Self-Knowledge and Self-Representation

ABSTRACT

In this paper I introduce a centrast between homomorphic and non-homomorphic aacriptiona of informational content to repreaentationa. In the former case there is a mapping from the parts of the representation onto the conatituents of the content. In the latter case, there is not; aome of the conatituenta of the content are settled by background factora . I centrast this diatinction with that between contezt dependent and contezt independent ascriptions of content. I note that in cases where the ascriber of content sharea the background with the agent, one is inclined to ascribe homomorphic content of a sort that does not have a fized truth-value to a representation. Thi1 leads to the notion of relative information. Soma UIIB for relative information are noted. Finally, the di1tinctiona developed are used to distinguish three type& of self-knowledge and account for their relations.

I HOMOMORPHIC AND NON-HOMOMOßPHIC

REPRESENTATIONS

Philosophars of mind and language and researcherB in artificial intelligence mu1t confront the que1tion of the content of representations. Philo1ophers want to understand what it is to know or believe or say something; reaearchera in AI want ~o build things that do know, believe and say things . In fact, moat researchtri in AI also 1hare the philoaophera goal; indeed, a1 far &I I can tell, most of them are philosopher1. But oven if thi1 were not so, underatanding the nature of knowing, believing and saying ought to be helpful in learning how to construct thing1 that have the1e capacitias.

At one time, the paradigm for representations was the utterance of an eterllal

SELF-KNOWLEDGE AND SELF-REPRESENTATION

John Perry

Department of Philoaophy and CSLI, Stanford

aentence, with a content fized by meaning alone . More recently. we have realized the importance of contezt. The meaning o! a repreaentation typically fizes content only relative to contezt . In this paper I diacuaa a further factor, background, that interacts with contezt in interesting ways.

Consider my utterance of SAM lAS SLEEPING . This utterance has the informational content that Sam was 1leeping. (I use informational content for that which i1 information if it get1 thinga right, and call it ·content· for short.) Thi1 content ha1, intuitively, three constituents: the individual Sam, the pa1t time in que1tion, and the property of being a1leep . The 1entence I used haa a number of constituent1: SAM, lAS, SLEEPING, and WAS SLEEPING. Among the1e are three that identify the constituents of the content. So, there is a homomorphiBm from the conatituent1 of the 1entence onto the

constituents of the content .

Thi1 might 111m inevitable. After all, how could constituents make their way into a proposition that an utterance ezpresses, unle11 some constituent1 of the uttered ezpr111ion identified them (at least in a 1uitably generous notion of conlitutents of an ezprenion)?

But now consider another ca11. George sticke his left han~ out the window of hi1 car, thereby signalling that he will turn left at the nezt corner. The content has as constituents George, the nezt corner, and the relation of turning left at. le might ~easonably say that the particular way he hold1 his arm stand& for the relational activity of turning left at. But the way he hold& his arm does not have constituents or aspects that identify the corner at which he i1 going tc turn . lhich corner is not, 10 to apeak, a question that the structure of the signal needa to nsolve . Once the anglo of bh arm identifies right or left . we have all the information we need . The rest of the propositional constituents are not supplied by aspects of the signalling movement, but by what I shall call the background .

The background is typically determined by the contezt. le are imagining the signalling to occur in the U.S. in the last half of the twentieth century. where there is an institution of signalling for turns. This institution provides a propositional function for each contezt. The function takes directions (left, right) as arguments and returns a proposition: that the driver involved will turn (left, right) at the nezt corner.

This suggests the following picture . Communication takes place against a background, determined by contezt . A background provides a propositional function, taking some sort of entities as arguments. The job of the representation is to provide the necessary arguments to get from the background to a proposition . The case of uttering a complete declarative sentence ·in vaccuo·. so to speak, is the limiting case. The background is null : it provides only the identity function, from propositions to propositions. The constituents of the proposition are supplied by the background and the representation; homomorphic representation is just a special case, where the background is null .

Backgrounds of this sort may be established in various ways. In the case of the driver's signal, there is an institution that establishes a background for each driver in th· vicinity of a corner . le may think of questions &I providing backgrounds . The question ·who shot Liberty Valence?· provides a function from individual· and types of individual· to propositions. An answer need only specify an individual (Jack Palance, say) or a type of individual (some man with a large gun) .

II BACKGROUND AND COHTEXT

The distinction between homomorphic and non-homomorphic representations cuts across that between eternal and contezt sensitive representations .

