1
SELF-KNOWLEDGE AND THE SENSE OF “I”
José Luis Bermúdez
In Self-Knowledge, edited by Anthony Hatzimoisis (Oxford University Press, 2011), 226-246
What does an understanding of the first person pronoun “I” contribute to the understanding of a
sentence involving “I”? To answer this question is, in the terminology I shall be employing, to
give an account of the sense of “I”. In posing the question thus I take myself to be faithful to
Frege’s original motivations for introducing the notion of sense, so that the question I am trying
to answer with respect to “I” is precisely the question that Frege was trying to answer with
respect to mathematical expressions and, derivatively, with respect to words and sentences in
non-mathematical language. But nothing that I say depends upon the exegesis of Frege.
Thinking about the sense of “I” reveals uncharted territory in current thinking about indexical
expressions, which is quite extraordinary given the amount of attention that they have received in
recent years. Philosophers of language have addressed important questions about how the
reference of indexicals is fixed; the evaluation of sentences involving indexicals in
counterfactual circumstances; the cognitive significance of indexicals; and the relation between
sentences involving indexicals and sentences without indexicals. But, as we shall see, there
remain questions about what it is to understand sentences involving “I” that none of the accounts
of indexicals currently available can easily answer.
It is natural to think that there will be close connections between theoretical accounts of selfknowledge, on the one hand, and what it is to understand the first person pronoun, on the other.
The canonical types of self-knowledge are typically expressed in sentences that use the first
person pronoun, and many have thought that the distinctiveness of this form of linguistic
expression is key to understanding what is distinctive about the knowledge it expresses.
Conversely, one might think that only thinkers capable of a distinctive type of self-knowledge
(only thinkers properly described as self-conscious) can master the first person pronoun. If this is
right, then understanding the characteristic forms of self-knowledge that make a thinker selfconscious is likely to be a central component in understanding linguistic mastery of the first
person pronoun.
On the other hand, however, the first person pronoun is typically used as a tool of
communication. We need to think not just about what it is to use the first person pronoun with
understanding, but also what it is to understand someone else’s use of the first person pronoun.
This leads to reflection on the relation between “I” and “you”. As we shall see, an extremely
plausible principle governing linguistic understanding via the conditions of adequacy upon
reporting speech entails what I call the symmetry constraint, which is that an account of the sense
of “I” must leave open the possibility of sentences involving “I” saying the same thing as
sentences involving “you”. One immediate consequence of the symmetry constraint is that we
need to make a distinction between token-sense and type-sense. The rationale for this distinction
is particularly clear in the case of “I” and other indexicals. But it can plausibly be extended to
other referring expressions. The symmetry constraint and the distinction between token-sense
and type-sense are discussed in sections 1 and 2.
The most interesting and surprising consequences of the symmetry constraint have to do,
however, with how we understand the forms of self-knowledge implicated in understanding the
sense of “I”. A number of authors, most prominently Gareth Evans in his influential book The
Varieties of Reference, have proposed that the sense of “I” needs to be elucidated, at least in part,
in terms of the speaker’s sensitivity to forms of self-specifying information that have the
2
property of being immune to error through misidentification relative to the first person pronoun.
The proposal has two motivations. The first is the general Fregean principle that the sense of a
referring expression is a mode of presentation of the expression’s referent. The second is the
claim, underwritten by the idea that the linguistic device “I” is expressive of self-conscious
thought, that the sense of “I” exploits ways of being presented to oneself that are distinctively
first personal. As will emerge in section 3, however, this way of thinking about the sense of “I”
is not compatible with the symmetry constraint, since it makes the sense of “I” private and
unshareable. This leads in section 4 to an alternative proposal for doing justice to the insight that
the sense of “I” reflects a distinctive way of being presented to oneself. On this proposal, the
distinctive way in which I am presented to myself can be systematically related to the distinctive
way in which I am presented to you. What matters is not that I am in receipt of distinctive types
of information about myself, but rather that I have a distinctive ability to locate myself in
objective space. On a plausible way of thinking about the individuation of cognitive abilities, this
is an ability that you also can have and that underwrites both your understanding of “you” when
you address me, and your understanding of “I” as uttered by me.
Introducing the symmetry constraint
A good place to start in thinking about linguistic understanding is with the basic principle,
recently stressed to good effect by Mark Sainsbury (Sainsbury 1998), that in understanding a
sentence I acquire knowledge of what is said by that sentence. Of course, this is just one of the
things that I can come to know through understanding a sentence. Other things that I can come to
know include what the speaker intends to communicate, or what result it is likely to have (an
inflammatory effect on the assembled crowd, for example). In each of these cases the knowledge
that I acquire is typically knowledge that I can express. In the central case, where what I express
is my knowledge of what is said, then I am, quite simply, reporting what is said. In the other,
more complicated cases, what I am expressing goes beyond a report of what is said. It might be a
report of communicative intent, or of perlocutionary effect.
There is a crucial difference between the central case and the others. As Sainsbury notes, if we
confine ourselves to a language that does not contain indexicals, then when I report what is said I
can typically do so using exactly the same words used in the original utterance (taking
“utterance” in a broad sense that includes written inscriptions, and so on). Suppose, for example,
that you say: “Harold is being characteristically damning”. I would be inclined to give a
homophonic report of what you said, where a homophonic report is one that uses exactly the
same words after the “that –” clause as in the utterance being reported (modulo cosmetic changes
in tense, and so on). This is plainly not the case, however, when what is reported is
perlocutionary effect. I might report the perlocutionary effect of your utterance as being that the
meeting rejects Harold’s assessment of the situation. Homophonic reports of perlucutionary
effect are rarely appropriate. Reports of communicative intent fall somewhere in the middle.
They can be homophonic, but the interesting cases tend to be the non-homophonic ones. This is
because the cases where we can give a homophonic report of communicative intent are those
where what the speaker intends to communicate is precisely what their utterance says, and here
the notion of communicative intent drops out of the picture.
Of course, as soon as indexicality enters the picture, homophonic reports of what is said can no
longer be expected in every case. Nonetheless, the bedrock status of homophonic reports in the
non-indexical case points to an interesting and important feature of the central cases of reported
speech, one that does extend to the indexical case. When a report is accurate, what is said by the
words in which the report is couched can be mapped straightforwardly onto what is said by the
words in the utterance being reported. The accuracy of a report is a function of the extent to
which what it says (or rather: what is said by the elements of the report that report what is said,
rather than who said it and when) is what was said by the utterance being reported. At the ideal
3
limit, accurate reports stand in the same-saying relation to what they report (or at least: are able
so to stand after cosmetic adjustments).
