Colin McGinn-Philosophy of Language_ The Classics Explained-The MIT Press (2015).pdf

Philosophy of Language

Philosophy of Language The Classics Explained

Colin McGinn

The MIT Press Cambridge, Massachusetts London, England

© 2015 Massachusetts Institute of Technology

 

All rights reserved. No part of this book may be reproduced in any form by any electronic or mechanical means (including photocopying, recording, or information storage and retrieval) without permission in writing from the publisher.

 

MIT Press books may be purchased at special quantity discounts for business or sales promotional use. For information, please email [email protected].

 

This book was set in Stone by the MIT Press. Printed and bound in the United States of America.

 

Library of Congress Cataloging-in-Publication Data

 

McGinn, Colin, 1950–. Philosophy of language : the classics explained / Colin McGinn.  pages  cm Includes bibliographical references and index. ISBN 978-0-262-02845-5 (hardcover : alk. paper) 1. Language and languages—Philosophy.  2. Language and languages—Philosophy— Textbooks.  I. Title. P107.M38 2015 401—dc23 2014021824

 

10 9 8 7 6 5 4 3 2 1

Contents

Preface ix 1

Frege on Sense and Reference  1

1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 1.10

Background 1 Identity 3 Additional Machinery  10 The Conception of Sense  12 Reference 16 Ordinary and Extraordinary Use  18 Further Points on Sense and Reference  20 Problems with Frege’s Theory  23 Extension of Frege’s Theory beyond Singular Terms  25 Further Aspects of Frege’s Theory  31

2

Kripke on Names  35

2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9

Background 35 Kripke’s Critique  39 Rigid Designation  42 Kripke’s Epistemic Objections  45 The Causal Chain Theory  48 Objections to Kripke’s Critique  49 The Social Character of Names  51 Essential Descriptions  52 Impure Descriptions  53

3

Russell on Definite Descriptions  55

3.1 3.2 3.3 3.4 3.5

Indefinite and Definite Descriptions  55 Three Theories of Definite Descriptions  60 Indefinite Descriptions and Identity  63 Russell’s Rejection of Meinong’s Ontology  65 The Details of Russell’s Theory of Descriptions  67

vi Contents

3.6 3.7

Problems with Russell  72 Primary and Secondary Occurrences  74

4

Donnellan’s Distinction  77

4.1 4.2 4.3 4.4 4.5 4.6 4.7

Introduction 77 Referential and Attributive Uses  78 Denoting and Referring  84 Truth-Value Gaps  85 Evaluating Donnellan’s Distinction  87 Implication and Implicature  90 Further Objections to Russell’s Theory  94

5

Kaplan on Demonstratives  97

5.1 5.2 5.3 5.4 5.5 5.6

Intension and Extension  97 Kaplan on Indexicals  100 The Two Principles of Indexicals  102 Context of Use and Conditions of Evaluation  105 Possible Worlds, Meaning, and Indexicals  109 Kaplan on “Today” and “Yesterday”  113 

6

Evans on Understanding Demonstratives  115

6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8 6.9

The Fregean Theory of Indexicals  115 The Point of Indexicality  118 Evans’s Theory of Sense and Reference for Indexicals  119 Saying versus Showing  122 Mock Sense  124 Empty Names  125 Evans’s View of Names  126 Evans on “Today” and “Yesterday”  128 Character, Content, and Information  130

7

Putnam on Semantic Externalism  133

7.1 7.2 7.3 7.4

Background 133 Twin Earth and “Water”  134 Meanings Are Not in the Head  135 Criticisms of Putnam  143

8

Tarski’s Theory of Truth  147

8.1 8.2 8.3 8.4 8.5 8.6

Background 147 Tarski’s Criteria of Acceptability  149 Aristotle and the Redundancy Theory  151 Object Language and Metalanguage  155 How to Derive the T-Sentences  157 Satisfaction 159

Contents vii

9

Davidson’s Semantics for Natural Language  165

9.1 9.2 9.3 9.4 9.5

Background 165 The Merits of Tarski’s Theory as Applied to Meaning  168 Applying Tarski’s Theory to Natural Languages  175 Empirical Truth Theory  181 Criticisms of Davidson’s Theory  185

10

Grice’s Theory of Speaker Meaning  191

10.1 10.2 10.3 10.4

Background: Speakers and Sentences  191 Two Types of Meaning  193 What Is Speaker Meaning?  195 Consequences and Criticisms  199

Appendix: Kripke’s Puzzle about Belief  203 Notes 211 Index 215

Preface

This book is intended as a student text suitable for undergraduates taking a typical philosophy of language course. But it takes an unusual form: it undertakes to explain ten classic works in the field as clearly as I know how. So it is not the typical general survey of issues, but instead focuses on individual authors. It could also be used as an introductory text for graduate students with no background in philosophy of language. The book is not geared specifically to students with a strong interest and background in analytic philosophy; it aims to include students who may not even be specializing in philosophy. The aim is to make difficult primary material accessible to people who might otherwise struggle with it. The book consists of ten chapters (plus an appendix), each of which discusses in detail a single classic article. It is intended to be used in conjunction with an anthology of classic texts. The anthology I have used is Philosophy of Language: The Central Topics, edited by Susana Nuccetelli and Gary Seahy (Roman & Littlefield, 2008). It could also be used in conjunction with A. P. Martinich’s The Philosophy of Language (Oxford University Press, 2006), though the selection of articles is somewhat different in the two books. I have found in teaching the subject that students need a thorough, clear explanation of the classic texts, which by themselves they find too difficult. Accordingly, the chapters in this book go through the classic texts carefully and systematically. There is no attempt to give a general survey of the literature, achieving complete coverage, and the book does not deal with some of the more recent literature. The instructor would use this book as a supplement to the original articles, sparing him or her a lot of arduous exegesis. I have generally included some evaluation and criticism of the views and theories being expounded, but this is more to stimulate the student’s

x Preface

thought (and class discussion) than to contribute to the subject to the satisfaction of my professional colleagues. I have always aimed to make the material as simple as possible, without sacrificing accuracy. Everything is explained from the ground up. The book had an unusual gestation. It began when a student in my class at the University of Miami, Colin Mayer, suggested that it would be useful if there were a book offering the kinds of explanations I provided orally. I agreed but was reluctant to write such a book myself, not wanting to give up the time. He then suggested that he could transcribe the lectures from recordings he had been making of my classes. We decided to give it a try. He diligently set about the work. My task was to go over what he had written and revise it. I did that, finding it necessary to make many revisions (almost every sentence). I did, however, try my best to preserve the original spokenword form of the lectures, thinking that this might make the material more accessible. In pure writing there is always a tendency to value succinctness and precision (not to mention elegance) over sheer comprehensibility. The end result is a mix of informality and careful formulation. I am grateful to Colin Mayer for undertaking this work, which could not have been easy, and for his original suggestion. I also had the assistance of Monica Morrison, who went over the raw material of the transcriptions, cleaning up and formatting the text. All the final text, however, is due to me. It was a much tougher job than I bargained for, but I think the resulting book should be a boon for students and teachers alike. I first taught philosophy of language some thirty-eight years ago, and this is my distillation of many years of experience teaching the subject. I hope it achieves its aim of conveying a rich body of thought in an accessible form.   Colin McGinn Miami, July 2012

1  Frege on Sense and Reference

1.1 Background Before we begin to expound Frege’s views on sense and reference, a few words about the general aims of the philosophy of language might be useful. The most general thing we can say is that philosophy of language is concerned with the general nature of meaning. But this is not very helpful to the novice, so let us be more specific. Language is about the world—we use it to communicate about things. So we must ask what this “aboutness” is: what is it and how does it work? That is, how does language manage to hook up with reality? How do we refer to things, and is referring to things all that language does? Further, is referring determined by what is in the mind of the referrer? If not, what else might determine reference? Some parts of language we call “names,” but is everything in language a name? How is a word referring to something connected to a person referring to something? Do expressions like “Tom Jones,” “the father of Shakespeare,” and “that dog” all refer in the same way? In what way do these types of expressions differ in meaning? How is a sentence related to its meaning? Is the meaning the same as the sentence or is it something more abstract? Can’t different sentences express the same meaning? What is a meaning? Are meanings things at all? How is meaning related to truth? Whether what we say is true depends on what we mean, so is meaning deeply connected to truth? How are we to understand the concept of truth? What is the relationship between what a sentence means and what a person means in uttering a sentence? These are the questions typical of the philosophy of language. In this book we will consider these questions by reviewing what the greatest philosophers of language have said about them, beginning with perhaps the greatest of them all—Gottlob Frege.

2 

Chapter 1

Frege’s article “On Sense and Reference,” published in 1892, is the beginning of modern philosophy of language, shaping the field ever since. We shall therefore pay particularly close attention to its content, returning to it in later chapters. But before entering into detailed discussion of the article it is important we gain some familiarity with two concepts: sentences and propositions. A proposition is what is expressed by a sentence: the proposition expressed by a sentence constitutes the meaning of the sentence. Thus it is possible for two different sentences to express the same proposition. Two sentences that are synonymous with one another will express the same proposition. Sentences can differ in their constituent words and be synonymous, having the same meaning, and thus express the same proposition. The following two sentences illustrate this point: (1)  John is a bachelor. (2)  John is an unmarried male. The terms “bachelor” and “unmarried male” are synonymous, that is, they have the same meaning; therefore, these two sentences express the same proposition. Hence two different, nonidentical sentences of English can express the same proposition. Two sentences from two different languages can also express the same proposition. Here we have two synonymous sentences of different languages, French and English: (3)  La neige est blanche. (4)  Snow is white. Despite the fact that these two sentences are made up of different words in two distinct languages, they still have the same meaning, and thus express the same proposition. With this understanding of how sentences relate to propositions, we can now ask what a sentence is. A sentence is a collection of shapes, signs, or acoustic signals. Different shapes of letters on paper or acoustic signals in the air can correspond to the same proposition. Propositions, then, are very different from sentences—more abstract than physical. A sentence is the perceptible vehicle that expresses a proposition, and in addition can be uttered by a person. When you utter a sentence like “Snow is white,” you thereby make a statement. A statement is a relationship between three things: the speaker, the sentence, and the proposition. When a person speaks he utters a particular sentence, and in so doing he makes a certain

Frege on Sense and Reference 

3

statement. If a Frenchman utters the sentence “La neige est blanche,” he is stating that snow is white, even though he did not utter that sentence of English. However, since the sentence “La neige est blanche” is synonymous with the English sentence “Snow is white,” the two different sentences express the same proposition. A sentence in one language can be used to report the same proposition expressed by a speaker who made a statement using a different language. Sentences, statements, and propositions are systematically correlated, but they are not the same thing. A sentence is a physical sequence, a statement is a human action, and a proposition is an abstract meaning. 1.2 Identity In “On Sense and Reference,” Frege is concerned with the relationship between a sentence and the proposition it expresses. He is concerned with discovering answers to the following questions: What exactly is the relationship between a sentence and the proposition that it expresses? When is one proposition the same as another proposition expressed by a different sentence? What constitutes a proposition? What is the meaning of a word? The questions that concern Frege lead one to wonder how a sentence, considered as an arrangement of shapes or a sequence of sounds, can be meaningful. That is, we are concerned with sentences and their meanings—how they are able to tell us things about the world. What kind of thing is meaning? Frege’s article discussing these questions is not straightforward—it contains certain obscurities that are seldom if ever brought up by commentators, because they are difficult to interpret. In what follows, however, we will bring out and clarify the obscurities in Frege’s article. First, let us examine the opening of “On Sense and Reference”: Equality gives rise to challenging questions, which are not altogether easy to answer. Is it a relation? A relation between objects, or between names or signs of objects? In 1 my Begriffsschrift, I assumed the latter.

Though Frege is not explicit about what he means by “equality,” he is using the term in a mathematical sense (not a social one!). The notion of equality can be illustrated with a mathematical statement: “4 × 5 = 20.” Contemporary philosophers use “identity” instead of “equality.” The example “4 × 5 = 20” would be called an identity statement, asserting that the

4 

Chapter 1

number 4 × 5 is identical to the number 20. It is these types of statements that Frege intends when he uses “equality.” Identity can also apply to other nonmathematical cases. There are a few things about identity that Frege does not mention. Philosophers often distinguish between numerical identity and qualitative identity. Qualitative identity occurs when two things are exactly alike. For example, two cars that come from the same assembly line and have the same color and so on would be said to be qualitatively identical. Frege, however, is primarily interested in numerical identity. Numerical identity is the relationship a thing has to itself. The relation is a very primitive and trivial one: everything has the relation of identity to itself. Furthermore, numerical identity does not obtain between one object and another object, even if the two objects are qualitatively identical. For example, two twins do not have the relationship of numerical identity to one another—that relationship exists only between one of the twins and himself. We can now ponder the following: Is identity a relation? There are all sorts of relations: left of, older than, belonging to a political party, or living in a certain place. Each one of these examples illustrates a nontrivial relation, and therefore tells us something substantial about reality. However, in the case of identity, it has been argued that the relation something has with itself is trivial, and therefore gives no substantial information but provides only a tautology. Frege continues his explanation of identity in the following passage: The reasons which seem to favor this are the following: a = a and a = b are obviously statements of differing cognitive value; a = a holds a priori and, according to Kant, is to be labeled analytic, while statements of the form a = b often contain very valuable extensions of our knowledge and cannot always be established a priori. The discovery that the rising sun is not new every morning, but always the same, was one of the most fertile astronomical discoveries. Even today the identification of a small planet or a comet is not always a matter of course.2

In the above Frege is concerned with statements that identify objects. An identity statement using different names will have this form: “a = b” (“a is identical to b”). There is one object that we have referred to with two names, “a” and “b.” For illustration, let “a” be “4  ×  5” and “b” be “20.” We have referred to the object, a number, with the numeral “20,” as well as with the expression “4 × 5,” and now we form a corresponding identity statement. Two names that refer to the same thing create a true identity

Frege on Sense and Reference 

5

statement when they are written down and have the symbol “=” between them. On the other hand, if “a” does not denote something identical to what “b” denotes, then we will produce a false identity statement. The essence of Frege’s point here is that when he wrote the Begriffsschrift, he thought that when we make a statement like “a = b” the relation expressed by “=” is a relation between the names themselves. In this case, the statement really is about the names “a” and “b” and not the objects to which the names “a” and “b” refer. The names of the objects are separate from the objects that they designate. During his Begriffsschrift days, Frege thought that when he made an identity statement he was concerned with the names that were in that statement. This is because the alternative view seems to lead to absurdity: Now if we were to regard equality as a relation between that which the names ‘a’ and ‘b’ designate, it would seem that a = b could not differ from a = a (i.e. provided a = b is true). A relation would thereby be expressed of a thing to itself, and indeed one in which each thing stands to itself but to no other thing.3

It would seem that taking “=” to relate objects, not names, would make “a = b” express the same proposition as “a = a.” To illustrate this point more clearly, we can use the example of the two names, “Hesperus” and “Phosphorous.” Venus is the planet that first comes up in the evening, and used to be called “Hesperus.” “Hesperus” is a proper name denoting Venus; it corresponds to the definite description “the evening star” (we discuss definite descriptions in more detail in chapter 3). In using the name “Hesperus” we thus refer to Venus. Understanding advances in modern astronomy that the ancients did not, we know that “Hesperus” refers to Venus. The ancients, however, knew neither the name “Venus,” nor that Venus is a planet and not a star. The same heavenly body is also seen in the morning—when viewed in the morning the ancients called it “Phosphorous, the bringer of light.” Frege points out that the two different acts of naming in fact correspond to the same object. In the example, the two different names “Hesperus” and “Phosphorous” in fact correspond to the same heavenly body—Venus. It appears at one point in the sky in the evening and at another point in the sky in the morning. The ancients did not know that they were applying two names to the same planet. We can then say that Hesperus is identical to Phosphorous, stating a substantial astronomical discovery. The ancient Babylonians were not able to assert that Hesperus is

6 

Chapter 1

identical to Phosphorous, nor did they have any reason to think that. They were ignorant of the identity. The example of Hesperus and Phosphorous is a general illustration of the following point: there are many cases where a single object has been given a name, and then in another time and another context, given another name, without anyone realizing that the object has been named twice. When the identity is discovered, what the observer has learned, intuitively, is that one thing has two appearances, and therefore that a = b. Therefore, the two different appearances correspond to the same object, thus producing substantial identity knowledge. In such a case “a = b” forms an informative identity statement. We have expressed a proposition that is not trivial and gives us genuine knowledge about reality. By contrast, an identity statement of the form “a = a” (“Hesperus is Hesperus”) is not an informative proposition—it is merely a tautology. The numerical identity—any numerical identity— can be seen to hold without any empirical observations about the world at all. In the case of Hesperus, if someone merely heard the name “Hesperus,” she could know without observation that the statement “Hesperus is Hesperus” is true. It is not possible to do the same with the statement “Hesperus is Phosphorus.” This statement is informative, whereas the previous statement is not. Thus “Hesperus is Phosphorus” has empirical content and is synthetic (from Kant); but “Hesperus is Hesperus” is analytic, or tautological, and is true simply in virtue of its meaning. To sum up, “a = a” expresses an analytic, a priori proposition; “a = b” expresses a synthetic, a posteriori proposition. In the passages above from “On Sense and Reference,” Frege explains how these two propositions (expressed by “a = a” and “a = b”) are completely different. For example, there could have been a time in the past when people thought that every morning, a different fiery heavenly body appears in the sky. Understanding that that heavenly body—the sun—is the same one that appears in the sky every morning is a substantial empirical discovery. We know that it has the same appearance, but sameness in appearance does not entail that it is the very same object. But Frege raises the following question: if equality is a relationship between an object and itself, how could there be any difference between the propositions expressed by “a = b” and “a = a”? Would they not both be saying the same thing, namely, that an object is identical to itself? In other words, wouldn’t “a = b” express the same thing as “a = a”? So isn’t it better to suppose that

Frege on Sense and Reference 

7

identity is really a relation between the names themselves, because they are clearly different? The sentence “a = a” expresses the proposition that a is identical to itself, and the statement “a is identical to itself” is analytic and a priori. However, there is no way to argue that the statement “a = b” gives us the same proposition as “a = a.” As we said, merely knowing a name allows one to say that the object named is identical to itself. Even the ancients knew that Hesperus is identical to Hesperus and that Phosphorous is identical to Phosphorous. What they did not know is that Hesperus is identical to Phosphorous. Making the assumption that identity is a relation between an object and itself appears to lead to paradox when thinking about identity propositions. Frege therefore thought when writing the Begriffsschrift that identity could not be a relation between an object and itself. To avoid the paradox, the two different sentences must state different propositions—but how? If identity is a relation between names and not objects, then something different is being stated in the two cases. Thus “a = a” informs us that the name “a” denotes the same thing as the name “a.” On the other hand, “a = b” informs us that the name “a” denotes the same thing as the name “b.” Here we are no longer concerned with the objects themselves but with the names of them. If we are really talking about the names, then we can see how the two sentences produce different propositions. Why? Because “a = a” contains the name “a” and only the name “a,” whereas “a = b” contains the name “a” and the name “b.” The second sentence accordingly refers to something the first one does not refer to, namely the name “b.” It contains the name “b” and in this analysis the sentence refers to that name. This explanation shows how these two sentences can express different propositions: they are about different things, because they are really about names, not objects. The latter proposition is about the names “a” and “b” whereas the former is only about the name “a.” This way of thinking is a natural way to think about identity statements: an identity statement says that one name denotes the same thing as another name—not that one object is identical to itself. It is not generally the case that sentences containing names are about those names. In fact, sometimes statements have nothing to do with names at all. Consider a statement where someone says, “Hesperus is bright”— here he does not appear to be talking about the name “Hesperus.” Rather, he is talking about a planet, which is Venus, and stating that it is bright.

8 

Chapter 1

He is not saying the name “Hesperus” is bright. It is, of course, still possible that the name “Hesperus” is bright (e.g., the name “Hesperus” is written as a neon sign). However, in general, if someone says, “Hesperus is bright,” he is not saying that the name “Hesperus” is bright. We are not generally talking about our words, but using them to talk about something else. Notice that there is a huge difference between a name occurring in an ordinary statement where it refers to its bearer, and a name occurring in quotation marks in a statement when it refers to the name. Generally speaking, statements that use a name do not refer to that name. Therefore, making the claim that an identity statement like “Hesperus is identical to Phosphorous” refers to the names is to say something quite revisionary about that sentence. In actuality, the speaker intends for that statement to refer to the planet Venus, and does not intend for that sentence to refer to names of the object at all. This is sometimes called the use–mention distinction: we use the name to mention an object; we don’t use the name to mention itself—except when we expressly want to talk about words, not things. Looking back on his view in the Begriffsschrift, Frege now thinks he was wrong to take the view that identity is a relation between names. He illustrates this point in the following passage: What is intended to be said by a = b seems to be that the signs or names ‘a’ and ‘b’ designate the same thing, so that those signs themselves would be under discussion; a relation between them would be asserted. But this relation would hold between the names or signs only in so far as they named or designated something. It would be mediated by the connection of each of the two signs with the same designated thing. But this is arbitrary. Nobody can be forbidden to use any arbitrarily producible event or object as a sign for something. In that case the sentence a = b would no longer refer to the subject matter, but only to its mode of designation; we would express no 4 proper knowledge by its means. But in many cases this is just what we want to do.

Frege had tried to avoid the problem in supposing that identity is a relation between an object and itself because that would make identity propositions trivial. Bringing in the names themselves was intended to solve this problem. The phrase “mode of designation” in the above passage is meant to include the names themselves. But then the statement would refer to a mode of designation, not to a state of affairs in the world. The mode of designation then becomes what he here calls the “subject matter” of the statement. Frege now finds this objectionable, because we would not be expressing what he calls “proper knowledge.” The reader will wonder what

Frege on Sense and Reference 

9

Frege means by the phrase “proper knowledge.” To learn that Hesperus is Phosphorous is to learn something substantial, empirical, and a posteriori. But what proposition have we learned? It is clearly not the proposition that a is identical to a. The proposition instead states that the name “a” denotes the same thing as the name “b,” according to the earlier theory. However, Frege raises the objection that knowing that one name co-denotes with another is not enough to acquire “proper knowledge.” If we suppose that proper knowledge is knowledge that goes beyond a tautology, does the knowledge that “a” co-denotes with “b” go beyond tautology? Contrary to what Frege implies, it can be informative to learn that one name refers to the same thing as another name—very informative. It would be impossible to possess this knowledge ahead of time by just knowing the names independently. Through knowing the name “Hesperus,” one also knows that Hesperus is identical to Hesperus. However, to discover that in addition to this the name “Hesperus” denotes the same thing as the name “Phosphorous” is to learn something previously unknown. Effectively, we have learned that two different symbols denote the same thing. Isn’t this “proper knowledge”? It certainly isn’t a tautology. But Frege is suggesting that learning that Hesperus is Phosphorous is not only learning a linguistic fact but also understanding something significant about reality and the objects in the world. This statement has revealed a genuine empirical fact about two heavenly bodies. Frege’s earlier theory does not capture the fact that the person who comes to know the statement has learned something about the world. It reduces the fact learned to a merely linguistic fact, but the fact learned is not merely linguistic in nature. What one learns is not merely that the names have the same reference, but that two appearances correspond to the same object. The object of one’s knowledge, then, is not the same as that of someone who learns that one name refers to the same thing as another name. That would be learning something about two names, not two appearances. The real knowledge in the sentence “Hesperus is Phosphorus” comes from understanding something empirical about reality, not just something about language. Frege’s idea of “proper knowledge” is knowledge of the world, and not merely linguistic knowledge. Thus he rejects the linguistic theory of the content of identity statements, as well as the simple object theory—the theory that identity statements are only about objects, not linguistic items.

10 

Chapter 1

1.3  Additional Machinery To capture what is grasped when someone learns that “a = b” is true, we need another analysis of the proposition expressed by that statement. So far, we have seen two propositions that “a = b” might express: (5)  a = a (the object is identical to itself). (6) “a” denotes the same thing as “b.” Of course, these two things are both things one can know, but they are not what one learns from the proposition expressed by the sentence “a = b.” It may seem that we have exhausted all the possibilities on this matter. If so, this leads to a huge logical problem, because it means that we cannot even explain such simple identity statements as “2 +2 = 4.” This logical problem is why Frege is faced with the task of trying to account for something that seemingly cannot be accounted for. The purpose of “On Sense and Reference” is to bring in extra machinery to account for the meaning of “a = b” beyond what we have talked about so far: If the sign ‘a’ is distinguished from the sign ‘b’ only as object (here, by means of its shape), not as sign (i.e. not by the manner in which it designates something), the cognitive value of a = a becomes essentially equal to that of a = b, provided a = b is true. A difference can arise only if the difference between the signs corresponds to a difference in the mode of presentation of that which is designated.5

Frege here introduces the notion of a “mode of presentation” without much fanfare or explanation, and he contrasts it with a “mode of designation.” For Frege, the mode of presentation is what is essential to the meanings of the names “a” and “b,” the modes of designation—where the mode of designation is simply the name considered as a sign. What is needed in this account is a mode of presentation associated with the objects where that mode is not to be identified with the objects themselves or with their names. Frege states: Let a, b, c be the lines connecting the vertices of a triangle with the midpoints of the opposite sides. The point of intersection of a and b is then the same as the point of intersection of b and c. So we have different designations for the same point, and these names (‘point of intersection of a and b,’ ‘point of intersection of b and c’) likewise indicate the mode of presentation; and hence the statement contains actual knowledge.6

Frege on Sense and Reference 

11

This is a mathematical example, but we can think of other examples, by returning to the evening star and the morning star, that illustrate the point more clearly. The description “the evening star” refers to the same thing as “the morning star,” because those things are just Hesperus and Phosphorous, respectively. There are many instances of this same possibility, where two descriptions pick out the same object. It need not be obvious to people that these descriptions do refer to the same thing. All Frege wants his readers to understand through the example is that two descriptions can refer to the same thing—the intersection of these two lines and the intersection of these other two lines are the same point. The reader would naturally infer at this point that the mode of presentation is connected to perception—it is the mode in which something perceptually appears, such that two different modes of presentation of something correlate with different perceptual appearances. It is natural to assume that two different ways an object is presented to somebody could produce two entirely different appearances of that object to that person. A famous example is of a mountain where someone approaches it from the east and upon seeing it calls it Atlan. The same explorer approaches the very same mountain from the west and calls it Athla. Of course, our explorer eventually discovers that he approached the same mountain twice, but from different perspectives. All of these examples illustrate the same point as Frege’s triangle intersection example. In addition to a name and its bearer, then, Frege has added the mode of presentation of the bearer to somebody who uses the name. This brings in additional machinery—some mode of presentation of both a and b. Let “a” be associated with the mode of presentation MP1 and let “b” be associated with the mode of presentation MP2. Frege is arguing, in effect, that if “a = b” is true the statement tells us truly that MP1 presents the same object as MP2. Here the modes of presentation have replaced the names. So understood, names are words with associated modes of presentation. Now we see the difference between “a = a” and “a = b.” In “a = a” there is only one mode of presentation, MP1, making that statement trivial. In “a = b” there are two modes of presentation, MP1 and MP2, thereby creating a nontrivial statement. It is nontrivial to find out that a single object has these two different modes of presentation. Hence, Frege’s solution to the problem of identity statements is to bring in modes of presentation as the missing ingredient.

12 

Chapter 1

1.4  The Conception of Sense The last sentence from the passage quoted above illustrates Frege’s view of what he calls “actual knowledge.” We have already discussed how actual knowledge is knowledge of the nonlinguistic world. It is not the names as such that are important in this case, but the references of the names and how they can appear or be “presented.” He continues: It is natural, now, to think of there being connected with a sign (name, combination of words, letter), besides that to which the sign refers, which may be called the reference of the sign, also what I should like to call the sense of the sign, wherein the mode of presentation is contained. In our example, accordingly, the reference of the expressions ‘the point of intersection of a and b’ and ‘the point of intersection of b and c’ would be the same, but not their senses. The reference of ‘evening star’ would be the same as that of ‘morning star,’ but not the sense.7

In addition to the term “mode of presentation,” Frege has now introduced another piece of theoretical machinery, the “sense.” He has so far explained the sense as connected to the mode of presentation of the reference. Thus in “a = b” the names “a” and “b” have the same reference but not the same sense. To account for the proposition expressed by a sentence it is not enough to look at the sentence itself or at the reference of the words in the sentence. To account for the proposition expressed by a sentence, another level of semantic reality must be recognized—that of the sense. So, in addition to the reference of an expression in a language, the expression also has a sense. At this point Frege has established to his satisfaction that the meaning of a name cannot be explained purely by its reference. Instead, the name must be assigned a particular mode of presentation of its reference, and the mode of presentation of the reference shows the true definition of the name. Although the name refers to an object in the world, the real meaning of the name comes not from what it refers to but from the mode of presentation. Therefore, Frege has shown us that a theory of language cannot have only reference—it must have sense over and above reference. So far, the word “sense” is merely a label. Frege has introduced this terminology so that there is a mechanism to differentiate the various names, since we have shown that it can be neither the reference nor the names themselves that play this role. The sense, then, accounts for the cognitive differences in names. But what is a sense? Frege uses the phrase “mode of

Frege on Sense and Reference 

13

presentation,” and given his example of the triangle, it is natural to suppose that the mode of presentation is a perceptual or psychological notion. Of course, it is possible to see an object from different angles and perspectives and not realize it is the same object you are seeing. The idea of sense can be generalized beyond what we have talked about with the examples of Hesperus and Phosphorous, or Frege’s own example of the triangle. But in our examples and his it seems as though sense has something to do with perceptual perspective—way of seeing. Notice in the previous passage that Frege is not saying that the sense is identical to the mode of presentation; rather, he says that the sense contains the mode of presentation. Strictly speaking, then, Frege has introduced two extra levels of meaning: sense and mode of presentation, where the former “contains” the latter. Not every expression in language that designates an object would naturally be considered a proper name. A proper name is normally considered an ordinary name, such as “Charles Dickens.” However, Frege also includes other expressions under the heading of proper name that are generally not called proper names. For instance, “the president of the United States in 2012” is said by Frege to be a proper name because it designates a particular person, Barack Obama. Usually, such expressions are called definite descriptions; however, Frege considers definite descriptions to be proper names. Consequently, he thinks that both proper names and definite descriptions have a sense and reference. In chapter 3, we will see that Bertrand Russell argues that definite descriptions are not proper names at all, and that logically proper names are completely different from definite descriptions. In his essay, however, Frege assumes that proper names and definite descriptions are logically the same. Frege’s main point is that every expression in either of these two categories—ordinary proper names and definite descriptions—has both a sense and a reference. Further, it is the sense that contains informative value for identity statements containing those proper names. Frege outlines this idea in the following passage: It is clear from the context that by ‘sign’ and ‘name’ I have here understood any designation representing a proper name, which thus has as its reference a definite object (this word taken in the widest range), but not a concept or a relation, which shall be discussed further in another article. The designation of a single object can also consist of several words or other signs. For brevity, let every such designation be called a proper name. The sense of a proper name is grasped by everybody who is sufficiently

14 

Chapter 1

familiar with the language or totality of designations to which it belongs; but this serves to illuminate only a single aspect of the reference, supposing it to have one. Comprehensive knowledge of the reference is not to be obtained.8

Here Frege attends to the fact that people who understand a language will grasp the senses of the names in that language. Hence the connection between sense and understanding—one who grasps the sense will understand the meaning of the names in the language. A close scrutiny of the paragraph just cited will help us to figure out the exact meaning of the term “sense.” There is a vital clue to the meaning of “sense” when Frege states that the sense is something that “illuminates only a single aspect of the reference.” From this, we can deduce that a sense is akin to a single aspect of an object. For it is natural up until this point for the reader to assume that senses are something like concepts or ideas in people’s minds. However, the above passage illustrates Frege’s rejection of the idea that senses are anything mental. If the sense is an aspect of an object, then it cannot be something in the person’s mind who understands the expression—it is a part of the object, not the individual cognizing it. Another way of interpreting this “aspect of an object” is viewing the sense as a certain property an object has. For example, one of the properties of the moon is that it is arid. Obviously, objects have many different properties, and different expressions can latch on to each one of those properties as distinct from others. The sense, then, consists in latching on to a particular property of the given object. As stated in the above passage, the mode of presentation is an aspect of an object. Those aspects will exist regardless of whether anyone is there to know them, perceive them, or apprehend them; objects have these properties, these aspects, independently of human minds. It is important at this point to note a natural interpretation of sense that is flawed. Take the example of the definite description “the president of the United States.” The reference of this definite description is a certain object with various properties. Each of those properties that the object has is (or corresponds to) a potential sense. In the case of this definite description, one of these properties is an actual sense, because we have an expression in our language that expresses that property—“the president of the United States.” That would seem to be the notion of sense that Frege has expressed so far. However, there is a hole in this seemingly natural interpretation. Since we know that the sense serves to illuminate this single aspect of the

Frege on Sense and Reference 

15

reference, is it correct to suppose that the sense is an aspect of the reference? No, because a thing that illuminates an aspect is not identical to that aspect. There is a distinction between the sense, the illuminator, and the thing illuminated, the aspect. The thing illuminated is an aspect of the object, a property. The sense is not identical to the aspect, though it is closely related to it. The purpose of the sense is to illuminate the aspect; it expresses it or contains it. To say that they are identical would be to ignore a vital point in the above passage. This distinction is a significant one for our purposes—if the sense were identical to the aspect and the aspect is not itself representational, then it follows that the sense is not representational. On the other hand, if the sense illuminates the aspect without being identical to it, then it can be a representational entity. With this interpretation, sense becomes something that represents an aspect of something. It is highly likely that this interpretation of sense is the one Frege was going for—sense is something that represents an aspect of an object. If we are trying to analyze an expression like “the president of the United States,” we thus have four levels to examine: (i) the linguistic expression, (ii) the sense that illuminates the aspect, (iii) the aspect illuminated by the sense, and (iv) the reference, an object. In fact, very strictly, we might identify five levels in Frege’s theory, because there is also the notion of a mode of presentation, which is contained in a sense without being identical to a sense, and which serves to present an aspect of the reference. The name expresses the sense, which contains the mode of presentation, which illuminates the aspect, which is possessed by the object of reference. Several questions arise concerning the possibility of a regress of explanation in trying to understand how reference works. If we think of sense as referring to an aspect, then the idea of referring is presupposed by the theory rather than explained. It matters whether or not we think that the sense represents something because representation is a form of reference. We must give a theory of reference to aspects before we can understand reference to objects. If the relationship between the sense and the aspect is one of representation, we may question whether the relationship of reference here is mediated by a further sense that presents the aspect. If the sense and the aspect were related in representation, it would appear that this relation would cause a regress. There is now something that lies between the sense and the aspect—the mode of presentation of the aspect, that is, an aspect

16 

Chapter 1

of an aspect. The possibility of regress raises an uncomfortable question for Frege: is the sense to be taken as an aspect or something that represents an aspect? Neither possibility appears satisfactory. If it is neither, then what is it exactly? We saw in the previous passage that the expression illuminates a single aspect of the reference but it does not illuminate every aspect of the reference. This is crucial to the whole picture Frege is painting, because a given object can have several aspects and two proper names can latch on to these different aspects. Therefore, when they are put together in an identity statement, the statement becomes informative. If we knew every aspect of every object, we would not gain information with identity statements, because we would already know everything. For example, we would know that the evening star is the morning star. But because we do not know a given object in all of its aspects we are in a position to be informed of something when we are told that a = b. I can know one thing about an object without knowing everything about it. 1.5 Reference We should examine the following passage to aid in the discussion of the relationship between signs, senses, and references: The regular connection between a sign, its sense, and its reference is of such a kind that to the sign there corresponds a definite sense and to that in turn a definite reference, while to a given reference (an object) there does not belong only a single sign. The same sense has different expressions in different languages or even in the same language. To be sure, exceptions to this regular behavior occur. To every expression belonging to a complete totality of signs, there should certainly correspond a definite sense; but natural languages often do not satisfy this condition, and one must be content if the same word has the same sense in the same context. It may perhaps be granted that every grammatically well-formed expression representing a proper name always has a sense. But this is not to say that to the sense there also 9 corresponds a reference.

The relationship as explained above is rather fluid—the very same sense can be expressed by two different signs, as in the case of synonymy. Synonymy can exist within a language or across different languages. For example, English speakers would say “snow” and French speakers would say “neige.” Further, because of ambiguity it is possible to have one sign that corresponds to two different senses—“bank” could mean a bank of a river or a bank for

Frege on Sense and Reference 

17

money. Ordinary proper names, such as “Bob,” in our language have a similar problem of ambiguity since many people have the same name. The same name has many different senses depending on whom or what it names. Concerning reference, Frege believes that a single reference can have many senses corresponding to it and can have many signs corresponding to it. However, there cannot be one sense that corresponds to several different things, since a sense uniquely determines its reference. In Frege’s system, the reference does not determine the sense, because there can be many different senses for the same reference. In contrast, the sense does determine the reference, because the same sense cannot fix two different references. A sense must always have one specific reference to which it corresponds. Therefore, the determination goes from sense to reference but not conversely. Furthermore, there is no determination from the sign to the sense. Although every expression should have a definite sense, it is possible for expressions not to have senses. For example, someone could make up words like “fedneep” that are nonsense—they are signs that lack a sense. However, to make a statement with meaning Frege states that the sign should have a sense: The words ‘the celestial body most distant from the Earth’ have a sense, but it is very doubtful if they also have a reference. The expression ‘the least rapidly convergent series’ has a sense; but it is known to have no reference, since for every given convergent series, another convergent, but less rapidly convergent, series can be found. In grasping a sense, one is not certainly assured of a reference.10

The general point may be lost to the reader since Frege’s examples are rather technical. Only astronomers would understand the former example, and mathematicians the latter. The general idea underlying the examples is that you can form definite descriptions that do not refer to anything. Take the following example of a definite description: “the polka dotted president of the United States.” There has never been a polka dotted President of the United States, so descriptions like those do not refer to anything at all. There is a reason why descriptions such as “the polka dotted president of the United States” must have a sense even though they do not have reference. For us to be able to construct meaningful and true statements such as “the polka dotted president of the United States does not exist,” the definite description itself must be meaningful. This is just one example, but there are infinitely many definite descriptions that have sense and are therefore meaningful, but lack reference. Therefore, it is possible to have

18 

Chapter 1

sense without reference, and to form proper names that have sense but no reference. 1.6  Ordinary and Extraordinary Use Frege applies his discussion of sense, signs, and reference to the ordinary use of words in our language, but not only to this: If words are used in the ordinary way, what one intends to speak of is their reference. It can also happen, however, that one wishes to talk about the words themselves or their sense. This happens, for instance, when the words of another are quoted. One’s own words then first designate words of the other speaker, and only the latter have their usual reference. We then have signs of signs. In writing, the words are in this case enclosed in quotation marks. Accordingly, a word standing between quotation marks must not be taken as having its ordinary reference.11

If words are used in an ordinary way, then in using a word one is intending to speak of the object the word refers to. For example, when someone uses the words “Barack Obama,” he will usually intend to speak of Barack Obama, and therefore Barack Obama is his reference. However, words are not always used in an ordinary way. Therefore, it is not in every case that we are speaking of the reference of a word. It is also possible that one can talk about only the words themselves. Likewise, one can talk about the sense of a word. For example, “the sense of ‘Barack Obama’” refers to the sense of that name, not to its reference. Take caution when parsing these sorts of sentences. For example, if one writes “the sense of Barack Obama” instead of “the sense of ‘Barack Obama’” one has confused the sense of a human being (whatever that may be) in the first case with the sense of a name in the second case. Barack Obama does not have a sense, because he is a person, not a piece of language. Quotations give us a device to prevent us from falling into such a logical error. When writing about the sense of an expression as opposed to the reference of an expression, quotation marks can be used to form the appropriate expression. Therefore, when talking about signs and the sense of signs we must be careful about our use of quotation marks so that what we say makes sense. Further, when reporting what someone else has said, words do not have their usual reference. In this case, the quoted words are signs of signs. Most of the time words are signs of objects, but in the case of quoting the words of another person the quoted words become signs within signs. Therefore,

Frege on Sense and Reference 

19

“‘Barack Obama’” is a sign of a sign. Let us look at two examples to further illustrate these points: (7)  The word man (8)  The word “man” The second example is correctly expressed because the quotation marks show that it is a word that is referred to. In the first example without quotation, “man” refers to a certain species or gender, not to the word itself. In spoken language, we can use such techniques as intonation of voice, body language, or say “quote” and “unquote.” Frege thought that ordinary natural language was quite defective in this way and that it should be clearer when one attempts to talk about words themselves and not what they are about. There are many other places in “On Sense and Reference” where Frege attempts to deal with how words function in normal and abnormal speech. He writes: In order to speak of the sense of an expression ‘A’ one may simply use the phrase ‘the sense of the expression “A”’. In reported speech one talks about the sense, e.g., of another person’s remarks. It is quite clear that in this way of speaking words do not have their customary reference but designate what is usually their sense. In order to have a short expression, we will say: In reported speech, words are used indirectly or have their indirect reference. We distinguish accordingly the customary from the indirect reference of a word; and its customary sense from its indirect sense. The indirect reference of a word is accordingly its customary sense. Such exceptions must always be borne in mind if the mode of connection between sign, sense, and reference in 12 particular cases is to be correctly understood.

Consider someone who says, “John said that Barack Obama is great.” Notice here that “that” has been inserted in the sentence with no quotation marks at all. This example illustrates indirect speech. Someone also could have said, “John said, ‘Barack Obama is great,’” and it would have served much the same purpose. But it may be that John, contrary to the latter statement, is not an English speaker. For example, John could have uttered an Italian sentence, “Barack Obama e meraviglioso” (translation: “Barack Obama is wonderful”). An English speaker would take the Italian words and translate them into an English sentence, thus forming a statement of indirect speech. Frege thinks that in indirect speech the expressions that follow a word like “that” do not have their ordinary reference. Instead, these words refer in that context to their ordinary sense not their ordinary reference.

20 

Chapter 1

To give you a better sense of what Frege has in mind, let us take an example of someone who utters a sentence containing an expression that has no reference. Suppose John says, “The polka dotted president of the United States is great.” In this case, that statement has no reference, and we have reported the sentence in the direct speech form. However, if we put it into the indirect speech form, then we might be taken to suppose that there is such a thing as a polka dotted president, contrary to our intentions. If the definite description were taken to refer to its normal reference, then that part of the sentence would have no reference at all. Furthermore, if the part of the sentence had no reference, something true could not have been said. To avoid these consequences, Frege thinks that we refer instead to the customary sense of the expression and use it abnormally in that particular context. Since the customary sense exists, there is no part of that sentence that lacks a reference. Paraphrasing the idea into explicit form, what is really being said when someone says “John said that Barack Obama is great” is “John said something expressing the proposition that Barack Obama is great.” It is almost as though the individual who utters these words is talking directly about the sense that someone’s words have and not the reference of what he says. When we are reporting what someone said, the interest does not lie in whether or not what the person said was true or really achieved objective reference. Rather, the interest lies in the content of what the person said, and therefore in the sense of the words he used. In this complex sentence, there is no reference to Barack Obama at all. The only thing that is referred to is the sense of the name “Barack Obama.” This solves the potential puzzle of reporting a thing that a speaker says that may not refer to any real object. So there may not be a reference for “the polka dotted president,” but there is a sense of that expression, and that is what matters in reporting the content of what someone said. 1.7  Further Points on Sense and Reference It is wrong to suppose that words can be used only to talk about their customary references. We have seen how it is possible to talk about words, and the sense of words, without talking about the reference of those words. Concerning this point, Frege states the following: The reference and sense of a sign are to be distinguished from the associated idea. If the reference of a sign is an object perceivable by the senses, my idea of it is an in-

Frege on Sense and Reference 

21

ternal image, arising from memories of sense impressions which I have had and acts, both internal and external, which I have performed. Such an idea is often saturated with feeling; the clarity of its separate parts varies and oscillates. The same sense is not always connected, even in the same man, with the same idea. The idea is subjective: one man’s idea is not that of another. There result, as a matter of course, a variety of differences in the ideas associated with the same sense. A painter, a horseman, and a zoologist will probably connect different ideas with the name ‘Bucephalus’. This constitutes an essential distinction between the idea and the sign’s sense, which may be the common property of many and therefore is not a part of a mode of the individual mind. For one can hardly deny that mankind has a common store of thoughts which is transmitted from one generation to another.13

In this passage, Frege sharply distinguishes between ideas present in people’s minds and the sense and reference of words. To reiterate a point made above, Frege does not think that the ideas present in people’s minds have anything essentially to do with sense and reference at all. A psychological idea may be necessary for a human being to grasp a sense, but that does not mean that the sense is the same thing as the idea. First, depending on who you are, a certain word will bring different ideas to mind. For example, an equestrian will have a different idea come to mind when he hears the word “horse” uttered than when a zoologist hears the same word. Frege thinks that the sense of the word “horse” is the same for both of those individuals—the only difference lies in the different mental associations each person has for that word. Furthermore, over time an individual can come to have different emotional associations with the same word. In that case, Frege does not think that the sense changes; rather, the mental associations do. Mental associations can change, but the sense will stay the same. The second reason he gives for making this distinction is that mankind acquires a stock of knowledge, a series of propositions we believe, and we pass those propositions on from generation to generation. Therefore, in a nonpsychological sense, the same thought (or proposition) is transmitted from one generation to another. This process concerns something that transcends the individual persons and the minds that are responsible for the transmitting. For example, consider Isaac Newton in the eighteenth century with various thoughts going through his mind. Suddenly, he states that gravity obeys the inverse square law and writes it in his Principia. After this event, everyone who reads Principia acquires that thought, down the ages, until the present day. Knowing such a thing is different from knowing

22 

Chapter 1

Newton’s subjective, psychological ideas. Hence, when Frege speaks of thoughts he refers to something that is objective and transcends time— a thought is the objective unchanging sense of a sentence. Thoughts, in Frege’s use, are abstract entities. Ideas are not the same as senses; rather, they are things that perish when the mind that has them perishes. Ideas are not really shared by people. Senses, however, are shared by people and do not perish with an individual mind. For Frege, senses have the same objectivity and mental independence as references. The sense of the word “gravity” existed back in Newton’s time and we grasp that same sense now. Therefore, many subjective ideas can correspond to the same objective sense. Frege’s general purpose in arguing for senses to be objective is to show the objective basis for mathematics and science in general. It is important to note that ideas can also be objects of reference. In normal speech, people do not typically talk about ideas. People have ideas all the time, but they do not usually refer to them. For example, if someone says, “It’s raining outside,” she is not saying anything about ideas at all. If she were talking about ideas, she would say something like, “My idea that it is raining outside is well founded.” Just as senses and words can be the objects of reference, so too can ideas be the object of reference. Frege constructs a complete picture for organizing all of these aspects of language by forming a system of levels—words, ideas, senses, and references. He illustrates his leveled system with an analogy: The reference of a proper name is the object itself which we designate by its means; the idea, which we have in that case, is wholly subjective; in between lies the sense, which is indeed no longer subjective like the idea, but is yet not the object itself. The following analogy will perhaps clarify these relationships. Somebody observes the Moon through a telescope. I compare the Moon itself to the reference; it is the object of the observation, mediated by the real image projected by the object on the glass in the interior of the telescope, and by the retinal image of the observer. The former I compare to the sense, the latter is like the idea or experience. The optical image in the telescope is indeed one-sided and dependent upon the standpoint of observation; but it is still objective, inasmuch as it can be used by several observers. At any rate it could be arranged for several to use it simultaneously. But each one would have his own retinal image. On account of the diverse shapes of the observer’s eyes, even a geometrical congruence could hardly be achieved, and an actual coincidence would be out of the question. This analogy might be developed still further, by assuming A’s retinal image made visible to B; or A might also see his own retinal image in a mirror. In this way we might perhaps show how an idea can itself be taken as an

Frege on Sense and Reference 

23

object, but as such is not for the observer what it directly is for the person having the idea. But to pursue this would take us too far afield.14

There is the telescope, the object observed through the telescope, the optical image on the lens of the telescope, and the retinal image on the eye of the observer. The retinal image is also an optical pattern that is projected through the lens of the eye and passes on to the retina. There appear to be three levels: the object out there, the optical image on the lens, and the retinal image. Frege compares the optical image to the sense, and the idea to the retinal image. The retinal image is different for each individual person who looks through the telescope because we all have different retinal structures. However, he thinks that the optical image is the same, even though people observe it with different retinas. Therefore, the sense is an objective thing in the same way that the optical image is an objective thing, and different from the retinal image, which is subjective and depends on an individual’s physiological makeup. 1.8  Problems with Frege’s Theory In an earlier section, we discussed how Frege explains that “a = b” could not state what he had previously held, namely that the name “a” denotes what the name “b” denotes. He argued that his earlier thoughts on this were incorrect because if the sentence says that “a” denotes what “b” denotes then it is about not the objects those names designate but the names themselves. His solution to this problem is to bring in the notion of sense, which contains the mode of presentation of the object. Associated with the name “a” and the name “b” there are particular modes of presentation, and this fact accounts for the informative value of “a = b.” To analyze “a = b” with Frege’s notions of sense and mode of presentation, we can consider a situation where MP1 is associated with the name “a” and MP1 presents what MP2, associated with the name “b,” presents. According to his theory, what makes a sentence such as “a = b” informative is that one mode of presentation presents the same thing as another mode of presentation. Some readers may wonder why the same objection Frege makes against the name theory could not be raised against his own theory. On the surface, the statement “a = b” appears to be about the objects a and b. However, Frege’s theory is focused not on the objects themselves but on the mode

24 

Chapter 1

of presentation of those objects. Common sense would tell us that “a = b” does not seem to be about modes of presentation at all but about objects. For example, few people would think that a statement involving the name “a” (e.g., “a is a planet”) is about a mode of presentation, unless the mode of presentation itself is explicitly under discussion. It is natural to assume that the statement is about an object and that the object is a planet. If names are generally not about modes of presentation, we may wonder how identity statements could be about modes of presentation. The problem is that the subject matter of “a = b” is not the name “a” or the name “b,” nor the mode of presentation of a and the mode of presentation of b, but the objects a and b. At no point are we talking about words or the modes of presentation they allegedly express. Frege raises no objections to himself in regard to this matter. However, the question is a rather uncomfortable one because it exposes a gaping hole in the theory he proposes in “On Sense and Reference.” If “a = b” is only about objects, then he has regressed to his original problem: “a = b” states that an object is identical to itself. Frege solves the problem of informative value, but the way he solves it seems to raise the same kind of objection he has against the names theory, which we discussed at the beginning of this chapter. The only difference between these two is that one theory deals with purely linguistic knowledge and the other deals with knowledge of modes of presentation. Through the latter theory, Frege has shown us that one mode of presentation can correspond to the same object as another, but that does not allow the identity statement “a = b” to be about the actual objects themselves. There is definitely a challenge here that Frege fails to address, considering that his own theory commits him to something objectionable by his own standards. Philosophers have approached this problem differently. In Tractatus Logico-Philosophicus, Ludwig Wittgenstein claims that these sorts of identity statements are ill formed. In natural language, Wittgenstein argues, we can make such statements, but they express trivial propositions and not substantial propositions. Wittgenstein thought that statements like this must be eliminated from an ideal language because they do not make any sense. However, Frege does not make any objection of that sort—instead he attempts to make a seeming triviality into something substantial. Though Wittgenstein’s solution to this problem is to eliminate that sort of sentence from an ideal language altogether, Frege tries to give a theory of it. He never considers Wittgenstein’s more radical eliminative suggestion.

