Sellars' Semantics

  • 72 41 10
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up

Sellars' Semantics

Gilbert Harman The Philosophical Review, Vol. 79, No. 3. (Jul., 1970), pp. 404-419. Stable URL:

844 190 274KB

Pages 17 Page size 595 x 792 pts Year 2007

Report DMCA / Copyright


Recommend Papers

File loading please wait...
Citation preview

Sellars' Semantics Gilbert Harman The Philosophical Review, Vol. 79, No. 3. (Jul., 1970), pp. 404-419. Stable URL: The Philosophical Review is currently published by Cornell University.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

JSTOR is an independent not-for-profit organization dedicated to and preserving a digital archive of scholarly journals. For more information regarding JSTOR, please contact [email protected]. Sat May 12 00:33:04 2007



CIENCE AND METAPHYSICS is a major work which deserves

to be, and will be, read and reread. I t is also a difficult work, particularly so in its opening chapters. The reader may wish to begin with Chapter 111, saving Chapters I and I1 for last. In this review, I shall ignore much in the book in order to concentrate on what seems to me its most important and controversial theses. These concern the nature of psychological states, meaning, and truth, and are discussed in Chapters 111-VI. I.

Meaning and Mind.

How is the way that thought represents related to the way that speech represents? One familiar answer is this: "Speech is the linguistic communication of thought. The sentence 'Sellars is wise' (S) can be used to represent that Sellars is wise only because the semantic and syntactic rules of English correlate S with the thought that Sellars is wise. Representation in speech derives from representation in thought." This answer does not account for linguistic representation in those cases in which, as we say, one "thinks out loud." I n such cases one does not use words to communicate one's thoughts. Indeed, it is difficult in such cases to distinguish the "thought" from its expression in speech. I n order to account for such cases, two strategies suggest themselves. One treats representation in thinking-out-loud as derivative from representation in linguistic communication, which in turn derives from representation in thought. A second strategy treats representation in thinking-out-loud on a par with representation in thinking that is not "out loud." Sellars' theory of meaning is an elaboration of this second strategy. Since all thought is representation, thought requires a system of representation which one might call the "language of thought." An occurrence (or state) is the thought (belief, desire, hope, fear, and so forth) for example, that Sellars is wise only if it represents that Sellars is wise. Therefore, a theory of the nature of psychological Science and Metaphysics: Variations on Kantian Themes. By Wilfrid Sellars. (New York, Humanities Press, 1968. Pp. 246. $5.25.)



states must include a theory of meaning for the language of thought. This theory may be considered a theory of meaning or a theory of the nature of psychological states, or both. It is the theory that meaning is role in one's conceptual scheme and the theory that psychological states are functional states. We are encouraged to view the language of thought by analogy with a game like chess. One's total psychological state at any moment represents his position at that moment. Certain moves are appropriate responses to observation, as when one accepts something as the result of observation. Other moves result in action, as when one does something because one has decided to do it. Still other moves are available, as in inference. The meaning of an expression in the language of thought is determined by the role it can play in the evidence-inference-action language game; the nature of a particular psychological state or occurrence is determined by its functional relations to the other states or occurrences that make up one's total psychological state (as affected by observation and inference and as leading to action). This view can be defended without appeal to thinking-out-loud; and it would continue to seem plausible even if one were to decide that thinking-out-loud is not really thinking at all but is a kind of rehearsal that derives from a more normal use of language in communication. The phenomenon of thinking-out-loud does, however, play a major role in Sellars' presentation of this theory, and I suspect that he is thereby led into error. Let me elaborate. Sellars advances his theory with an eye on the philosophical problem of our knowledge of other minds. His own solution to this problem treats the hypothesis that people have thoughts, beliefs, desires, hopes, fears, and so forth as a theory postulated to account for human behavior. Furthermore, he claims that theoretical postulation always involves the introduction of new theoretical predicates by analogy with predicates previously available. I n the present instance, the relevant properties have to do with representation, and the relevant analogy is with language. This cannot be language as used in communication, since anyone who has the concept of the communication of thought must already have the relevant psychological concepts. Therefore, we are to imagine a community whose inhabitants use language only to express their thoughts in thinking-out-loud, intending-out-loud, and so forth. The members of this community have no inkling of psychological occurrences or states but are aware of episodes of (what we call) thinking-out-loud, which they take' to have a representational character. Thus they take expressions of their language to have a meaning


