Inferentialism and Tacit Knowledge

Inferentialism and Tacit Knowledge

Inferentialism and Tacit Knowledge

Jon Cogburn

Louisiana State University

ABSTRACT: A central tenant of cognitivism is that knowing how is to be explained in terms of tacitly knowing that a theory is true. By critically examining canonical anti-behaviorist arguments as well as contemporary appeals to tacit knowledge, I am able to devise a more explicit characterization, where tacitly known theories must act as justifiers for claims that the tacit knower is capable of explicitly endorsing. In this manner, the new account is specifically tied to verbal behavior. In addition, if the analysis is correct, then it follows that a number of appeals to tacit knowledge by cognitivists are mistaken.

Key Words: Chomsky, cognitive explanations, Dummett, Evans, Fodor, procedural knowledge, prepositional knowledge, tacit knowledge

Tacit knowledge is a central concept in the cognitive sciences. For example, in the philosophy of language, Donald Davidson (1984), Michael Dummett (1996), and others (e.g. (Wright, 1987)) have long argued that we can best describe a competent language user in terms of that user’s tacit knowledge of a theory of meaning. With the advent of the Montagovian paradigm in linguistics, such “theories of meaning” became more than just philosophical hypostatizations. Montague (1974) showed how to rigorously and compositionally connect natural languages to interpreted formal languages, and how to empirically test the resulting structure in terms of the semantic properties of the interpreted formal languages in question. The indisputable empirical success of post Montague semantics (e.g. (Shalom, 1996)) provides inductive evidence of the cash value of the philosophers’ idea that linguistic ability is held, in part, in virtue of tacit knowledge of an empirical semantics.

Similarly, in the philosophy of mind, Jerry Fodor (1968), (1980) and others have argued that cognition should be thought of sententially. For example, intentional behavior can be thought of as involving sentences in a belief box and desire box being manipulated according to sensory input, inductive and deductive logics, and utility calculi. As with the case of Davidsonian philosophy of language and post Montague semantics, this philosophical view has proven empirically fruitful. The two most dominant “rule based” approaches to artificial intelligence (SOAR, (Newell, 1990) and ACT-R* (Anderson & Labierre, 1998)) can be thought of as empirical theories that realize Fodor’s philosophical views.

Both the Davidsonian meaning theorist and Fodorian cognitive scientist attempt to explain knowing how (procedural knowledge) in terms of knowing that (propositional knowledge). In both cases we have a kind of subconscious propositional knowledge that endows its possessor with cognitive abilities.

Other theories in the cognitive sciences are similarly epistemically motivated. Noam Chomsky has long criticized rival approaches to generative syntax for not being psychologically plausible. Likewise, Chomsky has motivated his successive approaches to syntax by arguing that they are psychologically real. For example, his most recent framework, Minimalism, attempts to characterize a natural language in terms of setting parameters of an initial syntax both invariant over all languages and somehow known by children prior to language learning. Here, even more strongly than in the case of post Montague semantics and rule-based approaches to AI, a philosophical view of competence secured by tacit knowledge of a theory determines the resulting empirical theory.

Given the manifest importance of the concept of tacit knowledge, it is exceedingly strange that it is so little philosophized about outside of very narrow discussions in the philosophy of language (e.g.(Miller, 1997)). This is doubly unfortunate, as all of the examples I have mentioned are controversial.

Chomskyan syntacticians and their philosophical allies such as Ludlow (1997), following Chomsky’s own strictures, tend to oppose semantics in the Montogovian tradition. Since Chomskyans have yet to devise rigorous theories of the systematic inferential role of parts of speech and morphological effects, it is hard to know what to make of this opposition.

In the case of rule-based approaches to AI and Minimalism in syntax, however, there are competing approaches with impressive successes. As discussed by Clark (1998), connectionist and situated agency approaches to AI and robotics have managed to replicate many of the successes of rule based approaches, as well as successfully model phenomena not yet modeled by rule based approaches. Likewise, there do exist serious challenges to Minimalism in syntax by influential linguists such as Carpenter (1989) and Morril (1994), and Pollard and Sag (1994) who work on, respectively, Categorial Grammar and Head Driven Phrase Structure Grammar.

