CSLI MONTHLY ----------------------------------------------------------------------- February 1987 Vol. 2, No. 5 ----------------------------------------------------------------------- A monthly publication of The Center for the Study of Language and Information ------------------ Contents The Wedge Is Realism for Real?---A Response to John Perry's Seminar 1 by Terry Winograd Winograd Versus Reality 2 by John Perry A Workshop on Semantical Paradoxes 3 by Jon Barwise and John Etchemendy A New CSLI Working Group 3 Position Available in the Symbolic Systems Program 3 CSLI Visiting Scholars 3 CSLI Publications 3 ------------------ THE WEDGE IS REALISM FOR REAL?---A RESPONSE TO JOHN PERRY'S SEMINAR Terry Winograd The Challenge In his recent seminar, John Perry defended his version of "critical realism" against the arguments presented in Winograd and Flores [1]. He argued that the only sensible position was some form of realism and that the radical rejection of realism in the book could not be taken seriously. He discussed at some length the example in chapter 5, concerning the question "Is there any water in the refrigerator?" He noted his agreement with the point made there that the appropriateness (or "felicity") of an answer had to be understood in terms of the context of the utterance and the background understanding of speaker and hearer. Let us briefly recapitulate the essence of the example in a compacted form: A friend walks into the kitchen where person A and person B are sitting and asks, "Is there any water in the refrigerator?" A says, "Yes," and B says, "No." The friend looks in and says, "I don't see any." A responds, "In the cells of all the vegetables." The point is that the correctness of the answer (or more abstractly, the truth value of the assertion "There is water in the refrigerator") cannot be judged in terms of some objective reality about water, but only in terms of the speaker's and hearer's understanding of the relevant background and purposes. In a different background the opposite answer might be better: The friend says, "I want to store some chemicals that are supposed to be kept away from water." A says, "Keep them out of the refrigerator, there's lots of water there---in the cells of all the vegetables." Perry argued that although he agreed with the thrust of this example, it was a non sequitur to take it as denying the existence or relevance of an objective reality. The appropriateness of either answer does seem to depend, he argued, on whether there really is a bottle of water in the refrigerator. Examples like this simply demonstrate that in order to account adequately for language use, other factors need to be considered in addition to objective reality. The Response The problem is that although Perry says he agrees, he doesn't take the point seriously enough to apply it to his own questions and answers. Consider the following variation on the example: A cognitive (or AI) researcher walks into CSLI and asks "Is there any objective reality out there?" Barwise and Perry say, "Yes" and Winograd and Flores say, "No" ... In asking who is right, we must be just as careful as in the first example to consider what the questioner is really thirsting after. Note that I have loaded the question here by assuming that the hearer has the concerns of a cognitive researcher. For a scholar of philosophy, a theologian, or a physicist, we would have to address other issues in assessing the appropriateness of the answers. Perry emphasized in his lecture that he is primarily a philosopher, but his choice of projects and associations over the last few years suggests that he expects what he says to be taken seriously by a variety of researchers concerned with language, cognition, and computation. I take it that statements about language, thought, and reality are being addressed to people who are concerned either with developing a scientific account of human language, or with designing artifacts (computer systems) that operate in relation to the human use of language. In [1] we emphasize (e.g., in the book's subtitle) that our concern is with `design'---with developing a theoretical base for creating meaningful artifacts and for understanding their use and effects. In this quest, we are concerned with how people's use of language is correlated with what they experience and do in the world. In this sense, we share the `realist' critique of cognitivism. We take the relation between language and the world as our primary concern and argue that psychological theories cannot provide the basis for explicating this relationship. However, we are not realists in Perry's sense, and his own example can serve to show why. Language in Action Perry describes what he imagines to be the procedure for admitting students in the Computer Science Department. There is a collection of folders, one for each candidate. The contents of the folders consist of statements in natural language, such as "Brown has a 4.6 GPA in his major," "Smith is a top athlete," and "Frankly, Jones is a first-class nincompoop." The committee carries on a discourse in which further statements are made, like "Brown is our top candidate" and ultimately sends acceptance and rejection letters. Perry's point is that although many contextual and background factors are involved in interpreting all of this language, there is indeed some contact with reality. If pranksters were to replace the folders with fictional ones, no matter how well-written and convincing, they would bear a different relationship to reality. According to Perry, sentences like "Brown is our top candidate" would still be equally valid, but those like "Smith is a top athlete" would no longer make statements about the real world. This fundamental difference, he argues, must be central to our theories of meaning. Of course, this distinction makes obvious sense. But so what? How is it relevant to anything we might care about as cognitive scientists or computer system designers? The first observation one can make is that the correspondence between language and reality in this case cannot make any difference to the cognitive processes of the admissions committee. They do not know whether they are reading a real or a fraudulent folder, and the processing can