Tacit Knowledge

Citation preview

Tacit Knowledge Christina Graves; Jerrold J. Katz; Yuji Nishiyama; Scott Soames; Robert Stecker; Peter Tovey The Journal of Philosophy, Vol. 70, No. 11. (Jun. 7, 1973), pp. 318-330. Stable URL: http://links.jstor.org/sici?sici=0022-362X%2819730607%2970%3A11%3C318%3ATK%3E2.0.CO%3B2-W The Journal of Philosophy is currently published by Journal of Philosophy, Inc..

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/about/terms.html. 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 http://www.jstor.org/journals/jphil.html. 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].

http://www.jstor.org Wed Jun 6 07:56:33 2007

THE JOURNAL OF PHILOSOPHY

T A C I T KNOWLEDGE *

T

HE controversy between rationalism and empiricism has recently taken a linguistic turn as a result of the development of the transformational theory of grammar.l Transformational theory led to the discovery of significant linguistic universals in phonology, syntax, and semantics previously beyond the range of grammatical description.2 With the discovery of such universals, the question arose of how to explain the fact that, even though environmental conditions under which languages arise, develop, and change seem to differ radically, every natural language coiiforms to a set of strong constraints on grammatical structure. Chomsky and other transformational linguists hypothesize that the common form of natural languages is part of the genetic capacity for language acquisition.3 But, even granting this hypothesis, the rationalist cannot claim the support of transformational linguistics without a further argument to connect this scientific claim to the philosophical claim about innate knowledge. In order for rationalists to argue from the existence of innate structures to the existence of innate knowledge, some notion of tacit knowledge is required so that they can claim that the grammatical structures represented in transformational grammars are objects of knowledge. Thus, the problem for both rationalists and empiricists is whether 'tacit knowledge' can be made sense of in a way that permits us to say that empirically adequate grammars explicate what a speaker knows about a language. In this paper, we take up this problem at the point its discussion has reached to date. We see this point as the intersection of two lines of argument: first, Fodor's recent attempt to explain the notion of tacit knowledge in terms of the idea of optimal simulation, and, second, Stich's criticism of Fodor and Stich's attempt to show that there is no knowledge that every speaker of a natural

* IVe want to thank the members of the MIT department of philosophy for their help, particularly Sylvain Bromberger and Ned Block. This work was supported in part by the National Institute of Mental Health, Grant 2 POI MH 13390-06. 1 See the Synthese symposium on innate ideas, XVII, 1 (March 1967): 1-28. 3 Taxonomic grammars have no means of representing deep structure. Thus, the existence of such universals was obscured by their exclusive concern with surface structure, where the most extensive differences between languages are found. 3 N. Chomsky, "Explanatory Models in Linguistics," in E. Nagel, P. Suppes, and A. Tarski, eds., Logic, Methodology, and Philosophy of Science (Stanford, Calif.: University Press, 1962), pp. 528-550.

