You are on page 1of 236

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

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 languagesPhilosophy. 2. Language and languagesPhilosophy
Textbooks. I. Title.




Frege on Sense and Reference 1


Additional Machinery 10
The Conception of Sense 12
Ordinary and Extraordinary Use 18
Further Points on Sense and Reference 20
Problems with Freges Theory 23
Extension of Freges Theory beyond Singular Terms 25
Further Aspects of Freges Theory 31

Kripke on Names 35


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

Russell on Definite Descriptions 55


Indefinite and Definite Descriptions 55

Three Theories of Definite Descriptions 60
Indefinite Descriptions and Identity 63
Russells Rejection of Meinongs Ontology 65
The Details of Russells Theory of Descriptions 67



Problems with Russell 72

Primary and Secondary Occurrences 74

Donnellans Distinction 77


Referential and Attributive Uses 78
Denoting and Referring 84
Truth-Value Gaps 85
Evaluating Donnellans Distinction 87
Implication and Implicature 90
Further Objections to Russells Theory 94

Kaplan on Demonstratives 97


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

Evans on Understanding Demonstratives 115


The Fregean Theory of Indexicals 115

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

Putnam on Semantic Externalism 133


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

Tarskis Theory of Truth 147


Tarskis Criteria of Acceptability 149
Aristotle and the Redundancy Theory 151
Object Language and Metalanguage 155
How to Derive the T-Sentences 157


Davidsons Semantics for Natural Language 165


The Merits of Tarskis Theory as Applied to Meaning 168
Applying Tarskis Theory to Natural Languages 175
Empirical Truth Theory 181
Criticisms of Davidsons Theory 185


Grices Theory of Speaker Meaning 191


Background: Speakers and Sentences 191

Two Types of Meaning 193
What Is Speaker Meaning? 195
Consequences and Criticisms 199

Appendix: Kripkes Puzzle about Belief 203



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. Martinichs 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 students


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

Before we begin to expound Freges 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 worldwe
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?
Cant 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 allGottlob Frege.

Chapter 1

Freges 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 sentencesmore 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

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.
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 meaningshow they are able to tell us things about the world. What kind of
thing is meaning?
Freges article discussing these questions is not straightforwardit 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 Freges 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
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: 45=20.
Contemporary philosophers use identity instead of equality. The example 45=20 would be called an identity statement, asserting that the

Chapter 1

number 45 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 anotherthat 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 45, 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

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 Freges 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 morningwhen 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 bodyVenus. 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

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 propositionit
is merely a tautology. The numerical identityany 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 bodythe sunis
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, wouldnt
a = b express the same thing as a = a? So isnt it better to suppose that

Frege on Sense and Reference

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 propositionsbut 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 namenot 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.

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 usemention distinction: we use the name to mention an object; we dont use the name to mention itselfexcept 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
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

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 namevery 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. Isnt this proper
knowledge? It certainly isnt 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. Freges 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 ones
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. Freges
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 theorythe theory that
identity statements are only about objects, not linguistic items.


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 designationwhere 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

Frege on Sense and Reference


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 thingthe 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 perceptionit 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 Freges
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 machinerysome 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, Freges solution to the
problem of identity statements is to bring in modes of presentation as the
missing ingredient.


Chapter 1

1.4 The Conception of Sense

The last sentence from the passage quoted above illustrates Freges 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 recognizedthat 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 referenceit 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


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 Freges own example of the triangle. But in
our examples and his it seems as though sense has something to do with
perceptual perspectiveway 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.
Freges main point is that every expression in either of these two categoriesordinary proper names and definite descriptionshas 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


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 understandingone 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 peoples minds. However, the above passage illustrates Freges 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 persons mind who understands
the expressionit 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 propertythe 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


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 purposesif 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 forsense 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 Freges 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 aspectthe mode of presentation of the aspect, that is, an aspect


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.
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
corresponds a reference.

The relationship as explained above is rather fluidthe 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 sensesbank could mean a bank of a river or a bank for

Frege on Sense and Reference


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 Freges 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 nonsensethey 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 Freges 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


Chapter 1

sense without reference, and to form proper names that have sense but no
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. Ones
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


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
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 persons 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
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.


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 someones 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


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 mans 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 signs 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 peoples minds and the sense and reference of words. To reiterate a point made
above, Frege does not think that the ideas present in peoples 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 individualsthe 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


Chapter 1

Newtons 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
Freges 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 Newtons time
and we grasp that same sense now. Therefore, many subjective ideas can
correspond to the same objective sense. Freges 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, Its 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 levelswords, 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 observers 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 As 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


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
individuals physiological makeup.
1.8 Problems with Freges 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 Freges 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,
Freges theory is focused not on the objects themselves but on the mode


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 sortinstead he
attempts to make a seeming triviality into something substantial. Though
Wittgensteins 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 Wittgensteins more radical eliminative suggestion.

Frege on Sense and Reference


1.9 Extension of Freges 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 Freges 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 Freges 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 sentenceswe 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 itthe 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
Truean objectbecause 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


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 Truean objectsince
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-valueit is either true or falsebut
it is still unclear why Frege claims that a sentence has as its reference a
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 Freges complete list of all the objects in the world, in addition to
ordinary objectsevery person, planet, elementary particle, and so onwe
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 puzzlingby 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 arisesthe extension of sense and reference to complex

Frege on Sense and Reference


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 Freges 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 Freges 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
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,
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.


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 statements
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 Freges 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 Fregesthat the truth-value of a thought is determined
by the references of the parts of the sentenceseems 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


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 Freges 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
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 considerswhat


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 truethat 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
Freges: in Freges 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 caseif 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 Freges proposal that
the truth-value of a sentence is an object. Contrary to Freges 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


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
expressionsthey 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 objectsomething 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 Freges 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


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 Janes 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


has to exist, given Freges 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 Freges 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 Freges 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
Freges 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 Freges 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

We will now jump forward eight decades. The reason for this is that Freges
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 Kripkes critique of it in Naming and Necessity.1 Since Frege is widely credited with inventing the description theory of
names, Kripkes critique is directed largely at Frege and those who followed
his lead. Freges 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
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


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 senseis 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 Kripkes 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 languagea 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


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 Aristotles 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
purposeto 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


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
Freges 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
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 ones 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 anotherone 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


description theory is in accordance with Freges 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 names 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 Kripkes 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. Kripkes 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 Kripkes critique of the theory, fully aware of
its merits and content.


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 statementone 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 analytictrue by definitionand 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
Given what the description theory entails, Kripkes 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


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 realitiespossible worldswhere 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 Platohe
wouldnt 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 Platos 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


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 Platos 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 Kripkes 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. Kripkes 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. Lets 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 Aristotles 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


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 wont 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 designatorsso 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


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 2no 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 Kripkes 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 Kripkes 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 Aristotles 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


Under this theory, if the brain of Aristotle were transplanted into the body
of Einstein, the resulting person would be Aristotle. Since Aristotles 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 Kripkes other objections to see if they
overcome this limitation.
2.4 Kripkes Epistemic Objections
One of Kripkes nonmodal objections has to do with whether or not something is a priori. If a statement is analytictrue by definitionthen it must
be a prioriknowable 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 Feynmans 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 anotherboth 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. Kripkes point is that
the same information will be associated with the names by our nonexpert


Chapter 2

speaker, but this information is not sufficient to pick out one physicist from
the other. Therefore, the descriptive information in the speakers mind does
not determine the reference of the namesyet the speaker does manage to
refer to specific distinct individuals. He does not know any definite description true of his reference, so he certainly doesnt 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 nave
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. Kripkes 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 GdelSchmidt example.
Many people who have heard of Kurt Gdel will know him as the mathematician who proved the incompleteness of arithmetic. Therefore, we
can refer to Gdel with the definite description the mathematician who
proved the incompleteness of arithmetic. Kripke asks us to suppose that
Gdel 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 Gdel had plagiarized his incompleteness theorem
from Schmidt, and Gdel had unjustly received the accolades for devising
the proof.
In Kripkes thought experiment, the man referred to when someone
says the mathematician who proved the incompleteness of arithmetic