depend only on what they read and discuss. If the committee considered the possibility of fraudulent folders, their beliefs as to whether they were dealing with reality or fiction might be relevant to how they did their work, but that is quite another thing. Perry argued that folders for real and fictional students bear different kinds of connection with reality. Real students who are admitted will later appear in the department, and on the basis of their performance one can, for example, discuss whether the admissions process is "working correctly." But, of course, this further discourse may be only second-hand as well. The admissions committee may never meet the students themselves, relying on grade reports, comments from their colleagues, and the like---all of which could equally well be fabricated. If we are trying to come up with causal explanations of mental processes, the claim of realism may not be wrong, but it is vacuous. It simply doesn't make any relevant difference. However, as we argue in [1], Perry's "critical realism" has a more pernicious effect as well. Instead of the reality of the candidates themselves let us consider the reality of their properties. How are we to interpret statements like "Smith is a top athlete"? According to Barwise and Perry [2], there is some objective (real) property which Smith exhibits (let us call it "top-athlete-ness"). Does this mean that he is highly competitive (the New England tiddly-winks champion), narcissistically devoted to his physique (spends eight hours a day working out), or what? Here's where the "critical" part of "critical realism" comes in. Perry acknowledges that there may be many different real properties all associated with the same natural-language phrase, and that the attribution of a particular one to Smith is highly context-dependent on who said it, with what expectations about the reader's purposes, etc. In fact, Barwise and Perry apparently feel that their position would still be valid if there turned out to be a different such property for every single instance in which the phrase "top athlete" was uttered. But now we are truly in the cells of the vegetables. Perhaps they can argue without logical contradiction that such properties are "real," but this provides no basis for the things we want from reality. Reality and Objectivity Let's look at that more closely. What do we want from reality? From the perspective of understanding how language works, we want some basis for intersubjective correlation. If I say something is a spade and you say it is a spade, we need to go beyond the trivial observation that each of us is identifying some (but not necessarily the same) property. Reality, when it works, gives us an objective standard against which individuals can be measured. There is some property of "really being a spade," which I may have right or wrong and you may have right or wrong. But if we both have it right, we are really referring to the same property, and our communication is successful. This seems obvious, so why is it pernicious? The problem is that it only works for the simple cases. It is a wild goose chase to look for the objective reality associated with "top athlete" or "first class nincompoop." There is no idealized real property that we could associate with one of these phrases if we "got it right." There are only contexts of use and relation to interpreted purposes. Further, in the cases where the quest for objective reality doesn't really work, it can lead to misguided attempts to establish objective grounds for things that need to be understood in terms of discourse and dialog. Imagine that we are building a computer system to aid in the admissions process. It will not be expected to make the decisions, [Note: Whether it should or could do so is a separate issue, also discussed in [1].] but will simply help organize the `data' for use by the admissions committee members. How do the statements above get entered into the data base? To take the most seemingly straightforward, we could have an entry for each student's GPA, and another one for "GPA-in-major." But what counts as "in the major"? What objective real property is being measured here? Some schools might count all courses offered by the major department, others count those courses required for the major, etc. Some schools might have a "computer hardware engineering" major, while others give degrees to computer specialists whose major is "mathematics." The point is not that these complexities are incomprehensible or could not be reduced to further distinctions. But when they are conglomerated into a single "property" (as they ultimately must be represented in a data representation or logical formalism), these distinctions are lost. The process of further distinctions is potentially endless (what distinguishes a "computer science" course?). All hope is lost when we get to assigning each student an appropriate data base entry for his or her "class" of "nincompoopity." Common Sense and Provisional Realism But what does our common sense tell us about this problem? Perry appeals to the common sense of realism. "Can you," he might ask, "say with a straight face that this chair you are sitting on is not objectively there---that it exists only through language? Maybe there are difficult borderline cases, but surely you must believe in the objective existence of your chair." This kind of appeal to common sense is dangerous and hardly the thing on which to ground philosophical theories. Common sense tells me that the ground beneath my feet is flat and that the sun and moon circle overhead. It tells me that running faster will not make me shorter in the direction I run, or make my watch slow down. Common sense is full of generalizations that work over some limited (and common) range of phenomena, but which have to be abandoned when we go further. Given this observation, of course, Perry is right about the chair. We deal with everyday objects and assign them labels like "chair" and "athlete" all the time, and we seem to get along pretty well. This is because we act in accord with a methodology I will call (until corrected by someone who knows the philosophical literature) "provisional realism." For the purpose of any particular discourse (e.g., the one carried on by the admissions committee) we take for granted some distinctions as being "objective"---that is, intersubjectively valid without further discussion. Most of our everyday language about chairs and the like falls into this category most of the time. Others (such as "top athlete" or "GPA-in-major") may be taken as objective for some purposes, and then open to discussion for others. The important thing is that we take the attribution of reality as provisional. When there is a breakdown because two people are not using natural-language terms comparably, the question isn't, "Which one of us (if either) has it right?" but, "Where do we locate the appropriate discourse about the interpretation of this term?" We may end up looking into the conversations about the computation of grades at some university, or be pushed into something as individual as a dialog with Professor X as to why he thought Jones wasn't a third-rate nincompoop instead. The reason the "real" folders must be understood as different from the fictional ones is that there are different potentials for further discourse. There could be conversations with the recommenders, university registrars, parents, and so on. Of course, any one of these could be made up as well---there is no final objectivity. But in the simple case of fabricated folders, the potentials would be drastically different, and even for the most sophisticated forgery, there would be some point at which the discourse appropriate for a "real student" would diverge from that for a fictitious one. Perry hinted at this in his discussion, when he said that in order to function we must be able to take some things as objective---that is, "independent of the discourse in question" (direct quote from his talk). This is a nice statement of provisional realism. It makes it clear that in any given analysis there is a particular discourse in question, and that there might be other discourses which take into consideration the things the first one takes for granted. Notice here that we are using "discourse" in a rather abstract sense. In a single stretch of language behavior (e.g., a meeting of the admissions committee) we might have some amount of discourse that takes the terms above as objective, some discourse in which they are questioned, some in which the terms in which they are articulated are in turn questioned, and so on. Perry's worry that, "If you don't have realism, everything is on a par" is true in that we always must act as though some body of grounding could be treated as objective, but false in that every element of that provisionally objective reality can be potentially open to rearticulation. Conclusion In casual asides during the talk, Perry suggested that some of the more provocative statements in [1] may be the result of the authors' not taking philosophical consequences seriously---of treating statements lightly, saying in effect, "that's just philosophy." I claim the opposite. The differences arise because we `do' take philosophy seriously---not as a structure of clever arguments and refutations, but as a basis for acting. We require of a philosophy that it do some work for us. In particular, it serves as a guide for how to proceed when our everyday unreflective bumbling along gets us into trouble. Our primary claim in the book is that in designing computer systems (whether AI-style or more conventional), the appropriate guide is one based on provisional realism grounded in an understanding of discourse and conversation.[Note: This paper is not the appropriate place to elaborate that approach in detail. It is outlined in [1] and further developed in [3],[4],and [5].] Faced with a problem in representing the contents of admissions folders, the right questions are neither realist ("What is a GPA, really?") nor cognitive ("What is in the concept of GPA?") but conversational ("What is the structure of the discourse in which the distinction "GPA" emerges"). This is not a particular kind of problem that arises in creating AI programs or data bases, but a general problem encountered in trying to relate natural language (or thought) to any precise symbolic representation. It applies equally to cognitive psychology, data-base design, and the formal semantics of natural language. As remarked in the beginning, one may not share this concern for relating language to symbolic representations, just as someone asking about water may not be interested in getting a drink. There are serious branches of philosophy in which such concerns are the farthest thing from the philosophers' minds. However, I think it is realistic to assume that the primary audience of people interested in CSLI are directly concerned with issues of representation and the formalization of meaning. It is these people whose concerns we have in mind when we truthfully answer "No, there is no objective reality." References [1] Winograd, T., and Flores, F. 1986. "Understanding Computers and Cognition: A New Foundation for Design." Norwood, NJ: Ablex. [2] Barwise, J., and Perry, J. 1983. "Situations and Attitudes." Cambridge, MA: Bradford Books/MIT Press. [3] Winograd, T. 1985. Moving the Semantic Fulcrum. "Linguistics and Philosophy" 8 (no. 1): 91--104. [4] Winograd, T. In press. Cognition, Attunement and Modularity. "Mind and Language," 1987. [5] Winograd, T. In press. On Representation. CSLI report. WINOGRAD VERSUS REALITY John Perry [Note: This note is based in part on a CSLI seminar and in part on a lecture given to Terry Winograd's class on language and cognition.] Terry Winograd, in his essay [1] and his book, with Fernando Flores, [2], has maintained the view that "nothing exists except through language." ([2] 68) Winograd and Flores reject the assumptions that we are inhabitants of a real world about which there are objective facts that do not depend on the interpretive acts of people ([2] 30, 33). I do believe that we are inhabitants of a real world made up of objects bearing properties and standing in relations, and that there are objective facts about that world that do not depend on the presence of people. So I think that what Winograd and Flores say is wrong. Naturally, I also think Winograd's criticisms of Barwise and Perry [3] based on these ideas are