TACIT KNOWLEDGE

3I9

language possesses qua speaker of the language.4 We argue that Stich is wrong in both his criticism of Fodor and his positive claim and that Fodor's explanation does not go far enough to connect linguistic nativism with philosophical rationalism. On the positive side, we try to provide an extension of Fodor's explanation that makes it reasonable to say that speakers of a natural language qua speakers know (tacitly) the rules of its grammar. Stich's thesis that there is no knowledge that speakers of a natural language must possess rests on three separate claims. H e denies that all speakers have knowledge tacit or otherwise of: (1) linguistic universals (2) the particular grammatical rules of their language (3) the grammatical properties of individual expressions of their language We will now try to show that Stich is wrong in his claims about (1) and (2) In order to make clear the nature of our disagreement with Stich's claims (1) and (2) we begin with an examination of claim (3). With respect to this claim, Stich argues that speakers may have knowledge of the structure of some of the expressions in their language, but that that knowledge is neither long standing nor tacit.6 He claims that "grammar is not the study of a body of beliefs-the speaker has no such beliefs. He has rather the capacity to acquire certain beliefs, to gain new knowledge, on hearing new sentences" 4 S. Stich, "What Every Speaker Knows", PhilosoPhical Review, ~ x x x 4 , (October 1971): 476-496; J. A. Fodor, "The Appeal to Tacit Knowledge in Psychological Explanation," this JOURNAL, LXV,20 (Oct. 24, 1968): 627-640. Parenthetical page references to Stich and Fodor will be to these papers, if not otherwise noted. 5 I n Stich's initial formulation of (2), he speaks of speakers knowing that particular rules of their language are rules of the grammar of their language. This is a slightly different claim from (2); however, in his discussions later in his paper, he employs (2) as above. We return to this distinction. 6 Stich wouldn't characterize this knowledge as knowledge of the grammatical properties of an expression. However, it does not matter for the argument that follows whether knowledge of the grammatical properties of sentences is characterized in this way or under some other description. e of gramFurthermore, Stich's reluctance to call this k n o ~ v l ~ d e"kno~vledge matical properties" seems to be based on a confusion. H e regards grammaticality, ambiguity, anomaly, etc. as theoretical properties. "To be grammatical is to be classified as grammatical by a correct grammar" (478). However, thiy is a mistake. T o be grammatical in L is to have the structnre of a sentence of L. T h e definition "x is grammatical in L = an optimal grammar of L generates x" gives an explication of the ordinary notion of grammaticality. It does not eliminate the ordinary notion. Such definitions state the condi~ions under which the theory of a language predicts that speakers will apply the definiendum to its expressions.

,,

320

THE JOURNAL OF PHILOSOPHY

(494). First, we agree with Stich that the knowledge presently under discussion is not tacit.7 We also agree that the knowledge that a particular expression has a certain grammatical property is acquired only upon encountering that expression, not when the language is learned by the child. Furthermore, we agree with Stich that speakers need not be able to justify their belief that a particular expression has a certain grammatical property. Stich thinks that this implies that such beliefs cannot be long standing. His argument is as follows: He thinks that beliefs that particular expressions have certain grammatical properties do not require justification, that only directly evident beliefs do not require justification, and that no long-standing belief is directly evident. He concludes that beliefs that particular expressions have certain grammatical properties are not long standing. Although this conclusion is validly drawn, the argument is not sound because at least the last of these premises is false.8 For example, many of us have believed for many years that "The cat is on the mat" is grammatical. This would seem to make it a long-standing belief. What is important here is that Stich's denial of (1) and (2) directly conflicts with Chomsky's claim that the best explanation of linguistic competence is a theory that posits tacit knowledge of linguistic universals and the grammar of the language. Our plan in this paper is to defend Chomsky's claim by arguing that, if speakers can have knowledge that particular expressions have certain grammatical properties, then speakers must also have knowledge of types (1) and (2). Stich argues that postulating tacit knowledge of types (1) and (2) is unjustified. He says:

. . . we must

ask whether there are any significant similarities or dissimilarities between the cases in question a n d unproblematic cases of knowledge. If there are significant dissimilarities we must ask whether the attribution of knowledge is nonetheless called for by some special feature of the situation, whether failure to attribute knowledge would leave something unexplained or whether attribution of knowledge would account for something otherwise unaccounted for (485).

Stich's argument must consist of two parts. First, he must demonstrate substantial dissimilarities between unproblematic cases of 7 Note that Chomsky does not claim that such knowledge is tacit; rather his critics, like Harman, do. Cf. G. H. Harman, "Psychological Aspects of the Theory of Syntax," this JOURNAL, LXIV,2 (Feb. 2, 1967): 75-87, p. 81. 8 We might also question the second premise on the grounds that a nondirectly evident principle like the principle of induction does not require a justification, although it would be nice to have a justification for it.

