratuitous introduction of tautologies at any stage in a proof is prevented. Suppose this policy did in fact succeed. Then we could be assured that the canonical theorems of the theory are T-sentences which satisfy Convention T. Since the interpretiveness of the axioms ensures the interpretiveness of the T-sentences derived from them, the constraint that the truth theory is interpretive seems to supply all we need to give the meaning of every sentence in the object language. But the catch in Lepore’s and Ludwig’s tempting proposal is its stipulation that in constructing a truth theory for a language we can simply exploit what we know about the meanings of the primitive expressions of the language. I think that Davidson was right to extend his initial project and to require that a substantive theory of meaning should be capable of illuminating the concept of meaning in general. A theory of meaning has to explain both the meanings of sentences and the meanings of words. A compositional theory of meaning that simply helps itself to the meanings of the primitive expressions is seriously incomplete. It neglects an essential aspect of meaning. The very first sentence of Davidson’s Inquiries into Truth and Interpretation poses the question “What is it for words to mean what they do?” But, as we have seen, his truth-theoretic semantics is mainly concerned with

32 Ibid. 109–112

280

Richard Schantz

the compositional aspect of meaning, with how our understanding of complex expressions depends on our understanding of their contained primitive expressions and the various ways they are put together. Thereby, we gain insight into the semantic form of many constructions of natural languages, into the combinatorics of language. Unfortunately, however, Davidson’s approach is only marginally concerned with the meanings of words. As we have seen, he firmly believes that sentence meaning automatically determines word meaning, because the meaning of a word is exhausted by its contribution to the meanings of the sentences in which it occurs. The axioms specifying the meanings, or rather the referents, of names and predicates do not correspond to anything in reality. There is no objective relation of reference in the world out there waiting to be discovered by us. In his article The Inscrutability of Reference, Davidson announces that there is no answer to the questions of what our singular terms refer to, and our general terms are true of.33 From his interpretive perspective, semantic reality is more a matter of imposing than discovering.34 Accordingly, Davidson and Davidsonians like early McDowell hold that the referential axioms, as pieces of the deductive machinery, have the purely instrumental function of helping to derive the T-sentences as theorems; they do not reflect any independent reality. The point of contact between a total semantic theory and the hard physical facts is not at the level of its axioms but exclusively at the level of its theorems.35 I find it very difficult to believe that the concept of reference is a purely theoretical concept. Truth, the utterance of true sentences, is not the only route to reference. Indeed, I think that examples of the kind given by proponents of the causal theory of reference or the direct reference theory in order to refute the traditional descriptivist paradigm show that reference has a certain autonomy with respect to truth, that reference cannot be exhaustively characterized in terms of the concept of truth. To find these examples convincing is not, as Davidson suggests, to be committed to a simple-minded building block approach, which starts with the meanings of names and simple predicates and then builds up the meanings of complex expressions.36 33 34 35 36

Davidson 2001g; cp. also Davidson 2001f Cp. Devitt 1984, 163–181 Cp. McDowell 1978 Cp. Davidson 2001f; Schantz 2001

From Truth Conditions to Structured Propositions

281

VI What is the alternative? Davidson and his numerous followers have for a long time nourished the hope that a theory of truth could qualify as a theory of meaning. But a theory of truth, howsoever embellished, does not deliver meanings; it only delivers truth conditions. This is the source of all the troubles of truth-theoretic semantics.37 The crucial point is that statements of meaning are stronger than statements of truth conditions. There have been many attempts to find a conception of truth conditions that is strong enough to capture meaning. Some have tried to read the “if and only if ” in sentences of the form “s is true in L if and only if p” as a strict or modal biconditional, expressing broadly logical equivalence, and, accordingly, to replace the notion of material truth conditions by the notion of strict or modal truth conditions.38 But this move is unpromising, too. For from s is true in L if and only if grass is green and the truth of Grass is green if and only if ( grass is green and 2 + 3 = 5) we obtain s is true in L if and only if ( grass is green and 2 + 3 = 5)

from which, evidently, the meaning of “s” cannot be deduced. Therefore, neither material nor strict truth conditions are sufficient conditions for the meanings of sentences. It is the task of a theory of meaning to tell us what sentences, and their subsentential constituents, mean. So far no one has developed a theory that permits us to deduce an instance of schema (M ) “s means in L that p” for each sentence of a language, from axioms about their constituents. My hopes lie elsewhere. I think that the best way for a theory of meaning to achieve its end is to pair sentences and their parts with their certain entities that are indeed their meanings. Traditionally, the meaning of a declarative sentence has been conceived of as the proposition it expresses. This view has regained popularity among contemporary philosophers of language and mind. As already indicated, propositions are said to be the things that are true or false, the things we believe, assert or doubt, the things synonymous sentences have in common, and the things that possess modal 37 Cp. also Horwich 2005, 198–221 38 Davies 1981, 27–8