Kripke on Names


is Schmidt, not Gdel. In this case, the speaker has a false belief about
Gdelhe thinks that Gdel invented the proof, but he did not. His false
belief about Gdel cannot then constitute the description that determines
the reference of the name Gdel whenever he uses it. He refers to Gdel
with Gdel, while the description refers to Schmidt.
Another example of a GdelSchmidt 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 viewers 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 viewers 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 persons qualitative experience.
Equally, we are familiar with perceptual illusions that mirror the Gdel
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 viewers experience by itselfthis can misrepresent the object. The internal nature of the viewers 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 ones 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 speakers mind in terms of descriptions. Rather, it is an external relationship between the speaker and an object of another kind. This


Chapter 2

relationship could be of the causal kind, as in the perception case. Kripkes

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 speakers
mind. This analogy to perception helps articulate the intuitive faults in the
description theory raised by the GdelSchmidt case and others like it.
If the objections Kripke raises through the Feynman and GdelSchmidt
thought experiments are correct, then it follows that the classic description
theory is incorrect. Descriptions in ones 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
GdelSchmidt case). Therefore, there is no such description that determines
the reference of the name. This concludes Kripkes 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 Kripkes 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 Freges 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


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
Kripkes 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 Kripkes 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 Kripkes 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


Chapter 2

value of Hesperus is Phosphorus? As an alternative theory, Kripke mentions John Stuart Mills view, that the meaning of a name is simply its bearer.
However, as we saw in considering Freges 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. Freges 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 theorythat just leads straight to Freges 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 Kripkes arguments really refute it. We have
already covered objections to aspects of Kripkes modal argument that
could resuscitate the description theory. However, Kripkes 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
GdelSchmidt 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 Kripkes theory, and a more psychological
part that characterizes what is in a persons 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
Kripkes counterexamples. However, we still face the problem of answering
Kripkes epistemic arguments against the description theory.

Kripke on Names


If Kripkes 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 GdelSchmidt
thought experiment, one individual in the linguistic community refers to
Gdel by using the name Gdel, 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 Gdel. If language is as social as Kripke takes
it to be, then an individual who believes the wrong description of Gdel
is connected to other individuals who know correct descriptions of him.
Therefore, an individuals reference is fixed by being part of a linguistic
community in which some people associate correct descriptions with the
namethough not all do.
2.7 The Social Character of Names
Kripkes 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 GdelSchmidt example, which Kripke also
mentions. Giuseppe Peano was an Italian mathematician who axiomatized
arithmetic, so there are various axioms that are called Peanos 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 Dedekinds work, but some
people wrongly attributed the axioms to Peano, and so they became known
as Peanos 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


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 Dedekinds 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
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 meaningas with DNA. Kripkes
epistemic arguments do not refute the description theory when the description theory is proposed as a theory of the language of a community. Kripkes
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


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 persons name. This type of description is not vulnerable to any of
Kripkes 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 subjectthe 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 Freges pressing problem of
informative identity statements), and can be accommodated to deal with
Kripkes 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 namesand that is how names
refer. The primary way to refer is through descriptions, and names are secondarily based on descriptions. We dont 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


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 Kripkes 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 didnt say much about the analysis of descriptions themselves. Frege
treats definite descriptions as belonging to the same category as proper
namesthey 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 discussinga chapter from Russells 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


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 Russells 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


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 bbut 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 arent any unicorns to meet.
Russells 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


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 instancethough 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 functionwithout 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, Russells 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 personfor if so, who? Even if someone is out there making

Russell on Definite Descriptions


true what the speaker is saying when he says, someone stole my bike, the
speaker is not naming that villain; if he were, hed 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. Freges 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 Russells 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 existthat 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 Meinongs view. This is the view that
in addition to the ordinary objects that exist, there are things that subsist, or


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 Freges view that
such terms have sense but not reference. With Meinongs 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 manthe 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
Meinongs. 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.
Russells new theory is developed against the background of Meinongs
theory, which is also a version of Freges theory in assuming that definite
descriptions function as proper names.
3.2 Three Theories of Definite Descriptions
Before continuing with a thorough analysis of Russells 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 Russells
Definite Descriptions. We can use Russells first example, the king of
France, to explain these three theories. The description the king of France
is an empty descriptionone with no referencebecause at the time Russell used the example, France had no king. Although this description is

Russell on Definite Descriptions


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.
Freges 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 Freges view. Some readers could potentially be confused reading
this excerpt alone, because Russell is constantly making assertions that
contradict Freges theory. Russell presupposes that Freges 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.
Meinongs 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 Meinongs 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,
Meinongs 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 Meinongs theory, such fictional characters have being without existencesubsistence. Therefore, a name like Hamlet refers to a
subsistent entity. With this theory, a referential theory of meaning can be
maintained, without adopting Freges 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.


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 isnt 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 its 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


uses his sensereference 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 centuryuntil 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. Russells 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
Russells 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


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 sentences 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 Russells 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 Russells 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 Russells theory is that we can explain how to speak
of nonexistent things without creating an entirely new ontology. In Meinongs view, we need subsistent golden mountains to analyze I climbed

Russell on Definite Descriptions


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
isis used in both statements of predication and statements of identity.
Russell believes that an ideal language would not have such ambiguities.
3.4 Russells Rejection of Meinongs Ontology
Russells stalwart rejection of the Meinongian ontology can be found in the
following impassioned passage:


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 Shakespeares 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: Shakespeares 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 Russells point here. To say that Hamlet
is an existent in Shakespeares 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 Shakespeares 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 meaningshe 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


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 ones imagination in the same way that a horse exists in a paddock.
As to whether the passage above refutes Meinongs 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 Russells theory, we will assume that he is correct about how we should deal
with definite descriptions that refer to these nonexistent entitiesthat is,
they have no reference at all.
3.5 The Details of Russells 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


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 Russells 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, Russells 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


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 Russells basic
ideait 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 namesthey 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


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. Russells 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 bearernot 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


he suggests that a logically proper name is a demonstrative, because a

demonstrative can refer directly to sense data. In Russells 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 dont 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 Russells Millian theory. For instance, how does Russells 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. Freges 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.
Russells 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 Russells strict
criteria for logically proper names. Thus Hesperus is not a name, for him,
but this sense datum of a luminous point is. In Russells system, there are
no identity statements that are informative and contain ordinary names.
One important consequence of Russells theory that generated much
discussion is how he handles truth-values. According to Russell, the


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 Russells analysis implies that it straightforwardly false.
3.6 Problems with Russell
Although in the previous sections we explained Russells 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
wordswith the exception of this and that and a few other words of which the
meaning varies on different occasionsno 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 descriptionsdefinite or indefinitethat 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


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.
Russells 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 Russells 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 partand 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 Russells 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. Russells


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 Russells
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 Russells 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


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
shes pregnant or as There exists a queen of England and only one and
shes 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 Russells theory of descriptions. In the
next chapter we will look at some possible criticisms of Russell.

4 Donnellans Distinction

To summarize our progress so far, we have examined two major theories of
descriptions: Freges theory and Russells theory. In Freges theory, descriptions are proper names referring to things. In Russells theory, logically
proper names refer, but descriptions do notthey 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.


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. Donnellans 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 Russells stance is that a definite description is
a quantifier, we could label Russells theory the quantifier view of descriptions. But Donnellan chooses to call it the attributive view. The following
passage outlines his understanding of these terms:

Donnellans Distinction


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 jobcalling attention to a person or
thingand 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 Donnellans notion of attributive use,
we could paraphrase the sentence The king of France is bald as Whoever is uniquely king of France is baldperhaps 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 Russells 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 audiences 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 audiences attention
to the right person most effectively. Her purpose was to identify a certain
individual and make a remark about himshe couldnt care less about the


Chapter 4

description itself. She wanted to point out the guys pensive look and any
mode of designation would do.
Donnellans 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, Smiths murderer is insane! When he said that, he did
not know the identity of the murderer. That statement could be rephrased
as Whoever Smiths 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, Smiths 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 Smiths murderer though he
is on trial and behaving erratically. Donnellan thinks the juror has still identified that individual even though he is not Smiths 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 Smiths 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 jurors 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

Donnellans Distinction


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 Russells 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, Whos the man drinking the martini? She is not
intending to identify a particular individual as you were in the previous
exampleindeed 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 martinishe 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 yourselfdoes 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.