TACIT KNOWLEDGE

321

knowledge and tacit knowledge of types (1) and (2). Second, he must show that failing to attribute such knowledge to the speaker would leave unexplained nothing that would otherwise be accounted for. In part one of his argument, Stich offers a description of a paradigmatic case of knowing. He says: "commonly when a person knows that p [(a)] he has occasionally reflected that p or has been aware that p; [(b)] he will, if inclined to be truthful and otherwise psychoIogically normal, assert that p if asked. More basic still, [(c)] he is capable of understanding some statement which expresses what he knows" (485/6). Now, if we take for p some individual rule of grammar or statement of linguistic theory, it is clear that, for most speakers at least, (a) and (b) will be false. The situation with respect to (c), however, is not clear. It is not clear that there exists any linguistic principle of which there is no formulation comprehensible to speakers. After all, fairly sophisticated principles of mathematics and science, once comprehensible to only a tiny elite, are now formulated in a manner that makes them understandable to high school children. More significantly, there are counterexamples to (c).9 For instance, someone who knows that his breakfast is on the table downstairs may not be able to understand any sentence that expresses what he knows because he is suffering from aphasia for written language and speech. And an animal, such as a dog, can know where its meal is located even though it is unable to understand any sentence expressing what it knows. Therefore, Stich has found only two undisputable dissimilarities between ordinary, unproblematic cases of knowledge and knowledge of types (1) and (2). Both of these dissimilarities would be predicted by proponents of tacit knowledge, since they constitute two basic conditions for the knowledge being tacit. But even if Stich is right about (c), nothing follows about the claim we will make about tacit knowledge. The existence of a dissimilarity between paradigm cases of explicit knowledge and the cases that proponents of tacit knowledge have in mind does not show that the latter are not cases of knowledge. Thus, the first part of Stich's argument doesn't get him very far; part two of his argument bears the burden of proof. Stich must show that nothing is explained by attributing tacit knowledge of types (1) and (2) to speakers of a natural language. The second part of Stich's argument is as follows. First, he assumes that individual grammars are theories that describe the 9 The authors wish to thank Ned Block for bringing this point to their attention.

322

THE JOURNAL OF PHILOSOPHY

linguistic intuitions of native speakers, and that linguistic theory describes a child's predispositions to acquire such intuitions. Next he points out that the evidence that tvould confirm a particular linguistic theory remains the same whether or not we attribute to the child knowledge of this theory. Last, he concludes, postulating tacit knowledge of linguistic universals accounts for nothing that is not accounted for by taking linguistic theory to be a theory about predispositions pure and simple. The same point is made about tacit knowledge of the rules of a grammar. There are a number of problems with this account. First, it seems to us misleading to say that a grammar is a theory of or about linguistic intuitions and that linguistic theory is a theory about predispositions of the child to acquire such intuitions.lO A grammar is a theory of a language. It enumerates the sentences of the language and marks their properties, syntactic, phonological, and semantic. A linguistic theory is a theory about the formal properties essential to all natural languages. Of course, given a grammar and some assumptions about performance mechanisms, we can predict that a speaker will find acceptable only those expressions with certain formal properties. Similarly, we can describe a child's predispositions to learn languages as predispositions to learn only languages with the properties described by linguistic theory. However, having done this descriptive job, there are further questions that need to be asked. Granted that a child does have a predisposition to learn only certain types of languages, why is it that the child has this predisposition? Granted that competent speakers find expressions with a certain property acceptable, how is it that they come to make these judgments? The general point is this: having described certain dispositions of an organism, it is typically appropriate to ask what features of the organism are responsible for its having the dispositions. Stich fails to see this point. According to him, the relationship between linguistic theory and the linguistic behavior of human speakers is analogous to the relationship between a theory describing the trajectory of projectiles and the behavior of projectiles (like rocks, bullets, and spears?). But this is wrong. The relevant difference between human beings and projectiles is that to explain the behavior of a human being one has to explain the role played by some internal structure, whereas in the case of projectiles of the kind Stich seems to have in mind, no mention of internal structure is necessary. The only features that need to be referred to to exl o Chomsky himself sometimes encourages this mistake; cf. his Aspects o f the Theory of Syntax (Cambridge, Mass.: MIT Press, 1965), p. 27.