Frege on Sense and Reference 

25

1.9  Extension of Frege’s Theory beyond Singular Terms With an understanding of how sense and reference apply to singular terms, we will now examine how Frege extends this theory of expressions beyond proper names and definite descriptions. In the text, Frege introduces his theory by giving some arguments for his principal tenets. An explanation of his overall theory is helpful before examining the text closely. Singular terms, as we have seen, are subsentential expressions. It is understandable to suppose that if Frege’s theory is applicable to singular terms, the parts of sentences, then it should also be applicable to full sentences. For example, consider the sentence “Hesperus is a planet.” Frege argues that his theory can be extended so that the entire sentence has a sense and reference. One of the odd things about Frege’s theory is that it is clear that singular terms have references, but he must persuade us that they have a sense in addition to reference; but the opposite problem holds for full sentences—we can all agree they have sense, but we have to be persuaded that they have reference. In the case of our example, the sense of the sentence is the nonpsychological thought expressed by it—the proposition that Hesperus is a planet. The claim of reference seems much more difficult for Frege to justify, and he produces a few different arguments for why the whole sentence has a reference. It is clear to the reader at this point what Frege means by the sense of a sentence, but what about the reference of a sentence? First of all, Frege thinks the reference of a sentence is its truth-value. The truth-value, for Frege, is an object. There are only two truth-values, true and false; Frege refers to them with the names “the True” and “the False.” If someone utters a true sentence such as “Hesperus is a planet,” then its truth-value is the True—an object—because it is true. If instead the speaker had said, “Hesperus is a man,” that statement would be false, so the truth-value would be the False. To reiterate, for Frege, all true sentences refer to the truth-value the True and all false sentences refer to the truth-value the False. Here the concept of truth-value has nothing to do with value or ethics. Sometimes, particularly in journalistic writing, “truth-values” have a completely different meaning pertaining to ethics. When Frege refers to truth-values, though, he is not speaking of values in ethics. Frege makes two stipulations concerning the truth-value of a sentence. The first is that a truth-value is the reference of a

26 

Chapter 1

sentence; the second is that the reference of a sentence is an object. Right away, we can see how peculiar both of these claims are. To say that a sentence refers to its truth-value seems to misuse the word “refers.” After all, this word “refers” is the same word he uses for a singular term referring to the object it designates (e.g., “Hesperus” refers to Venus). This type of relation of reference holds between names and objects, but to suppose that sentences refer to anything in the same way names do is to part from what we accept in ordinary language. People normally think that parts of sentences, the singular terms, refer to things, but not whole sentences. For instance, what is the reference of the sentence “Hesperus is a planet”? Naturally, it would seem that the reference of this sentence would have something to do with Venus, since it contains the name “Hesperus.” However, Frege thinks the reference of that sentence is the truth-value the True—an object—since the statement is true. To say that a true sentence refers to the truth-value the True is certainly not the ordinary use of the word “true.” It is logical to assume that a sentence has a truth-value—it is either true or false—but it is still unclear why Frege claims that a sentence has as its reference a truth-value. His second claim, that the truth-value is an object, is just as counterintuitive. In ordinary language, we would not assume the predicate “is true” refers to an object. Frege does not specify a special meaning for the word “object.” He seems to be using the word “object” in an ordinary way, as when it refers to an external thing in the world (e.g., a person, a planet, or a house). His claim that the True is an object too is quite strange. It means that in Frege’s complete list of all the objects in the world, in addition to ordinary objects—every person, planet, elementary particle, and so on—we would also include the True and the False. Hence, Frege considers the True and the False as entities to which one can intelligibly refer. Although these two doctrines seem strange, their purpose from the theoretical point of view is not as puzzling—by introducing these notions Frege can extend the theory of sense and reference to whole sentences. Then, not only will the singular terms have a sense and a reference, but the sentences the terms are a part of will also have a sense and a reference. The sense is the thought the sentence expresses, the reference is the truth-value, and the truth-value is an object. This is nice and neat, to be sure, but it sounds extremely far-fetched. Theoretically, in extending the whole apparatus to sentences, another possibility arises—the extension of sense and reference to complex

Frege on Sense and Reference 

27

sentences. Consider an example where someone says, “Hesperus is a planet and Mars is a planet.” In this case, the truth-value of the sentence depends on the truth-values of each of the component sentences. The application of Frege’s theory to this example would show that the sentence before the conjunction refers to the object the True and the sentence after the conjunction also refers to the object True. Therefore, the truth-value of a conjunction of two sentences both referring to the True would be the True. These examples illustrate how Frege attempted to extend the theory of sense and reference beyond the simplest case, where it seems very plausible, to more complex cases, where it seems less plausible. Now that we have examined more generally the two basic doctrines of Frege’s extension of sense and reference to full sentences, we can begin to look in detail at his arguments in the essay itself. Frege begins his discussion in the following passage: So far, we have considered the sense and reference only of such expressions, words, or signs as proper names. We now inquire concerning the sense and reference for an entire declarative sentence. Such a sentence contains a thought. Is this thought, now, to be regarded as its sense or its reference? Let us assume for the time being that the sentence has reference. If we now replace one word of the sentence by another having the same reference but a different sense, it can have no bearing upon the reference of the sentence. Yet we can see that in such a case the thought changes. For example, the thought in the sentence ‘the morning star is a body illuminated by the Sun’ differs from the thought in the sentence ‘the evening star is a body illuminated by the Sun.’ Anybody who did not know that the evening star is the morning star might hold the one thought to be true and the other false. The thought, accordingly, 15 cannot be the reference of the sentence but must rather be considered as the sense.

Here Frege assumes that the reader will question why sentences should have a reference. If we assume that the sentence has a reference, then it is possible for the sentence to refer to its expressed thought. Whatever the reference of the sentence is must be invariant under substitution of terms in the sentence that have the same reference. Therefore, whatever the reference is must be something that is uniquely determined by the references of each of the terms in the sentence. Take the following example: (9)  Hesperus is F and Phosphorus is F. (“F” stands for any property.) These conjuncts, according to Frege, express two different thoughts, where “Hesperus is F” expresses thought T1 and “Phosphorus is F” expresses thought T2. The question is whether the reference of “Hesperus is F” is T1.

28 

Chapter 1

For Frege, whatever the reference is, it will be preserved when anything is substituted with the same reference for any term in the original sentence— because the reference of the whole is a function of the reference of its parts. Suppose that in the statement above we now substitute the name “Hesperus” for the name “Phosphorus.” Since they have the same reference, swapping the names should be possible without affecting the statement’s truth-value. Of course, the resulting statement is still true, because Hesperus is F and Phosphorus is F. However, the sentence “Phosphorus is F” does not have the same sense as the sentence “Hesperus is F.” Since they do not express the same sense, it follows that they do not express the same thought. If they express two different thoughts, those thoughts cannot be the reference of the sentence. In other words, if the thought were the reference of the sentence, it could not be true that the reference of the sentence depends on the reference of each part of the sentence. Therefore, the thought is not the reference of the sentence. Despite all of our discussion so far, the question remains: why does Frege think that the sentence refers to anything? Why does he think that it refers to a truth-value, and why does he think that the truth-value is an object? The premise of concern in Frege’s argument is based on the example of the sentence “Odysseus is a brave man,” which contains the empty name “Odysseus,” a name with a sense but no reference. Such cases are common for scholars of epic poetry and mythology. In such cases, the thought itself is what is important and not the truth-value. However, if our interest lies in what is true in reality, then we must look at the reference of the sentence “Odysseus is a brave man.” Only by determining the reference is it possible to determine if the object referred to in the sentence, Odysseus, has the particular property attributed to it. Therefore, the truth-value of the thought lies not only in the thought itself but also in what the thought refers to, since the reference determines the truth-value. This premise of Frege’s—that the truth-value of a thought is determined by the references of the parts of the sentence—seems logically sound. He continues in the following passage with an explanation of how to extend this premise to the sentences themselves having references: The thought remains the same whether ‘Odysseus’ has reference or not. The fact that we concern ourselves at all about the reference of a part of the sentence indicates that we generally recognize and expect a reference for the sentence itself.16

Frege on Sense and Reference 

29

Frege does not qualify his statement here, and he makes a huge logical leap. Unless he is able to give a more thorough defense of his reasoning, there is absolutely no reason why sentences should have reference just because their parts do. Furthermore, if our interest is in the truth-value of the sentence, and the truth-value can be known through the parts of a sentence, there is no reason to concern ourselves additionally with the reference of the sentence. If the term in the sentence (e.g., “Odysseus”) refers to something real, that makes the truth-value of the sentence the True, assuming the denoted object has the attribute predicated. Frege gives no explanation for why we should also acknowledge that the sentence itself has a reference, and the passage above is the only place where he tries to defend this position. The sentence may indeed have the property of being true, but it is a further question whether sentences refer to the True. Although this part of Frege’s argument is flawed, he makes two further claims that should be investigated. First, he claims that sentences have truth-value, and then he claims that the reference of a sentence is its truthvalue. He concludes that the reference of a sentence is its truth-value in this passage: We have seen that the reference of a sentence may always be sought, whenever the reference of its components is involved; and that this is the case when and only when we are inquiring after the truth-value. We are therefore driven into accepting the truth value of a sentence as constituting its reference. By the truth value of a sentence I understand the circumstance that it is true or false.17

Frege concludes here that the reference of a sentence must be its truthvalue. The only reasoning behind his conclusion is that the truth-value of a sentence is something that is determined by the reference of its parts. This point can also be made through our earlier examples of substitution arguments. When substituting coreferential singular terms, the truth-value is preserved. The truth-value of “Hesperus is F” remains the True when we substitute “Phosphorus” for “Hesperus.” Therefore, if the reference of the sentence is preserved by substituting coreferential singular terms, then it can be argued that the truth-value is the reference. However, some problems arise from this conclusion. Although something may in fact be preserved under substitution of coreferential terms, there is no reason to call what is preserved the reference of the sentence. Furthermore, in addition to the truth-value, there is something else such a substitution preserves that Frege never considers—what

30 

Chapter 1

we call a fact, a state of affairs that makes a statement true. In this respect, the fact stated by “Hesperus is a planet” would be the same fact as stated by “Phosphorus is a planet,” because facts concern objects and properties, not the words used to describe them. The fact that makes the first statement true is the fact that makes the second statement true—that a certain object has a certain property. When we substitute one coreferential name for another, truth-value is preserved, but so is the fact that makes those statements true. In other words, the state of affairs that corresponds to the sentence is preserved. So why not say this is the reference? In addition to the truth-value, then, the fact is also invariant under substitution of coreferential terms. This proposal is less counterintuitive than Frege’s: in Frege’s view, every true sentence has the same reference and every false sentence has the same reference. However, it is not true that every true sentence corresponds to the same state of affairs. Therefore, the state of affairs is a much more useful concept than truth-values in this case—if sentences have references at all, states of affairs would seem a better choice. If we only suppose that the reference of a sentence is its state of affairs, then our only components are the sense and the state of affairs. There is no need to speak of truth-values as objects of reference. This proposal is much more logically sound than making the odd claim that the sentence refers to its truth-value, and that all true sentences have the same reference. Another way to challenge his argument is to propose that the sentence has no reference at all and only expresses a thought. It is clear why singular terms should have reference, but his reasoning for why thoughts should have reference is absent of any intuitive or argumentative justification. A problem also emerges if we take a closer look at Frege’s proposal that the truth-value of a sentence is an object. Contrary to Frege’s proposal, the truth-value appears to be a property of something that the predicate “is true” ascribes. Why does he think that “is true” is a singular term for an object, the True? In fact, he has to completely deny the way languages are structured when using this concept of truth. Instead of a sentence standing in relation to an object called “the True,” why not just say that being true is a matter of a sentence having the property of being true? Transforming the truth-value from a property to an object is an unnecessary step that Frege takes in attempting to extend his theory of sense and reference to sentences. Sentences are just not like singular terms.

Frege on Sense and Reference 

31

There is still, possibly, one explanation that Frege could produce by drawing on an earlier theory he has about complete expressions, incomplete expressions, and objects. His view is that a complete expression always designates an object whereas an incomplete expression always designates a concept. His notion of object is extremely broad and is whatever is referred to by a complete expression. Singular terms are complete expressions and sentences are complete expressions. It is clear why sentences are complete expressions—they are used to make statements. Why he thinks a singular term is a complete expression is more obscure, because a singular term cannot be used to make a statement. But, since Frege thinks that proper names are complete expressions and complete expressions designate objects, and since he thinks that sentences are complete expressions, he concludes that both must designate objects. He argues they must, because that is what he means by an object—something that is designated by a complete expression. Therefore, the object that a sentence must designate is a truth-value (even though it could have been a state of affairs). The natural objection to this idea is that he uses a completely technical sense of the word “object,” since he claims that an object is to be defined as whatever a complete expression denotes. Of course, it is possible to define “object” that way, but he has shifted the sense of the word “object” from its ordinary sense to his own technical sense. In the same way he stipulated a new meaning for the word “object,” he could have stipulated that whatever is denoted by a complete expression is a dog. Frege could then argue that he has a technical interpretation of the word “dog” such that “dog” means whatever is designated by a complete expression. In doing so, Frege would have completely changed the meaning of the word “dog” and used it to refer to the truth-value, in the same way he used the word “object.” The suspicion is that he has taken over the meaning of the word “object,” which has a well-established meaning and use. Someone can stipulate whatever he likes, but this does not mean he has discovered anything significant, such as that truth-values are objects (or dogs). 1.10  Further Aspects of Frege’s Theory A sentence does not, for Frege, always refer to a truth-value, any more than a singular term always refers to its customary reference, because sometimes it shifts its reference. Remember that if a name is quoted in a sentence, it

32 

Chapter 1

does not refer to its customary reference but to the name itself. In the same way, quoting a sentence would result in a reference to the sentence itself, and not its truth-value. According to Frege, that is not the only case of reference shift, or the most interesting case. Sentences refer to something other than their truth-value when they occur in what are called opaque contexts. Consider this example: “John said that Hesperus is a planet.” Now in this example there is a subsentence: “Hesperus is a planet.” Here Frege thinks we are referring neither to the truth-value of that subsentence nor to Hesperus. In such opaque contexts, “Hesperus is a planet” now refers to the thought John expresses when the sentence occurs outside that context. On the other hand, if it occurs on its own, it expresses its customary sense and refers to a truth-value. But when it occurs in the opaque context the reference shifts. The name “Hesperus” refers now to the sense it normally has, the customary sense, and the whole sentence no longer refers to its truthvalue but to its customary sense, which is a thought. So it is not true that for Frege the sentence always refers to its truth-value (this may make us wonder why he is so convinced that it ever refers to a truth-value). The basis for this shift of reference lies in the fact that when a sentence occurs in this kind of context, its truth or falsity does not matter to the truth or falsity of the whole sentence. For example, if Jane says, “John said that Hesperus is cream cheese,” Jane said something true even though John said something false. Whether what John said is true or not does not matter so far as Jane’s report is concerned, as long as Jane quotes him properly. Since the truth-value of her statement depends only on the accuracy of the quotation, Frege thinks that the truth-value of this opaque context sentence depends solely on the sense of those words. All words then refer to at least two things, according to Frege: ordinary uses of words refer to their ordinary reference, but in opaque contexts they refer to their ordinary sense. Although words in opaque context all have references, we may wonder whether or not they all have distinct senses. The sense of the name “Hesperus” in an ordinary context cannot be the sense of the name “Hesperus” in an opaque context. Otherwise, the sense would be identical to the reference, since the reference now is its ordinary sense. To solve this problem, Frege proposes that there must also be an indirect sense. Now, in addition to every name having two references depending on the context, it also has two senses. The name has its ordinary sense and also the sense it has when it occurs in an opaque context. We can understand why the indirect sense

Frege on Sense and Reference 

33

has to exist, given Frege’s assumptions, but we do not know what it is. Since it is referred to, there must be a sense that refers to it. The sense is a mode of presentation, so the indirect sense is a mode of presentation of a mode of presentation. What kind of creature is that? Another way to explain Frege’s proposal is to consider someone looking at an object from a particular perspective. Frege would introduce the notion of an indirect perspective, a perspective on a perspective. But what is that exactly? It is not possible to have two perspectives on a perspective, because movement (a different position in front of the object) would cause a new perspective. Further, Frege does not tell us what this new perspective-on-aperspective might be. Is it possible to perceive a perceptual perspective from a specific perspective? He explains the ordinary sense well enough with the examples of the triangle and planets, but he never gives an example of the senses that correspond to these words when they occur in opaque contexts. We are left wondering how there can be a mode of presentation of a mode of presentation. At this point the theory is generating consequences that are completely detached from anything that has a clear articulation. If we give Frege the benefit of the doubt, then there must be cases where there is a mode of presentation of a mode of presentation of a mode of presentation (e.g., Jane says, “John said that I said that Hesperus is cream cheese”). There is no explanation as to what that third-order mode of presentation might be. The multiple modes of presentation are all meant to be distinct from each other, but we do not know what they are. Even with these difficulties in his theory, we must not overlook how attractive Frege’s theory is from a theoretical point of view. It has a simple structure, with only a few components. Further, it is a unique semantic theory that had not existed prior to its introduction in his essay. Frege has attempted to build a kind of mathematical theory of meaning, elegant and economical. However, he runs into trouble when he tries to apply the theory to natural language, which is not so streamlined. He tries to squeeze too many disparate things into his mathematically inspired model. But Frege’s contribution to the philosophical understanding of the semantics of language is tremendous. In many ways “On Sense and Reference” was the essay that began the discussion on how to develop a rigorous theory of language. Though some of Frege’s doctrines in this essay are highly questionable, his idea of sense and reference for singular terms influenced philosophers far into the future, and we will often return to it.

2  Kripke on Names

2.1 Background We will now jump forward eight decades. The reason for this is that Frege’s theory of sense for names received its most sustained criticism in 1972, though criticism had been brewing for a while. Thematic continuity thus trumps chronological continuity. In this chapter we will discuss the description theory of names and Saul Kripke’s critique of it in Naming and Necessity.1 Since Frege is widely credited with inventing the description theory of names, Kripke’s critique is directed largely at Frege and those who followed his lead. Frege’s essay “On Sense and Reference” contains a footnote that states the theory that Kripke is criticizing. So here is footnote 4 of that essay: In the case of an actual proper name such as ‘Aristotle’ opinions as to the sense 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 sense to the sentence ‘Aristotle was born in Stagira’ than will a man who takes as the sense of the name: the teacher of Alexander the Great who was born in Stagira. So long as the reference remains the same, such variations of sense may be tolerated, although they are to be avoided in the theoretical structure of a demonstrative science and ought 2 not to occur in a perfect language.

The point Frege makes in this footnote is that when different people speak a language containing a single proper name they can associate different descriptions with that name. Since that is possible, the proper name to which speakers assign those two or more different descriptions is ambiguous. Such ambiguity is a defect of natural language. In a properly constructed scientific language, the same proper name would not be allowed to have two or more different senses by being associated with two or more different descriptions. Still, in ordinary language, people may well assign

36 

Chapter 2

different descriptions to the same name. Frege is here assuming that what people mean by a name is expressed in a definite description, and he is concerned that the descriptions can vary, thus producing unwanted ambiguity. In Naming and Necessity Kripke is interested not in the ambiguity issue but rather in the underlying theory of the meaning of names. He is interested in the theory of names that supposes that the meaning of a name— its sense—is given by a definite description. Frege writes the footnote as though this theory does not warrant a discussion and merely raises the specter of ambiguity in natural languages. Perhaps he regards the description theory as self-evidently true, so in need of no defense. Before we discuss any of Kripke’s critical points, it is important to have a basic understanding of the description theory of names. Take an example of a proper name like “Aristotle.” The name “Aristotle” refers to a long-dead individual. In the present day, someone can say, “Aristotle was a great philosopher,” and refer to that long-dead individual, and there is no ambiguity as to whom he means. There was a certain individual back in Ancient Greece and that very man is the man we refer to when we say “Aristotle” today. Of all the billions of people who have lived we manage to pick out just one of them with the name “Aristotle.” Remarkable! How do we do that? Certainly not just in virtue of the sound the name makes when you utter it. Also, we can make true statements about this man like “Aristotle wrote The Metaphysics.” We refer to a unique individual and then we say something true about him. Thus names permit a remarkable feat of linguistic time travel, homing in on a man who existed over two thousand years ago. The question arises: how can we refer to such a long-dead individual by using a name? We do not see evidence of how in the name itself. The name is just a piece of language—a shape or a sound. It would be impossible to scrutinize the name as it is written or pronounced and somehow deduce the identity of the man to whom it refers. To answer this question, philosophers following Frege have wheeled in the description theory. The description theory uses definite descriptions that can be applied to a certain individual and no one else to enable a speaker to refer to that individual. Aristotle can be referred to with the definite description “the best pupil of Plato.” Definite descriptions enable the speaker or writer to refer to a certain individual by combining a number of different words, such that that combination of words refers only to that particular individual. In addition to “the best pupil of Plato,” other examples of definite descriptions

Kripke on Names 

37

would be “the tallest man in Australia” or “the president of the United States.” The key point here is that the description must refer to one individual and one individual only. There is only one tallest man in Australia, just as there is only one president of the United States and one best pupil of Plato. Such descriptions are uniquely identifying. The definite description “the best pupil of Plato” refers uniquely to Aristotle in virtue of the fact that Aristotle alone fits that description. In other words, he uniquely satisfies the terms in that description. He was a pupil of Plato, and he was the best pupil of Plato, and this definite description expresses those properties. Therefore, when that definite description is used it does not refer to anybody else but Aristotle. Definite descriptions contain a predicate (“is the best pupil of Plato”) and only one object (Aristotle) satisfies that predicate. Initially, it may seem as though the name “Aristotle” is not made up of the terms in the definite description and that the name does not express any of Aristotle’s properties. After all, it does not on its face express any properties that a certain individual had way back in ancient Greece. Therefore it cannot refer in the way the definite description refers, because it does not have the same semantic nature. But according to the description theory, the name “Aristotle” does work in the same was as a definite description. According to this theory, the name is in fact synonymous with the description. The name “Aristotle” is used as a short form of the definite description “the best pupil of Plato” for purely practical reasons. It is inconvenient to continually refer to someone with a lengthy definite description. Instead of repeatedly saying “the best pupil of Plato,” we abbreviate this definite description to a synonymous name, “Aristotle.” We could shorten it even further if we liked (e.g., to the name “Ari”), but it all accomplishes the same purpose—to make it easier to refer to that particular individual. Thus names are just condensed definite descriptions, and their mode of reference is the same as that of descriptions. In other words, the definite description defines the name “Aristotle.” The name “Aristotle” is therefore a disguised form of the definite description. Notice that this theory is a surprising theory because on the face of it the name is not a definite description; this is why it is thought to be a disguised definite description. Now we know that the name “Aristotle” refers to Aristotle because it is short for a definite description of Aristotle. Since the definite description refers to him, the name “Aristotle” also refers

38 

Chapter 2

to him. So, if John says to Jane, “Whom do you mean by ‘Aristotle’?” she could reply, “I mean the best pupil of Plato,” and her statement would be an example of the description theory of names. To understand the description theory, it is important to see how it works and what its commitments are. The first thing to consider is that according to this theory the sense of the name “Aristotle” is expressed by the definite description “the best pupil of Plato,” so that when names differ in their sense, they are short for different definite descriptions. Since the sense of the definite description constitutes the sense of the name, we can apply Frege’s account of the sense of definite descriptions in terms of modes of presentation, as discussed in chapter 1. Therefore, a definite description gives a mode of presentation containing a specific aspect of the reference. Two names with the same reference can express different definite descriptions. The sense is what is understood when a name is spoken or written. In understanding the name “Aristotle” one grasps the sense of the name and therefore the sense of the associated definite description. The theory of descriptions, then, is a theory of what understanding the name consists in, and what one’s grasping the meaning of a name is a grasping of. The theory also tells us what constitutes the information value of the name. Informative identities can be formulated with names, and the associated defining definite descriptions give their information value. In the case of the names “Hesperus” and “Phosphorus,” the descriptions are “the evening star” and “the morning star,” respectively. In our discussion about identity statements using names in chapter 1, we saw that the information value of these two names differs, since the two definite descriptions are not synonymous with one another—one says “evening star” and one says “morning star.” To determine what proposition is expressed by “Hesperus is Phosphorus” we must substitute the descriptions for the names. Since the two descriptions are not synonymous, those types of descriptions differ in their information value; so the names that abbreviate them have a different information value. Further, the theory of descriptions explains what uniquely determines the reference of a name. The definite description refers to only one particular individual. For example, the definite description “the best pupil of Plato” is a unique condition that only Aristotle satisfies. Therefore, the definite description determines the reference of the name. This part of the

Kripke on Names 

39

description theory is in accordance with Frege’s theory of sense and reference, as discussed in chapter 1, since sense is held to determine reference. The sense incorporates the description, the description determines the reference, and so sense determines the reference. For these reasons, when someone utters the name “Aristotle” she is referring to one individual only. The description is what “targets” the name’s reference to a particular individual. Finally, the theory explains how name reference comes to be introduced. When a particular name becomes introduced in a language, it can be introduced through a definite description. By contrast, we can imagine a situation a couple of thousand years ago where a baby is about to be baptized, and the priest asks, “What is the name of the baby I am about to baptize?” The mother says, “Aristotle,” and the priest continues, “Let the baby before us now be called henceforth ‘Aristotle.’” Another example is a definite description that denotes an individual who is not in close proximity to the speaker. For instance, one could say, “I will call the tallest man in Australia by the name ‘Herbert.’” The point is that descriptions can be used to introduce names and bring them into the language. 2.2  Kripke’s Critique The description theory was extremely popular among philosophers for a long time, and the main tenets of the theory remained virtually unchallenged from the time Frege introduced it until Kripke raised objections in 1972. Naming and Necessity contains a series of lectures that generated a considerable amount of controversy because Kripke claimed that the description theory was completely wrong. Moreover, he seemed to prove that it was completely wrong, which was shocking to philosophers, since the theory had been well established for over seventy years. Kripke’s arguments were received with a good deal of surprise by the philosophical community, because the description theory seems like such a natural theory. It just has so much going for it. It is important to note that this theory describes the psychological condition of a person who understands or uses a name. The idea is that if the name is synonymous with a description, then a description must be psychologically present in the mind of the person who utters the name. The theory tells us what it is to know the meaning of a name. We must now turn to Kripke’s critique of the theory, fully aware of its merits and content.

40 

Chapter 2

The description theory says that a name “A” is synonymous with a description “the F.” Now consider the sentence “A is the F.” This sentence will have several properties. First, it will be known to be true a priori. Without any empirical investigation, this sentence can be known to be true simply by understanding the name “A.” If “A” is synonymous with “the F,” all one needs to know is the meaning of the name “A” to know that A is F. Compare “Bachelors are unmarried males”: there is no need to know anything more than what “bachelor” means to know that bachelors are unmarried males. However, if someone said, “Bachelors are unhappy,” that illustrates an example of an a posteriori statement—one that requires research into the empirical world to determine if it is true. The truth of that statement cannot be shown just by virtue of the definition of “bachelor.” According to the description theory, “A = the F” is analytic—true by definition—and a priori, because the description gives the meaning of the name and no more. A further property of “A = the F” is that it must be a necessary truth. If a truth is analytic, it is true in all possible worlds. Given that the two terms are synonyms in that statement, the statement is a necessary truth, just as “A = A” is a necessary truth. It will thus follow that A is F in every possible world, just because “A” means “the F.” Thus, according to the description theory, the proposition expressed by “A is the F” will be a priori, analytic, and necessary. These are straightforward consequences of the description theory. Notice that not every description you couple with the name will have those consequences, because not every description is supposed to be synonymous with the name. Only certain descriptions are synonymous with the name. When someone says, “Aristotle,” he might mean the best pupil of Plato, but he can go on to attribute other properties to Aristotle that are not contained in the meaning of “Aristotle,” for example, having a mole on his left elbow. Some definite descriptions, then, will give rise to statements that are a posteriori, synthetic, and contingent. Clearly, some of the things that are true about Aristotle are true about him only contingently. The main point to understand, though, is that some of the descriptions are true of him analytically and a priori, according to the description theory. Given what the description theory entails, Kripke’s question is the following: is it true that there is a description “the F” such that it generates a proposition expressed by “A is the F” that has these three characteristics?

Kripke on Names 

41

So, is it true that “Aristotle was the best pupil of Plato” is a priori, analytic, and necessary? If this is true, then the description theory is correct; but if not, it is incorrect. Kripke argues that there is no description, or cluster of descriptions, regularly associated with a name that generates these three characteristics. Thus the description theory of names has to be false. Kripke first argues against the necessity of the description. He uses the same example as Frege (“Aristotle”), so we can use our definite description of Aristotle here as well (“the best pupil of Plato”). He attempts to show that the fact that Aristotle was the best pupil of Plato is a contingent truth and not a necessary truth. Of course, it is not being disputed that Aristotle was the best pupil of Plato, because he wrote a number of the formative texts of Western philosophy and is one of the most influential philosophers of all time. In the real world, there is not much debate about Aristotle being the best pupil of Plato. In our world, Aristotle was indeed the best pupil Plato ever had (he always got an A+). However, Kripke asks us to consider alternative realities—possible worlds—where this may not have been the case. There is the actual world, the world we reside in now, where things are a certain way. In this world Aristotle was a philosopher, the sun rises in the east, and a man walked on the moon. Then, there are possible worlds, which are alternatives to the actual world, in which different things are the case. Imagine Aristotle was born in the same year, had the same parents, and lived in the same household. However, in this alternate reality, he has an accident as a child where he banged his head on a Greek sculpture and suffered enough brain damage to prevent him from any further academic pursuits. Although this did not (thankfully!) happen in our world, it could have happened in another world. Such events could contingently happen. If that had been so, Aristotle would not now be called the best pupil of Plato—he wouldn’t have been a philosopher at all. There are less extreme examples of possible worlds in which the Aristotle we know could have turned out differently. If Aristotle had strong musical interests, he could have attended another school that was not Plato’s academy to develop his musical talents. Therefore, Kripke argues, it is quite contingent that he became a philosopher and not something else, such as a harp player. The point is that there are many contingent facts about people that can be expressed in definite descriptions. It is not necessary that we pursue one particular path in life (e.g., a philosopher). We could have easily pursued

42 

Chapter 2

other paths, just as Aristotle could have. These facts are contingent; they are not necessary facts like 2 + 2 = 4 or all bachelors are unmarried males. They could have been otherwise. Since it is only a contingent fact that Aristotle was Plato’s best pupil, the statement “Aristotle was the best pupil of Plato” expresses only a contingent fact and not a necessary fact. But if “A = the F” is not necessary, then the name “A” does not mean the same thing as the description “the F.” Therefore, the description theory is false. This argument of Kripke’s can be called the “modal argument” because it deals with questions of modality, that is, what is necessary and contingent. Frege (and later Russell) thought that when using a name like “Plato” or “Aristotle” we have in mind some famous deeds of the individual denoted. Eventually, the description of those famous deeds becomes synonymous with the name. Kripke’s objection to this proposal is that when a person performs those famous deeds, he has not necessarily performed them. It is conceivable that he might not have performed such deeds, and therefore it is not a necessary truth that he performed those deeds. 2.3  Rigid Designation At this point, Kripke explains his concept of rigid designators and nonrigid designators. To begin, we can first discuss the non-rigid designator. Again, Kripke brings up the idea of possible worlds. Let’s consider the definite description “the most famous pupil of Plato.” In the actual world, it designates Aristotle, but it does not designate him in every possible world. In some possible worlds, Aristotle does not even exist, since it is not true in every possible world that Aristotle’s mother gave birth to him. Therefore, the definite description “the most famous pupil of Plato” is a non-rigid designator, meaning it designates different objects in different possible worlds from what it designates in the actual world. The non-rigid designator itself stays the same when considering every world, but in different worlds it designates different individuals or objects, depending on who does what in that world. A rigid designator, then, is one that designates the same object in every possible world. Kripke argues that, for example, proper names are rigid designators. Before we explain what that means, let us examine a consequence of that for the description theory of names. If it is true that definite

Kripke on Names 

43

descriptions are non-rigid designators, and if it is true that names are rigid designators, then it cannot be true that names are synonymous with definite descriptions, because they are semantically different. If Kripke shows that names are rigid designators and definite descriptions are non-rigid designators, he will have shown the description theory to be false. In other words, he will show that names refer to the same thing in all possible worlds, but definite descriptions refer to different things in different possible worlds. The reason Kripke holds that a name is a rigid designator is that a name refers to one specific individual and only to that individual from world to world. He holds that the name “Aristotle” designates the same person in all possible worlds. Suppose in the actual world the only person with the name “Aristotle” was that particular Greek philosopher. Now could “Aristotle” have denoted anyone other than the actual Aristotle we refer to with that name? That is, could Aristotle be anyone other than Aristotle? Clearly not. Given the meaning of “Aristotle” as it now exists, it cannot denote anyone other than the person it actually denotes. Someone other than Aristotle could have been denoted by “the most famous pupil of Plato,” but no one else could be Aristotle himself. We use the name to pick out a specific individual and this reference stays constant from world to world. It is as if the name gets hold of a certain individual and won’t let him go as we traverse modal space, whereas descriptions allow us to vary our reference as we travel from world to world. Kripke makes this point using a number of different names (e.g., “Moses”), but the same point applies in every case. We can summarize his argument in the following way: if the description that is held to be synonymous with the name is a description that records famous deeds of the bearer of the name, and those famous deeds are contingent properties of the bearer, then they cannot hold with necessity of that individual. Therefore, they cannot be synonymous with that name. To put it differently, the descriptions of famous deeds give rise to non-rigid designators like “the most famous pupil of Plato,” but names are rigid designators—so the latter cannot mean the same as the former. It is important to note a couple of things about the force of this argument so far. The first point is that the argument works only if the description expresses a contingent property of the object denoted. However, questions may arise as to whether or not every description in a language gives only a contingent property of the object. Kripke himself acknowledges

44 

Chapter 2

that descriptions are not always non-rigid designators and that there are instances where they are rigid designators. To illustrate this point, consider “3 is the successor of 2.” This sentence has the same logical form as “A = the F.” The numeral “3” is a name of the number 3, and that number must be identical to the successor of 2—no other number than 3 could be the successor of 2. This statement is a necessarily true one, not a statement of contingent fact. It could not have been the case that in other worlds 3 is the successor of the number 82. Since the successor of 82 is 83, 3 cannot be 83, because it is built into the nature of 3 that it is not 83. Therefore, the definite description “the successor of 2” is a rigid designator for the number 3. There is no possible world in which that description can designate anything other than the number 3. The modal point Kripke makes about the description theory is based on descriptions that designate famous deeds that are rooted in contingency. But what if the description described aspects of the reference that are not merely contingent? In that case Kripke’s modal objection would not apply. If there are properties of human beings that are necessary properties of them in the same way that being the successor of 2 is a necessary property of 3, that would show the theory of descriptions to be less vulnerable than Kripke claims. In some of Kripke’s other work, he discusses something called the necessity of origin. This idea stipulates that the essence of a person comes from the origin he actually has. In other words, there is no possible world where Aristotle existed and came from different parents than the ones from which he actually came. In different possible worlds, even if there were an individual who resembles Aristotle down to the last detail, he would not qualify as being Aristotle unless he had the actual Aristotle’s origins. We can express this essentialist claim in a definite description: “the person with origin O.” Now we can say, “A is (necessarily) the person with origin O,” or “Aristotle is (necessarily) the person who came from parents A and B.” We can agree with Kripke that this statement expresses a necessary truth. In that case, there is no refutation of that version of the description theory on the basis of nonrigidity and contingent properties, because now in every possible world Aristotle satisfies that description: he is necessarily the person with origin O. The modal argument works only if the description is contingent, but not all are. In addition to the necessity of origin, there are different theories about personal identity. One theory is that a person is identical to his brain.

Kripke on Names 

45

Under this theory, if the brain of Aristotle were transplanted into the body of Einstein, the resulting person would be Aristotle. Since Aristotle’s brain carries his identity, it does not matter what body his brain has been transplanted into. Take a person with brain B. If Aristotle is the person with brain B, nobody could be Aristotle without having brain B, and anybody would be Aristotle who has brain B. Therefore, the description “the person with brain B” always designates Aristotle in every possible world, so that that description is necessary or rigid. That description will not result in these modal objections, objections having to do with the contingency of the property expressed. In Naming and Necessity, Kripke never considers these types of rigid descriptions. He does build a convincing argument against the famous deeds version of the description theory, but we have no reason to take the famous deeds theory to constitute the entire scope of the description theory. Even if Frege and Russell were fixated on famous deeds, many other examples of descriptions do report something non-contingent about an individual. We must next consider Kripke’s other objections to see if they overcome this limitation. 2.4  Kripke’s Epistemic Objections One of Kripke’s nonmodal objections has to do with whether or not something is a priori. If a statement is analytic—true by definition—then it must be a priori—knowable without examining the world. If it is not a priori, then it is not analytic. If it is not analytic, then the terms are not synonymous; and if they are not synonymous, then the description theory is false. Kripke gives the example of the physicist Richard Feynman. He supposes that someone knows that Feynman is a physicist but does not understand his specific contribution to physics. Most people are not experts in physics and will not be able to tell you what Feynman’s unique discoveries were but can still say, “Feynman was a famous physicist.” If the same person was asked who Gellman was, he could say, “Gellman was a famous physicist too.” It is evident that with these two descriptions, nothing distinguishes the two physicists from one another—both are simply “a famous physicist.” The person who made those statements does not have sufficient knowledge in his mind to descriptively define Feynman or Gellman. Kripke’s point is that the same information will be associated with the names by our nonexpert

46 

Chapter 2

speaker, but this information is not sufficient to pick out one physicist from the other. Therefore, the descriptive information in the speaker’s mind does not determine the reference of the names—yet the speaker does manage to refer to specific distinct individuals. He does not know any definite description true of his reference, so he certainly doesn’t know any such description a priori. Even though the speaker cannot distinguish between Feynman and Gellman, he is not referring to Gellman when he uses the name “Feynman.” In this case, the speaker does not have the kind of knowledge that the description theory says he should have in order to understand the name. Therefore, he does not know a priori that Feynman is the F for some F that uniquely identifies Feynman. The speaker does not know the description a priori that Feynman is the F, because he does not know that Feynman is the F at all. So it cannot be descriptions in his mind that fix the reference of the name as he uses it. Now consider a case where someone comes along and tells our naïve speaker, “Feynman is the man who originated the parton model.” Our speaker clearly learned something from his informant, contained in the definite description about Feynman. However, as Kripke points out, this knowledge is not a priori. According to the description theory, if a description is synonymous with a name the corresponding statement should be known a priori. But the person who heard that Feynman is the man who originated the parton model knows something empirically about Feynman, not a priori. Kripke’s point is that for any description that a person associates with a name, the description is always known empirically, not analytically. Statements that report such famous deeds are always synthetic, never analytic. The second point Kripke makes is based on the Gödel–Schmidt example. Many people who have heard of Kurt Gödel will know him as the mathematician who proved the incompleteness of arithmetic. Therefore, we can refer to Gödel with the definite description “the mathematician who proved the incompleteness of arithmetic.” Kripke asks us to suppose that Gödel had not proved that theorem at all, but that it was instead proved by an obscure figure named Schmidt. He also asks us to suppose, hypothetically speaking, that Gödel had plagiarized his incompleteness theorem from Schmidt, and Gödel had unjustly received the accolades for devising the proof. In Kripke’s thought experiment, the man referred to when someone says “the mathematician who proved the incompleteness of arithmetic”

Kripke on Names 

47

is Schmidt, not Gödel. In this case, the speaker has a false belief about Gödel—he thinks that Gödel invented the proof, but he did not. His false belief about Gödel cannot then constitute the description that determines the reference of the name “Gödel” whenever he uses it. He refers to Gödel with “Gödel,” while the description refers to Schmidt. Another example of a Gödel–Schmidt type situation that Kripke does not use is the case of perceiving an object. A description theory of seeing would maintain that a description in the mind of the perceiver is what determines which object is seen. Imagine the description here is very closely related to the appearance of what is being seen. The appearance is like the description, and the object and the viewer’s relation to it can be likened to the object being referred to with the name. This description theory tries to analyze the relation of seeing an object. That is, the object being seen is determined by an appearance that is in the viewer’s mind, which translates into a description. The first objection to this theory is that there could be another object in the world that is exactly similar to the one the viewer originally saw. So the perceptual experience of the viewer cannot be the only determinant of the object being seen, since there could be many such objects. The seen object cannot be uniquely fixed by the person’s qualitative experience. Equally, we are familiar with perceptual illusions that mirror the Gödel– Schmidt case. Suppose someone views an object and he experiences a perceptual illusion with respect to that thing. Does that mean he is not really seeing that thing? No; he sees it, but his experience misrepresents it. Nor is it the case that he really sees some remote object that fits his experience better. The lesson is that what determines the object of perception is certainly not the internal nature of the viewer’s experience by itself—this can misrepresent the object. The internal nature of the viewer’s experience plays a role, but it is not the only factor that fixes the perception relation. The object you are seeing is rather the one that actually causes you to have the visual experience. The causal theory of perception proposes that the object being seen is the thing that causes the perceptual experience. The object that best fits one’s experience need not be the cause of the experience. Consider referring with proper names along the lines of our perceptual example. What fixes the object of reference is not merely what is going on in the speaker’s mind in terms of descriptions. Rather, it is an external relationship between the speaker and an object of another kind. This

48 

Chapter 2

relationship could be of the causal kind, as in the perception case. Kripke’s own theory later defends the view that the object of reference is what causes one to use a name, not what best fits the description in the speaker’s mind. This analogy to perception helps articulate the intuitive faults in the description theory raised by the Gödel–Schmidt case and others like it. If the objections Kripke raises through the Feynman and Gödel–Schmidt thought experiments are correct, then it follows that the classic description theory is incorrect. Descriptions in one’s mind cannot determine reference because one might not have any definite description in mind (as in the Feynman case), or the description might not fit the actual reference (as in the Gödel–Schmidt case). Therefore, there is no such description that determines the reference of the name. This concludes Kripke’s argument against the description theory, which includes the modal part and the epistemic part. Though we have already considered some possible counterarguments to the modal part of Kripke’s argument, the epistemic part looks extremely convincing. However, since the description theory solves so many semantic conundrums concerning names, we must ask what alternative theory might to put in its place. 2.5  The Causal Chain Theory If the description theory is incorrect, then the first question we must address is how to solve Frege’s problem of the informative value of identity statements, discussed in chapter 1, which Kripke barely mentions. But he does put forward the chain of communication theory of naming. He argues that we do not refer to something with a name by having a description in our minds that picks out that object. Rather, naming is a much more social, interactive phenomenon than that picture would suggest. Kripke suggests that we must consider these social realities when someone is given a name. We can refer back to our example of Aristotle being baptized. The baby, Aristotle, is given a name, and then people who were present at his baptism begin using his name. Five years later, say, people who have never seen Aristotle may refer to him by name. Then, after decades of interaction with people, Aristotle one day dies, but people still refer to him. Kripke thinks that the reason why people can still talk about Aristotle after his death is that they spoke to people who knew Aristotle, and then picked up their reference from those people.

Kripke on Names 

49

Kripke describes a historical situation in which each speaker is a link in a chain, each intending to refer to the same person with the name “Aristotle” as the previous person in the chain did. Here, the reference is preserved by the intention to refer to the same person as the speaker referred to from whom we originally got the name. This chain continues on and on through the centuries, down until the present time, where any of us can say, “Aristotle is a great philosopher.” So, we can refer back to Aristotle because of this long chain of linguistic connections stretching back to his baptism. Notice that Kripke emphasizes that it is not that a speaker has a description of this chain in mind; rather, being a link in the causal chain makes one refer to that original individual. In other words, when referring to Aristotle one does not need to have a description of him in mind but just to be a link in the right causal chain. This example is somewhat like our example of the perception case, except that it is social. In the case of perception, the objects out there are causing the experiences in the viewer. Similarly, in Kripke’s view, an object out there is causing this long chain of communication that causes one to say the name “Aristotle.” Because of that long causal chain, anyone suitably connected to it can now refer to that person. The description an individual has in mind does not matter in this case; rather, being embedded in this causal chain with other speakers is what matters. These speakers form a long chain going back in time to the point where Aristotle was first called by the name “Aristotle.” This is the alternative picture Kripke paints for us as to how reference works and what determines it. 2.6  Objections to Kripke’s Critique Kripke knows he is not giving a theory of necessary and sufficient conditions, because the causal chain theory faces some prime facie problems. However, he still believes that it paints a better picture of reference than the description theory. He acknowledges the fact that the causal chain could be interrupted at various points. There are many examples of this. Someone along the chain might not intend to refer to the same person, or she might make a mistake with the name, or somehow shift the reference of the name. But the really troublesome issues that arise if we accept Kripke’s theory are the problems about the sense of names, raised by Frege. If Kripke rejects the description theory, then he does not believe that the sense of a name is equivalent to a description. How then does he account for the informative

50 

Chapter 2

value of “Hesperus is Phosphorus”? As an alternative theory, Kripke mentions John Stuart Mill’s view, that the meaning of a name is simply its bearer. However, as we saw in considering Frege’s work, this view cannot handle the case of “a= b,” where “a” and “b” refer to the same object (e.g., “Hesperus” and “Phosphorus”). If the Millian view is true, then “a = b” has the same cognitive content as “a = a”. Frege’s description theory solves that problem; but Kripke, in rejecting the description theory, appears to be left with only the Millian view, which does not adequately explain the sense of a name. It is not as if in rejecting the description theory we can embrace a nice alternative theory, the Millian theory—that just leads straight to Frege’s problem. We are thus left with a nasty dilemma on our hands. Because of these difficulties, a second look at the description theory is warranted to determine if Kripke’s arguments really refute it. We have already covered objections to aspects of Kripke’s modal argument that could resuscitate the description theory. However, Kripke’s epistemic arguments require a different set of considerations. First, we could decide that the description theory is a theory of sense, but not reference. Kripke has refuted the use of the description theory to determine reference with the Gödel–Schmidt example, but we could still suppose that the description constitutes the sense of a name so far as its cognitive content is concerned. On this approach, two names can have two different cognitive values, contained in descriptions, without supposing that the descriptions that constitute the cognitive value also determine the reference of the name. We can think about it just like the case of perception. When one sees an object, there is a cognitive, psychological component of experience and an extrinsic component of an object causing the experience. In the same way, there could be a two-factor structure with names. The descriptions could be considered the cognitive, psychological content of the name, and the causal chain could be what determines reference. According to this solution, we take a two-factor approach to the meaning of names: the reference-determining part, along the lines of Kripke’s theory, and a more psychological part that characterizes what is in a person’s mind when he understands the name. Thus the description constitutes the psychological side of meaning, but the referential side is determined by a Kripkean causal chain. This twofactor approach solves the problems raised by Frege, while still accepting Kripke’s counterexamples. However, we still face the problem of answering Kripke’s epistemic arguments against the description theory.

Kripke on Names 

51

If Kripke’s epistemic arguments refute the description theory in its classical form, it would only be possible to have a description theory that somehow accommodates the force of those arguments. In the Gödel–Schmidt thought experiment, one individual in the linguistic community refers to Gödel by using the name “Gödel,” despite having in mind an incorrect description of the reference. However, Kripke does not mention the fact that certain members of the community do have in mind a uniquely identifying, correct description of Gödel. If language is as social as Kripke takes it to be, then an individual who believes the wrong description of Gödel is connected to other individuals who know correct descriptions of him. Therefore, an individual’s reference is fixed by being part of a linguistic community in which some people associate correct descriptions with the name—though not all do. 2.7  The Social Character of Names Kripke’s epistemic objections deal primarily with descriptions on the level of an individual. But if the description theory is focused on the level of the community instead of the individual, then the objections that applied only to an individual with an incorrect description fall apart. According to the socialized description theory, the reference of a name is fixed by the people who have in their minds the correct description. We thus come to the idea of linguistic deference. The people who are least knowledgeable about the reference of a particular name will defer to those who are most knowledgeable. To illustrate deference and the social description theory, let us consider a historical case similar to the Gödel–Schmidt example, which Kripke also mentions. Giuseppe Peano was an Italian mathematician who axiomatized arithmetic, so there are various axioms that are called “Peano’s axioms.” However, according to authorities, Peano was not in fact the man who invented those axioms. Richard Dedekind, another nineteenth-century mathematician, proposed this collection of axioms, and Peano published a more precise version of them. Peano had cited Dedekind’s work, but some people wrongly attributed the axioms to Peano, and so they became known as Peano’s axioms. Many people in our linguistic community thus have a false belief about Peano. If someone uses the name “Peano” thinking that he satisfies the definite description “the man who axiomatized arithmetic,” that does not mean he is referring to Dedekind with “Peano.” The reason

52 

Chapter 2

is that there are people in the community who know other correct descriptions that apply to Peano, such as “the man who cited Dedekind’s invention of the axioms.” In this way, the description theory can be true for the primary users of the name and the mathematical authorities, the people to whom others defer in using the name “Peano.” The descriptions used by the authorities overrule those of the odd misinformed speaker. The descriptive beliefs of the authorities fix the reference of the name, not those of the uninformed. Another example that illustrates this point involves the scientific terms that are used by nonexperts. Certain terms like “DNA” find their way into popular culture, even though most people have a poor understanding of the terms. Although people use the term “DNA” all the time, few people can refer to DNA with a unique scientific description and truly understand it. However, the people who do not understand “DNA” borrow their reference from people who do have in mind an adequate description. If no one had in mind a correct description of DNA, nobody could refer to it. When a name comes into a language, its reference is fixed by the description that introduces it into that language. Kripke himself does not deny this possibility, because he accepts that names can be introduced by means of a description. The fact that some people do not really know what names mean does not show that those names do not have meaning—as with “DNA.” Kripke’s epistemic arguments do not refute the description theory when the description theory is proposed as a theory of the language of a community. Kripke’s arguments do not refute the description theory as modified to include this social element, though they do refute the individualistic form of the theory. We can say that a definite description determines the reference of the name in a community, because people can defer linguistically. 2.8  Essential Descriptions Given the additions and modifications to the classic description theory, you may be wondering how it is possible to formulate the right kind of description. Consider a person with brain B, such that whoever has that brain is that person. The description “the person with brain B” cannot fail to apply to the person who has that brain. Someone could say, “Aristotle might not have been a famous philosopher,” and that is a true statement because it expresses a contingency; but it is not contingent that Aristotle

Kripke on Names 

53

had a particular brain. He must have that brain in all possible worlds since it is a part of his individual essence. This argument can be made with a variety of personal identity theories. Consider the following descriptions: “the person with soul S,” “the person with consciousness C,” “the person with memories M,” “the person with personality P.” These all express theories about what a person essentially is. So, we can pick whichever personal identity theory best describes the essence of a person, according to our metaphysical views, and express it in a description. For example, if an individual consciousness is indeed the essence of a person, then a description “the person with consciousness C” can be chosen as constituting the meaning of a person’s name. This type of description is not vulnerable to any of Kripke’s modal arguments. In the case of the epistemic arguments, there is always the option to defer to those members of the community who are authorities on the subject—the metaphysicians of personal identity. In our example above, the people who have not met the person with brain B will be able to defer to those who have enjoyed such acquaintance. In summary, we can generate descriptions that determine the reference of the name, provide necessary truths concerning the bearer of the name, give the sense of the name (thus solving Frege’s pressing problem of informative identity statements), and can be accommodated to deal with Kripke’s epistemic objections. The underlying thought is that descriptions refer to objects in the word descriptively, and then names are introduced on their backs as abbreviations of those names—and that is how names refer. The primary way to refer is through descriptions, and names are secondarily based on descriptions. We don’t need a separate account of name reference. There is, however, a further objection to the description theory to consider that Kripke does not bring up at all. 2.9  Impure Descriptions Let us return to our example of the name “Aristotle” and the definite description “the best pupil of Plato.” Notice that this description contains a name, “Plato.” Many of these uniquely identifying descriptions contain such names. But according to the description theory, all names are equivalent to descriptions. What, then, is meant by the name “Plato”? The name “Plato” cannot abbreviate the definite description “the teacher of Aristotle” because that definition would be circular. To refer to Plato, we must create

54 

Chapter 2

a new definite description. We could say, “the most famous philosopher of ancient Greece,” but then the question would arise as to what the name “Greece” means. The point is that the uniquely identifying definite descriptions themselves contain another name. To explain what that name means, the descriptions continue to regress to descriptions containing other names. This issue raises serious problems for the description theory, since names are supposed to depend ultimately on descriptions for their reference. One type of description that can be used embeds a demonstrative, such as “the owner of that dog.” Here we secure reference to the owner by referring demonstratively to her dog. No name is used. So such a description might give the sense of a name without itself containing a name. Demonstratives such as “this” and “that” are very important in language and are often used to provide descriptive reference without the use of names. Without this use of demonstratives, reference by means of descriptions would be crippled. So descriptive reference depends upon and presupposes demonstrative reference. That means that demonstrative reference is basic. It cannot be analyzed in terms of purely descriptive reference. Therefore, demonstratives are not short for demonstrative-free descriptions. We will be considering demonstratives in detail in later chapters; for now we must note that the description theory of names is not applicable to demonstratives. Our conclusion, then, is that though it may be true that proper names are equivalent to descriptions, those descriptions always in the end embed demonstratives. Since demonstratives cannot be explained in terms of descriptions, reference is not fundamentally descriptive. Even if the description theory is true of names, this does not show that the way we basically refer to things in the world is through descriptions. The basic way we refer to the world is by means of demonstratives, which are not equivalent to descriptions. The victory of the description theory over Kripke’s attack is therefore a Pyrrhic one. In the end, we must accept that some referential terms function nondescriptively.