For our purposes, the most interesting fact is that what is at issue between such theories often involves the notion of tacit knowledge as well. For example, Chomsky (1993, p. 22) argues that the concept in our mind that corresponds to the word “London” cannot have its reference secured in the manner reference is treated in truth conditional semantics.

is there any object London to which I am referring? If so, it is a very curious thing. . . We can regard London with or without its population: from one point of view, it is the same city if its people desert it; from another, we can say that London came to have a harsher feel to it through the Thatcher years, a comment on how people act and live. Referring to London, we can be talking about a location, people who sometimes live there, the air above (but not too high), buildings, institutions, etc., in various combinations. A single occurrence of the term can serve all these functions simultaneously, as when I say that London is so unhappy, ugly, and polluted that it should be destroyed and rebuilt 100 miles away. No object in the world could have this collection of properties.

After noting that semantics in the Montogovian tradition usually assigns a single denotation to London, Chomsky (ibid., p. 27) finishes his discussion by concluding,

Recall that Frege was speaking of a “logically perfect language,” one that will allow for a “common science.” He regarded natural language as not only imperfect, but even “in principle incoherent,” Dummett argues. If so, his specific project is unaffected by any of these considerations. But we may ask what it has to do with the inquiry into language and thought. Perhaps very little.

Truth conditional semantics is not real linguistics for Chomskyans precisely because, pace Davidson, Chomskyans hold that truth conditional semantics fails to explain the psychological reality of grasp of language.

Almost certainly because of Geoffrey Pullum’s (1989) infamous and influential critique of Chomsky’s recent approaches to syntax as being woefully nonrigorous, Chomsky’s criticisms of approaches that strive for greater formal precision than Government and Binding Theory and Minimalism is somewhat truculent. He (1995, p. 16) writes,

Another notion that appears commonly in the literature is “formal language” in the technical sense: set of well-formed formulas. . . Call such a set an E-language, where E is to suggest “external” and “extensional”. . . It is a question of empirical fact whether natural language has any counterpart to this notion. . . Apart from expository passages, the concept of E-language scarcely appears in the tradition of generative grammar that we are considering here. As distinct from the notions discussed earlier, it has no known status in the study of language. One might define E-language in one or another way, but it does not seem to matter how this is done; there is no known gap in linguistic theory, no explanatory function, that would be filled were such a concept presented. Hence, it will play no role in our discussion.

A practitioner of Categorial Grammer would no doubt point out the successes of the approach: its role in yielding natural language translation algorithms, its nice fit with semantics in the Montagovian tradition, and the elegant and simple way in which it models distressingly complex syntactic phenomena such as pied piping. However Chomsky should not be understood as denying these successes.

Rather, he is claiming that such successes are merely engineering, not reflecting what is really in the head of the language user. Of his own conception of language, Chomsky (ibid., p. 15) writes

let us refer to it as I-language, where I is to suggest “internal,” “individual,” and “intensional.” The concept of language is internal, in that it deals with an inner state of Jones’s mind/brain, independent of other elements in the world. It is individual in that it deals with Jones, and with language communities only derivatively, as groups of people with similar I-languages. It is intensional in the technical sense that the I-language is a function specified in intension, not extension: its extension is the set of SDs (what we might call the Structure of the I-language).

While defenders of alternate theories such as Sag and Wasow (1999) would not agree, Chomsky portrays the difference in terms of which theory can be claimed to be psychologically real.

In a similar manner, connectionists such as Patricia Churchland and Terrence Sejnowski (1989) criticize Fodor’s language of thought hypothesis (and rule based AI architecture) and defend connectionism (and neural net architecture) in terms of psychological reality. The claim is that even if rule based AI systems are successful as engineering, they don’t shed much light on human thinking. Unlike rule based AI, human thinking is massively parallel and distributed across the brain, degrades gracefully as the brain is destroyed, and is more different in degree than kind to animal cognition.

Of course these debates cannot be settled here. However, clarity about the notion of tacit knowledge is a necessary part of settling them. In what follows, I will first show how a characterization of tacit knowledge due to Gareth Evans has the counterintuitive result that a crane has tacit knowledge of aerodynamics. This failure is tied to Michael Dummett’s worry that his own earlier conception of tacit knowledge merely amounted to calling practical abilities a kind of knowledge. Dummett worries that his earlier view might be subject to canonical criticisms of behaviorism, the validity of which have largely been assumed by the philosophical community. Given the failure of Evans’ view, and Dummett’s worry, it would seem reasonable to examine the phenomena commonly taken to undermine behaviorism. Unfortunately though, it is easy to show that the critiques of early forms of behaviorism do not in any way motivate characterizing mental abilities as a species of propositional knowledge. Given this, phenomena such as domain-specificity, innateness, and the importance of getting inside the “black box” are of no help in characterizing tacit knowledge.