TACIT KNOWLEDGE

323

plain the trajectory of such a projectile are its initial position and velocity, its shape, mass, and other observable features. An example closer to the case of human behavior would be a "smart bomb" or guided missile. T o explain the trajectory of such a weapon, some account of the operations of the internal guidance mechanism is necessary. On an analogy with this model, the hypothesis that speakers have internalized the rules of the grammar constitutes an explanation of specific features of linguistic behavior in just the manner in which a left turn on the part of a "smart bomb" or guided missile might be accounted for by the operations that instantiate features of its targeting program. If Stich were now to reply that reference to internalized rules can be replaced by talk about dispositions of the elements of such an internal mechanism, then we would regard this as a mere facon de parler for talking about internalized structures.ll Stich's argument does not distinguish between the claim that speakers know the grammatical rules of their language and the claim that these rules are internally represented. We have just shown the inadequacy of his argument that nothing is explained by positing an internalized structure of grammatical rules. But since not all internalized structure need be counted as knowledge, we now need to argue that speakers tacitly know the grammatical rules that they have internalized. At this point, we turn to the constructive part of our discussion. We must determine whether all, some, or none of this internal structure is knowledge. One proposal for settling this question has been made by Fodor.lZ It says that any principles that are assumed in an optimal simulation of the speaker's linguistic performance are tacitly known. Stich's criticism of this proposal is that, even if Fodor is right, it doesn't follow that the principles employed in an optimal simulation of performance will be the rules of the grammar of the speaker's language (490). There are two things wrong with this criticism. First, Stich claims that there is no knowledge that speakers, qua speakers, must possess. Thus, it doesn't matter whether the rules of the grammar are incorporated in the performance 11 It is important to see that the problem for which attribution of knowledge of types (1) and (2) constitutes a partial answer is not a problem about the form or content of linguistic theory, but rather a problem about the relationship between linguistic theory and a psychology of human abilities. Thus it is not surprising that "evidence that would confirm or disconfirm a particular linguistic theory remains the same with or without the assumption that speakers tacitly know the theory." That Stich thinks that this fact supports a claim that attribution of tacit knowledge has no explanatory value shows that he is confused about what needs to be explained. 1 2 O p . cit. Also cf. Katz, Semantic Theory (New York: Harper & Row, 1972), pp. 24-29.

3*4

THE JOURNAL OF PHILOSOPHY

model, since some principles must be incorporated, and these, according to Fodor's proposal, would be knowledge. Second, there is every reason to believe that the optimal performance model would include the principles of the grammar. What distinguishes a model of performance from a model of competence is that, in addition to the rules of the grammar, a performance model includes principles representing a mechanism for processing sentences in real-time (such as hypotheses about computation space).13 But Fodor's proposal is too liberal. It would lead us to accept as tacit knowledge the differential equations that determine how a cyclist maintains balance. We might accept Fodor's proposal as an account of the computation procedures the cyclist goes through. However, the problem with the proposal is that it does not distinguish between computation procedures and knowledge about operations and output. The reason we want to make this distinction is that, given an account of computation procedures that explains an organism's behavior, nothing further is explained by postulating tacit knowledge of aspects of these computation procedures. Thus, the essential difference is that, in the case of the cyclist, we merely need to explain behavior, whereas in the case of the speaker, we need to explain explicit knowledge of grammatical properties of sentences. We need a less liberal proposal that will include the wanted cases and exclude cases like that of the cyclist. In order to obtain such a proposal, we might ask ourselves how we explain knowledge in situations where no issue of tacitness arises. Consider, for example, a linguist in the field, studying an exotic language. After some months of study, he is able to make correct judgments about the grammatical properties and relations of novel sentences. How does he know that the sentences have these properties and relations? Clearly, his study of the language has led him to formulate explicitly certain principles of grammar from which he can deduce that the sentences in question have the properties and relations. Accordingly, the linguist's type (3) knowledge is explained by the explicit use of more general rules of grammar. In the above case, we explain the linguist's type (3) knowledge by saying that the linguist deduces the relevant statements from the rules (and definitions) of the grammar.14 The question now is 1 3 Chomsky and G . Miller, "Finitary Models of Language Users," in R. D. Luce, R. R. Bush, and E. F. Galanter, eds., Handbook of Mathematical Psychology (New York: Wiley, 1963). 1 4 We do not mean to exclude the use of various psychological strategies from the theory of speech performance, nor do we assume any particular relation between the form of the grammar and the form of a performance model.