3  Russell on Definite Descriptions

3.1  Indefinite and Definite Descriptions In the previous chapter we considered the description theory of names, but we didn’t say much about the analysis of descriptions themselves. Frege treats definite descriptions as belonging to the same category as proper names—they are “singular terms,” whose function is to denote a particular object for the rest the sentence to comment on. They have both sense and reference. Russell, however, disagrees: he denies that definite descriptions are singular terms, analogous to proper names. He thinks they belong to a quite separate semantic category. In particular, he denies that they have reference. He thus believes that their surface grammatical form is misleading. In this chapter we will see why he says these things. In the text we will be discussing—a chapter from Russell’s Introduction to Mathematical Philosophy (written while he was in prison during the First World War for treason)—Russell builds up to his theory of definite descriptions by first considering indefinite descriptions. Once he establishes the right logical analysis of indefinite descriptions, his analysis of definite descriptions comes out as a simple addition. Though he does not use this terminology, his essential thesis is that definite descriptions are quantifiers (if you are not familiar with this concept already, it will be explained as we proceed). His first example in the text is the sentence “I met a man.” An indefinite description is one formed with the indefinite article “a,” whereas a definite description is one formed with the definite article “the.” His famous example of a definite description is “the king of France”; an indefinite description would be “a king of France.” The sentence “I met a man,” then, is formed using the indefinite description “a man” attached to the verb “met” and the indexical singular term “I” (indexical terms are

56 

Chapter 3

discussed in later chapters). Another example of a sentence using the same indefinite description is “Socrates is a man.” Frege believed that an expression of the form “the F” is a proper name that functions as the subject of a subject-predicate sentence. It is possible to substitute an indefinite description in its place and preserve grammaticality. This makes it natural to suppose that “an F” is also a proper name that constitutes the subject of a sentence. Russell addresses himself to the question of whether “a man” in “I met a man” is a proper name. In the following passage, Russell wonders if in the sentence “I met a man,” “a man” refers to Jones: Our question is: What do I really assert when I assert “I met a man”? Let us assume, for the moment, that my assertion is true, and that in fact I met Jones. It is clear that what I assert is not “I met Jones.” I may say “I met a man, but it was not Jones”; in that case, though I lie, I do not contradict myself, as I should do if when I say I met a man I really mean that I met Jones. It is clear also that the person to whom I am speaking can understand what I say, even if he is a foreigner and has never heard of Jones.1

Russell here makes a simple objection to “I met a man” being synonymous with “I met Jones”: suppose I met Jones, but I lie and say, “I met a man who was not Jones.” Or maybe I forgot I met Jones and do not lie, but just say something false. Regardless of my motivation, though I make a false statement, it is not the case that I am contradicting myself. If “I met a man” meant the same thing as “I met Jones,” then I would be saying “I met Jones but I did not meet Jones.” This would be a very poor way of lying. Russell rightly claims that I am not contradicting myself when I say, “I met a man but it was not Jones,” even if I did meet Jones. So it cannot be that “a man” means the same thing as “Jones” in this sentence, even though Jones was the man I met. The meaning of “a man” cannot be given by the meaning of a name for the man I met. This is Russell’s first proof to show that an indefinite description is not a name of an individual. The relation between “a man” and “Jones” cannot be a synonymy relation, or else I would be contradicting myself when I said, “I met a man who was not Jones.” Looking at the matter grammatically, one would not suppose that “a man” is a proper name, since grammatically it is quite a different expression from “Jones.” However, when thinking in terms of reference, it would be natural to think this way about how to determine the truth conditions of the sentence. For example, for the sentence to be true, there has to be a

Russell on Definite Descriptions 

57

meeting relation between someone referred to as “I” and someone referred to as “a man.” This statement would express a relational proposition relating me to the person I met. It should have the form “a R b”—but if that is true, and “a” and “b” are names, then contrary to appearances, “a man” should be a name. Thus we might suppose that logically “a man” is a name, though grammatically it clearly is not. But Russell thinks that this reasoning is incorrect; otherwise, as he says, the statement “I met a man but it was not Jones” would be a contradiction, on the assumption that I met Jones. The second point Russell makes is to the same end. Consider the sentence “I met a unicorn.” If we thought that indefinite descriptions were names, then there must be something that the name names in order to make the name meaningful. In this case, there are no unicorns to name, so the phrase “a unicorn” cannot function in that sentence as a name of something, or else it would be meaningless instead of merely false. In the previous example (“I met a man”) there was an actual man being met who could possibly be the bearer the name. With the unicorn example, nothing in reality can bear that name, so it would have to be a meaningless sentence. You could never meet a unicorn, because there aren’t any unicorns to meet. Russell’s point here is that if “a unicorn” were a name of something then the name could be meaningful only if something were named. Since nothing is named, it would lack meaning; but it does not lack meaning. The only way the sentence can be false is if it is meaningful. Therefore, it cannot be that “a unicorn” is a name of something. The thing that enters into the proposition expressed by these words is not an object named. Instead, it is the concept of a unicorn that is the constituent of the proposition expressed by the sentence “I met a unicorn.” With respect to the “I,” what enters into that proposition is not a concept but an object, because I am not a concept. However, sentences like “I met a unicorn” or “I met a man” bring the concept of a unicorn or a man into the proposition, not an actual man or unicorn. According to Russell, then, in the example “I met a man,” “a man” refers to a general concept, not to a particular man. Russell uses the term propositional function to describe what is left in a proposition when part of it is deleted. If I say, “I met Jones,” this is an ordinary proposition whose constituents are me and Jones. However, if we delete the name and put in its place the letter “x,” then the letter “x” does not refer to any individual at all. It is a placeholder that indicates a part of the sentence has been deleted and left blank. The phrase “x is a

58 

Chapter 3

man” is called a propositional function, because when something specific is added to replace “x” (usually called a variable) the entire sentence expresses a proposition. In essence, it is the abstract form of a proposition, without being a particular determinate proposition. In ordinary logic, “x” here would be referred to as a free variable. The phrase with “x” will not become a proposition until a name is plugged in to replace the variable. Propositional functions can be simple or complex. Russell discusses the sentence “I met x and x is human,” and takes it to mean “I met someone or something and that someone or something is human,” or, more simply, “I met something and it is human.” He explains that such a propositional function is “sometimes true” if a proper name is inserted to replace “x.” He suggests that we replace the relational form (“a R b”) with the form of this propositional function (“I met x”). Thus the propositional function “I met x” is said to have an instance in which the resulting sentence is true. If I met Jones and plug “Jones” in to the propositional function, the sentence would be true. When someone says, “I met a man,” he is not really talking about a particular man, according to Russell. Instead, Russell thinks that when someone says, “I met a man,” he is talking about a propositional function and saying that it has an instance—though he may not know what this instance is. It is important to note that any name could be plugged into this propositional function. As long as the name refers to a real person, the function has an instance, and is therefore true. Therefore, there are two relations that Jones can have to a proposition to make it true. One is that Jones can be named by a name in that proposition. But in the other relation, Jones can be an instance of a propositional function—without being named by it. To put it differently, Jones can either be explicitly named or he can fall under a general term or predicate like “man I met.” Falling under a predicate is not the same kind of relation as being named. If I say, “Everyone in this room is a philosopher,” I have not named anyone, even though several people fall under the predicate “person in this room who is a philosopher.” If we put it in more contemporary terms, Russell’s view is that indefinite descriptions are quantifiers. Now we realize that quantifiers and names are semantically not at all the same. For example, take the quantifier phrase “no one”: that cannot be a name of someone! If it were, “no one is over ten feet tall” would entail “someone is over ten feet tall.” But neither is “someone” a name for a person—for if so, who? Even if someone is out there making

Russell on Definite Descriptions 

59

true what the speaker is saying when he says, “someone stole my bike,” the speaker is not naming that villain; if he were, he’d know who did it. All of this has to do with the revolution in traditional logic that stretched all the way back to Aristotle. In the old days, everything was just terms and predicates. Russell rejects this traditional logic, just as Frege argued that quantifier expressions (“something,” “everything,” etc.) should not be assimilated to names. Frege’s position is that a quantifier word is a “secondlevel concept.” He thought that these words were neither names of objects nor concept expressions like “is a man.” A second-level concept applies to a first-level concept. When a person says, “Someone is a man,” the quantifier word is like a second-order propositional function: it is a comment about the first-level concept expressed by “man.” If a person says, “Jack is a man,” then he is speaking of Jack and stating that he is a man. But if someone says, “Someone is a man,” he is now talking about a propositional function, stating that it has an instance. He is saying this: “The first-level concept expressed by ‘is a man’ has at least one instance.” In Russell’s example of “I met a man,” the correct analysis is this: “the propositional function ‘I met x and x is human’ has at least one instance.” In this there is no mention of Jones by name, even if he is the instance in question. This has a bearing on statements about existence. When an atheist says, “God does not exist,” what he is really saying is “The propositional function ‘x is a god’ has no instance.” He is not saying about some individual named “God” that he does not exist—that would be self-defeating. Russell argues that a person cannot make a true negative existence statement about a named individual because he was never talking about any individual in the first place. Instead, the speaker was really talking about a propositional function and asserting that it has no instances. By paraphrasing the statement into a statement about a propositional function, we are not misled into believing that terms like “a man” or “someone” or “no one” are somehow functioning like names that require a reference. The only thing referred to with a propositional function is a concept, about which we state that it has, or lacks, instances. The point that Russell is ultimately building up to is that a definite description is also a quantifier, not a name. In adopting this approach, Russell thereby resolves many puzzles that arise with definite descriptions, particularly when they are empty. Russell had previously held Alexius Meinong’s view. This is the view that in addition to the ordinary objects that exist, there are things that subsist, or

60 

Chapter 3

have a peculiar quasi existence. Things that people normally do not think exist, such as unicorns and golden mountains, have this quality of subsistence. Because of this subsistent category, Meinong thinks that expressions like “the golden mountain” do refer to things, and because they have reference they also have sense. This view is in contrast to Frege’s view that such terms have sense but not reference. With Meinong’s view, “the golden mountain” is meaningful because it refers to the golden mountain, which is a subsistent thing. In his system, such expressions can be endowed with reference, so long as we accept this expanded ontology of subsistent entities. Russell now avoids this view by developing a theory of descriptions that does not postulate any Meinongian ontology in order to give meaning to empty definite descriptions. Russell believes that such phrases do not denote anything, even when they have an existent correlate. It is the same point he makes about the phrase “a man”—the definite description is not a phrase that functions like a name at all. Cases where there is no entity to denote (e.g., “the golden mountain”) do not require an extra ontology like Meinong’s. Rather, we say that the expression is not a denoting phrase to start with, but something completely different from that, just as “a man” is not a denoting phrase. Russell argues that definite descriptions also express propositional functions that do not refer to or denote or name objects. As Frege would put it, they function as quantifiers. Therefore, since quantifiers are different from names, definite descriptions are different from names. Russell’s new theory is developed against the background of Meinong’s theory, which is also a version of Frege’s theory in assuming that definite descriptions function as proper names. 3.2  Three Theories of Definite Descriptions Before continuing with a thorough analysis of Russell’s theory, it is important to note that Russell does not follow the rules on when things should be quoted or not. Indeed, Russell is notorious for his misuse of quotations. We should be more careful. There are three theories about definite descriptions relevant to Russell’s “Definite Descriptions.” We can use Russell’s first example, “the king of France,” to explain these three theories. The description “the king of France” is an empty description—one with no reference—because at the time Russell used the example, France had no king. Although this description is

Russell on Definite Descriptions 

61

empty, it is just as meaningful a description as “the queen of England,” though that description does have reference. The fact that there are meaningful empty descriptions refutes the idea that the meaning of a definite description is identical to its reference. If reference and meaning were identical, then our first example would have no meaning. Frege’s theory is consonant with this fact, because it allows that such expressions have sense but no reference. Of course, the sense is where the meaningfulness lies. As far as we can tell from Frege, he believes that every meaningful expression has a sense, and there are no expressions whose meaning is simply their reference. Every expression that exists in natural language is something whose meaning consists in its sense, where the sense is independent of the reference. Russell, in his discussion, never takes into account Frege’s view. Some readers could potentially be confused reading this excerpt alone, because Russell is constantly making assertions that contradict Frege’s theory. Russell presupposes that Frege’s theory is wrong without overtly stating his rejection of the theory of sense and reference. Instead, Russell holds a referential theory of meaning, where he believes that the meaning of an expression must be its reference. Meinong’s view is that “the king of France” has a reference to a peculiar, subsistent entity. Its reference is not something that exists in the same way that the reference of “Queen Elizabeth II” exists. In Meinong’s ontology, the world is divided into existent things and nonexistent things, and nonexistent things also have a kind of being. Given his distinction between existence and subsistence, it is possible for Meinong to argue that “the king of France” refers to a subsistent being. By considering fictional characters, Meinong’s view becomes easier to understand. According to him, the name “Hamlet” refers not to any existent Prince of Denmark but to a fictional character. In Meinong’s theory, such fictional characters have being without existence—subsistence. Therefore, a name like “Hamlet” refers to a subsistent entity. With this theory, a referential theory of meaning can be maintained, without adopting Frege’s distinction between sense and reference. If an expression is meaningful because of its reference, we have no need to bring in sense to establish meaning, because we always have subsistent references when existent references are lacking. According to Russell, every proper name or singular expression has a meaning determined by its reference. He does not accept a two-level theory of reference and sense; he thinks he can get by with reference alone.

62 

Chapter 3

Contrary to appearances, he argues, a definite description is not a singular term at all and does not denote an object. Frege thinks that empty descriptions like “the king of France” have no reference but that such expressions are meaningful because they have a sense. Meinong thinks they refer to subsistent entities and are meaningful that way. Russell thinks they are not referential expressions, so their emptiness isn’t a problem. As mentioned above, Russell was a Meinongian in his earlier years. But since he liberated himself from trying to find a reference for empty descriptions, he does not have to reconcile himself to accepting shady subsistent entities. He thinks that ordinary language is logically misleading, because it makes definite descriptions occupy the place of names. For example, in ordinary language, the sentences “The king of France is bald” and “Bertrand Russell is bald” are both subject-predicate sentences. The first one has a definite description as the subject while the second has a name for the subject. Ordinary language makes it seem as though definite descriptions function as proper names, even though logically they do not. Quantifier expressions also illustrate this point. The sentence “Someone is bald” appears to express a subject-predicate proposition in the same way that “Bertrand Russell is bald” does. These two expressions look grammatically and syntactically the same. However, it would be very strange to think that “someone” is a name (“Someone, come here!”). Consider the claim that “someone” denotes Jones in the sentence “Someone is bald,” where Jones is in fact bald. But “someone” cannot be the name of Jones, because the statement “Someone is bald but it’s not Jones” is not a contradiction, even though Jones may be the only bald person. The appearance of subjectpredicate status for “Someone is bald” has to be misleading. At the same time, it is not plausible to think that “someone” refers to some shadowy, ideal, possible bald individual, as Meinong supposes. Russell argues that terms like “someone” are logically not singular terms. Part of his purpose is to explain what their logical role is. Since we have seen that these sorts of terms are not referring expressions at all, their meaning cannot be constituted by reference. Because of the defectiveness of ordinary language, these sorts of statements are misinterpreted as having subjectpredicate form. However, the fact that such terms lack a singular reference does not mean that they lack meaning. Frege and Meinong have their own explanations as to why such terms as “the king of France” lack an existent reference but have meaning. Frege

Russell on Definite Descriptions 

63

uses his sense–reference distinction, and Meinong postulates an existence– subsistence distinction. Russell rejects both of those ideas. He thinks that every expression that is referential has a meaning that is determined by its reference, but these sorts of expressions are not referential at all. However, Russell accepts that these sorts of expressions appear to be referential, owing to the deceptiveness of natural language. This point about the deficiencies of natural language was very important to Russell, because it showed that ordinary language can be logically misleading and bears on the question of constructing an ideal logical language. In Principia Mathematica, Russell and Alfred North Whitehead formed an ideal language that is essentially the same as predicate logic. The formation of this logical language led to the idea that natural language was adequate for practical purposes but deficient for logical ones. This view was the standard one for a long time and shaped philosophy for the first half of the twentieth century—until Ludwig Wittgenstein came along and argued against this view, which he had also held in his Tractatus Logico-Philosophicus. So this issue about descriptions had wide philosophical implications. It is important to understand the context within which Russell produced this work. Much of the correct methodology in twentieth-century philosophy and expectations about language hung on the theory of descriptions, in addition to its contributions to pure logic. Russell’s theory practically shaped the whole of twentieth-century analytical philosophy. The resulting dialogue of twentieth-century philosophy revolved around whether philosophers agreed with it or were against it. So, the theory was of massive importance at the time Russell developed it. 3.3  Indefinite Descriptions and Identity Russell’s position is that statements containing descriptions like “a man” must be paraphrased to reveal their meaning. This will involve changing their form quite dramatically, and also introducing logical symbols. To paraphrase such statements, he uses propositional functions to take particular expressions out of a sentence and substitute a variable “x.” In this case, he would insert an “x” into “a man,” creating the propositional function “I met x and x is human.” This propositional function is then said to have at least one instance, meaning that it applies to at least one thing in the world. Jones is the instance out of all those things in the world that might

64 

Chapter 3

make that propositional function true. Although the sentence appears to refer to a certain individual in the world with the expression “a man,” the original sentence’s form is logically misleading. For Russell, what the sentence really says is that the particular propositional function has at least one instance. He uses this apparatus of explanation to make it philosophically clear that this sentence is about a propositional function. Today we routinely use quantifiers to express Russell’s point. Take, for example, the long logical form: (1)  There is an x such that I met x and x is human. The same propositional function can have several variations. It could also be read existentially: (2)  There exists an x such that I met x and x is human. Different theories of the quantifiers correspond to the ways in which such a statement is read. A useful thing to remember about interpreting existential quantifiers is that the variable “x” can be replaced by a name. After such a substitution, there will be at least one instance where the substitution will make the statement true. In our particular case, “Jones” would make the statement true. Such an analysis is often called the “substitutional interpretation” of the existential quantifier, because a particular substitution into the open sentence expressing a propositional function makes the resulting sentence true. Russell tends to assume the substitutional interpretation. The best way to understand his interpretation is with the sentence “I met something and that something is human.” The only term in this sentence that refers to an individual is “I.” The phrase “a man” becomes a part of the existential quantifier. Then, there is a conjunction of the predicates giving us the assertion about my meeting a human. The only things that are brought in by the quantifier phrase are concepts. To better illustrate this point, we can use a statement involving a nonexistent entity: “I met a unicorn.” Since there are no unicorns, I could not have met a unicorn. However, when using Russell’s apparatus to analyze this sentence, we see that the proposition contains only me and the property of being a unicorn. The sentence is actually saying (falsely) that there is an instance of that property and that I met that instance. In this form, no unicorn is named. The advantage of Russell’s theory is that we can explain how to speak of nonexistent things without creating an entirely new ontology. In Meinong’s view, we need subsistent golden mountains to analyze “I climbed

Russell on Definite Descriptions 

65

the golden mountain.” Russell avoids creating an entirely new ontology of subsistent things, because he thinks that the statement is really about a propositional function. Russell argues that genuine names that are empty are indeed meaningless, but “the golden mountain” is not a genuine name. Russell just assumes that Frege is wrong, because he assumes that the meaning of a name comes from its reference if it really is a name. In contrast to Frege, Russell also sharply distinguishes between names and descriptions. He thinks that descriptions, definite and indefinite, do not function in the way that names do. Russell includes a few paragraphs on the important distinction between the “is” of predication and the “is” of identity, which we shall briefly pause to explicate. Although these points are not really essential to his main line of argument, they have major significance in analytic philosophy. He says there are two kinds of “is”: the “is” of identity and the “is” of predication. The “is” of identity is used in sentences that could be paraphrased as “a = b,” like “Hesperus is Phosphorous.” Russell points out that we do not always use “is” in the sense of identity. Take the statement “This table is brown.” The table has the color brown, but the identity of the table is not brown. There are a great many things in the world that are brown and not just this table. It would be absurd to claim that this table is identical to the color brown! According to Russell, the “is” that is present in “this table is brown” is the “is” of predication. The “is” used in the sentence “Socrates is human” is thus entirely different from the “is” used in the sentence “Socrates is a man.” The former is the “is” of predication, and the latter is the “is” of identity. He gives us the following paraphrase of the sentence with the identity “is”: (3)  There is an x such that Socrates is identical to x and x is human. His general point is that we must be aware of the two different forms of “is” in language. Also, the ambiguity of “is” adds further evidence to his point that ordinary language is logically misleading, because this one word— “is”—is used in both statements of predication and statements of identity. Russell believes that an ideal language would not have such ambiguities. 3.4  Russell’s Rejection of Meinong’s Ontology Russell’s stalwart rejection of the Meinongian ontology can be found in the following impassioned passage:

66 

Chapter 3

For want of the apparatus of propositional functions, many logicians have been driven to the conclusion that there are unreal objects. It is argued, e.g., by Meinong, that we can speak about “the golden mountain,” “the round square,” and so on; we can make true propositions of which these are the subjects; hence they must have some kind of logical being, since otherwise the propositions in which they occur would be meaningless. In such theories, it seems to me, there is a failure of that feeling for reality which ought to be preserved even in the most abstract studies. Logic, I should maintain, must no more admit a unicorn than zoology can; for logic is concerned with the real world just as truly as zoology, though with its more abstract and general features. To say that unicorns have an existence in heraldry, or in literature, or in imagination, is a most pitiful and paltry evasion. What exists in heraldry is not an animal, made of flesh and blood, moving and breathing of its own initiative. What exists is a picture, or a description of words. Similarly, to maintain that Hamlet, for example, exists in his own world, namely, in the world of Shakespeare’s imagination, just as truly as (say) Napoleon existed in the ordinary world, is to say something deliberately confusing, or else confused to a degree which is scarcely credible. There is only one world, the “real” world: Shakespeare’s imagination is part of it, and the thoughts that he had in writing Hamlet are real. So are the thoughts that we have in reading the play. But it is of the very essence of fiction that only thoughts, feelings, etc., in Shakespeare and his readers are real, and that there is not, in addition to them, an objective Hamlet. When you have taken account of all the feelings roused by Napoleon in writers and readers of history, you have not touched the actual man; but in the case of Hamlet you have come to the end of him. If no one thought about Hamlet, there would be nothing left of him; if no one had thought about Napoleon, he would have soon seen it that someone did. The sense of reality is vital in logic, and whoever juggles with it by pretending that Hamlet has another kind of reality is doing a disservice to thought. A robust sense of reality is very necessary in framing a correct analysis of propositions about unicorns, golden mountains, round squares, and other such pseudo-objects.2

We can clearly see the force of Russell’s point here. To say that Hamlet is an existent in Shakespeare’s imagination or our own imaginations is a confused way of speaking. Hamlet, Russell argues, does not have the same existence in our imaginations as you have as you are reading the text. At most, the sentence “Hamlet has an existence in Shakespeare’s imagination” can mean that Shakespeare invented the fictional character of Hamlet. The sentence does not mean that we can go to a place called “Imagination” and find Hamlet skulking there, existing like one of us does in reality. Herein lies another misleading aspect of ordinary language. The sentence “There exists a dog in the next room” allows the listener or reader to understand its meaning—she will see a dog in the next room if she goes into that room. However, the sentence “There exists a dog in my imagination” makes it

Russell on Definite Descriptions 

67

seem as though imagination is a place to which people can travel and, upon arriving there, we will find a dog, barking and wagging its tail. This notion, Russell argues, is ludicrous; a dog or a unicorn does not exist in one’s imagination in the same way that a horse exists in a paddock. As to whether the passage above refutes Meinong’s position, we cannot yet say. Meinong never says that phrases like “the golden mountain” refer to things that have an existence. His whole argument is based on the thesis that they have only subsistence. Meinong never states that things exist in the imagination in the way normal people exist in towns and cities. Really, Russell is arguing against what he thinks Meinong is proposing, not what Meinong is actually proposing. However, for the sake of understanding Russell’s theory, we will assume that he is correct about how we should deal with definite descriptions that refer to these nonexistent entities—that is, they have no reference at all. 3.5  The Details of Russell’s Theory of Descriptions The theory of descriptions is now quite simple. An indefinite description like “a man” is equivalent to the existential quantifier. The reader may now be wondering how Russell distinguishes a definite description from an indefinite one. Suppose we start with the indefinite description in “A present king of France is lucky.” We could paraphrase that sentence in the following way: “There exists someone x such that x is a present king of France and x is lucky.” Russell then asks us to consider a case where the sentence has “the king of France” as a component. The difference lies in whether uniqueness is implied. In the sentence “I met a man” the speaker of the sentence does not logically imply that he met just one man. Such descriptions with “a” can apply to many men. On the other hand, a definite description with “the” (e.g., “the king of France) can only apply to one individual if it applies to any. Therefore, uniqueness is what is added when “a” is replaced by “the.” Russell thus argues that definite descriptions should be analyzed in basically the same way that indefinite descriptions are analyzed; the only difference in the analysis of definite descriptions is that uniqueness is added. Keeping these considerations in mind, we will first examine an analysis of an indefinite description; then we will examine an analysis of a definite description. So consider “An F is G” and “The F is G.” The former is true if and only if at least one thing is both F and G. The

68 

Chapter 3

latter is true if and only if at least one thing is F and that thing is G and at most one thing is F and that thing is G. Both imply existence, expressed by “at least,” but only the second implies uniqueness, expressed by “at most.” If we analyze the sentence “The queen of England is happy,” we would say that there is a queen of England, and that there is only one queen of England, and that she is happy. There are three conjuncts in this analysis of “The F is G”: (1) there exists something that is F, (2) there is only one thing that is F, and (3) that thing is G. Therefore, if you uttered the sentence “The king of France is bald,” you would be saying that there exists something that is a king of France and that there is at most one thing that is a king of France and that thing is bald. That is Russell’s general form for the analysis of the statement “The F is G.” His theory is fairly straightforward. The basic idea is that the word “the” means existence and uniqueness. Existence means at least one, but uniqueness means at most one, and then the particular predication (“is bald”) follows. Thus, Russell’s interpretation of definite descriptions begins in grammatical form with the simple phrase “the F.” It is then paraphrased as a conjunction of existence and uniqueness, thus producing a more complex linguistic form. This logical form is quite different from the apparent form in ordinary language, where “the F” is not a conjunction at all. The definite description disappears as a singular term in this analysis, and so it has no reference assigned to it. A side note on a slightly technical part of the Russellian analysis: there are two ways of logically analyzing uniqueness. One is with this notation: “∃!x (Fx and Gx),” read “There is a unique x such that Fx and Gx.” This is a very easy and convenient way to build uniqueness into the quantifier. In that way, we have specified uniqueness without an analysis: we just use “!” as a primitive symbol expressing uniqueness. But there is also another nice way of analyzing uniqueness in logical vocabulary. Consider the following: (4)  There is an x such that Fx and for all y if Fy, then x = y and Gx. In plainer language, this analysis is saying the following: “There is an x such that x is a king of France, and for any object y, if y is a king of France then y is identical to x, and x is bald.” This is a way of saying that someone is uniquely king of France and bald. We are saying, intuitively, that if there is anything else in the world that is a king of France, then it is identical to the first thing. That implies that there is not more than one thing, because

Russell on Definite Descriptions 

69

anything else that is a king of France is just the first thing. Such is the standard way for expressing uniqueness using ordinary quantifier logic with identity. It is not essential to understanding the theory. Rather, it is one way to analyze what uniqueness means. Uniqueness just means “at most.” This part of the theory, using standard logic, is not essential to Russell’s basic idea—it is just one explanation of what uniqueness is. As we have seen, Russell thinks that definite descriptions are not proper names, despite the fact that in some ways they appear to be proper names. Once the philosopher of language realizes that grammar is logically misleading, he or she can then have a theory that is not logically misleading. According to Russell, we do not need to postulate in our theory of meaning anything more than the reference of terms, once our sentences have been fully analyzed. Russell is a kind of Millian about genuine proper names, because he believes that ultimately expressions mean what they do in virtue of referring to what they refer to. If Russell does not believe that definite descriptions are proper names, we may wonder what proper names are for him. Russell does think there are proper names, but he has a peculiar set of criteria for them. As before, one of his points is that the words that appear in language to be proper names are not actually proper names, because language is logically misleading. Therefore, a name like “Bertrand Russell” will occur in a language though it is not a proper name at all. Russell advocates the description theory of names and considers such names to be the equivalent of a description. He takes a name and gives a paraphrase of it, turning it into a description (e.g., “the author of Principia Mathematica”), and then analyzes the description by his theory of descriptions, thereby eliminating the name as a name. According to Russell, none of the names of ordinary language is a logically proper name. They are all fake names—they all appear to be names, but they are not actually names. His view is that all the standard words we consider to be proper names in natural language are all disguised definite descriptions, and those descriptions are all analyzed away by the theory of descriptions. Following his theory, these descriptions do not have their meaning in virtue of their reference; so neither do ordinary proper names. Russell does believe that there are words that can have meaning in virtue of their reference, but those are what he calls logically proper names. Logically proper names are meaningful in virtue of what they refer to. Our ordinary names are not logically proper names, however, because they do not

70 

Chapter 3

have meaning in virtue of what they refer to. There is the logical category of proper names, but none of the ordinary expressions called names belong in that category. Russell’s view is rather peculiar when compared to the more grammatically conservative views of Frege and Meinong. He thinks that language is so misleading that, despite appearances, it does not even contain real proper names! In the following excerpt, Russell describes what he means by names: A name is a simple symbol whose meaning is something that can only occur as subject, i.e., something of the kind that we defined as an “individual” or a “particular.” And a “simple” symbol is one which has no parts that are symbols. Thus “Scott” is a simple symbol, because, though it has parts (namely, separate letters), these parts are not symbols. On the other hand, “the author of Waverly” is not a simple symbol, because the separate words that compose the phrase are parts which are symbols. … We have, then, two things to compare: (1) a name, which is a simple symbol, directly designating an individual which is its meaning, and having this meaning in its own right, independently of the meanings of all other words; (2) a description, which consists of several words, whose meanings are already fixed, and from which results whatever is to be taken as the “meaning” of the description. A proposition containing a description is not identical with what that proposition becomes when a name is substituted, even if the name names the same object as the description describes. “Scott is the author of Waverly” is obviously a different proposition from “Scott is Scott”: the first is a fact in literary history, the second a trivial truism. And if we put anyone other than Scott in place of “the author of Waverly,” our proposition would become false, and would therefore certainly no longer be the same proposition.3

His idea here is that a proper name is a simple symbol having no analysis and no parts. It means what it does simply in virtue of what it designates. Definite descriptions are not proper names in that sense at all, because the proposition expressed cannot be preserved by substituting a name for a description (or vice versa). This substitution is not plausible because definite descriptions and names are completely different types of expressions, having quite different sorts of meanings. Russell also employs the idea of “direct designation.” Direct designation characterizes how a real name directly designates its bearer—not via any description. A name does not express a description that then picks out an object. Instead, a name directly designates its bearer, and the bearer is the meaning of the name. Russell is a Millian, then, because he believes that names have their meaning in virtue of their reference and their reference alone. One thing to notice is that in “Definite Descriptions” Russell fails to say anything about what would be a proper name. But in other writings

Russell on Definite Descriptions 

71

he suggests that a logically proper name is a demonstrative, because a demonstrative can refer directly to sense data. In Russell’s view, one cannot refer directly to material objects since the material object might not exist (the viewer could be hallucinating the object). Therefore, the only logically proper names are phrases like “that black patch you are now seeing,” where this refers to a subjective sense datum. According to Russell, the only logically proper names are demonstratives, and they can only refer to sense data. This certainly seems odd; we don’t usually classify demonstratives as names. When did you last call one of your sense data by its proper name? Have you ever referred to a sense datum as, say, “Phil”? If we look back at our discussion of Frege, we may have a few questions in regard to Russell’s Millian theory. For instance, how does Russell’s idea of logically proper names work with identity statements? Russell never talks about that, perhaps because he is very concerned by the question of existence. Frege’s main concern is with identity. Russell does not have anything to say here about identity statements. He assumes that two logically proper names of the same thing have the same meaning, because the meaning of a proper name is its bearer. Russell is committed to the position that an identity statement linking two logically proper names must be a tautology. Russell avoids an obvious objection here by avoiding the question of Hesperus and Phosphorus. Russell’s position as to how to handle an identity statement that links two logically proper names is that two nonsynonymous logically proper names, in his system, cannot designate the same object. Names can differ in their meaning while referring to the same thing only if they are not really names. If they are names, as Russell defines logically proper names to be, then they cannot differ in their meanings while co-denoting. The identity statement must contain demonstratives that refer to sense data. Of course, it will be a false identity statement if the reference is to two different appearances. For the viewer, Hesperus elicits different sense data in the morning than Phosphorus does in the evening. Because these represent two entirely different pieces of sense data, they do not fit Russell’s strict criteria for logically proper names. Thus “Hesperus” is not a name, for him, but “this sense datum of a luminous point” is. In Russell’s system, there are no identity statements that are informative and contain ordinary names. One important consequence of Russell’s theory that generated much discussion is how he handles truth-values. According to Russell, the

72 

Chapter 3

truth-value of the sentence “The king of France is bald” is false. It is natural to assume that this statement would be false only if the subsistent Meinongian king of France had hair. Russell does not think along these lines at all. He believes that any statement containing that description is false, because the king of France does not exist. In his handling of truth-values, the sentence “Sherlock Holmes is a detective” is false, because it logically implies the real existence of Sherlock Holmes. In his famous article “On Referring,” P. F. Strawson objected to this point, arguing that such a statement is neither true nor false, because there is no king of France to be bald or not be bald. The only way for that sentence to be true would be by the king of France being bald, and the only way it could be false is by the king of France having a good head of hair. Since neither of those things is the case, the statement “The king of France is bald” must to be neither true nor false. But Russell’s analysis implies that it straightforwardly false. 3.6  Problems with Russell Although in the previous sections we explained Russell’s analysis, we have not yet discussed whether or not this analysis is correct. The following passage is an excellent summary of what we discussed in the previous sections: We may even go so far as to say that, in all such knowledge as can be expressed in words—with the exception of “this” and “that” and a few other words of which the meaning varies on different occasions—no names, in the strict sense, occur, but what seem like names are really descriptions. We may inquire significantly whether Homer existed, which we could not do if “Homer” were a name. The proposition “the so-and-so exists” is significant, whether true or false; but if a is the so-and-so (where “a” is a name), the words “a exists” are meaningless. It is only of descriptions—definite or indefinite—that existence can be significantly asserted; for, if “a” is a name, it must name something: what does not name anything is not a name, and therefore, if intended to be a name, is a symbol devoid of meaning, whereas a description, like “the present king of France,” does not become incapable of occurring significantly merely on the ground that it describes nothing, the reason being that it is a complex symbol, of which the meaning is derived from that of its constituent symbols. And so, when we ask whether Homer existed, we are using the word “Homer” as an abbreviated description: we may replace it by (say) “the author of the Iliad and the Odyssey.” The same considerations apply to almost all uses of what look like proper names.4

In this passage, Russell makes three major points. He defines a name as a simple symbol whose meaning is its reference. A name without reference

Russell on Definite Descriptions 

73

would lack meaning. Calling a name “empty” is a contradiction in terms, because a name without reference is not a name. Russell also believes that descriptions are quantifiers and that ordinary “names” are equivalent to descriptions. The only reason why ordinary proper names appear to be names is because of the infirmities of natural language. Russell’s conception of genuine names has an obvious consequence for existential statements. He believes that existential statements are highly misleading because they appear to contain names when they do not. Statements like “a exists” look like they contain the proper name “a.” There are two possibilities for this type of statement. First, if the name does refer to something, then the meaning of the name guarantees that the name has a reference. Therefore, adding “exists” to the name is stating a tautology, because names in Russell’s system will refer only to things that exist. We can create an example to illustrate this point. If someone looks up outside and says, in reference to the color of the sky, “That shade of blue exists,” he knows that that shade of blue exists, because it is an aspect of a sense datum. To add that the color exists is unnecessary, since that is understood in virtue of grasping the name alone. The second possibility arises if the name does not refer to anything. If the name does not refer to anything, then the statement containing it must be a meaningless statement with a meaningless part—and hence not a real statement. Take the sentence “a does not exist.” Since the name “a” does not refer, we can say that it is empty. The problem with that alleged statement, “a does not exist,” is that it cannot be true since the name lacks reference and is therefore meaningless. According to Russell, existential statements cannot be applied to names. On the other hand, existential statements can be applied to descriptions, because in the case of descriptions they do not need reference in order to have meaning. Existential statements will never contain names. In Russell’s system, names must refer to have meaning, so it is trivial to say that their reference exists because it will always have to exist. Russell is making a very radical proposal. The thought behind this proposal is that there are propositions that lurk behind sentences and each proposition has a kind of intrinsic logical form. It is as if these propositions are clothed in the sentences of ordinary language, but the clothing is very misleading as to the real form of the proposition. The job of the philosopher is to slip beneath the clothing and discern the real nature of the proposition. Then, he is able to devise a notation to reflect that nature. Russell’s

74 

Chapter 3

proposal led to the idea that philosophers needed to devise a logically perfect language that reveals the actual structure that is hidden behind ordinary language. In our example of “a exists,” the sentence looks like a subjectpredicate sentence like “a is red,” but in actuality it is a quantifier sentence. Therefore, the underlying proposition is of a completely different kind than that expressed by the sentence “a is red.” One of the reasons why Russell’s analysis of descriptions was so important was that it initiated discussions on the possibility of creating a logically perfect language. Many philosophers believed that such a logically perfect language could solve all philosophical problems. In particular, a completely logical language could solve ontological problems, ridding us of the shadowy ontology of Meinong. For example, consider the ontological proof for the existence of god: God has all perfections, and one of those perfections must be existence, and therefore God must exist. According to Russell, this presupposes that existence is a predicate. In other words, subject-predicate sentences like “God exists” would assign a predicate to something named by “God.” According to both Russell and Frege, that sentence is not a subject-predicate sentence at all, because the word “exists” is not a predicate. The idea is that existence is not a predicate or a property of things, like being red. Rather, it is a secondorder concept that is really a property of a propositional function. Therefore, the ontological argument is unsound. To resolve philosophical problems, we must reform language so as to reflect the hidden form of propositions. 3.7  Primary and Secondary Occurrences So far, we have only considered sentences of the form “The F is G.” We may wonder how Russell handles sentences of the form “The F is not G.” He argues that such sentences are ambiguous. To understand his point, we can consider a case where the “not” applies to a predicate, for example, “The queen of England is not pregnant.” Here we are attributing nonpregnancy to Her Majesty. But instead of placing the negation sign immediately before “G,” we could place it at the beginning, creating the sentence “It is not the case that the queen of England is pregnant.” If we translate this into Russell’s analysis, we get the negation of the existential clause: “It is not the case that at least one thing is a queen of England.” This sentence is expressing the proposition that it is not the case that a queen of England exists.

Russell on Definite Descriptions 

75

Let us now consider an example where the description is empty: “It is not the case that there is at least one king of France.” By negating the existential statement that there is a king of France, the statement becomes true. Since it is not the case that there is at least one king of France, the sentence “The king of France is not bald” will be true when interpreted this way. But under the first interpretation, the sentence will not be true. These two propositions have different truth-values. Thus, the truth or falsity of the sentence depends on at what point the negation is inserted. In the latter case, the entire sentence was negated; in the former case, only the predicate was negated. Consider the sentence “It is not the case that there is a queen of England and she is pregnant.” Since there is a queen of England, this sentence is false. On the other hand, if “not” were placed before the predicate, the sentence would be true (since the queen of England is not pregnant). To handle this kind of ambiguity, Russell brings in the concepts of primary and secondary occurrence. A primary occurrence of the description happens when the negation occurs before the predicate. A secondary occurrence of the description happens when the negation applies to the whole sentence including the description. To illustrate this point more clearly, we can bring in the concept of the scope of negation from logic. In the primary occurrence, negation has narrow scope; in the secondary occurrence, negation has wide scope and thus applies to the description. The scope merely tells you what is included in the negation. Are we negating the whole proposition or just the part of it that corresponds to the predicate? This point about negation also applies to necessity. Like negation, necessity has a similar kind of ambiguity. One might wonder how to read the sentence “The queen of England is necessarily pregnant.” It could be read either as “Necessarily there exists a queen of England and only one and she’s pregnant” or as “There exists a queen of England and only one and she’s necessarily pregnant.” In the former case the modal operator has wide scope; in the latter, narrow scope. These can have different truth-values. When these sorts of operators like negation, necessity, or possibility occur in sentences containing descriptions, the scope determines the logical interaction between the operator and the description. This interaction can get quite complex if the sentence contains multiple operators. This concludes our discussion of Russell’s theory of descriptions. In the next chapter we will look at some possible criticisms of Russell.

4  Donnellan’s Distinction

4.1 Introduction To summarize our progress so far, we have examined two major theories of descriptions: Frege’s theory and Russell’s theory. In Frege’s theory, descriptions are proper names referring to things. In Russell’s theory, logically proper names refer, but descriptions do not—they are analyzed in terms of quantifiers. In a case in which a description fails to apply to anything, these two theories have different consequences. For Russell, statements made using descriptions without reference (e.g., “The king of France is bald”) are always false, since they assert existence. Since the sentence expresses in part the proposition that there exists a king of France, and there is no king of France, the truth-value of the sentence is false. According to Frege, such a sentence would be neither true nor false. If the description refers to something, the sentence is true if the predicate applies to the object to which the description refers. The condition for it to be false is that the thing referred to by the description does not satisfy the predicate. However, if the description does not refer to anything, it can be neither true nor false. Therefore, it is not true of every proposition that it is either true or false. P. F. Strawson is famous for making this idea of “truth-value gaps” explicit in his paper “On Referring.” The point becomes clearer when we consider an example involving names. Take an ordinary proper name used in a statement. If that name refers to nothing at all, we would not conclude that the statement is false, because there is no reference to fail to satisfy the predicate. It is neither true nor false. These two theories are intended to give a uniform account of the meaning of definite descriptions whenever they occur. They are theories of the “inner logic” of descriptions.

78 

Chapter 4

We shall see that Keith Donnellan disagrees with both of these camps. According to Donnellan, these uniform accounts of the semantics of definite descriptions cannot give an analysis of definite descriptions as they are used in every statement. He proposes that definite descriptions can function in two different ways. In some statements, they function in the way Russell claims, and in other statements they function in the way Frege and Strawson claim. Donnellan does not reject their views completely, but he thinks that neither one nor the other theory covers the semantics of all definite descriptions. According to Donnellan, there is a third possibility as far as truth-values are concerned. Russell thinks that an empty description gives rise to a false sentence. Frege thinks it gives rise to a sentence that is neither true nor false. Donnellan thinks that an empty description can give rise to a true statement. Thus, he offers a third possibility. His reasons will emerge as we go on. The general point Donnellan makes with his examples is that descriptions can work in more ways than the uniform ones recognized either by Russell or Frege/Strawson. The theories we have examined so far analyze only the semantics of language. Donnellan believes that to have a more complete theory of language, we must include the pragmatics of language. Semantics is about the abstract analysis of language independently of speakers, whereas pragmatics examines language in relation to speakers in concrete speech situations. Donnellan’s critique forms part of a more general movement toward the analysis of speech acts in the understanding of language. We have to look at what speakers do with words and not just at the words themselves. Donnellan believes that our views about how descriptions function in acts of communication will change if we examine the role of descriptions in speech acts. 4.2  Referential and Attributive Uses Donnellan calls the view of Strawson and Frege the referential view of descriptions, because they take the position that descriptions are referential, namelike devices. Since Russell’s stance is that a definite description is a quantifier, we could label Russell’s theory the quantifier view of descriptions. But Donnellan chooses to call it the attributive view. The following passage outlines his understanding of these terms:

Donnellan’s Distinction 

79

I will call the two uses of definite descriptions I have in mind the attributive use and the referential use. A speaker who uses a definite description attributively in an assertion states something about whoever or whatever is the so-and-so. A speaker who uses a definite description referentially in an assertion, on the other hand, uses the description to enable his audience to pick out whom or what he is talking about and states something about that person or thing. In the first case the definite description might be said to occur essentially, for the speaker wishes to assert something about whatever or whoever fits that description; but in the referential use the definite description is merely one tool for doing a certain job—calling attention to a person or thing—and in general any other device for doing the same job, another description or a name, would do as well. In the attributive use, the attribute of being the so-andso is all-important, while it is not in the referential use.1

The attributive use is shown in sentences where the predicate “F” in the description is used to apply to whatever satisfies it, not to a particular thing. The fact that a thing in the world actually satisfies the predicate is essential and all-important. With Donnellan’s notion of attributive use, we could paraphrase the sentence “The king of France is bald” as “Whoever is uniquely king of France is bald”—perhaps asserted in the belief that being king of France induces baldness in whoever occupies that position. To determine if this sentence is true, we would have to find whoever in the world satisfies the description “the king of France” and then determine whether that person is bald. This is clearly along the lines of Russell’s analysis of the semantics of descriptions. The referential use occurs when the description picks out a particular object in order to identify something for an audience, where the description is just a tool for directing the audience’s attention in the right way. In the simplest case, the object of interest is right in front of the speaker and in plain sight of the audience. The description is used to show the audience the particular object the speaker has in mind. The description here is not essential and all-important, because many other modes of identification would work as well. Imagine a classroom full of students in which one of the male students is wearing a green shirt. A female student could make a statement about him in the following ways: “The guy in the green shirt has a pensive look,” “He [pointing] has a pensive look,” “Billy has a pensive look.” The speaker then chooses one but could have used the others, depending on what she thinks will direct the audience’s attention to the right person most effectively. Her purpose was to identify a certain individual and make a remark about him—she couldn’t care less about the

80 

Chapter 4

description itself. She wanted to point out the guy’s pensive look and any mode of designation would do. Donnellan’s point is that these are very different speech situations in which the speaker has very different communicative intentions. According to him, the description functions differently depending on the intention behind the speech act. He uses a thought experiment to illustrate this point more clearly. Imagine that a detective on a crime scene has encountered the body of a dead man, Smith. The condition of the body is so mutilated that the detective says, “Smith’s murderer is insane!” When he said that, he did not know the identity of the murderer. That statement could be rephrased as “Whoever Smith’s murderer is, he is assuredly insane.” This is an excellent example of the attributive use. For that statement to be established as true, the detective would have to find the person who murdered Smith and determine whether or not he is insane. He certainly had no specific individual in mind; hence the use of the quantifier phrase “whoever is the murderer of Smith.” The same description could also have a referential use. Suppose Jones is being tried for the murder of Smith and one of the jurors notices that Jones is behaving erratically the whole time. The juror then points at Jones and says, “Smith’s murderer is insane.” The juror has thus succeeded in identifying Jones. He wanted to single that man out and make a remark about him; here the quantifier phrase would be quite inappropriate. Now consider a case where Jones is not in fact Smith’s murderer though he is on trial and behaving erratically. Donnellan thinks the juror has still identified that individual even though he is not Smith’s murderer, because the audience understands that he intends to refer to Jones and to say he is insane. It could be the case that Jones is insane but not Smith’s murderer. In that case, the juror has still said something true about Jones because he is insane and the speaker has singled him out. Regardless of the situation and the truth or falsity of the juror’s description, the juror has succeeded in identifying the individual in question by using that definite description. The description itself is not all-important in the reference that the juror has achieved and it is not essential that the referent actually satisfy it. Although the description may be defective if it does not apply to Jones (depending on the situation), the juror has still managed to identify a particular individual using that description. It is as though the description can function either as a quantifier phrase or as a demonstrative that points out someone. The juror has succeeded in his

Donnellan’s Distinction 

81

referential intention by identifying an individual and making a statement about him. The detective, on the other hand, is best interpreted as saying something analyzable along the lines of Russell’s theory. There is another thought experiment Donnellan uses that illustrates the same point. Imagine that you are at a party and there is a man apparently drinking a martini who is a famous philosopher. Seeing this man, you say, “The man drinking a martini is a famous philosopher.” However, suppose that although the man is a famous philosopher, he is drinking water from a martini glass, not a martini. You have said something true about him, but your identifying description does not apply to him. Nevertheless, it can still perform its function of indicating to whom you meant to refer. Now we can consider a similar case that illustrates the attributive use. Imagine that the woman running the party does not want people drinking alcohol and says, “Who’s the man drinking the martini?” She is not intending to identify a particular individual as you were in the previous example—indeed she is trying to discover who the martini drinker is. If it turns out that the man apparently drinking a martini is not drinking a martini, she will not be concerned. Her speech act requires that there be somebody who satisfies that description. If there is somebody at the party who fits that description, she would have accomplished her aim by using that description. She is using the description to mean “whoever is drinking a martini”—she has no particular individual in mind. It is also possible that there is in fact another man at the party who is drinking a martini, is in another room, but is not a famous philosopher. The statement “The man drinking a martini is a famous philosopher” would then be false if the description is interpreted attributively. Although the man drinking the martini was not your intended reference, he happened to fit your description. Your reference was to the person you incorrectly described as a martini drinker, though you also said something true about him. The best way to understand both of these examples is to determine the intention of the speaker. Ask yourself—does the speaker intend to identify a particular individual or just to speak of whoever satisfies a particular description? Sometimes the use of the definite description will have a general (attributive) intention behind it, and sometimes it will have a singular (referential) intention behind it. It all depends on what the speaker intends to communicate.