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 Donnellans 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 Donnellans 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 Smiths 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. Donnellans main argument deals with these two
uses of descriptionsthe 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
Smiths murderer case, Jones might not be the murderer but a juror can still
identify Jones by saying, Smiths 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

Donnellans Distinction


Donnellans 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 Smiths 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


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 Smiths 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 Smiths murderer denotes Brown. In that case,
the juror refers to Jones by saying Smiths 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 notiona strict and literal interpretation of the
phrase Smiths 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 Russells theory, but there are
pragmatic uses in which a speaker refers to something other than the
As a matter of fact, at one point in Reference and Definite Descriptions
Donnellan clearly states that he is not arguing against Russells semantic
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

Donnellans Distinction


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 speakers 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 speakers intentions.)2

This is a very important passage in terms of the significance of Donnellans

arguments. He claims here that there is no semantic ambiguity in descriptions. By semantic ambiguity he means what the words actually mean in
the languagetheir 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 Russells 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, Smiths 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


Chapter 4

thinks the onlooker is still referring to the man there even though the man
carrying a rifle does not fit the speakers 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 whateverman,
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 speakers 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 Russells 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 Russells. Although Donnellan believes that
Russells theory is incomplete because it does not recognize the referential
use of descriptions, he still thinks that his conception of descriptions is
analogous to Russells 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 Russells
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 Russells 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 objectthey are namelike.
Whether or not the description correctly describes the object is irrelevant,

Donnellans Distinction


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
Empirea 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 Donnellans referential uses.
4.5 Evaluating Donnellans Distinction
When assessing the cogency of Donnellans 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. Lets 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 ariseswhom 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 animateddid she refer to him


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: hasnt 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 Donnellans 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
speakers misuse of the demonstrative is exactly like the hostesss 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

Donnellans Distinction


interested observershe is just impartial. Someone may say, Im just completely disinterested in that subject matter, and the hearer, knowing this
common error, can infer from the speakers 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 notalmost 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 oneso 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 Donnellans 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


Chapter 4

words actually mean. Thus a speaker can express a belief about Jones by
using words that denote Brown (Smiths murderer). Donnellans point is
a purely pragmatic one and does not undermine any semantic theory. Since
Russells and Strawsons theories were offered as semantic theories, Donnellans 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 Donnellans position, we will now bring in some points
made in an excerpt from Stephen Neales 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.
Professor Horatio Handwavy, PhD

The committee reviewing John Smiths 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 Smiths 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
Smiths 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 didnt straightforwardly lie, but he certainly gave

Donnellans Distinction


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.
Neales main problem with Donnellan is that he disregards this distinction. Donnellan is suggesting that Russells 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 Donnellans pragmatic points as having any implications for semantics. Although Neale
never mentions it, we have discussed a passage from Donnellans original


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 Russells 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 Russells semantic analysis, but he himself accepts that his considerations do not bear on semantics.
Let us suppose that Russells 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 Russells 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 Russells 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 himthat there is no semantic ambiguity. Yet he doesnt
seem to appreciate the significance of this admission.
Neale believes that Donnellans arguments show the necessity of bringing to bear Grices distinction between the proposition expressed and the
proposition meant. To understand why this distinction matters, we shall
return to Donnellans examples. Let us consider again the Smiths murderer case where Jones is the man in the dock. The juror sees Joness erratic
behavior and wants to express his belief that Jones is insane, so he says,
Smiths 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 Donnellans referential use.
However, the proposition expressed by the sentence itself (Smiths murder
is insane!) is that Smiths 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 Donnellans
examples with this Gricean distinction allows us to see that there are two
different propositions associated with the utterance of the sentence in

Donnellans Distinction


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 expressand 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 meantfor 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 Russells theory. Even though Donnellans critique of Russell seems misguided for the


Chapter 4

reasons given, other objections to Russells theory persist. Let us quickly

survey these objections.
4.7 Further Objections to Russells Theory
The first objection is Strawsons: empty descriptions give rise to statements
that are neither true nor false. According to Russells theory, The F is G
expresses an existential propositionthat there exists an F. If there is no F,
the sentence expresses a false proposition. Strawsons point is that Russells
assignment of truth-value is intuitively wrongit 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 Russells theory it would be simply false, because of the lack of golden
mountains. Such a statement does not seem to fit Russells theory at all.
Russell might reply that this is just a matter of ordinary languagehe has
shown that contrary to appearances the statement is false. There is something to be said for Russells 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, Russells 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 hasnt
said anything. According to Russell, however, descriptions are complete

Donnellans Distinction


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 Russells 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 Russells way, then simply uttering a name is already
expressing a complete propositionto 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 sentencethe subject part
not a whole sentence. Again, this makes us wonder if Russell has the correct
Nonindicative sentences raise more worries about Russells theory. Consider an example of an imperative sentence: Kill the King of France! Using
Russells 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 Russells paraphrase, the sentence comes out as
nonsense. Russells theory cannot be applied mechanically in this case. Russell never discusses how he would handle cases where descriptions occur in
imperatives. It doesnt 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 existswhich 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 Russells 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 cigarshe took it for granted that said author exists. If


Chapter 4

the definite description occurs in a propositional attitude context (in this

case wonder whether), then we obtain the wrong analysis when applying
Russells theory. So not all occurrences of descriptions fit Russells 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 Russells 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 Russells 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 Russells theory. But then we have admitted
that not all descriptive phrases fall under Russells 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
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 descriptionsand 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 Russells theory. Try it and see.
Russells 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 sentencethe proposition it
expressesis 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 Freges 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.


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


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 Kripkes 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


Chapter 5

inside propositionsthere is no room for them inside propositions. Propositions, for the Fregean possible worlds theorist, are made of intensions, not
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 Russells 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


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 thats 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 Kaplans 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.
Kaplans 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 Freges 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 Freges 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


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,
Im hot, he is referring to himself because the context of that utterance
includes the speaker. If you, the reader, say, Im 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 nameyou dont 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, Im 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.
Kaplans view of indexicals is rather like Kripkes 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


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 beforea 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 Kripkes 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 rigidthe
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 Kaplans 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
propositional component.

Kaplans 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.


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 evaluationthe what-is-said in a given contextas propositions. Dont 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 Russells. 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


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
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 propositionas 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 cant 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


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. Dont we have variation in extension with constancy of intension in both cases? Kaplans 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 cant 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. Descartess 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


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 languageits 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


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, Im hot, and you say, Im 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 contentcharacter 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 Im 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 Kaplans picture, indexicals have two
sides to their meaning, whereas in Freges 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 Kaplans 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. Freges theory works

Kaplan on Demonstratives


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 Englands 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
Kaplans 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


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 extensionsit 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, assertedand 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 truethat depends on the circumstance of evaluation. In the case of

Kaplan on Demonstratives


character, the function takes contexts as arguments and produces contents

as valueswhereas 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 Kaplans 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 theoryFregean 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.


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, Im
pregnant, and another speaker says, Shes 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 Russells theory, Freges 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 Russells 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 Freges 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 Freges view reference is not constitutive of meaning. In Freges 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 Freges theory recognizes a level of meaning above reference, his
theory of meaning is still one-dimensional, because sense does all the work.
Kaplans theory can be characterized as having two levels or three levels,
depending on how each level is understood. Kaplans theory of meaning
has two levelscharacter and contentand 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 Freges 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


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 its raining. How do I say the same thing as I said
today tomorrow? Suppose I say tomorrow Today its raininghave I said
the same thing as I said the day before when I said, Today its 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 its 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 speakers context. So, in Kaplans 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 cant 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 Freges 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 cant be character, but can it be content? No, the sense cant be identical to the content either, in Kaplans sense, because in Freges 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 Kaplans scheme for Evans to
equate with Fregean sense.