TACIT KNOWLEDGE

325

whether this paradigm of explanation should be transferred to the case of the untutored speaker. T h e position of the untutored speaker is comparable to that of the linguist. T h e untutored speaker has the same information, in explicitly knowing that certain sentences have certain grammatical properties. But we can assume that the speaker, unlike the linguist, does not explicitly know the principles from which it follows that these sentences have these grammatical properties. Why not then extend the paradigm of explanation to the speaker, by assuming that the speaker performs the same deduction as the linguist, only tacitly, and thus that the principles in the deduction are tacitly known? T h e argument that we should extend the explanation is simply that by so doing we explain this explicit knowledge. Let us look more closely at this claim to explain knowledge by extending the paradigm of explanation to the case of unconscious knowledge. Consider an example. T h e explanandum is the range of knowledge that English speakers have about the form of reflexive objects of verbs. For example, English speakers know that (4), (5), and (6): (4) John overestimated himself (5) Mary praised herself (6) T h e bugs fought with themselves are well-formed, grammatical sentences, whereas (7), (8), and (9): (7) "John overestimated themselves (8) "Mary praised himself (9) +The bugs fought with herself are not. T h e essential point is that this knowledge is open-ended; that is, for virtually any such string, the speaker will know whether it is grammatical or ungrammatical. In this case, extending the paradigm would consist in taking the explanans to include the grammatical rule that the number, gender, and case of reflexive objects of verbs are identical with the number, gender, and case of the subject (together with the other grammatical rules required for deriving these facts in a transformational grammar).l5 It is natural to ask whether such an explanation is the only kind available or whether there is another kind that does not involve the positing of a tacit deduction from tacitly known principles. As we see this question, the choice before us is whether to extend a familiar paradigm of explanation by postulating an unfamiliar type of knowledge or to save ourselves this postulation by adopting 1 5 P. Postal, "On So-called 'Pronouns' in English," in D. A. Reibel and S. A. Schane, eds., Modern Studies in English (Englewood Cliffs, N.J.: Prentice-Hall, 1969), pp. 201-224.

326

THE JOURNAL OF PHILOSOPHY

a new paradigm of explanation. We have taken the former alternative because there appears to be no plausible paradigm of explanation available that avoids the postulation of tacit knowledge. T o avoid this postulation, one would have to say that internalized ruies by themselves account for the speaker's type (3) knowledge. But this amounts to saying that knowledge could be explained in terms of the mechanical operations of a piece of computing machinery. The operations of a "wheels-and-pulleys" device (including modern electronic computers) transform physical inputs into physical outputs by a series of physical events which, because they bear some structural relation to the steps that people go through in constructing arguments, playing chess, etc., can be construed as representing some intellectual achievement. What we do when we speak metaphorically of the intellectual achievements of computers is to impose an interpretation under which the sequence of physical events in the computer can be viewed as corresponding to a sequence of propositions, and the workings of the device can be viewed as corresponding to certain mental operations. But in precisely the same sense we could also treat the changes in the positions of a certain cluster of heavenly bodies (from one time of year to another) as a theorem-proving computer. Computers and appropriately chosen regions of the heavens are isomorphic; so if the computers can bear an interpretation under which it can be said to perform various intellectual tasks, the heavenly regions can as well. Since no one would be much tempted to say that beliefs can arise out of the dynamics of some collection of heavenly bodies, one ought not be tempted to say this about computer operations. Therefore, we know of no plausible paradigm that avoids postulating an unfamiliar type of knowledge (although, of course, not too unfamiliar since the same kind of postulation is found in other disciplines, e.g., psychoanalysis, perceptual theory, etc.) while at the same time affording a realistic explanation of the processes by which we come to have type (3) knowledge. The problem is that there is no analysis of knowledge in purely physicalistic terms under which computers can be literally said to know propositions. Until such an analysis can be given, the notion ot a computer explanation is vacuous and too problematic to afford us with a new paradigm of explanation. Thus, we don't object to the claim that it is possible in principle to build a computing device that might be said to know propositions. And we note that if there were such a device and we had a physicalistic analysis of knowledge, then it would also be appropriate on our account to say