82 

Chapter 4

Donnellan continues the article by reiterating the main line of argument. Each of his subsequent examples illustrates the difference in intention in the attributive and referential uses. That is Donnellan’s fundamental way to understand any of these cases. If it does not matter that the description fits the object, it is a referential use. If it does matter, then it is an attributive use. Therefore, we can actually refer to something using a description without truly describing what we are referring to. Referential success does not depend on accurate description. To sum up: The core of Donnellan’s argument is the distinction between referential and attributive uses. He illustrates the distinction by means of the thought experiments we have just described. A speaker uses the description attributively when she says “Smith’s murderer” with a general intention. The speaker has no particular person in mind when using this description. She could have equally said, “The murderer, whoever he may be, must be insane.” The referential use occurs when the speaker has a particular individual in mind and uses her description to pick out the individual she has in mind. Donnellan’s main argument deals with these two uses of descriptions—the generality of the attributive use and the particularity of the referential use. A consequence of the distinction, according to Donnellan, is that in the referential use the speech act can be successful regardless of the truth or falsity of the description. Referring back to the Smith’s murderer case, Jones might not be the murderer but a juror can still identify Jones by saying, “Smith’s murderer is insane.” Unlike the attributive use, the descriptive content is not all-important in the referential use. The description in the referential use is incidental, a mere instrument to identify an individual. Donnellan thinks that the theories of Russell and Frege/Strawson are incorrect because they do not acknowledge the duality of uses of descriptions. In the rest of his paper, he brings up various consequences of this basic point. By understanding the distinction between these two uses, we can understand his main line of argument. One obvious point is that the referential use occurs when a particular thing is pointed out, and in the attributive use a remark is made involving some general notion. It is the difference between a quantified proposition (as in “whoever”) and a particular proposition (as in “this individual”). The distinction is analogous to the distinction Russell discusses when he talks about the difference between a name and a description. Using our understanding of Russell is another way to explain

Donnellan’s Distinction 

83

Donnellan’s distinction. He thinks that some uses of definite descriptions are namelike in the Russellian sense, but others are propositional-functionlike. Yet the expressions themselves look uniform from use to use. One of the consequences of this distinction is that even though in both uses the speaker presupposes the individual she is referring to (or trying to refer to) satisfies the description, there are different outcomes of the individual not satisfying the description. If the description is attributive and no one satisfies it, then the statement cannot be true. According to Russell, the statement would be simply false. For example, according to the theory of descriptions, “The king of France is bald” is false because there exists no such thing as a king of France. If we use this description attributively, and the implication of there being something that fits the description turns out to be false, then the statement cannot be true and must be false. On the other hand, according to Donnellan, in the case of referential use the statement can still say something true regardless of whether the intended referent satisfies the description. It might be that Jones really is insane even though Jones is not in fact Smith’s murderer. There may also be instances where the speaker does not even believe the description she uses to refer to the individual is true of that individual. In most instances, the speaker will think that the description applies (e.g., that Jones in the dock is a murderer or that the man over there is drinking a martini). However, Donnellan suggests that there can be instances where the speaker knows the description is not true but uses it to identify the individual anyway. Consider the example he gives of a spurious king. The speaker may believe that this alleged king is a usurper and so not really king, but because everyone else in the country thinks that the man is the rightful king, the speaker still refers to him as such (e.g., “Is the king in the counting house?”). The speaker does not believe that the individual she wishes to speak of is the king but uses the royal description anyway. She makes a successful referential use out of a piece of false description. The hearer of the sentence may or may not believe the description as well. For example, instead of all of the people around the alleged king thinking that he is the king, they could all think he is a usurper. They may still refer to him as “the king” to avoid any trouble. Everyone in the court will refer to the usurper with the description “the king” and know he is not the king but still use that description anyway. In this case, if our original speaker asks, “Is the king in the counting house?” everyone in the court will

84 

Chapter 4

understand to whom the speaker refers, even though they do not believe that impostor to be the king. The description can still refer even if it is false— even if the speaker and hearer know it to be false. 4.3  Denoting and Referring Having said all this, Donnellan makes a further distinction between denoting and referring. He does not deny that there may be a sense in which the description “Smith’s murderer” denotes somebody other than Jones, assuming Jones is innocent. The juror is referring to Jones with that false description, but Donnellan accepts that the description may have a denotation that is other than Jones. If we suppose that Brown is the man who actually killed Smith, then “Smith’s murderer” denotes Brown. In that case, the juror refers to Jones by saying “Smith’s murderer,” but his description denotes Brown. Donnellan borrows this idea of denotation from Russell. According to Donnellan, the speaker can refer to somebody with a description who is not the person denoted by that description. Thus referring is to be distinguished from denoting. Denotation is a semantic notion—a strict and literal interpretation of the phrase “Smith’s murderer.” It is not the pragmatic notion of what or whom the speaker is referring to in using that phrase. This marks the distinction between a pragmatic question and a semantic question. In effect, Donnellan is admitting that he is primarily interested in the pragmatic question of how individual speakers convey a message to hearers on particular occasions. He accepts that the description, considered in itself, denotes (semantically) whatever fits the description, and so functions “attributively.” So a speaker can use a description that semantically denotes a particular individual (Brown) to pragmatically refer to another individual (Jones). Donnellan is therefore not arguing that there are two interpretations of semantic denoting. He thinks that denoting follows Russell’s theory, but there are pragmatic uses in which a speaker refers to something other than the denotation. As a matter of fact, at one point in “Reference and Definite Descriptions” Donnellan clearly states that he is not arguing against Russell’s semantic theory: It does not seem possible to say categorically of a definite description in a particular sentence that it is a referring expression (of course, one could say this if he meant

Donnellan’s Distinction 

85

that it might be used to refer). In general, whether or not a definite description is used referentially or attributively is a function of the speaker’s intentions in a particular case. “The murderer of Smith” may be used either way in the sentence “The murderer of Smith is insane.” It does not appear plausible to account for this, either, as an ambiguity in the sentence. The grammatical structure of the sentence seems to me to be the same whether the description is used referentially or attributively: that is, it is not syntactically ambiguous. Nor does it seem at all attractive to suppose an ambiguity in the meaning of the words; it does not appear to be semantically ambiguous. (Perhaps we could say that the sentence is pragmatically ambiguous: the distinction between roles that the description plays is a function of the speaker’s intentions.)2

This is a very important passage in terms of the significance of Donnellan’s arguments. He claims here that there is no semantic ambiguity in descriptions. By semantic ambiguity he means what the words actually mean in the language—their logical analysis. There is no semantic ambiguity in descriptions even though speakers may use those descriptions in two different ways. He thus in effect admits that descriptions are always semantically attributive, that is, Russellian. One of the major criticisms of Donnellan, to be considered later, is that his critique of Russell’s theory is ineffective because he tries to apply a pragmatic distinction to a semantic question. Therefore, understanding the import of this passage is particularly important to this discussion. 4.4  Truth-Value Gaps Donnellan makes some of his main points against Strawson toward the end of his article. He argues that Strawson is wrong to suggest that when using an empty description referentially the speaker says something neither true nor false. According to Donnellan, the speaker can say something true by using a description that fails to refer. If there was no murderer of Smith at all, but just a gruesome accident, and the speaker shouts, “Smith’s murderer is insane!” referring to Jones, Strawson thinks that the utterance would be neither true nor false. But Donnellan argues that the speaker would have said something true of Jones, assuming that he is in fact insane. He goes on to say that in certain instances he agrees with Strawson. There might be cases where you do fail altogether to refer to an object using a description. Imagine first a case where an onlooker sees a man apparently carrying a stick and says, “The man carrying a stick is out of breath.” Now suppose a man is there, but he is carrying a rifle instead of a stick. Donnellan

86 

Chapter 4

thinks the onlooker is still referring to the man there even though the man carrying a rifle does not fit the speaker’s description. However, there could be a case where the onlooker has completely hallucinated a walking man. The onlooker could also have mistaken a tree or a rock for a man with a stick, in which case Donnellan believes the onlooker would have still successfully referred to something. But this referential ability comes to an end at a certain point. If the onlooker has completely hallucinated a man with a stick and there is only empty space there, not even a tree or a rock, then Donnellan thinks he has failed to refer to anything whatever—man, rock, tree, chunk of space. He is totally out of luck, referentially speaking. Strawson would have been right to say that in such a case the utterance is neither true nor false. Here the speaker’s intention to refer would have been completely nullified. The question of truth-value would not arise in that sort of case. Therefore, Donnellan does think that there are cases of intended reference to something where it turns out that no reference takes place. The consequence of such radical reference failure is that the speaker will be stating something neither true nor false. Of course, in Russell’s theory, such a statement would express a straightforwardly false proposition. Donnellan takes a middle ground. He does not think that it is always the case that the speaker says something true or false, but he also thinks that Strawson overstates the frequency of such truth-value gaps. He thinks both Russell and Strawson are wrong about certain cases of reference failure, though they are right about others. As in his concluding points on Strawson, Donnellan sees some commonality between his views and Russell’s. Although Donnellan believes that Russell’s theory is incomplete because it does not recognize the referential use of descriptions, he still thinks that his conception of descriptions is analogous to Russell’s conception of names. Russell regards genuine names as labels for particular objects and not as descriptions of objects. He thus makes a sharp distinction between names and descriptions. In Russell’s system, a genuine name acts merely as a tag for an object and does not describe the object at all. Donnellan suggests that he can map his distinction onto Russell’s distinction, because he thinks that descriptive content does not play a role in the referential use of descriptions. Donnellan views descriptions used referentially as mere labels for object—they are namelike. Whether or not the description correctly describes the object is irrelevant,

Donnellan’s Distinction 

87

because the object has been successfully labeled. In his system, these descriptions only appear to be descriptions, because they do not refer by describing. They only label or point. Descriptions can thus act like names in the Russellian sense, and so it is not important if the object satisfies the description, since they succeed in referring even if inaccurate. For Donnellan, the descriptive content of the description is incidental and dispensable to the role it plays in referring, in the case of referential uses. There is another class of examples that Donnellan does not cover in his paper, but which illustrates this point well. In these examples, the descriptions function as names, and it is entirely obvious that they do not accurately describe the things to which they refer. Consider “the Holy Roman Empire”—a description that notoriously refers to something that was neither holy, nor Roman, nor an empire. The description in that example is not referring by means of its descriptive content. These words stand for something that is entirely removed from their actual predicative meaning. Compare “the European Community” or even “the United States” or “the Grand Order of Pig Farmers” (this one is invented!). The combinations of words in these descriptions have become labels that refer but the descriptive meaning is beside the point. These are like Donnellan’s referential uses. 4.5  Evaluating Donnellan’s Distinction When assessing the cogency of Donnellan’s arguments, it is important to consider certain cases that may arise when using other types of expression in sentences. Consider a situation similar to his thought experiment about the famous philosopher at the party apparently drinking a martini. This time imagine that the famous philosopher at the party is a certain individual, say, Jerry Fodor. Let’s suppose that the hostess of the party has heard of the philosopher Saul Kripke and heard him described; moreover, she has reason to believe he is at the party. Suppose now that she sees Fodor chatting to a group of people about philosophy. She forms the belief that this must be Kripke and says, “Kripke is very animated.” Of course, she is wrong about who is in front of her, but the question arises—whom did she refer to with that name? We might be tempted to say that she succeeded in referring to Fodor with “Kripke” and made a true remark about him, even though her referent does not “fit” the name she used. Kripke himself might be in another room passed out, so not at all animated—did she refer to him

88 

Chapter 4

and make a false statement about him? Following Donnellan, we might say that such an example illustrates a referential use of names, in which accuracy counts for little: hasn’t she in some sense referred to the man in front of her, that is, Fodor? Semantically, the name denotes Kripke, but pragmatically our hostess seems to have referred to Fodor. She has referred to somebody who is not Kripke with the name “Kripke,” which has a specific meaning that makes it denote only Kripke. In other words, our hostess has used the name in a way that does not fit its actual conventional meaning. It thus seems that Donnellan could have written a paper called “Reference and Names” and said all of the same things about names that he says about descriptions. There are two uses of names, referential and attributive, and referring must be distinguished from denoting, and so on. But something seems to be going wrong with this argument if the ways in which speakers misuse words show that particular semantic theories of names are false. If we wonder whether Donnellan’s objections apply to theories of proper names, then we must also consider if they apply to demonstratives. Suppose a tourist in front of an animal at a zoo says, “That antelope is brown.” However, the animal is not an antelope but a different species of deer. Although the speaker has in some sense succeeded in referring, the animal he has pointed out does not fit the demonstrative he used. The speaker’s misuse of the demonstrative is exactly like the hostess’s misuse of the name “Kripke.” The intended reference of the tourist is the animal in front of him, but it is not an antelope as he supposed. It is possible, then, to use a demonstrative to refer to something other than the denotation of that demonstrative, if it has one. According to Russell and Strawson, such a demonstrative would be empty since it lacks denotation. But the tourist could still succeed in saying something true about the animal in front of him, even if the animal were not an antelope. Since it holds of names and demonstratives, it seems as though we could apply the Donnellan treatment to any expression. There are myriad examples in popular culture of linguistic misuse, particularly when speakers use certain terms to try to sound sophisticated but instead demonstrate their ignorance. Certain speakers treat the words “disinterested” and “uninterested” as though they are interchangeable. But “uninterested” means to lack interest in something whereas “disinterested” means to be impartial about something. A disinterested observer of a tennis match need not be an uninterested observer. On the contrary, a disinterested observer could be a very

Donnellan’s Distinction 

89

interested observer—she is just impartial. Someone may say, “I’m just completely disinterested in that subject matter,” and the hearer, knowing this common error, can infer from the speaker’s misuse of the word the point she is trying to make, namely, that she lacks interest in that subject matter. True things can thus be conveyed by the misuse of words. If we were very ingenious about it we could create Donnellan cases for quantifier words, or even words like “and” or “not”—almost anything. All you need to do is produce a case where a speaker utters a word that has a certain conventional meaning (denotation) and uses that word in an incorrect way. Even though the word will not apply to the thing the speaker is applying it to, the audience understands what the speaker meant to convey and so the speech act is successful. Any expression of language at all could be used in that deviant way. If you know that I have a tendency to confuse my quantifier words (perhaps I am new to the language I am speaking), you can on occasion interpret my use of “someone” to mean “no one”—so when I say “someone is in that room” you interpret me to be intending to convey my impression that no one is in that room (especially if the room is palpably empty). The significance of that point concerns whether or not the production of Donnellan cases can undermine semantic theories of certain classes of expressions. If there is an established, semantic definition of a word, captured by a particular theory, can that theory be undermined by pointing out that people sometimes misuse the word? No. The misuse of a word does not change its semantic status, or show that a particular theory of its meaning is incorrect. People can misuse words in the way Donnellan describes, but that does not mean that these misuses establish an interesting linguistic duality. If a foreign speaker of English does not understand the language and he uses the word “and” when he means “all,” his misuse of the word “and” would not change the meaning of “and,” or show that the theory of “and” as a truth-functional sentence connective is mistaken or oversimplified. Would we say that the meaning of “and” is ambiguous because a foreigner used the word incorrectly? No. Nor would we say that “and” has two uses, as a sentence connective and as a universal quantifier. As Donnellan admits in the passage cited earlier, he is not pointing out any semantic ambiguity. But then Donnellan’s considerations could not even be relevant to a question of semantics, because they are purely a matter of pragmatics. The pragmatic point he is making is that it is possible for speakers to use words to convey something that is completely divorced from what those

90 

Chapter 4

words actually mean. Thus a speaker can express a belief about Jones by using words that denote Brown (“Smith’s murderer”). Donnellan’s point is a purely pragmatic one and does not undermine any semantic theory. Since Russell’s and Strawson’s theories were offered as semantic theories, Donnellan’s point is irrelevant to those theories. For all Donnellan has said, Russell is completely correct about the semantics of descriptions. Descriptions will always semantically denote what fits them. Speakers can use such descriptions incorrectly to make singular reference, but that does not show Russell is wrong in the semantic theory he provides. 4.6  Implication and Implicature To further evaluate Donnellan’s position, we will now bring in some points made in an excerpt from Stephen Neale’s book Descriptions.3 In this excerpt, Neale makes use of some ideas developed by Paul Grice. Since these ideas are independently important, we will spend a little time explaining them. Perhaps the most well-known idea covered in his article is that of “conversational implicature.” To explain the notion of conversational implicature, we can use the example of a professor who is asked to write a reference letter for one of his graduate students: To whom it may concern, John Smith has very good handwriting. Sincerely, Professor Horatio Handwavy, PhD

The committee reviewing John Smith’s application would not infer that he has outstanding philosophical ability from this recommendation letter. They would infer that Professor Handwavy thinks poorly of Smith. Suppose that after reviewing Smith’s entire application and interviewing him the committee decides that Smith is an excellent candidate. Then one of the committee members asks the letter writer why he said that John was a poor student. Handwavy indignantly replies, “I did not say that he was a poor student, I just said that he has good handwriting. In fact I think Smith is a brilliant student.” And it is true that he never said anything false about Smith’s philosophical ability. In fact, he said something quite true, because John is also an excellent calligrapher. But he certainly implied something false, irresponsibly so. He didn’t straightforwardly lie, but he certainly gave

Donnellan’s Distinction 

91

an extremely false impression. He is morally at fault, even if not factually at fault. This example illustrates conversational implicature, which relates to what a statement suggests given its context. Nothing said in the letter as written logically implies that John Smith is a bad philosophy student. However, given the context of the recommendation letter, the professor did conversationally imply it. We could reasonably paraphrase the original sentence into its conversational implicature: in that context, saying “John Smith has good handwriting” is tantamount to saying “John Smith is a bad philosophy student.” The notion of conversational implicature reveals the distinction between what a speaker strictly says when uttering a sentence and what he or she implies in uttering it. What the speaker means, and can be reasonably taken to mean, can depart quite dramatically from the literal meaning of the sentence that is uttered. When a speaker utters a sentence, therefore, there is the proposition conversationally meant and the proposition literally expressed. These may coincide, but they may not. Neale outlines this distinction in his book. The proposition expressed is closely connected to the meaning of that sentence in a particular language, whereas the proposition meant depends on the context and expectations about the speech act. The proposition expressed and the proposition meant can be completely different propositions that are not logically related to each other. Therefore, in conversational implicature, propositions are implied conversationally that are not directly expressed by words. The point is very important philosophically because it undermines various philosophical claims made about various subjects. It is really important to distinguish between whether an utterance of a sentence is strictly speaking false and whether it is misleading to say it in a certain context. The fact that something is misleading to say in a certain context does not show that it is false. It is misleading to say, “It looks to me as if there is a dog in the doorway” if you are not in any doubt about there being a dog there, but it might still be true that this is just how things look to you. Neale’s main problem with Donnellan is that he disregards this distinction. Donnellan is suggesting that Russell’s analysis of definite descriptions is inadequate because it does not handle his cases of referential use. Neale rejects that form of argument, because he does not see Donnellan’s pragmatic points as having any implications for semantics. Although Neale never mentions it, we have discussed a passage from Donnellan’s original

92 

Chapter 4

article that admits this very distinction. In that passage, Donnellan clearly states that there is no syntactic or semantic ambiguity in sentences containing definite descriptions. Nevertheless, he still thinks there is something wrong with Russell’s account of the meaning of definite descriptions. The question is how he can make the admission and persist with the argument. He thinks that his two pragmatic uses somehow show that something is wrong with Russell’s semantic analysis, but he himself accepts that his considerations do not bear on semantics. Let us suppose that Russell’s account is correct for attributive uses, so that descriptions are quantifier expressions when used attributively. According to Donnellan, there is no semantic ambiguity in definite descriptions. Therefore, when definite descriptions are used referentially, they have exactly the same meaning as when they are used attributively. If that is the case, then we must suppose that Russell’s theory gives the correct meaning in both cases. We have seen how the misuse of words cannot undermine an analysis of their semantics. So Donnellan has not pointed to anything that could threaten Russell’s semantic theory. If Russell is correct about the attributive use, then he must be correct about the referential use as well. The curious thing is that Donnellan already admits the point that Neale is urging against him—that there is no semantic ambiguity. Yet he doesn’t seem to appreciate the significance of this admission. Neale believes that Donnellan’s arguments show the necessity of bringing to bear Grice’s distinction between the proposition expressed and the proposition meant. To understand why this distinction matters, we shall return to Donnellan’s examples. Let us consider again the “Smith’s murderer” case where Jones is the man in the dock. The juror sees Jones’s erratic behavior and wants to express his belief that Jones is insane, so he says, “Smith’s murderer is insane!” The proposition meant is that Jones, the man in the dock, is insane, even though Jones did not in objective fact murder Smith. The proposition meant is in line with Donnellan’s referential use. However, the proposition expressed by the sentence itself (“Smith’s murder is insane!”) is that Smith’s murderer is insane, which may or may not be true. In the case that Jones is insane, the proposition meant (that Jones is insane) would be true, but the proposition expressed would be false, assuming the real murderer (Brown) is not insane. The analysis of Donnellan’s examples with this Gricean distinction allows us to see that there are two different propositions “associated with” the utterance of the sentence in

Donnellan’s Distinction 

93

this instance. These propositions are about different individuals and can differ in truth-value. The handwriting case also illustrates the distinction between proposition expressed and proposition meant. In that case, the proposition expressed is that John Smith has good handwriting and the proposition meant (or apparently meant) is that John Smith is not a good philosopher. One proposition is entirely different from the other. Although the words can be used by a speaker to convey a particular proposition, the actual words spoken may not mean that proposition. What Donnellan really showed was that sentences can be used by speakers to mean propositions that those sentences do not express—and hence to convey information that the words do not themselves contain. Thinking about this idea in a more general way, we can see many uses of language have much the same character. Take irony, for example. If a speaker says something ironically, the proposition expressed is the opposite of the proposition meant—for example, “You are so smart,” said sarcastically. However, it would be strange to claim that the possibility of irony somehow changes the semantic analysis of the sentence. Irony depends on the fact that the proposition expressed is not the same thing as the proposition meant. Irony, then, is another example of this type of distinction working itself out, where the relationship between literal meaning and speaker meaning is complex. In this case, one proposition is actually the negation of the other. Hyperbole and exaggeration also illustrate these distinctions. Hyperbole uses exaggeration to convey a particular point. It would be misguided for someone to interpret a hyperbolic statement as literal. If you were to describe someone as extremely tall by saying, “That guy is like twenty feet tall,” most listeners would not think that the man is actually twenty feet tall. There is a difference between what a sentence means and what the speaker means by using that sentence in a particular way. Metaphors also demonstrate this point. If Romeo says, “Juliet is the sun,” it would be strange for someone to claim he has discovered a hidden semantic ambiguity in the word “sun.” We must not conflate the message conveyed in using language with what the words themselves literally mean. It is indeed of the essence of language that we can sometimes use words to mean what they do not in fact mean. This concludes our discussion of Donnellan, but not of Russell’s theory. Even though Donnellan’s critique of Russell seems misguided for the

94 

Chapter 4

reasons given, other objections to Russell’s theory persist. Let us quickly survey these objections. 4.7  Further Objections to Russell’s Theory The first objection is Strawson’s: empty descriptions give rise to statements that are neither true nor false. According to Russell’s theory, “The F is G” expresses an existential proposition—that there exists an F. If there is no F, the sentence expresses a false proposition. Strawson’s point is that Russell’s assignment of truth-value is intuitively wrong—it sounds more natural to say that the sentence fails to express a proposition that has a truth-value. We do not want to say that “The king of France is bald” is false when there is no such king. It could be false only if there were a king of France and he had a decent head of hair. Thus Strawson contends that when the description is empty, the statement is neither true nor false. Another type of example makes the criticism even clearer: “The golden mountain is golden.” This statement looks to be true a priori, but according to Russell’s theory it would be simply false, because of the lack of golden mountains. Such a statement does not seem to fit Russell’s theory at all. Russell might reply that this is just a matter of ordinary language—he has shown that contrary to appearances the statement is false. There is something to be said for Russell’s response. It is always possible to insist that sentences like “The golden mountain is golden” are strictly speaking false. We do not ordinarily say they are false, but they are false. Skepticism tries to show that nobody knows anything. According to skepticism, it would be false to say, “I know I am reading these words.” It seems fairly strange to say that sentence is false, but it is possible to argue that it is strictly speaking false. In the same way, with statements like “The golden mountain is golden,” we might insist that the statement really is false though to common sense it seems true. But still, Russell’s position does strike one as hard to accept and makes one wonder if his analysis is correct. The second objection is that “the golden mountain” and “the king of France” are phrases, not sentences. They are parts of sentences, not actual sentences. Grammatically, these phrases constitute the same part of speech as a name or demonstrative. If a speaker only says, “that dog,” or “Saul Kripke,” he has only uttered a fragment of a sentence and hence hasn’t said anything. According to Russell, however, descriptions are complete

Donnellan’s Distinction 

95

sentences because they expand into assertions of existence and uniqueness. If a speaker just says “the man outside,” we would think he has yet to express a complete proposition; but on Russell’s theory, he has already said that there exists a man outside and only one man outside. This seems strange since the speaker never actually completed a sentence. Note, in addition, that if we apply the description theory to names and then analyze the description Russell’s way, then simply uttering a name is already expressing a complete proposition—to the effect that an F uniquely exists. But did I say anything with truth-value just by saying “Eric Clapton”? Both these objections suggest that definite descriptions are more namelike than Russell allows. They are used as subject terms to identify something of which an attribute is predicated. Whether the statement is true or false depends on whether the thing identified with the descriptive term has the attributed property. The description looks more namelike than sentencelike. The description looks like part of a sentence—the subject part— not a whole sentence. Again, this makes us wonder if Russell has the correct analysis. Nonindicative sentences raise more worries about Russell’s theory. Consider an example of an imperative sentence: “Kill the King of France!” Using Russell’s theory, we would have to paraphrase this sentence as “Kill there is a unique king of France!” The first thing to be said about this paraphrase is that it is meaningless, ill formed, and ungrammatical. If the definite description is simply replaced with Russell’s paraphrase, the sentence comes out as nonsense. Russell’s theory cannot be applied mechanically in this case. Russell never discusses how he would handle cases where descriptions occur in imperatives. It doesn’t help to render the imperative as “Make it the case that the king of France is dead!” because then the imperative will order the addressee to make it the case that a unique king of France exists—which is rather contrary to the command to kill him. A closely related problem to the problem posed by imperatives is illustrated by the sentence “George IV wondered whether the author of Waverley smoked cigars.” Replacing the description with Russell’s paraphrase, we have it that George IV wondered whether there exists an author of Waverley, and only one author of Waverley, and whether he smoked cigars. But George IV may have never wondered if there existed an author of Waverley and only one such. He only wondered whether or not the author of Waverley smoked cigars—he took it for granted that said author exists. If

96 

Chapter 4

the definite description occurs in a propositional attitude context (in this case “wonder whether”), then we obtain the wrong analysis when applying Russell’s theory. So not all occurrences of descriptions fit Russell’s theory. A third objection stems from the fact that descriptions can function and yet be radically incomplete. Take the description “the table” and consider the sentence “The table is bare.” Now if we analyze this sentence according to Russell’s theory, there is a problem in the second conjunct, “There exists only one table.” The original statement surely did not imply that there is only one table in the world! If it did, it would be false. When incomplete descriptions are analyzed by Russell’s theory, the uniqueness clause is shown to be dramatically false. Certain maneuvers could help Russell escape these problems. Some people have suggested that a phrase such as “the table” is actually a demonstrative. Thus “The table is bare” means “That table is bare.” If we use this paraphrase, the problem of uniqueness falls away because the context singles out the object of reference. Such descriptions turn out to be demonstratives and so are not analyzed by Russell’s theory. But then we have admitted that not all descriptive phrases fall under Russell’s analysis. Demonstratives are singular referential devices that pick out an object; they are not quantifier phrases. Since some grammatical definite descriptions are not quantifier-like, Russell is wrong to have claimed that all definite descriptions are quantifier-like. Then we have degenerate namelike descriptions like “the Fonz,” “the Ace,” and “the Situation.” Presumably Russell would deny that these are descriptions at all, but they look like descriptions—and they are clearly namelike. What about “the GOP”? A final problem for Russell concerns “the former” and “the latter.” How do we analyze these as quantifier phrases? “Jack and Jill went up the hill and the former fell over while the latter sat down”: it is quite impossible to paraphrase these “the” phrases using Russell’s theory. Try it and see. Russell’s theory seems to contain a strong element of truth, but difficulties emerge if we try to apply it across the board. How to deal with these difficulties is an unsolved problem in the philosophy of language.

5  Kaplan on Demonstratives

5.1  Intension and Extension We have had occasion to mention demonstratives in our earlier explorations of names and descriptions, noting their basic role in linguistic reference. Now we will move on to focus on demonstratives explicitly, concentrating our discussion on the work of David Kaplan. But before doing that we must make a tour of possible world semantics. To introduce this topic, we can attend to an example of an ordinary contingently true sentence: (1)  Rafael Nadal was the number one tennis player in the world in 2010. The sentence is true. However, it might not have been true, because someone else could have been the number one tennis player in the world in that year (e.g., Roger Federer). If we consider all possible worlds, there are possible worlds in which Nadal is not the number one player. In a possible world where Federer is still number one in 2010, our sentence about Nadal is false. A contingent sentence can be true in the actual world but might be false in other possible worlds. Logicians and philosophers use a certain terminology to talk about contingent sentences and the possible worlds in which they have truth-value. The truth-value of a given contingent sentence at a world is called the extension of the sentence. The meaning of the sentence—the proposition it expresses—is called the intension of the sentence. For the one intension the sentence has in English in the actual world, there are variable extensions in respect to possible worlds. These ideas of intension and extension are analogous to Frege’s ideas of the sense of a sentence (a thought) and the reference of a sentence (a truth-value). The truth-value extension varies from world to world, while the intension remains the same.

98 

Chapter 5

Kaplan employs a slightly more theoretical way of explaining intension and extension. He characterizes the intension of a sentence as a function from possible worlds to truth-values. Thus intensions act like mathematical functions, taking worlds as arguments and giving truth-values as values. For example, in an addition equation like 2 + 3 = 5, 2 and 3 are the arguments for the addition function and the value of the function for those arguments is 5. In the same way, the value of the function that is the intension of “Nadal was the number one tennis player in the world in 2010” is True for the actual world as argument, but for other worlds as argument the value of the function is False. Sentence meanings are thought of as functions from worlds to truth-values. Intensions determine extensions with respect to worlds. In specifying the function a given sentence expresses from worlds to truth-values, we determine the truth conditions of the sentence. The truth conditions of a sentence are the set of worlds in which it is true: so our sample sentence is true in just those worlds in which Nadal is number one. A possible world semantics theorist explicates meanings as functions from worlds to truth-values, which is to say in terms of truth conditions. This idea can be extended to parts of sentences, such as definite descriptions. Take the following definite description: “the inventor of bifocals.” This description, like a whole sentence, has a particular intension and an extension, which is the reference of the description. In the actual world, Benjamin Franklin is the reference (extension) of that description. However, in a different possible world he might not be the extension, because he might not have invented bifocals and someone else did. The intension of the description determines a different object as extension in different worlds, just as the intension of the sentence determines different truth-values in different worlds. The meaning of the definite description is a function from worlds to extensions in the same way that the meaning of a sentence is a function from worlds to extensions. The difference lies in the fact that for a sentence, the extension is its truth-value, whereas for a description, the extension is an object. In the case of this particular definite description, the extension corresponding to the intension with respect to the actual world is Benjamin Franklin, but that same intension with respect to a different world could give Thomas Jefferson as the extension. The extension varies from world to world, while the intension stays fixed. This is a way of talking about contingency: it is only contingent that the inventor of bifocals was Benjamin Franklin.

Kaplan on Demonstratives 

99

But there is also necessity to consider. The sentence “2 + 2 = 4” expresses an intension that has the same extension with respect to every world, because the proposition is necessarily true. There are no worlds where 2 + 2 equals anything but 4. The function gives the same value as output regardless of what world goes into it as input. Any world you go to, you will see that 2 +2 = 4 in that world. The intension here is a constant function from worlds to truth-values, because it never varies in the output of the function from world to world. On the other hand, if we had written “2 + 2 = 5,” that would have the truth-value of being false in every world, because there is no world where 2 + 2 = 5. There are also cases in which definite descriptions are necessarily true of their bearers. We talked about one of these when we were discussing Kripke in chapter 2. For example, “the successor of 3” refers to only one number from world to world because in every possible world the successor of 3 has to be 4. To put it in Kripke’s vocabulary, that description is a rigid designator, because it has the same designation in every world. Using that terminology, we could say “Nadal is number one” is a non-rigid designator of the truth-value True, and “2 + 2 = 4” is a rigid designator of the truth-value True. Thus there are definite descriptions that are rigid designators, and they function in essentially the same way that the non-rigid ones do, that is, they are associated with intensions that operate as functions from worlds to extensions. The difference is that the rigid designators express constant functions, whereas the non-rigid designators express variable functions. Suppose we make a representation of the proposition expressed by a sentence containing a definite description. The thought expressed by the sentence, the proposition, will consist of the intensions of the various terms of the sentence. The intension for the description will be something like the concept of being F. So the component of the proposition corresponding to “the F” will be the concept of being uniquely F, and then there will be other components for other expressions in the sentence. Such a proposition would be in accordance with possible world semantics. The extension is determined by determining the object that uniquely satisfies the concept F in a world, which in our example will be Benjamin Franklin in the actual world. However, Benjamin Franklin is not a component of that proposition, only the concept F is; the man himself is a component of the world. The proposition is made up of concepts or intensions or senses, not references or extensions. References exist in the objective world, not

100 

Chapter 5

inside propositions—there is no room for them inside propositions. Propositions, for the Fregean possible worlds theorist, are made of intensions, not extensions. 5.2  Kaplan on Indexicals Kaplan disagrees with the picture of meaning painted by possible world semantics because of the presence of indexicals in language. He thinks indexicals must be analyzed in a different way and that a very different conception of meaning is needed to represent the meaning of indexicals. In the very beginning of the article, Kaplan introduces the idea of the semantics of direct reference: If there are such terms, then the proposition expressed by a sentence containing such a term would involve individuals directly rather than by way of the “individual concepts” or “manners of presentation” I had been taught to expect. Let us call such putative singular terms (if there are any) directly referential terms and such putative propositions (if there are any) singular propositions. Even if English contained no singular terms whose proper semantics was one of direct reference, could we determine to introduce such terms? And even if we had no directly referential terms and introduced none, is there a need or use for singular propositions?1

Kaplan defines a singular proposition in contrast to the traditional definition. A singular proposition will not contain the concept or intension corresponding to “Benjamin Franklin.” It will contain the actual man Benjamin Franklin. The real Benjamin Franklin is a constituent of the singular proposition in the same kind of way that a concept can be a constituent of a general proposition. This is very much opposed to the classic Fregean model, because there are now actual concrete individuals in the proposition. The notion is more in line with Russell’s view that certain terms (genuine names) introduce into the proposition the reference of the term. Russell made a sharp distinction between a term that introduces a concept (e.g., a description) and a term that introduces an object (e.g., a logically proper name). Kaplan is advocating a return to Russellian semantics as against Fregean semantics, because he thinks of a singular proposition as one that contains concrete individuals. If a directly referential term occurs in a sentence, the singular proposition contains the object of reference without the intermediary of a Fregean sense. Kaplan thinks that when it comes to indexicals, this view is the right one.

Kaplan on Demonstratives 

101

In the Fregean story, the word expresses the sense and that sense determines the reference, which is a particular individual. Therefore, when the word refers to the individual it does so indirectly by expressing the sense. The sense is the propositional component, the thing that enters into the proposition. The sense determines the reference by being the concept of a certain individual, but that individual is not a component of the proposition. As an indirect result of this relationship of expressing, the word denotes the individual. The direct reference story is very different. There is the word and the referring relation and the individual, and that’s all. The expressing relation and the sense, with sense determining reference, are now cut out of the story. Kaplan does bring in more linguistic machinery later, but the propositional component is constituted simply by the individual. The individual is the propositional component, which is why he writes that the relationship is identity. The individual thing referred to is literally identical to the propositional component. The word does not refer in a mediated way via the sense; it refers directly to the individual. The propositional component is the meaning, and so the meaning turns out to be an individual inhabiting the world outside language. One very big difference between the Fregean model and the direct reference model is that in the Fregean model many senses can correspond to the same reference. This cannot happen in Kaplan’s model, because the individual determines the sense, not the other way around. The propositional component is the meaning, which is determined by the reference, and that relation is simply identity. Therefore, there can only be one sense for each reference, so that coreferential terms must have the same sense. Kaplan’s model does not recognize Fregean cases with two terms having the same reference but different senses. However, as we have discussed several times, this account of the meaning of names runs into Frege’s problem of identity. Although the direct reference model is attractive in some ways, Frege thought that his apparatus of sense and reference is needed to solve the problem of identity. Kaplan does not attempt to confront Frege’s problem in this paper, concentrating on other questions, but we must keep it in mind as we proceed. On the face of it, it seems impossible to deal with cases like “Hesperus” and “Phosphorous” in terms of reference alone. At any rate, this is a challenge to direct reference theories. What is an indexical? Demonstratives are a subclass of indexicals. A demonstrative is a word like “that” or “this” that is typically accompanied

102 

Chapter 5

by a pointing gesture. Indexical words include ones like “here,” “there,” “you,” “he,” “I,” and “now.” The basic idea of an indexical is that it is a word that is used in a particular context and which depends for its reference on that context. Accordingly, we can call an indexical a context-dependent expression. Indexical words therefore differ from names and definite descriptions, even though some definite descriptions contain indexicals. Kaplan also makes the qualification that he does not mean to include indexicals that are used anaphorically, as in “John went to the shops, and he bought a sandwich there.” Instead, he is interested in indexicals as they are used without borrowing their reference from a previous singular reference (as with “he” and “John”). His concern is to understand the semantics of these words. The notion of direct reference will play a big part in this. 5.3  The Two Principles of Indexicals Kaplan tells us that two basic principles about indexicals will guide his discussion. First, indexicals are context dependent: the reference of an indexical depends on the context in which it is uttered. If Rafael Nadal says, “I’m hot,” he is referring to himself because the context of that utterance includes the speaker. If you, the reader, say, “I’m hot,” the context is a different one, and so it refers to you, the reader. Definite descriptions and proper names do not have this property of context dependence: if you say “Rafael Nadal,” you refer to the same person that Nadal does when he says that name—you don’t refer to yourself! The second principle is that indexicals are directly referential. A directly referential term is one in which the proposition expressed by the indexical sentence is a singular proposition. If a speaker says, “I’m hot,” the proposition expressed by that sentence will consist of the speaker (the person “I” refers to) and the property of being hot. Kaplan thinks that indexicals are directly referential in the way that Russell and Mill thought that names are directly referential. The reference is not mediated by a descriptive concept that identifies an object uniquely. Kaplan’s view of indexicals is rather like Kripke’s view of names: both go against description theories of what determines the reference of these expressions. Kaplan thinks both names and indexicals are directly referential. So semantically, indexicals are namelike in the Russellian sense. Since names are rigid designators, it would follow that indexicals are also rigid

Kaplan on Demonstratives 

103

designators. Kaplan does think that indexicals are rigid designators. However, he thinks using that terminology confuses two distinct notions that he believes need to be kept separate. A description that is a rigid designator does not differ semantically from a description that is a non-rigid designator. It is not directly referential. The propositional component is the same as it was before—a concept. The component of the proposition expressed by the rigid designator “the successor of 3” will be the concept of the successor of 3. It will not be the number 4 itself. In the case of the rigid description, the propositional component is a concept (not an individual), so a rigid description is not a directly referential device. It does not create a singular proposition, but rather creates a general proposition. Its components consist of a general concept (the meaning of the description) and whatever is predicated. This becomes clearer if we consider Kripke’s necessity of origin example. Consider a person with origin O. The propositional component corresponding to “the person with origin O” is just the general concept of having origin O. Semantically, the description functions the way it does when the description is not rigid—the propositional component is a general concept. Therefore, it does not follow from the fact that the expression is a rigid designator that it is also directly referential. Descriptions can be rigid without being namelike. Another passage from Kaplan’s text explains this point: For me, the intuitive idea is not that of an expression which turns out to designate the same object in all possible circumstances, but an expression whose semantic rules provide directly that the referent in all possible circumstances is fixed to be the actual referent. In typical cases the semantic rules will do this only implicitly, by providing a way of determining the actual referent and no way of determining any other 2 propositional component.

Kaplan’s idea of direct reference is not the idea that the term designates the same thing in all possible circumstances. Rigid designation can arise from individual essence independently of the rules of language. It can result from the facts of metaphysics. Origins are metaphysically necessary, but that is not a semantic idea. Rather, individual essence is something that comes from the nature of numbers and the nature of people. Direct reference is meant to be a property of an expression that results from its status as a piece of language. The semantic rules that are part of the very meaning of the expression determine that it is directly referential.

104 

Chapter 5

Kripke uses some terminology in Naming and Necessity that is relevant to our present discussion. A de facto rigid designator is said to be one that designates the same object in every possible world as a matter of metaphysical fact (e.g., “the successor of 3” or “the person with origin O”). On the other hand, a de jure rigid designator is one that designates the same object in every possible world because of its meaning or the semantic rules that govern it. For Kripke, names are de jure rigid designators, but rigid descriptions are de facto rigid designators. Kaplan believes in a similar distinction between rigidity and direct reference. Rigidity is not the same notion as direct reference because there are rigid descriptions without direct reference. Here is Kaplan again: If I may wax metaphysical in order to fix an image, let us think of the vehicles of evaluation—the what-is-said in a given context—as propositions. Don’t think of propositions as sets of possible worlds, but rather as structured entities looking something like the sentences which express them. For each occurrence of a singular term in a sentence there will be a corresponding constituent in the proposition expressed. The constituent of the proposition determines, for each circumstance of evaluation, the object relevant to evaluating the proposition in that circumstance. In general, the constituent of the proposition will be some sort of complex, constructed from various attributes by logical composition. But in the case of a singular term, which is directly referential, the constituent of the proposition is just the object itself. Thus it is that it does not just turn out that the constituent determines the same object in every circumstance, the constituent (corresponding to a rigid designator) just is the object. There is no determining to do at all.3

This passage clearly distinguishes between rigidity and direct reference. The proposition that corresponds to a directly referential term is a singular proposition. The proposition that corresponds to a rigid description is a general proposition, because descriptions are not directly referential. The terminology Kaplan uses is similar to Russell’s. Russell would say that a sentence that contains a definite description expresses a general proposition because it is equivalent to a quantifier sentence. The general proposition that is expressed by that sentence may look like a singular proposition, because it is a grammatically singular sentence; but that is a grammatical illusion, because it is logically a general proposition. But there is also a class of expressions Russell calls names (Kaplan calls them directly referential), where the proposition expressed is a singular proposition instead of a general proposition. The idea of singularity of propositions is captured in the representation of propositions as containing individual things as

Kaplan on Demonstratives 

105

constituents. Rigidity is just the idea of having the same reference in every world, but direct reference is the idea of what constitutes the corresponding proposition. Rigidity is a modal notion, but direct reference is a semantic notion. If we consider this issue from the perspective of the speaker, we can ask what he understands when he grasps propositions of different types. In the case of descriptions, rigid or non-rigid, the speaker is grasping something general constituted by concepts. In the case of the directly referential term, though, the speaker grasps an individual, and that individual occurs in the very content of the proposition that has been grasped. If a speaker says, “This room is nice,” the proposition before his mind at that moment contains a certain actual room. There is a sense in which that room is a part of his mind, a part of the proposition he grasps. One consequence of this is that if there is no such room (e.g., he is hallucinating), then there is no such proposition. Since the speaker has used a demonstrative, he has directly referred (ostensibly) to a room that does not exist. There is then no singular proposition that he has succeeded in expressing. Thus it is possible to think that one is expressing a singular proposition when one is not really expressing such a proposition—as when one hallucinates a particular object and says, “That is F.” For instance, you might hallucinate a tiger and say, “That tiger looks fierce.” But since there is no tiger there, you have failed to express a proposition containing a particular existent tiger. Singular propositions are object dependent, so they fail to exist if the intended object fails to exist. Direct reference thus can give rise to illusions of propositions. But that can’t happen in the case of purely general propositions. 5.4  Context of Use and Conditions of Evaluation To further distinguish rigid designation from direct reference, Kaplan outlines the distinction between context of use and conditions of evaluation. This is a very important distinction. Context of use consists of the person, the time, and the place in which a given sentence can be uttered. A circumstance of evaluation is a possible world where a given proposition can be either true or false. We have to distinguish quite clearly between the two concepts. The reason we might fail to see the distinction is that different contexts of use produce different references. When I say “I,” I refer to me, and when you say “I,” you refer to you. Therefore, different contexts of

106 

Chapter 5

use with the same indexical term produce different references. Accordingly, they can produce different truth-values, because I might be what I say I am while you might not be what you say you are. We may wonder if this is the same as what happens in the case of a description (e.g., “the inventor of bifocals”) with different references in different possible worlds. Don’t we have variation in extension with constancy of intension in both cases? Kaplan’s point is that we should not conflate those two kinds of dependence of extension. We should not confuse dependence on context and dependence on world. Consider the sentence “I do not exist.” If a speaker says, “I do not exist,” that can never be truly uttered, because it can’t be uttered by someone without that someone existing. Take any context of use and it will always be false, because the context includes the speaker. In any context in which somebody says, “I exist,” it is always true (cf. Descartes’s point in the Cogito). It is necessarily true in the sense that in any context in which the sentence is uttered it must be true. However, it is not a necessarily true proposition that the speaker who uttered the sentence “I exist” exists. Along with anyone else who utters that sentence, he might not have been born. There are possible worlds in which that speaker would not be alive to utter the sentence “I exist.” No one exists necessarily (except perhaps God, if he does exist). Therefore, there is a big difference between the context of use and the circumstances of evaluation. The circumstances of evaluation concern the extension of the proposition expressed once it has been expressed, but the context of use concerns which proposition gets expressed to start with. Thus context determines which proposition is expressed by a use of “I,” but circumstance determines whether a particular proposition so expressed is true in a world. For this reason, Kaplan makes a firm distinction between context of use and circumstances of evaluation. And the first point that Kaplan makes against possible world semantics is that it blurs that distinction. It does not recognize the difference between circumstances of evaluation and contexts of use because it just talks about descriptions and intensions in relation to possible worlds. All we have in possible world semantics are circumstances of evaluation, where different circumstances give different extensions for a given intension. It does not have the notion of a context of use. It deals with the modal notion of variation of extension with possible circumstances, not the notion of a context that fixes what was said on an occasion. In effect, it treats all of language as context independent (this is not

Kaplan on Demonstratives 

107

surprising, given that it deals with languages modeled on standard formal logic and these languages contain no indexicals). This discussion of context dependence leads up to a distinction that Kaplan draws between what he calls character and content. This is the heart of his theory. All the points we have just made can be formulated by means of these notions of character and content. Fortunately, this distinction is easier to understand than some of the earlier points Kaplan makes. Consider a word like “I” or “here” or “now” and look at its meaning. The meaning such words have whenever they are uttered is called “character.” It is what the word means in the language—its lexical meaning. Roughly, this meaning or character specifies that if one utters the word “I,” it refers to the speaker, whoever he or she is. “Here” is the word that you use to refer to the place you are in, wherever that place is. Similar definitions hold of “there” and “now.” Character captures the meaning of those indexical expressions, because it determines what is referred to by those expressions when they are uttered in a particular context. Essentially, character is the dictionary meaning of the word. It is important to note that the word has the same character no matter what context it is used in. If Jack says the word “I” and John says the word “I,” there are two different contexts of utterance, but the word “I” has the same meaning in both and thus the same character. Character seems close to the Fregean sense of a word, because the sense of a word corresponds to its linguistic meaning. But there is a big difference between character and Fregean sense. Character does not by itself determine reference, whereas Fregean sense does. Character does not determine reference because when John says “I” and Jack says “I” they utter the word with the same character but not the same reference. Therefore, the meaning of an indexical is not a sense in the Fregean understanding of the term. The context in which an indexical is used also works to determine its reference. It cannot be done by the character alone. Obviously, a speaker cannot say the word “I” and succeed in referring to a certain place. He must use the word with the correct linguistic meaning. But the character is too general and unspecific to tie down a unique reference without contextual supplementation. Consequently, both character and context determine reference. Both of those factors operating together fix what the speaker refers to. Character, then, is very different from sense. In the case of sense, sense determines reference, and there is no need to bring in the context of use. With Frege, we learned that sense determines reference regardless of the

108 

Chapter 5

context of use. Unlike sense, however, character requires interaction with the context of use to determine reference. The full meaning of an indexical utterance cannot consist of its character alone; if it did, the full meaning of a sentence would not determine the proposition it expresses. The proposition that is expressed is something separate from the character. Kaplan calls the proposition expressed by the sentence its content. If I say, “I’m hot,” and you say, “I’m hot,” we express different contents, because we are talking about different people. The sentence we both uttered has exactly the same character, because the same character is expressed by the sentence no matter what the context is. But a different propositional content is expressed by the sentence in the two contexts. The content is a product of both the character and the context. Content includes reference, but character does not. Content is what has truth-value in different possible worlds, but character is what interacts with context to produce content—character by itself cannot have truth-value. Another reason why the content is separate from the character is that the same content can be expressed by a sentence with a different character. An utterance of the sentence “I’m hot” expresses a content with a certain character, but the same content could be expressed by someone who utters the sentence “you are hot” in reference to the person who uttered the first sentence. In those two utterances, there is the same proposition and the same content, but a different character. Therefore, character does not determine content and content does not determine character. They are two independent semantic dimensions of an indexical utterance. Thus the total meaning of an indexical utterance has two parts or aspects: the character and the content. There is no single straightforward entity called “meaning” or “sense” because the indexical utterance has two different semantic dimensions. In Kaplan’s picture, indexicals have two sides to their meaning, whereas in Frege’s picture there is only one side, the Fregean sense. The reason for this is that Fregean sense is supposed to determine reference. However, in the case of indexicals, their ordinary lexical meaning does not determine their reference, because their reference is context dependent. Context dependence is the central pillar of Kaplan’s theory of indexicals. All the other aspects of his theory stem from that one major point. Kaplan is saying that Frege is wrong to suppose that the linguistic meaning of an expression is always a reference-determining sense. Frege’s theory works

Kaplan on Demonstratives 

109

perfectly when applied to context-independent definite descriptions. The thing that determines the reference of a definite description is the very thing that constitutes the linguistic meaning of it. But in the case of indexicals, they do not coincide. Fregean sense, and its descendant, possible world intensions, cannot accommodate indexical expressions. These are modeled on the case of the pure definite description, but indexical terms are nothing like pure descriptions. They are directly referential and context dependent, whereas descriptions are neither of these things. 5.5  Possible Worlds, Meaning, and Indexicals Consider the two sentences “The queen of England is pregnant” and “I am pregnant.” To better understand the semantics of these two sentences, imagine that the second sentence is uttered by Queen Elizabeth II. She refers to herself with the word “I,” and she is also the denotation of “the queen of England,” so we have coincidence of reference. We have already talked about the many reasons why these two sentences are not synonymous. Now we are interested in what Kaplan believes to be the essential difference between the two sentences. The first sentence expresses a meaning and that meaning is an intension. The intension is a function from possible worlds to truth-values. If we consider just the definite description, it will express a function from possible worlds to objects. In the actual world, that function gives us the object Queen Elizabeth II. But in other possible worlds the description might designate a different individual. It is not necessarily the case that Elizabeth II is England’s present queen. Since “the queen of England” is a non-rigid designator, the intension corresponding to the meaning of that description will determine a different object in different possible worlds. Notice that this description is completely context independent. It does not matter in what context it is said; it will have the same reference. All that matters is that the intension is something that determines a certain object when given a certain possible world as argument. To use Kaplan’s terminology, certain circumstances of evaluation fix which object that description refers to, and these can vary. Kaplan is arguing that this model applies only to certain types of expressions. Indexicals are a class of words to which this model does not apply. If we return to our example above, Kaplan believes that the description “the queen of England” is a non-rigid designator that does not directly

110 

Chapter 5

refer. The propositional component corresponding to the description is an individual concept, not a particular object (an actual thing in the world). It is not directly referential (in the Russellian sense). Kaplan suggests that indexicals cannot express such intensions, which are context independent, so their meaning cannot be understood as functions from possible worlds to extensions. The meaning of the sentence “I am pregnant” is a character (in the technical sense that Kaplan gives to “character”). Character is not an intension from possible worlds to extensions—it is not something that can be applied to a world to determine what the extension is of that term in the world. The meaning of the word “I,” for example, is common to everybody who uses the word “I,” so it is impossible to look into a possible world and determine what the reference of the word “I” will be in that world. It has none, considered out of context. Character is nothing like a classic intension in possible world semantics. The sentence “I am pregnant” considered by itself does not express a proposition at all. A proposition must be something that is true or false. That sentence by itself is not true or false but must be uttered in a context first. If a man says, “I am pregnant,” that would certainly not be true. If a woman who is pregnant says that sentence, then it is true. The character alone fails to determine a proposition. The character is thus not a function from worlds to extensions. An indexical sentence can of course express a proposition on an occasion, but the context must be added to the character to produce anything propositional. The combination of the character and context determines the proposition. Kaplan gives us the equation: Character + Context = Content Content is what is said, stated, asserted—and that is a proposition. The content is not the same as the character. It is something produced by the character when it is combined with the context. The content is what the speaker states when he uses the particular sentence in a specific context. This content corresponds to the classic notion of intension, but the character that produces it does not. Rather, character is best conceived as a function from context to content. The function here is not from worlds to truth-values. Rather, it is something that expresses a relationship between the context and what is said when the expression is uttered. Character determines (with context) what you say; it does not determine whether what you say is true—that depends on the circumstance of evaluation. In the case of

Kaplan on Demonstratives 

111

character, the function takes contexts as arguments and produces contents as values—whereas contents are functions that take worlds as arguments and produce truth-values as values. So two very different functions are involved in an indexical utterance, and Kaplan’s entire point in his paper is that we should not confuse these two functions. In one case (“The queen of England is pregnant”), a fixed intension for the description combines with different circumstances to give a particular extension (e.g., whoever “the queen of England” refers to in a particular world). In the other case (“I am pregnant”), there is no fixed intension, and the reference of “I” can vary as different propositions are expressed in different contexts. We must not confuse the way context contributes to extension with the way circumstance contributes to extension. Definite descriptions like “the queen of England” are context independent, but indexicals like “I” are context dependent. Therefore, what is said when using an indexical depends on the context, but this is not so for descriptions. Descriptions float free of context, but indexicals are steeped in context. A number of consequences flow from the distinction between character and content. One is that not all meanings are intensions. There cannot then be a complete theory of meaning based on classic possible world semantics. There are two kinds of lexical meaning: character-type meaning and content-type meaning. There is only one type of meaning in the classic intension-based semantic theory—Fregean sense. But according to Kaplan, there are two irreducibly distinct types of meanings. Thus the meaning of an utterance of the sentence “I am pregnant” is given in two stages. One stage gives the character, which is a function from contexts to contents, and the other stage gives the content, which is a function from worlds to truth-values. This type of theory is sometimes called dual-aspect semantics. It is a rejection of the one-dimensional Fregean picture of things. Frege did not consider indexicals when writing “On Sense and Reference.” In a later essay called “The Thought,” Frege does discuss indexicals and picks up on some of these issues. But he did not originally design the theory of sense and reference in “On Sense and Reference” by looking carefully at indexicals. He was mainly interested in mathematical language, which is a context-independent language. Hence his examples are all context-independent names and descriptions, for which a one-dimensional semantics is adequate.