282

Richard Schantz

properties. In light of the central role propositions play in many regions of philosophy, I think it is time to learn to live with the fact that there really are such things. But what is the metaphysical nature of propositions? According to one familiar proposal, propositions are functions from possible worlds to truth values, so that the proposition expressed by a given sentence can simply be identified with the set of possible worlds or circumstances in which it is true.39 It is a characteristic feature of this conception that propositions lack the kind of semantic structure which the sentences that express them possess. Indeed, propositions are defined independently of language and mind altogether. It is, however, an unacceptable consequence of this feature of propositions that they are individuated extremely coarsely. Indeed, on the possible worlds view, all necessarily equivalent propositions turn out to be identical just because they are true in all possible worlds. Since all necessary truths are necessarily equivalent to each other, this analysis implies that there is only one necessary truth. Thus, all mathematical truths, no matter whether of arithmetic or geometry, express the same proposition, although they may express it in many different ways. Moreover, this view implies that a person cannot believe a certain proposition without also believing all the propositions that are necessarily equivalent to it. This cannot be right. Obviously, a person may believe, for example, that Paris is the capital of France without believing that Paris is the capital of France and that 7 is a prime number. These consequences of the view that propositions are sets of possible worlds are very hard to swallow. It is no wonder, therefore, that many philosophers came to believe that there must be something wrong with the proposed identity conditions for propositions. They share the view that we are in need of a more fine-grained conception of the things that are in the first instance true or false, and that we assert, believe or doubt. In this dialectical situation, several authors focused their attention once again on the classic conception of Frege and Russell, according to which a declarative sentence expresses a structured proposition, a complex entity, whose parts or constituents are the meanings or semantic values of the words occurring in the sentence. In sharp contrast to the possible worlds account, Fregean and Russellian propositions exhibit structures similar to the syntactical structures of the sentences expressing them.40 This is an important advantage of this approach. Since structured propositions encode both syntactical structures and the meanings of subsentential expressions, 39 Cp. Montague 1974; Stalnaker 1984 40 Cp. Frege 1892a; 1892b; Russell 1903

From Truth Conditions to Structured Propositions

283

they enable us to recognize that sentences exhibiting significantly different structures can express different propositions, although they are true in exactly the same set of possible worlds. For example, the propositions expressed by the sentences “Bachelors are unmarried men” and “Drakes are male ducks” are both necessarily true. But manifestly the propositions expressed by these sentences consist of different constituents and hence are distinct.41 We have excellent reasons to believe that structured propositions of this kind really exist. For it is standard practice in semantics to associate singular terms, predicates, transitive verbs, and logical expressions with their semantic values, with objects, properties, relations and logical operations. These semantic values are the constituents of structured propositions. The structured proposition consists of its constituents standing in certain more or less complex relations to one another. And it seems to be this fact, the fact that the constituents are configured in a certain way in the proposition that enables the proposition to represent the world as being a certain way. This was a key idea of Wittgenstein’s Tractatatus.42 In general, advocates of structured propositions prefer Russellian propositions, which are compounded out of objects, properties, and functions, to Fregean propositions or thoughts, which are compounded out of complete and incomplete senses. Unfortunately, however, neither Frege nor Russell provided a satisfactory explanation of the mechanism that binds the parts of structured propositions together. They did recognize that a proposition is more than the sum of its parts. But they did not succeed in solving the problem of the unity of propositions. Frege claims that the sense of a sentence, the thought or proposition it expresses, consists of the senses of the words contained in the sentence. What is supposed to hold the whole proposition together is the fact that at least one of its senses is incomplete or “unsatured”, as Frege likes to say. Hence other senses are needed to complete the incomplete senses and so to form the unified proposition. I think that Davidson’s critical commentary, mentioned above, that “this doctrine seems to label a difficulty rather than solve it” is indeed apt. And Russell’s suggestion that the semantic value of the verb furnishes the required unity founders for the cogent reason that it implies that there are no false propositions. So we have to go beyond the classic conception of propositions laid out in the writings of Frege and Russell. It is the credit especially of Scott 41 Cp. Soames 2009d 42 Wittgenstein 1921

284

Richard Schantz

Soames, Nathan Salmon, and Jeffrey King to have developed novel theories of structured propositions that make substantial progress in our attempts to solve the puzzles that neither the classic account nor the possible worlds account of propositions were able to solve.43 There are considerable differences between the accounts these philosophers offer which, however, I can largely ignore for present purposes. The fundamental idea underlying their position is that the structure of propositions is inherited from the semantically significant aspects of the syntactic structures of the sentences by means of which they are expressed. Thus, propositions are not intrinsically representational; they do not possess their truth conditions independently of languages and minds. Rather, they themselves stand in need of interpretations. It is an important aspect of this new conception of structured propositions that traditional explanatory priorities are radically reversed. Sentences and mental states do not acquire their representational capacity in virtue of their standing in certain relations to intrinsically representational propositions. Rather, propositions acquire their representational capacity in virtue of the relations in which they stand to the sentences we use to express them. This reversal of explanatory priorities opens up the possibility of giving an entirely naturalistic account of propositions. After all, if this view is correct, the representational properties of propositions are grounded in the interpretive practices of us, of speakers of natural languages. Of course, this is only a rough sketch of an explanation of the nature of structured propositions and of our cognitive relations to them. Although grand strides have already been made, a lot of work still remains to be done in order to arrive at a full-fledged metaphysical, epistemological and formal theory of structured propositions. But there seem to be no principled obstacles to carrying out this ambitious enterprise. So, in the face of the many troubles for the truth-theoretic perspective, I think that this is currently the most promising project to pursue in studies on the foundations of semantics.