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, Im 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 Russells theory of descriptions to the description associated with the indexical, thereby combining Freges view with
Russells. 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, Im a philosopher, what Im 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


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 otherhence 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-styleby
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 its 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. Evanss
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


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 toosay, 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 didnt realize it for a second or
two. Suddenly, you are thunderstruck with the realization, Oh, its me Im
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 cant remember anything. I am going to suppose
that this unfortunate individual is myself. The doctor asks me, Where
do you live? and Whats your name? and I dont know because I cant
remember. I cant remember any facts about myself. I might say, I cant
remember anything about myselfyet I successfully refer to myself. So
there I am in the hospital and I dont 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


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 cant be that I succeed in
making this kind of first-person reference in virtue of knowing true descriptions about myself. I certainly dont 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 dont 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 Evanss 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 Freges 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,


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 Freges 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 Freges 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 Freges system is meant to be a theory of a speakers 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 referenceit is the mode in which the object is presented
to the persons 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 ithow it enters my thoughts.
Evanss point in regard to this specific part of Freges 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


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 Freges 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?). Evanss 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. Evanss 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 notthough 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.
Evanss 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. Evanss 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.


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 isthere is no way to specify sense
directly. We dont speak about senses in specifying them. If we say, The
reference of Hesperus is Hesperus, intending to convey the names 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 Evanss theory, the sense of
a word is not specified by giving the sense of another word. At this point
Evans invokes a suggestion of Dummetts that involves using a distinction
of Wittgensteinsthe 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 penhe just shows it to you.
As a result of showing the pen, the observer gains knowledge without the
mediation of language. Evans is using Wittgensteins 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
iswithout 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 Im English. I get the
point across without stating it in so many words.
Evanss 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 sayingshowing
distinction promises to get Frege out of a tight theoretical corner. It makes

Evans on Understanding Demonstratives


sense of the elusiveness of senseat least it attempts to do that. Senses

belong to the realm of that which cannot be said but can nevertheless be
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 expressions 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, Evanss 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 Russells 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 Evanss 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


Chapter 6

of Russell about names while still trying to answer Freges 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 Freges 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


(so fictional names dont 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 isnt 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 godsand 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. Meinongs view gives a smooth


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 arent part of factual discourse; but
isnt 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 thoughtand 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 Evanss 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?
Dont I still enjoy a perceptual statea way of seeingeven 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


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 Evanss 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 ones 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. Evanss 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 Russells 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 referenceas general concepts do. The extra ingredient of sense cannot itself be
reference dependent.


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 Freges 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 cant 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 yesterdayI 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
waythere 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.
Evanss 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


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 Freges 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


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 Kaplans character and
content, which is closer to Freges 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 speakers 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


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 dayas 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 dayit 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 others 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

Our earlier discussions of indexicality will help us to understand the force
of Hilary Putnams arguments in Meaning and Reference. For indexical
expressions, the classic theory of descriptive intensions that determine
extensions looks highly implausibleas 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 dont 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 roleand not simply what occurs descriptively
inside the speakers mind. What you refer to can depend on who and where
you are, not just on what you thinkit depends on external context, not
internal descriptions. That is, indexical reference is fixed externally by the
speakers 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 speakers internal concepts do suffice to fix what he
refers to. Thus externalism is correct for indexical reference, but internalism
is correct for (pure) descriptive referenceas 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.
Putnams focus is on natural kind terms like water, aluminum, and
tigerwords that stand for types of object found in naturenot words for
human artifacts like table, computer, or president. He wants to know
what these words mean, particularly how they determine their reference. At


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 termswe 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 termsthat is, internal psychology does not determine a speakers reference. He thus rejects the old view to
the effect that a speakers 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 usthey 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


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 cant 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 liquidsthey would give the same
descriptions of itbut the context of use is different, so the reference is different too. The speakers dont 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
Putnams 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 persons psychological state does not determine what he means by his


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 dont
mean the same thing by their term water because they dont refer to the
same thing. The meaning of the word cannot be deduced from the speakers
psychological state. It depends on other extrinsic factors, and we will soon
see what those are. The speakers 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
Let us restate Putnams 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


Earth, when the speakers use the name Aristotle they refer to their Aristotlenot 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. Putnams 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 exampleit 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


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 themthey
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 Kripkes example of Feynman and
Gellman (see chapter 2). Not familiar with the particulars of their work, a
speakers 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
speakers 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 speakers meaning cannot be read off his psychological state,
but can only be gleaned from his contextspecifically, 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


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 headswhat 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.
Putnams recipe for producing these outside-the-head cases is very straightforward: we just vary the speakers 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 propositionswe 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 Putnams Twin Earth cases but question
whether they prove that meaning is not in the head and hence not


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
statebut 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
extensionthat sense determines referencethat Putnams 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 Kaplans 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 headinstead, 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 Putnams view of what his examples show.
One thing he concludes is expressed in the following passage:
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


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 beechelm 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 Putnams 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 speakers 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 itselfnot 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 anybodys


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 someones head, individually or socially. Rather, meaning is
a function of the speakers actual external environment. The environment
itself determines what words mean. So, Putnam concludes, meaning is not
in the headit 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 speakers 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 speakers 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
demonstrativeas in Water refers to that liquid, said while pointing at
H2Oand 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


7.4 Criticisms of Putnam

What is the best way to characterize the upshot of Putnams 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 speakers 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 contentas 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 Putnams 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 Putnams 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 Putnams,
wondering whether we should conclude instead that psychological states


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. Lets 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 eitherand for essentially the same reason. The content of
psychological states is also fixed by the persons actual environment. That
is, the full propositional content of psychological states is partly fixed by

Putnam on Semantic Externalism


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 himto 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 headbecause 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 persons 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.


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 Tarskis Theory of Truth

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 Tarskis theory to propose
a theory of meaning for natural languages, as we will see in the next chapter.
Tarski had tamed truth, rendered it scientificquite a feat. The adjective
Tarskian took on the canonical stature of the adjective Fregeanas in
the Tarskian theory of truth and the Fregean theory of meaning.
Still, there is controversy about what Tarskis 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


Chapter 8

for that we cannot do better than to attend to Tarskis own words. This will
be our procedure in what follows.
Let us first talk a bit about the background to Tarskis 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 ones 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 propositionfacts 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 ones 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
Lets 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

Tarskis Theory of Truth


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 doesnt
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 realitynot 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 Tarskis Criteria of Acceptability
Tarskis 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 truthwhich 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
achieveonly 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


Chapter 8

of truth must be materially adequate and formally correct. To start with,

material adequacy means simply that the definition should (in Tarskis 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.
Doesnt 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 Tarskis 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

Tarskis Theory of Truth


(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 anybodys 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 languagesthe 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 languagethey 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 truthintuitions which find their expression in the wellknown words of Aristotles 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


Chapter 8

For the sake of simplification, we can cut out the negation part of Aristotles
formulation and still express the essence of Tarskis view. Truth is saying
of what is so that it is sothis is Aristotles 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 saysand thats 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, Thats true. By saying Thats 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 dont 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 Einsteins 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 Tarskis 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 theorywhich is sometimes
also called the disappearance theory or the disquotational theory. It is as if the

Tarskis Theory of Truth


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. Tarskis 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 Tarskis theory we do not have to bother with such
questions. There is no need to bring in the concepts of correspondence
and factswe just repeat, Snow is white after if and only if. And snow
being white is not philosophically problematic, because we know what that
isthere 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:


Chapter 8

If we wished to adapt ourselves to modern philosophical terminology, we could

perhaps express this conception [Aristotles] 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. Tarskis actual theory has a very different form, as we shall see.
To get started on the details of Tarskis 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
namesay, 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 sentenceit is just a sentence in use, referring to
snow and whiteness. For this reason, in logic the variables conventionally

Tarskis Theory of Truth


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 sentenceand 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 Tarskis theory. This is the distinction between the object


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

Tarskis Theory of Truth


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 dont 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 beliefsthe 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 dont 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


Chapter 8

the ultimate goal of Tarskis 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
Tarskis requirements, as so far stated.
That would be a perfectly adequate definition of truth by Tarskis 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 conjunctionand 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 Tarskis theory actually

Tarskis Theory of Truth


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 dont have to form an infinite
conjunction of all the T-sentences, even though that would have satisfied
Tarskis condition of material adequacy. Strictly speaking, we should amend
Tarskis 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.
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 Cant
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
satisfactionobstacles 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