in the sense of role in the language game of thinking-out-loud. Equivalently, they are able to give low-level explanations of human behavior in terms of observed regularities involving observation, thinking-outloud, and action. One day, a great genius, Jones, postulates the existence of inner occurrences and states that are analogous to occurrences of and dispositions to think-out-loud. These inner occurrences and states are taken to be representational in the same sense in which occurrences of thinking-out-loud were taken to be representational. In other words, Jones postulates an inner language game of thought analogous to the outer language game of thinking-out-loud. In doing so he hopes to be able to explain all behavior that could be explained on the older theory as well as other behavior that could not be previously explained. Indeed, the theory works so well that it is taught to the other members of the community. Jones then discovers that, for certain thoughts, beliefs, and so forth, one can be trained so that, when he thinks, believes, and so forth that something is the case, he will be disposed to think that he thinks, believes, and so forth that it is the case. So Jones adds this fact to his theory. Sellars does not offer this story as historical truth; nor does he suggest that it represents part of the development of a child's conceptual scheme. He purports, instead, to "reconstruct" part of our ordinary conceptual scheme, comparing his enterprise with that of contract theorists in political philosophy. Presumably Sellars has in mind the plight of a philosopher who has become skeptical about the very intelligibility of ordinary talk about mental states and occurrences. To bring such a philosopher back to sanity, Sellars offers the above story to show how our ordinary conceptual framework could have arisen in a perfectly rational and intelligible way. Sellars can then say that we talk exactly as descendants of the community in his story would talk and that our talk is therefore just as intelligible as theirs would be. I think that Sellars' general approach to this sort of skepticism is correct; but I also feel that doubts can be raised about his heavy reliance on the phenomenon of thinking-out-loud. I t is not obvious that thinking-out-loud is a kind of thinking. It seems equally plausible that thinking-out-loud should be a kind of play-acting in which one, as it were, pretends to be talking to someone. If the latter supposition is correct, then thinking-out-loyd will not have the connections with observation and action needed for Sellars' argument. Furthermore, even if thinking-out-loud is a kind of thinking, there

SELLARS' SEMANTICS is not enough of it. One could not begin to account for human action in terms of observation and particular instances of thinking-out-loud-or, to make the same point in a different way, one cannot get a theory of meaning for sentences uttered when one thinks-out-loud without mentioning unexpressed psychological states and occurrences. It is true that Sellars has not suggested inferring a psychological theory from the present meager supply of instances of thinking-outloud. It would be enough for his purposes if we could imagine a community in which thinking-out-loud were common enough to allow the sort of moves he has in mind. But he underestimates what is involved in imagining such a community. How a person responds to observation, what inferences he makes, how he acts, and so forth depend on his entire psychological state. There are no simple correlations between particular occurrent thoughts, observations, and/or actions. And the relevant states far outstrip any dispositions one might have to think certain thoughts. So presumably we must imagine that the inhabitants of Sellars' Rylean community express aloud a great deal of their psychological occurrences and states. This means that we must not only envision them constantly saying things-in general, many things at the same time-but we must also envision, for example, that a rough and ever-changing list of their beliefs, desires, hopes, fears, and so forth is attached to them (perhaps as tattoos). My point is that, if we really must imagine all this in order to make Sellars' reconstruction work, that particular reconstruction no longer seems the best way to answer the skeptic. Why not instead point out that we can (partially) explain human behavior if we postulate the existence of states and occurrences functionally related to each other, to observation and to action? One would develop a n account of representation for these states and occurrences as a n integral part of one's theory of human behavior. This might conflict with Sellars' view that theories must introduce new predicates that are analogous to old predicates, since the representational nature of psychological states and occurrences would not on the revised account be analogous to anything that went before. If so, that is a problem for Sellars' theory of theories. He needs to assume that one can come to conceive instances of thinking-out-loud as representational without having previous views about representation; and it is difficult for me to see why one should suppose it harder thus to conceive theoretically postulated states and occurrences. I think that Sellars' general theory of meaning is correct: meaning,


or at least one kind of meaning, is role in one's conceptual schemethat is, role in the evidence-inference-action language game of thought. I also approve of Sellars' general method of hypothetical theory construction for resolving the doubts of a philosopher who has become worried with the thought that psychological concepts are meaningless, although I a m less confident about this than about the theory of meaning. O n the other hand, I am very doubtful about the details of Sellars' particular example of hypothetical theory construction, with its emphasis on thinking-out-loud. Furthermore, this emphasis on thinking-out-loud can lead to an extremely misleading picture of thought as talking to oneself. I n particular, it suggests that when one thinks in language the words must "pass through one's head," something which need not happen. Therefore I want to emphasize the independence of Sellars' general theory of meaning from his epistemological theses. We can have the former without the latter. 2.