quite misplaced. But, at least for now, I shall concentrate on the less parochial issue of whether anything exists outside of language. In what follows, I first use an example of my own choosing to explain the reasons why I think that any reasonable theory of language, thought, discourse, and the like should recognize objects and facts that are independent of language. I then turn to an argument of Winograd and Flores that uses a somewhat more complex example. I believe that the reader will find that the points I make will serve to undermine all of the various and sundry arguments they put forward in the book against language independent reality. Reasons for Realism Imagine this simple example. I ask my wife, "Where is Felix?". She says, "Felix is on the table." I go to the only table in our house, find Felix, and put him outside. This is an example of successful cooperative action. In particular, my wife, with her remark, achieved her goal of being helpful to me in achieving my goal, of having our cat out for the night. It seems to make sense to try to analyze, at least partially, which factors were relevant to the success of my wife's remark. The ordinary way of doing this is to try to isolate one factor, and see how varying it would change things, keeping the other factors constant. This may require experiments, although in the case of things we use a lot we may make some progress by just drawing on our own knowledge of what would happen if various things were changed. This is a pattern of thinking Winograd and Flores use, so I don't see how there can be any objection to my using it here. For example, it seems clear that the time of my wife's utterance was relevant to its success, and also the place. If she had said what she did where I couldn't hear it, or waited an hour until I was sound asleep, it would not have been successful. Similarly, it seems that the sentence she uttered was relevant to the success of her utterance. Let's vary the sentence she uses, but hold everything else we can the same. So, with everything else exactly the same, including the cat, as it was in the above example where her remark was helpful, she says instead, "Felix is already outside." Then my action, which was not independent of the words she spoke, would have been different. Rather than going to the table, picking up the cat, and putting it out I would get in bed and go to sleep. But then, I would not achieve my purpose, of having the cat out, for the cat would still be inside. And so my wife's action would not have been helpful to me in achieving my goal. So, our method has shown us that the helpfulness of her remark was dependent on when and where she uttered the words she did, and what those words were. It also seems that the helpfulness of her remark depended in part on where Felix was. If he had been outside, rather than where he was, the remark she made wouldn't have been helpful. And, if he had been outside, the second remark we imagined her making would have been helpful. This is my candidate for a fact that is not dependent on language, is perfectly objective, and needs to be recognized by any theory that will cope with the helpfulness of my wife's remark. I do not need to deny, of course, that some of the factors on which the success of my wife's remark depends are dependent on language. Intuitively, it seems that some of the factors that are relevant to the success of my wife's utterance are independent of language, and others are not. And that is exactly the position that I would subscribe to. I am not claiming that all facts are outside of language, only that some of them are. As an example of a fact that depends on language, consider the fact that our cat is named "Felix." This is a fact about language, although not about language alone, for it is also a fact about our cat. On the other hand, the fact that Felix is on the table seems not to depend on language at all. I also don't need to deny, and don't want to deny, that for some kinds of discourse, there aren't any such extralinguistic facts that are relevant to its success or lack of success. More to the point, I think that there is considerable variation, of an interesting sort, as to how and in what way the success or failure of a piece of discourse in realizing the aims of its participants depends on extralinguistic facts. One of the great advantages of recognizing extralinguistic facts, is being able to make such distinctions. That is my argument for the view that there are language-independent facts, and that a theory of language needs to recognize them. Without recognition of such facts, I do not see how we could understand the fact that some utterances are successful in achieving the intentions of their makers, helpful to those to whom they are made, and the like, while others are not. Do They Mean It? One might wonder, at this point, if I have not simply taken Winograd and Flores much too literally. Perhaps they really don't mean to deny that there are facts, like whether my cat was on the table or not, that do not depend on language. Perhaps they simply want to distance themselves from ordinary ways of talking about such facts, in order to help make the point that in the complex modern world, particularly those parts of it that have to do with computers, the objects, properties and relations that depend on discourse of various kinds are extremely important. The following sort of passage suggests this interpretation: The distinctions made by language are not determined by some objective classification of "situations" in the world, but neither are they totally arbitrary. Distinctions arise from recurrent patterns of breakdown in concernful activity. There are a variety of human activities, including drinking, putting out fires, and washing, for which the absence or presence of "water" determines a space of potential breakdowns. Words arise to help anticipate and cope with these breakdowns. ([2] 69) Winograd and Flores here seem to be adverting to the fact that, water is some places and not others, and that this is sometimes relevant to the success---nonbreakdown---of linguistic activities. It is not likely that they suppose that, say, the fact that a remark like, "Drill down a couple of hundred feet right here if you