43 Soames 2009c; 2009d; 2002; Salmon 1986; King 2007

From Truth Conditions to Structured Propositions

285

References Carnap, Rudolf: 1942, Introduction to Semantics, Harvard University Press, Cambridge/MA Davidson, Donald: 1990, “The Structure and Content of Truth”, Journal of Philosophy 87, 279–328 Davidson, Donald: 1999, “Reply to James Higginbotham”, in Hahn, Lewis Edward (ed.), The Philosophy of Donald Davidson, Open Court, Chicago, 687–9 Davidson, Donald: 2001a, “Theories of Meaning and Learnable Languages” in Inquiries into Truth and Interpretation, Clarendon Press, Oxford, 3–15 Davidson, Donald: 2001b, “Truth and Meaning”, in Inquiries into Truth and Interpretation, Clarendon Press, Oxford, 17–36 Davidson, Donald: 2001c, “Radical Interpretation”, in Inquiries into Truth and Interpretation, Clarendon Press, Oxford, 125–139 Davidson, Donald: 2001d, “Belief and the Basis of Meaning”, in Inquiries into Truth and Interpretation, Clarendon Press, Oxford, 141–54 Davidson, Donald: 2001e, “Reply to Foster”, in Inquiries into Truth and Interpretation, Clarendon Press, Oxford, 171–79 Davidson, Donald: 2001f, “Reality without Reference”, in Inquiries into Truth and Interpretation, Clarendon Press, Oxford, 215–25 Davidson, Donald: 2001g, “The Inscrutability of Reference”, in Inquiries into Truth and Interpretation, Clarendon Press, Oxford, 227–41 Davidson, Donald: 2001h, “On Saying That”, in Inquiries into Truth and Interpretation, Clarendon Press, Oxford, 93–108 Davies, Martin: 1981, Meaning, Quantification, Necessity, Routledge, London Devitt, Michael: 1984, Realism and Truth, Blackwell, Oxford Foster, John: 1976, “Meaning and Truth Theory”, in Evans, Gareth & McDowell, John (eds.), Truth and Meaning, Oxford University Press, Oxford, 1–32 Frege, Gottlob: 1892a, “Über Sinn und Bedeutung”, in Patzig, Günter (ed.), Funktion, Begriff, Bedeutung (1980), Vandenhoeck & Ruprecht, Göttingen, 40–65 Frege, Gottlob: 1892b, “Über Begriff und Gegenstand”, in Patzig, Günter (ed.), Funktion, Begriff, Bedeutung (1980), Vandenhoeck & Ruprecht, Göttingen, 66–80 Horwich, Paul: 2005, Reflections on Meaning, Clarendon Press, Oxford King, Jeffrey: 2007, The Nature and Structure of Content, Oxford University Press, Oxford Lepore Ernie & Ludwig, Kirk: 2005, Donald Davidson. Meaning, Truth, Language, and Reality, Clarendon Press, Oxford Lepore Ernie & Ludwig, Kirk: 2007, Donald Davidson’s Truth-Theoretic Semantics, Clarendon Press, Oxford Loar Brian: 1976, “Two Theories of Meaning”, in Evans, Gareth & McDowell, John (eds.), Truth and Meaning, Oxford University Press, Oxford, 138–61 McDowell, John: 1978, “Physicalism and Primitive Denotation: Field on Tarski”, Erkenntnis 13, 131–52 Montague, Richard: 1974, Formal Philosophy: Selected Paper of Richard Montague, Richard Thomason (ed.), Yale University Press, New Haven Russell, Bertrand: 1903, Principles of Mathematics, W.W. Norton, New York

286

Richard Schantz

Salmon, Nathan: 1986, Frege’s Puzzle, MIT Press, Cambridge/MA Schantz, Richard: 2001, “Truth and Reference”, Synthese 126, 261–81 Soames, Scott: 2002, Beyond Rigidity, Oxford University Press, New York Soames, Scott: 2003, Philosophical Analysis in the Twentieth Century, vol. 2, The Age of Analysis, Princeton University Press, Princeton Soames, Scott: 2009a, “Truth, Meaning, and Understanding”, in Philosophical Essays, vol. 1, Natural Language: What it means and how we use it, Princeton University Press, Princeton, 208–24 Soames, Scott: 2009b, “Truth and Meaning—in Perspective”, in Philosophical Essays, vol. 1, Natural Language: What it means and how we use it, Princeton University Press, Princeton, 225–47 Soames, Scott: 2009c, “Semantics and Semantic Competence”, in Philosophical Essays, vol. 1, Natural Language: What it means and how we use it, Princeton University Press, Princeton, 182–201 Soames, Scott: 2009d, “Direct Reference, Propositional Attitudes, and Semantic Content”, in Philosophical Essays, vol. 2, The Philosophical Significance of Language, Princeton University Press, Princeton, 33–71 Soames, Scott: 2010, What is Meaning?, Princeton University Press, Princeton Stalnaker, Robert: 1984, Inquiry, MIT Press, Cambridge/MA Stich, Stephen: 1976, Davidson’s Semantic Program, Canadian Journal of Philosophy 6, 201–27 Tarski, Alfred: 1956, “The Concept of Truth in Formalized Languages”, in Logic, Semantics, and Metamathematics, Clarendon Press, Oxford Wittgenstein, Ludwig: 1921, Tractatus Logico-Philosophicus, in Schriften Bd.1 (1960), Suhrkamp, Frankfurt

Five Flies in the Ointment: Some Challenges for Traditional Semantic Theory Gabriel M.A. Segal 1. Five challenges for traditional semantic theory Truth-conditional semantic theory for natural language appears to be flourishing. The work deploys apparatus from the tradition of Frege, Tarski, Carnap, Davidson and Montague.1 But is all as well as it seems? Work in this tradition relies, obviously, on notions like truth, reference, satisfaction and extension. It assumes that many sentences of natural language, or utterances of them, are true. It assumes that many words of natural language refer to, extend over, or are true of things in the world. If those assumptions are faulty, then all is not well with semantics for natural language. And these core assumptions do face serious challenges. In this chapter, I will consider a few well-known ones that are particularly interesting, mostly emanating from the work of Noam Chomsky. I will list them first, then look at how they might be met. Challenge 1: A watermelon is green on the outside and red on the inside. I visit a greengrocer’s shop. I want to know if the watermelon I have picked up is one of those that is red on the inside, as opposed to some other colour. In response to my query the greengrocer says: ‘Yes it is red’. I buy the watermelon. I visit my friend, the artist. I have the watermelon with me. He asks to borrow it so that he can include it in his still-life. He explains that this work hinges on contrasts between red and other colours. ‘I shall place this tomato, which is red, next to the watermelon, which is not red’. Challenge: the watermelon both is and is not in the extension of ‘red’, so the whole idea of a predicate having an extension is kaput.