Semantic Regularities.

With these preliminaries out of the way, we can turn our attention to certain details of Sellars' theory. It treats mental states and occurrences as representations that are functionally related to each other. One's beliefs represent the world as one believes it is; one's desires represent the world as one desires it to be; and so forth. Observation affects one's beliefs; beliefs and desires are modified in inference and lead to action; and so forth. Sellars speaks here of various "semantic regularities" of which there are three basic sorts: observationthought regularities, thought-thought regularities, and thoughtaction regularities. (He calls them world-language, language-language, and language-world regularities, where the relevant language is the language of thought.) These regularities are semantic in the sense that their existence is what gives various mental states and occurrences whatever representational character they have. We are not to think of the regularities as deterministic laws of the form "Whenever one observes so-and-so, one comes to believe such-and-such" or "Whenever one thinks this and also thinks that, one infers thus." Semantic regularities are, roughly speaking, statements of what it is possible for one to do and escape intellectual criticism (from oneself or others). Thus rules of inference do not tell one what under certain conditions he must infer but rather what he may infer. A problem arises here. If one believes that there are deductive and inductive inferences, one is likely to conclude that deductive and


inductive rules of inference are semantical rules in Sellars' sense. That may be so; but what rules are they? Some things that Sellars says suggest rnodus ponens would become the principle that: if one believes P and also believes If P, then Q, one may believe Q . And it can seem plausible that the meaning of logical terms such as "if" and "all" should be a function of rules of inference, so construed. A difficulty here is that there is no such thing as deductive inference. So-called deductive rules of inference are nothing of the sort. They are simply transformations which take one or more statements into another statement so that if the original statements are all true the resulting statement is also true. Modus ponens does not tell one what to believe. I t is not the case that, if one believes P and also believes If P, then Q, one may always believe Q, for perhaps one should give up P or IfP, then Q . In deductive "logic," a contradiction entails anything. That does not mean that, if some of one's beliefs are contradictory, one is justified in believing anything. It is a pun to speak of inductive and deductive rules of inference in the same breath. If rules of inference correspond to legitimate thought-thought connections, there are no deductive rules of inference. Perhaps there are no rules of inference at all. One simply tries to maximize "coherence" by making the least change necessary in one's antecedent beliefs, desires, and so forth. A different approach would associate rnodus ponens with the principle that one should not simultaneously believe P, P, then 0,and not-Q, and similarly for other so-called deductive rules of inference. Furthermore, a law of nature such as lightning is alwaysfollowed by thunder could be said to correspond to the principle that one should not both believe that at some time lightning occurred and also believe that no thunder occurred shortly after that time. Sellars suggests just such an analysis for this law of nature (V, 5).2 SO this way of understanding his view may be truer to his intentions than the one deplored above. Presumably, the basic principle on this second approach would be that one should not believe things that cannot all be true. Logic or sciences tell us that certain things cannot be true together and we thereby learn that we ought not to believe all of them. This suggests a connection between the theory of meaning and the theory of truth, where before we were envisioning a connection between the theory of meaning and what we might call the theory of evidence. Thus the theory of meaning is concerned with semantic regularities. These are "n this and subsequent references, the roman numeral refers to the chapter, the arabic to the paragraph.


normative principles telling one what one ought and ought not to believe: one ought to believe what is true and one ought not to believe what is not true. There are, however, two different ways to answer the question "What should one believe?" In one sense one should believe what is true. I n another sense one should believe what one has good reason to think is true. Thus one might answer the question by providing either a theory of truth or a theory of evidence. Either sort of theory can be taken to be a theory of meaning, for one can plausibly maintain (at different times, in different contexts) both that meaning is determined by truth conditions and that meaning is determined by evidential role in one's conceptual scheme. This is not to say that theories of truth and evidence can be identified with each other. Something can be true without being evident; something can be evident without being true. I now want to consider the following puzzle about Sellars' view. Sometimes he speaks of semantic rules as if they were principles of the theory of truth. At other times he speaks of them as if they were principles of the theory of evidence. The puzzle is whether he has confused two different sorts of principle or has rather found a way to exploit certain connections between the two sorts of principle. I think he has tried to do the latter; but I am not convinced that he has succeeded. Sellars' definition of truth is this: for a proposition to be true is for it to be . . . correctly assertable . . . in accordance with the relevant semantical rules, and on the basis of such additional, though unspecified, information as these rules may require. . . "True" then means semantically assertable .. . and the varieties of truth correspond to the relevant varieties of semantical rule [IV, 261.