112 

Chapter 5

There are now two kinds of semantic compositionality, as Kaplan points out. The two ways that the meaning of a complex expression can depend on its parts are through character compositionality and content compositionality. An example will illustrate this point. If the queen of England says, “I’m pregnant,” and another speaker says, “She’s pregnant,” the indexical has been changed. The character of “I am pregnant” is different from the character of “She is pregnant.” However, the content is the same, so the content of the whole thing, the proposition expressed by it, does not depend on the specific character of the words. Here there is the same content but different character, but there are also cases where the same character can have different content. The two are not connected to each other in any simple way, at least not in the way Frege had supposed. There are two sorts of compositionality because there are two different levels of meaning. Different sorts of semantic unit are combined together to produce complex expressions. A terminological issue arises here: one might assume that the Fregean theory of meaning is two leveled. Relative to Russell’s theory, Frege’s theory is two leveled, because Russell thinks there is only one level, the level of reference. Russell handles everything concerning meaning beyond the simple level of name reference with the theory of descriptions. To him, every primitive expression means what it does by virtue of denoting something. In Russell’s system, predicate expressions denote universals (e.g., the predicate “red” denotes the universal of redness). Russellian semantics is onedimensional because ultimately there are only references. In Frege’s view, there is the sense and the reference, so it seems right to suppose that his theory is two leveled. However, such an assumption is ill founded, because in Frege’s view reference is not constitutive of meaning. In Frege’s theory, the sense is the meaning, and only the sense. Reference is outside of meaning, which is why words can be meaningful even if there is no reference. Although Frege’s theory recognizes a level of meaning above reference, his theory of meaning is still one-dimensional, because sense does all the work. Kaplan’s theory can be characterized as having two levels or three levels, depending on how each level is understood. Kaplan’s theory of meaning has two levels—character and content—and both of these correspond to the intuitive idea of what somebody meant when he uttered a sentence; but there is also the level of reference. So we might speak of three levels in the same spirit as thinking of Frege’s theory as having two levels. What is important is that Kaplan breaks Fregean sense into two and hence introduces an extra semantic level.

Kaplan on Demonstratives 

113

5.6  Kaplan on “Today” and “Yesterday” Finally, Kaplan talks a bit about the words “today” and “yesterday.” This discussion raises a tricky problem for Kaplan in the end. Suppose I say on a certain day, “Today it’s raining.” How do I say the same thing as I said today tomorrow? Suppose I say tomorrow “Today it’s raining”—have I said the same thing as I said the day before when I said, “Today it’s raining”? Suppose the first day was Tuesday: then the first use of “today” referred to Tuesday and the second to Wednesday. So I have not said the same thing. I have referred to Tuesday in the first case and Wednesday in the second case. The same indexical word cannot refer to the same day on consecutive days. To say the same thing on Wednesday as I said on Tuesday, I have to say, “Yesterday it was raining.” Clearly the words “today” and “yesterday” are not synonyms of one another. They have different meanings even while referring to the same thing. And yet, in a very intuitive sense, these two sentences manage to say the same thing. They do not say the same thing in the sense that they have the same linguistic meaning, because the sentence “Today it’s raining” and the sentence “Yesterday it was raining” do not have the same linguistic meaning. However, each can say the same thing as the other depending on the speaker’s context. So, in Kaplan’s terminology, two sentences with different character can say the same thing. What makes this thing said the same? Kaplan might suggest that it is the identical reference of the two terms. But as we have repeatedly seen, just because the reference of two terms is the same does not mean they have the same propositional content. We know, for example, from “Hesperus” and “Phosphorus” that these names do not say the same thing. If somebody says, “Hesperus is a planet,” it would be wrong to report him as having said that Phosphorus is a planet. But in the case of indexicals for days, it is necessary to use a word (“yesterday”) that has a different meaning from the first word used (“today”) in order to say the same thing. We have to change the meaning to keep what is said the same! Something strange is going on because the meaning of the word is being pulled apart quite radically from what is said by using the word. The question is whether Kaplan has the resources to capture this idea of what is said: is it character or is it content? It can’t be character because the characters are different; but how can it be content if content is just a matter of reference? We will go into this in more detail in the next chapter.

6  Evans on Understanding Demonstratives

6.1  The Fregean Theory of Indexicals Kaplan takes indexicals to refute Frege’s theory of meaning, at least for their case. In particular, the Fregean notion of sense does not apply to indexicals. Gareth Evans, however, questions this conclusion, arguing that it is possible to develop a Fregean interpretation of indexicals. In such a theory, we would be able to apply the theory of sense and reference to indexicals. We know that it is not possible to do this by equating the sense of the indexical with the conventional linguistic meaning (character) of the indexical, because then sense would not determine reference. Different people can use the same indexical word with the same meaning and yet make different references. Sense cannot be identified with the standard conventional meaning of an indexical word if we are to create a theory of indexicals where sense determines reference. To establish a Fregean theory of indexicals, we must find a sense for them that goes beyond their conventional meaning, that is, their Kaplanian character. What would this sense be like? It can’t be character, but can it be content? No, the sense can’t be identical to the content either, in Kaplan’s sense, because in Frege’s system senses are never identical to references and there are always many senses corresponding to the same reference. Content for Kaplan is just a singular proposition, constituted by reference alone. Therefore, it would be impossible for the sense to be identical to the reference, because then there would be only one sense for a given reference. The sense of the indexical as uttered by someone will be identical neither to its character nor to its content. And there does not seem to be anything left in Kaplan’s scheme for Evans to equate with Fregean sense.

116 

Chapter 6

A possible answer is that the sense of an indexical is neither the character nor the content but rather the description a speaker has in mind when she uses an indexical. This suggestion borrows from the description theory of names. When a name is used it is held to be synonymous with a description that the speaker has in mind that uniquely applies to the bearer of the name. Analogously, we might propose a description theory for indexicals, suggesting that when a speaker uses an indexical, she has a description in mind that is synonymous with the indexical as then used, and that this description uniquely applies to the object of reference. Suppose I say, “I’m a philosopher.” Then we might suggest that the description I have in mind is “the author of The Subjective View,” since I wrote that book. So on this Fregean description theory of indexicals, when I use the word “I” the sense of it is expressed by “the author of The Subjective View.” When you, the reader, use the word “I” you have a description in mind that uniquely applies you, and thus you refer to yourself with “I” in virtue of this mediating description. Just as with the description theory of names, the proposition expressed by a sentence of the form “I am F” would be represented using a general concept expressed by a particular definite description. This indexical sense would function just like a classical intension in possible world semantics. We might go on to apply Russell’s theory of descriptions to the description associated with the indexical, thereby combining Frege’s view with Russell’s. We then have a description theory of the meaning of individual occurrences of the word “I” that takes these occurrences to be equivalent to quantified propositions of the Russellian form. When I say, “I’m a philosopher,” what I’m saying is “There exists an author of The Subjective View and there is only one such and he is a philosopher.” There is no Kaplanian direct reference in this paraphrase, just quantifiers and predicates. Evans uses some terminology that might not be familiar to some readers. The word “I” as uttered on a particular occasion is called a token of the word. The word “I” that is common to all tokens of it is called the word type. You use the same word type when you say “I” as when I say “I,” but you utter a different token of that type. If I say “I” at a given time, that is a different token from my saying “I” at a later time. Nevertheless, each utterance consists of a token of the same type. Tokens are events that occur at particular times and places, but types are more abstract. The Fregean theory of indexicals claims that we should analyze tokens of indexicals as expressing

Evans on Understanding Demonstratives 

117

Fregean senses, and then equate each of those tokens with descriptions (at least according to this first style of Fregean theory). The description might be constant from token to token, as it is with token utterances of names. However, we are trying to accommodate the point that someone else can use the word “I” and refer to a different person, not me, so we will need a different description with a different reference. We can conclude from this that the word “I” is ambiguous, according to this theory, because it has different senses on different occasions. It would be rather like having a room full of people all named “John Smith.” No John Smith would be identical to any other—hence “John Smith” would have variable sense and reference across this room of people. The name “John Smith” in that case would be ambiguous. Similarly, “I” would be ambiguous, having different sense and reference in different contexts. The type is ambiguous, that is, though the tokens would all have a specific sense and reference. A definite description for each of them would give the sense of the token, but as a type the word would be ambiguous. That is one possible idea for how to handle indexicals Frege-style—by proposing a description theory of the sense of indexical tokens. The semantics of indexicals would consist of three elements: character, content, and a description that captures the sense on a particular occasion of utterance (the token sense). In this picture, indexicals are not directly referential. The word is synonymous with a description, and the description has an intension that is context independent. The role of context is just that different individuals use the same (type) word and they associate different descriptions with it, with these descriptions determining what they refer to. It is necessary here to distinguish between the descriptive sense and the character. The word has the same conventional meaning (character) in its different uses, but the sense varies from context to context. So it’s not that once sense is admitted we can do without character: we will have character, sense, and reference in our final semantic theory. The author that Evans is mainly criticizing, John Perry, assumes that the theory we have just outlined is the correct Fregean model, because he thought it must be some kind of description theory of sense. Evans’s reply is that Perry has overlooked a different kind of Fregean theory, one not built around definite descriptions. There could be other ways to think about sense than descriptively, he believes, and these other ways are equally Fregean. Not all sense has to be descriptive sense, he contends. He agrees

118 

Chapter 6

with Perry that a description theory of indexical sense is a very implausible idea. It does not seem at all attractive to suppose that people have uniquely identifying descriptions in mind when they use these terms. It is unappealing to think that context plays no substantive reference-determining role. Perry gives us a very neat argument against this type of position, as follows. 6.2  The Point of Indexicality The point and essence of indexicality is best understood by considering two types of examples: mirror examples and amnesia examples. Let us first examine mirror examples. Suppose you are sitting in a restaurant and you see the reflection of a man or woman in the mirror in front of you, and you have the following thought about the individual in the mirror: “That person is very good-looking.” You could have certain other beliefs about the person in the mirror too—say, that he (or she) seems rather pleased with himself (herself). Though this is improbable, it is perfectly conceivable that the person in the mirror is you, but you didn’t realize it for a second or two. Suddenly, you are thunderstruck with the realization, “Oh, it’s me I’m seeing!” You had referred to yourself without realizing it. This tells us that when you refer to yourself with “I” it cannot be via the kinds of descriptions that truly apply to you in the mirror reflection, because then you would have to realize the truth of “I am the person reflected in the mirror.” The word “I” cannot mean these descriptions. It is informative to discover the truth of “I am the person in the mirror,” so this cannot be a tautology, which it would if “I” (that token) were synonymous with “the person in the mirror.” Almost any description is such that it is potentially a discovery that you are the person so described. Another, more extreme example that makes this point even more clearly is the amnesia case. Imagine a person who sustains a trauma to the head, and when he wakes up can’t remember anything. I am going to suppose that this unfortunate individual is myself. The doctor asks me, “Where do you live?” and “What’s your name?” and I don’t know because I can’t remember. I can’t remember any facts about myself. I might say, “I can’t remember anything about myself”—yet I successfully refer to myself. So there I am in the hospital and I don’t know about my past history, and I start reading a book called The Subjective View. As I read I say to myself, “The author of The Subjective View is not much of a philosopher.” I report

Evans on Understanding Demonstratives 

119

my opinion to the doctor, who smiles indulgently and replies, “You are the author of The Subjective View.” I thereby make a substantial discovery, thus showing that “I” in my mouth does not mean “the author of The Subjective View.” And we could have anticipated that because I succeed in referring to myself with “I” even though I have amnesia. So it can’t be that I succeed in making this kind of first-person reference in virtue of knowing true descriptions about myself. I certainly don’t refer to myself with “I” by knowing my famous deeds and well-known facts about myself. Perry gives this argument, and Evans agrees with him. The upshot is sometimes called the indispensability of the indexical “I” or the essential indexical. The idea is that the word “I” cannot be removed from the language and replaced by descriptions, because indexical sentences express a different kind of proposition from nonindexical sentences (e.g., sentences involving the descriptions we used in the mirror and amnesia examples). This poses a serious problem for the description theory version of a Fregean account of the meaning of indexicals. Evans agrees that descriptions don’t work to give the meaning of an indexical because of this kind of argument. If indexicals have sense, it cannot be descriptive sense. But what other kind is there? 6.3  Evans’s Theory of Sense and Reference for Indexicals Since Evans agrees with this point, we may wonder how it would be at all possible to have a Fregean theory of the meaning of indexicals. There seems to be nothing else the sense could be except some sort of descriptive concept. We have already explained how the sense of an indexical can be neither a character nor a reference, and now we see it cannot be a description either. To approach the question, Evans tells us what he thinks a theory of meaning should look like. That is, he tells us how sense is related to reference. He spends the first half of his paper talking about this relationship. First, we will examine his conception of a theory of reference, then outline his theory of sense, and finally explain how he thinks the two are related. Then we can discuss whether or not this overall theory applies to indexicals. First of all, a semantic theory is founded on a theory of reference. A theory of reference is an assignment of reference to every meaningful expression in a language. And we know that Frege’s position on assigning reference has two main parts. One part is that if the expression is a proper name, it will be assigned an object as reference. Proper names, for Frege,

120 

Chapter 6

can be ordinary names or definite descriptions or even whole sentences. Ordinary objects are assigned to ordinary singular terms and truth-values are assigned to sentences as their reference. In the second part of the theory, Frege also assigned concepts to predicate expressions. In Frege’s system, a concept is a function from objects to truth-values. In the sentence “Socrates is a man” the concept corresponds to the word “man” and the argument is the reference of “Socrates.” When you apply that concept to the argument, the value of the function for that argument is the True (an object for Frege). The value of the function would have been the False if we had inserted the argument Cleopatra into the function, because Cleopatra is not a man. A truth function is a function from truth-values to truth-values. Connectives and predicates are logically the same because they both map objects into truth-values. Since truth-values are objects, they function as arguments for functions into truth-values. Thus in Frege’s system there is an assignment of objects to complete singular terms, where complete singular terms can be proper names, definite description, and whole sentences; but there is also an assignment of reference to incomplete expressions, like predicates or sentence connectives, which are assigned concepts. Quantifier expressions are all that is left, and these are classified as denoting second-level concepts, since they map first-level concepts onto truth-values. The general point is that the theory of reference in the Fregean model is an assignment to every expression in the language of a semantic value that is its reference. The notion of reference is taken very broadly. It is correlative with the truth conditions of a sentence. But Frege’s system is meant to be a theory of a speaker’s understanding, not merely of the truth conditions of his sentences. A theory of sense is then needed to account for how we apprehend references. It is a theory of how references come before the mind, and how they are represented in the mind. A sense, as Frege tells us, is a mode of presentation, and the mode of presentation is the relationship between an object in the world and the person who makes reference—it is the mode in which the object is presented to the person’s mind. The way Evans explains this idea is that a sense is a “way of thinking” of a reference: not so much how it presents itself to me, but how I think of it—how it enters my thoughts. Evans’s point in regard to this specific part of Frege’s theory of sense is that it has not stipulated anything to the effect that senses must be descriptions. We have just stated in a very abstract way the idea that senses are ways

Evans on Understanding Demonstratives 

121

we have of apprehending things. Whether or not these ways are descriptions is a completely different question. What is built into the notion of sense is just that the sense is something that presents the reference. The next question is how to specify what the sense is. From our exploration of Frege’s work, we know that senses are different from references, but we have not established how to specify them. Frege himself says little about this question. Fregean senses seem rather elusive in and of themselves (can you point to them, stub your toe on them, examine them from different angles?). Evans’s view is that the sense of an expression is specified by saying what the reference of the expression is. Suppose we want to give the sense of the word “Hesperus.” Evans thinks we can give the sense of this word by saying “The reference of ‘Hesperus’ = Hesperus.” This certainly gives the reference of the name, and certainly the statement is true. Compare that sentence with the following sentence: “The reference of ‘Hesperus’ is Phosphorus.” Is that sentence true or not? Yes, that sentence is also true, since Hesperus is Phosphorus. Evans’s claim is that these two sentences both correctly say what the reference is of “Hesperus” but only one also specifies the sense. The sentence “The reference of ‘Hesperus’ = Hesperus” specifies the sense, while the sentence “The reference of ‘Hesperus’ is Phosphorus” does not—though both state the reference. In the first sentence we have an example of what Evans calls a sense-specifying reference assignment. The sense is given by stating the reference, but only some statements of reference succeed in giving the sense. Evans’s idea is that we can specify the sense of a name by saying what its reference is, as long as we do it using the right kind of ascription of reference. In the second sentence, we have stated the reference but have not specified the sense. Though he does not explicitly say so, the right way to state reference if we want to specify sense is by using a synonym of the name we are talking about. The reference can be stated in two different ways, by using a name with the same sense as the name mentioned or by using a name with a different sense, that is, by using a synonymous name or a nonsynonymous one. In the first way, the sense has been specified, but in the second way it has not. Evans’s position, then, is that senses can be specified only by assigning references, but not all ways of assigning reference convey sense. Here we have said nothing about senses as descriptive concepts. A sense is a way of thinking of an object, but there is no way to specify a sense except by talking about the object.

122 

Chapter 6

Notice that in this way of formulating specifications of sense, nothing is said like “The sense of ‘Hesperus’ is so-and-so.” In specifying what the sense is, we must say what the reference is—there is no way to specify sense “directly.” We don’t speak about senses in specifying them. If we say, “The reference of ‘Hesperus’ is Hesperus,” intending to convey the name’s sense, nothing has been said directly about the sense of “Hesperus” itself. This is different from saying that the sense of the word “bachelor” is given by the sense of the words “unmarried male.” In Evans’s theory, the sense of a word is not specified by giving the sense of another word. At this point Evans invokes a suggestion of Dummett’s that involves using a distinction of Wittgenstein’s—the distinction between “saying” and “showing.” In Wittgenstein this distinction is a matter of obscurity, and we will not cover it here in detail. Basically, there is an intuitive idea of saying versus showing that we can illustrate in the following example. 6.4  Saying versus Showing Imagine someone holding a pen behind his back. He can say, “I have a pen in my hand,” or he can just reveal his hand and show you the pen. Either way you come to know that he has a pen in his hand. In the showing gesture he does not say anything about a pen—he just shows it to you. As a result of showing the pen, the observer gains knowledge without the mediation of language. Evans is using Wittgenstein’s general intuitive idea of saying and showing as illustrated in this simple example. The claim is that reference clauses say what the reference is and show what the sense is—without stating it directly. In the pen example, an individual came to know something without it being communicated to her verbally. In the same way, these reference clauses are supposed to show the sense of “Hesperus” without actually saying what the sense of “Hesperus” is. It is a bit like my wishing to convey to you that I am English by opening my mouth and speaking in an English accent, without ever saying “I’m English.” I get the point across without stating it in so many words. Evans’s claim is that it is not possible to say directly what senses are; it is only possible to show what they are. Evans claims this for good reason. It is difficult to see how Frege can ever specify what a sense is independently of the reference of a particular expression. Invoking the saying–showing distinction promises to get Frege out of a tight theoretical corner. It makes

Evans on Understanding Demonstratives 

123

sense of the elusiveness of sense—at least it attempts to do that. Senses belong to the realm of that which cannot be said but can nevertheless be shown. The second point Evans wants to make about sense, which follows from the first point, is that the sense of an expression is reference dependent. Given the reference-stating way of thinking about sense, an expression’s having a sense will require that it have a reference. According to Evans, it is not possible to give a clause specifying the sense of “Hesperus” if there is no such thing as Hesperus. By asserting “The reference of ‘Hesperus’ = Hesperus” we presuppose that there is such a thing as Hesperus. We are using the name “Hesperus” to refer to Hesperus, so we must be assuming that Hesperus exists. Thus, Evans’s mode of specification of sense presupposes the existence of the reference. For this reason, he thinks, there cannot be senses without references. Senses are ontologically dependent on references. This idea of reference dependency, we will recall, derives from Russell. It is the idea that some expressions have a meaning that depends on the fact that the expression actually refers to something. In Russell’s theory, the meaning of a name is the actual object denoted. If there is no such object, there is no such meaning. Evans argues, like Russell, that the sense of names is reference dependent. He accordingly calls such terms “Russellian.” For these Russellian terms, there cannot be sense without reference. Names have a meaning, or sense, which depends on their having an existent reference. The next point Evans makes is that even though there are referencedependent senses, as Russell conceives them, names can have the same reference and different senses. The sense can be reference dependent, but that does not mean it is strictly identical to the reference. We can have a divergence of sense between two coreferring names that are nevertheless Russellian. Frege would say that “Hesperus” and “Phosphorus” have different senses, and that the sense is something that does not depend on reference. By contrast, Evans believes that even though those two names have different senses the senses are reference dependent. There cannot be a sense without reference (hence they are Russellian), but sense is something over and above reference and not identical to reference (hence they are Fregean). In Evans’s semantics, names can be both Fregean and Russellian at the same time. The meaning is not reducible to the bearer, but the meaning depends on the bearer. Evans is attempting to absorb the insights

124 

Chapter 6

of Russell about names while still trying to answer Frege’s worries about identity statements. 6.5  Mock Sense If names do not have sense unless they have reference, then what about empty names? Evans argues that, despite appearances, Frege does not really believe that there can be sense without reference. Evans attributes this position to Frege on the basis of what he says about fictional names. A fictional name like “Sherlock Holmes” apparently has sense and can therefore occur in meaningful sentences. But such a fictional name has no reference. So it seems that its sense is not reference dependent. Evans does not accept that conclusion. He tries to give textual evidence to support his interpretation of Frege. Frege says: “The logician does not have to bother with mock thoughts, just as a physicist, who sets out to investigate thunder, will not pay any attention to stage-thunder. When we speak of thoughts in what follows we mean thoughts proper, thoughts that are either true or false.” Evans defends the idea that the sense of an empty fictional name is defective because such names have a merely quasi sense, a mock sense. He suggests the comparison of empty names with vagueness. Frege himself made this defectiveness point about vagueness. The predicate “bald” says that someone is lacking in hair, but it is not precise about a specific threshold of hairs one must not have to qualify as bald. Frege held that such vague predicates lack genuine sense. Since there are borderline cases of baldness, there are sentences containing the word “bald” that are neither true nor false. However, in Frege’s system sentences cannot express a thought that is neither true nor false. Therefore, Frege was prepared to insist that vague predicates lack sense. Vague sentences express merely a quasi sense, not a proper scientific sense. There can be no vague predicates in science (e.g., math and physics). Vagueness is a defect of natural languages. Frege thus makes a distinction between words that have a proper scientific sense and words that lack such a scientific sense. He was ready to say that a vague predicate may appear to have a robust sense but not to possess such a sense when logically examined. Evans argues, analogously, that a fictional name may have a kind of degraded sense but does not have a strict proper sense. He takes the position that all proper senses are reference dependent, but the improper mock senses are not reference dependent

Evans on Understanding Demonstratives 

125

(so fictional names don’t have real sense). There is thus a class distinction between two kinds of sense. There is authentic non-nonsense sense and there is specious phony sense. Evans thinks that Frege has the resources to maintain both that upper-class sense is reference dependent and that there are lower-class expressions whose sense is independent of reference. In the case of empty names, the putative sense is always low-class, irresponsible, reference-indifferent sense. 6.6  Empty Names Philosophers have taken several views about empty names; the question is vexed. Accept it as a given that there is no such god as Zeus, that is, “Zeus does not exist” is true. What should we say about the sense of that name? The strict Millian view is that a name has sense only if it has reference, so in this case the name “Zeus” would have no sense. Indeed, it could not even be a name if it lacked reference, because it would be rendered meaningless. But if a name lacks sense, sentences containing it must be meaningless, which would make “Zeus does not exist” meaningless, instead of true. Another view is that “Zeus” does have a sense and that sense is contained within a synonymous definite description. The sense of an empty name is thus no different from that of a name of something that does exist. We could give Zeus the description “the most powerful of the Greek gods,” and then the sense of the name would be no more lacking than a name defined by “the most powerful man on Wall Street.” A third possibility, noted above, is that the empty name has a kind of sense, but it is only has a mock or apparent sense. This would be rather like an impostor pretending to be a big shot: he isn’t really one, but he puts on a good show. The name has pretend-sense, make-believe sense. A fourth possibility is that “Zeus” lacks an existent reference but instead has a Meinongian subsistent reference. The name denotes the most powerful of the Greek gods—and though this being does not exist, he nevertheless subsists. The sense of the name can be constituted by this shadowy subsistent reference. This theory of empty names is Mill meets Meinong. Each of these theories has its attractions and drawbacks. The Millian view, though nice and simple, makes some true sentences come out meaningless. The description theory saves meaning for empty names but runs into objections as a general theory of names. Meinong’s view gives a smooth

126 

Chapter 6

and comprehensive theory, but the ontology has struck many as hard to stomach. The pretend-sense theory looks reasonable for fictional sentences like “Zeus smote the Cyclops,” which aren’t part of factual discourse; but isn’t it a plain scientific fact that “Zeus does not exist” is true? The thought expressed here is not a kind of mock phony thought lacking in truth-value but a straightforwardly true thought—and how can this be if “Zeus” has only a mock sense? Evans has described another approach to empty names, but it is hard to see how it captures the linguistic data adequately. 6.7  Evans’s View of Names In the next part of his paper, Evans sets out to defend the thesis that names are Russellian. He writes: Therefore, on the present conception, the sense of a singular term is a way of thinking about a particular object: something that obviously could not exist if that object did not exist to be thought about.1

He asserts here that if a sense is a way of thinking about an object, there could not be a sense without the object existing. Let us first consider this assertion in application to perceiving. Suppose I visually perceive a certain object, for example, a pen. My perceptual state could be specified by saying which thing I perceive: “CMG is seeing that pen.” In this case, the perceived object is referred to in the course of characterizing my perceptual state. My perceptual state is a way of seeing that pen. You might have a different way of seeing the pen because you have a different perspective, but we all see the same pen. But is it strictly necessary for the pen to be there in order for me to have a way of seeing it? What if I am hallucinating a pen? Don’t I still enjoy a perceptual state—a way of seeing—even though there is nothing there? How can we characterize the perceptual state of someone hallucinating a pen? Not by saying “He sees that pen,” which presupposes there is a pen. Rather, we say something like “It appears to him that there is a pen in front of him.” This kind of sentence does not commit us to the proposition that there really is a pen in front of the person hallucinating. There is no reference to any particular pen here. We can thus ascribe a perceptual content to him without specifying a reference for that perceptual content. This is fortunate, because there is no such reference.

Evans on Understanding Demonstratives 

127

In general, it is not true that a way of seeing an object can exist only if the object exists. There can be modes of presentation of objects without the objects existing. So Evans’s argument that senses are reference dependent begins to run into trouble. Consider an ordinary definite description, say, “the queen of England.” The meaning of that description can be characterized as a way of referring to an object. In one’s thoughts it is a way of thinking of an object (thinking of Elizabeth II as the queen of England). But Evans does not think that descriptions are reference dependent, because there can clearly be meaningful expressions like “the queen of England” without there being a queen of England. For instance, Evans would agree that “the king of France” is a meaningful description fully endowed with sense, even though there is no reference for that description. Evans’s general argument here would imply that since the sense of the description is a way of thinking of an object, there must be an existing object of which it is a way of thinking. But it is just a non sequitur to suggest that where there is a way of thinking there must be an object thought about. There are obviously ways of painting mythical beasts, but that does not imply that there are mythical beasts that are painted. So Evans has not shown that senses are reference dependent and hence Russellian. Evans also contends that Russellian terms can be Fregean. That is, he thinks that coreferring terms can have reference-dependent sense and still differ in their sense. However, that raises the question: What is the difference between two Russellian terms that differ in their sense? What does that difference consist in? It certainly cannot consist in their having different references, because they have the same reference. There has to be something that goes beyond reference to generate the distinction of sense. Whatever this is, it cannot consist in reference alone. But if there is some semantic dimension to the name that goes beyond its reference, it should be possible to have some conception of what that difference is. Is it perhaps the way the reference is conceptualized? But now we are moving in the direction of a description theory, and descriptive concepts are not reference dependent. The semantic difference cannot be explained in purely Russellian terms, because in Russell’s theory that would be just the reference. If you say there is no difference, then the terms are not Fregean after all. If there is a Fregean distinction between them, it must float free of reference—as general concepts do. The extra ingredient of sense cannot itself be reference dependent.

128 

Chapter 6

The upshot of all this is that Evans has not succeeded in describing a coherent alternative to the description theory of sense that might provide a viable Fregean treatment of demonstratives. He accepts that a description theory of indexical sense has to be wrong, and then he tries to construct an alternative nondescriptive Fregean theory. But it remains unclear that there is any such nondescriptive Fregean alternative, so indexicals appear to refute Frege’s general semantic principles after all. 6.8  Evans on “Today” and “Yesterday” Evans makes an important point later in his paper about the words “today” and “yesterday.” Suppose on a certain day D1 I say truly, “Today is cold.” Now the next day D2 I want to express the same thought I expressed on D1. I can’t do this by uttering, “Today is cold” on D2, because that will refer to D2. To express the same proposition as I did on D1 requires the use of the word “yesterday”—I have to say, “Yesterday was cold.” Intuitively, I have expressed the same thing on D2 as I did on D1 by using that sentence. The same thought is expressed on two different days by using these different sets of words. These forms of words are related in a certain systematic way—there are rules for which word to use in a different context to express the same thing. When we understand these words, we grasp these rules. There is a very similar linguistic structure in the case of spatial indexicals (as well as personal indexicals). For example, if I say, “Here is cold” and I then move away from that place, I must say, “There is cold” in order to say the same thing: the same proposition is expressed about the original location from different locations using different words. The indexical used must be changed when the context of utterance changes. Evans’s point about these cases is that they apparently require a Fregean notion of sense, because the sense of the word “today” when it is used on D1 is the same as the sense of the word “yesterday” when it is used on D2. As noted at the end of the previous chapter, “today” certainly does not have the same character (or conventional meaning) as “yesterday.” To capture what the two words have semantically in common, Evans thinks we need to invoke Fregean sense. We clearly need a piece of semantic machinery to capture the commonality when those two different indexicals are used to express the same thing in two different contexts. Character is not suitable, because the character is different in the two cases. We might suppose that

Evans on Understanding Demonstratives 

129

although the character is different, the Kaplanian content is the same. In other words, the reference is the same. The reference of “today” on D1 is D1 and the reference of “yesterday” on D2 is also D1. The sense in which the same thing has been said on those two successive days is captured (it may be said) by the fact that those two tokens of the indexicals have the same reference. Notice that this view is a non-Fregean view of having the same thought, because it makes no distinction between sense and reference. Frege does not think that having the same reference is ever the same thing as expressing the same sense. But at least the content is the same on both days, unlike the character. Suppose that D1 is a Tuesday and so “today” is tied down to a specific Tuesday. D2 would then be a Wednesday. Now there is a relationship between those two names of days and the two indexical terms. We can say that Tuesday is identical to the reference of “today” when said on D1, which is identical to the reference of “yesterday” when said on D2. So D1 can be referred to with “Tuesday,” “today,” and “yesterday.” Now consider the relationship between saying “Today is cold” on Tuesday and saying “Tuesday is cold.” The word “Tuesday” here refers to the same day to which “today” refers. We have the true identity statement “Today is Tuesday.” There is a truth-value relationship between “Today is cold” and “Tuesday is cold,” such that if one statement is true the other is also. Each of these statements has the same Kaplanian content, because “Tuesday” refers to the same day as “today.” But, intuitively, “Today is cold” does not say the same thing as “Tuesday is cold.” Each word refers to the same day, but they have different senses. We can see this from the fact that someone might not actually know that today is Tuesday when he uses the word “today” to refer to Tuesday. He might assent to “Today is cold” but dissent from “Tuesday is cold” because he disbelieves that today is Tuesday. If he later discovers that today is Tuesday, he would have learned an a posteriori synthetic truth. So “Today is cold” cannot express the same thought as “Tuesday is cold” even though the reference is to the same day. Those two statements (“Tuesday is cold,” “Today is cold”) do not say the same thing according to Frege’s test for the identity of thoughts. Also, they intuitively do not say the same thing. However, they have the same content in the Kaplanian sense. This case is different from where “today” is said on D1 and “yesterday” is said on D2. In that case, each of the sentences does say the same thing, because no new information is acquired when

130 

Chapter 6

one discovers that those sentences are related in the way they are. There is an analytical, logical connection between those two indexicals, written into the rules for their use. We know that if “Today is cold” is true on D1 then “Yesterday was cold” has to be true on D2. But we do not know that if “Today is cold” is true on D1 then “Tuesday is cold” has to be true, because “Today is cold” can be truly uttered on days that are not Tuesday. These two sentences are not synonymous in the ordinary sense of making the same statement. The word “yesterday” said on D2 captures the same sense as “today” said on D1, but “today” and “Tuesday” do not express the same sense. Therefore, the identity of sense between the former pair is not captured by Kaplanian content, because that content is in common between the latter pair of statements too. Sameness of content is not enough for sameness of sense. So we need an extra semantic ingredient to capture what is common to “today” and “yesterday” but not to “today” and “Tuesday.” We are thus driven to accept a third level, beyond Kaplan’s character and content, which is closer to Frege’s idea of sense. 6.9  Character, Content, and Information Now we can combine three semantic elements to explicate the full meaning of an indexical sentence as used on an occasion. The first is character, the second is content, and the third corresponds to the sameness of sense that exists between “today” and “yesterday.” Let us call this third layer information. The same information is conveyed by saying “Today is cold” on D1 as would be conveyed by saying on D2 “Yesterday was cold.” The speaker acquires the information from his sense experience on D1 that the day is cold and that information is stored in his memory. On D2 when he says “Yesterday was cold,” he is merely referring back to the information he acquired from the previous day that is stored in his memory. The speaker has the same piece of knowledge acquired the previous day but he expresses it by using different words. Therefore, the same information is in the speaker’s mind over the two days and he expresses it using these two different sentences. This notion of information is not reducible to either character or content. Content is too wide a notion and does not capture the exact meaning of what the speaker says. To avoid confusion, we might rename Kaplanian content real-world correlate. The real-world correlate of the indexical is the object to which the speaker refers. We can still regard this as a

Evans on Understanding Demonstratives 

131

component of the proposition expressed. We could also naturally rename character perspective. Perspective incorporates the two different temporal perspectives the speaker has on the given day—as present and as past. Let us insert this into the proposition too. The same information is expressed from two different perspectives. It is information about the same real-world correlate. We should not say that there is only the real-world correlate and the perspective, because then we could not understand the relationship between “today” and “yesterday” in the right way. The information is preserved over time and then expressed from two different perspectives, but the information is more like a cognitive state than a real-world correlate. This can be bundled into the proposition along with the other two elements. None of these propositional components determines any of the others, so none is redundant. If we think of the informational component as descriptive, which is natural, then we shall not insist that the descriptive information determines a particular day—it might be available on other days too (so it is not equivalent to reference-determining Fregean sense). We have three irreducibly distinct and indispensable semantic ingredients: real-world correlate, perspective, and information. According to this triple-layer semantics, it turns out that everybody is a little bit right and little bit wrong about this subject. Kaplan is right to introduce character and content but wrong to think that character and content are all that is required. Evans believes that only Fregean sense is needed. He is right to think that there is something common to “today” and “yesterday” but wrong to suppose that nothing separates them (character). Evans leaves no room in his theory for this semantic difference: he needs character in the full meaning of an indexical utterance as well as sense. The same information is indeed expressed by these two words on successive days, but each term has a different conventional meaning. Kaplan and Evans both offer incomplete theories because they each need something from the other’s arsenal to fill out the full account of indexical meaning. We need both character and content, but we also need to recognize that indexicals with different character can have something in common (what we have been calling information) that is not reducible to content. The next task would be to inquire more closely into what this notion of information amounts to (a task we shall leave for homework). All we can say now is that information is an epistemic notion: it relates to what someone knows. What is clear by now is that the topic of indexical semantics bristles with complexity and difficulty, and no current theory has all the pieces worked out.

7  Putnam on Semantic Externalism

7.1 Background Our earlier discussions of indexicality will help us to understand the force of Hilary Putnam’s arguments in “Meaning and Reference.” For indexical expressions, the classic theory of descriptive intensions that determine extensions looks highly implausible—as Kaplan argues. The meaning of an indexical when used on an occasion is not equivalent to a definite description of the object or type of object referred to. As Putnam notes toward the end of his paper, two people can use the word “I” to refer to themselves even if they don’t differ in the descriptions they would ascribe to themselves; so the difference of reference cannot stem from uniquely identifying knowledge possessed by the two speakers. Here context plays an indispensable reference-determining role—and not simply what occurs descriptively inside the speaker’s mind. What you refer to can depend on who and where you are, not just on what you think—it depends on external context, not internal descriptions. That is, indexical reference is fixed externally by the speaker’s objective context, not by what he has in his mind subjectively. This is in contrast with descriptive reference, which is context independent, because here the speaker’s internal concepts do suffice to fix what he refers to. Thus externalism is correct for indexical reference, but internalism is correct for (pure) descriptive reference—as with “the first dog to be born at sea.” In the case of “I” we just need to know who is uttering the word to determine its reference, not what that person thinks about his reference. Putnam’s focus is on natural kind terms like “water,” “aluminum,” and “tiger”—words that stand for types of object found in nature—not words for human artifacts like “table,” “computer,” or “president.” He wants to know what these words mean, particularly how they determine their reference. At

134 

Chapter 7

the end of his paper he says, “Our theory can be summarized as saying that words like ‘water’ have an unnoticed indexical component: ‘water’ is stuff that that bears a certain similarity relation to the water around here” (note the indexical “here”). In other words, the semantics of natural kind terms mirrors the semantics of indexical terms. Such terms do not conform to the classic Fregean model of the definite description and its reference. Putnam tells us that it used to be thought that for any meaningful expression there is an intension that determines the extension in every possible world, and that when a speaker understands the term she grasps the intension of the term. But he argues that this is not true of natural kind terms—we do not understand them by grasping such intensions. We understand them in the same way we understand indexicals, where context plays an indispensable role. Putnam puts this by saying that the psychological state of the speaker is not the sole determinant of the reference of her terms—that is, internal psychology does not determine a speaker’s reference. He thus rejects the old view to the effect that a speaker’s reference can be extracted from what is in her mind as she speaks. We shall now examine his arguments for this conclusion. 7.2  Twin Earth and “Water” Putnam proceeds by constructing his Twin Earth thought experiment. Imagine a time on Earth before chemistry fully developed but people still used the word “water.” Because of the lack of development in chemistry, people did not know the actual chemical composition of water, which is H2O. The word “water” as it is used on Earth refers to water. Now imagine an exact duplicate of Earth, Twin Earth, where there is no water. However, there is a liquid on Twin Earth with many of the same apparent features that water has even though it is not water. Putnam stipulates that it has a chemical composition of XYZ. Of course, it is possible for liquids to have the same appearance without having the same chemical composition. The thought experiment is all perfectly metaphysically possible. Now suppose that there are people on Twin Earth that are exactly like us—they are in fact identical molecular duplicates of us, perfect twins. They speak a language we would call English and one of the words they use is “water.” However, in Twin Earth English, “water” refers to the liquid on Twin Earth (XYZ), not the liquid on Earth (H2O). The term has a different extension on the two planets. However, since the time period we are considering is before the

Putnam on Semantic Externalism 

135

advent of chemistry, no one on either Twin Earth or Earth knows that the liquid around him or her is XYZ or H2O, respectively. The two words have different references, but the speakers can’t distinguish them chemically. Our word “water” does not refer to XYZ but only to H2O. At the same time, their word “water” does not refer to H2O but only to XYZ. Even though our Twin Earth counterparts are molecular duplicates of us, they still use the word “water” to refer to something different from what we use it to refer to. Since our twins are molecular duplicates of us, we are both in the same psychological state, but the extensions of our terms differ. What goes on in our minds when we use “water” also goes on in their minds when they use “water,” because the two liquids present the same subjective appearance. So psychological state does not determine reference or extension, according to Putnam. What a speaker means by his words is not determined by his internal psychological state; it is determined by his actual external environment, his context. Both sets of speakers have the same information about their respective liquids—they would give the same descriptions of it—but the context of use is different, so the reference is different too. The speakers don’t know enough chemistry to distinguish the liquids, but since they are distinct, the reference is also distinct. If we now assume that meaning determines reference, we can conclude that “water” does not have the same meaning on Earth and Twin Earth. The words thus have the same descriptive content but they do not have the same meaning. The words are functioning much like a device of direct reference where the reference itself enters into the meaning. We could think of the word “water” on Earth as a proper name that denotes H2O, and the word “water” on Twin Earth as a proper name that denotes XYZ. As Kaplan would say, the proposition expressed contains different entities. The term “water” is not short for a description, because the same descriptions that are in our minds are in the minds of our Twin Earth counterparts, yet our references diverge. And this implies that the meanings diverge, on the assumption that meaning determines reference. 7.3  Meanings Are Not in the Head Putnam’s conclusion is that “meanings are not in the head.” What does he mean by that? He means that we can see from his thought experiment that a person’s psychological state does not determine what he means by his

136 

Chapter 7

words. Putnam thinks that what is in your head does not determine your meaning because it does not determine reference. The people on Earth and Twin Earth have the same things going on in their heads, but they don’t mean the same thing by their term “water” because they don’t refer to the same thing. The meaning of the word cannot be deduced from the speaker’s psychological state. It depends on other extrinsic factors, and we will soon see what those are. The speaker’s inner state of understanding does not necessarily fix what he is referring to, so the meaning of his term cannot be “read off” his state of understanding. Therefore, Putnam concludes, meaning is not “in the head.” His point is that meaning is not a psychological phenomenon. Let us restate Putnam’s argument with all the pieces now in place. The essential point of the Twin Earth thought experiment is that we would be right to say that “water” in Twin Earth English refers to XYZ and that “water” in Earth English refers to H2O. Since the inhabitants of Twin Earth are molecular duplicates of us, this point has serious philosophical consequences for what constitutes meaning. As molecular duplicates, their brain states are exactly the same as ours. Consequently, if we could gaze into the minds of our molecular duplicates while they say the word “water,” we would see exactly the same experiences, beliefs, emotions, and desires that we would see if we gazed into our own minds when we said that word. So we can see that on these two different planets the word “water” has a different reference and therefore a different meaning, despite the fact that the speakers who use that word are in the same psychological state when they use it. Since the same descriptions are associated with the word by the two sets of speakers (“the colorless, tasteless liquid that flows in rivers” etc.), the speakers are in the same psychological state, even though the word “water” has different reference in the two cases. Since meaning determines reference, as Putnam assumes, following Frege, the two words must have different meanings. Thus “water” on Twin Earth does not have the same meaning (or sense) as “water” on Earth. Nevertheless, the speakers are in the same psychological state when using this word. Another easy way to see how the argument works is to consider the case of ordinary names. Take the name “Aristotle” and suppose that on Twin Earth there is no Aristotle, because it is too far away for Aristotle to have ever visited Twin Earth. However, there is a person on Twin Earth who looks and behaves exactly the same way as Aristotle, without being him. On Twin

Putnam on Semantic Externalism 

137

Earth, when the speakers use the name “Aristotle” they refer to their Aristotle—not to our Aristotle. To avoid ambiguity and confusion, we can call their Aristotle “Albert.” When they use the name “Aristotle” they refer to Albert (as we call him), because the name “Albert” is our name for the man they refer to with “Aristotle.” Putnam’s point holds here because the speakers on Twin Earth are exact psychological and physical duplicates of us, but they refer to a different person when they use the name “Aristotle” from the person we refer to when we use that name. They refer to Albert (though their name for him is “Aristotle”), whereas we refer to Aristotle. Since meaning determines reference, it cannot be that the meaning of our word “Aristotle” is in our heads. The Twin Earthlings’ psychological state is exactly the same as ours and yet they do not refer to Aristotle but to Albert with “Aristotle.” There is a different reference yet the same internal psychology. It is important to note that when the speakers either on Earth or Twin Earth say the word “water,” there are no experts on these planets as to what water is. We are assuming in this first example that the thought experiment concerns a time before the rise of chemistry. No one on Twin Earth or Earth knows the molecular composition of the liquid they refer to with the word “water.” So the case is not like the contemporary world. In addition to the example of the word “water,” Putnam also gives us the case of molybdenum and aluminum. It is essentially the same situation as with Twin Earth “water” except Putnam assumes that there are some experts who can tell the difference between aluminum and molybdenum. He supposes that there are some metallurgists who can determine this fairly easily (on Twin Earth pots and pans are made of molybdenum and on Earth they are made of aluminum, and the metallurgists can tell the difference by a simple test). They look very similar and are used for the same purposes, but a metallurgist would quickly be able to determine what type of metals they were. There is nothing really new in this second example—it is exactly the same as the first one, but Putnam just happens to bring in some experts. In this case we also have duplicate speakers referring to different things with the same term, so it is not a matter of what is going on inside you that determines what you refer to but what kind of environment you are in. The third example Putnam mentions is the use of the words “elm” and “beech” to refer to different species of trees. This example does add something to the original story because Twin Earth is not required to see this point. It is a point about Hilary Putnam himself, stuck here on Earth. In his

138 

Chapter 7

idiolect, when he uses the word “elm” he does not associate any descriptions with that word that he does not associate with the term “beech,” because he confesses that he cannot tell the difference between elms and beeches. Since most of us are similarly (and shamefully) ignorant of the differences between elms and beeches, we could not supply a description to distinguish one from the other. Nevertheless, the words “elm” and “beech” mean something different in our idiolects as we use them—they do not have the same reference or extension. Although there is nothing in our minds that would allow us to make a distinction between the two, one term refers to a tree that is an elm, and the other refers to a tree that is a beech. This should remind us of Kripke’s example of Feynman and Gellman (see chapter 2). Not familiar with the particulars of their work, a speaker’s description of each of these physicists might be that they are both famous twentieth-century physicists. Even though the speaker possesses no descriptions to distinguish Feynman from Gellman, he still refers to a different person when he uses “Feynman” from when he uses “Gellman.” We may wonder how we can use these words to refer to different natural kinds of trees, even though what is in our head may be the same in respect to these two words. The speaker means something different by “elm” and “beech,” even if the stuff in his head is the same. This is a question about a speaker’s idiolect in a specific linguistic community, as opposed to comparing two linguistic communities (Twin Earth and Earth). Earlier in the book (chapter 2) we talked about the division of linguistic labor in connection with Kripke and names. That same division of linguistic labor, in which experts determine what particular words refer to, is present here. When we ignoramuses use “elm” or “beech,” we really intend that our reference with those words should depend on our relation to the arboreal experts in our midst. When we use the words, we intend to refer to what the experts refer to when they use the words “elm” and “beech.” In this case, too, the individual speaker’s meaning cannot be read off his psychological state, but can only be gleaned from his context—specifically, the experts in his linguistic community. There are a few other examples that Putnam does not go into detail about that are useful to our discussion. Toward the end of the article, he begins to talk about indexicality and that notion seems to play a central role in these cases. Many of them directly involve indexicals. Imagine somebody pointing to an elephant. When the speaker says “that elephant,” imagine that his

Putnam on Semantic Externalism 

139

brain is in a certain state and that he perceives the elephant in a certain way (as big, gray, etc.). Now imagine on Twin Earth or some other place on Earth that there is an exact twin of that person who says “that elephant,” pointing to a different elephant. He is a molecular twin of the first speaker, so everything is internally the same in both of their minds. However, when the first speaker says “that elephant” he refers to a different elephant from the one his twin refers to. They refer to different animals even though they are in exactly the same psychological state, because they are pointing at different elephants. The context fixes the reference, not the perceptions and ideas in their minds. They refer to what they see, and they see different elephants. Another example is the word “I.” I say, “I am hungry”; now consider an exact duplicate of me who says, “I am hungry.” He does not refer to me, he refers to him, but he is in exactly the same psychological state as me because he is a molecular duplicate of me. By uttering the word “I” he refers to an object a, whereas I refer to an object b, but we are both in the same internal psychological state. Therefore, if meaning determines reference, our meanings are not in our heads—what we say cannot be extracted from an examination of what is happening inside us. The context, that is, who is actually uttering the word on that occasion, determines what we say. Putnam’s recipe for producing these outside-the-head cases is very straightforward: we just vary the speaker’s environment while keeping his head the same, and we find that the semantics varies. It is not difficult to generate similar cases for “now” and “here.” The trick of the examples is simply that context can vary while internal states stay constant. Let us make something explicit. Toward the end of the article, Putnam hints at this point, but it has a much greater significance than he recognizes. He is really arguing for a disjunction: either meaning is not in the head or meaning does not determine reference. His thought experiments are neutral between these two propositions—we can interpret them either way. However, Putnam assumes that meaning determines reference, and hence concludes that meaning is not in the head. If meaning determines reference, then meanings are not in the head. But what if meaning does not determine reference? Then meaning can stay in the head, while failing to determine reference. He has shown that meaning does not determine reference, on this alternative interpretation. We could therefore accept Putnam’s Twin Earth cases but question whether they prove that meaning is not in the head and hence not

140 

Chapter 7

determined by psychological state. Could it be that meaning is in the head and hence is determined by psychological state, but meaning does not determine reference? There are thus two theoretical possibilities: (1) meanings are not in the head and hence are independent of psychological state, or (2) meanings are in the head and hence are dependent on psychological state—but meaning is not sufficient to fix reference. Why does Putnam choose one of these interpretations over the other? We could interpret the Twin Earth case as showing that the people on Twin Earth mean the same thing by the word “water” as we mean by the word “water” but that their reference for that word is different from our reference for that word. What they mean is what is in their heads, the descriptions they would give. But what they mean does not uniquely determine what they refer to. It is only on the assumption that intension determines extension—that sense determines reference—that Putnam’s cases establish that meaning is not in the head. To better illustrate this point, let us return to the indexical examples. In the elephant case, when each speaker says “that elephant” he refers to a different animal when pointing to each elephant. It is indisputable that they refer to something different, but it does not follow that they mean something different. It largely depends on our definition of “meaning.” There is a lot of complexity in the notion of meaning, particularly in the case of indexicals. We learned in earlier chapters that we need at least a twodimensional theory of indexical meaning. Using Kaplan’s idea of character as the meaning of “meaning,” the words “that elephant” have the same character and hence the same linguistic meaning for the first speaker as for the second. That character, however, does not determine reference. What determines reference is character plus context, not character by itself. So meaning, construed as character, does not determine reference. For this reason, it would be a strange interpretation to say that this example shows that meaning is not in the head—instead, it shows that meaning (character) does not determine reference. As Kaplan would say, it shows that character does not determine content. We will come back to this point later, but first we must cover Putnam’s view of what his examples show. One thing he concludes is expressed in the following passage: HYPOTHESIS OF THE UNIVERSALITY OF THE DIVISION OF LINGUISTIC LABOR: Every linguistic community exemplifies the sort of division of linguistic labor just described; that is, it possesses at least some terms whose associated “criteria” are

Putnam on Semantic Externalism 

141

known only to a subset of the speakers who acquire the terms, and whose use by the other speakers depends upon a structured cooperation between them and the speakers in the relevant subsets.1

This is the familiar idea of the experts. Experts can differentiate one object or type of object from another, and other members of the linguistic community rely on the ability the experts have. Thus the references of the words “elm” and “beech” are the species of trees that the experts decide those names designate (the experts might be scientists or just observant country people). In the beech–elm type of case, the division of linguistic labor is the explanation of why meanings are not in the head of individual speakers, because the meaning depends on your relation to the experts, not your own imperfect information. Those experts are not “in your head,” and they have knowledge that is not in your head either. And you rely on them in such a way that in your idiolect the word refers to a certain kind of thing not in virtue of anything you know personally but in virtue of the fact that you defer to them, the experts. We can summarize that by saying that meaning is a social phenomenon. What you mean depends on the competence of others. Thus Putnam’s arguments are said to establish an “anti-individualist” view of meaning. But notice that this explanation cannot account for the original “water” type of case, because there are no experts in that thought experiment. It cannot be that the difference of reference between Earth and Twin Earth depends on experts that are deferred to in those communities. No one can tell the difference between the two liquids. In that kind of case the semantic difference would not depend on the division of linguistic labor. Instead, in the Twin Earth thought experiment, meaning depends on the fact that the speaker causally interacts with the actual natural kinds occurring in the world in which he is embedded. The speaker’s use of the words is tied to his causal interactions with those natural kinds, as they act on his senses and he acts on them, and those interactions fix what the speaker refers to with the word. On Earth, when we use the word “water” we are always interacting with water, that is, H2O. On Twin Earth, when they use the word “water,” they are always interacting with XYZ. The thing that determines that they refer to something different with the term “water” is the surrounding world itself—not the presence of experts in that world. The meaning is not in the heads of the experts either, since there are none. The meaning comes from the world itself, unmediated by anybody’s

142 

Chapter 7

psychological states. Speakers are embedded in a world and they interact with various things: it is the fact that they have those interactions that determines what their words mean. What words mean is not just a function of what is in someone’s head, individually or socially. Rather, meaning is a function of the speaker’s actual external environment. The environment itself determines what words mean. So, Putnam concludes, meaning is not in the head—it emerges from interactions with the environment. This doctrine came to be known as semantic externalism, because it says that meanings are externally determined. As noted earlier, Putnam thinks the case of natural kind terms is similar to the case of indexicals. In the case of indexicals, we can clearly see that the reference depends on the speaker’s way of being embedded in his or her environment. We can see the operation of context. What determines the thing I refer to when I say “that woman,” pointing to a particular woman in front of me? Not what is lurking in my head, but the fact that a certain woman is in my environment right now in front of me and I am pointing straight at her. In the case of indexicals, it is very clear that reference is fixed by the speaker’s location in the world. Here externalism seems obvious, because indexicals are so clearly context dependent. Putnam now makes a direct link between indexicals and natural kind terms like “water.” He suggests that there is an indexical element in natural kind terms. We can explain the reference of our word “water” by using a demonstrative—as in “‘Water’ refers to that liquid,” said while pointing at H2O—and that is how we tie down the reference of the word. As we have discussed earlier, indexicals play a crucial role in determining the reference of words that are not themselves indexical (e.g., proper names and definite descriptions like “the father of that baby”). On Earth when we say “water” the reference is determined by the indexical “that liquid.” On Twin Earth when they say “water,” the reference is also determined by “that liquid,” but here the demonstrative picks out a different natural kind. Hence the word “water” has different reference on the two planets. Given this referential link between indexicals and natural kind terms, we would expect to find natural kind terms functioning in the same way indexicals do. The meaning of indexicals is not in the head, nor is the meaning of the natural kind terms that are linked to them. Externalism thus holds for terms like “water” because they have an indexical component.