TACIT KNOWLEDGE

327

that the device had tacit knowledge of a least some of the rules built into it. Now, it might be thought that Stich himself offers a conclusive objection to our position in claiming that judgments of type (3) are directly evident perceptual judgments. However, it is important to note that such an objection might embody two quite separate criticisms. I t might be thought that (10): (10) Judgments of type (3) are directly evident, and if a judgment is directly evident, then it is not deduced. or it might be thought that (11): (11) Judgments of type (3) are perceptual judgments, and if a judgment is perceptual, then it is not the result of a deduction. In support of (10) Stich might claim that directly evident jutlgments are not deduced because, on the one hand, such judgments cannot be justified, and, on the other, judgments that are the result of a deduction from true premises can always be justified by appealing to the deduction itself. Our reply to this objection is that judgments of type (3) are not directly evident in the required sense. What might be directly evident in Stich's sense are judgments about ~vhatsound sequences seem well-formed in our own ideolect.l6 But judgments of type (3) are not of this sort; they are judgments about the grammatical properties of sentences in a natural language (like English). With respect to these judgments there is always evidence that bears on their truth or falsity, e.g., evidence regarding the intuitions and behavior of other speakers, comparison with other sentences, etc. Hence, judgments of type (3) are capable of justification. We note that eliminating objection (10) undermines one strong reason for holding (11). Stich, for example, argues that the purported fact the judgments of type (3) are directly evident is explained by the fact that they are perceptual judgments fully on a par with judgments like (12): (12) I seem to see a yellow object But since judgments of type (3) are not directly evident, Stich has no argument to establish that they are perceptual. 16 We say "seem" rather than "are" because any sentence in a speaker's idiolect is related to infinitely many other sentences, past, present, future, and possible, so that there is always evidence available about whether a sound sequence is \\.ell-formed in an ideolect.

328

THE JOURNAL OF PHILOSOPHY

One way, then, of looking at the dispute between Stich and ourselves is to see that the two sides are talking about different kinds of judgment. Whereas we claim that speakers of English can make correct judgments of the form "S is grammatical in English," "S is ambiguous in English," etc., Stich seems to think that they can make only judgments like "S sounds (to me) more similar to Sf than it does to S"." We think that Stich's claim is not only unwarranted but also in direct conflict with the explicit practice of the transformational linguists he is attempting to reconstruct. We think that these claims of Stich's are based upon two previously noted confusions. First, because he thinks that properties like grammaticality, ambiguity, anomaly, etc. are exclusively theoretical properties defined by a grammar or a theory of grammars (cf. fn 6 of this paper), he thinks that speakers cannot be expected to have intuitions regarding them. Thus, he chooses the alternative of supposing that the intuitions upon which linguistic theory and particular grammars rely for their empirical confirmation are intuitions about similarity relations among sentences. Second, because he fails to distinguish between the data for a theory and the subject matter of the theory, he thinks that the facts that grammars must account for are nothing more than intuitions that certain sentences sound similar to certain individual speakers. However, these views are simply mistaken. T h e second, in particular, involves a quite general confusion (not atypical of empiricist treatments). For to say that, because speakers' judgments constitute evidence for linguistic theory, therefore such judgments must be the subject matter of that theory (what the theory is about) is like saying that, because sensory experience (of meters, etc.) constitutes evidence in physics, therefore the theories in physics are theories about our sensory experience. But just as the subject matter of physical theories is the structure of the physical universe, so the subject matter of the theory of grammar is the structure of language. Thus, knowledge of this structure, unlike knowledge of internal psychological states of one's own, is never directly evident. Now we come directly to objection (11). Of course, showing that judgments of type (3) are not directly evident does not show that they are not perceptual. Although we have no longer any reason to think that such judgments are perceptual, they certainly may be. But even if they are perceptual in some appropriate sense, still there is no reason to think that no tacit deductions underlie them. Until we are given some reason to think that perceptions do not involve tacit deduction of the kind postulated here, objection (11) is a non sequitur.