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
satisfieswe cant 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

Tarskis Theory of Truth


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 Tarskis 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
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


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

Tarskis Theory of Truth


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 functionassuming 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
Tarskis construction, as we examine Davidsons claim that Tarski-style truth
theories provide a framework for doing the semantics of natural languages.
Here we will ask what general significance Tarskis 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 Davidsons Semantics for Natural Language

Tarskis intention was to define the concept of truth for formalized languages. Donald Davidsons aim is to use Tarskis theory of truth for formalized languages to give a theory of meaning for natural languages. Davidson
is therefore using Tarskis theory for a purpose that he did not originally
intendas 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, Tarskis 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 Davidsons 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 Freges writings, because when Frege is discussing sense and reference one of his concerns is the reference of parts of


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 dont 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 cant

Davidsons Semantics for Natural Language


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 sentences 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. Havent 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


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 Davidsons 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.
Davidsons 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 itas 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 anotheras when I say John is quick and Jill is quick.
We human speakers spend our lives recombining old words into new patternswe 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 Tarskis Theory as Applied to Meaning
Davidsons 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

Davidsons Semantics for Natural Language


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 Chomskys 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


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
speakers 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 theorybecause 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 Davidsons
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 dont talk about the things that words and sentences mean.
Davidsons 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 Tarskis theory is right for the job.

Davidsons Semantics for Natural Language


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 childs 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
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 Tarskis truth theory fulfills these conditions and
accommodates the general features of meaning we have articulated. Thus
Davidson claims that Tarskis 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


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 Tarskis 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-

Davidsons Semantics for Natural Language


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
A salient feature of Tarskis 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. Shouldnt we do a bit more to say what the meaning of the
sentence Snow is white is? Shouldnt 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 dont 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 whiteand 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


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 worldsas 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 dont 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 philosophers inventionit 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

Davidsons Semantics for Natural Language


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 Tarskis Theory to Natural Languages
When dealing with a language based on ordinary predicate logic, Davidsons use of Tarskis 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.
Hasnt the theory dealt with only a fragment of language as we have it? This
is already a problem for Tarskis 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
Tarskis theory to natural languages in their entirety. If Tarskis 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 Tarskis methods to different areas of language. How do we give the meaning
of the parts of language that dont 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. Lets 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 Tarskis


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
cant say that objects like John satisfy quicklythat 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. Tarskis theory just specifies references for singular terms by means
of designation axioms. Sense is ignored. For Tarskis 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

Davidsons Semantics for Natural Language


to all the linguistic constructions of natural languageand 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
That3). 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 Tarskis extensional theory and not
run into any problems. Likewise, the sentence John said that is completely extensionalin 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 Davidsons, 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 Tarskis 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 treatmentsince the paraphrase does have truth conditions.


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 Tarskis 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 hotwithout me being hot. Davidsons 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 hotthe 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?
Davidsons 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

Davidsons Semantics for Natural Language


it wont 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
universalthe 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) snowit 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 categorythat
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 termsas
names or predicates.) Similarly, in Davidsons 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 modestits 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 planewas it just the boat that was


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 dont 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 removedRbank 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 resultsthe theoremsbut 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 sentences 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 Tarskis 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. Its the journey as well as the
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 its 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 wordsand 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

Davidsons Semantics for Natural Language


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 predicateno 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 anywaythe spatiotemporal objects that
are white. Similarly, Tarski does not interpret the connectives by specifying a reference for themhe doesnt 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 meaningswithout 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


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 dont.
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
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 Davidsons 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 elses truth conditions right? We cant consult our own

Davidsons Semantics for Natural Language


competence in the language, because we dont have any. We have to find

out what the alien speakers mean by their words. At this point in the paper
Davidson alludes to Quines 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. Quines 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
Quine gives the example of the word gavagai in the radical translation
thought experiment. The traveler immerses himself in the tribes 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 wont understand
what they say, and they dont speak his language either. The first thing he
would do is to discover when and where they utter gavagaiin 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 meansit 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


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 gavagaiin
addition to rabbitsis 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 dont 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 natives 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 Quines thesis. His concern
here is with Quines general picture of how we set about devising and testing interpretations of the language of others. This brings us to Davidsons
theory of what he calls radical interpretation. Davidson has a whole article
(Radical Interpretation5) 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 charitythat is, the interpreter must

Davidsons Semantics for Natural Language


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 gavagaihe 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 Davidsons Theory
Let us assemble some criticisms of Davidsons theory of meaning. First we
can ask whether Davidson says enough about what meaning isand about
what our grasp of meaning consists in. His guiding idea is that a theory of
meaning assigns truth conditions to sentences, and a speakers 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 conditionsespecially when we restrict ourselves to
homophonic statements of truth conditions? Isnt this just way too minimal? Cant we ask what this knowledge of truth conditions itself involves?
There are different options to take in response to this line of criticism.
Davidsons 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


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 speakers mind when she understands sentences.
Alternatively, we might borrow an idea from Wittgensteins 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 sentences
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 blackI would have misunderstood the sentence. This
Wittgensteinian account of knowledge of truth conditions goes beyond
Davidsons 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 Davidsons 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 lookwe 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 dont, you are clueless. Some philosophers (often describing themselves as positivists) try to elevate these truisms into a theory of what knowing truth conditions isit comes down to

Davidsons Semantics for Natural Language


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. Tarskis 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 Tarskis 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 Davidsons 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 dont 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 cant 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 Davidsons 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


Chapter 9

predicates, but he is an enthusiast of logical form attributions. But consider Russells theory of descriptions (see chapter 3): here we break the word
the down into a complex quantified conjunction. Why isnt 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, Davidsons 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 isnt 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 Quines rejection of the
analyticsynthetic 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 thatnot by a long chalk. In elementary

Davidsons Semantics for Natural Language


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 Davidsons 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. Cant 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
Tarskis 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.


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 dont have meaning in virtue of their identity as marks or soundstheir 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

10 Grices 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 isnt
much reiteration of points, so it calls for careful reading. Lets 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 meanhow 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 handsthere 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 Godit 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.


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 meaningwords 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 uswe 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 speakerspeople 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 statethat
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

Grices Theory of Speaker Meaning


explain semantics in terms of psychology. Sentence meaning comes down

to psychological facts about speakers. This seems to conflict with Freges
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 Grices 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 couldnt 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
Grices 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


Chapter 10

for common words: trouble and strife for wife, apples and pears for
stairs, and so on. The speaker says, I cant 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 hasnt 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 cant 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 indispensablehe 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 cant do
this for natural meaning. We can say that the conductor meant, The bus is
full by his three rings, but we cant 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 measlesthey 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 meaninga 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 meaningwe cannot
refer to what was meant by the spots.
Grices terminology here is not altogether perfect, though it has become
entrenched. He talks of non-natural meaning, but there is really nothing

Grices Theory of Speaker Meaning


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 meaningthat 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. Stevensons 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 Xs
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.


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. Grices point is that no such thing was meant. The agent did
not mean anything by his actionhe 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 ones
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 anybodyI wasnt 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 areand 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 thingsall
the things that people might believe who hear me speak. The conditions
proposed by the causal theory are hopelessly weak.

Grices Theory of Speaker Meaning


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 intentionspecifically, 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 Bs 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 intentionthe 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 agents 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


Chapter 10

Salome should recognize this intention. He does not attempt to conceal his
intentionno 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 Grices 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. Xs 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 audiences
reason for forming the infidelity belief is the evidence contained in the
photograph itselfit makes no difference how Mr. X views my intention
in showing it to him. He might find the photograph in his wifes closet, so
there is no act of showing at all. But in the drawing case, Mr. Xs reason for
forming the infidelity belief is not the drawing itselfthe drawing by itself
is not good evidence for that beliefbut 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 beliefand 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

Grices Theory of Speaker Meaning


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
audiences 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 speakers meaning that p is equated with the intention to induce
a belief in an audience by means of the audiences 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 intentionsthat 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 Grices picture of language and its raison dtre. We all have a great many beliefs about the world, often acquired
by observation. Imagine a time before language evolved, but when people


Chapter 10

still had a stock of beliefs. As a social species, we might wish to induce

some of our beliefs in othersthat 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
othersso 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 cant 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 wasnt your reasonyou 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 dont believe me, but let me show you thisand then we whip out
the clinching piece of hard evidence. Sentences, on this conception, are

Grices Theory of Speaker Meaning


really evidence surrogateswhat we resort to when we cant just point to

the facts or produce the smoking gun. Sentences take up the evidential
slack. This is the lesson buried in Grices account of speaker meaning: you
dont 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 Grices 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 audiences 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 sameso 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 dont 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 EnglishI might have been a
French speaker, say.