This can work only if "semantically assertable" means truly assertable (in the language of thought) rather than justifiably assertable, and the semantical rules are principles of the theory of truth rather than principles of the theory of evidence. Other things that Sellars says, however, make it seem that semantical rules are principles of the theory of evidence after all. For example, there are supposed to be semantical rules connecting observation and thought. Why bring in observation if these rules are rules of truth rather than rules of evidence ? Everything depends on the exact nature of the rules connecting observation and thought. Evidential rules would correlate stimulations


(or how it looks to one) with specific thoughts. Truth rules might correlate the actual (and not just apparent) observation of something with a thought of that thing. Presumably Sellars has something like the latter sort of rule in mind (cf. V, 36, 56). But whether he can thus escape the charge of confusing truth and evidence is a question that can be answered only after some consideration of his complex and interesting theory of truth.

3. Truth. Roughly speaking Sellars' theory of truth is a synthesis of the correspondence, coherence, and pragmatic theories of truth. He holds that in one sense truth is always truth in a particular language or conceptual scheme. Certain representations are true in a particular conceptual scheme C if and only if they correctly picture particular facts according to rules of projection in C. Other representations are true not because they directly picture the facts but because of the way in which they are related to representations that do picture the facts. Sellars also claims that conceptual schemes may be more or less adequate in their picturing resources and that it is the job of science to find better and better theories, which is to say more and more adequate conceptual schemes. Ideally, science aims at theory that would provide completely adequate pictures. Such a theory would give us the ultimate truth. So on Sellars' view we can distinguish three sorts of truth: truth as adequate picturing of facts in C; truth as coherence with the facts in C; and truth as C's degree of adequacy in providing accurate pictures of facts compared with that of other conceptual schemes. For example, according to a rather simple atomic physics, the basic physical facts would be facts about atoms, their size, shape, temporalspatial position, spinning, and so forth. Now consider all possible representations (in some system of representation) of one or more atoms of this or that shape, spatially-temporally related in such and such ways, spinning one way or another, and so forth. We may think of all such representations as the atomic sentences of the language of the theory. They are true if and only if they correctly picture the world-that is, if and only if they directly correspond to the atomic facts. Furthermore, in a n important sense these atomic facts are all the facts there are. There are no facts "over and above" the atomic facts. The atomic facts completely determine all other facts. Thus the truth of all nonatomic representations is determined by the set of true atomic representations. The point is obvious for truth-


functional compounds and universal and existential quantifications. Consider also the way in which the truth of representations of macroobjects is a function of the truth of atomic representations. Given the sort of atomism here envisioned, one could not imagine two possible universes such that the set of true atomic representations is the same in both universes, although it is true in one universe but not in the other that there is a table in such and such a place. That would not be possible. The truth of representations of macro-objects is completely determined by the atomic facts. (This does not mean that representations of macro-objects are translatable into atomic representations or even into quantificational and truth-functional compounds of atomic representations.) Now consider lawlike statements about atoms. Should one say that there are facts about laws over and above facts about the position and so forth of the atoms? This becomes the question whether there could be two universes that contained the same atoms in the same spatiotemporal positions and so forth, but that had different laws. In the two universes, the same atomic representations, truth-functional compounds, and quantifications would be true, but different contrary-to-fact conditionals would hold. Can we envision two different universes of this sort? I take it that Sellars thinks not. And if not, the laws are determined by the atomic facts. Perhaps the set of laws is the "simplest" or "most coherent" set that accounts for all the true atomic representations, which is of course not to say that the set of laws is the simplest or most coherent set that accounts for all the evidence. Thus, with respect to the atomic theory just considered, one seems to be able to make sense of the idea that certain representations are true because they correspond to or "picture" the facts, whereas others are true because of their relationships to representations that picture the facts. Sellars holds that in general both sorts of truth are relevant to any conceptual scheme, since he holds that any such scheme distinguishes atomic representations from others; and atomic representaions are true if and only if they correctly picture the facts, whereas the truth of other representations is completely determined by the set of true atomic representations. Sellars contrasts this much of his theory with what he calls "TarskiCarnap semantics." The idea behind the latter approach is that a theory of truth ought to offer a recursive characterization of denotation, satisfaction, and truth which meets Tarski's condition T. Consider this formula: x is true if and only if@.