want water" is more likely to be helpful if made while standing in Nebraska over the Ogallala Aquifer, than if made while standing at most spots in Arizona, is to be totally explained without appealing to facts that do not depend on language. On this interpretation, while Winograd and Flores may wax a bit rhetorical at times, their point is that `in addition' to language-independent facts, and to a great degree that is usually recognized, the success of discourse depends on language-dependent facts. Unfortunately, this interpretation doesn't hold up. Winograd and Flores are not just encouraging us to recognize factors other than language-independent facts. They are not just minimizing such facts by talking about the presence or absence of "water" rather than the presence or absence of water. As we shall see in the next two sections, they really do argue that we do not need language-independent facts. Impatient Rationalists In my original example, I seem to exhibit the impatience that Winograd and Flores expect from readers in the "rationalistic tradition." Such readers, they note, are likely to concentrate on examples about dogs and "dog." ([2] 60) Cats like Felix are probably equally bad if not worse. According to such rationalists, "Nature (or at least nature as perceived by the human organism) comes carved up into objects of various kinds, and the role of language is to give them labels and to state facts about them. A language can be arbitrary in using the words `dog' and `on' or `chien' and `sur,' but it is constrained by the nature of the world to group a certain set of objects or properties together under whichever names it uses." ([2] 61). I think this is looking at things a little topsy-turvey. I can't imagine what would `constrain' languages to have a word like the English for "dog." A language could be perfectly coherent, and have no word at all that applied to dogs, or have just one word for dogs and cats and other things. How about a language that grouped dogs and odd numbers together as one sort of thing, and cats and even numbers as another? It's not as incoherent as it might sound---consider the way gender works in many languages. All the realist needs is the recognition that `some' groupings work better for `some' purposes than others, `partly' because of facts that are independent of language. That is, what is at issue is not whether there are facts that any language, of any culture, serving any purpose must recognize, but whether such facts help explain the utility of those languages that recognize them for certain purposes for certain cultures. Any higher standard is simply unfair to facts. I think this sort of unfairness is ubiquitous in [2]. A recurring pattern of argument is to jump from the point that the success or breakdown of discourse cannot be explained by facts independent of language `alone', to the conclusion that the latter are not needed `at all', (and that those who believe in them are "naive"). I now turn to one example of the sort of argument in question. The Case of the Refrigerator and the Smart Aleck Perhaps my example about Felix is too simple. Let's consider one from Winograd and Flores, and the argument they base on it. It is impossible to establish a context-independent basis for circumscribing the literal use of a term even as seemingly simple as "water," as shown by the following dialog: A: Is there any water in the refrigerator? B: Yes C: Where? I don't see it. B: In the cells of the eggplant. ... In making the statement, "There's some water in the refrigerator," a person is not stating an objective fact. Every speech act occurs in a context, with a background shared by speaker and hearer. The "felicity conditions" depend on mutual knowledge and intentions. Here we have a fact about a bit of discourse. B's answer "Yes" to A's question, "Is there any water in the refrigerator?" was, Winograd says, "infelicitous." He indicates that there might be some disagreement as to whether it was true or not, and that most semantic theories would defend B on this point. I don't know about that, but the point I want to make doesn't depend on the concept of truth. Let's just say that in the imagined circumstances, B's answer was unhelpful. A wanted to get a glass of cold water to drink. B was just being a smart aleck. I take it that we all agree that our tools for the analysis of language needs to make some sense of this fact about B's remark. I take the last part of the quotation from Winograd and Flores to indicate that they think we do not need to cite any objective facts to account for the infelicity of B's reply. I think this is wrong. Let's first look at the argument. Winograd and Flores say: Most semantic theories in the rationalistic tradition provide formal grounds to support B, but a theory of language as human phenomenon needs to deal with the grounds for A's complaint as well---i.e., with the "infelicity" of B's reply. One surely must agree with the dependent clause here. So the dialectic is to see if a semantic theory in the rationalistic tradition can deal with the grounds for A's complaint. Now here we must be careful. "The rationalistic tradition" is a technical term in [2]. It stands for a list of assumptions that Winograd and Flores wish to replace. But, in general, it is not valid to argue from the failure of a set of assumptions to the failure of any one of them. This is particularly so when the list of assumptions have at best a rather impressionistic unity, as in the present case. So the failure to find a semantic theory that accounts for the infelicity of B's remark, while continuing to embrace all aspects of "the rationalistic tradition," would show very little. In particular, the fact that commitment to objective facts is a part of the rationalistic tradition, combined with the failure of a theory embracing all the tenets of that tradition to deal with the infelicity of B's remark, would show precisely nothing about whether dealing with this infelicity requires the postulation of objective facts. At any rate, Winograd and Flores continue by considering whether the definition of "water" could be expanded so as to make B's remark untrue. This way of handling the infelicity would be acceptable within the rationalistic tradition. But it will not work. For example, if we defined water as, "water in its liquid phase in sufficient quantity to act as a fluid," then we couldn't handle the following possible responses of B: 1. B: Yes, condensed on the bottom of the cooling coils. 