For ezample, if you ask me the square of 2 and I say ·Four·. my answer is eternal. ·Four· stands for four in any contezt (or so we may assume) . But the representation is non-homomorphic, because the informational content, that the square of two is four, picks up constituents from the background .

lhen I say ·I am sitting·. the representation is homomorphic but contezt sensitive . If I say ·John Perry is sitting on May 14, 1986·, the representation is homomorphic and (arguably) eternal. If you aak ·who wanta a chocolate milkahake?· and I reply ·Me·, my answer is both contezt sensitive and non-homomorphic.

lhich background ia relevant is generally a matter of contezt · but it need not be . le can imagine a convention of language. for ezample, that placing one'a hand vigorously over one's heart supplies the answer ·yea· to ehe question ·Ia America the home of the brave and the land of the free?· no matter who does it and when . In this caae, we imagine the background being Iupplied by the meaning of the representation, and not its contezt . (For 1om~ purpo1e1, however, it is uaeful to think of the language being used as a conteztual fact.)

Thus, the contezt of a representation is just the larger situation of which its uae is a part . The contezt may contain all sorta of facts relevant to the content of the representation . The background for a representation ia not determined by the meaning of any part of the representation, but by the meaning of the whole . The meaning may determine the background ·eternally· as in our patriotic ezample. More commonly the meaning determines the background only relative to contezt. In the signalling case. the system of signalling determines a relation from contezts to backgrounds. For a signaller A approaching a corner C, the background is the partial function that takea a direction as argument and returns the proposition that A will turn in that direction at C.

III RELATIVE INfOßMAIIOH

lhen a background is supplied by contezt. the background may be more or less sensitive to shifts of contezt. For ezample . the statement ·It's four o'clock· said by me now, sitting in Palo Alto, has the informational content that it is four o'clock Pacific Coast Time . le can imagine that ·It· refers to a time . Then there really is not a property of timea, being four o'clock, but only a relation between times and zones. So my representation is non-homomorphic. The zone supplied as background does not vary over wide geographical regions, although it changes suddenly when it does .

For people in a single time zone talking to one another, there is no point in mentioning the time-zone . Indeed, people who grow up in one time-zone and do little traveling and don't watch TV, may function perfectly well without ever realizing that there are time- zones, and that being four o'clock is really a relation and not a property . They don't need the concept of a time-zone at all to make use of information about the time of day and to detect and communicate such information themselves .

In general, the meaning of a representation, a sentence of English, for example, can be seen as a relation between the sorts of situations in which it occurs and the content of those situations . Thus we can tako the meaning of I AM SITTING to be that relation between utterances and contents that obtains if the content requires the speaker to be sitting . Given this picture, contents provide an equivalence relation among utterances that employ sentences· with different meanings . So, the utterance where I utter I AM SITTING and the one where you utter YOU ARE SITTING while talking to me have the same content . This equivalence relation is important in understanding the flow of information, where i nformational content is preserved across changes of context . It is also important for understanding such psychological notions as continuing to believe the same thing (Barwise and Perry. 1983; Perry. 1980; Perry. 1985).

The phenomenon of non-homomorhic representation suggests that we need to broaden the notion of informational content, to include not just "propositions·, that are true or false absolutely. but also various types of relative informational content. that are true and false relative to a background .

Consider communication about the time t>f day . So lang as this takes place wi thin a time-zone , there is little need to worry about the propositions expressed . If I ask Ingrid what t ime it is and she says "It's four o'clock". I do not need tothink "She is in the Pacific Zone, so that means that it is fou: o'clock Pacific Coast Time . · I just think

"It's four o'clock". We can say that she communicated the information that it's four o'clock, where this is relative information, information that is true or false only relativ\ to a background . The transaction is information-preserving only if the participants share a background . If Ingrid is talking to me long-distance from New York City and I don't realize it, the transaction will not be information preserving.

We might suppose (and I did for a long time) that we do not need to recognize relative information, since meanings. relationally conceived, give us entities that are true or false relative to context . But there are transactions in which meanings change systematically to preserve relative information . Suppose Ingrid says "It's four o'clock by my watch," and I tell you "It's four o'clock by Ingrid's watch . " Meaning has not been preserved , and the shift in meaning does not automatically preserve non-relative information . What is preserved is information relative to a time-zone . We seem to need the notion of relative information to think about this transaction, and to be able to carefully characterize just what goes wrong when she made the original remark long-distance from New York , and my remark to you was false .

IV SELF-KNOWLEDGE AND SELF-REPRESENTATION

I think the notions of non-homomorphic representation and relative information are crucial in a number of areas of epistemology and semantics . For example, Jon Barwise (Barwise. 1985) has recently analyzed conditionals as providing relative information about the three place relation among types of situations, T involves T' relative toT''. T'' is supplied by the background, and Barwise shows how some of the puzzles about conditionals involve i nferences t hat are only valid when the background is kept fixed . I suspect these notions will be important in dealing with subjunctives, unbounded dependencies of various sorts, and other troublesome topics .

I want to end the paper by focussing on a particular topic, however, that will suggest a more or less deep reason why relative information and non-homomorphic representations are so important .

folk. ie saw that they can communicate perfectly well about the time of day without having any words for or even concepts of time-zones. Because they live within a certain constant background, they have no need to worry about it . In particular their thought about the time of day can be keyed to certain perceptions and actions in a perfectly workable way, even though they lack such concepts . They look at the clock and think ·rt·s five o'clock". and so they close up shop and go home for supper .