Tarski's condition requires that one's theory of truth should entail, for every sentence of the language in question, an instance of the abovedisplayed formula with a name of that sentence replacing x and with a translation of that sentence replacing p. These entailments give the truth conditions for every sentence of the language. Thus, "Snow is white" is true if and only if snow is white; " 2 2 = 4" is true if and only if 2 2 = 4; "Stealing is wrong" is true if and only if stealing is wrong; and so forth. Now it is sometimes said that, because of these entailments, Tarski's theory of truth is a version of the correspondence theory. Sellars offers two reasons for rejecting that suggestion. First, since relevant entailments hold for every sentence in the language, this way of interpreting Tarski's theory yields the claim that all truth is correspondence. But Sellars wants to say that only some representations are true because they correspond to (or picture) the facts and that others are true, not because they directly correspond to the facts, but because of their relationships to other representations that do directly correspond to the facts. Because it attempts to offer a recursive characterization of truth, however, Tarski's theory can also be interpreted as providing a correspondence theory somewhat similar to Sellars' picture theory of truth. Things are complicated by the fact that Tarski envisions a recursive characterization of satisfaction as well as of truth where Sellars envisions a nonrecursive characterization of truth apart from satisfaction. Still, in a recursive characterization of satisfaction and truth, some expressions will be said to be satisfied by certain sequences or be true if and only if certain conditions obtain, where these conditions do not refer to satisfaction or truth; and other expressions will be said to be satisfied by certain sequences or be true if and only if certain conditions obtain, where these conditions do involve the satisfaction or truth of other expressions. Obviously one can say that the former expressions are satisfied or are true because they directly correspond to (or picture) the facts, whereas the latter expressions are satisfied or are true only because of their relationships to other expressions. This, however, brings us to Sellars' second reason for denying that "Tarski-Carnap semantics" provides any sort of correspondence theory of truth. Sellars claims (in effect) that the relation of satisfaction, which plays an essential role in Tarski's account of truth for quantified sentences, is not really a semantic relation at all. That is, he denies that satisfaction is a relation between language and the world. Now it seems to me that Sellars must be mistaken here, possibly because he




has not fully appreciated Tarski's theory. I shall sketch his argument and say what I think is wrong with it. Strictly speaking, one should say that an open sentence is satisfied by a sequence of objects. For our purposes, however, we may think of a predicate expression as satisfied by certain objects when it is true of them. Thus, "is wise" is satisfied by Tarski. Now Sellars says, in effect, that what I have just said in the preceding sentence only appears to assert a relation between language and the world. He argues that my claim really asserts a relation between the expressions "is wise" and "Tarski," and he suggests that it would be more accurate to write it in any of the following (for him equivalent) ways: "Is wise" is satisfied by "Tarski." "IS wise" is true of "Tarski." "Tarski"-"is wise" is true. "Tarski is wise" is true. He concludes that, since on his analysis satisfaction is not a relation between language and the world and since "Tarski-Carnap semantics" must make essential use of satisfaction (or some equivalent notion), "Tarski-Carnap semantics" does not show how language is related to the world. Notice that Sellars' analysis works only if every object that satisfies any expression has a name; and recall that Tarski introduced the notion of satisfaction in order to give truth conditions for quantified sentences. He would not have had to do so if he had been able to assume that all objects in the range of a quantifier have names in the language, for if they do, one can make do with so-called substitutional quantification. And indeed, Sellars' version of quantification is the substitutional variety. Tarski did not want to assume, however, that all objects in the range of the quantifiers have names and he therefore had to introduce a notion of satisfaction that of necessity cannot be analyzed in the way that Sellars suggests. So Sellars is just wrong when he says that the relation of satisfaction in "TarskiCarnap semantics" does not relate language and the world. O n the contrary, it works only because it does relate language and the world. So far, I have explained the sense in which Sellars defends a correspondence theory of truth for certain representations and a coherence theory for others; and I have ,pointed out similarities between this part of Sellars' theory and "Tarski-Carnap semantics." I should now mention some differences. First, since Tarski makes use of the notion of