1 Not all the major contributors to the tradition believed that natural language has a formal semantics. In my view, Tarski and Carnap did not. Montague and Davidson did. And Frege did too, at least in his later years. See Segal (2006) for discussion.

288

Gabriel M. A. Segal

Here is a very similar example involving ‘water’. I am in a restaurant, I ask the waiter for a glass of water. He brings me a cup of tea. I say ‘That isn’t water’. Some tea leaves have been dumped into the reservoir that supplies water to my house, as a purifier. What comes out of my tap is chemically indistinguishable from the tea in the restaurant. But I still call it ‘water’. Challenge: tea both is and is not in the extension of ‘water’. The whole notion of extension is kaput.2 Challenge 2: Theseus’s ship sails the seas. As its timbers age, they are replaced with new ones. The old timbers are kept, restored and combined to form a ship. Call the original ship ‘A’, the ship that had all its timbers replaced, ‘B’, and the ship that was, at the end of the story, made from the original planks, ‘C’. The individuation conditions we associate with ‘ship’ entail that A is the same ship as B and that A is the same ship as C. But B is not the same ship as C. So ‘x is the same ship as y’ extends over and over , but not over . The whole notion of extension is kaput. Challenge 3: Like ships like people. Individuals A and B are in the hands of a mad scientist. She explains that she has an apparatus that will copy all the information in A’s brain into B’s and vice versa.3 After that, one person will receive a million pounds and the other will be tortured. Call the person whose body is A’s body at T1, the “A-body-person-at-T1”, the person whose body is B’s body at T1, the “B-body-person-at-T1”, etc.. Then we have this picture: T1 info transfer at T2 T3 A-body-person-at-T1- - - - - - - - - - - - - - - - - - - A-body-person-at-T3 B-body-person-at-T1- - - - - - - - - - - - - - - - - - - B-body-person-at-T3

2 The watermelon example is adapted from Travis (1997). The ‘water’ example is from Chomsky (1995, 22). Here is another one. In an anthropology class the lecturer says ‘Humans drink many varieties of impure water. In poor countries the water is sometimes dirty or contaminated. In more prosperous ones, impurities are deliberately added to enhance the taste or other beneficial properties—tea leaves for example’. I find this example very compelling. But it might be objected that the anthropologist is using ‘water’ in a technical, scientific sense and that his term is not the same as the natural-language homonym. I am not convinced. The anthropologist could be speaking ordinary language. One type of folk language game involves adopting terms from scientific discourse. 3 We can assume that the entire wiring pattern is copied. 4 The example is from Williams (1970)

Five Flies in the Ointment

289

The scientist asks B to decide which body-person will be tortured and which will receive the million pounds.4 B chooses the A-body-person-at-T3. Upon awakening, the A-body-person-at-T3 recalls having made the decision as the B-body-person-at-T1 and is very pleased. Hence are in the extension of ‘x is the same person as y’. A person is a human being. The A-body-person-at-T1 is the same human being as the A-body-person-at-T3. Hence the A-body-person-atT1 is the same person as the A-body-person-at-T3. Hence are in the extension of ‘x is the same person as y’. But of course are not in the extension of ‘x is the same person as y’. The whole idea of extension is kaput. Let us suppose the A-body-person-at-T1 is called ‘Fred’. The name ‘Fred’ refers to the A-body-person-at-T3. And it refers to the B-body-personat-T3. But it cannot refer to both. The whole idea of reference is kaput. Challenge 4: An utterance of (1) could easily be true. (1) John gave a book to Mary, but she already had it, so he read it himself then shredded it. But then ‘book’ extends over objects that are both abstract and concrete and ‘it’ refers to something that is both abstract and concrete.5 But nothing is both abstract and concrete. So the ideas of extension and reference are kaput. Challenge 5: An utterance of (2) could be true. (2) The average American family has 2.3 children. But traditional semantic theory is committed to treating the sentence as having something like the form partially depicted in (2 ): (2 ) (x)(family(x) & average-for-American(x) & has ) And utterances of (2 ) are, if meaningful at all, then false. 5 Chomsky (2003). See Ludlow (2003), to which Chomsky is replying, for discussion of this example and the general topic of this paper.