Suppose, then, that you say to me: “Your position is either trivial or false”. On being asked by a
third party what you said to me I reply: “She said that my position is either trivial or false”. This
report is accurate. If the third party is aware of the background to the exchange, knows whom the
indexicals pick out, and so on, then the knowledge that she gains through hearing my report of
your utterance is precisely the knowledge that she would have acquired had she heard your
original utterance. Were I to utter as a self-standing sentence the words following the “that –”,
namely “My position is either trivial or false”, then I would make myself a samesayer with you,
at least relative to that particular context of utterance. This is entirely intuitive, given the logical
relation between your saying “Your position is either trivial or false” and my saying “My
position is neither trivial nor false”. Plainly, what I am doing is denying what you are saying – in
which case removing the negation sign makes us same-sayers.
This has interesting consequences for how we think about what is said by the indexicals in the
respective utterances. If we are same-sayers then my token-utterance of “I” must stand in the
same-saying relation to your token-utterance of “you”. So, what we both understand when I say
“I” (in that particular conversational context) must be the same as what we both understand when
you say “you” in that context. And, since we are taking the notion of sense to be correlative to
the notion of understanding, this means that the sense of “I” in that context must be the same as
the sense of “you” in that context.
The point generalizes to give a basic constraint upon any satisfactory account of the sense of “I”.
What I will term the symmetry constraint requires that we preserve the token-equivalence of “I”
and other personal pronouns with respect to same-saying and relative to a particular context. So,
for example, any satisfactory account of the sense of “I” must allow a speaker to use the first
person pronoun to contradict what another speaker says in the same conversational context using
the second person pronoun. And it must make it possible for me to use “you” to report what you
say using “I” (where the report and the original utterance fall in a single conversational context).
It is easiest to appreciate the symmetry constraint in the case of “I” and “you”, but the constraint
extends to other personal pronouns, most obviously “he” in the singular case, and “we”, “you”,
and “they” in the plural case. Each of these personal pronouns can, in suitable contexts, be tokenequivalent to another personal pronoun of the same number.
The qualification of context-dependence is important. It is not in general true, for example, that
an accurate report of a speaker’s indexical utterance can generate a single sentence that says the
same thing as that indexical utterance. Sainsbury elegantly describes the characteristic feature of
the non-indexical case as follows: “Within the simple picture of a single language and no
indexicality, one who gave an adequate report could make himself a samesayer with the original
speaker merely by uttering the words which follow “said that” in his report” (1998, 139). As he
himself illustrates, nothing like this holds generally once indexicals are brought into the picture.
Accurately reporting indexical sentences involves setting the scene in ways that often prevent
detaching the sentence that comes after the “that–” to give an utterance that says the same as the
utterance being reported. To give a simplified version of an example that Sainsbury discusses at
some length, we might report Gareth Evans’s utterance on July 4 1968 of “today is July 4” by
saying something along the lines of: “On 4 July 1968, Gareth Evans said that it was July 4”. This
report captures the indexicality of the original utterance, but there is no prospect of detaching the
sentence following the “that–” to give an utterance that says the same as the original utterance.
We cannot detach the sentence that gives the content of the sentence being reported because it is
anaphorically dependent upon the preliminary scene-setting material. For an example involving
“I” consider how we might report Dr Lauben’s utterance on 1 January 1900 of the sentence “I
am hungry”. An accurate report would be “On January 1 1900, intentionally speaking of himself,
Dr Lauben said that he was hungry”. Again, anaphoric dependence precludes detachability.
4
The token-equivalence that the symmetry constraint is intended to preserve applies primarily in
cases where fully explicit scene-setting is not required, because the reporter and the person to
whom the report is addressed already possess the relevant information. Suppose that X says to
me, on 1/1/2007: “My position is neither trivial nor false”. I might report this at a later date to
someone who was not present by saying something along these lines: “On 1/1/2007 X told me
that her position was neither trivial nor false”. We have the expected anaphoric dependence and
failure of detachability. If you and I both know that we are talking about a conversation between
X and me on 1/1/2007, however, then there is no need for scene-setting. I might simply say, “She
said that her position is neither trivial nor false”. The second indexical pronoun is not
anaphorically dependent upon the first. Rather, the very same contextual facts and shared
information that fix the reference of the first pronoun fix the reference of the second. So, I can
detach the sentence “her position is neither trivial nor false” in a way that makes me a samesayer with X’s original utterance. This is exactly the sort of case to which the symmetry
constraint applies. The account we give of how an understanding of X’s use of “my” contributes
to what is understood by X’s utterance of “My position is neither trivial nor false” needs to allow
me to be a same-sayer by uttering “His position is neither trivial nor false in the relevant context.
Reflection on how we report utterances is not the only motivation for the symmetry constraint.
There is also an epistemological motivation. Just as we need to think about how utterances are
reported as expressions of knowledge, so too should we think about communication as a way in
which knowledge is transmitted.1 If I ask you “Where am I?” and you reply “You are in
Copenhagen, near the Tivoli Gardens” you have transmitted to me your knowledge of our mutual
location in a way that, in the normal course of things, makes me come to know that I am in
Copenhagen, near the Tivoli Gardens. Why should this be? The epistemology of testimony raises
deep issues (Jack, 1993; McDowell, 1993) and any full account of how warrant is transmitted
through testimony is likely to be very complex. But it is natural to think that a core component of
any such account will be the thesis that what you say is what I come to know – and this, once
again, requires the token-equivalence of “I” and “you” in the relevant context. Suppose that there
is no such token-equivalence. Then, assuming a fairly standard way of thinking about sameness
and difference of what is said, my acceptance of what you say would be rationally cotenable with
denying that I am in Copenhagen. It would be hard to square this with the thesis that, simply in
virtue of hearing you say that I am in Copenhagen, I come to know that I am in Copenhagen. If it
is rationally permissible for me to understand your utterance of “You are in Copenhagen, near
the Tivoli Gardens” while at the same time denying that I am in Copenhagent, near the Tivoli
Gardens, then my understanding your utterance can hardly give me knowledge. But what more
could be required for knowledge in this case, besides understanding what you have said?