2. B: There's no water in the refrigerator, but there's some lemonade. 3. B: Yes, there's a bottle of water in the refrigerator, with a little lemon in it to cover up the taste of the rust from the pipes. Winograd and Flores say that response 1 might be appropriate if person A were checking for sources of humidity that ruined some photographic plates being stored in the refrigerator. Similarly, the difference between the felicity and infelicity of 2 and 3 depends on cultural issues about whether having lemon suspended in it disqualifies something as being water. Then they say, We cannot come up with a situation-independent definition of what qualifies as water, since after any amount of fiddling with the definition, one can always come up with a new context in which it is inadequate. In making the statement "There's some water in the refrigerator," a person is not stating an objective fact. Now this all seems to me to be a complete non sequitur. What seems to follow, from the evidence adduced, is that which objective facts one is stating, depends not only on the words used and their meaning in the language, but cultural factors, and facts about the intentions of the person to whom one is talking, and the like. That is, Winograd and Flores show that factors `in addition to the language used and its meaning or meanings' are relevant to the felicity or infelicity of B's utterance. And they show that, among those factors, are cultural factors and specific facts about the intentions and desires of the conversants. That is all fine. But it does not show that the felicity or infelicity does not depend at least in part on facts that are independent of language. Now the method Winograd and Flores use, to show that cultural facts and facts about the conversants are relevant, is just an application of the method we used above. They note that you keep B's words and their meaning constant, and vary the cultural facts and the intentions and needs of the conversants, and different of B's responses come out felicitous and infelicitous. That's fine. Good, sound methodology. But exactly the same procedure will lead to the conclusion that objective facts are also among the parameters of infelicity. For why do we suppose that B's original remark, given the original desires of A, was infelicitous? It seems to me it is because we suppose, as is clearly implied (although never explicitly stated by Winograd and Flores), that there isn't, say, a bottle of water in the refrigerator. (Or at least a "bottle" of "water" in the "refrigerator.") And why do we suppose that B's response would be felicitous, if A's intention had instead been to discover the source of the humidity that ruined his photographic plates? It seems to me that it is because we suppose that it is an objective fact, that there is water condensed on the bottom of the cooling coils. Are these facts objective? That is, are they independent of discourse? Are they facts because certain nonlinguistic objects have certain nonlinguistic properties? It seems to me that the answer to these questions is clearly yes, once one has distinguished the question of the objectivity of these facts, from the question whether their relevance to the felicity of the utterance is determined only by language, or also by intentional and cultural factors. Have I Begged the Question, by Using Words to Criticize? Perhaps there are those who think otherwise. How can you say that these facts are objective, they might ask, when in identifying these putatively objective facts, you have yourself used words and categories that reflect your interests, cultural background, and the like? It is a very old idea, and one that I like a lot, and one that was emphasized in [3], by the way, that there are different ways of breaking up reality into objects, properties and relations. Different organisms, different cultures, different languages might break the same reality into different chunks to serve as objects, and single out different uniformities to serve as properties and relations, totally ignoring uniformities that are crystallized in some other form of life or language. It is a further very old idea that these schemes of individuation and classification might reflect facts about the nature of the organism, culture, or language as well as independently existing facts about reality. Locke, for example, advanced all of these ideas. One need only note his doctrine of secondary qualities, and his (extremely controversial) claim that our notion of a person was basically a forensic (i.e., legal) notion. Further, it seems plausible that all thought and assertion is mediated by such schemes of classification and individuation. Therefore, it is certainly correct that in saying that one of the factors relevant to the felicity or infelicity of B's reply, was whether or not there was a bottle of water in the refrigerator (in the first case), or water on the coils of the refrigerator (in the second), I was using a scheme of individuation and classification, that reflects facts about me, my language and culture, as well as facts about the external world. But so what? The point being made is that to analyze my discourse, as well as the discourse I analyze, factors `in addition to' language-independent facts are needed. But that is beside the point. For the fact that other facts ARE needed, does not show that language-independent facts are NOT needed. The criticism simply repeats the same basic fallacious argument one level up. We need to make a distinction here analogous to the use-mention distinction. It is a fact about the sun that it is hot. It is a fact about English and the sun that the sentence, "the sun is hot," given its meaning in English, is true. My first statement uses English, to say something about the sun. My second uses English to say something about English and the sun. To confuse the two is to confuse use and mention. The second fact would not exist without persons, culture, English, and the like. The first would. I