290

Gabriel M. A. Segal

Similarly, an utterance of (3) could be true: (3) Joe Sixpack’s priorities are changing But according to traditional semantics, all such utterances should be false since the proper name doesn’t refer. Likewise for (4) and (5): (4) Sherlock Holmes is more famous than any real detective (5) Sherlock Holmes is a fictional detective And (6) too, because traditional semantics treats ‘some’ as an existential quantifier, implying real-world existence: (6) Some fictional detectives are more famous than any real detective And finally: (7) There is a flaw in the argument. An utterance of (7) could be true. According to the traditionalist, this could only be so if flaws were things. But they are not.6 The problems we have with the individuation of the denizens of the manifest image: ships, water, books, people and so on might lead one to think of the manifest image as ‘a dream modulated by sensory input.’7 (Llinas 1987, quoted with approval in Chomsky 1995). According to Chomsky, all these problems go away if we refrain from thinking of manifest kinds and particulars as real-world objects. They are not things out there for our commonsense words and concepts to extend over or refer to. Real kinds and particulars are to be found only in the scientific image.8 And there is no overlap between the two. It is, according to Chomsky, ok to say things like ‘books exist and so do people, but unicorns don’t’ using ordinary language. In Wittgensteinian spirit: language is alright as it is. But

6 Chomsky (1981, 324), Hornstein (1984, 58). 7 The actual quote is ‘perception is a dream modulated by sensory input’, from Llinas 1987. 8 The expressions ‘manifest image’ and ‘scientific image’ are from Sellars (1962). I don’t think Chomsky uses the terms. But they are apt.

Five Flies in the Ointment

291

this is a relatively non-committal usage (‘without metaphysical import’). It would not do, for example, to translate the claim into (8) and then give it a standard semantics: (8) (x)(z)(book (x) & person (y) &-(z)(unicorn z)). For Chomsky “the semantic properties of an expression focus attention on selected aspects of the world as it is taken to be by other cognitive systems, and provide intricate and highly specialized perspectives from which to view them, crucially involving human interests and concerns even in the simplest cases.” For example if I am inside a house I can clean it, affecting the inside, but I cannot see it, unless an exterior surface is visible. And if I am inside it, I cannot be near it, even though, in the unmarked case, it is surface (like a cube, to which the same point applies). The semantic properties of an expression provide “instructions” to “conceptual-intentional” systems for building up these intricate representations, which can then be used to interpret speech acts. (Chomsky 2005, 20). Well maybe that is right. It is an appealing picture. But what is the alternative? Certainly, objects and kinds in the manifest image are not reducible to objects and kinds in the scientific image. Even manifest material particulars are not physical objects, in the sense of being objects that could be identified in physics. Ships and cups are not bundles of molecules. Remove a molecule from a healthy ship and you are left with the same ship, but not with the same bundle of molecules. The same goes for orangutans, planets, elm trees and steel girders. But all of these feature in the special sciences: geology, astronomy, botany and material science respectively. These sciences seem to be perfectly respectable. If orangutans, planets and steel girders can be the subject of serious scientific study without being reducible to objects in physics, then maybe the same could be said for some of those things that appear in the manifest image only. Maybe— if, that is, we can get around the problems of individuation introduced in the Challenges and their kin.9 So how should a traditional semantic theorist who is a realist about the manifest image respond to the challenges? The next section outlines what I take to be the most promising package of answers.

9 Such as sorites paradoxes, about which I have nothing much to say here.

292

Gabriel M. A. Segal

1.2. Addressing the challenges 1.2.1 Watermelons and water Let us begin with the problem of red melons and glasses of impure water. The solution I recommend is offered by Rothschild and Segal (2009). The traditional semanticist has two main options. He could deny the apparent data and hold that the sentences don’t have context-dependent truth values after all. He might argue that the greengrocer’s utterance of ‘It is red’ isn’t really true. Only objects that are largely red on the outside are really red, he might say. He might argue that tea really is water, but that we are reluctant to call it so in a restaurant because that would be confusing. Or, indeed, he could take the opposite line and argue that in the tap context, I am wrong to call the tea ‘water’. Denying the apparent data is methodologically questionable. Speakers’ intuitions pretty consistently opt for context-sensitivity of truth values in these and a wide variety of similar cases. Of course, it is theoretically possible that these intuitions rest on some kind of confusion of semantic and pragmatic factors, or some other confounding factor. But there doesn’t appear to be any evidence for this. The alternative is to claim that there is some sort of indexicality going on. There are three ways to pursue this option. One: one can suppose that there are one or more variables present at logical form, in the underlying syntax of the sentences. Zoltan Szabo, for example, has argued that ‘red’ associates with two variables at logical form, one designating a comparison class to specify how red the object has to be and one designating the part of the object that has to be red: ‘red(c, p)’ (Szabo 2001). Two: one can suppose that one or more variables are supplied by the context. These are not present in the object language. But they are present in the meta-language and enter into the interpretation of the target sentences. Third: one can suppose that words like ‘red’ and ‘water’ are themselves indexical: they have different extensions in different contexts of utterance. Thus in the greengrocer’s context the watermelon is in the extension of ‘red’ and in the context of the artist’s studio it is not. In my view, only the third option is viable. Let’s switch to ‘tall’. Suppose that Fred is 6ft tall. Then, in a typical conversational context in the Netherlands, an utterance of ‘Fred is tall’ would be false. In a typical conversational context in China, an utterance of ‘Fred is tall’ would be true. Suppose now that ‘tall’ means something like ‘tall[c]’, ‘c’ being a variable that picks out a comparison class.10 In the each context, ‘c’ picks out some