Chapter 10

Fair enough, it may be saidEnglish 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 languagewhat 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 languagecombinatorial, finitely based, recursive, referential. Arent 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 meaningbecause it cant 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

Appendix: Kripkes Puzzle about Belief

Finally, let us consider Kripkes 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. Kripkes 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 doesnt 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


the same person. And so the subjects beliefs are all about a single person
throughout the experiment, though he thinks they are about different people. Let us suppose that the subjects 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 Kripkes 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 doesnt 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 individuals face in the picture, the subject says yes. The psychologist then shows the subject another picture, and based on that persons
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 subjects 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 Kripkes
Pierre example. The subject believes that that person is trustworthy and
that that person is untrustworthyyet these are same person. Suppose the
experimenter says to the subject, Just for convenience Im 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


person. This is a perfectly intelligible situation, just as Kripkes 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 sameone 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 persons 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 Albertthis 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 Kripkes, but it is
more streamlined because it requires that we use only one language. We
also make explicit the subjects 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 doesnt 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


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 secondsspecifically, 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 electronsdespite 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 doesnt think so, because he doesnt
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 clashand 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
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 Larryyou 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


really called Larry because you have heard others refer to them by that
name, but there is nothing impossible theredifferent people often have
the same name. The problem is that you have a false identity belief concerning Larryyou 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 genteelhe 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 namea 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 separatelyit 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. Kripkes 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


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 arisesits 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 Leibnizs 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 dont 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 Kripkes example, Pierre does not believe the identity Londres is
identical to Londonhe 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 beliefthose 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


identical to London. That is, if we confronted Pierres 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
couldnt 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 dont 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 statementa true identity statement.
This is not meant to solve or remove Kripkes 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 subjects 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.


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),
6. Ibid., 113114.
7. Ibid., 114.
12. Ibid., 114115.
13. Ibid., 115.
14. Ibid., 115116.
15. Ibid., 116.
16. Ibid., 117.


2 Kripke on Names
1.Discussion in this chapter follows the excerpt from Saul Kripkes Naming and
Necessity (Lecture II) in Philosophy of Language: The Central Topics, 128146.
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,
2. Ibid., 148.
3. Ibid., 150151.
4. Ibid., 153154.

4 Donnellans 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,
2. Ibid., 187.

6 Evans on Understanding Demonstratives

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


7 Putnam on Semantic Externalism

1. Hilary Putnam, Meaning and Reference, in Philosophy of Language: The Central
Topics, 275.

8 Tarskis 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, 3031.
4. Ibid., 32.
5. Ibid., 38.

9 Davidsons 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 Grices Theory of Speaker Meaning

1. Discussion in this chapter follows H. P. Grices paper Meaning in Philosophy of
Language: The Central Topics, 6976.
2. Ibid., 7273.


Appendix: Kripkes Puzzle about Belief

1. The discussion in this appendix concerns an excerpt from Saul Kripkes A Puzzle
about Belief, in Philosophy of Language: The Central Issues, 257263.


Agent meaning, 195. See also Meaning

Analyticsynthetic distinction, 4, 67,
4041, 45, 46, 129, 188
A posteriori propositions, 6, 9, 40
Appearances, 6
A priori propositions, 4, 6, 7, 4041, 45, 46
Aristotle and redundancy theory,
Aspect of object and sense, 1416, 145
Atheism, 59
Attributive view of descriptions, 7884
Axiomatized arithmetic, 51
Base axioms, 175, 178
Belief, Kripkes puzzle about, 203, 209.
See also Kripke, Saul
Belief ascription, 203207, 209
Biconditionals, 152156, 188189. See
also T-sentences
material, 189
strict, 189
Brain and language, 202
Causal chain theory, 4849
Causal theory of meaning, 195, 196
Causal theory of perception, 47
Character, 107108, 110111, 117,
130131, 140
content and, 107108, 110, 111, 130,
context and, 107108, 110

defined, 107
as function from context to content,
nature of, 107, 110
Character compositionality, 112
Character- vs. content-type meaning,
111. See also Character: content and
Charity, principle of, 184185
Circumstances of evaluation. See Conditions of evaluation
Cockney rhyming slang, 193194
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, 104110
context of use and, 105109
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, 106109, 111, 142


Context-dependent singular term, 179

Context independence, 106107, 109
111, 117, 133
Context of use, 105109
Contingency, 4045, 5253, 98
Contingent facts, 4142
Contingent properties, 43
Contingent sentences, 40, 97
Contradictions, 5657
Contradictory beliefs, 203209
Conversational implicature, 90, 91. See
also Implications and implicature
Coreference, 29, 30, 101, 123, 127
Correspondence theory of truth, 148,
149, 153154
Creationism, repeat, 205206
Davidson, Donald
applying Tarskis theory to natural languages, 175181
empirical truth theory and, 181185
and the merits of Tarskis theory as applied to meaning, 168175
semantics for natural language,
theory of meaning, 175, 178181
criticisms of, 185190
Definite descriptions, 40, 41, 46, 9596,
context dependence and, 102, 108
109, 111
defined, 13
impure descriptions and, 5354
vs. indefinite descriptions, 5560, 65,
indexical expression and, 102, 109, 111
meaning and, 17, 61, 70, 77, 95, 98,
126127, 143
mode of presentation and, 38
names and, 36, 43, 60, 62, 65 (see also
under Description theory of names)
proper names and, 13, 5355, 60, 62,
6971, 102, 119120

as quantifiers, 55, 59, 60, 78, 96, 104

reference and, 13, 14, 16, 17, 20, 48,
5152, 61, 67, 68, 79, 8485, 91,
109, 117, 126127, 134, 143
referential and attributive uses of,
rigid designation and, 4244, 99
Russell on, 5575, 104
semantic ambiguity and, 92
sense and, 13, 14, 17, 38, 117, 125
as singular terms, 55, 6162, 68
theories of, 6063, 77
Definite Descriptions (Russell), 60, 70.
See also Definite descriptions; Russell,
Demonstrative reference, 54
Demonstratives, 54, 88, 96, 142
defined, 101102
Donnellan on, 80, 88
Evans on understanding, 127 (see also
Evans, Gareth)
Frege and, 127
identity propositions that connect,
indexicals and, 101102
Kaplan on, 97 (see also Kaplan, David)
nature of, 71, 96
proper names and, 54, 71, 88
Putnam on, 142
Russell and, 7071, 88, 96
Strawson on, 88
Denoting, 8485
vs. referring, 84
Descriptions. See also Definite descriptions; Indefinite descriptions
essential, 5253
impure, 5354
as reference dependent, 127 (see also
Reference dependency)
Russells theory of, 6772, 116
semantics of, 90
theory of (see Description theory of


Description theory of (indexical) sense.