The distinction between type-sense and token-sense
The symmetry constraint has clear implications for how we think about the sense of “I”. It holds
that what I understand by your utterance of “you” in a given context is what you understand by a
similarly embedded utterance of “I” in that same context. So, to return to the terminology with
which we began, it must be possible for the sense of a token utterance of “I” to be the same as
the sense of a suitably related token utterance of “you”. An immediate consequence of this is that
we need to make a distinction between what might be termed the token-sense of “I” and the typesense of “I”. The token-sense of “I” is what a speaker or hearer understands when they utter or
hear an utterance involving “I”. The type-sense of “I”, on the other hand, is that in virtue of
which a speaker or hearer can properly be said to understand the expression “I”.
1
Compare Evans: “It is a fundamental, though insufficiently recognized point, that communication is essentially a
mode of the transmission of knowledge” (1982: 310).
5
The distinction between token-sense and type-sense may have a more general application outside
the realm of indexicals – to proper names, for example. It might be that we only need a very
minimal account of what it is to understand the expression “ψ” where “ψ” is a proper name.
Perhaps it is enough simply to understand that “ψ” has the syntactic role of a proper name,
together with mastery of a disquotational principle to the effect that “ψ” names ψ.2 It seems right
to say that I can understand an utterance of the sentence “ψ is F”, even when I know nothing
about ψ except that he or she has a name and is being said to be F. I would have no difficulty in
continuing the conversation, for example, or reporting what has been said to a third party. On the
other hand, there are reasons to think that such understanding would be very partial. It would
give me very little grip on the sentence’s semantic value. If I wanted to put myself in a position
to determine whether the sentence is true or false, or even to know what would count as evidence
for its being true or false, then I have to find out something more about ψ. It may be helpful to
think of this additional element in terms of the token-sense of the proper name “ψ”, while the
type-sense of “ψ” is what provides the minimal understanding of how “ψ” functions in English.
Be that as it may, the distinction between type-sense and token-sense is plainly called for in the
case of indexicals. Understanding how the expression “I” functions in English is very different
from understanding how the expression “you” functions, even though (as we have seen) the
token-sense of “I” can be identical to the token-sense of “you” in appropriate contexts. The same
holds for other pairs of indexicals such as “here” and “there”; “now” and “then”; or “this” and
“that”. In suitable contexts I can use “there” or “then” to deny what you say using “here” or
“now”. Consider the sentence:
(1) In the basement on Monday morning she said, “It’s cold here now”, but I happen to
know that it was warm there then.
Here I express my understanding of what was said, making explicit enough of the context to
anchor the indexicals, and then deny what I have just reported. The words that I use in the denial
share token-sense but not type-sense with the words in oratione recta that I am denying.
So what is the relation between type-sense and token-sense? Here is one answer. The type-sense
of an indexical expression is something like its character, as discussed by Kaplan (Kaplan,
1977). That is, the type-sense is a rule that tells one in general how the reference of the indexical
is fixed in a given context. The character of “I”, roughly speaking, is that a token of “I” refers to
the utterer of that token. If someone knows the type-sense of “I”, together with enough details of
the context to enable the reference-fixing rule to be applied, then she is in a position to determine
the reference of the token utterance of “I”. Fixing the reference of the token utterance of “I”
determines the content of that utterance. We could, therefore, simply identify type-sense with
what Kaplan calls character and token-sense with what he terms content.
One advantage of this proposal is that it plainly satisfies the symmetry constraint. Relative to a
particular context, the content fixed by my use of “I” is exactly the same as the content fixed by
your use of “you”. Both utterances share a single truth condition and, for Kaplan, that truth
condition is their content. A comparable equation of token-sense and content would have the
same desirable consequence for the other pairs of indexicals we have discussed.
But the notion of token-sense is suppose to be correlative with the notion of understanding, and
the notion of content (in Kaplan’s sense) is too coarse-grained to do the job required. Suppose
that “I” corefers with the proper name “ψ”. Then, in the appropriate context, there is no
difference in content between the sentences “I am F” and “ψ is F”. But there is generally an
important gap between understanding these two sentences. This is the indisputable upshot of
discussions of essential indexicals (Perry, 1979), reflected in the twin facts (a) that “ψ is F”
2
An early version of this minimal view is suggested in Sainsbury 1979.
6
cannot generally be embedded in reports of “I am F” and (b) that “ψ is not F” is not in general
immediately understood as a denial of “I am F”, even holding the context constant. It is not hard
to imagine contexts in which I am addressing Y but where I am unaware that his name is “Y”,
and so fail to take a third party’s utterance of “Y is not F” to be the denial of Y’s own assertion
“I am F”. But this difference between understanding “I am F” and understanding “ψ is F” is
precisely what the notion of token-sense is being called upon to explain.
Perhaps, then, we went wrong in locating token-sense at the level of content? Might we find
token-sense at the level of character? Something like this seems to be suggested by Kaplan, when
he proposes, in section XVII of “Demonstratives”, that the cognitive significance of an utterance
involving “I” be understood in terms of the character of “I”. Alluding to a familiar passage from
Frege’s “The thought: A logical inquiry” Kaplan writes:
What is the particular and primitive way in which Dr Lauben is presented to himself?
What cognitive content presents Dr Lauben to himself, but presents him to nobody else?
Thought determined this way can be grasped by Dr Lauben, but no one else can grasp
that thought determined in that way. The answer, I believe, is, simply, that Dr Lauben is
presented to himself under the character of “I”. (Kaplan 1977, 533)
One problem with this suggestion is that the character of “I” is completely different from the
character of “you”, whereas the symmetry constraint demands that the token-sense of “I” be
equivalent to the token-sense of “you”, modulo a given context. This point, were it the only
difficulty, could be finessed, however. After all, it is fairly easy to map the character of “I” onto
the character of “you” in a way that might be thought sufficient for the symmetry constraint. If I
am addressing you then, in virtue of giving a rule for determining the referent of “I” as the
utterer of the relevant token of “I”, the character of “I” to all intents and purposes fixes the
referent of “you” (as the person whom this person, the utterer of “I”, is addressing).