Five Flies in the Ointment

293

salient group. Fred is not tall relative to the Dutch. Fred is tall relative to the Chinese. So everything is fine. Except it isn’t. The proposal gets the extensions right. But not the meanings. An utterance of ‘tall’ in a context that determines a specific comparison class doesn’t mean what an utterance of ‘tall[c]’ with ‘c’ referring to that class would mean. Let me explain why. One might know, e.g., that Fred is tall relative to Chinese people without knowing what it is to be tall relative to Chinese people: for example, simply because one doesn’t know how tall the Chinese people are. Now, please imagine Fred in China. He is in a supermarket with some local friends. They want to purchase an item that is on a high shelf and they can’t quite reach it. One of them says (8): (9) Fred can reach it. He is tall. Compare (9) with (10), assuming that the variable refers to the group of friends: (10) Fred can reach it. He is tall[c] Someone who didn’t know how tall members of the group were could know that the thought expressed by ‘He is tall[c]’ in (10) is true, without knowing that the thought expressed by ‘He is tall’ in (9) is true.11 Hence the utterances express different thoughts, hence they mean different things. Notice also that there is a certain ad hocery about the extra-variable approach. What is the relevant parameter for counting as water in the restaurant context versus counting as water in the tap context? It is not comparison class. In the restaurant, the water in the water glass might, for all I know, contain a lower percentage of H2O than the water in my water glass. In the restaurant, the relevant classification system is one for distinguishing different kinds of drink. Chemical composition is only one relevant factor among

10 As proposed by Ludlow (1989). 11 Let me put the point more clumsily but less ambiguously. Let us call the thought expressed by (10) ‘T1’ and the thought expressed by (9), ‘T2’: someone could stand an attitude of propositional knowledge to T1 without standing in an attitude or propositional knowledge to T2. So, using the normal Fregean method of distinguishing contents, T1 and T2 are different thoughts.

294

Gabriel M. A. Segal

others.12 So what would the extra variable pick out? ‘C’ for ‘classification system’ perhaps? Or simply ‘c’ for ‘context’. Again the proposal gets the extension right, but not the meaning. One could know that something is water[c] without knowing what it is to be water[c].13 Thus what is expressed by ‘water’ in a particular context of utterance differs from what is expressed by ‘water[c]’, with ‘c’ referring to that context.14 Let us consider option three, treating the target terms as indexicals. The next subsection simply expounds Rothschild and Segal’s proposal.15

1.2.1.1

Rothschild and Segal’s account of indexical predicates is an extension of the standard approach taken by T-theoretical semanticists to singular indexicals, particularly ‘that’. On this approach, one proves what are sometimes called

12 See Malt (1994) for an empirical study of some of the uses of ‘water’. Szabo (2006) denies that ‘water’ shifts its extension in the way Chomsky claims it does. Responding to Chomsky’s example, he writes: “I disagree: I think many of us would be reluctant to stand by both judgments upon learning the chemist’s verdict [that the stuff coming out of the tap and tea in a tea cup are chemically indistinguishable]; we might not know which one to give up, but that does not mean that they must have the same standing.” I disagree: most of my informants are happy with both judgments. 13 Could ‘c’ work like a special kind of indexical, such that understanding its use in a specific context entails knowing what the relevant parameter is, under the right description? Could ‘tall[c]’ mean, something like ‘tall for this context’, where one is only in a position to understand ‘this context’ if one knows what it is to be tall for the context? I doubt it very much. Suppose that you are in the supermarket in Beijing and you encounter the group of friends. You haven’t met them before. But you have had a few words with them about the shop and its goods. You are part of the conversation, when one of them says (9). You are blind and you don’t know how tall the friends are. Then you understand (9) in the manner of ‘He is tall for this context’. But you don’t know what it is to be tall for the context. So you don’t understand (9) in the way the speaker and the other participants do. You don’t know which thought the utterance expresses. 14 Any attempt to extend the variable approach will have to contend with many different kinds of parameter. Consider for example an utterance of ‘the shoes are under the bed’ when they are eight floors below. And, for another, ‘the door must be kept shut at all times’, which sometimes means: except when you are passing through it, and sometimes doesn’t. Thanks to Travis for both. 15 Rothschild and Segal have other arguments against the extra-variable approach.

Five Flies in the Ointment

295

‘conditionalized T-sentences’. These are items along the lines roughly sketched in (11): (11) If u is an utterance of ‘that is a watermelon’, and the speaker uses ‘that’ in u to refer to x, then u is true iff x satisfies ‘is a watermelon’. The information specified in (11) is context-independent. Now suppose that you are in a particular context of utterance and you know that the speaker used ‘that’ to refer to watermelon D, then you can move on to (12): (12) u is true iff D is a watermelon. The extension of the basic idea to predicates, not an entirely straightforward affair, proceeds as follows. An utterance of ‘Fred is tall’ is true in a certain Chinese context, but false in a certain Dutch context. This is because different standards are operative in the different contexts: to be tall by the standards of the Chinese context is easier than to be tall by the standards of the Dutch context. We distinguish tokens of ‘tall’ in different contexts by subscripting. And we keep contexts and tokens in line using numbers. Thus: if the Chinese context is the kth context, all tokens of ‘tall’ governed by the standards of that context are subscripted with ‘k’: “tallk”. “Tallk” is a syntactic type (and ‘k’ is a numeral). All and only utterances of ‘tall’ in the kth context are utterances of “tallk” (although the numeral remains unpronounced). An object satisfies “tallk” iff the object is tall by the standards of the kth context. Correspondingly, we distinguish the sentences “Fred is tallj”, “Fred is tallk” etc.. And we think of these as evaluated relative to their contexts of utterance. Suppose again that the Chinese context is the kth. Then “Fred is tallk” is true relative to the context because Fred satisfies “tallk”, because Fred is tall by the standards of the context. Finally, we treat utterances as bearers of truth values absolutely: an utterance of a sentence in a context is true, absolutely, iff the sentence uttered is true relative to the context. When we say “Fred is tall by the standards of the Chinese context, but not by the standards of the Dutch context”, we are using “tall” in a rather abstract and general way. This is how we use “tall” when we are theorizing about height in general and related matters. Let us call this “tallg” (the gth context is the general one).