satisfaction, he is able to give truth conditions for referentially quantified sentences, whereas Sellars, who does not make use of the notion of satisfaction, will be able to give truth conditions only for substitutionally quantified sentences. Now, Sellars claims that substitutional quantification gives a better account of generalizations expressed in ordinary language than referential quantification does. For one thing, it allows quantification of the predicate without commitment to universals. So perhaps no real objection to Sellars' theory arises here. Second, Tarski requires that a truth characterization satisfy condition T, referred to above. Sellars does not require this; and if I understand his analysis of laws, the characterization of truth for laws will not by itself entail all the necessary instances of the formula " x is true if and only if@," where x is replaced by the name of a law and p is replaced by a translation of that law into the language in which one states the truth characterization. Since, however, there is no other generally accepted analysis of laws that satisfies condition I, this cannot be held against Sellars' analysis, which on other counts seems rather plausible. Sellars does pretend to explain why Carnap is wise if and only if it is true that Carnap is wise. He points out that "It is true that Carnap is wise" means that "Carnap is wise" is (truly) assertable. (I suppress complications involving Sellars' use of dot-quotes.) Therefore, consider the following sequence of assertions (or thoughts): "It is true that Carnap is wise. (So) Carnap is wise." The first of these assertions authorizes the assertion of the second. That, according to Sellars, explains why we must agree that Carnap is wise if it is true that Carnap is wise. But even if that much of Sellars' argument could be accepted, he offers no further argument to show why we must agree that Carnap is wise only i f i t is true that Carnap is wise. So he does not offer any account of why Carnap is wise ifand only ifit is true that Carnap is wise. That concludes my discussion of Sellars' reconstruction of the correspondence and coherence theories of truth and his discussion of "Tarski-Carnap semantics." I now want to say something about Sellars' version of the pragmatic theory. He argues that whole conceptual schemes may be more or less adequate, more or less true, in the sense that they may provide more or less adequate pictures. That is, from the point of view of some conceptual scheme that science leads us to accept, we may find that there are no objects such as those pictured in an earlier, more conimonsensical scheme. The atomic sentences of our earlier view describe the familar tables and chairs


around us. Let us assume that science tells us that only atoms existthat is, let us suppose that the "atomic" sentences of science picture atoms and do not picture any tables and chairs. If there are only atoms, there are no tables or chairs, since a table is not an atom. A table is not a set or even a sum of atoms. Perhaps it is a sum of appropriate time slices of atoms; but, by hypothesis, science tells us there are only atoms and no sets, sums, or slices. The basic particulars of the common-sense framework do not correspond to any basic particulars in the scientific framework. I n other words, the atomic sentences about tables in the common-sense framework do not correspond to atomic sentences in the scientific framework. Presumably they correspond to nonatomic sentences; and whereas the truth conditions for the sentences of the common-sense framework are a matter of whether or not these sentences directly picture the facts, the truth conditions for the corresponding sentences in the scientific framework are complicated and refer to other sentences of that framework. Thus, according to Sellars, there is a sense in which certain atomic representations of tables in the common-sense framework are true, since corresponding representations in the scientific framework are true. But in another sense, the atomic representations of tables in the common-sense framework are inadequate or not completely true, since they represent tables as basic particulars whereas the (truer) representations of science represent tables as constructions. Sellars claims that this vindicates the Kantian view that tables and chairs and so forth are only appearances and not really things in themselves. For Sellars, the things in themselves are whatever would be pictured in the ideally adequate scientific theory; appearances are basic particulars in the common-sense framework that are not basic particulars in the ideal scientific framework but survive as constructions in that framework. This vindication of Kant is ingenious and has a certain amount of intuitive appeal. I am, however, skeptical of the application of notions such as atomic sentence to actual languages, theories, or conceptual schemes apart from a particular way of formalizing that language, theory, or conceptual scheme. Theories can always be formalized in a number of different ways; and sentences that turn out to be atomic in one formalization can be nonatomic in another. Similarly, there will be various equivalent truth characterizations for a language (if there is one) such that on one characterization a sentence can turq out true because it directly corresponds to the facts but on another characterization the sentence can be true only because of its relationships to other sentences. I do not know what