Putnam on Semantic Externalism 

143

7.4  Criticisms of Putnam What is the best way to characterize the upshot of Putnam’s examples? What do they show about meaning? Putnam says they show that meaning is not in the head, but we can (as noted earlier) equally conclude that they show that meaning does not determine reference. Which characterization is better? If we start with an indexical case like “I,” then on any reasonable notion of meaning, the word “I” has the same meaning for whoever uses it. The reference is not the same, to be sure, but the meaning clearly is. A speaker refers to a certain individual when he uses the word “I” on an occasion, but that fact is not reflected in what the word means, because the reference depends on the meaning plus the context (character plus context). It thus makes perfect sense to say that the meaning (character) of “I” is in the head, because what is present in the speaker’s mind determines what character the indexical has. However, the conventional meaning of “I” is not enough to determine its reference on an occasion. If we get fixated on the description case, we will think that meaning has to determine reference, because meaning does determine reference for definite descriptions. But this is not so for indexicals. Indexicals require a more complex semantics, as Kaplan demonstrates, in which we must distinguish different dimensions of semantic significance. To say simply “Meaning is not in the head” is therefore incomplete and ambiguous. Do we mean meaning as character or as content—as conventional linguistic meaning or propositional content? Nothing in Putnam has shown us that character is not in the head, so one sort of meaning is in the head; all we have shown is that propositional content is not in the head. Given Putnam’s own indexical interpretation of his cases, he should have concluded that part of meaning (character) is in the head, though part is not in the head (content). A further question concerns Putnam’s notion of a psychological state. He simply assumes from the outset that psychological states are in the head. This enables him to conclude that meaning is not psychological, because the former is not in the head while the latter is. He thus takes it for granted that the psychological state of the molecular duplicates on Twin Earth is the same as the psychological state of people on Earth. He takes it that people cannot have different psychological states if they are physically identical. But is that so obvious? Some have questioned this assumption of Putnam’s, wondering whether we should conclude instead that psychological states

144 

Chapter 7

are not in the head either. Let us ask ourselves what people believe on Earth and Twin Earth. What do I believe when I say “This water is warm”? Obviously I believe that this water warm. On Twin Earth my molecular duplicate would also say “This water is warm,” referring to some XYZ. Does he believe that this water is warm? Well, he clearly does not believe that this water is warm, because this water is here on Earth, not there on Twin Earth. But does he have any belief involving the concept of water at all? No, he does not: he has no beliefs about water at all. He has beliefs about another liquid, not water. Let’s call his XYZ liquid “retaw”; then we can say he has beliefs about retaw. What he believes is that some retaw is warm. His belief is about something different from what my belief is about. He has the concept retaw whereas I have the concept water. Clearly he perceives something different from what I perceive, because I am in the perceptual state of seeing water and he is not in that perceptual state. He is never in that perceptual state, because he never sees any water; he sees only retaw. We cannot report his perceptual state with the words “He sees that water is in the well.” The psychological state of seeing water is not a psychological state anyone on Twin Earth is ever in. Neither does anyone there have the concept water, nor the belief that there is water about. The psychological states associated with the word “water” on Twin Earth are therefore not the same as the psychological states we have on Earth. They have different psychological states from us. To be sure, they share some psychological states with us, namely the descriptive beliefs they apply to the liquid on their planet. But it does not follow that they share all their psychological states with us, and it seems plainly false to use our word “water” to describe their psychological states. Have you ever had any beliefs involving the concept retaw before hearing about Twin Earth? Hardly; all your beliefs involve the concept water. They no more think about the natural kind water than they think about the particular pool of water I referred to here on Earth with “this water.” So there are psychological states correlated with the use of “water” on Earth and Twin Earth that differ in their content, even though the speakers are molecular duplicates. Therefore, these psychological states are not in the head. When Putnam says that meanings are not in the head, he should have added that psychological states are not in the head either—and for essentially the same reason. The content of psychological states is also fixed by the person’s actual environment. That is, the full propositional content of psychological states is partly fixed by

Putnam on Semantic Externalism 

145

interactions with the environment. Thus we have externalism about mind as well as meaning. But this changes the whole picture. If the psychological states of Earthlings and Twin Earthlings are different, then those psychological states do determine the meaning of the terms used, even when this meaning is taken to incorporate something like Kaplanian content. The psychological state of my counterpart involves the concept retaw, while the psychological state that I am in involves the concept water. Those two concepts are not determined purely by our internal states but by our embedding in the world. These externally determined psychological states accordingly do determine what we mean by the term “water.” Thus there is no divorce of semantics from psychology, only a divorce of psychology from neurophysiology. Neither meaning nor mind is reducible to internal neurophysiology. To return to the indexical case involving the elephant: one speaker says “That elephant is big” while pointing to elephant A, while the other speaker points to elephant B and says “That elephant is big.” The first speaker believes that A is big, while the second believes that B is big. A and B could be animals on different continents. Each speaker only has beliefs about the elephant in front of him—to the effect that that elephant is big. The content of the belief a person has when he uses an indexical term like this is determined by his environment, so his beliefs are not in his head. This is just to apply the lessons of direct reference to beliefs as well as meanings. Belief and meaning march in parallel, as we would expect. Thus, psychological states are not in the head and meanings are not in the head. Or better, an aspect of both is not in the head—because there is also an aspect that is in the head (the aspect corresponding to character). If psychological states are not in the head, they can determine meaning, even assuming that meaning determines reference. My psychological state could determine the reference of my terms even if we accept Twin Earth cases, because the psychological states of the people on the two planets differ, despite their molecular identity. The psychological state mirrors what is in the person’s environment too. As soon as we realize that psychological states are not in the head, we see that Putnam is misstating his conclusion. He is right to say that what is internal to us cannot determine our reference, but that does not entail that our psychology does not determine our reference. Rather, our psychology is not (purely) internal. We need to accept psychological externalism too.

146 

Chapter 7

In sum: Putnam is wrong to claim that meaning is totally outside the head, because of the existence of an internal component to meaning, namely character; and he is wrong to claim that meaning is not fixed by psychological state, because his own arguments imply that psychological states are as externally determined as meaning. What he is right about is that external context plays a vital role in fixing reference. This may not sound like the resounding and revolutionary conclusion he originally announced, especially once we have properly investigated the semantics of indexicals, but it does contain an important truth.

8  Tarski’s Theory of Truth

8.1 Background We have invoked the concept of truth at several points, but we have said nothing much about how this concept is to be understood. What is truth? The theory of truth we are about to examine was originally proposed in 1933 in a very long and difficult article called “The Concept of Truth in Formalized Languages” by Alfred Tarski, a Polish mathematical logician. The article we will examine, however, is “The Semantic Conception of Truth,” published in 1944, and meant as an easier presentation of the same ideas as the original, more daunting article. This one is quite daunting enough. As Tarski remarks at the beginning, it is a return to the subject matter of his earlier article, which is really a treatise in formal logic. The original article is tough reading if one does not have a solid grounding in mathematical logic. It was a very important contribution to pure logic. It has also been philosophically important. Historically, people thought it was a great breakthrough in the philosophical theory of truth. It finally made the study of truth rigorous and subject to logical treatment. It made philosophy into mathematics! Many philosophers felt that we no longer needed to have any qualms about employing the notion of truth, because Tarski had given us such a tight, precise definition. Donald Davidson later took up Tarski’s theory to propose a theory of meaning for natural languages, as we will see in the next chapter. Tarski had tamed truth, rendered it “scientific”—quite a feat. The adjective “Tarskian” took on the canonical stature of the adjective “Fregean”—as in “the Tarskian theory of truth” and “the Fregean theory of meaning.” Still, there is controversy about what Tarski’s theory really accomplishes, both as a theory of truth and a theory of meaning. But before we get to that, we need a good understanding of what the theory actually says, and

148 

Chapter 8

for that we cannot do better than to attend to Tarski’s own words. This will be our procedure in what follows. Let us first talk a bit about the background to Tarski’s own proposal. In the history of philosophy, a number of different theories have been proposed about truth: the coherence theory of truth, the correspondence theory of truth, and the pragmatic theory of truth. The coherence theory states that a proposition is true if and only if that proposition coheres with the other propositions that one believes. By the standards of the coherence theory, a belief is true if and only if that belief is consistent with one’s other beliefs. Truth is then a matter of the logical relations between believed propositions. The correspondence theory states that a belief is true if and only if that belief corresponds to the facts. A formulation Tarski uses for the correspondence theory is that a proposition is true if it designates a particular state of affairs: that is, it refers to the actual state of reality. It is called the correspondence theory because it talks about the relationship between a proposition and something in the world outside of the proposition—facts or states of affairs or something of the sort. There are these things out there in the world and a true proposition is one that corresponds to them. The notion here is not coherence among beliefs but correspondence to something that lies outside of them. The third theory is usually associated with American pragmatism: hence the pragmatic theory of truth. It is that a proposition is true if and only it is useful to believe that proposition. That is, a proposition is true if and only if one’s plans and projects go better by believing it than by not believing it. Truth is utility. A true belief increases utility and a false belief decreases utility. For example, if I hold the false belief that I can jump off a tall building and fly away into the sky, that will likely result in a decrease of utility as I plummet to the ground. Therefore, true beliefs are the ones that maximize well-being. Let’s quickly go over the standard objections to each of these theories. The problem with the coherence theory is that a belief could be consistent with my other beliefs and yet the whole lot could be false. Consistency alone will never make a belief true, because false propositions can be mutually consistent (the belief that the Earth is flat is consistent with the belief that you will drop off the edge if you travel far enough, but neither belief is true). Coherence is just about the relationship between one belief and another, not about whether any of them fit objective reality. A person could

Tarski’s Theory of Truth 

149

have a perfectly coherent belief set and yet all her beliefs are false. To get truth we need to bring in something external to belief. The pragmatic theory of truth has a very similar problem. I could have a belief about something that is useful to me but that belief is false. We can imagine a case where someone lives in a society where having certain beliefs is rewarded and having other beliefs is punished. In Communist Russia, for instance, if you had the belief that the bourgeoisie were wicked, that belief was likely to be rewarded. If you had the belief that the bourgeoisie are meritorious, you would hold a belief that is likely to be punished. It is more beneficial to hold the former belief than the latter, but it doesn’t follow that the former belief is true and the latter false. Therefore utility does not always coincide with truth. At best the two properties are generally correlated. Most philosophers think that the correspondence theory is the best theory. It captures the idea that truth depends on objective reality—not on us. The problem the correspondence theory has concerns more technical issues such as what a fact is and what the correspondence relation amounts to. Are facts complexes of objects and properties? How do we count them? How exactly do they differ from true propositions? Are there general facts and negative facts? It is also difficult to find a clear and correct formulation of the underlying notion of correspondence to reality. Is it a kind of denoting, or some sort of isomorphism? To clarify the correspondence theory is largely the task Tarski sets himself, so let us proceed directly to that. 8.2  Tarski’s Criteria of Acceptability Tarski’s theory is supposed to sweep away all of this confusion and obscurity about truth and replace it with a logically hygienic theory with none of the above-mentioned problems. It is meant to be a nice, clean logical definition about truth—which is why everyone liked it (almost everyone). He says at the very beginning of the article that to arrive at a satisfactory definition of truth we first need to know what the definition is meant to achieve—only then can we properly judge the definition. He then immediately dives into his approach to defining truth. We need to determine what we want the theory to do and what conditions make it acceptable. Here he distinguishes two tests for whether the theory of truth is acceptable. He calls them material adequacy and formal correctness. A good theory

150 

Chapter 8

of truth must be materially adequate and formally correct. To start with, material adequacy means simply that the definition should (in Tarski’s own words) “catch hold of the actual meaning” of the word “true.” In other words, it should not be a theory that stipulates a new meaning for the word “true,” or seeks to reform its meaning; rather, the definition should actually capture what the word “true” means when we all use that word. You might think that this is a trivial requirement, because surely if we are trying to define a word of ordinary language we should try to capture what it actually means. And you would be right: if we are trying to define “know,” for example, we also want to catch hold of the actual meaning of that word. Doesn’t every philosopher interested in defining a word want his definition to be “materially adequate,” that is, correspond to what the word actually means? Sometimes people think that there is a mysterious technical aura surrounding Tarski’s concept of material adequacy, but really he just means capturing the concept of truth that we actually have. Later we will see that he has a more technical formulation of material adequacy, but to begin with he just means that the definition should be accurate. By “formally correct” Tarski means that there should be no logical errors in the definition and that we must specify the formal structure of the language we are using. For example, the definition must not commit use– mention confusions. The theory must be formulated in such a way that it is not guilty of any logical infelicities or lack of clarity. Again, this is a familiar requirement that we should place on any philosophical definition of any concept. No definition can be permitted to be formally incorrect! In the case of truth, Tarski is particularly concerned with the paradoxes that can arise with the word “true” (as with the paradox of the Liar who announces “Nothing I say is true”), so he is especially concerned to avoid logical pitfalls. The next point he brings up concerns the application of the word “true.” We have the predicate “is true” and from the point of view of grammatical form it looks exactly the same as a predicate like “is red.” The predicate “is red” ascribes the property of being red to an object. Similarly “is true” appears to ascribe a property to the thing being referred to. Thus truth is a property expressed by a predicate just as redness is a property expressed by a predicate. But what is it a property of? As Tarski says, it can be applied to different things, three of which he mentions. First, it can be applied to beliefs, which are psychological states: we can say that our beliefs are true

Tarski’s Theory of Truth 

151

(or false). It can also apply to propositions, which are the abstract contents of beliefs. For example, we can say that the proposition that snow is white is true; here we say nothing about anybody’s beliefs. If we apply “true” to a proposition we apply it to something that is not dependent on any particular language or on any believer. The same proposition can be expressed by different sentences in different languages—the sentences that are synonyms or exact translations. A proposition is a kind of abstract entity and we can ascribe truth to that. But we can also ascribe truth, Tarski says, to sentences, which are concrete linguistic entities. We can say that the sentence “Snow is white” is true, where the sentence is conceived as a series of marks or sounds, that is, a perceptible physical entity. The previous sentence contains a reference to a sentence, unlike the one before it. Using quotation marks, we refer to the English sentence “Snow is white.” When we apply the predicate “is true” to a sentence we have to put that sentence in quotation marks. We thereby create a name of a sentence, to which we attach the predicate “is true.” Tarski does a lot of naming of sentences in his theory. The thing about sentences is that unlike propositions they are dependent on language—they are not common between languages like propositions. It therefore changes the logic of the word “true” when you apply it to sentences instead of propositions. Now you are applying it to the tangible vehicle of propositions, not the shadowy propositions themselves. We can also apply “true” to speech acts performed by uttering sentences, as with statements and assertions. All these things can be said to be true or false, despite their variety. Tarski announces that he will take “true” to apply to sentences, so that he is defining truth as it applies to sentences. Thus the extension of the predicate “true” will be the class of true sentences. This affects the form of his definition, as we shall see, particularly with regard to the use of quotation. 8.3  Aristotle and the Redundancy Theory Tarski explains how he came up with the inspiration for his own theory by tracing it back to Aristotle: We should like our definition to do justice to the intuitions which adhere to the classical Aristotelian conception of truth—intuitions which find their expression in the wellknown words of Aristotle’s Metaphysics: To say of what is that it is not, or of what is not that it is, is false, while to say of what is that it is, or of what is not that it is not, is true.1

152 

Chapter 8

For the sake of simplification, we can cut out the negation part of Aristotle’s formulation and still express the essence of Tarski’s view. Truth is saying of what is so that it is so—this is Aristotle’s fundamental idea. If it is so that this table is brown, then it is true to say that the table is brown. That sounds right and it is the basis of what is now called the redundancy theory of truth. To say that a sentence is true is just to say that things are as the sentence says—and that’s it. We could equally have repeated the sentence. Tarski himself never mentions this type of theory by name even though the theory that he proposes is clearly a version of the redundancy theory. Suppose a speaker says, “Snow is white,” and his audience replies, “Yes, that is true.’” What does the second speaker mean when he says this? He could have also said, “Yes, snow is white.” It is bit more long-winded to say, “Yes, snow is white,” because then he would have to repeat exactly what the speaker says. It is much easier to say, “That’s true.” By saying “That’s true” he can reassert everything that the first speaker said in a shorthand form. Thus we can abbreviate our agreement with what someone said by using the simple predicate “is true.” We don’t need to go to the trouble of asserting the whole thing over again. This piece of linguistic machinery prevents us from needing to repeat everything that someone else says. It can be very useful in making a statement like “Einstein’s theory of relativity is true”— sparing us the need to recite the whole theory of relativity! Tarski takes it that sentences containing “true” are equivalent to the sentences to which it is applied. The word adds nothing to the content of the sentences to which it applies. The idea is that the word “true” is strictly speaking redundant. We have it in our language, and we use it for practical purposes, but we could in principle do without it. And so we come to Tarski’s celebrated biconditional: (1)  “Snow is white” is true if and only if snow is white. The predicate “is true” is strictly speaking redundant because the result of applying it to a sentence produces something equivalent to that sentence itself. We could say “The sentence ‘Snow is white’ is true” or simply “Snow is white.” Either way we have said the same thing. The sentence “The sentence ‘Snow is white’ is true” means the same thing as the sentence “Snow is white.” That is the insight behind the redundancy theory—which is sometimes also called the disappearance theory or the disquotational theory. It is as if the

Tarski’s Theory of Truth 

153

predicate “is true” strips a sentence of the quotation marks that enclose it, and then disappears into the night. We disquote the sentence and write it again after “if and only if,” thereby obtaining the definition of “true” as it applies to “Snow is white.” But before getting to the Tarskian technicalities involving these disquotational biconditionals, let us talk a bit more about the Aristotelian view of truth, as Tarski understands it. Actually, the view is very often attributed to Frege, on the strength of this passage from “On Sense and Reference”: “The thought that 5 is a prime number is true” contains only a thought, and indeed the same thought as the simple “5 is a prime number.” It follows that the relation of the thought to the True may not be compared with that of subject to predicate.2

Frege is claiming that a sentence of the form “S is true” expresses the same thought as S. Of course, saying they express the same thought is another way of saying they are synonymous. Thus the sense of the sentence “The sentence ‘Snow is white’ is true” is identical to the sense of the sentence “Snow is white,” because they express exactly the same thought and are exact synonyms of one another. Tarski’s truth biconditionals are just a regimented expression of this Fregean insight. By contrast, the correspondence theory tells us that “Snow is white” is true if and only if it corresponds to the fact that snow is white. Here we invoke, in addition to snow and whiteness, entities called “facts” and a relation called “correspondence.” These raise tricky philosophical and logical questions. With Tarski’s theory we do not have to bother with such questions. There is no need to bring in the concepts of correspondence and facts—we just repeat, “Snow is white” after “if and only if.” And snow being white is not philosophically problematic, because we know what that is—there is no particular philosophical problem about snow being white! This is a very simple streamlined account of what truth is, using no murky notions. We have boiled truth down to the basics. The only real question is a technical one about how we would apply this definition to the full range of sentences. There is in the end no more to the concept of truth than ordinary sentences and what they are ordinarily about. The beauty of this theory lies in its triviality. It does not involve us in any complicated conceptual analysis or controversial notions. Tarski actually rather misstates this aspect of his theory. He seems to think that his theory is a form of the correspondence theory. He explains himself in the following passage:

154 

Chapter 8

If we wished to adapt ourselves to modern philosophical terminology, we could perhaps express this conception [Aristotle’s] by means of the familiar formula: The truth of a sentence consists in its agreement with (or correspondence to) reality.3

Most philosophers want firmly to distinguish the Aristotle/Frege conception of truth from the correspondence theory just stated. The view he here describes is aptly called a correspondence theory, because it talks about a relation of “agreement” between sentences and something called “reality,” but his own theory makes no use of such notions. The idea is to avoid all of that by adopting a redundancy view of truth. Tarski seems to be confusing the classic correspondence theory with the redundancy theory. The latter theory treats “true” as an essentially redundant device, but the former takes truth to be a substantial relation of correspondence between sentences, on the one hand, and facts/existing states of affairs/reality, on the other. Tarski’s actual theory has a very different form, as we shall see. To get started on the details of Tarski’s theory, we must first analyze the basic logical form of his truth biconditionals. Their abstract logical form is the following: x is true if and only if p. The letter “x” is typically reserved in logic for individual variables. Individual variables are what stand in place of names or descriptions or pronouns. The letter “x” is thus a variable that stands in place of a singular term. Of course, the singular term is a part of the sentence and not the whole sentence. Looking at only the left-hand side of the biconditional, for example “The sentence ‘Snow is white’ is true,” we see that it has the form “x is T.” The part where we quote the sentence is a singular term and hence can be replaced by a variable. If we wanted we could give that sentence a name—say, “Burt.” We could stipulate, “Burt is the English sentence ‘Snow is white.’” Then we could formulate the biconditional as “Burt is true if and only if snow is white.” Logically, quotation converts a sentence into a singular term designating itself. Therefore, the logical form of “‘Snow is white’ is true” is “x is T.” In standard logical notation that would be “Fa,” where “a” is a name and “F’ is a predicate (as in “John is bald”). In other words, it is a subject-predicate sentence. However, the sentence on the other side of “if and only if” does not contain a singular term for a sentence—it is just a sentence in use, referring to snow and whiteness. For this reason, in logic the variables conventionally

Tarski’s Theory of Truth 

155

used are “p” and “q.” Traditionally, these letters stand for propositions or whole sentences, and they are not singular terms. That is why you will see truth functions that link together the letters “p” and “q,” as in “p and q.” It would be completely ill formed to put the sentence connective “and” between singular terms that designate sentences, because “and” is a sentence connective and so joins only sentences. It would not be appropriate to put a variable like “x” on one side of “and” and “y” on the other side. If we interpret “x” and “y” in the conventional way, these are variables that stand in the place of names of things. Of course, names and sentences are not in the same semantic category. So the thing on the right-hand side is a sentence and therefore the appropriate variable for that is going to be “p.” Sometimes in logic the “p” is called a schematic letter. So the “x” on the left is an individual variable ranging over sentences and the “p” on the right is a sentence variable or sentence schematic letter. That is the logical form of the sentences Tarski calls “equivalences of the form (T).” “T” is for truth, obviously. Thus we have the generalized form: “x is T if and only if p.” Of course, this biconditional statement has the logical form “q if and only if p.” This is because “x is true” is a sentence and so must be replaced by a sentence variable, but it contains an individual variable “x” that stands in the place of names of sentences. The essential point here is just that on the left-side we have a name of a sentence embedded in a sentence but on the right-side we just have a sentence—and yet these two are equivalent. In other words, “‘Snow is white’ is true” is equivalent to “Snow is white.” The logical form “x is T if and only if p” simply generalizes on this case. One of the things that Tarski is given credit for is how fastidious he is about use and mention: that is, the distinction between using a sentence in the ordinary way to make a statement and referring to a sentence (mentioning it). Employing that terminology, we can say that on the left-hand side of the biconditional, “Snow is white” is mentioned and not used, whereas on the right-hand side the sentence is used and not mentioned. This is all by way of making sure the definition of truth is “formally correct.” 8.4  Object Language and Metalanguage One other piece of logical terminology is important to grasp in coming to grips with Tarski’s theory. This is the distinction between the object

156 

Chapter 8

language and the metalanguage. The object language is the language we are speaking about when we formulate our definition of truth for that language. So far, the object language has been English, because the sentence “Snow is white” is an English sentence. But it could be French or Italian or Chinese. It is just whatever language we are talking about, to whose sentences “true” can apply. We typically refer to sentences of the object language by using quotation marks, though that is not the only way. The metalanguage, on the other hand, is the language that we are using to talk about another language. So far, that has also been English, but it could be any language. A Frenchman interested in defining truth for English will take English as his object language but use French as his metalanguage. The distinction is simply between the language that we talk about and the language that we use to talk about that language. So far, our metalanguage and our object language have been the same language, namely English. But that is not always the case. We could have an object language that is French and a metalanguage that is English. For example, we could say “‘La neige est blanche’ is true if and only if snow is white.” We could equally talk about a Martian language in Swahili when formulating our Tarskian truth theory for Martian. This terminology helps us keep straight what language we are talking about (note that we can also talk about the metalanguage, though we are now using a meta-metalanguage). Just because we use English as both our object language and our metalanguage does not mean that we can ignore the distinction. Most philosophers call the Tarskian biconditionals “T-sentences.” Adopting this terminology, we can say that a T-sentence is a sentence of the metalanguage that mentions (on the left-side) a sentence of the object language. Thus we use the metalanguage to mention the object language when we write down a T-sentence. One other point Tarski makes in this connection is that since we apply the word “true” to sentences, and not to propositions or statements or beliefs, we have to relativize the truth predicate. The sentence “Snow is white” can in principle be true in one language but not true in another language, because that string of marks or sounds could mean different things in different languages. In English, the sentence “Snow is white” means that snow is white, and snow is white, so that sentence is true in English. But suppose some other language contains exactly the same sentence either acoustically or graphically, yet with a different meaning, say, that snow is black. Then in that language the sentence “Snow is white”

Tarski’s Theory of Truth 

157

means that snow is black, but snow is not black, so that sentence is false in that language. Strictly, then, we need to write our T-sentences as follows: “x is T in L if and only if p.” Now we are logically pucker! The T-sentence for the second language (call it Twenglish) will then read: “‘Snow is white’ is true in Twenglish if and only if snow is black.” We don’t have to relativize truth when applying it to statements or beliefs or propositions, because they are not language dependent. The proposition that snow is white is true if and only if snow is white, period. Here meaning is built in. A proposition does not vary in meaning between languages, because it is not a piece of language (similarly for statements and beliefs—the proposition is built in). But if we are defining “true” as it applies to sentences, construed as marks or sounds, then we need to relativize the truth predicate, because of potential variations of meaning from language to language. This is simply because sentences in themselves are just meaningless scribbles or noises. 8.5  How to Derive the T-Sentences What we have on the table so far are two things: a philosophical motivation, deriving from Aristotle and Frege, for focusing on the T-sentences; and some clarification of the logical status of the T-sentences and how they should be analyzed. But we don’t yet have a theory of truth. Here in a nutshell is what Tarski proposes: a definition of the word “true” for a language is materially adequate and formally correct if it entails all the T-sentences for that language. In other words, take all the (indicative) sentences of English and write out a T-sentence for each of these sentences. We then have all of the T-sentences corresponding to all of the sentences in English. A satisfactory definition of truth, Tarski proposes, is a theory that entails all these T-sentences. Here is where he introduces the idea of a “partial definition.” What he is saying is that a T-sentence for (say) “Snow is white” defines the word “true” partially with respect to that sentence. We have given a partial definition of the word “true” for the sentence “Snow is white.” If we now take the sentence “Grass is green” and write out its T-sentence, then we have partially defined “true” for that sentence. And so on. Each of these is a partial definition, and the totality of them is a complete definition of the word “true” for English. If we had such a complete compilation, we would have shown what it means for every sentence in English to be true. That is

158 

Chapter 8

the ultimate goal of Tarski’s theory. The correct and complete definition of the word “true” is something that would entail all those partial definitions. We just need to add them all up to get what we seek. A logically adept student might jump up at this point and say that there is an easy way to get the desired result. We could simply form the logical conjunction of all of these T-sentences. We take all of the individual T-sentences for English and join them together with “and” (“‘Snow is white’ is true if and only if snow is white and ‘Grass is green’ is true if and only if grass is green and …”). The conjunction of sentences entails each conjoined sentence, because in elementary logic “p and q” entails “p” (and also entails “q”). If we had a conjunction of the full range of T-sentences, that conjunction would entail each of those T-sentences. The conjunction would thus entail all the partial definitions, so it would be the complete definition. So start conjoining! The conjunction of all the T-sentences would meet all of Tarski’s requirements, as so far stated. That would be a perfectly adequate definition of truth by Tarski’s standards, except for one small point. There are infinitely many sentences in English. We can generate an infinite number of sentences in a natural language like English, because these languages contain certain devices that enable the speaker to form ever more complex sentences. The most obvious one is “and.” Whenever we have a sentence we can always add another sentence by conjoining it to the previous sentence. If we start with a conjunction, no matter how long the conjunction is, we can always create another sentence by conjoining something else to it. It is the same with negation. We can negate “p” to get “not-p,” and then we can negate that again to get “not-not-p,” and so on. The rules of the English language allow us to negate as many times as we like and therefore produce as many sentences as we wish. Therefore, a conjunction of all the English sentences would be an infinite conjunction—and so in consequence would be the conjunction of all the T-sentences. To use a more precise logical terminology, the resulting theory of truth would not be finitely axiomatized, which means it could not be written down (or even formulated in thought). It would clearly be much better to have a finitely axiomatized theory that entails all the T-sentences. Then we could at least have a look at it! It will turn out that such a theory must analyze each sentence into its constituent parts, and that is why it is of interest to those engaged in semantic theory (see the next chapter). The way Tarski’s theory actually

Tarski’s Theory of Truth 

159

works is that we do not take each sentence as primitive; instead we provide a structural analysis of each sentence, and based on that analysis we generate a T-sentence for every sentence. So we don’t have to form an infinite conjunction of all the T-sentences, even though that would have satisfied Tarski’s condition of material adequacy. Strictly speaking, we should amend Tarski’s condition to read: the theory must entail all the T-sentences from a finite number of axioms. How do we produce something that generates all the infinitely many T-sentences without conjoining them into an infinite conjunction? Tarski proposes that what we want is something that is in effect a logical conjunction of all the T-sentences. He makes this point in the following paragraph: Now at last we are able to put into a precise form the conditions under which we will consider the usage and the definition of the term “true” as adequate from the material point of view: we wish to use the term “true” in such a way that all equivalences of the form (T) can be asserted, and we shall call a definition of truth “adequate” if all these equivalences follow from it. … The general definition has to be, in a certain sense, a logical conjunction of all these partial definitions.4

“In a certain sense” it has to be the logical conjunction of the partial definitions, but not in the straightforward sense of simply conjoining them as they stand. What Tarski has is a technical way of constructing something that is in effect a logical conjunction without being an actual logical conjunction. Soon we will see what that way is. 8.6 Satisfaction Tarski next makes a few points about semantic notions and formal languages. Semantic notions he defines as relational, with the two most important semantic notions being designation and satisfaction. I doubt the Rolling Stones had anything Tarskian in mind when they wrote their song “(I Can’t Get No) Satisfaction,” but the lyrics fit quite nicely. It is indeed not easy to get no satisfaction. As Tarski shows, you need to be quite ingenious to get satisfaction—obstacles must be overcome. These two semantic notions are relational because they connect language to things in the world (I suspect the Stones were singing of relational connections too). An example would be the name “Mick Jagger” designating the writhing entity that is Sir Mick Jagger. Satisfaction is very similar, but satisfaction is a semantic relation that applies to predicates and not singular terms. Satisfaction is a relation

160 

Chapter 8

between objects and predicates. The predicate “white” is satisfied by all objects that are white. More formally, an object x satisfies “white” if and only if x is white. This is very similar to a T-sentence in its form but now we are talking about an object satisfying a predicate, not a sentence being true. These then are semantic notions. Tarski eventually defines truth in terms of these semantic notions, designation and satisfaction. For this reason, he calls his definition the semantic conception of truth. The concept of truth itself is not on the surface a semantic notion, because it is not relational. The predicate “true” is what is called a oneplace predicate. The word “true” is not a relational term like “designates” or “satisfies”—we can’t say “x trues y.” Although Tarski speaks of the semantic conception of truth, the concept of truth is not strictly a semantic notion. However, if Tarski is right, it is definable in terms of semantic notions, and so it turns out to have a kind of semantic deep structure. Truth, for Tarski, reduces to designation and satisfaction. To really understand his construction, we must figure out what satisfaction is and how it works. Tarski also lays out the idea of a formal language. This idea is important to the whole philosophical significance of the theory. English is not a formal language and it is not reducible to the formal languages typically studied by logicians. It has various constructions in it that are not the same as the constructions in any standard logical system. For example, predicate calculus, which is what Tarski is talking about, does not contain any intentional operators (like “believes” and “necessarily”), whereas natural language does contain intentional operators. Tarski is only defining truth for a specific type of formal language, not a natural language such as English (yet the word “true” applies to many sentences of English that cannot be rendered into a standard formal language, as he admits). We can think of a formal language like predicate calculus as a fragment of natural language, containing various stilted idioms and some unfamiliar symbols. Let us take a classic predicate calculus language as our formal language, following Tarski. The point of calling it formal is that you can specify its properties completely formally. Such a language will contain finitely many individual constants symbolized by letters like “a,” “b,” and “c”. It will also contain finitely many predicate constants symbolized by the letters “F,” “G,” and “H.” We can then stipulate that any combination of something in the first list with something in the second list, such as “Fa” or “Hc,” is well formed and counts as a sentence. If there are only three constants in each

Tarski’s Theory of Truth 

161

list, that means there would be nine possible well-formed sentences. Combinations like “abc” or “GHb” are not well formed. This is a toy language where we have specified the primitive vocabulary and the formation rules. Intuitively, we are talking about a language that can generate sentences like “John is bald.” We could now add another category of expressions to our toy language: sentence connectives. We will add these two: “not” and “and.” These are stipulated to produce well-formed sentences when “not” precedes a sentence and “and” sits between two sentences. Thus “not-Fa” is well formed and “Gb and Hc” is well formed. This is how we specify a formal language. We list each of the primitives in the language and then we specify the permissible modes of combination. Finally let us add two quantifier expressions, “all” and “some,” with associated variables and a device for bracketing, so that we get sentences like “For some x, (x is F and x is not-G).” We have now specified a classic predicate calculus language such as can be found in any introductory logic text. The reason we are covering this material is that Tarski’s theory of truth is built around the sentence structures specified in a formal language of this kind. We see how Tarski goes about defining truth for a formal symbolic language in section 11 of his article, entitled “The Construction (in Outline) of the Definition.” He begins: A definition of truth can be obtained in a very simple way from that of another semantic notion, namely, the notion of satisfaction. Satisfaction is a relation between arbitrary objects and certain expressions called “sentential functions.” These are expressions like “x is white,” “x is greater than y,” etc. Their formal structure is analogous to that of sentences; however, they may contain the so-called free variables (like x and y in “x is greater than y”), which cannot occur 5 in sentences.

What he calls a sentential function we have been calling a predicate, which can be satisfied by objects. Satisfaction is a semantic relation between objects and such sentential functions. His explanation sounds very technical, but it is actually straightforward. Satisfaction is really the converse of the relation expressed by “true of.” If I say that the predicate “white” is true of snow, I am talking about satisfaction. I could equally have said that snow satisfies “white.” It is just the converse of “true of.” To specify the satisfaction conditions of a predicate we just write out something of the form “x satisfies ‘F’ if and only if x is F.” This resembles a T-sentence in that on the

162 

Chapter 8

left we mention an expression and on the right we use that same expression (if the metalanguage is the same as the object language). We could call this an S-sentence, by way of analogy with a T-sentence. An S-sentence tells us under what conditions a given predicate is satisfied by an object. We could also say that each S-sentence is a partial definition of satisfaction for the language in question. All the S-sentences together give a complete definition of satisfaction for the language. There are only finitely many basic S-sentences, because there are only finitely many primitive predicates in the language (three, to be precise). These are sometimes called satisfaction axioms. (We can also write out “designation axioms” for the individual constants, which will have the form: “‘a’ designates a.”) We have taken something that is a part of a sentence, the predicate, and then defined the semantic relation of satisfaction for that part, which is analogous to the way we would define truth for a whole sentence. We are left with the following formula for “white”: “x satisfies the predicate ‘white’ if and only if x is white.” We define satisfaction for each of the predicates by using in the metalanguage the expression we refer to in the object language. But from this finite formulation, we can generate infinitely many S-sentences. This is because we can use devices like “not” and “and” to produce arbitrarily complex predicates, such as “x is white and x is cold and x is not ice cream.” This operation is called a recursive procedure, which Tarski explains thus: In defining the notion of a sentential function in formalized languages, we usually apply what is called a “recursive procedure”; i.e., we first describe sentential functions of the simplest structure (which ordinarily presents no difficulty), and then we indicate the operations by means of which compound functions can be constructed from simpler ones. Such an operation may consist, for instance, in forming the logical disjunction or conjunction of two given functions, i.e., by combining them by the word “or” or “and.” A sentence can now be defined simply as a sentential func6 tion which contains no free variables.

He is making the point here that we must remember that in addition to primitive predicates there are also complex predicates built up by using connectives. Consider the complex predicate “is white or red.” An object satisfies “is white or red” if and only if that object satisfies “white” or it satisfies “red.” We can then generalize this over all predicates to get a general rule for “or”: for any predicates “F” and “G,” x satisfies “F or G” if and only if x satisfies “F” or x satisfies “G.” Now we have covered every possible

Tarski’s Theory of Truth 

163

disjunction of predicates with this one axiom. Here is how Tarski explains the idea: To obtain a definition of satisfaction we have rather to apply again a recursive procedure. We indicate which objects satisfy the simplest sentential functions; and then we state the conditions under which given objects satisfy a compound function—assuming that we know which objects satisfy the simpler functions from which the compound one has been constructed. Thus, for instance, we say that given numbers satisfy “x is greater than y or x is equal to y” if they satisfy at least one of the functions “x is greater than y” or “x is equal to y.”7

Once we have a recursive definition of satisfaction we can generate S-sentences for any complex predicate in the language. This means that we get an infinite number of these S-sentences from a finite number of axioms— axioms for each primitive predicate and axioms for each of the connectives used to form complex predicates. In other words, we get the effect of an infinite disjunction of S-sentences from a finite basis. We have analyzed the complex predicates into their parts and then said something general about the parts. This solves the problem raised by the infinity of complex expressions in the language. The theory has been finitely axiomatized. The final stage of the truth definition consists in linking satisfaction to truth. Tarski writes: “Hence we arrive at a definition of truth and falsehood simply by saying that a sentence is true if it is satisfied by all objects, and false otherwise.” In effect, Tarski has recursively defined “true of” by using disquotational S-sentences and then linked “true of” to “true” by invoking the idea of a sentence being true of all objects. This is all really just a technical way to implement the underlying idea of the T-sentences, which themselves already contain partial definitions of truth. Thus Tarski achieves his stated conditions of adequacy. In the next chapter we will look in more depth at the scope and limits of Tarski’s construction, as we examine Davidson’s claim that Tarski-style truth theories provide a framework for doing the semantics of natural languages. Here we will ask what general significance Tarski’s theory has, beyond that of recursively defining “true” for particular formal languages. From a purely logical point of view, it appears that Tarski has accomplished what he set out to accomplish. The more difficult question is the philosophical upshot of his work, if any.

9  Davidson’s Semantics for Natural Language

9.1 Background Tarski’s intention was to define the concept of truth for formalized languages. Donald Davidson’s aim is to use Tarski’s theory of truth for formalized languages to give a theory of meaning for natural languages. Davidson is therefore using Tarski’s theory for a purpose that he did not originally intend—as a form of semantic theory for a natural language. Tarski was restricting his definition of truth to a limited formal language, taking the concept of translation (sameness of meaning) for granted, while Davidson is redeploying his theory to give a theory of meaning for a full natural language. Tarski was explaining the nature of truth; Davidson is using truth to explain the nature of meaning. If Davidson is right, Tarski’s theory has a much greater significance than its originator realized. It is both a theory of truth in a limited setting and a theory of meaning in an unlimited setting. Before we discuss Davidson’s article “Semantics for Natural Language,” let us make a few background comments. Two thoughts about meaning are very much in the air during the whole of twentieth-century philosophy of language, beginning with Frege. One thought is that meaning and truth are somehow intimately connected. A second thought is that meaning is essentially compositional: that is, the meaning of a sentence results from the meanings of its parts. Meaning is something that works constructively, proceeding from simpler elements to determine the meaning of more complex elements by means of rules. Putting the two thoughts together, meaning is something that operates compositionally to generate sentences that are true or false. These ideas are present in Frege’s writings, because when Frege is discussing sense and reference one of his concerns is the reference of parts of

166 

Chapter 9

sentences, and the reference is what determines the truth-value of a sentence. Moreover, sense is the “route to reference,” so that sense is understood in terms of the concept of reference. In his view the reference of a sentence is its truth-value. So sense is what contributes to truth-value via reference. Obviously, whether a sentence is true depends on what it means. The connection between meaning and truth is thus quite overt in Frege and was made more explicit by later philosophers. A simple formulation of the connection is that the meaning of a sentence is its truth condition. Let us talk about that for a moment, so that we get a grip on the underlying ideas before we discuss what Davidson has to say. Take a sentence like our old friend “Snow is white”: it means something. If we wanted to say what it means, the most straightforward way would be to declare: “‘Snow is white’ means that snow is white.” Again, do not suppose that this is trivial just because I have written the same sentence twice. The proposition expressed is no tautology but a contingent, informative proposition. If you know that “Snow is white” means that snow is white, you know something substantial about that sentence. A person who does not know English could know this proposition. I could say of a monolingual Frenchman “Pierre knows that ‘Snow is white’ means that snow is white,” thereby attributing to him knowledge of the meaning of a single English sentence (he need not know what “means” means in English to know that proposition). You need not know the metalanguage in order that this language can be used to describe what you know. I can use English to ascribe knowledge to animals, but I don’t suppose that they know English. Notice that the sentence “‘Snow is white’ means that snow is white” has the characteristic structure that we talked about in connection with Tarski. It both mentions and uses the same sentence. It does not have the same form as “‘Snow is white’ means ‘La neige est blanche,’” in which two sentences are mentioned. This sentence reports the correct translation of an English sentence into a French sentence. So there are two different ways to “give the meaning” of a sentence: one is by mentioning a sentence that has the same meaning as the given sentence (giving a translation), the other is by using a sentence to state the meaning of a mentioned sentence. In the latter case you can know the proposition expressed without knowing the language that is used to express it. Thus we can say of a completely monolingual Frenchman “Pierre knows that ‘La neige est blanche’ means that snow is white” without imputing any knowledge of English to him. But you can’t

Davidson’s Semantics for Natural Language 

167

do that if you quote “Snow is white” after “means,” because then you are ascribing knowledge of an English expression to him. So in our sample meaning-ascription (“‘Snow is white’ means that snow is white”), a sentence is mentioned on the left-hand side and then used on the right-hand side, just like a T-sentence (see the previous chapter). The idea that meaning and truth are connected comes from the very simple observation that we can substitute for the words “means that” the words “is true if and only if.” We thereby obtain something that is well formed grammatically and that duplicates the pattern of use and mention that we have noted. This suggests that to know what a sentence means is to know the conditions under which it is true. To know the meaning of a sentence is to know its truth condition. At the least, to know a sentence’s truth condition is to know something about its meaning. Acquiring such knowledge is removing semantic ignorance to some degree. You may be wondering what a particular sentence in a foreign language means, and then someone tells you that the sentence in question is true if and only if the sky is blue. Haven’t you learned that the sentence means that the sky is blue? Learning the truth condition of a sentence is learning what the sentence means, apparently. At any rate, it is learning something important about its meaning. Let us then entertain the hypothesis that when a person understands a sentence he or she knows what its truth conditions are. Knowledge of meaning is knowledge of truth conditions. Many philosophers embraced this idea about meaning throughout the twentieth century (Wittgenstein in the Tractatus most prominently). Davidson is in the same tradition. He is supposing that meaning and truth conditions are at the very least intimately connected. An issue we will discuss later is whether truth conditions are sufficient for meaning, but they do seem to be necessary, because it would not be possible to know the meaning of a sentence without knowing its truth conditions. How could I know what “Snow is white” means if I was completely ignorant that “Snow is white” is true if and only if snow is white? Still, we may wonder whether knowing the truth conditions of a sentence is sufficient for knowing the meaning of the sentence. To give you an intuitive sense of things, it seems very natural to suppose that “Hesperus is a planet” has the same truth conditions as “Phosphorus is a planet,” because truth conditions are determined by reference. The truth condition that makes both these statements true is that a certain object, namely Venus, is a planet. As we know from Frege, those two names do not

168 

Chapter 9

have the same meaning, however; so identity of truth conditions is not sufficient for synonymy. Referential truth conditions do not add up to sense. We will come back to this. In any case, it seems as though truth conditions are very intimately connected to meaning, because they involve reference, which is fixed by sense. If we did not grasp the truth conditions of a sentence, we would not know its meaning. Thus Davidson’s first idea is that meaning and truth conditions are connected. A theory of truth conditions would therefore be a theory of meaning, or something close to it. Davidson’s second idea is the compositional idea. It is hardly deniable that language forms a compositional structure. There are a finite number of primitive elements (“words”) that crop up in various combinations. These elements join together according to syntactic rules that generate phrases, and these phrases in turn combine to form sentences. A sentence is a complex entity, made up of parts, and these parts can recur in other sentences. It seems just obvious that the meaning of a sentence in a language is derived from the meanings of the elements that compose it—as obvious as the fact that buildings are put together from simpler parts. The units of language are, moreover, exceptionally mobile, because they can jump from one sentence to another—as when I say “John is quick” and “Jill is quick.” We human speakers spend our lives recombining old words into new patterns—we seem to get a kick out of it. Now put those two ideas together and you get the following: the truth conditions of a sentence depend compositionally on the words that compose the sentence. The compositionality of meaning is the compositionality of truth conditions. The meaning of a sentence is its truth condition, and the compositionality of meaning is the compositionality of truth conditions. Thus, if we had a compositional theory of truth conditions, we would have a compositional theory of meaning. The question then is what a compositional theory of truth conditions would look like. 9.2  The Merits of Tarski’s Theory as Applied to Meaning Davidson’s proposal emerges from the background outlined above. He presupposes that background when arguing for the relevance of Tarski to the theory of meaning. Let us trace how he comes to his conclusion. He begins by declaring that a theory of meaning should give the meaning of every meaningful expression. He says this as if it were obvious, but it is not

Davidson’s Semantics for Natural Language 

169

so obvious. Many philosophers have given theories of meaning without supposing that a theory of meaning must actually specify the meaning of every meaningful expression. They proceed at a more abstract theoretical level, saying variously that meaning is an image in the mind or a behavioral disposition or a social practice or a certain kind of intention. Davidson is influenced here by linguistics and Chomsky’s conception of what a theory of syntax should look like. A syntactic theory is a theory that specifies (finitely, recursively) which strings of words are grammatical. It supplies a set of rules that determine which strings are grammatical or well formed. Such a theory is deemed adequate if and only if the rules correctly decide which strings of words are grammatical. It gets pretty detailed and specific. Davidson thinks that a theory of semantics should likewise encompass the whole language, giving meaning rules for every expression. A syntactic theory tells us, for any string of words, whether it is meaningful; a semantic theory (for Davidson) tells us what each meaningful string of words means. However, there is the question what form such a specification of meanings might take. In other words, how should we specify the meaning of every meaningful expression? Davidson does not in this article give any examples of alternatives to the theory that he himself favors, but we can give a few illustrations of what he has in mind. One thing we could do is to specify meanings by presenting what is called a translation manual. We could give a specification of meaning for English by providing a translation of every word and sentence of English into some other language. Thus we say things like “white” means “blanche.” We could also provide synonyms from within the same language, as in “bachelor” means “unmarried male.” We could even give the trivial identity translation: “white” means “white.” The form of these translation manuals is always the same: there is a pair of quoted expressions linked by the relation-word “means” or “means the same as.” If we wanted to do it seriously, we would devise a translation manual that is compositional. We would not want to provide translations for every sentence outright, because they are infinitely many sentences. There would be a finite set of rules by which sentences are translated from one language to another. Davidson does not think that a good theory of meaning should take the form of a translation manual, but that is one obvious way in which we might set about giving the meaning of every meaningful expression. One might wonder whether there is any other feasible

170 

Chapter 9

way. How else can we give the meaning of an expression other than by providing a synonym for that expression? A Fregean might suggest that we assign a sense to every expression in the language. Thus we say things of the form: “The word ‘w’ has sense S.” We saw when discussing Frege that there are problems with that, because there are questions as to how we assign a sense to a word. We need somehow to refer to a sense, but how exactly do we refer to senses? The only way to refer to senses appears to be to link them with expressions, as in “the sense of ‘white.’” Then we end up saying, “The word ‘w’ has the sense of the word ‘w*,’” where “w*” is a synonym of “w.” But this is the translation manual again. So it is difficult to see how we might implement a systematic assignment of Fregean sense to all meaningful expressions of the language that gets beyond a translation manual. Still, this would be one conceivable framework for assigning meanings to expressions. Another approach might be to wheel in psychology. Locke and others thought that the meaning of a word is an image in the speaker’s mind when she utters the word. A specification of meaning might then involve a specification of which images are associated with the word. Thus: “The meaning of ‘w’ is image I.” For example, the meaning of “red” is an image of red. Here the problem is not so much the form of the specification but the plausibility of the underlying theory—because the image theory has been thoroughly discredited (how does it work for the meaning of “not” and “number” and “believes”?). At any rate, those are some possibilities about how we might specify meanings, to be set beside Davidson’s positive proposal. His proposal is quite different and avoids altogether the locution “Word ‘w’ means X,” whatever X may be. His is a theory of meaning in which we don’t talk about the things that words and sentences mean. Davidson’s first point about the proper form of a meaning specification is that it must be one that is structurally based, finitely stated, and capable of generating an infinite output. A natural language like English has infinitely many sentences, so whatever your theory of meaning is it has to specify meanings for all of those infinitely many sentences. It should not perform this function one sentence at a time, because then it would be an infinite specification. What is desirable is a finite number of axioms with an infinite array of consequences, so that the theory of meaning will be something that works recursively. Davidson believes that a theory of meaning must have this recursive shape, and this is one main reason he thinks that Tarski’s theory is right for the job.