296

Gabriel M. A. Segal

Putting these ideas together we get: (A1) (x)(n)(x satisfies “Fred”, cn iff x = Fred) (A2) (x)(n)(x satisfies “is tall”^n, cn iff x is tallg, cn)16 (A3) (S)(NP)(VP)(If S = NP^VP, then ((n)(S is true, cn iff (x)(x satisfies NP, cn and x satisfies VP, cn))))17 (A4) (u)(n)(S) (if u is an utterance of S in cn, then (u is true iff S is true, cn)) (A1)-(A4) provide the context-independent T-theory. To get beyond this, you need to be in a particular context and to know what the standards of the context are. If you are in the kth context, you need information of this sort: (A5) uk is an utterance of Sk = ‘Fred is tallk’ in ck (A6) (x)(x is tallg, ck iff x is tallk) You can then combine the T-theory with the contextual information to get you to: (A7) uk is true iff (x)(x = Fred x is tallk) From which you might infer: (A8) uk is true iff Fred is tallk 1.2.2 Ships Given the way we individuate ships (artifacts generally, and other kinds of thing too), a ship can survive gradual replacement of all of its planks. A ship can also be gradually dismantled, have its planks removed, and then be put back together. So it can be that what began as a single ship can survive as two distinct ships. Given the logic of identity, this causes us a 16 For all x, for all n, x satisfies the expression composed of ‘is tall’ concatenated with the nth index relative to the nth context iff x is tall by the standards of the nth context. 17 For any sentence S, any noun phrase NP and any verb phrase VP, if S is the concatenation of NP and VP, then, for any n, S is true relative to the nth context iff for some x, x satisfies NP relative to the nth context and x satisfies VP relative to the nth context.

Five Flies in the Ointment

297

problem. We need to revise our individuation criteria for ships. A ship1 can survive gradual replacement of all of its planks, but cannot be dismantled and put back together. Once taken to bits, it’s history. A ship2 can be dismantled and reassembled. But it cannot survive gradual replacement of its parts. Once you have replaced fifty percent of its parts, it is history, and a new ship2 has come to be. It would be convenient if by ‘ship’ we sometimes meant ship1 and other times meant ship2 and the traditional semanticist could treat ‘ship’ like ‘tall’ and ‘red’. Life, though, is rarely convenient. Ship1 and ship2 are artificial concepts, not concepts deployed by Jane Winecooler. So how do we deal with Jane Winecooler’s utterance of “Aristotle owns lots of ships.”? I suggest a supervaluationist approach. It is true iff “Aristotle owns lots of ship1s” and “Aristotle owns lots of ship2s” are both true, false iff they are both false and indeterminate if exactly one is true.

1.2.3 People I believe that we are born dualists. We are born with an innate ‘theory of mind’, according to which minds are immaterial entities that could exist disembodied or move from body to body. There is decent (though not 100% conclusive) evidence for this from anthropology and psychology. (A) All natural human groups believe in ghosts (Boyer 2003). (B) Babies don’t expect humans to obey physical laws (Kuhlmeier et. al. 2004). (C) Untutored children tend to believe that people survive the death of their bodies (Bering and Bjorklund (2004)). (d) The mind/body problem persists: after all these years of physicalism, we remain baffled at the idea that a physical thing can think and feel.18 Cartesian dualism is false. The folk concept needs to be revised. But a workable concept of mind could retain many of its aspects. Minds cannot exist disembodied. But they can exist embodied. It seems to me that the concept of a mind that can move from one brain to another is not defective. We can coherently conceive of three-dimensional continuants individuated by relations of psychological continuity. In the Williams scenario there are things that fall under this slightly physicalist concept of mind: A’s mind, which moves into B’s body and B’s mind, which moves into A’s.

18 For further discussion see Segal (2007).

298

Gabriel M. A. Segal

The Williams scenario also, of course, features human beings. Does ‘person’ pick out the mind, or the human being? Does ‘Fred’ pick out the mind, or the human being? The lack of agreement amongst philosophers about these issues suggests that the answer is: “it is indeterminate”. It is indeterminate whether Fred is the A-body-person-at-T3 or the B-bodyperson-at-T3. I used to think that that meant that it was then open to each of us to choose which concept we wished to express by ‘person’ (and ‘self ’) and which object one would pick out with ‘I’. And I want to be a mind. If I were the B-body-person-at-T1, I would want the A-body-person-at-T3 to get the reward. But I no longer think I can choose to be a mind. If it is indeterminate whether I am a mind or a human being, then I can’t resolve the indeterminacy by ditching my old concept of self and replacing it with the slightly physicalist concept of a mind. The human being is a thinking thing, just as the mind is. And it can’t choose to be other than it is. Equally the mind already is the mind, and it doesn’t need to make a choice in order to be self-identical. If this is right, then it gives Fred in the Williams scenario a serious problem. When the scientist asks Fred to choose the fates of the later body-people, the human being should choose for the benefit of the B-body-person-at-T3 (or so it seems to me, at least) and the mind should choose for the benefit of the A-body-person-at-T3. And of course the two have not yet divided: the human being and the mind think and want and speak as one. If Fred chooses happily, without a feeling of conflict, then something loses out. If he chooses to benefit the A-body-person, then the human being loses out, if he chooses to benefit the B-bodyperson, then the mind loses out.19 There might be a problem here for persons, but that doesn’t mean that there is one for semantic theorists. Again, the theorist might reasonably opt for a supervaluationist approach to ‘person’ and ‘Fred’. I suggest that the supervaluationist approach recommended here for ‘ship’ and ‘person’ might extend to a wide range of similar cases, cases in which lay concepts don’t have fully determinate extensions. Consider, for example, the extensions of ‘hammer’, ‘dog’ and ‘tooth’ in the following scenarios. On a distant planet with no carpentry, there naturally occur

19 For extensive discussion see and a different view, see Noonan (1998). For general discussion see Eric Olson’s entry ‘Personal Identity’ in The Stanford Encyclopaedia of Philosophy.