could be meant by the claim that one way of formalizing the theory, or of giving the truth characterization for the language, is correct and others incorrect; but Sellars will have to defend some such claim if his theory of truth is to have the interesting philosophical implications that he wants it to have. The following problem seems especially serious. How can Sellars tell whether atoms are basic particulars of the common-sense conceptual framework? Ordinary size physical objects can count as such particulars; why not very small physical objects? We know what transparent glass spheres are like. If such colorless physical objects are basic particulars in our common-sense conceptual framework, why not extremely small colorless physical objects ? Why should the introduction of atomic theory be considered the introduction of a new framework of basic particulars ? A defender of Sellars might reply that, even if science does not add new basic particulars, it subtracts some of the old ones, since it tells us that only atoms exist. But how is one to decide whether science denies the status of basic particular to clusters, configurations, sets, time slices, sums, and even sums of time slices of atoms? Nothing that Sellars says seems to answer this question. T o be sure, there are scattered cryptic remarks that may seem to bear on the subject (for example, V, 1 0 1 ) . But cryptic remarks do not make a n argument; indeed they make one suspicious that no argument can be given. So much, then, for Sellars' theory of truth. I want now to return to the issue raised at the end of the preceding section: does Sellars confuse the theory of evidence with the theory of truth? I want to argue that, because Sellars identifies semantical rules both as the rules of truth and as the rules of the language game of thought, he is guilty of exactly this confusion. T o identify the rules of truth with those of the language game of thought is ( I ) to identify rules which indicate how the truth of certain representations depends on the truth of other representations with rules of "inference" that permit one to go from certain thoughts to others. It is also (2) to identify rules which indicate how the truth of certain representations depends on whether they correspond to or picture the facts with rules that permit one to have certain thoughts given certain observations. (Recall that the rules of the language of thought must be understood to tell one what is truly to be thought and not what one might be justified in thinking.) Sellars wishes, however (in V, 92), to deny that picturing is always associated with observationality, since he wishes to assert' that "singular statements in the language of microphysics . . . constitute pictures of microphysical


objects and events." He agrees that "no singular statement about microphysical particles can occur in a language entry transition, or observation"; but he claims that this is not an objection to the picturing of microparticles.

. .

This objection assumes . that statements which are basic as the constituents of pictures must also be epistemically basic in the sense that they formulate observable states of affairs. I t is, indeed, true of the common-sense framework that statements which are basic in one sense are also basic in the other. Yet the two senses of "basic" are different, and a transcendental philosophy which rises to a level of abstraction which distinguishes the generic character of epistemic concepts (e.g. language entry transition, conceptual picture, object) from the specific forms they take in common-sense discourse will not assume that the basic constituents of conceptual pictures must be statements of the kind which occur as conceptual responses to sensory stimulation.

But this also shows that a transcendental philosophy must not identifjr the rules of truth with the rules of the language game of thought, if these latter rules must include rules for "language entry transitions" expressing appropriate responses to observation. I can sum up my objection as follows. Any conceptual scheme must be representable as a language game of thought that includes, among other things, appropriate responses to observation. The rules of the game that determine which responses are in this sense appropriate are epistemic rules rather than truth rules. The game, its epistemic rules, and his position in the game determine the meaning of representations in a person's language of thought, and thereby determine what will and what will not count as a good translation of representations or sentences of the language of thought into some other language. Given all that, one might attempt to give a characterization of truth for the language of thought that satisfies a condition similar to Tarski's condition T. (Recall that one can obtain a characterization of truth-inL that satisfies condition T only if one can translate sentences of L into the language in which one gives the truth characterization.) The truth characterization provides truth rules for the language (game) of thought. Both epistemic rules and truth rules have a right to be considered "semantical rules"; so both sorts of rule might be called the semantical rules of the language (game) of thought. I t is not clear, however, which rules Sellars wants to call the semantical rules. He has a choice. First, he could say that semantical rules in his sense are epistemic rules. That would allow "language entry transitions" to be identified with responses to observation; but it would also mean that truth could not be identified with semantic assertability (assertability


in accordance with the semantical rules). Second, he could say that semantical rules in his sense are truth rules. Then truth would be semantic assertability; but "language entry transitions" would represent picturing rather than responses to observation. Furthermore, it is somewhat less clear in what sense the semantical rules in this second sense would be rules a person follows, and can be observed to follow, in playing the language game of thought; for this second sort of rule is a rule of the game only in the sense that it can be associated with the game. I cannot say what choice Sellars would make between these alternatives; and I am not sure that he does not have another alternative in mind. I hope he will clarify this matter in future writing^.^

Princeton University

3 Work on this review was supported in part by the National Science Foundation.