Davidson’s Semantics for Natural Language 

171

A point that Davidson brings up in this connection has a Chomskian flavor: the theory must be finite because human languages are learnable. An ordinary child with a finite brain can learn a language that contains infinitely many sentences. So the child’s potentially infinite mastery of the language must be finitely based, that is, grounded in a finite number of semantic principles. The child, being finite, can only learn something that is finitely specifiable. If it were only infinitely specifiable, then no finite being could learn it. Learnable languages must be finitely based. Thus the language must be based on repeatable rules that govern infinitely many potential cases. At any moment you might hear a sentence you have never heard before and understand it in an instant. You never learn the meaning of a sentence by learning the meaning as a whole. The way you understand novel sentences is by analyzing them into their constituent words. Once you understand the rules that combine those words, you can generate from that basis what the sentence means. Our understanding of language is a compositional operation. For a language to be learnable and represented in a finite brain, the language itself must have finitely many basic semantic structures with generative potential. Any theory of meaning must reveal what that generative semantic structure is. If it did not perform that function, it would treat every sentence as a semantic primitive. Such a theory would be inadequate because it would not represent an essential feature of the semantics of natural language, and hence our understanding of language. Meaning must be compositional and languages learnable, so we need a finite semantics. Meaning is also closely linked with truth conditions. So we need a finite statement of truth conditions if we are to capture the essence of what meaning is. This is what we know about meaning before we settle on any specific theory. Davidson is saying that these are general facts about meaning that any theory of meaning should respect. His bold proposal, then, is that Tarski’s truth theory fulfills these conditions and accommodates the general features of meaning we have articulated. Thus Davidson claims that Tarski’s theory has the right form to qualify as a theory of meaning. It is a finite, structural, recursive assignment of meanings (i.e., truth conditions) to sentences, capable of generating a potential infinity of semantic assignments. Let us examine in a particular case how the theory recursively generates truth conditions by analyzing the structure of sentences. We take an

172 

Chapter 9

ordinary English sentence like “Snow is white” and we analyze it into a singular term “snow” and a one-place predicate “is white.” Then we give a designation axiom for “snow”: “‘Snow’ designates (in English) snow.” Now we give a satisfaction axiom for “is white”: “An object x satisfies ‘is white’ (in English) if and only if x is white.” We have broken down the sentence into its constituent parts and assigned semantic properties to the parts. Now we need to derive truth conditions for “Snow is white” based on our axioms. Since this is a subject-predicate sentence, we have a rule that says that such a sentence is true if and only if the designation of the subject term satisfies the predicate term. We then consult our axioms to ascertain what the designation of the subject term “snow” is and what are the satisfaction conditions of the attached predicate “is white.” We find that these are as specified just now. We can then deduce that “Snow is white” is true if and only if snow is white. We simply substitute “snow” for “the designation of ‘snow’” and “is white” for “satisfies ‘is white.’” We broke the sentence into its syntactic parts and then derived the truth conditions from our axioms dealing with those primitive parts. Thus we derived the truth conditions of the whole sentence from the semantic properties of its parts. Since meaning coincides with truth conditions, we have derived the meaning of the whole from the meanings of its parts. If we now add axioms for connectives like “and” and “not,” as outlined at the end of the last chapter, we can derive truth conditions for complex sentences compounded from these connectives, such as “Snow is white and grass is not blue.” And now we have a language with potentially infinitely many sentences in it. The primitive expressions recur in different sentences, so we just need to have axioms that cover those expressions; the full range of sentences results simply from repetition. Thus Davidson thinks that Tarski’s theory performs one of the key functions of a semantic theory: it shows how the meaning of a sentence depends on the words that make up the sentence, because it shows how truth conditions result from the structure of a sentence. Here is a quotation from Davidson summing up much of the above: What properties do we want [of a theory of meaning]? An acceptable theory should, as we have said, account for the meaning (or conditions of truth) of every sentence by analyzing it as composed, in truth-relevant ways, of elements drawn from a finite stock. A second natural demand is that the theory provides a method for deciding, given an arbitrary sentence, what its meaning is. (By satisfying these two condi-

Davidson’s Semantics for Natural Language 

173

tions the theory may be said to show that the language it describes is learnable and scrutable.) A third condition is that the statement of truth conditions for individual sentences entailed by the theory should, in some way yet to be made precise, draw upon the same concepts as the sentences whose truth conditions they state.1

One of the things that Davidson aims to do is articulate the conditions for what a theory of meaning should accomplish. Other philosophers have often neglected this point. He wants us to be clear about what a theory of meaning should aim for, and he gives a set of criteria to determine whether some proposed theory is a good theory or not. We have talked about the first two of those conditions, but we have not yet talked about the third one. A salient feature of Tarski’s theory is that it often evokes a sense of triviality. It is always saying things like “‘Snow is white’ is true if and only if snow is white.” If the same sentence is repeated on the right of the biconditional as appears on the left, this does not appear to be saying much that is interesting about the original sentence. It is not trivial if the object language sentence comes from another language, but in our own language it does seem quite trivial. Shouldn’t we do a bit more to say what the meaning of the sentence “Snow is white” is? Shouldn’t we try to be more ambitious, more informative, and more analytical? We already knew perfectly well that “Snow is white” means that snow is white. Tell me something I don’t know! Davidson thinks that this alleged drawback is actually a virtue of the theory. He thinks it is good that the theory does not draw on any conceptual resources not contained in the sentence with which we started. He thinks the theory should not draw on any fresh or innovative conceptual resources. He does not give any particular arguments or reasons for that position, but his basic thought is that the one thing every speaker indisputably knows is that the sentence “Snow is white” is true if and only if snow is white—and also that it means that snow is white. If our aim is to provide a meaning specification that captures what the speaker meant by uttering a sentence, there are no questions or doubts about that specification when using Tarskian T-sentences. By being conservative in our meaning-ascriptions, we do not go beyond what the speaker ordinarily knows in knowing the meaning of a sentence. We are not attributing to the speaker dubious items of knowledge that he may simply not possess. There is a word for this conservative approach that Davidson does not use in the article we are discussing: “homophonic.” That word means that what is on the right-hand

174 

Chapter 9

side should be the very same sentence as what is mentioned on the left, or a straightforward translation of it. What it should not be is an analysis or reduction or paraphrase or spinning out of the object language sentence (that is, heterophonic). If the T-sentence is homophonic, then we can be sure that it does not attribute to the speaker more knowledge than he actually possesses regarding the truth conditions of the sentence whose meaning he grasps. The only concepts he needs to understand “Snow is white” are the concepts snow and white, so our description of his knowledge should be restricted to those concepts. We may wonder what the homophony requirement rules out. Davidson gives the example of modal expressions. Suppose we are interested in sentences like “Necessarily 2 + 2 = 4” and we want to provide a T-sentence for that. A homophonic T-sentence will simply repeat that sentence neat on the right-hand side. We just peel the quotes off it. But many philosophers have supposed that the semantics of modals should be more adventurous. They suppose for various reasons that it is illuminating to invoke the apparatus of possible worlds. Thus we can analyze the modal operator “necessarily” as a quantifier over possible worlds—as in “for all worlds w.” Adopting this analysis, we can write our T-sentence accordingly: “‘Necessarily 2 + 2 = 4’ is true if and only if, for all worlds w, 2 + 2 = 4 in w.” Davidson would protest that bringing in the ontology of possible worlds introduces conceptual resources not contained in the original sentence. The original sentence said nothing about possible words, nor did it have a quantifier in it. The sentence we began with has been enriched or expanded by bringing in these alien concepts. The speaker might even complain when presented with such a T-sentence: “But I don’t believe in the ontology of possible worlds, and that is not what I meant by ‘necessarily’!” But this issue is controversial because it is not clear at what point we have introduced alien concepts into our T-sentence. A possible world theorist might insist that he has not introduced alien concepts because the ontology of possible worlds is implicitly contained in our ordinary talk about necessity. It is not just a philosopher’s invention—it is the underlying meaning of modal sentences. Do we add alien concepts if we write a T-sentence for “John is a bachelor” by using on the right the sentence “John is an unmarried male”? It becomes rather difficult to argue the issue because it is not always clear what people ordinarily mean by the sentences

Davidson’s Semantics for Natural Language 

175

they use. This is no doubt why Davidson hedges his homophony requirement with the phrase “in some way yet to be made precise.” 9.3  Applying Tarski’s Theory to Natural Languages When dealing with a language based on ordinary predicate logic, Davidson’s use of Tarski’s theory of truth to provide a theory of meaning is quite straightforward. It can be essentially the same theory as Tarski constructed. The Davidsonian theory of meaning will consist of the Tarskian apparatus of base axioms, recursive axioms, and rules of combination. But Tarski recognized that his theory applied only to precise formalized languages, not messy natural languages. Of course, that limited type of language is not the whole of language, so there is a question about the rest of language. Hasn’t the theory dealt with only a fragment of language as we have it? This is already a problem for Tarski’s aim of defining truth, because the word “true” applies to many sentences of English that go beyond the resources of predicate logic languages. He is therefore unable to say what “true” means when it applies to sentences not translatable into the formal language. But the problem has special force for Davidson because he is claiming to apply Tarski’s theory to natural languages in their entirety. If Tarski’s methods do not apply to certain sentences in natural language, then Davidson cannot rely on Tarski to give a complete theory of meaning for natural languages. So Davidson has some obligation to explain to us how we can extend Tarski’s methods to different areas of language. How do we give the meaning of the parts of language that don’t fit the forms of classical predicate logic? Davidson is fully aware of this potential problem. He writes of his style of semantic theory: What would emerge as the deep problems are the difficulties of reference, of giving a satisfactory semantics for modal sentences, sentences about propositional attitudes, mass terms, adverbial modification, attributive adjectives, imperatives, and interrogatives, and so on through a long list familiar, for the most part, to philosophers.2

As he sees it, we need to find ways to assimilate these idioms to semantic forms that are already susceptible to Tarskian treatment. Let’s now consider some of these idioms. Adverbs provide a particularly instructive case. A theory of truth for sentences containing adverbs needs to specify how adverbs contribute to the truth conditions of sentences. We need suitable semantic axioms for adverbs. There is no obvious way to apply Tarski’s

176 

Chapter 9

apparatus to sentences such as “John ran quickly,” simply because there are no adverbs in the formal languages he is concerned with. We certainly can’t say that objects like John satisfy “quickly”—that makes no sense. It is necessary to give a different sort of theory of how such adverbial sentences might work. Davidson accomplishes this by paraphrasing adverbial sentences into sentences that quantify over events and then recasts the adverbs as predicates of events. For example, he would paraphrase the sentence “John ran quickly” as “There was an event e such that e was a running by John and e was quick.” We have thereby replaced the adverb “quickly” with the adjective “quick” and applied it to an event (not to John himself). Then we can give a satisfaction axiom for the predicate “quick” in the usual way: an event e satisfies “quick” if and only if e is quick. What Davidson has done here is translate the grammatically adverbial sentence into a sentence without adverbs and replaced them with adjectives (predicates) that apply to events. Thus the forms familiar from predicate logic are shown capable of including the adverbial constructions of English and other natural languages. Another example involves the so-called intensional operators. These go all the way back to Frege. John believes that Hesperus is a planet, but John does not believe that Phosphorus is a planet, despite the identity of Hesperus and Phosphorus. Since “Hesperus” denotes the same planet as “Phosphorus,” we find that we cannot substitute co-denoting names inside belief contexts. Contexts of this kind are called opaque. As Frege pointed out, the truth conditions of sentences containing intensional operators like “believes that” depend on the sense of the embedded name, not the reference. We therefore cannot have a comprehensive axiom for a name that simply gives its reference, because that cannot capture the contribution that the name makes to sentences containing intensional operators. Sometimes the name affects the truth-value of a sentence in a way that goes beyond its reference and brings into play what Frege calls sense. For this reason, our account of the semantics of names is incomplete if it only gives their reference. Something else must be added to that, and it is not clear how we can accommodate these cases within the framework Tarski laid down. Tarski’s theory just specifies references for singular terms by means of designation axioms. Sense is ignored. For Tarski’s own purposes, that is fine, because he is only interested in defining truth for languages without intensional operators. But Davidson wants the Tarskian framework to apply

Davidson’s Semantics for Natural Language 

177

to all the linguistic constructions of natural language—and that is a much taller order. How can a semantics devised for purely extensional languages deal with intensional languages? As it happens, Davidson does have a theory of intensional contexts that is ingenious and purports to solve the problem (see his paper “On Saying That”3). Consider the sentence “John says that the sky is blue.” Davidson proposes that we analyze that sentence in the following way: “The sky is blue. John said that.” We divide the original sentence into two, separated by a period, but linked by the demonstrative “that” that refers back to the first sentence. It is just like you saying something and I reply, “I just said that!” The point of this analysis (sometimes called the paratactic theory) is that by removing the embedded sentence we abolish the intensional operator. There is no opaque context left anymore. In the sentence “The sky is blue” by itself we can substitute any co-denoting term and not alter the truth-value of the sentence. It does not occur within an intensional context as part of a complex sentence. It is a separate sentence, so everything here is extensional. We can therefore apply Tarski’s extensional theory and not run into any problems. Likewise, the sentence “John said that” is completely extensional—in particular, we can substitute any term that refers to the same thing as “that” and not alter truth-value. This demonstrative can be taken to refer to the proposition expressed by the first sentence, so any term that refers to the same proposition will not change truth-value. So, by ingenious paraphrase, we can bring these apparently intensional contexts into the Tarskian fold: they turn out to be extensional after all. (There is a lot more that can be said about this proposal of Davidson’s, as well as about his theory of adverbs, but we are just trying briefly to provide a flavor of how he would go about extending Tarski’s framework to natural languages.) There is also the matter of nonindicative sentences, which apparently lack truth conditions altogether. The imperative “Shut the door!” does not appear to be true or false. A straightforward method here would be to translate these sentences into indicative sentences. We paraphrase “Shut the door!” into “I order you to shut the door.” The latter sentence can be true or false, depending on whether I did indeed order you to shut the door. And it would generally be true because in saying “I order you” I did order you (these kinds of speech acts are called performatives). Again, we find a suitable paraphrase of the original sentence that can be subjected to the Tarskian treatment—since the paraphrase does have truth conditions.

178 

Chapter 9

These examples illustrate the kind of massaging that needs to be done on natural language sentences in order to make the Tarskian semantic framework applicable to natural languages. At the least, Davidson would say, it is not obvious that we cannot extend Tarski’s truth theory further than initially appears. Trying to do so constitutes a research program (this means it will keep eager graduate students busy for a few years). The case of indexicals poses an awkward problem for the homophony requirement. Suppose we give a homophonic T-sentence for “I am hot”— that is, “‘I am hot’ is true (in English) if and only if I am hot.” The problem is obvious: no one can say truly “I am hot” unless I (Colin McGinn) am hot. But people other than me can be hot and can truly say they are with the sentence “I am hot”—without me being hot. Davidson’s homophony condition clearly fails here. We obviously need to write our T-sentence along the following lines: “‘I am hot’ is true for speaker S at time t if and only if S is hot at t.” That is the correct truth condition for the English sentence “I am hot.” Fine, but that T-sentence is not homophonic, because the right-hand side does not repeat the sentence mentioned on the left. We must eliminate the word “I” altogether and add in “S” and “t.” We thus use conceptual resources not present in “I am hot”—the right-side is just not a synonym of the sentence mentioned on the left. Homophony violated! That seems the right way to go, but then we wonder how Davidson can stipulate his homophony requirement in the first place. How can he formulate it so as to rule anything out, while making an exception for indexicals? Add to this the point that his treatment of adverbs also seems to violate homophony, with the added quantifiers and ontology of events, and the requirement begins to lose any bite. How can it rule out possible world paraphrases of modal idioms, say, if it lets in nonhomophonic T-sentences for indexicals and adverbs? Davidson’s theory of meaning makes no attempt to define the semantic primitives. There is only an assignment of logical form. He makes a sharp distinction between defining the primitive expressions and giving the logical form of sentences. In his way of looking at things, base axioms for primitive terms would be something like the following: “‘Snow’ designates snow,” and “An object satisfies ‘white’ if and only if it is white.” His theory analyzes the logical structure of sentences but it does not analyze individual words. The theory will tell us that a sentence is made up of a singular term and a one-place predicate or that a complex sentence is a conjunction, but

Davidson’s Semantics for Natural Language 

179

it won’t tell us that “bachelor” means “unmarried male,” say. This kind of theory has been described as modest, because it refrains from venturing into the analysis of word meaning. The label is not very apt, because giving logical form is not something that is trivial or obvious or uncontroversial. But the idea is that giving logical form is something quite different from analyzing individual expressions. The former is thought necessary and admirable, while the latter is taken to be optional and vaguely taboo. Describing logical form involves determining the semantic categories of words. This can be quite nontrivial. Consider again the word “snow” and the sentence “Snow is white.” If we treat that sentence as having the logical form of a straightforward subject-predicate sentence, as we have so far, then we are treating “snow” as a singular term, a name of snow, whatever snow may be (is it the totality of all lumps of snow or is it more like a Platonic universal—the Form of Snow?). Then, we would write an axiom for “snow” that would be just like the axiom for a name like “Hesperus” (“‘Snow’ designates snow,” “‘Hesperus’ designates Hesperus”). But if we thought that the word “snow” is not a singular term but really a predicate, then we would formulate its axiom in the following style: “x satisfies ‘snow’ if and only if x is (a piece of) snow”—it would get a satisfaction axiom, not a designation axiom. Such a semantic classification would award a different logical form to the sentence “Snow is white.” Instead of having the logical form “Fa,” a singular term plus a predicate, it would have the logical form of a universal quantification, as in “For all x, if x is (a piece of) snow, then x is white.” The word “snow” would then be placed in a different semantic category—that of predicate, not singular term. (In fact, “snow” is what is called a “mass term,” and we have just described two ways of handling mass terms—as names or predicates.) Similarly, in Davidson’s treatment of adverbs, a word like “quickly” is transformed into a predicate in the assignment of logical form. In his treatment of indirect discourse, the word “that” in “John said that the sky is blue” gets classified as a demonstrative and hence as a context-dependent singular term. None of this semantic classifying is especially modest—it’s pretty bold. Formal languages are supposed to be unambiguous, so what about ambiguity in natural languages? For instance, the word “bank” is ambiguous, meaning either the bank of a river or a bank for money. This is called lexical ambiguity. But there is also syntactic ambiguity, as in the example Davidson cites: “They came by slow boat and plane”—was it just the boat that was

180 

Chapter 9

slow or also the plane? Clearly truth conditions will vary when there are ambiguities, so we will need to resolve ambiguities before constructing our T-sentences. We don’t want to end up with monstrosities like “‘Samantha lay down on the [river] bank” is true if and only if Samantha lay down on the [money] bank.” Here we might simply index the word “bank” so that the ambiguity is removed—“Rbank” and “Mbank.” For the syntactic ambiguity, a bracketing device would suffice, as in “They came by (slow boat and plane)” and “They came by (slow boat) and plane.” (This bracketing device is used in standard logic to indicate scope.) It is important to note that the T-sentences themselves are not the whole story. They assign truth conditions, and hence meanings, but the meat of the theory does not reside in the T-sentences alone. There is also the proof of the T-sentences. Davidson makes the point that we have to derive the T-sentences from a finite set of axioms that reflect a recursive structure— the repeatable occurrence of semantic primitives. The illumination comes not just from the final results—the theorems—but also from the process of deriving the theorems from an analysis of the semantic structure of the sentences. We see how the constituent words generate the sentence’s truth condition. The point for Davidson is that the theory must be structural and hence explain how an infinite language can derive from a finite foundation. There is much more to Tarski’s theory than the output of T-sentences, lovely as these are; there is the whole complex apparatus of axioms and derivations that generates that output. It’s the journey as well as the destination. Another merit Davidson sees in this theory is that it allows us to give a theory of meaning without postulating meanings as entities. W. V. O. Quine lurks in the background here. Quine is notorious for his antipathy to meanings as entities (“creatures of darkness,” he calls them, threats to clean living, etc.). Quine wonders how we might set about counting these elusive entities, distinguishing them one from another. How many meanings are there in this book, for example? Davidson thinks it’s a big advantage of Tarskian semantics that there is no need to assign any “meanings” to words in the theory of meaning. This is a theory of meaning that does without any special entities called meanings (senses, intensions). Instead, it assigns references to words—and references are honest, civilized citizens, not murky shadows hovering in the vicinity of words. We say, “‘Hesperus’ refers to Hesperus” with confidence in our theory, but we say nothing

Davidson’s Semantics for Natural Language 

181

about any alleged semantic wraiths describing themselves as “senses.” Yet we succeed in saying what sentences mean (supposedly: see below). In the case of predicates, the theory assigns no entity at all, not even a reference. We simply reuse the predicate in our satisfaction axiom. Consider again an axiom like the following: “x satisfies ‘white’ if and only if x is white.” Notice that there is no reference here to anything that might be denoted by the predicate “white.” We could have said, “‘White’ designates Whiteness,” but we said no such thing. Instead, we say that an object satisfies “white” if and only if it is white, with no reference to any supposed abstract entity named “Whiteness.” There is no singular term in this statement for anything that is being assigned to the predicate—no properties, universals, senses, or the like. The axiom gives a condition under which the predicate is satisfied but without committing us to any strange entities of the kind that so horrified the fastidious Quine. The only entities referred to in the satisfaction axiom are the ordinary objects we need anyway—the spatiotemporal objects that are white. Similarly, Tarski does not interpret the connectives by specifying a reference for them—he doesn’t say “‘And’ designates conjunction.” He just says a sentence of the form “p and q” is true if and only if “p” is true and “q” is true. Using the word “and” on the right-hand side does not mean we must assign any reference to the word. It is a theory of meaning without things called “meanings”—without peculiar semantic entities. Words and sentences mean something, and we can say what they mean, but there are no meaning entities such that words and sentences mean them. So Quine need not worry that talk of “theories of meaning” threatens to unleash a disreputable ontology of meanings that besmirch his clean and tidy universe. 9.4  Empirical Truth Theory With meanings safely out of the way, Davidson broaches the question of the empirical status of Tarski-type truth theories. That is, how would you verify that a particular theory is correct? There are two cases to consider: the case where the object language and the metalanguage converge, and the case where the object language is different from the metalanguage. Let us take the simple case where we are giving a truth theory for our own language. How could we verify that its theorems are correct? Davidson suggests that it is quite simple to do so, because we can look at the theorems

182 

Chapter 9

and see from their very form whether they are correct. If the theory says “‘Snow is white’ is true if and only if snow is white,” we can see immediately that it has to be correct. But if it said “‘Snow is white’ is true if and only if the stock market is about to collapse,” we would know it had gone badly wrong somewhere, since that is a far cry from what “Snow is white” means. Our own semantic competence enables us to judge whether the theory has got the truth conditions right for a given sentence. The T-sentence is empirically correct just when the sentence used on the right is the same as the sentence mentioned on the left. It is thus easy to tell in our own case whether the T-sentences are correct. (Actually, he seems to be forgetting here that not all T-sentences are homophonic. Is it so easy to judge that a T-sentence incorporating his own theory of adverbs is correct? Certainly we cannot just check to see if we have the same sentence twice, since we don’t. It might be quite controversial whether the following T-sentence is true: “‘John ran quickly’ is true if and only if there is an event e such that e was a running and e was by John and e was quick.” But it is at least true that we can decide such questions by consulting our own competence, since we do understand “John ran quickly.”) Davidson makes the rather interesting observation that it can be easier to judge the truth of a T-sentence than to judge the grammaticality of the sentence at issue. He writes: It may in fact be easier in many cases for the speaker to say what the truth conditions of a sentence are than to say whether the sentence is grammatical. It may not be clear whether “The child seems sleeping” is grammatical; but surely “The child 4 seems sleeping” is true if and only if the child seems sleeping.

This implies that it is easier to know what a sentence means than to know whether it is meaningful. One might have thought that we first decide if a sentence is meaningful and then inquire into its meaning, but if Davidson is right it can be the other way around. But how far can this go? Do I know that “The ocean swims nightly to itself” is true if and only if the ocean swims nightly to itself, even if I doubt that this sentence makes sense? What about “‘Dawn and not sun upward grim’ is true if and only if dawn and not sun upward grim”? Or “‘The’ is true if and only if the”? Repetition is surely not enough if the sentence is totally whacky. These are Davidson’s reflections on the domestic case, but what about verifying a truth theory for a foreign language? How do we know when we have got someone else’s truth conditions right? We can’t consult our own

Davidson’s Semantics for Natural Language 

183

competence in the language, because we don’t have any. We have to find out what the alien speakers mean by their words. At this point in the paper Davidson alludes to Quine’s discussion of radical translation. Quine has a famous thought experiment where a traveler goes to a foreign land and comes across a tribe of speakers whose language has never been translated. The traveler fancies himself a field linguist. He engages in radical translation, translation from scratch, with no dictionary. Quine’s question is how he would begin the process of radical translation and whether it is possible to arrive at a definitively correct translation scheme for the language. Davidson takes over that question because he is interested in how we might verify a truth theory for a radically foreign language. In other words, he wants to determine how we could empirically assign truth conditions to sentences. Quine gives the example of the word “gavagai” in the radical translation thought experiment. The traveler immerses himself in the tribe’s culture, observing their linguistic behavior, and sets out to discover what they mean when they utter the word “gavagai.” The traveler has no dictionary because this is radical translation. How could our traveler figure out the meaning of the word? There is no use asking the natives because he won’t understand what they say, and they don’t speak his language either. The first thing he would do is to discover when and where they utter “gavagai”—in response to what immediate sensory presentations. What, say, are they looking at when they say “gavagai”? Suppose our traveler notices that the natives say “gavagai” just as a rabbit runs by them. Our traveler might conclude that he knows what “gavagai” means—it means “rabbit.” The general idea is that he looks around the natives when they utter the word and makes a hypothesis about its meaning. We might agree with the traveler that the correct translation of the word “gavagai” is indeed “rabbit,” because the natives are observed to utter that word if and only if there is a rabbit running by. Since our traveler is a keen student of Tarski, he records his hypothesis in the form of a satisfaction axiom: “x satisfies ‘gavagai’ if and only if x is a rabbit.” Quine would put the situation by saying that a rabbit is part of the stimulus meaning of the word. The natives are stimulated to utter “gavagai” just when there is a rabbit in their sensory vicinity. If you trace the stimulus back from their sense organs to the environment, you will find a rabbit at the other end. Now Quine makes his killer point: even though the natives might intone “gavagai” when and only when they see a rabbit, that does

184 

Chapter 9

not necessarily imply that “gavagai” means “rabbit.” As logicians say, it does not imply that the set of rabbits constitutes the extension of “gavagai.” Even though it is quite true that rabbits are included in the stimulus meaning, there are other things that are included in the stimulus meaning too. One of the things that is included in the stimulus meaning of “gavagai”—in addition to rabbits—is parts of rabbits, say, the ears. So “gavagai” might mean “rabbit ears.” Whenever there is a rabbit present, rabbit ears are also present. Of course, there could be a case where our traveler holds in his hand a pair of severed rabbit ears, and he finds that the natives don’t utter “gavagai” in the presence of the ears alone. He can then rule out the rabbit ear hypothesis. But then it occurs to our wily translator that “gavagai” could still mean, “rabbit ears still attached to a living rabbit.” And then he realizes that it could also mean “temporal stage of a rabbit” or “retinal cause of my sensations of rabbit” or even “visual percept of a rabbit” (the native never utters “gavagai” unless he is perceiving a rabbit). It might indeed mean “rabbit flea” so long as rabbits and their fleas keep close enough company. The point is that there may be many things in the causally operative environment (or even in the native’s own heads) that the word might mean. We cannot easily determine just what specifically the word means (what its extension is). As a result of these reflections, Quine comes to the startling conclusion that what the native means is radically indeterminate (and he even extends the indeterminacy thesis to what we mean by our words). There is no “fact of the matter” as to what “gavagai” means (or what our word “rabbit” means come that that). Davidson is not in this paper concerned with indeterminacy, though elsewhere he expresses general agreement with Quine’s thesis. His concern here is with Quine’s general picture of how we set about devising and testing interpretations of the language of others. This brings us to Davidson’s theory of what he calls radical interpretation. Davidson has a whole article (“Radical Interpretation”5) that delves into this question. Here we must be brief. Roughly, he thinks that we need to assign truth conditions according to the external environmental causes of utterances. If a native “holds true” a sentence just when a certain state of affairs objectively obtains in the environment, we must suppose that the sentence is true just when such a state of affairs obtains. If this leaves gaping indeterminacies, then so be it. As a way to constrain our interpretations further, Davidson advocates what is called the principle of charity—that is, the interpreter must

Davidson’s Semantics for Natural Language 

185

interpret his subjects in such a way that their beliefs and assertions come out mainly true. We cannot suppose that our native is massively in error, totally deluded, riddled with false beliefs. Of course, the native could be mistaken about there being a rabbit in front of him when he utters the word “gavagai”—he might be prone to rabbit hallucinations (that strange plant he insists on smoking all the time). But Davidson maintains that we have to attribute mainly true beliefs to our native if we are to understand him at all. Speakers are uninterpretable (indeed, for Davidson, impossible) unless the principle of charity applies to them. Since we ourselves are interpretable (and apparently possible) that means that we cannot be massively in error either. And that implies that skepticism about our beliefs has to be wrong: we must have mainly true beliefs, no matter what the skeptic says. There is a whole discussion about this set of issues, ranging into philosophy of mind and epistemology, but we will not cover it here. We have said enough to indicate how Davidson views the project of verifying theories of meaning for alien speakers. 9.5  Criticisms of Davidson’s Theory Let us assemble some criticisms of Davidson’s theory of meaning. First we can ask whether Davidson says enough about what meaning is—and about what our grasp of meaning consists in. His guiding idea is that a theory of meaning assigns truth conditions to sentences, and a speaker’s understanding of a sentence consists in his knowing what the truth conditions are. Thus, to understand “Snow is white,” the speaker merely needs to know that this sentence is true if and only if snow is white. This account of meaning raises the question: is it enough to say that knowledge of meaning is knowledge of truth conditions—especially when we restrict ourselves to homophonic statements of truth conditions? Isn’t this just way too minimal? Can’t we ask what this knowledge of truth conditions itself involves? There are different options to take in response to this line of criticism. Davidson’s response would be that we do not need to dig any deeper into linguistic understanding to have an acceptable theory of meaning. Maybe a psychologist could say more about linguistic understanding, but from the point of view of philosophical semantics we have achieved our aim, namely, to specify meanings systematically and show how an infinite mastery can proceed from a finite basis. To venture further is to stray into vague

186 

Chapter 9

and swampy territory. If we stick with Tarski, pure and simple, we have all the security of rigorous formal logic, without speculations about what might be secretly on the speaker’s mind when she understands sentences. Alternatively, we might borrow an idea from Wittgenstein’s Tractatus. He there takes the view that when a speaker understands a sentence he grasps what possible state of affairs would make the sentence true. To understand “Snow is black” you must grasp the state of affairs that would make that sentence true. That is a merely possible state of affairs, not an actual state of affairs. We apprehend possibilities by the faculty of imagination. We imagine such a state of affairs when we grasp the meaning of “Snow is black.” So when I understand the sentence “Snow is black” what I do is imaginatively conjure up a possible state of affairs where snow is black. I might form a mental image of black snow. The fact that I imagine that state of affairs and not some other state of affairs is what my grasp of the sentence’s meaning consists in. If I were to imagine the state of affairs of snow being blue, I would not have imagined the state of affairs that corresponds to the sentence “Snow is black”—I would have misunderstood the sentence. This Wittgensteinian account of knowledge of truth conditions goes beyond Davidson’s austere minimalist account. It is Tarski plus the modal imagination. The speaker has to employ her modal imagination in order to get her mind around meaning. This is a richer psychological story than Davidson’s proudly modest account. It attempts to illuminate nontrivially what knowledge of truth conditions involves psychologically. Another approach that many philosophers have favored is to bring in the notion of verification. The ability to verify the sentence “Snow is white” is what knowing its truth condition amounts to. To verify this sentence we need to seek out some snow, check it over, and decide what color it is. We need to see with our own eyes that it is white. To do that, we have to know where to look—we have to know that snow falls from the sky and covers the hills and dales in winter. If someone tried to verify the sentence “Snow is white” by examining the lava spewing from volcanoes, he would show himself not to understand “Snow is white.” The ability to verify the sentence in the right way is clearly connected to knowing its truth condition. If you know the truth condition of a sentence, you generally have a pretty shrewd idea of how to verify it. If you don’t, you are clueless. Some philosophers (often describing themselves as positivists) try to elevate these truisms into a theory of what knowing truth conditions is—it comes down to

Davidson’s Semantics for Natural Language 

187

knowing what kind of evidence counts in favor of the truth of a sentence. This turns knowledge of truth conditions into knowledge of verification conditions. The view may be horribly misguided, but at least it attempts to spell out what knowing truth conditions might be. (The right view is that we generally have two pieces of knowledge regarding a sentence: knowledge of what state of affairs would render it true, and knowledge of the kind of evidence that would warrant assent to it.) A second line of criticism brings us back to Frege. Tarski’s axioms for names are designation axioms, and so they only assign reference to names. For Tarski that is fine, because sentences containing names are true only depending what the names refer to. If we are only interested in defining truth, it does not matter what name we use, so long as denotation is preserved. If “Hesperus is a planet” is true, then “Phosphorus is a planet” is also true. But those sentences do not mean the same. This is why Frege brought in sense to beef things up. We need to assign more than reference to a name if we are to capture its full meaning. We need something like sense. But Tarski’s semantic apparatus does not specify sense. How then can it function as a theory of meaning? At best, it is a theory of reference. A third criticism is that Davidson’s theory provides no explanation of how words come to have semantic properties. The axioms say things like “‘Hesperus’ denotes Hesperus,” but nothing in the theory tells us how it is that a word like “Hesperus” acquires reference. Similarly for predicates and satisfaction. The axioms don’t explain what gives marks and sounds the semantic features they have. What constitutes reference? Many philosophers of language have felt that we need an explanation of relations like denotation. We can’t just accept them as primitive. In other words, a satisfactory theory of meaning must propose an account of denotation. Some tough-minded philosophers have even undertaken to explain reference and satisfaction in physical terms. But in Davidson’s Tarski-based theory, denotation is taken for granted. At the least, we need to supplement the Tarskian semantics with some sort of explanatory theory of denotation; it is not by itself a complete account of meaning in natural languages. Fourth, Davidson distinguishes sharply between giving logical forms for sentences and giving analyses of individual words. But how robust is that distinction? The intuitive idea Davidson is working with is that in attributing logical forms we do not break words down into parts, but in lexical analyses we do. He is skeptical about the whole idea of analyzing primitive

188 

Chapter 9

predicates, but he is an enthusiast of logical form attributions. But consider Russell’s theory of descriptions (see chapter 3): here we break the word “the” down into a complex quantified conjunction. Why isn’t this lexical analysis? It certainly involves taking a unitary word and analyzing its meaning into separate and more primitive parts. How does this differ from analyzing “bachelor” as “unmarried male”? Similarly, Davidson’s own theory of adverbs construes sentences containing adverbs as quantifications over events with predicates of events. The logical form here is quite different from the superficial syntax of the sentence. The paraphrase finds hidden semantic complexity in adverbs. Why isn’t this a case of lexical analysis? What about modal words like “possibly”? A standard account has it that “possibly” means “There exists a possible world.” The modal adverb goes over into an existential quantifier over worlds. This looks like an exercise in conceptual analysis, but it is also an attribution of logical form. If we want to know what the logical form of “possibly p” is, we are told that this sentence means the same as “There exists a world w such that p in w.” But this is at the same time a conceptual analysis of “possibly.” Again, there is no clear distinction between accounts of logical form and lexical analyses. The alleged distinction evaporates on closer inspection. Yet Davidson seems wedded to ruling out lexical analysis while championing logical form assignment. One suspects that he has bought into Quine’s rejection of the analytic–synthetic distinction, while seeing the merits of certain theories of the meaning of syntactically simple terms. In fact, these two stances are in tension. However, this is a subject beyond our current purposes, so we will not pursue it. We must finally scrutinize a rather fraught passage from Davidson: A theory of truth entails, for each sentence s, a statement of the form “s is true if and only if p” where in the simplest case p is replaced by s. Since the words “is true if and only if” are invariant, we may interpret them if we please to mean “means that.” So construed, a sample might then read “‘Socrates is wise’ means that Socrates is wise.”6

Davidson seems to believe that we can substitute “means that” for “is true if and only if” in a Tarskian T-sentence (“if we please”) and say essentially the same thing (what this has to do with the “invariance” of “if and only if” remains a mystery). Thus a truth theory can do duty as a meaning theory. The gap from truth to meaning has been crossed by this simple substitution. If Davidson does believe this, he is wrong. The biconditional “is true if and only if” does not mean “means that”—not by a long chalk. In elementary

Davidson’s Semantics for Natural Language 

189

logic “if and only if” is called a “material biconditional” and a sentence containing it is true when the two sentences on either side of it are true. Thus “Snow is white if and only if grass is green” is a true sentence. By the same token, “‘Snow is white’ is true if and only if grass is green” is true, if “if and only if” is the material biconditional (i.e., is a truth function). Now perform Davidson’s substitution, replacing “if and only if” with “means that.” We obtain the sentence “‘Snow is white’ means that grass is green.” That is false, egregiously so. The English sentence “Snow is white” most certainly does not mean that grass is green! If Davidson were right, any sentence of English would mean the same as any other sentence sharing its truth-value. That would be the total collapse of meaning and disqualifies any theory of meaning entailing it from serious consideration. But it may be replied that this is only if we adopt the material biconditional interpretation of “if and only if.” Davidson does appear to be doing just that, but maybe this is a slip. Can’t we offer him a stronger biconditional? Not the material biconditional but the strict biconditional. The strict biconditional requires not just actual identity of truth-value for the joined sentences but identity of truth-value in all possible worlds, that is, necessary coincidence of truth-value. The sentences “Snow is white” and “Grass is green” have the same truth-value in the actual world but not in all worlds, because in some worlds grass may be blue and snow is still white. But we can quickly see that this will not solve the basic problem. Suppose we have a sentence like “2 + 2 = 4 if and only if 3 + 3 = 6.” The joined sentences are both true in all possible worlds, so this biconditional sentence is true under a strict modal interpretation of “if and only if.” Now we can run through the same argument again. If we replace “if and only if” with “means that” in a T-sentence we get: “‘2 + 2 = 4’ means that 3 + 3 = 6.” That is no better than before. This ascription of meaning is quite wrong. The simple fact is that “means that” is far more restrictive about substitutions within its scope than “is true if and only if,” no matter how strict you are about the biconditional. The only way to get something that adds up to “means that” out of “is true if and only if” is to stipulate that you are going to mean the former by the latter. But that would be a futile verbal ruse, gaining us nothing. It would also completely wreck the idea of using Tarski’s theory of truth as a theory of meaning, since the words “is true if and only if” would no longer mean what they now do. In sum, the passage from Davidson is an error.

190 

Chapter 9

Davidson proposes that a theory of meaning should specify the meanings of all meaningful expressions. But he does not try to explain how words and sentences come to have the meaning they have. He just takes it that they have meaning. But they certainly don’t have meaning in virtue of their identity as marks or sounds—their meaning somehow comes from outside of them. Where does it come from? How do words mean what they do? Does God miraculously confer meaning on them by some sort of divine intervention? That seems far-fetched. Surely words and sentences have meaning in virtue of their relation to us, the users of those words and sentences. But what exactly is that relation? How do the words we use come to have meaning in virtue of our use of them? That is the topic of the next chapter.

10  Grice’s Theory of Speaker Meaning

10.1  Background: Speakers and Sentences We now move to a discussion of a short but groundbreaking paper by H. P. Grice, called simply “Meaning.”1 The paper is very dense and there isn’t much reiteration of points, so it calls for careful reading. Let’s begin by explaining the larger project that Grice was trying to make progress on in this paper. He is really interested in how words and sentences come to mean what they do mean—how word meaning and sentence meaning arise. What makes bits of language express meaning? Grice has a very natural, intuitive answer to that question: it has something to do with the way speakers mean things. It is not that words mean what they do because there is some fact of nature that makes them do so. It is not as though meaningful words came preformed in nature and humans decided to exploit this naturally given fact. Words are not like apples on a tree, patiently waiting for us to pick them. Meaningful language is not an independent phenomenon that we tap into. Language does not predate the existence of speakers. English was not just lying around and we discovered it. Words in themselves are just sounds or marks that we produce with our voices or our hands—there is nothing about them intrinsically that determines what they mean, or that they mean anything. The meaning of words is arbitrary and conventional, the result of a kind of decision. Meaning is conferred on words. But it is not conferred by nature or by God—it is conferred by us. It is what we humans do with words that makes them mean what they do. Presumably this brings in the human mind somehow, because it can hardly be the human body that gives meaning to words (the kidneys, the toes). Grice focuses on the notion of an agent meaning something by his or her actions. More specifically, he introduces the notion of speaker meaning.

192 

Chapter 10

Not only do words and sentences mean things; speakers mean things by their words. We use the word “means” in both ways. We can say that the sentence “Snow is white” means that snow is white, and we can also say that a speaker meant that snow is white by uttering the sentence. We must distinguish sentence meaning from speaker meaning—words do the former, human agents do the latter. But we must also inquire into how the two types of meaning are connected. Grice proposes that sentence meaning derives from speaker meaning. It is because people mean things by their words that those words come to mean what they do. We have not yet explained or analyzed the notion of speaker meaning, but the notion is already familiar enough for us to understand the thesis that speaker meaning is the basis and origin of sentence meaning. Words come to mean what they do in virtue of the fact that we mean various things by them. We confer meaning on them by meaning something by them. Linguistic meaning thus comes from us—we create it by acts of speaker meaning. Following that initial thought, Grice is proposing that we analyze word meaning in terms of speaker meaning. If we can do that, we will have explained how words mean what they do. This will be a major philosophical achievement. But first we need to know exactly what speaker meaning is, as well as how it connects with sentence meaning. Sentence meaning is properly described as semantic meaning: it relates to words considered independently of speakers. We make no reference to a speaker when we say “‘Snow is white’ means that snow is white.” But speaker meaning is properly described as pragmatic meaning because it explicitly refers to speakers—people mean things in this sense. The word “pragmatic” here has nothing to do with the doctrine called “pragmatism,” still less with the notion of the merely practical. It merely connotes the fact that speaker meaning is about the relation between agents and language. Semantics is about words themselves and what they mean, but pragmatics is about speakers and what they do with language. (Syntax is about words considered independently of their meaning.) Put in these terms, Grice is suggesting that pragmatic meaning has primacy over semantic meaning. We could state his position in another way: semantic meaning is ultimately psychological, because for a sentence to mean something is for speakers to use the sentence while in a certain psychological state—that of meaning something by the sentence. We shall see later just what this psychological state is. What Grice is proposing, in effect, is that we can

Grice’s Theory of Speaker Meaning 

193

explain semantics in terms of psychology. Sentence meaning comes down to psychological facts about speakers. This seems to conflict with Frege’s doctrine (expounded in chapter 1) that senses are not psychological. Senses are abstract entities, according to Frege, objective things that do not depend on minds at all. The Gricean approach to meaning appears to be rejecting that Fregean position. In effect, Grice takes the meanings of words to be reducible to psychological facts, contra Frege. That is the program that hovers in the background of Grice’s paper “Meaning.” In subsequent papers, he labors to develop the program of trying to reduce semantics to psychology, and others joined him. In the present paper he focuses on understanding just what speaker meaning is. Let us turn to this seminal paper. 10.2  Two Types of Meaning Grice begins by distinguishing between two types of meaning, which he calls “natural meaning” and “non-natural meaning.” He then devotes the rest of the paper to explaining non-natural meaning. It is easy to grasp this distinction at an intuitive level. As an example of natural meaning, Grice gives “Those spots mean (meant) measles,” which might be paraphrased “Those spots are a symptom of measles.” We can infer measles from the spots, so the spots mean measles. The spots are a natural sign of measles. Another example: “The recent budget means that we will have a hard year ahead.” Given the tightness of the budget, money will be scarce in the coming year. We can infer hard times ahead from the budget. A third example not given by Grice would be: “Those clouds mean rain.” This says something like, “There is a natural association between clouds and rain, so we can infer the latter from the former.” We can contrast these examples of natural meaning with the following examples of non-natural meaning: “Those three rings on the bell (of the bus) mean that ‘the bus is full,’” and “That remark, ‘Smith couldn’t get on without his trouble and strife,’ meant that Smith found his wife indispensable.” These are very British examples, perhaps not familiar to all readers. In Grice’s day (circa 1957) conductors on buses would ring the bell three times to indicate that the bus is full (they had different numbers of rings for starting and stopping). The second example involves what is called “Cockney rhyming slang,” a dialect of east London that substitutes charming phrases

194 

Chapter 10

for common words: “trouble and strife” for “wife,” “apples and pears” for “stairs,” and so on. The speaker says, “I can’t get on without my trouble and strife” and he means that he finds his wife indispensable. We can see intuitively that “means” is being used differently in these two types of case, and Grice suggests some marks that distinguish the cases. Spots “mean” measles in a different sense from the sense in which three rings “mean” that the bus is full. In the measles case we cannot say, “Those spots mean measles, but he hasn’t got measles,” but in the three rings case we can say, “Those three rings meant the bus is full, but the bus is not full.” The bus conductor may have made a mistake when he thought the bus was full, but the spots can’t make mistakes. The fact that the conductor meant it does not entail that it is true. The Cockney made his colorful statement and meant to compliment his wife, but this does not entail that he really does find his wife indispensable—he might be quite able to survive without her. Just because someone makes an assertion and means something by it does not imply that the assertion is true. Another difference is that in the case of non-natural meaning we can substitute an expression in quotation marks after “means” but we can’t do this for natural meaning. We can say that the conductor meant, “The bus is full” by his three rings, but we can’t say that the spots meant, “The patient has measles.” What this really comes to is that three rings are synonymous with the sentence “The bus is full,” but spots are not synonymous with the sentence “The patient has measles”—they are not synonymous with anything, even though they naturally mean something. Spots are not words. A third difference is that in the case of reports of natural meaning, there is no indication that an agent or person is involved in the fact of meaning. When spots mean measles, there is no agent or person who is meaning anything. But in the case of non-natural meaning, there is always the implication of an agent or a person involved. When there is non-natural meaning, there is always an agent of that meaning—a bus conductor or an uxorious Cockney, say. People mean things in the non-natural sense, but objects or events mean things in the natural sense. This is connected to the further point that in non-natural cases we can speak of “what was meant” (by an agent), but we cannot speak this way about natural meaning—we cannot refer to “what was meant” by the spots. Grice’s terminology here is not altogether perfect, though it has become entrenched. He talks of “non-natural meaning,” but there is really nothing

Grice’s Theory of Speaker Meaning 

195

non-natural about it. We typically use the word “non-natural” to indicate things that are supernatural or lie outside of nature. But Grice does not mean to suggest that idea when he speaks of non-natural meaning. He is not using the word “non-natural” as G. E. Moore used it when describing the property of being good as non-natural, that is, not part of the natural causal order. It is not a very descriptive label and it has some misleading connotations. We could instead call it “semantic meaning” or “speaker meaning” or “agent meaning.” At any rate, it is good to keep these alternative labels in mind when using the phrase “non-natural meaning.” It is, in fact, not easy to come up with the perfect nomenclature for the distinction Grice is making, despite the clarity of the distinction. 10.3  What Is Speaker Meaning? The question is what constitutes this so-called non-natural meaning. Here Grice is looking for the necessary and sufficient conditions for instances of non-natural meaning—that is, he seeks an analysis of the notion. His procedure then is to try out various analyses and see if there are any counterexamples. He begins by examining a suggestion of C. L. Stevenson’s that he calls the causal theory of meaning. This theory is very tempting because it reflects some obvious facts about language. Let us take an ordinary assertion, such as my asserting to you, “Nadal won the French Open in 2012.” By making this assertion I meant precisely that Nadal won the French Open in 2012. Why does the speech act mean that? Well, here are two obvious facts: my utterance of that sentence tends to produce in my audience the belief that Nadal won the French Open in 2012, and the utterance itself was produced by my having that same belief. The utterance expresses my belief and it induces the same belief in you. I have a tendency to say it given my beliefs, and you have the tendency to believe it because you hear me say it. The assertion has these causes and effects, and they seem bound up with what I meant. We might then propose the following definition of non-natural speaker meaning: X means that p by uttering s if and only if X’s uttering s is caused by his belief that p and his uttering s causes in his audience the belief that p. Less formally, you mean that p by an action if and only if that action causes observers of the action to believe that p. Grice gives a counterexample to this analysis, questioning its sufficiency. He describes a man putting on a tailcoat when he is about to go to a dance.

196 

Chapter 10

This causes an observer to believe that the man is about to go to a dance. The observer believes this because putting on a tailcoat is very good evidence that someone is going to a dance. The tailcoat wearer also believes he is going to a dance. Given the causal theory of meaning, we should be able to conclude that putting on the tailcoat meant that he was going to a dance. Indeed, we should be able to conclude that in putting on the tailcoat, the wearer meant that he was going to a dance. We should be able to report that “what was meant” in performing the action was that the agent was going to a dance. Grice’s point is that no such thing was meant. The agent did not mean anything by his action—he was just preparing for the dance. His action was not any kind of assertion, not a case of speaker meaning. He was not trying to communicate anything. So inducing beliefs in others by one’s actions is not sufficient for those actions to be cases of non-natural meaning. This is really quite obvious, because most of your actions are not cases of meaning anything to anybody, even though observers will form beliefs about you given your actions. I may comb my hair to keep it neat, and you may be caused to believe that I am trying to keep my hair neat by observing me comb it, but my action of combing was not a case of my meaning anything to anybody—I wasn’t trying to tell you anything. These kinds of examples put paid to the causal theory of speaker meaning. Another type of case Grice gives that is equally devastating to the causal theory involves the sentence “Jones is an athlete.” What I meant by that utterance is that Jones is an athlete. Hearing this, my audience might form the belief that Jones must be tall, because athletes usually are—and maybe it is true that Jones is tall, and that I believe it. But did I mean that Jones is tall when I said, “Jones is an athlete”? No, I did not. The utterance of “Jones is an athlete” has a tendency to induce the belief that Jones is tall, but it does not mean that. Again, this point is obvious and generalizes. Whenever I utter a sentence of English, my utterance has a tendency to induce the belief that I am speaking English, but I certainly do not mean that I am speaking English whenever I open my mouth to speak in that language. The utterance may also cause the belief in my audience that I am alive, but again this is not something I can be said to mean. If this were sufficient for speaker meaning, every time I speak I would mean hugely many things—all the things that people might believe who hear me speak. The conditions proposed by the causal theory are hopelessly weak.