More serious is the problem that the notion of character just seems too general. One thing that we
want the notion of token-sense to do is distinguish the way in which I understand what I read
when I stumble across a piece of paper in the street on which all that remains legible is the first
line of a will (“I write this document in full possession of my faculties. . . ”) from the way in
which I understand those very same words when I am present on an occasion when the will is
being read. Yet, from the point of view of character, there is no relevant difference here. Both the
inscription and the utterance of “I write this document. . . ” exploit the reference-fixing rule
given by the character of “I” and I apply that rule in exactly the same way when I read the piece
of paper by an unknown author as when I hear the lawyer read out the words written by a
familiar hand.
The distinction here has to do with different ways in which we can understand the truth
conditions of an utterance. It is agreed on (almost) all sides that we understand an utterance by
grasping its truth condition (by understanding how the world would have to be for that utterance
to be true). But I can understand the truth condition of an utterance in a way that does not put me
in any position to know how to go about establishing whether or not that condition holds. This is
exactly what is happening in the first case. I know, in what one might think of as an abstract
sense, what it would be for the inscription to be true. It is true just if the writer was in full
possession of his faculties at the time of writing. But, not knowing who the writer is or how to go
about identifying him, let alone the time of utterance, I am not in a position to make any progress
in determining the utterance’s truth value. This level of understanding contrasts with what one
might think of as a practical grasp of an utterance’s truth condition, where a practical grasp is
one that that put one in a position to determine the utterance’s truth value.
This idea that there are different levels in understanding truth conditions, particularly for
utterances involving indexicals, has been taken up and developed by John Perry. Perry has
7
proposed for indexical expressions a distinction between different types of content that is
intended to be more fine-grained than Kaplan’s distinction between character and content (Perry,
1997). Whereas Kaplan simply distinguishes content from character, Perry distinguishes between
what he calls content-M (more or less corresponding to Kaplan’s character) and two additional
notions of content, content-C and content-D, that between them do the work of Kaplan’s content.
Strikingly, he supports his three-way taxonomy with considerations from same-saying not unlike
those raised earlier in this paper. Perhaps this is the place to look for an elucidation of tokensense?
The twin ideas behind Perry’s taxonomy are, first, that content is given in terms of truth
conditions and, second, that truth conditions are assessed relative to a background of facts taken
as fixed. The three types of content differ in what background facts are taken as fixed. In effect,
we can view them as becoming increasingly fine-grained and determinate. The most general type
of content is content-M, which corresponds to the truth-conditions of the utterance given the
basic syntactic and semantic facts about the utterance. So, for example, the truth condition in
content-M terms of my utterance “I am now in the best university in Denmark” is roughly
speaking that the utterer of the token sentence is, at the time of uttering it, in the best university
in Denmark. The next most determinate type of content is content-C, corresponding to the truthconditions relative to the relevant contextual information required to fix the reference of the
indexical expressions. The content-C truth conditions are that JLB is in the best university in
Denmark on the morning of Thursday 21 September, 2006. Finally, the content-D of an utterance
corresponds to the truth-conditions relative to all of the above information, in addition to
whatever information is required to fix the reference of other expressions. For the content-D of
my utterance we merely need to identify a particular university as the best university in
Copenhagen.
Perry’s taxonomy can be characterized in the following way. The content-M of a typical
indexical utterance of the form “I am F”, where F includes a descriptive component, is a singular
proposition about the utterance itself, while content-C and content-D are both singular
propositions whose subject is the referent of “I”. In the case of content C the singular proposition
retains an element whose referent remains unspecified. The omission is remedied in content-D.
According to Perry, the cognitively relevant content of an indexical utterance is content-M, while
Content-D is a complete characterization of the utterance of the truth-condition. What he terms
the “official” content is given by content-C, where the official content of an utterance is what it
says. One of Perry’s two arguments for locating official content at the level of content-C is that it
allows for utterances of “I am F” and “you are F” to have the same content in an appropriate
context. In so doing, of course, the notion of content-C satisfies the symmetry constraint. But the
problem, as with Kaplan’s notion of content, is that the price is too high. The C-content common
to “I am F”and “you are F” is also shared with “ψ is F”, where “ψ” corefers with “I” and “you”
in the relevant context. There certainly is, according to Perry, a level of content at which “I am
F” and “ψ is F” have different contents. This is the level of content-M. The problem, however, is
that, at the level at which “I am F” and “ψ is F” have different contents, “I am F” and “you are
F” also have different contents (even in the type of contexts we have been discussing).
Moreover, content-M simply gives what we earlier characterized as an abstract grasp of a truthcondition, whereas for the notion of token-sense we need a practical grasp that would put us in a
position to determine (or to know how to go about determining) the utterance’s truth value.
The upshot, then, is that the distinction between token-sense and type-sense, although well
motivated by considerations from what is involved in understanding utterances involving
indexicals, does not map onto any of the distinctions made in the most prominent theories of
indexicals.
8
Token-sense and indexical thoughts
Putting the discussion in the previous two sections together gives a fairly specific “job
description” for an account of the token-sense of “I”. First, the token-sense of an indexical must
obey the symmetry constraint where appropriate. That is, our account of the token-sense of
indexicals must make clear why and how pairs of indexicals such as “I” and “you”, “I and he”,
“now” and “here”, or “this” and “that” can be token-equivalent relative to a context. Second, our
account of token-sense must be more fine-grained than the way of thinking about truth
conditions reflected in Kaplan’s notion of character or Perry’s content-C or content-D. “I” and
“you” can be token-equivalent in a particular context, even though neither is token-equivalent to
a co-referring proper name. Third, our account must be more closely tied to the thinker’s
knowledge of how to go about determining an utterance’s semantic value than is yielded by the
notion of character.
My project for the remainder of this paper is to think through some of the implications of this job
description. Let me begin by considering a natural way of thinking about the token-sense of “I” –
natural because it is an obvious way of developing some ideas about the relation between
indexicals and behavior that have become almost platitudinous. According to a venerable
philosophical tradition, which goes back at least to Kant, there is a particularly tight connection
between self-reference and self-knowledge – between the machinery we use to refer to ourselves
in speech and the knowledge of ourselves that this ability to refer to ourselves allows us to
express. In recent philosophy, thanks to the work of Castaneda, Perry and others, we have
become familiar with the idea that there are certain forms of self-knowledge that can only be
linguistically expressed using the first person pronoun. These forms of self-knowledge are
essentially indexical, to use the standard phrase. If the capacity for self-reference really is
integral to the expression of these types of self-knowledge, then it is reasonable to think that we
might work backwards from essentially indexical self-knowledge to an understanding of the
linguistic device of self-reference.