think it is fairly obvious that persons, culture, and English would not exist if the sun was not hot, and will certainly cease to exist when it cools down, if they are not already long gone. Similarly, the fact that there is no bottle of water in the refrigerator is a fact about the refrigerator. If a plague develops and all life ends on earth, there still won't be a bottle of water in the refrigerator. The fact that, in stating and thinking about the refrigerator, I use a scheme of individuation that takes refrigerators as things, bottles as things, water as a liquid, liquid as a kind of substance, and the like, and that, relative to this mode of individuation and classification, the state of affairs of there being a bottle in the refrigerator with water in it isn't a fact, is a complex fact, about that mode of individuation classification, me, and the refrigerator. These two facts need not, and should not be confused. I have a feeling that these last two paragraphs are going to be unconvincing to many. Somehow the fact that in describing language-independent facts, one uses language, strikes some people as conceding the issue right there. One is reminded of a passage from Berkeley [4]. Philonous believes that `nothing exists outside of the mind'. He believes that this doctrine really doesn't go against common sense, only against what a tradition of mixed-up philosophers have said. Throughout [4] Philonous puts forward a variety of clever arguments for his view. But at one point, he thinks he has an argument that should settle the things for once and for all: Philonous: But (to pass by all that hath been hitherto said, and reckon it for nothing, if you will have it so) I am content to put the whole upon this issue. If you can conceive it possible for any mixture or combination of qualities, or any sensible object whatever, to exist without the mind, then I will grant it actually to be so. Hylas: If it comes to that, the point will soon be decided. What more easy than to conceive a tree or house existing by itself, independent of, and unperceived by, any mind whatsoever? I do at this present time conceive them existing after that manner. Philonous: How say you, Hylas, can you see a thing which is at the same time unseen? Hylas: No, that were a contradiction. Philonous: Is it not as great a contradiction to talk of `conceiving' a thing which is `unconceived'? Hylas: It is. Philonous: The tree or house therefore which you think of, is conceived by you. Hylas: How should it be otherwise? Philonous: And what is conceived is surely in the mind. Hylas: Without question, that which is conceived is in the mind. Philonous: How then came you to say, you conceived a house or tree existing independent and out of all minds whatsoever? Hylas: That was I own an oversight; but stay, let me consider what led me into it.---It is a pleasant mistake enough. As I was thinking of a tree in a solitary place, where no one was present to see it, methought that was to conceive a tree as existing unperceived or unthought of, not considering that I myself conceived it all the while. But now I plainly see, that all I can do is to frame ideas in my own mind. I may indeed conceive in my own thought the ideas of a tree, or a house, or a mountain, but that is all. And this is far from proving, that I can conceive them `existing out of the minds of all spirits'. Hylas is presented as a bit slow-witted in [4], but even he gets the point when Philonous spells it out to him. Indeed, elsewhere, referring to the same argument, Berkeley says it should demonstrate his point "in a line or two to anyone that is capable of the least reflection." ([5] 22) So, in admitting that I am quite unmoved by Philonous' argument, and moved even less by contemporary versions of it deployed in favor of the principle that nothing exists outside of language, I risk coming across as even slower-witted than poor Hylas. Enough risks for one day! I end by recommending the study of Berkeley to those tempted to be Winogradian antirealists. References [1] Winograd, T. 1985. Moving the Semantic Fulcrum. "Linguistics and Philosophy" 8 (no. 1): 91--104. [2] Winograd, T., and Flores, F. 1986. "Understanding Computers and Cognition: A New Foundation for Design". Norwood, NJ: Ablex. [3] Barwise, J., and Perry, J. 1983. "Situations and Attitudes". Cambridge, MA: Bradford Books/MIT Press. [4] Berkeley, G., Bishop of Cloyne. 1713. "Three Dialogues between Hylas and Philonous." London. [5] Berkeley, G., Bishop of Cloyne. 1710. "A Treatise Concerning the Principles of Human Knowledge." Dublin. ------------------ A WORKSHOP ON SEMANTICAL PARADOXES Jon Barwise and John Etchemendy In spite of all the work in semantics in recent years, there is still no consensus about one of the most fundamental foundational questions in semantics: how to understand the semantic paradoxes. Rather, most semantic theories of natural language simply avoid the problem by restricting attention to fragments of language that cannot express sentences, like the Liar sentence (`This is not true'), that give rise to paradox. Simple as such sentences are, they in fact amount to one of the most recalcitrant problems facing rigorous semantic theories of languages, like English, that contain their own truth predicate. Such paradoxes seem to allow the derivation of a contradiction from intuitively obvious principles that describe the behavior of the property of truth and its interaction with negation and reference. The two of us, as part of our work on the STASS project, have recently finished a book on the matter, "The Liar: An Essay on Truth and Circularity," to be published by Oxford University Press next month. In this book we argue that the paradox arises out of a shift in subject matter that takes place in our reasoning about the Liar's truth value, a shift captured in the Austinian conception of truth at the heart of situation semantics. Treating truth as a property of propositions, not sentences, we give mathematical models of two quite distinct conceptions of propositions; first the standard notion used by Bertrand Russell, among others, then one based on J. L. Austin's work on truth. By contrasting and comparing these two accounts, we claim to show that the standard Russellian