See Description theory of names: indexicals and
Description theory of names, 5354, 69
definite descriptions and, 3639
demonstratives and, 54
Evans and, 117, 119, 127
Frege and, 35, 36, 39, 50, 116, 117,
historical perspective on, 35, 39
indexicals and, 102, 116119, 127
John Stuart Mill and, 50, 125
Kaplan and, 102
Kripkes critique of, 35, 36, 3945, 52,
54, 102 (see also Kripke, Saul: critique
of Frege)
objections to, 4952, 54
names and, 46, 48
objects and, 47
overview, 36, 38
Perry and, 117119
reference and, 49, 50, 54, 102
Russell and, 69, 95, 116
social/socialized, 51
Designation, 159
Designation, mode of, 8, 1011
Designation axioms, 162, 172, 176, 187
Direct designation, 70
Direct reference, 101105
vs. rigid designation, 103106
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, 9194
distinction between denoting and referring, 8485
evaluating, 8790
distinction between referential and attributive uses, 7884

on implication and implicature, 9091

Neales criticism of, 9092
P. F. Strawson and, 82, 8586, 90
Russell and, 78, 8387
truth-value and, 78
on truth-value gaps, 8587
Dual-aspect semantics, 111
Empirical truth theory, 181185
Empty names, 125126
Equality (mathematics), 34
Essential indexical, 119. See also I
Evans, Gareth
description theory of names and, 117,
119, 127
Frege and, 115, 117, 120125, 128
129, 131
on indexicals, 115118, 131
John Perry and, 117119
Kaplan and, 115, 131
on reference dependence, 123127
on Russellian terms, 123, 127
on saying vs. showing, 122123
on senses, 119128
terminology, 116
theory of sense and reference for indexicals, 119122
thesis that names are Russellian, 126
on today and yesterday, 128130
on understanding demonstratives, 127
view of names, 123124, 126128
Exaggeration, 93
Existential quantifiers, 64, 67, 188
Existential statements, 7375
Extension, 135
character and, 110
context and, 110, 111
defined, 97
of Freges theory beyond singular
terms, 2529
two types of dependence of, 106
Extensional contexts, 177
Extensional languages, 177


Extensional sentences, 177

Extensional theory, Tarskis, 177
Externalism, 133. See also Semantic
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, 160161, 179
Frege, Gottlob, 12. See also Kripke,
Saul: critique of Frege
description theory of names and, 35,
36, 39, 50, 116, 117, 119
on ideas (in peoples minds) vs. sense
and reference (of words), 2023
on identity, 39
on indexicals, 108109, 115118
leveled system, 2223, 112
mock sense and, 124125
on mode of presentation, 1011
On Sense and Reference, 2, 3, 10,
24, 33, 35, 111
on opaque contexts, 3233
on ordinary and extraordinary use,
on proper names, 1314, 16, 22, 27,
31, 35, 55, 56, 60, 77, 119120
on reference, 35, 823, 121
Russell and, 55, 127
on sense(s), 35, 816, 1923, 120,
121, 123
theory of definite descriptions, 61
theory of meaning, 112
theory of sense and reference
compared with Russells theory, 77
extension beyond singular terms,
problems with, 2324
theory of truth, 153, 154
on truth-value, 2532, 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, 191199
consequences and criticisms,
Homophony, 173175, 178
Hyperbole, 93
I, 105107, 110, 111, 116119, 133,
139, 143, 178
Ideal language, 24, 63, 65
Ideas, 2023
compositional, 168
as objects of reference, 22
vs. senses, 22
Identicals, indiscernibility of, 208
Identity, 39
law of, 67
as a relation between names, 8
theories about personal, 4445, 53
types of, 4
Identity statements, 35, 710, 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, 9094
Incomplete expressions, 31, 120
Indefinite descriptions, 6768
vs. definite descriptions, 5560, 65, 67
identity and, 6365
as quantifiers, 58
Indeterminacy, radical, 184


Indexical expression and definite descriptions, 102, 109, 111

Indexical I. See also I
indispensability of, 119
Indexicality, the point of, 118119
context dependence, 102, 106109,
111, 142
description theory of names and, 102,
116119, 127
Evans on, 115122, 131
Fregean theory of, 108109, 115118
Kaplan on, 100105, 115
nature of, 101102
possible worlds, meaning, and,
two principles of, 102105
Indirect perspective, 33
Indirect reference, 19
Indirect sense, 19, 3233
Indirect speech, 1920
Indiscernibility of identicals, law of, 208
Information, 130131
as an epistemic notion, 131
Intension, 109111, 133, 134, 140
defined, 97
extension and, 97100, 106, 110, 111,
133, 134, 140
Intensional contexts, 177
Intensional languages, 177
Intensional operators, 176, 177
Intension-based semantic theory, 111
Intention, 8082, 197202
Intentional operators, 160
Internalism, 133
radical, 184
substitutional, 64
Irony, 93
Kaplan, David, 97
on character and content, 107108,
131, 140

on context of use and conditions of

evaluation, 104109
and Fregean model, 101
on indexicals, 100105, 115, 131, 133
and possible worlds and meaning,
on intension and extension, 98100,
return to Russellian semantics, 100
on rigid designators, 103
on rigidity and direct reference,
on today and yesterday, 113
Kripke, Saul, 3536, 99, 138
causal chain theory, 4849
critique of description theory of
names, 35, 36, 39, 52, 54, 102
objections to, 4952, 54
critique of Frege, 3942
epistemic objections, 4548
objections to, 4951
essential descriptions, 5253
impure descriptions, 5354
modal arguments, 4245, 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, 4245
social character of names, 5152
Language, philosophy of. See also specific topics
questions in, 1
Language acquisition. See Learning
Language of thought, 202
Learning, 910
Learning language, 171, 173
Levels, Freges system of, 2223, 112
Lexical ambiguity, 179. See also Semantic ambiguity
Lexical meaning, types of, 111


Linguistic deference, 51
Linguistic labor, universality of the division of, 140141
Logical form, 73, 154, 155, 178179,
Logically proper names, 6971, 77, 100
Material adequacy, 149150, 157, 159
Material biconditional, 189
Meaning, 108, 181, 191. See also Speaker
as compositional, 165, 168, 171
definite descriptions and, 17, 61, 70,
77, 95, 98, 126127, 143
definitions, 140
meanings are not in the head, 135142
merits of Tarskis theory as applied to,
as a social phenomenon, 141
theories of, 112, 181, 185, 188189
(see also Causal theory of meaning)
Freges theory, 33, 112, 115, 147
Kaplans 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
Russells theory, 61, 69, 112
theories of truth and, 147, 165, 168
173, 175, 185, 189
types of, 111, 193199
Meaning and Reference (Putnam),
Meaning-ascription, 167
Meaning (Grice), 191, 193. See also
Grice, H. Paul
Meinong, Alexius, 5963, 125
theory of definite descriptions, 6163
Meinongian ontology, Russells rejection of, 6567

Metaphors, 93
Mill, John Stuart, 50, 125
Mock sense, 124125
Mode of designation, 8, 1011, 80
Mode of presentation, 8, 226
of aspect, 1516
as an aspect of an object, 14
definite descriptions and, 38
Frege and, 1013, 1516, 2324, 33,
38, 120
identity statements and, 11, 12, 2324
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, 1516, 23, 33, 120
Names. See also Description theory of
names; Kripke, Saul
Evanss view of, 126128
genuine, 65, 69, 73, 86, 100
Russell on, 104
as Russellian, 126
social character of, 5152
Naming and Necessity (Kripke), 35, 36,
39, 45, 104. See also Kripke, Saul
Natural kinds, 141
Natural kind terms, 133134, 142
Natural language(s), 63, 160, 165, 202.
See also Davidson, Donald
applying Tarskis theory to, 175181
Natural vs. non-natural meaning,
Neale, Stephen
criticism of Donnellan, 9092
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


Non-natural meaning, 193195

Numerical identity, 4
Object language and metalanguage,
Object(s). See also Aspect of object and
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, 3233, 176177
Ordinary and extraordinary use, 1820
Paratactic theory, 177
Partial definitions, 157159, 162, 163
Peanos axioms, 5152
causal theory of, 47
and mode of presentation, 11, 13
Performatives, 177
Perry, John, 117119
Personal identity theories, 4445, 53
Perspective, 130131. See also Character
on a perspective, 33
Positivists, 186187
Possible worlds
analyticsynthetic distinction and,
40, 41 (see also Analyticsynthetic
content and, 108
idea of, 42
indexicals, meaning, and, 109112
intension, extension, and, 97100,
106, 109111, 134
ontology of, 174
possibly and, 188
rigid designation and, 4245, 99,
truth-value and, 97, 98, 108, 189