The distinction between type-sense and token-sense is very relevant here. It is most implausible
that we need bring in essentially indexical self-knowledge to elucidate the type-sense of “I”,
since grasping the type-sense of “I” is simply a matter of understanding how the expression “I”
functions in English. But the token-sense of “I” is context-sensitive and occasion–sensitive. We
do not grasp the token-sense of “I” tout court. What we grasp is the token-sense of “I” on a
particular occasion and in a particular context – specifically, what “I” contributes to the sense of
the sentence as used on that occasion and in that context. Now, what “I” contributes to the sense
of a sentence is what an understanding of “I” contributes to the thought that the sentence
expresses and this contribution is going to be most interesting and distinctive in precisely the
cases where the thought expressed is essentially indexical. And so, one might think, we can
elucidate what it is to understand utterances involving “I” through understanding the indexical
thoughts that can only be expressed by means of such utterances.
I use the expression “indexical thought”, without prejudice, for those thoughts, whatever they
might turn out to involve, that have the familiar and distinctive cognitive dynamics discussed by
Perry. To cut a long story short (see Ch. 1 of Bermúdez 1998 and Bermúdez 2003 for less short
versions), the characteristic cognitive dynamics of an important class of indexical thoughts is
closely connected to the fact that those thoughts are based on sources of information that can
only provide information about oneself. So, for example, if I am driving and, noticing stopped
traffic looming in front of me, suddenly realize: “I’m going to crash if I don’t slow down”, there
is no question but that it is me who is going to crash. This, of course, is why the need to take
evasive action is so immediately compelling, as soon as I notice the looming traffic.
The defining feature of this class of indexical thoughts is that they have the property, originally
named by Shoemaker, of being immune to error through misidentification (IEM) relative to the
first person pronoun (Shoemaker 1967). Judgments have this property relative to the particular
9
sources of information on which they are based. A judgment “∂ is ψ”, exploiting pronoun ∂ and
made on the basis of information of type F, has the IEM property just if that judgment is not
susceptible to a particular type of error – the error of being mistaken about whom or what it is
that is ψ. If the judgment turns out to be false because the relevant object is not ψ, then there is
no possibility of retreating to a weaker claim to the effect that someone is ψ (compare Wright
1998 at p.19). In the case of judgments expressed using the first-person pronoun and based on
sources of information that are IEM there is no gap between learning that someone is F and that
one is oneself F. The sources of information that warrant the claim that ψ-ness is instantiated can
only give me information that ψ-ness is instantiated in me. So, for example, if I come to believe
through introspection that I am having a certain thought, I may well be mistaken, but I cannot be
mistaken about whom it is whom I am taking to be having that thought. Likewise, my field of
view is structured in a way that offers information about my spatial relations to perceived objects
and about the route I am taking through the perceived environment (Bermudez, 2002, 2003).
This information can be inaccurate but it cannot (in the normal course of events) be information
about anyone else’s spatial location or trajectory.
But not all indexical thoughts have the IEM property. Testimony is quite plainly not a source of
information that generates judgments with the IEM property. And yet testimony-based
judgments can qualify as indexical thoughts in virtue of having the characteristic cognitive
dynamics of indexical thoughts. So, for example, if I overhear someone saying “Your cart is
trailing sugar” and think that they are addressing me, I may perfectly reasonably come to the
conclusion that I am the shopper trailing sugar. The ensuing judgment would have the
characteristic functional role of an indexical thought, but would certainly be susceptible to error
through misidentification. So why, one might ask, should we take the IEM property to be the key
to explaining indexical thoughts?
The answer is that judgments with the IEM property are epistemologically prior to judgments
that are susceptible to error through misidentification. The basic point is made by Gareth Evans
in The Varieties of Reference. Any judgment of the form “I am F” that does not have the IEM
property can be analyzed in terms of a pair of judgments “I am ∂” and “∂ is F”, where ∂ is some
way of picking out or identifying a person. On pain of regress, the judgment “I am ∂” must either
itself be IEM or analyzable at some point into a pair of judgments one of which has the IEM
property. The notion of analysis here can be explicated as follows: A judgment ϕ can be
analyzed in terms of two judgments ϕ* and ϕ** just if the warrant for the judgment that ϕ
depends upon the warrant that she has for the judgments ϕ* and ϕ**.
We can see how this works for the example just given. If I overhear someone saying “Your cart
is trailing sugar” and as a result judge that I am the shopper trailing sugar, then my warrant for
that judgment stands or falls with my warrant for the pair of judgments “I am the person being
addressed” and “the person being addressed is the shopper trailing sugar”. The first judgment
does not have the IEM property. My coming to realize that I am not the person being addressed
would be perfectly compatible with my judging that nonetheless someone is being addressed.
Yet that judgment can be analyzed into a further pair of judgments. Perhaps: “I hear the sentence
‘You are the shopper trailing sugar’” and “The hearer of the sentence ‘You are the shopper
trailing sugar’ is the shopper trailing sugar”. The first of these certainly has the IEM property,
since it is derived through perceptual channels that can only inform me about what I am hearing.
(If I come to learn that I didn’t hear the sentence “You are the shopper trailing sugar”, then I
would have no grounds for shifting to the judgment that someone heard the sentence “You are
the shopper trailing sugar”.) And it is because of this that the original judgment that I am the
shopper trailing sugar has the characteristic cognitive dynamics of an indexical thought.
We have two ideas in play. The first is that we can characterize the sense of “I” through
understanding those indexical thoughts that can only be expressed using the first-person
pronoun. The second is that the IEM property is fundamental to the cognitive dynamics of
10
indexical thoughts. Putting these two ideas together gives us one way of glossing Frege’s well
known suggestion that “everyone is presented to himself in a special and primitive way in which
he is presented to no one else” (Frege, 1918). The connection comes swiftly, given the tendency
of Frege and many Fregeans to think of the sense of a referring expression as a mode of
presentation of an object, where each mode of presentation “presents” a unique object. The
information channels that yield information that is immune to error through misidentification
certainly present each individual to herself in a way that she is presented to noone else, and so
one might think that a crucial element in any explanation of the sense of “I” will be suitable
sensitivity to information of the appropriate type.