Five Flies in the Ointment

299

objects that are intrinsically physically identical to hammers. On a distant planet, there are animals that are intrinsically physically just like Alsatians, but that share no ancestor with any Earth creature. On a distant planet, there occur naturally things exactly like your teeth. They grow in the ground by themselves and are not associated with eating. An android has objects just like dentures permanently in place in its mouth. He uses them to chew food. Arguably, these examples reveal indeterminacy in the extension of the terms: ‘hammer’, ‘dog’ and ‘tooth’. Supervaluationism can make sense of our quotidian use of these terms. 1.2.4 Books (1) means something like (1’): (1’)

John bought [(a copy of ) [a book]i]j for Mary. But Mary already had (a copy of ) iti. So he read itj then shredded itj

But then nothing in the logical form of (1) needs to extend over anything that is both abstract and concrete. Sometimes ‘book’ extends over abstract things, and sometimes it extends over concrete things. Likewise some occurrences of ‘it’ that are anaphoric on ‘book’ refer to abstract objects and others refer to concrete ones. But no occurrence of ‘book’ or ‘it’ extends over anything that is both abstract and concrete. In (1) ‘book’ extends over book types, which are abstract. (Peter had already bought it for her, which is why she already had it.) The first ‘it’ refers to the type. The second and third refer to a concrete thing, the copy that John bought, read and shredded. I don’t know how to get the meaning expressed in (1’) out of (1). I don’t know what the logical form of (1) is. I have no argument that it looks like (1’). But, as far as I can see, the problem of how to get the meaning (1’) out of (1) remains whether one is doing truth-theoretic semantics or describing instructions to the conceptual-intentional systems. And there is no apparent reason to suppose that the problem would be any easier to solve if embedded in the latter enterprise.20 20 Chomsky says very little about what a theory of instructions to conceptualintentional systems would look like. He thinks we know very little about the matter. Paul Pietroski (2006) has sketched a small proposal about how to do Chomsky semantics. It appears to be fairly widely agreed that if Chomsky is right about the nature of semantics, then a lot of what truth-theoretic theorists actually do could be recast in the theory of instructions. This looks plausible in the case of the little T-theory for ‘tall’ presented above.

300

Gabriel M. A. Segal

Chomsky (1995) has objected to this proposal. Suppose I am holding a book and say (13): (13) This book won the Pulitzer Prize The traditional semantic theorist has to treat ‘This book’ as referring both to the type and to the copy. I am not convinced. What won the prize was the type. So it is reasoanble to hold that ‘this book’ refers to the type. But it doesn’t also have to refer to the copy. It is true that the speaker draws attention to the copy as part of the pragmatic mechanism that secures reference to the type. But that doesn’t enter the semantics.21

1.2.5 Nonexistents There are three standard strategies that a semantic theorist can deploy in relation to (2)–(7): (i) posit funny objects, such as average American families, existing yet fictional objects and flaws, (ii) monkey around with the sentence’s logical forms in an effort to show that the apparent ontological commitment disappears under analysis, (iii) adopt a fictionalist approach, according to which utterances of the sentences are not literally true, but they relate in appropriate ways to thoughts that are literally true. I think that (i) offers a plausible account of (7) and (iii) offers a plausible account of (3)–(6). (2), though, requires more extended discussion, as follows. (2) can appear extremely problematic for a traditional semantic theory because none of (i)–(iii) look promising. The matter deserves some discussion. Notice first that there is an innocent and relatively unproblematic use of the ‘the average American’:

21 The problematic phenomenon is not restricted to books, nor to an abstract/ concrete duplex. We also have ‘I painted the door brown then walked through it’ and ‘The baby finished the bottle then broke it’ (both adapted from Chomsky 1995, drawing on Pustejoski 1993). It looks as though when two kinds of things are systematically related—figure and ground, contents and container—we can use the same term to pick out either one. Again, the response is: these phenomena are hard to explain. But as far as one can tell, they can be explained just as well within a traditional framework as within the proposed alternative.

Five Flies in the Ointment

301

(14) The average American is concerned about falling fertility rates. Here the ‘the’ is generic ‘the’, as in ‘the tiger likes to hunt at night’ and ‘average’ just means: typical (Stanley 2001, Higginbotham 1985). This use of ‘the average’ appears to be different from ‘the average’ in (2). Consider (15) a and b: (15) a. The average American man’s tastes are changing. He used to prefer beer to wine. Now he prefers wine to beer. b. The average American man is concerned about falling fertility rates. He used to have 2.3 children. Now he only has 2.1 One can’t combine the numerical constructions together with the typical use of ‘average’. The problematic ‘the average N’ construction appears to appear only in relation to numbers. The most successful account of this construction appears to be that of Stanley (2007).22 According to Stanley’s theory, the sentences at issue have the structure: [[NP The average ) N][VP P]]. [NP The average ) N] denotes a set of properties, derived as follows. ) denotes a function from properties to functions from dimensions to measure functions, defined only over the domain of those properties, where a dimension is a way of scaling things along a numbered scale (e.g. height in feet). So ) denotes a function of type