Even those of us with a well-developed conceptual apparatus for dealing with time-zones operate, for the most part. in ways that allow our perceptions of the time of day and the actions we take in virtue of the time of day to ignore time-zones . Like the folk above. we go home when the clock shows five o'clock (or so) .

I think these facts about time-zones are suggestive about certain basic facts of the human condition. The information we pick up through perception is always relative information . ihen I see a terminal before me. there need be nothing in my perceptual state that is indicative of its being me that the terminal is in front of. Not only is there not a constituent of the perception that ·eternally" stands for me. there need not even be a constituent that picke me out in context. I am the background for my perceptions, and you are the background for yours. ihen you are in the very same perceptual itate I am now in, you know that there is a terminal in front of you, not that there is one in front of me.

Imagine now a somewhat simpler organism than myself. perceiving a potato rather than a terminal in front of it . le can imagine this perception leading to a cognitive state, that in turn leads to the action of seizing the potato and jamming it into one's mouth . le can ascribe content to the action, in terms of its result . Each such organism, taking the action in question. will jam the potato into its own mouth. not someone else's. So we can think of actione &I having non-homomorphically determined content, or homomorphically determined relative content . These organisms do not need to have any concept of themeelvee. They surely do not need any ·eternal· idea of themselves, but they also do not need any internal indexical either. Since they are always in the background of their perceptions and actions, they need not be represented in the cognitions that intervene between them.

ie can suppose, then, that relative information is systematically connected with types of perception and action . Equally important, we cannot imagine that non-relativ~ information is systematically tied to them. That Elwood has a potato in front of him can't be something that all our little organisms know on the basis of being in the potato-in-front percpetual state, and can't be something the cognition of which leads them all to grab and shove . At any rate, if things worked this way. Elwood is the only member of the group that wouldn't go hungry .

These reflections suggest the following picture of our cognitive make-up . At the 1 bottom· level. we have cognitions that have no r cpresentation of ourselves (or the present moment), which are tied pretty directly to cognition and action . This gives us self-knolwedge of a eort: we know the world from our perspective . At the ·top· level we have representations that are not systematically tied to perception and action (or at least not to the same sorts of perception and action). in virtue of which we have relatively context insensitive cognitions that homomorphically determine propositions about ourselves. This is self-knowledge of another sort. I have it if I read a note ·John Perry must call home,· where "John Perry· designates me. Note that I would have it even if I had forgotten my name. and didn't realize. as I might put it, that I was to call home . The real purpose of indexicals is to mediate between these levels, yielding full-blown self-knowledge (although not yet Socratic self-knowledge). ihen I read a note. ·John Perry please call home.· I think "I must call home· and then go into a state that we might express with ·must call home· . The step from the top level to the intermediate level varies from person to person; the step from tne intermediate level to the bottom level, and from it to action (modulo procrastination) is universal . The purpose of indexicals is to align the homomorphic representations we get through language and other forme of communication, at eome kinds of memory, with a more basic, selfless, cognitive syetem.

One who has the bottom and top levels ;orrectly linked knowe who he is (where he is , what time it is) . This is still oversimple. in a variety of ways . le would need more levels, for example, to get at what happene when one realizes that it is four o·clock Eastern Time, hence one o'clock real time, hence one o'clock here, hence time to go to lunch . But I hope I have said enough to suggest that the topics of non-homomorphic representation and relative information are worth careful thought, whether ~r not the thought I have provided is careful enough .

ACKHOWLEDGEMENTS

Recognition of the n·ed for something like relative information, and the consequent need to distinguish between non-homomorphic and context dependent representation, was forced on me by Joseph Almog and Bob Moore in the course of conversations about the motivation for propositions with truth values relative to times , as are found in David Kaplan's work on demonstratives (Kaplan. 1979) . The present ~pproach is the result of conversations with

Jon Barwise. David Israel, Moore, and others.

REFEREHCES

1. Barwise, J . 1 The Situatio~ Conditionals and Conditional CSLI Report Ho . 21. 1985 . in Logic-II : Information·.
2. Barwise. J . and J. Perry. Situation& and Bradford Books. Cambridge, Attitude&. Massachusetts : MIT Press, 1983 .
3. Kaplan. D. ·on the Logic of Demonstrat­ ives· , in Perspectives in the Philosoph! of Language. ed . French, P. A. , T. E. Uehling, Jr . , and H. K. lettstein . Minneapolis, 1979, pp . 383-412 .
4. Perry, J . 1 A Problem about Continued Belief· . Pacific Philosophical Suarterly 61, 1980, pp . 317-332.
5. - - . "Percept.ion, Action. and The Structure of BeliH 5 (· :;. in Grandy & larner : Philosophica) Grounds of Rationality. Oxford : Oxford University Press, 1985, pp . 330- 359 .