The clearest example of this approach is Gareth Evans’s account of the sense of “I” in The
Varieties of Reference (Evans, 1982). Evans applies to the first person pronoun the general
model that he develops for all referring expressions on the basis of his distinctive reading of
Frege.3 The sense of a referring expression, for Evans, is a way of thinking about the object that
the expression picks out, where this requires knowing which object one is thinking about. Evans
interprets this requirement in a very strong way, so that we can only think about objects of which
we have discriminating knowledge. Different types of referring expression are distinguished in
terms of different forms that this discriminating knowledge can take. In the case of “I”, the
discriminating knowledge is, of course, knowledge of oneself and, for Evans, we have that
knowledge to the extent that we are able to pick up, and act upon, information that gives rise to
judgments with the IEM property. In effect, Evans is proposing an account of the sense of “I” in
terms of its functional role, where the input side is given by sensitivity to sources of information
that are IEM (where a source of information is IEM just if it typically gives rise to judgments
that have that property), and the output side is given by the ability to act upon information of the
relevant type (and upon the judgments to which such information gives rise).
As we will see further below, there is a third component to Evans’s account, but the two
components just mentioned illustrate how Evans elucidates what is “special and primitive” in
terms of the capacity to think thoughts, and to make statements, that are immune to error through
misidentification relative to the first person pronoun. The sense of the first person pronoun is the
ability to keep in touch with the person that one in fact is, the person referred to by one’s use of
the first person pronoun. One has that ability as a function of sensitivity to sources of
information about oneself that have the IEM property.
This picture of the sense of “I” is incompatible with taking the symmetry constraint seriously,
however, for the simple reason that my possessing sources of information about myself that are
immune to error through misidentification is irrelevant to your ability to understand and to report
what I say using the first person pronoun. In fact, the very idea of explaining the sense of “I” in
terms of a “special and primitive way” of being presented to oneself is fundamentally alien to the
symmetry constraint, for it leads to a conception of indexical thoughts as incommunicable. The
thought that I express by saying “I am hungry” cannot (on this way of thinking about the sense of
“I”) be the thought that you grasp when you hear me say it, since you have no access to my
special and primitive way of being presented to myself.4
In order to see what has gone wrong we need to look more closely at Evans’s account of selfreference and the sense of “I” (Ch. 6 of Evans 1982). Evans applies to the first person pronoun
his general model of what it is to understand a referring expression, which in turn rests on the
3
For an overview of the principal themes in Evans’s approach to the philosophy of language see the Introduction to
Bermúdez 2005. Bermúdez 2005a, in the same volume, is a detailed discussion of Evans’s account of the sense of
“I”.
4
Evans is prepared to bite the bullet here. He accepts that, on his understanding, “I”-thoughts are incommunicable.
He attempts to soften the blow by arguing, against Frege, that private and incommunicable thoughts can
nevertheless be perfectly objective. For further discussion see section 5 of Bermúdez 2005a.
11
basic principle that understanding referring expressions is a matter of standing in what he calls
an information-link to the referent of the expression. Information links allow the subject to
identify, or at least to know how to go about identifying, a particular object as the reference of
the referring expression. According to Evans, genuine linguistic understanding requires there to
be some such information-link on the part both of hearer and of speaker.
This is not the time to defend Evans’s use of what he terms Russell’s Principle, which has been
the subject of considerable discussion (Millikan, 2000) – although, as remarked earlier, the
natural reaction that the requirement of an information link is quite simply too stringent can be at
least partly allayed by appealing to the distinction between token-sense and type-sense. Instead I
want to point out that, even if one thinks that the information link requirement does hold in
general and that it applies to the first person pronoun, the type of information link provided by
sources of information that are immune to error through misidentification is completely unsuited
to do the job. The distinguishing feature of judgments based on such information sources is
precisely that they do not either involve or require any identification of a particular person as
oneself. They are identification-free.
Identification-free judgments have a distinctive epistemological property that explains both why
they are canonical examples of genuinely self-conscious judgments and why they are not strictly
relevant to the sense of “I”. Identification-free judgments are judgments whose truth-conditions
can be understood simply in terms of the accuracy or otherwise of the deliverances of sources of
information that are immune to error through misidentification. So, for example, if I judge on the
basis of introspection that I am thinking about the election then my judgment is typically true just
if my introspectively derived information is accurate – and my understanding of the truth
condition just is an understanding of what it would be for that introspectively derived
information to be accurate. Likewise for my judgment that I am extending my arm, when that is
based on the deliverances of somatic proprioception. This judgment is typically true just if my
proprioceptively derived information is accurate.
What does this tell us about what it is to utter with understanding a sentence of the form “I am
thinking about my car” or “my arms are crossed”? The conclusion to draw, I think, is that we
should think about understanding these utterances in terms of particular ways in which the
predicates “ – thinking about –” and “– extending –” can be satisfied. Plausibly, we understand
such complex predicates by understanding what it would be for an arbitrary object to satisfy
them. (This, in essence, is Evans’s generality constraint.) In the case of predicates that can be
judge to hold on the basis of information sources that are immune to error through
misidentification, such an understanding will typically have a first person and a third person
component. It is hard to see how a language user could be credited with understanding the dyadic
predicate “ – is thinking about –” if she were unable to distinguish between the introspective
grounds on which one might apply the predicate to oneself and the very different grounds on
which one might apply it to others. The same holds for any predicate that can be applied to
oneself on the basis of information sources that are immune to error through misidentification.
The point to extract from this is that, although uttering certain sentences involving “I” with
understanding does indeed require a degree of sensitivity to the types of identification-free
information highlighted by Evans, this has nothing to do with the token-sense of the first person
pronoun. Rather, the appropriate sensitivity is already built into the sense of the relevant
predicates. Immunity to error through misidentification is irrelevant to the token-sense of “I”.
The sense of “I”: A different kind of self-knowledge
So, where should we look for the token-sense of “I”? Is the token-sense of “I” a type of selfknowledge at all? In this section I argue that the token-sense of “I” should indeed be understood
12
in terms of self-knowledge, but the relevant type of self-knowledge is fundamentally different
from those considered in the last section.