Possible world semantics, 97100,

Pragmatic meaning, 192
Pragmatics, 78, 84, 85, 8892. See also
Donnellan, Keith
Pragmatic theory of truth, 148, 149
Predication, 65, 68
Presentation. See Mode of presentation
Pretend-sense theory, 125126
Primary and secondary occurrences (of
descriptions), 7475
Proper knowledge, 89
Proper name(s), 47, 119120
ambiguity, 17, 35, 36, 71
aspects and, 1314, 16
as complete expressions, 31
context dependence and, 102
definite descriptions and, 13, 5355,
60, 62, 6971, 102, 119120
demonstratives and, 54, 71, 88
as descriptions, 54
as designating objects, 31
Frege on, 1314, 16, 22, 27, 31, 35, 55,
56, 60, 77, 119120
identity statements and, 71
Kripke and, 35, 36, 42
logically, 6971, 77, 100
Meinongs theory and, 60
multiple entities with the same, 17
ordinary names and, 13, 6970,
reference and, 1314, 1618, 22, 27,
55, 56, 61, 69, 77, 119, 135, 142
as rigid designators, 42
Russell on, 13, 55, 56, 58, 6062, 69
73, 77, 100
Russells criteria for, 69, 71
sense and, 13, 1618, 27, 35, 55
as singular terms, 55
Propositional function, 5760
contradictory, 207208 (see also Contradictions; Contradictory beliefs)


Propositions (cont.)
the proposition expressed vs. the
proposition meant, 9193 (see also
Implications and implicature)
singularity of (see Singular
Putnam, Hilary
meanings are not in the head,
on semantic externalism, 133146
Twin Earth thought experiment,
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, 5859, 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, 152155
Aristotle and, 151155
Reference, 12, 1618, 181, 187. See also
specific topics

definite descriptions and, 13, 14, 16,

17, 20, 48, 5152, 61, 67, 68, 79, 84
85, 91, 109, 117, 126127, 134, 143
levels of, 112
meaning as determining, 139140
proper names and, 1314, 1618, 22,
27, 55, 56, 61, 69, 77, 119, 135, 142
sense and, 1618, 2023, 129, 166
of a sentence, Frege on, 25
theories of, 15, 61, 119120, 187
truth-value and, 2532, 97, 106, 120,
Reference and Definite Descriptions
(Donnellan), 8485. See also Donnellan, Keith
Reference dependency, 123127
Referential theory of meaning, 61
Referential view of descriptions, 7884
Referring vs. denoting, 84
Regular use, 199
Relational connections and semantic
notions, 159160
Repeat creationism, doctrine of,
Representational entities, 15
Rigid descriptions, 45, 103, 104
Rigid designators
de facto, 104
de jure, 104
indexicals as, 102103
vs. non-rigid designators, 4244, 99,
103, 109110
Rigidity/rigid designation, 4245, 104
definite descriptions and, 4244, 99
vs. direct reference, 103106
possible worlds and, 4245, 99, 104
Russell, Bertrand
on definite descriptions, 5575, 104
demonstratives and, 7071, 88, 96
description theory of names and, 69,
95, 116
Donnellan and, 78, 8387
Donnellans critique of, 78, 9194


Frege and, 61, 116

on indefinite descriptions and identity, 6365
on indefinite descriptions as quantifiers, 5859
John Stuart Mill and, 50
Meinong and, 5963
Russells rejection of Meinongs ontology, 6567
on names, 100, 104105 (see also under
Proper name(s))
on primary and secondary occurrences, 7475
Principia Mathematica, 63
problems with, 7274
on propositional function and instance, 5759
reference dependency and, 123
referential theory of meaning, 61, 69
theories of definite descriptions, 6063
(see also Definite descriptions)
theory of descriptions, 6772, 116
compared with Freges theory, 77
objections to, 9496
problems with, 7274
theory of meaning, 112
truth-value and, 7172, 77, 78, 94, 95
Russellian terms, 123
as Fregean, 127
that differ in their sense, 127
Satisfaction, 159163
Satisfaction axioms, 162163, 172, 176,
179181, 183
Saying vs. showing, 122123
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, 1617, 3536, 65,
74, 75, 85, 89, 92, 93, 117

Semantic compositionality, types of,

Semantic conception of truth, 160
Semantic Conception of Truth, The
(Tarski), 147. See also Tarski, Alfred
Semantic denoting, 8485
Semantic externalism, 142
Putnam on, 133146
Semantic meaning, 192, 195. See also
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, 1416, 73
authentic non-nonsense sense and
specious phony sense, 124125
conception of, 1216
definite descriptions and, 13, 14, 17,
38, 117, 125
definition and meaning of the term,
Evans on, 119128
Freges conception of, 1216 (see also
under Frege, Gottlob)
Freges introduction of the term/concept, 20
as a label, 12
proper names and, 13, 1618, 27, 35,
as the route to reference, 166 (see also
Reference: sense and)
Sense-specifying reference assignment,
Sentence connectives, 89, 155, 161
Sentence meaning. See Semantic
Sentences and propositions, 23
Sentential functions, 161
Showing vs. saying, 122123


Simple object theory (identity statements), 9

Singular propositions, 100, 102105,
defined, 100
Singular terms, 30, 31, 33, 120, 154
155, 176, 178179
context-dependent, 179
coreferential, 29 (see also Coreference)
definite descriptions as, 55, 6162
extension of Freges theory beyond,
Kaplan and, 100, 104
nature of, 25, 55, 126
Russell and, 55, 6162, 68, 90
Skepticism, 94
Speaker meaning, 93, 191193, 195. See
also Meaning
nature of, 192, 195199
Speakers and sentences, 191193
S-sentences, 162, 163
Statements, 3. See also Sentences and
Stimulus meaning, 183184
Strawson, P. F.
critique of Russell, 94
on demonstratives, 88
Donnellan and, 82, 8586, 90
On Referring, 77
referential view of descriptions,
truth-value and, 72, 77, 85 86, 94
on truth-value gaps, 8586
Subjective ideas, 22
Subjective View: Secondary Qualities and
Indexical Thoughts, The (McGinn),
116, 118119
Subsentential expressions, 25, 32
Substitutional interpretation, 64
Synonymy, 16, 121
Syntactic ambiguity, 85, 179180. See
also Semantic ambiguity
Syntactic theory, 169

Tarski, Alfred
criteria of acceptability, 149151 (see
also Truth theory(ies): criteria of
object language, metalanguage, and,
theory of truth, 147163, 165
Tarskian biconditionals, 152156. See
also Biconditionals; T-sentences
Tautology, 6, 9, 71, 73. See also Analyticsynthetic distinction
Temporality. See Today and
Testimony, 200
The, 55, 68. See also Definite
Thought, language of, 202
Thought, The (Frege), 111
Today and yesterday, Evans on,
Translation, 165, 166
radical, 183
Translation manual, 169, 170
Transparency condition, 197
Truth, definitions of, 147, 149151, 153,
155161, 163, 165. See also Partial
definitions; Truth theory(ies)
Truth biconditionals. See Tarskian
Truth conditions, theory of, 168. See
also Truth theory(ies)
Truth function, 120
Truth theory(ies), 148149, 158. See
also Redundancy theory of truth;
Semantic conception of truth; Tarski,
Alfred: theory of truth; Truth, definitions of
criteria of acceptability, 149151, 175,
empirical, 181185
Truth-value, 75, 120, 166, 177
character, context, content, and, 105
106, 108, 110111


concepts and, 120

contingent sentences and, 97
Davidson and, 189
Donnellan and, 78 (see also Truthvalue gaps)
Frege on, 2532, 78, 97, 120
intensions, extensions, and, 9799
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, 2532, 97, 106, 120,
Russell and, 7172, 77, 78, 94, 95
sense and, 166
of a sentence, 2526
Strawson and, 72, 77, 85, 86, 94
Truth-value gaps, 77, 8587
T-sentences, 156157, 161163, 174,
178, 180, 182. See also Biconditionals
how to derive, 157159
Twin Earth (thought experiment) and
water, 134145
Uniqueness, 69
Use-mention distinction, 8
Verification, 181183, 185187
Whitehead, Alfred North, 63
Wittgenstein, Ludwig, 24, 63, 186
on identity statements, 24
on saying vs. showing, 122
Tractatus Logico-Philosophicus, 63, 167,
Words. See also specific topics
having a proper scientific sense, 124
ordinary and extraordinary use of,
terms that introduce concepts vs. objects, 100