conception of the relation between sentences, propositions, and truth is crucially flawed in limiting cases. The Austinian conception, on the other hand, can be seen as a refinement of the Russellian, one that avoids the paradox while providing a straightforward understanding of the semantical intuitions that gave rise to it. In order to exhibit this view to the scrutiny of those who had thought most seriously about the semantic paradoxes, STASS held a small, intensive workshop on the paradoxes at the nearby Buck Estate on 5--7 February. It was organized by the two of us along with the help of Tony Martin of UCLA. Each day was organized with three talks (two on the final day) followed by an extensive discussion session, a format that successfully encouraged fruitful discussion. The program was as follows: Apart from Barwise and Etchemendy, speakers included Peter Aczel (University of Manchester), Nicholas Asher (University of Texas at Austin), Nuel Belnap Jr. (University of Pittsburgh), Tyler Burge (University of California at Los Angeles), Haim Gaifman (Stanford University), Anil Gupta (University of Illinois at Chicago), David Israel (SRI International), Hans Kamp (University of Texas at Austin), Tony Martin (University of California at Los Angeles), Peter Woodruff (University of California at Irvine), and Stephen Yablo (University of Michigan, Ann Arbor). By the end of the meeting, a certain pattern emerged in the work presented, or so it seemed to us. Accounts of the Liar can be ranked on two orthogonal dimensions. On one dimension, we find the degree to which an account tries to pinpoint the difficulty on some sort of shift in context or not. The accounts of Martin, Gupta, and Yablo, for example, do not see this as the problem, nor does the Russellian account offered by us in our book. On the other hand, Burge and Gaifman both see the problem as one involving a shift in context, as does our preferred, Austinian account. The other dimension along which one can judge an account has to do with what one might call a bottom-up or a top-down approach. Bottom-up approaches view the problem as one of coming up with a satisfactory inductive definition of truth. Top-down approaches do not attempt to define truth, but rather view the problem as one of investigating consequences of some very basic intuitions about the truth, negation, and reference. Kripke's 1975 theory is usually viewed as the former, but is in fact the latter, as are both of our Russellian and Austinian treatments. By contrast, the accounts of Martin, Gaifman, and Burge are bottom-up accounts. Yablo's account is an interesting mixture of the two. While there was no more consensus than one would have predicted, the meeting left us all feeling that this is an exciting, timely area of research. Difficult problems surrounding the semantic paradoxes are leading to important new mathematical and philosophical ideas. We hope to assemble a CSLI Lecture Notes volume in which the various participants at the meeting give short overviews of their theories. ------------------ A NEW CSLI WORKING GROUP A new CSLI working group, Problems in Authentic Discourse, began this month under the leadership of Herbert Clark, a CSLI researcher and Professor in the Psychology Department. Clark stated the purpose of the group as follows: Those of us at CSLI who study natural language ordinarily work from sanitized examples of language we have thought up to test this or that point of theory. We rarely put our ideas to test on language that has actually occurred in conversation, extemporaneous narratives, or even literature. With the emphasis on situated language at CSLI, it is especially important that we do this. The situations in which language is actually used are likely to have many features we cannot discover through traditional intuitive methods. The proposal is to begin with segments of authentic discourse and examine the problems they raise for the theories and models we are currently working on. The group will begin by testing different accounts of personal pronouns on occurrences in a spontaneous story introduced into a conversation between two academics. ------------------ POSITION AVAILABLE IN THE SYMBOLIC SYSTEMS PROGRAM Position: Program Coordinator, Symbolic Systems Undergraduate Program (half-time, replacement position, 8/1/87--6/30/88) Job description: The position involves a wide variety of activities from academic planning and advising of undergraduates to the administrative management of the program office. These activities include: Academic responsibilities: Advise undergraduate majors, assist in curriculum design and planning. Coordinate faculty committees. Liaison with affiliated departments. (Supervise Honors Students, teach course---depending on academic qualifications.) Administrative responsibilities: Budget preparation and control, supervise clerical staff (one), write copy for university publications. Qualifications: Knowledge of subject area. Organizational skills, effective oral and written communication, ability to work independently. Some administrative experience, familiarity with university environment. To apply: Send copy of resume with cover letter briefly describing work experience and academic background relevant to the position, plus two or three letters of recommendation covering both academic and administrative abilities to: Yvette Klemm, Building 1, Stanford University, Stanford, CA 94305-2070. ------------------ CSLI VISITING SCHOLARS There are no new CSLI visitors (see last month's Monthly); however, Gordon Plotkin and Marilyn Ford have both left. ------------------ CSLI PUBLICATIONS CSLI Reports may be obtained by writing to Trudy Vizmanos, CSLI, Ventura Hall, Stanford, CA 94305--4115. ------------------ OVERHEARD Every once in a while I have to say something to remind myself of why I usually don't. -------------------------------------------------------------------- Editor's note Selected commentary about Monthly articles or other matters will be published in future issues. Please send correspondence to the Editor of the Monthly at CSLI or by electronic mail to Monthly-Editor@csli.stanford.edu. --------------------------------------------------------------------- - Elizabeth Macken Editor