Once again we can follow Gareth Evans’s lead. As mentioned earlier, there is a third element in
Evans’s account of the sense of “I”. This third element departs both from Frege and from
information-links that have the IEM property. The following passage explains why Evans thinks
that the considerations of functional role cannot be all that there is to say about the sense of “I”.
Indispensable though these familiar ingredients (an information component and an action
component) are in any account of the Ideas we have of ourselves, . . . they cannot
constitute an exhaustive account of our “I”-ideas. So long as we focus on judgments
which a person might make about himself on the basis of the relevant ways of gaining
knowledge, the inadequacy may not strike us. A subject’s knowledge of what it is for the
thought “I am in pain” to be true may appear to be exhausted by his capacity to decide,
simply upon the basis of how he feels, whether or not it is true – and similarly in the case
of all the other ways of gaining knowledge about ourselves. However, our view of
ourselves is not Idealistic: we are perfectly capable of grasping propositions about
ourselves which we are quite incapable of deciding, or even offering grounds for. I can
grasp the thought that I was breast-fed, for example, or that I was unhappy on my first
birthday, or that I tossed and turned in my sleep last night, or that I shall be dragged
unconscious through the streets of Chicago, or that I shall die. (Evans 1982, 208-209)
Recall the basic epistemological property of judgments of the form “I am F” with the IEM
property. These are precisely those judgments where the grounds on which they are based are
such that there is no gap between determining the presence of F-ness and determining that one is
oneself F. But we can entertain many thoughts about ourselves for which there are not (and could
not be) such grounds – thoughts about our distant past, for example, or about our future. These
thoughts concern states of affairs whose holding or otherwise cannot be determined by us on the
basis of information channels with the IEM property. In order to think such thoughts the subject
must be able to think of himself in a much richer way than is required in order to grasp the truthcondition of sentences that are immune to error through misidentification.
Since this class of “I”-thoughts does not have the IEM property, we need a different account of
how thinkers of such thoughts can have the relevant discriminating knowledge of themselves.
For Evans this knowledge is to be understood in terms of the subject’s ability to locate
themselves within the objective spatio-temporal world.
It seems to me clear that as we conceive of persons, they are distinguished from one
another by fundamental grounds of difference of the same kind as those which
distinguish other physical things, and that a fundamental identification of a person
involves a consideration of him as the person occupying such-and-such a spatio-temporal
location. Consequently, to know what it is for [δ = I] to be true, for arbitrary δ, is to know
what is involved in locating oneself in a spatio-temporal map of the world. (Evans 1982,
211)
As Evans understands it, being able to locate oneself within the objective spatio-world is a nontrivial achievement. It requires being able to integrate the egocentric space within which one acts
with a non-egocentric, third-personal representation of space. It is, in short, a matter of
possessing, and being able to situate oneself within, what psychologists term a cognitive map of
the world.5
5
This aspect of Evans’s view has been taken up by Campbell (1994) and Cassam (1997). See also Bermúdez 1998.
The essays in Eilan et al 1993 give philosophical and psychological perspectives on spatial representation.
13
For Evans, this ability is a third component in the sense of “I”, complementing the two functional
role dimensions that we have already discussed. My proposal, in contrast, is that this is all that
we need in order to understand the token-sense of “I”. Let me clarify the claim, showing how it
satisfies the desiderata identified earlier for an account of the sense of “I”.
We are looking for an account of what it is to understand sentences involving the first person
pronoun. Specifically, we need to elucidate what an understanding of “I” contributes to the
understanding of sentences in which it features. In thinking about this we have to start from the
obvious fact that the referent of “I” is the utterer of the sentence. This, after all, is the type-sense
of “I” and our account of token-sense must conform to it. In standard circumstances an utterer of
“I” grasps, as a function of the rule encapsulated in the type-sense of “I”, that he himself is the
referent of “I”, and so understands that the sentence expresses a thought about the particular
object that he himself is. As we have seen, this understanding is not purely extensional. The
comprehending user of “I” thinks about himself in a particular way, where this particular way of
thinking constitutes the token-sense of “I”. We have rejected the view that the speaker thinks of
himself as the unique recipient of certain types of information (coming from sources of
information with the IEM property). The proposal instead, following one strand in Evans’s
account, is that the thinker thinks of himself as an object uniquely located in space and following
a single path through space-time. But what is it to think of oneself in this way? This way of
thinking of oneself should be understood as a practical capacity. This is the capacity to situate
oneself within a cognitive map of the world – the ability to superimpose one’s egocentric
understanding of space (an understanding of space on which spatial orientation is fixed relative
to one’s behavioral capacities and, more generally by one’s practical engagement with the world)
upon a non-egocentric representation of how the world is spatially configured.
It should be clear how thinking about the token-sense of “I” in this way satisfies the second of
the three desiderata identified earlier. It is far more fine-grained than Kaplan’s notion of content.
The third desideratum is also satisfied. We can see how grasping the token-sense of “I”
contributes to a practical capacity to determine the truth-value of sentences using “I”. My being
able to situate myself within a cognitive map of the world is a necessary first step for knowing
how to go about determining the truth-value of sentences about myself that cannot be
immediately decided on the basis of information from information channels that have the IEM
property. And even those sentences that can be so decided are sentences about a physical object
located in space and following a single trajectory through space-time, so that I cannot grasp their
truth conditions without grasping that I myself am such a thing.
The first desideratum is the least obvious, however. How does the proposed account of the sense
of “I” satisfy the symmetry constraint. The symmetry constraint requires it to be the case that,
within a specified context, what you say using “you” or “he” can be token-equivalent (say the
same thing) as what I say using “I”. On the face of it, however, this is not possible if we
understand the token-sense of “I” in terms of an ability to situate oneself within a cognitive map
of the world. How can that ability be exercised by anyone but me? Do we not find ourselves
once again with a private and unshareable sense of “I”?
In order to see why not we need to recall the connection between grasping the token-sense of “I”
and grasping the truth-conditions of a sentence involving “I”. Grasping the token-sense of “I”
involves much more than simply grasping its type-sense. It gives a practical grasp of the truthconditions of the sentence. In virtue of such a grasp the thinker knows how to go about
determining the truth-value of the sentence. The first step in seeing how the symmetry constraint
is satisfied is noting that someone else who understands sentences that I utter using “I” in the
same context in which I utter them (and hence who is a candidate for the symmetry constraint)
will, in virtue of understanding them, also gain knowledge of how to go about verifying them.