Grice’s Theory of Speaker Meaning 

197

Grice now switches to a different kind of theory. Instead of using the idea of a causal tendency to induce a belief in an audience, this new theory brings in the idea of intention—specifically, the intention to produce a belief in the audience. So the speaker means something by an action if he intends to produce a certain psychological effect. This intention was lacking in the tailcoat and “athlete” examples. To mean something you have to intend to get a belief across to an audience, not merely get a belief across in any old way. If I assert that p, I intend to get you to believe that p by making that utterance. This sounds on the right track, because in meaning something I surely do intend to have a certain effect on my audience. But Grice produces a counterexample to this analysis: the handkerchief case. I leave B’s handkerchief at the scene of a murder in order to induce the detective to believe that B is the murderer. Thus I intend to produce in the detective the belief that B committed the murder by leaving the incriminating handkerchief around the murder scene. I might thereby fulfill my intention of producing in the detective the belief that B is the murderer. But did I, by this action, mean that B is the murderer? Clearly not: all I did was intentionally fabricate evidence from which the detective inferred that B is the murderer. What is intuitively missing in this example is that the detective does not know that I intended him to form the belief by leaving the handkerchief there. I concealed my intention completely by secretly depositing the handkerchief. If he knew that I had left it there, he would not have formed the belief that B is the murderer, because he would know that I was trying to frame B. So let us add the condition that the agent must not only intend to produce the belief but must also intend that the audience should recognize this intention. Now we have an extra intention—the intention that the first intention should be out in the open. The agent intends to produce the belief in the audience and he intends the audience to realize that he has the former intention. Thus there is a double intention, where the second intention refers back to the first. We might call this the transparency condition. The agent’s belief-inducing intentions must be intentionally transparent to the audience, if the agent is to mean something by the action. This is beginning to sound pretty good, but Grice is not through yet with his counterexamples. He gives the grisly example of Herod presenting Salome with the head of John the Baptist on a charger. He thereby intends her to form the belief that John the Baptist is dead, but he also intends that

198 

Chapter 10

Salome should recognize this intention. He does not attempt to conceal his intention—no doubt because he is not afraid of her knowing that he has that intention. The severed head is ample evidence that John the Baptist is dead and Herod transparently proffers this evidence to Salome, so that all his intentions are out in the open. But, Grice insists, this act by Herod is not a case of meaning that John the Baptist is dead. It is not a way of telling her he is dead. So we have still not captured what is distinctive of non-natural speaker meaning. It is not like saying, “John the Baptist is dead.” Now we reach the crux of Grice’s argument, which is contained in the following passage: The way out is perhaps as follows. Compare the following two cases: (1) I show Mr. X a photograph of Mr. Y displaying undue familiarity to Mrs. X. (2) I draw a picture of Mr. Y behaving in this manner and show it to Mr. X. I find that I want to deny that in (1) the photograph (or my showing it to Mr. X) meantNN anything at all; while I want to assert that in (2) the picture (or my drawing and showing it) meantNN something (that Mr. Y had been unduly familiar), or at least that I had meantNN by it that Mr. Y had been unduly familiar. What is the difference between the two cases? Surely that in case (1) Mr. X’s recognition of my intention to make him believe that there is something between Mr. Y and Mrs. X is (more or less) irrelevant to the production of this effect by the photograph. Mr. X would be led by the photograph at least to suspect Mrs. X even if instead of showing it to him I had left it in his room by accident; and I (the photograph shower) would not be unaware of this. But it will make a difference to the effect of my picture on Mr. X whether or not he takes me to be intending to inform him (make him believe something) about Mrs. X, and not to be just doodling or trying to produce a work of art.2

The distinction to which Grice is drawing attention is clear enough (despite his rather convoluted grammar). In the photograph case, the audience’s reason for forming the infidelity belief is the evidence contained in the photograph itself—it makes no difference how Mr. X views my intention in showing it to him. He might find the photograph in his wife’s closet, so there is no act of showing at all. But in the drawing case, Mr. X’s reason for forming the infidelity belief is not the drawing itself—the drawing by itself is not good evidence for that belief—but rather the reason is that Mr. X infers that I intend him to form the infidelity belief. In this case, if we ask Mr. X why he has that belief he will say that it is because he knows that I intended him to form the belief—and he goes by my intention because he knows me to be trustworthy about such matters. Nothing like this holds in the photograph case: here his knowledge of my communicative intentions

Grice’s Theory of Speaker Meaning 

199

plays no role in his belief formation. What I intend in the drawing case is that Mr. X should form the belief because of my intention to get him to believe it, and not because my drawing is somehow solid gold evidence for the belief. The drawing is relevant only because it is evidence of my communicative intention, but this is not so for the photograph. It is the audience’s recognition of my belief-inducing intention that supplies him with a reason for forming that belief, not any independently plausible evidence. In short, his sole reason for forming the belief is that he sees that I am intending him to form it. Thus for the agent to mean something, it is necessary that he should intend the audience to form a belief by means of the recognition that the agent has such an intention. The agent intends the audience to engage in a piece of reasoning of the following form: the speaker intends me to form the belief that p, therefore I will form the belief that p. This is quite unlike the photograph and severed head cases, because in these cases the audience reasons as follows: I have solid evidence that p based on a photograph/severed head; therefore I will believe that p. 10.4  Consequences and Criticisms So now we know what speaker meaning is. It is intending people to form beliefs based on a recognition that that is what you intend. What can we do with this information? We can use it to define sentence meaning. A sentence s means that p if and only if people regularly use s to mean that p, where a speaker’s meaning that p is equated with the intention to induce a belief in an audience by means of the audience’s recognition of that intention. No doubt we will have to say more about the notion of “regular use,” but the thrust is clear: a sentence means what it does because people utter sentences with the kinds of intentions specified by Grice. Meaning something non-naturally is a matter of performing actions with Gricean intentions, and semantic meaning has its origin in speaker meaning. So semantics reduces in the end to intentions—that is, to a certain kind of psychological state. Languages like English exist because humans are capable of Gricean communicative intentions. It is in virtue of these intentions that words have meaning. It is worth spelling out a bit further Grice’s picture of language and its raison d’être. We all have a great many beliefs about the world, often acquired by observation. Imagine a time before language evolved, but when people

200 

Chapter 10

still had a stock of beliefs. As a social species, we might wish to induce some of our beliefs in others—that is, we want to share our knowledge (this can be useful in child rearing, among other things). How do we set about doing this? The obvious way would be to present other people with the evidence that led you to form your beliefs and let them come to their own conclusions. If you want others to know where the succulent fruit is, you might take them there so that they can see it for themselves. Alternatively, you might preserve the evidence you had and bring it with you to show others—so you bring them a piece of fruit as evidence that you know the location of fruit and then have them follow you. But this is not always feasible, because evidence is often perishable and nonportable. You had the evidence but you can’t present it to others to induce the same belief in them. So you have a problem of belief transmission: how do you get them to share your belief? The only apparent solution is that you have to present them with evidence that you have the belief in question, and then rely on their ability to reason that if you believe it, there must be a reason to believe it. In other words, their reason to believe that p is that you believe that p. Of course, that wasn’t your reason—you had solid evidence, but the evidence has long since perished. So you need to intend to produce a belief in others by getting them to recognize that you have the belief yourself, so that they can reason that if you believe it, that is a reason for them to believe it. That is, you need Gricean intentions if you are to solve the problem of perishable, nonportable evidence in belief transmission. Since Gricean intentions constitute meaningful language, you need to invent language to fill the evidential gap. So language exists because evidence vanishes or is unobtainable for other reasons. Your beliefs can persist through time and space, even though the evidence on which they are based is confined to a particular time and place. So you can use the existence of your beliefs to persuade others to believe as you do. When you do that, the stage is set for speaker meaning and hence language. Thus language exists to let people know what we believe so that they can form the same beliefs. Gricean intentions are substitutes for actual hard evidence. They allow us to transmit our beliefs by testimony, instead of by lugging evidential stuff around. On occasion, our audience may refuse to form the belief we want them to, perhaps distrusting our own belief-forming powers; then we might say, “You don’t believe me, but let me show you this”—and then we whip out the clinching piece of hard evidence. Sentences, on this conception, are

Grice’s Theory of Speaker Meaning 

201

really evidence surrogates—what we resort to when we can’t just point to the facts or produce the smoking gun. Sentences take up the evidential slack. This is the lesson buried in Grice’s account of speaker meaning: you don’t have the photograph, so you produce the drawing, intending the audience to infer the belief from the fact that you have the belief. What objections might be raised against Grice’s account of meaning? The actual analysis of speaker meaning looks pretty strong, so it is hard to object to that. But there are questions about the precise philosophical significance of his analysis. If we are to provide an explanation of sentence meaning in terms of speaker meaning, it must be that speaker meaning does not presuppose sentence meaning. Since speaker meaning consists in a complex array of intentions and beliefs, these intentions and beliefs must not presuppose sentence meaning. That is, the intentions and beliefs must not be linguistic in character. But two sorts of argument suggest that they do build in sentence meaning. One line of thought is that it is not possible to possess Gricean intentions without already being a language user: the intentions must be formulated in the very language the speaker is using. Thus, when I utter, “Snow is white” with Gricean intentions, I must be thinking something like, “I intend to produce the belief that snow is white by means of my audience’s recognition of my intention.” But this is itself an English sentence, so my intention presupposes the notion of sentence meaning. In other words, if thoughts are inherently expressed in language, they cannot be used to explain language. The natural reply to this is that thoughts are not inherently expressed in language. There can be thought without language. Animals have intentions and beliefs, but they do not speak a language. Human infants have thoughts before they acquire their native tongue. So thought itself does not presuppose mastery of a language. Moreover, people who speak different languages can have the same thoughts, even though their sentences are not the same—so there must be a psychological level that is independent of particular spoken languages. Perceptual states are surely not inseparable from spoken language, so why should thoughts be? I surely don’t see in English, so why should my thoughts have English written into their identity? I can express my thoughts to others in English, but they are not themselves English sentences running through my mind. I could have had the same thoughts and yet never have learned English—I might have been a French speaker, say.

202 

Chapter 10

Fair enough, it may be said—English is not essentially the medium of my thoughts, even though I am an English speaker. But might my thoughts not have a more subtle connection to language—what about the idea of a language of thought? True, I do not think in English, but my thoughts must exist in some sort of symbolic medium; and this medium must have the characteristics of a language—combinatorial, finitely based, recursive, referential. Aren’t my concepts essentially symbolic entities that join together to make thoughts? Thus the brain has a language of its own in which beliefs and intentions are encoded. This is not a conventional natural language but a universal, species-wide language that the brain employs to carry out cognitive operations. When I think that snow is white, my brain activates its special words for snow and whiteness, maybe in the form of a binary code that neurons can embody. These brain symbols will have reference, and maybe even sense, and they can combine to produce strings that have truth conditions. So having a mind depends on having a brain language. But then, sentence meaning is basic after all, because Gricean intentions are grounded in brain sentence meaning. Natural language sentence meaning may be explicable in terms of psychological states, but psychological states themselves must be explained in terms of a universal language of thought— so at the bottom of the whole thing, we find sentence meaning staring up at us. There is then still the question of what gives these brain sentences their meaning—because it can’t be that they are uttered with certain kinds of intentions. How do brain symbols come to mean what they do? That question remains unanswered. At this point we stray into the area of philosophy of mind. We are now inquiring into the semantics of thought. That is a subject for another kind of book. What we can say here is that these questions are not going to be easy. But no matter how the deep questions are resolved, Grice does at least provide an illuminating and plausible account of speaker meaning. Its precise significance for the general nature of meaning, however, remains debatable.

Appendix: Kripke’s Puzzle about Belief

Finally, let us consider Kripke’s paper “A Puzzle about Belief,”1 mainly because of its intrinsic interest, impact, and connection to issues already discussed. It is also fun to think about. I confine this topic to an appendix because the issue is more about the nature of belief than the nature of language, and because Kripke is not offering any theory but presenting a puzzle. I am going to describe my own version of the puzzle, which I think reveals its essential structure, without irrelevant distractions. Kripke’s version involves a bilingual speaker, Pierre, who speaks only French at one time, and on the basis of his verbal behavior we ascribe to him the belief that London is pretty. He assents to “Londres est jolie” on the strength of what he has read about London in rosy travel books. Then he comes to London and learns English, living in a seedy part of the city. He now assents to “London is not pretty.” Of course, he doesn’t realize that the place he is living in is actually the reference of the French word “Londres.” On the basis of his assent behavior, we now ascribe to him the belief that London is not pretty. So we have ascribed contradictory beliefs to him. Yet he is guilty of no logical blunder. He has manifested no irrationality. His situation is perfectly intelligible. I will now describe a case that exhibits much the same structure but without the reliance on two different languages (Kripke himself recognizes that his puzzling cases do not require two different languages). A psychologist is conducting experiments on the interpretation of faces. She asks the subjects to respond to photographs of faces according to whether the subject finds the person photographed trustworthy or untrustworthy, going by their facial expression. Then the psychologist tells the subjects that although the pictures may appear to be of the same person they are all in fact pictures of different people. In reality, however, the pictures are all of

204 Appendix

the same person. And so the subject’s beliefs are all about a single person throughout the experiment, though he thinks they are about different people. Let us suppose that the subject’s response takes the form of checking the sentence “That person is trustworthy” or “That person is untrustworthy.” The experiment is run and the data show that subjects systematically vary in their responses according to facial expression. Logically, this example is just like Kripke’s Pierre example: “Londres” and “London” refer to the same city, but Pierre does not realize this. He may in fact believe explicitly that they are not identical. In the experiment, the subject keeps seeing a picture of the same person but doesn’t realize it and even disbelieves it. To begin the experiment, the psychologist shows the subject the first picture and asks him if that person is trustworthy. Based on the expression of the individual’s face in the picture, the subject says yes. The psychologist then shows the subject another picture, and based on that person’s expression, the subject responds that that person is not trustworthy. Keep in mind that the subject thinks there is a different person in each picture. The experimenter goes on to show the subject ten different pictures and based on the subject’s judgments ascribes beliefs to the subject. Using the normal method of belief ascription, the experimenter would ascribe contradictory beliefs to the subject in exactly the way we would in Kripke’s Pierre example. The subject believes that that person is trustworthy and that that person is untrustworthy—yet these are same person. Suppose the experimenter says to the subject, “Just for convenience I’m going to call all these different people photographed ‘Albert,’ so I want you to respond to the sentence ‘Albert is trustworthy.’” In fact, the single person in all the pictures is named Albert. Then after the first photograph is presented, the experimenter asks, “Do you think that Albert is trustworthy?” The subject responds affirmatively, thus showing that he believes that Albert is trustworthy. Now, on the second trial, the subject responds negatively, thus showing that he believes that Albert is untrustworthy. Already in two presentations of the pictures the subject has exhibited contradictory beliefs: he believes that Albert is trustworthy and he believes that Albert is untrustworthy. The subject could continue to form contradictory beliefs about the same person throughout the experiment. What is going on, intuitively, is that the subject does not realize it is the same person in each photograph, so he feels free to form different beliefs from trial to trial. However, the experimenter knows that the subject is forming beliefs about the same

Appendix 205

person. This is a perfectly intelligible situation, just as Kripke’s Pierre case is, and what makes it intelligible is that people can fail to realize they are forming contrary beliefs about the same thing. It is not a given that perceived objects are the same—one might have false beliefs about this. Even if the objects are presented in a qualitatively identical way, and in fact really are the same object, a person can intelligibly suppose that two numerically distinct objects are being presented. It is intelligible that someone might take a single individual to be really twins, and hence freely form contradictory beliefs about this single person. We could imagine an alternative experiment where the experimenter tells the subject that all the photographs are of the same person. Now consider what happens. The experimenter presents the first photograph and asks if the person depicted (“Albert”) is trustworthy. The subject assents to this proposition, thus showing that he believes that Albert is trustworthy. Then she presents the second photograph and asks the same question. The subject replies, “But I already told you I think Albert is trustworthy.” The experimenter persists with her question, pointing out the extremely shifty expression on the person’s face. She asks: “Are you so sure now that Albert is trustworthy?” The subject may hesitate and then comment, “Perhaps I should revise my belief about Albert—this expression could only come from an untrustworthy man.” The subject, as we say, changes his mind, forming a new belief and rejecting the old belief. He is rationally committed to changing his earlier belief when he acquires good contrary evidence. It would be highly irrational of him to persist with the first belief in the light of acquiring the second. Why? Because he believes (truly) that the same person is being presented, and it is irrational to attribute contrary predicates to the same individual, given that you know it is the same individual. This thought experiment is structurally the same as Kripke’s, but it is more streamlined because it requires that we use only one language. We also make explicit the subject’s beliefs about the identity of the things he has beliefs about. In both cases, however, we wind up ascribing contradictory beliefs to the subject. We are now starting to see what these kinds of examples depend on. Here is another example. Consider someone who has some eccentric metaphysical views about the world. He doesn’t think objects persist for more than two seconds. He subscribes to the doctrine of repeat creationism, maintaining that God actually re-creates the world every two seconds. But God

206 Appendix

makes it look as if there is seamless continuity to the human senses. In fact, he thinks, God annihilates the particles that compose objects and creates new particles ab initio every two seconds. He is omnipotent, after all, and he likes to keep himself busy. (Note: we are supposing that this metaphysical system is false.) In addition to this belief, our eccentric metaphysician believes that objects change their natures in important ways every two seconds—specifically, they become made up of different kinds of particles every two seconds. So suppose that at time t he assents to “This table is made of electrons” but at time t plus two seconds he assents to “This table is not made of electrons”—despite the fact that he refers to the same table both times (contrary to his metaphysical beliefs). Does he not have contradictory beliefs? Of course, he doesn’t think so, because he doesn’t think the same table is referred to by the two demonstratives; but from our point of view, we can see that he both believes that this table is made of electrons and believes that this table is not made of electrons. We arrive at these ascriptions of belief simply by taking seriously his assent to “This table is made of electrons” at t and his assent to “This table is not made of electrons” at time t plus two seconds. If we gave the table a name, say “Bill,” then we could convict our metaphysician of believing both that Bill is made of electrons and that Bill is not made of electrons. He sees no clash in his beliefs, because he thinks they are about different objects; but we know better, so we detect a clash—and we are right, because objects do persist over time. It is rather as if Pierre were to assert outright that “Londres” and “London” do not refer to the same city, were we to suggest to him that perhaps they are the same. He has a false nonidentity belief, just like our metaphysician. Suppose you have used the name “Larry” to refer to someone of your acquaintance, assuming (truly) that there is only one Larry you have been calling by that name. Perhaps you have noted that Larry seems rather a changeable kind of man. Then you come to the conclusion that there is no single Larry—you have been calling two men by the same name. This conclusion, however, is false. You might now start to feel a new liberation in your assent to sentences containing “Larry,” because now you can attribute his varying characteristics to two different men. But using our usual practice of belief ascription, we find ourselves ascribing contradictory beliefs to you, because you are in fact referring to the same man with “Larry” though you think you are not. You might believe that both men are

Appendix 207

really called “Larry” because you have heard others refer to “them” by that name, but there is nothing impossible there—different people often have the same name. The problem is that you have a false identity belief concerning Larry—you believe that Larry1 and Larry2 (as you put it to yourself) are not identical, when they are. Here is one final example. Consider Peter, a born-and-bred Londoner. Peter was raised in Hackney, a less than salubrious part of London. As a result of his experiences in Hackney, he concludes (a little hastily) that London is not genteel—he assents readily to “London is not genteel.” But then, at age eighteen, he is kidnapped and taken to Hampstead, also a part of London. Hampstead is so different from Hackney that Peter does not think he is in the same city. He notices that people refer to the city of which Hampstead is a part as “London,” but he assumes that this is just a case of different places having the same name—a common occurrence, as he has learned from geography classes. If you ask him what he thinks of the sentence “London is genteel” after the move to Hampstead, he enthusiastically assents to it. Of course, he thinks that this “London” refers to a different city from the other “London.” Given our usual practices of belief ascription, we would have to conclude that he believes both that London is not genteel and that London is genteel. Certainly, his assent behavior in the two places would warrant such ascriptions of belief considered separately—it is only the fact that we can make both ascriptions that might make us hesitate. The word “London” in his idiolect refers only to one city, which is why we can make the contradictory ascriptions, but Peter does not realize this, which is why he can so smoothly believe both things. It should be clear that in none of these cases are we talking merely about contradictions between de re beliefs. There is nothing puzzling or paradoxical about attributing to someone the belief of Harvey that he is shady and also the belief of Harvey that he is not shady. You just need to observe Harvey acting suspiciously in one situation and observe him acting irreproachably in another, without realizing you have observed the same individual twice. In this kind of case, there is no de dicto attribution of the form “X believes both that Harvey is shady and that Harvey is not shady.” All we have is the de re attribution “X believes of Harvey that he is shady and also of Harvey that he is not shady.” Kripke’s examples involve contradictory de dicto beliefs, not just contradictory de re beliefs. The latter are not puzzling at all. There is no suggestion in these cases that the subject believes

208 Appendix

contradictory propositions. But in the Kripke cases that is precisely the situation. The same is true of the further examples I have described. Although we cannot resolve the paradox, we can at least examine how it arises—its inner logic. There are two kinds of case in which a person has contradictory beliefs: there is the case where a person has contradictory beliefs because he is irrational, and there is the case where a person has contradictory beliefs without being irrational. What is the difference between those two cases? Suppose you ask someone, “Do you think that a is F?” and he replies, “Yes.” Now you ask, “Do you think that a is identical to b?” and he again replies, “Yes.” Then you ask, “Do you think that b is F?” He replies, “No.” This case is an example of outright irrationality, because it logically follows from “a is F” and “a is identical to b” that “b is F” is true. This is a simple consequence of Leibniz’s self-evident law of the indiscernibility of identicals, that is, that if a is identical to b then anything true of a must be true of b. If someone were to reply as just described, you would be within your rights to protest, “Then you don’t really believe that a and b are identical!” But, of course, it is not irrational to refuse to make the inference to “b is F” from “a is F” if you do not believe “a is identical to b.” You simply lack the identity premise that would make the inference go through. Indeed, it would be irrational to make the inference without the identity premise. You are not guilty of any irrationality if you refuse to infer that Phosphorus is a planet just from the premise that Hesperus is a planet, but you are if you refuse to make that inference given that premise and the premise that Hesperus is identical to Phosphorus. These are just totally different kinds of psychological situations, not to be confused or assimilated. In Kripke’s example, Pierre does not believe the identity “Londres is identical to London”—he will not assent to this sentence. The same is true in all the cases I described. The subject lacks belief in a crucial identity premise. So he is not being irrational; in fact, he is being perfectly rational. There are cases of rationally contradictory belief—those in which the subject believes that p and believes that not-p without violating any principle of logical inference. These arise when the subject does not believe an identity proposition that connects two names or demonstratives or descriptions. It is not irrational to have the beliefs that Pierre has, because he formed them perfectly rationally. What would be irrational is to believe that London is pretty and that London is not pretty while assenting to “Londres is

Appendix 209

identical to London.” That is, if we confronted Pierre’s assent to “London is not pretty” and his assent to “Londres est jolie” with the information that “Londres” refers to the same city as “London,” and he accepted that identity but refused to give up either assent, then he would be irrational. He couldn’t rationally suppose that the place he called “Londres” is the same as the place he calls “London,” and yet the former place is pretty and the latter is not. Everything depends on his view of a certain identity question. Pierre and his puzzling kindred spirits are really no more irrational than someone with contradictory de re beliefs, that is, not irrational at all. It is not irrational to believe of a that it is F and of a that it is not F, because in such a case you don’t subscribe to an identity judgment regarding the objects of your beliefs. You fail to realize that your beliefs are about the same thing. You only lapse into irrationality if you accept the identity “a is identical to b” but persist in assenting to “a is F” and “b is not F.” In all the puzzling Pierre-type cases we have described there is a crucial nonacceptance of an identity statement—a true identity statement. This is not meant to solve or remove Kripke’s puzzle, which does reveal something strange in our normal practice of belief ascription, only to diagnose how it arises. We need to see clearly the difference between irrational contradictory beliefs and rational contradictory beliefs. That difference turns on the role of identity judgments in the subject’s reasoning. What is surprising is that a nonparadoxical rejection of a true identity statement can lead so quickly to a puzzling assignment of contradictory beliefs, given that we insist on sticking to our usual practice of belief ascription. Being logical can lead to an appearance of illogicality. This appearance is the same as in genuine irrationality. But the underlying state of mind is quite different.

Notes

1  Frege on Sense and Reference 1.  Gottlob Frege, “On Sense and Reference,” in Philosophy of Language: The Central Topics, ed. Susana Nuccetelli and Gary Seay (New York: Rowman & Littlefield, 2008), 113. 2. Ibid. 3. Ibid. 4. Ibid. 5. Ibid. 6.  Ibid., 113–114. 7.  Ibid., 114. 8. Ibid. 9. Ibid. 10. Ibid. 11. Ibid. 12.  Ibid., 114–115. 13.  Ibid., 115. 14.  Ibid., 115–116. 15.  Ibid., 116. 16.  Ibid., 117. 17. Ibid.

212 Notes

2  Kripke on Names 1. Discussion in this chapter follows the excerpt from Saul Kripke’s Naming and Necessity (Lecture II) in Philosophy of Language: The Central Topics, 128–146. 2.  Gottlob Frege, “On Sense and Reference,” in Philosophy of Language: The Central Topics, 126.

3  Russell on Definite Descriptions 1. Bertrand Russell, “Descriptions,” in Philosophy of Language: The Central Topics, 147. 2.  Ibid., 148. 3.  Ibid., 150–151. 4.  Ibid., 153–154.

4  Donnellan’s Distinction 1. Keith Donnellan, “Reference and Definite Descriptions,” in Philosophy of Language: The Central Topics, 157. 2.  Ibid., 164. 3. Stephen Neale, Descriptions, excerpted in Philosophy of Language: The Central Topics, 170.

5  Kaplan on Demonstratives 1. David Kaplan, “Demonstratives,” in Philosophy of Language: The Central Topics, 181. 2.  Ibid., 187. 3. Ibid.

6  Evans on Understanding Demonstratives 1. Gareth Evans, “Understanding Demonstratives,” in Philosophy of Language: The Central Topics, 201.

Notes 213

7  Putnam on Semantic Externalism 1.  Hilary Putnam, “Meaning and Reference,” in Philosophy of Language: The Central Topics, 275.

8  Tarski’s Theory of Truth 1.  Alfred Tarski, “The Semantic Conception of Truth,” in Philosophy of Language: The Central Topics, 30. 2.  Gottlob Frege, “On Sense and Reference,” in Philosophy of Language: The Central Topics, 117. 3.  Alfred Tarski, “The Semantic Conception of Truth,” 30–31. 4.  Ibid., 32. 5.  Ibid., 38. 6. Ibid. 7. Ibid.

9  Davidson’s Semantics for Natural Language 1.  Donald Davidson, “Semantics for Natural Languages,” in Philosophy of Language: The Central Topics, 58. 2.  Ibid., 62. 3.  Donald Davidson, “On Saying That,” in his Inquiries into Truth and Interpretation (Oxford: Oxford University Press, 2001). . 4.  Davidson, “Semantics for Natural Languages,” 61. 5.  Davidson, “Radical Interpretation,” in Inquiries into Truth and Interpretation.. 6.  Davidson, “Semantics for Natural Languages,” 60.

10  Grice’s Theory of Speaker Meaning 1.  Discussion in this chapter follows H. P. Grice’s paper “Meaning” in Philosophy of Language: The Central Topics, 69–76. 2.  Ibid., 72–73.

214 Notes

Appendix: Kripke’s Puzzle about Belief 1.  The discussion in this appendix concerns an excerpt from Saul Kripke’s “A Puzzle about Belief,” in Philosophy of Language: The Central Issues, 257–263.

 

Index

Agent meaning, 195. See also Meaning Analytic–synthetic distinction, 4, 6–7, 40–41, 45, 46, 129, 188 A posteriori propositions, 6, 9, 40 Appearances, 6 A priori propositions, 4, 6, 7, 40–41, 45, 46 Aristotle and redundancy theory, 151–155 Aspect of object and sense, 14–16, 145 Atheism, 59 Attributive view of descriptions, 78–84 Axiomatized arithmetic, 51 Base axioms, 175, 178 Belief, Kripke’s puzzle about, 203, 209. See also Kripke, Saul Belief ascription, 203–207, 209 Biconditionals, 152–156, 188–189. See also T-sentences material, 189 strict, 189 Brain and language, 202 Causal chain theory, 48–49 Causal theory of meaning, 195, 196 Causal theory of perception, 47 Character, 107–108, 110–111, 117, 130–131, 140 content and, 107–108, 110, 111, 130, 131 context and, 107–108, 110

defined, 107 as function from context to content, 110 nature of, 107, 110 Character compositionality, 112 Character- vs. content-type meaning, 111. See also Character: content and Charity, principle of, 184–185 Circumstances of evaluation. See Conditions of evaluation Cockney rhyming slang, 193–194 Cogito ergo sum (Descartes), 106 Coherence theory of truth, 148 Complete expressions, 31 Compositional idea, 168 Compositionality, semantic, 112 Compositional theory of meaning, 168 Compositional theory of truth conditions, 168 Conditions of evaluation, 104–110 context of use and, 105–109 Conjunctions. See Sentence connectives Content, 117 defined, 108 Content compositionality, 112 Content- vs. character-type meaning, 111. See also Character: content and Context dependence, 102 definite descriptions and, 102, 108– 109, 111 of indexicals, 102, 106–109, 111, 142

216 Index

Context-dependent singular term, 179 Context independence, 106–107, 109– 111, 117, 133 Context of use, 105–109 Contingency, 40–45, 52–53, 98 Contingent facts, 41–42 Contingent properties, 43 Contingent sentences, 40, 97 Contradictions, 56–57 Contradictory beliefs, 203–209 Conversational implicature, 90, 91. See also Implications and implicature Coreference, 29, 30, 101, 123, 127 Correspondence theory of truth, 148, 149, 153–154 Creationism, repeat, 205–206 Davidson, Donald applying Tarski’s theory to natural languages, 175–181 empirical truth theory and, 181–185 and the merits of Tarski’s theory as applied to meaning, 168–175 semantics for natural language, 165–168 theory of meaning, 175, 178–181 criticisms of, 185–190 Definite descriptions, 40, 41, 46, 95–96, 99 context dependence and, 102, 108– 109, 111 defined, 13 impure descriptions and, 53–54 vs. indefinite descriptions, 55–60, 65, 67 indexical expression and, 102, 109, 111 meaning and, 17, 61, 70, 77, 95, 98, 126–127, 143 mode of presentation and, 38 names and, 36, 43, 60, 62, 65 (see also under Description theory of names) proper names and, 13, 53–55, 60, 62, 69–71, 102, 119–120

as quantifiers, 55, 59, 60, 78, 96, 104 reference and, 13, 14, 16, 17, 20, 48, 51–52, 61, 67, 68, 79, 84–85, 91, 109, 117, 126–127, 134, 143 referential and attributive uses of, 78–84 rigid designation and, 42–44, 99 Russell on, 55–75, 104 semantic ambiguity and, 92 sense and, 13, 14, 17, 38, 117, 125 as singular terms, 55, 61–62, 68 theories of, 60–63, 77 “Definite Descriptions” (Russell), 60, 70. See also Definite descriptions; Russell, Bertrand Demonstrative reference, 54 Demonstratives, 54, 88, 96, 142 defined, 101–102 Donnellan on, 80, 88 Evans on understanding, 127 (see also Evans, Gareth) Frege and, 127 identity propositions that connect, 208 indexicals and, 101–102 Kaplan on, 97 (see also Kaplan, David) nature of, 71, 96 proper names and, 54, 71, 88 Putnam on, 142 Russell and, 70–71, 88, 96 Strawson on, 88 Denoting, 84–85 vs. referring, 84 Descriptions. See also Definite descriptions; Indefinite descriptions essential, 52–53 impure, 53–54 as reference dependent, 127 (see also Reference dependency) Russell’s theory of, 67–72, 116 semantics of, 90 theory of (see Description theory of names)

Index 217

Description theory of (indexical) sense. See Description theory of names: indexicals and Description theory of names, 53–54, 69 definite descriptions and, 36–39 demonstratives and, 54 Evans and, 117, 119, 127 Frege and, 35, 36, 39, 50, 116, 117, 119 historical perspective on, 35, 39 indexicals and, 102, 116–119, 127 John Stuart Mill and, 50, 125 Kaplan and, 102 Kripke’s critique of, 35, 36, 39–45, 52, 54, 102 (see also Kripke, Saul: critique of Frege) objections to, 49–52, 54 names and, 46, 48 objects and, 47 overview, 36, 38 Perry and, 117–119 reference and, 49, 50, 54, 102 Russell and, 69, 95, 116 social/socialized, 51 Designation, 159 Designation, mode of, 8, 10–11 Designation axioms, 162, 172, 176, 187 Direct designation, 70 Direct reference, 101–105 vs. rigid designation, 103–106 semantics of, 100, 101 Direct reference model, 101 Disappearance theory. See Redundancy theory of truth Disquotational theory of truth. See Redundancy theory of truth Donnellan, Keith critique of Russell, 78, 91–94 distinction between denoting and referring, 84–85 evaluating, 87–90 distinction between referential and attributive uses, 78–84

on implication and implicature, 90–91 Neale’s criticism of, 90–92 P. F. Strawson and, 82, 85–86, 90 Russell and, 78, 83–87 truth-value and, 78 on truth-value gaps, 85–87 Dual-aspect semantics, 111 Empirical truth theory, 181–185 Empty names, 125–126 Equality (mathematics), 3–4 Essential indexical, 119. See also “I” Evans, Gareth description theory of names and, 117, 119, 127 Frege and, 115, 117, 120–125, 128– 129, 131 on indexicals, 115–118, 131 John Perry and, 117–119 Kaplan and, 115, 131 on reference dependence, 123–127 on Russellian terms, 123, 127 on saying vs. showing, 122–123 on senses, 119–128 terminology, 116 theory of sense and reference for indexicals, 119–122 thesis that names are Russellian, 126 on “today” and “yesterday,” 128–130 on understanding demonstratives, 127 view of names, 123–124, 126–128 Exaggeration, 93 Existential quantifiers, 64, 67, 188 Existential statements, 73–75 Extension, 135 character and, 110 context and, 110, 111 defined, 97 of Frege’s theory beyond singular terms, 25–29 two types of dependence of, 106 Extensional contexts, 177 Extensional languages, 177

218 Index

Extensional sentences, 177 Extensional theory, Tarski’s, 177 Externalism, 133. See also Semantic externalism psychological, 145 Famous deeds theory, 45 Fictional characters, 61, 66 Fictional names, 124 First-level concepts, 59, 120 Formal correctness, 149, 150, 155, 157 Formal languages, 160–161, 179 Frege, Gottlob, 1–2. See also Kripke, Saul: critique of Frege description theory of names and, 35, 36, 39, 50, 116, 117, 119 on ideas (in people’s minds) vs. sense and reference (of words), 20–23 on identity, 3–9 on indexicals, 108–109, 115–118 leveled system, 22–23, 112 mock sense and, 124–125 on mode of presentation, 10–11 “On Sense and Reference,” 2, 3, 10, 24, 33, 35, 111 on opaque contexts, 32–33 on ordinary and extraordinary use, 18–20 on proper names, 13–14, 16, 22, 27, 31, 35, 55, 56, 60, 77, 119–120 on reference, 3–5, 8–23, 121 Russell and, 55, 127 on sense(s), 3–5, 8–16, 19–23, 120, 121, 123 theory of definite descriptions, 61 theory of meaning, 112 theory of sense and reference compared with Russell’s theory, 77 extension beyond singular terms, 25–29 problems with, 23–24 theory of truth, 153, 154 on truth-value, 25–32, 78, 97, 120

Genuine names, 65, 69, 73, 86, 100 God, question of the existence of, 59 Grammar as logically misleading, 69 Grice, H. Paul, 90, 92 theory of speaker meaning, 191–199 consequences and criticisms, 199–202 Homophony, 173–175, 178 Hyperbole, 93 “I,” 105–107, 110, 111, 116–119, 133, 139, 143, 178 Ideal language, 24, 63, 65 Ideas, 20–23 compositional, 168 as objects of reference, 22 vs. senses, 22 Identicals, indiscernibility of, 208 Identity, 3–9 law of, 6–7 as a relation between names, 8 theories about personal, 44–45, 53 types of, 4 Identity statements, 3–5, 7–10, 123, 129, 209. See also Frege, Gottlob: theory of sense and reference Evans and, 123 false, 5, 71 Frege on, 11, 123 informative, 6, 13, 16, 38, 48, 53, 71 linguistic theory of the content of, 9 logically proper names and, 71, 77 as modes of presentation, 24 Russell and, 71 Wittgenstein on, 24 Implications and implicature, 90–94 Incomplete expressions, 31, 120 Indefinite descriptions, 67–68 vs. definite descriptions, 55–60, 65, 67 identity and, 63–65 as quantifiers, 58 Indeterminacy, radical, 184

Index 219

Indexical expression and definite descriptions, 102, 109, 111 Indexical “I.” See also “I” indispensability of, 119 Indexicality, the point of, 118–119 Indexicals context dependence, 102, 106–109, 111, 142 description theory of names and, 102, 116–119, 127 Evans on, 115–122, 131 Fregean theory of, 108–109, 115–118 Kaplan on, 100–105, 115 nature of, 101–102 possible worlds, meaning, and, 109–112 two principles of, 102–105 Indirect perspective, 33 Indirect reference, 19 Indirect sense, 19, 32–33 Indirect speech, 19–20 Indiscernibility of identicals, law of, 208 Information, 130–131 as an epistemic notion, 131 Intension, 109–111, 133, 134, 140 defined, 97 extension and, 97–100, 106, 110, 111, 133, 134, 140 Intensional contexts, 177 Intensional languages, 177 Intensional operators, 176, 177 Intension-based semantic theory, 111 Intention, 80–82, 197–202 Intentional operators, 160 Internalism, 133 Interpretation radical, 184 substitutional, 64 Irony, 93 Kaplan, David, 97 on character and content, 107–108, 131, 140

on context of use and conditions of evaluation, 104–109 and Fregean model, 101 on indexicals, 100–105, 115, 131, 133 and possible worlds and meaning, 109–112 on intension and extension, 98–100, 133 return to Russellian semantics, 100 on rigid designators, 103 on rigidity and direct reference, 104–105 on “today” and “yesterday,” 113 Kripke, Saul, 35–36, 99, 138 causal chain theory, 48–49 critique of description theory of names, 35, 36, 39, 52, 54, 102 objections to, 49–52, 54 critique of Frege, 39–42 epistemic objections, 45–48 objections to, 49–51 essential descriptions, 52–53 impure descriptions, 53–54 modal arguments, 42–45, 48, 50, 53 Naming and Necessity, 104 “Naming and Necessity,” 35, 45 necessity of origin example, 103 puzzle/puzzling cases, 203, 209 on rigid designation, 42–45 social character of names, 51–52 Language, philosophy of. See also specific topics questions in, 1 Language acquisition. See Learning language Language of thought, 202 Learning, 9–10 Learning language, 171, 173 Levels, Frege’s system of, 22–23, 112 Lexical ambiguity, 179. See also Semantic ambiguity Lexical meaning, types of, 111

220 Index

Linguistic deference, 51 Linguistic labor, universality of the division of, 140–141 Logical form, 73, 154, 155, 178–179, 187–188 Logically proper names, 69–71, 77, 100 Material adequacy, 149–150, 157, 159 Material biconditional, 189 Meaning, 108, 181, 191. See also Speaker meaning as compositional, 165, 168, 171 definite descriptions and, 17, 61, 70, 77, 95, 98, 126–127, 143 definitions, 140 “meanings are not in the head,” 135–142 merits of Tarski’s theory as applied to, 168–175 as a social phenomenon, 141 theories of, 112, 181, 185, 188–189 (see also Causal theory of meaning) Frege’s theory, 33, 112, 115, 147 Kaplan’s theory, 112 for natural languages, 147, 165, 175 (see also Davidson, Donald: theory of meaning) possible world semantics and, 111 properties/conditions for, 119, 168– 173, 175, 185, 187, 189, 190 referential, 61 Russell’s theory, 61, 69, 112 theories of truth and, 147, 165, 168– 173, 175, 185, 189 types of, 111, 193–199 “Meaning and Reference” (Putnam), 133 Meaning-ascription, 167 “Meaning” (Grice), 191, 193. See also Grice, H. Paul Meinong, Alexius, 59–63, 125 theory of definite descriptions, 61–63 Meinongian ontology, Russell’s rejection of, 65–67

Metaphors, 93 Mill, John Stuart, 50, 125 Mock sense, 124–125 Mode of designation, 8, 10–11, 80 Mode of presentation, 8, 226 of aspect, 15–16 as an aspect of an object, 14 definite descriptions and, 38 Frege and, 10–13, 15–16, 23–24, 33, 38, 120 identity statements and, 11, 12, 23–24 vs. mode of designation, 10 nature of, 10, 13, 14, 120 perception and, 11, 13 of reference, 12, 15 second- and third-order, 33 sense and, 12, 13, 15–16, 23, 33, 120 Names. See also Description theory of names; Kripke, Saul Evans’s view of, 126–128 genuine, 65, 69, 73, 86, 100 Russell on, 104 as Russellian, 126 social character of, 51–52 Naming and Necessity (Kripke), 35, 36, 39, 45, 104. See also Kripke, Saul Natural kinds, 141 Natural kind terms, 133–134, 142 Natural language(s), 63, 160, 165, 202. See also Davidson, Donald applying Tarski’s theory to, 175–181 Natural vs. non-natural meaning, 193–195 Neale, Stephen criticism of Donnellan, 90–92 Descriptions, 90 on Donnellan and Grice, 92 Necessity, 75, 99, 103, 174. See also Naming and Necessity Necessity of origin, 44, 103 Neurolinguistics. See Brain and language Nonindicative sentences, 95, 177

Index 221

Non-natural meaning, 193–195 Numerical identity, 4 Object language and metalanguage, 155–15 Object(s). See also Aspect of object and sense Frege on, 26, 31 terms that introduce, 100 truth-value as, 25, 26, 28, 30, 31, 120 “On Sense and Reference” (Frege), 2, 3, 6, 10, 19, 24, 33, 35, 111, 153. See also Frege, Gottlob purpose, 10 Opaque contexts, 32–33, 176–177 Ordinary and extraordinary use, 18–20 Paratactic theory, 177 Partial definitions, 157–159, 162, 163 Peano’s axioms, 51–52 Perception causal theory of, 47 and mode of presentation, 11, 13 Performatives, 177 Perry, John, 117–119 Personal identity theories, 44–45, 53 Perspective, 130–131. See also Character on a perspective, 33 Positivists, 186–187 Possible worlds analytic–synthetic distinction and, 40, 41 (see also Analytic–synthetic distinction) content and, 108 idea of, 42 indexicals, meaning, and, 109–112 intension, extension, and, 97–100, 106, 109–111, 134 ontology of, 174 “possibly” and, 188 rigid designation and, 42–45, 99, 104 truth-value and, 97, 98, 108, 189

Possible world semantics, 97–100, 104–106 Pragmatic meaning, 192 Pragmatics, 78, 84, 85, 88–92. See also Donnellan, Keith Pragmatic theory of truth, 148, 149 Predication, 65, 68 Presentation. See Mode of presentation Pretend-sense theory, 125–126 Primary and secondary occurrences (of descriptions), 74–75 “Proper knowledge,” 8–9 Proper name(s), 47, 119–120 ambiguity, 17, 35, 36, 71 aspects and, 13–14, 16 as complete expressions, 31 context dependence and, 102 definite descriptions and, 13, 53–55, 60, 62, 69–71, 102, 119–120 demonstratives and, 54, 71, 88 as descriptions, 54 as designating objects, 31 Frege on, 13–14, 16, 22, 27, 31, 35, 55, 56, 60, 77, 119–120 identity statements and, 71 Kripke and, 35, 36, 42 logically, 69–71, 77, 100 Meinong’s theory and, 60 multiple entities with the same, 17 ordinary names and, 13, 69–70, 119–120 reference and, 13–14, 16–18, 22, 27, 55, 56, 61, 69, 77, 119, 135, 142 as rigid designators, 42 Russell on, 13, 55, 56, 58, 60–62, 69– 73, 77, 100 Russell’s criteria for, 69, 71 sense and, 13, 16–18, 27, 35, 55 as singular terms, 55 Propositional function, 57–60 Propositions contradictory, 207–208 (see also Contradictions; Contradictory beliefs)

222 Index

Propositions (cont.) the proposition expressed vs. the proposition meant, 91–93 (see also Implications and implicature) singularity of (see Singular propositions) Putnam, Hilary “meanings are not in the head,” 135–142 on semantic externalism, 133–146 Twin Earth thought experiment, 134–145 “Puzzle about Belief, A” (Kripke), 203. See also Kripke, Saul Qualitative identity, 4 Quantified proposition vs. particular proposition, 82 Quantifier expressions, 59, 62, 92, 120 Quantifier phrases, 58–59, 80, 96 Quantifiers, 64 definite descriptions as, 55, 59, 60, 78, 96, 104 descriptions as, 73 existential, 64, 67, 188 indefinite descriptions as, 58 vs. names, 58 names as, 73 Quantifier view of descriptions, 78. See also under Quantifiers Quantifier words, 59, 89 Radical indeterminacy, 184 Radical interpretation, 184 Radical translation (thought experiment), 183 Rationally contradictory beliefs, 208, 209 Real-world correlate, 130, 131 Recursive procedure, 162, 163 Redundancy theory of truth, 152–155 Aristotle and, 151–155 Reference, 12, 16–18, 181, 187. See also specific topics

definite descriptions and, 13, 14, 16, 17, 20, 48, 51–52, 61, 67, 68, 79, 84– 85, 91, 109, 117, 126–127, 134, 143 levels of, 112 meaning as determining, 139–140 proper names and, 13–14, 16–18, 22, 27, 55, 56, 61, 69, 77, 119, 135, 142 sense and, 16–18, 20–23, 129, 166 of a sentence, Frege on, 25 theories of, 15, 61, 119–120, 187 truth-value and, 25–32, 97, 106, 120, 166 “Reference and Definite Descriptions” (Donnellan), 84–85. See also Donnellan, Keith Reference dependency, 123–127 Referential theory of meaning, 61 Referential view of descriptions, 78–84 Referring vs. denoting, 84 Regular use, 199 Relational connections and semantic notions, 159–160 Repeat creationism, doctrine of, 205–206 Representational entities, 15 Rigid descriptions, 45, 103, 104 Rigid designators de facto, 104 de jure, 104 indexicals as, 102–103 vs. non-rigid designators, 42–44, 99, 103, 109–110 Rigidity/rigid designation, 42–45, 104 definite descriptions and, 42–44, 99 vs. direct reference, 103–106 possible worlds and, 42–45, 99, 104 Russell, Bertrand on definite descriptions, 55–75, 104 demonstratives and, 70–71, 88, 96 description theory of names and, 69, 95, 116 Donnellan and, 78, 83–87 Donnellan’s critique of, 78, 91–94

Index 223

Frege and, 61, 116 on indefinite descriptions and identity, 63–65 on indefinite descriptions as quantifiers, 58–59 John Stuart Mill and, 50 Meinong and, 59–63 Russell’s rejection of Meinong’s ontology, 65–67 on names, 100, 104–105 (see also under Proper name(s)) on primary and secondary occurrences, 74–75 Principia Mathematica, 63 problems with, 72–74 on propositional function and instance, 57–59 reference dependency and, 123 referential theory of meaning, 61, 69 theories of definite descriptions, 60–63 (see also Definite descriptions) theory of descriptions, 67–72, 116 compared with Frege’s theory, 77 objections to, 94–96 problems with, 72–74 theory of meaning, 112 truth-value and, 71–72, 77, 78, 94, 95 Russellian terms, 123 as Fregean, 127 that differ in their sense, 127 Satisfaction, 159–163 Satisfaction axioms, 162–163, 172, 176, 179–181, 183 Saying vs. showing, 122–123 Schematic letter, 155 Scientific sense of words, 124 Secondary occurrences. See Primary and secondary occurrences Second-level concepts, 59, 120 Self. See “I” Semantic ambiguity, 16–17, 35–36, 65, 74, 75, 85, 89, 92, 93, 117

Semantic compositionality, types of, 112 Semantic conception of truth, 160 “Semantic Conception of Truth, The” (Tarski), 147. See also Tarski, Alfred Semantic denoting, 84–85 Semantic externalism, 142 Putnam on, 133–146 Semantic meaning, 192, 195. See also Meaning Semantics, 78. See also specific topics of natural languages, 163 (see also Natural language) Semantic theory, 119 Semantic value, 120 Sense, 108, 181. See also specific topics aspect and, 14–16, 73 authentic non-nonsense sense and specious phony sense, 124–125 conception of, 12–16 definite descriptions and, 13, 14, 17, 38, 117, 125 definition and meaning of the term, 14 Evans on, 119–128 Frege’s conception of, 12–16 (see also under Frege, Gottlob) Frege’s introduction of the term/concept, 20 as a label, 12 proper names and, 13, 16–18, 27, 35, 55 as the route to reference, 166 (see also Reference: sense and) Sense-specifying reference assignment, 121 Sentence connectives, 89, 155, 161 Sentence meaning. See Semantic meaning Sentences and propositions, 2–3 Sentential functions, 161 Showing vs. saying, 122–123

224 Index

Simple object theory (identity statements), 9 Singular propositions, 100, 102–105, 115 defined, 100 Singular terms, 30, 31, 33, 120, 154– 155, 176, 178–179 context-dependent, 179 coreferential, 29 (see also Coreference) definite descriptions as, 55, 61–62 extension of Frege’s theory beyond, 25–29 Kaplan and, 100, 104 nature of, 25, 55, 126 Russell and, 55, 61–62, 68, 90 Skepticism, 94 Speaker meaning, 93, 191–193, 195. See also Meaning nature of, 192, 195–199 Speakers and sentences, 191–193 S-sentences, 162, 163 Statements, 3. See also Sentences and propositions Stimulus meaning, 183–184 Strawson, P. F. critique of Russell, 94 on demonstratives, 88 Donnellan and, 82, 85–86, 90 “On Referring,” 77 referential view of descriptions, 78 truth-value and, 72, 77, 85– 86, 94 on truth-value gaps, 85–86 Subjective ideas, 22 Subjective View: Secondary Qualities and Indexical Thoughts, The (McGinn), 116, 118–119 Subsentential expressions, 25, 32 Substitutional interpretation, 64 Synonymy, 16, 121 Syntactic ambiguity, 85, 179–180. See also Semantic ambiguity Syntactic theory, 169

Tarski, Alfred criteria of acceptability, 149–151 (see also Truth theory(ies): criteria of acceptability) object language, metalanguage, and, 155–158 theory of truth, 147–163, 165 Tarskian biconditionals, 152–156. See also Biconditionals; T-sentences Tautology, 6, 9, 71, 73. See also Analytic–synthetic distinction Temporality. See “Today” and “yesterday” Testimony, 200 “The,” 55, 68. See also Definite descriptions Thought, language of, 202 “Thought, The” (Frege), 111 “Today” and “yesterday,” Evans on, 128–130 Translation, 165, 166 radical, 183 Translation manual, 169, 170 Transparency condition, 197 Truth, definitions of, 147, 149–151, 153, 155–161, 163, 165. See also Partial definitions; Truth theory(ies) Truth biconditionals. See Tarskian biconditionals Truth conditions, theory of, 168. See also Truth theory(ies) Truth function, 120 Truth theory(ies), 148–149, 158. See also Redundancy theory of truth; Semantic conception of truth; Tarski, Alfred: theory of truth; Truth, definitions of criteria of acceptability, 149–151, 175, 188 empirical, 181–185 Truth-value, 75, 120, 166, 177 character, context, content, and, 105– 106, 108, 110–111

Index 225

concepts and, 120 contingent sentences and, 97 Davidson and, 189 Donnellan and, 78 (see also Truthvalue gaps) Frege on, 25–32, 78, 97, 120 intensions, extensions, and, 97–99 Kaplan and, 98 Kripke and, 99 names and, 176 as object, 25, 26, 28, 30, 31, 120 possible worlds and, 97, 98, 108, 189 reference and, 25–32, 97, 106, 120, 166 Russell and, 71–72, 77, 78, 94, 95 sense and, 166 of a sentence, 25–26 Strawson and, 72, 77, 85, 86, 94 Truth-value gaps, 77, 85–87 T-sentences, 156–157, 161–163, 174, 178, 180, 182. See also Biconditionals how to derive, 157–159 Twin Earth (thought experiment) and “water,” 134–145 Uniqueness, 69 Use-mention distinction, 8 Verification, 181–183, 185–187 Whitehead, Alfred North, 63 Wittgenstein, Ludwig, 24, 63, 186 on identity statements, 24 on saying vs. showing, 122 Tractatus Logico-Philosophicus, 63, 167, 186 Words. See also specific topics having a proper scientific sense, 124 ordinary and extraordinary use of, 18–20 terms that introduce concepts vs. objects, 100