TACIT KNOWLEDGE

329

Implicit in our above account is a narrower criterion for tacit knowledge; a proposition is tacitly known if it is one of the propositions appealed to in a tacit deduction; that is, if it is one of the propositions from which the person has tacitly deduced a (in this case explicitly) known proposition. It must be clear that this criterion for tacit knowledge avoids counterexamples like the cyclist case. In cases of that kind, there is no knowledge to be explained.17 We now turn to the question of tacit knowledge of linguistic universals. Supposing that we have established the existence of tacit knowledge of type (2), we have, clearly, yet to show that tacit knowledge of type (1) exists. The question of the existence of tacit knowledge of type (1) is, moreover, of more philosophical importance, because it is the crucial question in the controversy between rationalism and empiricism (since type (2) knowledge can be expected to vary from language community to language community and hence be ruled out as a candidate for the status of a linguistic universal). One way to argue for the existence of tacit knowledge of type (1) is to argue that many of the principles of particular grammars are also principles of linguistic theory and so, by our previous argument, objects of tacit knowledge. This is especially clear in connection with the definitions of grammatical properties and relations, since all such definitions are expressed in linguistic theory.ls But it is also true of certain principles of particular grammars. Some principles in grammars of specific languages are there by virtue of the fact that they are part of universal grammar, that is, of linguistic theory. One type of case includes both base rules, like the rule for the recursive expansion of N P to contain sentence-structures, and transformations, like the rule that forms a relative clause.1gAnother type of case includes conditions on the application of rules in a grammar, e.g. Chomsky's A-over-A condition, which imposes restrictions on the ability of transformational rules to extract a phrase from a more inclusive phrase of the same type.20Thus, having shown that speakers know the rules of their grammar, we have also shown that they know these universal rules and conditions.Z1 17 It should also be made clear that our criterion does not commit us to an infinite regress. We do not claim that anything a speaker knows is to be explained by a tacit deduction. We leave open the possibility of "ultimate princlples" of language, that is, principles which by their nature cannot be explained. 1s Katz, Semantic Theory, pp. 11-12. 1 9 Chomsky, Aspects of the Theory of Syntax, pp. 63-127. 20 Chomsky, "Conditions on Rules," in S. Anderson and P. Kiparsky, eds., Festschrift for Morris Halle (New York: Holt, Rinehart & Winston, 1973). 2 1 Note that this argument does not establish that all linguistic universals are known by a speaker who knows all the rules of the grammar. This is be-

33O

THE JOURNAL OF PHILOSOPHY

Another argument for tacit knowledge of type (1) can be constructed in a manner similar to our argument for tacit knowledge of type (2). We suppose that speakers have tacit knowledge of the grammatical principles of their language. We may then ask, as in the previous case, for an explanation of how they acquired this knowledge. If the best account of how speakers learn grammatical rules says that they learn them on the basis of certain innate linguistic principles PI, . . . , P, that determine a grammar G with respect to some class D of primary linguistic data, then there is a deduction of G from D and P I , . . . , P,. As before, we can take P,, . . . , P, to be tacitly known.22 Note, however, that establishing that speakers know some principles of linguistic theory does not show that they know that these principles are principles of linguistic theory. This is because such knowledge would be knowledge that the principles in question are principles of every possible grammar. Clearly, this is not something that speakers qua speakers can be supposed to know. Knowing a universal condition or grammatical principle is knowing what it states; e.g., knowing the A-over-A principle is knowing the condition for extracting phrases in certain contexts. But the relevance of linguistic theory to issues like the rationalist-empiricist issue depends only on whether speakers have tacit knowledge of the universal linguistic principles upon which language acquisition depends, not whether they know what may never be known even by future linguists. CHRISTINA GRAVES JERROLD J. KATZ YU JI NISHIYAhlA

SCOTT SOAhlES ROBERT STECKER

PETERTOVEY

R/Iassachusetts Institute of Technology cause some universals may not pertain to some natural languages. Universals can take the form of conditionals whose antecedent refers only to some family of languages (having some abstract property). 2 2 Note that one of these principles is a metric for evaluating grammars 011 the basis of linguistic data; cf. Aspects of the Theory of Syntax, pp. 30-47. I n a more recent paper, "Grammar, Psychology, and Indeterminacy," this JOURNAL, LXIX, 22 (Dec. 7, 1972): 799-818, Stich calls such deductions into question on the grounds that there is no justification for choosing a metric for evaluating grammars. For a refutation of Stich's argument, see Chomsky and Katz, "What the Linguist Is Talking about," to appear.