This knowledge is a practical ability. My understanding of “I am F” is a function of my
understanding of “I” and my understanding of the predicate “– is F”. On the proposed account,
14
what my understanding of “I” contributes to my understanding of “I am F” is the ability to locate
a particular person (who of course is me) in space and time in such a way that I know how to go
about determining whether or not I am F. Now consider your understanding of my utterance of “I
am F”. Here too it looks as if what you understand in virtue of your understanding of my
utterance of “I” is the ability to locate a particular person (namely, me) in such a way that you
know how to go about determining whether or not that person is F. Of course, the way in which
you go about determining whether I am F may be very different from how I go about
determining whether I am F. The first-person and second-person ways of establishing F-ness
may differ. But this, as suggested earlier, has no bearing on the relation between your
understanding of “I” and my understanding of “I”. Both of these are to be understood, I suggest,
in terms of a single ability.
The argument here depends upon a particular view of how to individuate abilities, which should
be made explicit. I am assuming that abilities can be, and in many cases should be, individuated
in terms of their ends. So, my ability to locate myself is the same as your ability to locate me,
because there is a single person who is the target of both abilities. Once again, we need to
relativize to contexts. There are many circumstances in which two people might be trying to
locate the same person, but where there is no temptation to think of them both exercising a single
ability. What makes it the case that you and I exercise the same ability is that we are both starting
from a particular individual in a particular context (me in this room in the university, for
example) and what we are each doing is situating that individual relative to a more general and
non-egocentric understanding of space.6
What this gives us, then, is a clear way in which the token-sense of “I” is shareable (within a
context). There is a single ability that you and I have in virtue of each of us understanding my
utterance of “I am F” in the relevant context. The next step in seeing how the symmetry
constraint is satisfied is to observe that this ability is the same ability that you exercise, not
simply in your understanding of my utterance of “I am F”, but also in your understanding of your
own utterance of “you are not F”. Again, when you contradict my utterance of “I am F” by
saying “you are not F” we are both starting from a particular individual in a particular context,
whom we both understand to have a particular location in non-egocentric space, and predicating
contradictory properties of that person. Finally, this very same ability is exploited again when I
understand your utterance of “you are not F”.
*
*
*
There is a close connection between the sense of “I” and a particular form of self-knowledge.
Initial resistance to this idea (surely we do not need self-knowledge to understand how “I”
functions to refer to the person uttering it. . .) can be overcome through the distinction between
token-sense and type-sense. It is token-sense that we are interested in, where the token-sense of
an expression is a function of how an understanding of that expression contributes to a practical
grasp of the sentence’s truth conditions (as opposed to an understanding of how the expression
functions in the language). As we saw, an account of the token-sense of “I” must allow sentences
involving “I” and sentences involving “you” to be token-equivalent in particular contexts. This
symmetry constraint is decisive against proposals to understand the token-sense of “I” in terms
of one type of self-knowledge – namely, the type of self-knowledge derived from informationsources that have the IEM property. However, it is compatible with the proposal, derived from
one strand in Gareth Evans’s account of the sense of “I”, that the token-sense of “I” be
understood in terms of a very different type of self-knowledge – namely, the self-knowledge
manifested in the ability to locate oneself within a non-egocentric conception of space and time.
This ability can be shared by different individuals within a context, and can be deployed in
6
For more on this way of individuating abilities see Bermúdez 2005a, particularly section 3.
15
comprehending utterances both of “I” and “you”. It is here, I would suggest, that we can find the
deep connection between self-reference and self-knowledge.
Bermudez, J. L. (1998). The Paradox of Self-Consciousness. Cambridge MA: MIT Press.
Bermudez, J. L. (2002). Sources of self-consciousness. Proceedings of the Aristotelian Society,
102, 87-107.
Bermudez, J. L. (2003). The elusiveness thesis, immunity to error through misidentification, and
privileged access. In B. Gertler (Ed.), Privileged Access. Aldershot: Ashgate.
J. L. Bermudez (Ed.), Thought, Reference, and Experience: Themes from the Philosophy of
Gareth Evans. Oxford: Oxford University Press.
Bermudez, J. L. (2005a). Evans and the sense of 'I'. In Bermúdez 2005.
Campbell, J. (1994). Past, Space, and Self. Cambridge, MA: MIT Press.
Cassam, Q. (1997). Self and World. Oxford: Oxford University Press.
Eilan, N., Brewer, B., & McCarthy, R. (Eds.). (1993). Spatial Representation: Problems in Philosophy
and Psychology. Oxford: Blackwell.
Evans, G. (1982). The Varieties of Reference. Oxford: Oxford University Press.
Frege, G. (1918). The thought: A logical inquiry. In P. F. Strawson (Ed.), Philosophical Logic.
Oxford: Oxford University Press.
Jack, J. (1993). The role of comprehension. In B. K. Matilal, Chakrabarti, A. (Ed.), Knowing
from Words. Dordrecht: Kluwer.
Kaplan, D. (1977). Demonstratives. In J. Almog, Perry, J., and Wettstein, H. (Ed.), Themes From
Kaplan. Oxford: Oxford University Press.
McDowell, J. (1993). Knowledge by hearsay. In B. K. Matilal, Chakrabarti, A. (Ed.), Knowing
from Words. Dordrecht: Kluwer.
Millikan, R. G. (2000). On Clear and Distinct Ideas. Cambridge: Cambridge University Press.
Perry, J. (1979). The essential indexical. Philosophical Review, 86, 874-897.
Perry, J. (1997). Indexicals and demonstratives. In R. Hale, Wright, C. (Ed.), Companion to the
Philosophy of Language. Oxford: Blackwell.
Sainsbury, R. M. (1979). Russell. London: Routledge Kegan Paul.
Sainsbury, R. M. (1998). Indexicals and reported speech. In R. M. Sainsbury (Ed.), Departing From
Frege. London: Routledge.
Shoemaker, S. (1967). Self-reference and self-awareness. Journal of Philosophy, 65(19), 555-567.
Wright, C. (1998). Self-knowledge: The Wittgensteinian legacy. In C. Wright, B. C. Smith & C.
MacDonald (Eds.), Knowing Our Own Minds. Oxford: Oxford University Press.