Then, I pick up another facet of Dummett’s characterization of tacit knowledge, that is, the requirement that a person be able to recognize the tacitly known theory as true. While this is promising, and consonant with the epistemic talk of many linguists, it too is beset by problems. Most people simply aren’t able to recognize correct theories in cognitive science. Furthermore, those able to do so tend to disagree with one another!

The Dummettian should respond that the requirement only applies to idealized people. However, this form of idealization again renders the definition non-necessary for tacit knowledge. If a rock were much smarter, then that rock would be able to recognize the laws of gravity that govern its behavior.

By examining linguists’ actual use of the notion of tacit knowledge, I am able to suggest a principled, and broadly behavioristic, way to limit the Dummettian idealization. I suggest that the theory tacitly known should justify propositions non-tacitly known by the person in question. Though the resulting characterization is free from obvious counterexamples, it ends up being far from trivial, as it entails that language of thought and rule based explanations will work best for cognitive phenomena that we are relatively articulate about (e.g. planning, reasoning, linguistic phenomena), and competing approaches will work best for phenomena of which we are relatively inarticulate (e.g. sensation, bodily movement). Thus, though my proposed characterization is arrived at mostly by conceptual analysis, it is subject to empirical falsification.

Evans’ Proposal

The clearest explication of the notion of tacit knowledge in the philosophical literature remains the discussion in “Semantic Theory and Tacit Knowledge,” where Gareth Evans (1985, p. 336) explicates the notion in this manner.

(1) At the level of output, one who possesses the state of tacit knowledge that p is disposed to do and think some of the things which one who had the ordinary belief that p would be inclined to do and think (given the same desires).

(2) At the level of input, one who possesses the state of tacit knowledge that p will very probably have acquired that state as a result of exposure to usage which supports or confirms (though far from conclusively) the proposition that p, and hence in circumstances which might well induce in a rational person the ordinary belief that p.

We can perhaps best understand this if we consider syntax. A correct syntax for a language will generate all and only the grammatical sentences of that language. At the level of output, this matches up with the ability of competent language users to discriminate between grammatical and ungrammatical sentences. For the case of input, the usage to which a competent speaker is exposed as a child is the same as that to which the syntactician uses in testing her theory.

Thus, we can start to make good sense of how the attribution of tacit knowledge of a correct syntax to competent speakers is both explanatory and testable. Given Evans’ characterization of tacit knowledge, we can represent it in this manner.

(1) At the level of output, one who possesses the state of tacit knowledge that a correct syntax S for a language L is correct is disposed to make the same grammaticality and distributional judgments as one who has the ordinary belief that S is a correct syntax for L (given relevantly similar desires).

(2) At the level of input, one who possesses tacit knowledge that S is a correct syntax for L will very probably have acquired that state as a result of exposure to grammaticality and distributional judgments which support or confirm (though far from conclusively) the proposition that S is a correct syntax for L, and hence in circumstances which might well induce in a rational person the ordinary belief that S is a correct syntax for L.

It should be somewhat clear that characterization of tacit knowledge does provide plausible (if vague) necessary conditions for the correct attribution of tacit knowledge (at least when concerned with syntax) of the truth of a proposition to a person. However, Evans’ characterization certainly does not provide adequate sufficient conditions.

It is relatively easy to come up with situations where we would not say that a person has tacit knowledge of a set of propositions, while both the input and output conditions of Evans’ definition of tacit knowledge are satisfied. It is easy to show cases of procedural knowledge that satisfy Evans’ description.

Consider a crane, call him Leonard, who is quite proficient at flying. We don’t want to say that Leonard has tacit knowledge of the textbook Principles of Aeronautical Engineering, but Evan’s criterion, taken as a sufficient condition, forces us to say that Leonard does. Consider.

(1) At the level of output, one who possesses the state of tacit knowledge that Principles of Aeronautical Engineering is correct is disposed to do and think some of the things which one who had the ordinary belief that Principles of Aeronautical Engineering is correct would be inclined to do and think (given the same desires).

(2) At the level of input, one who possesses the state of tacit knowledge that Principles of Aeronautical Engineering is correct will very probably have acquired that state as a result of exposure to usage which supports or confirms (though far from conclusively) the proposition that Principles of Aeronautical Engineering is correct, and hence in circumstances which might well induce in a rational person the ordinary belief that Principles of Aeronautical Engineering is correct.

To see why this doesn’t work, let Amelia be the author of Principles of Aeronautical Engineering. Also note that Amelia had wanted to design and fly planes since she was a child. Again, at the level of output, both Leonard and Amelia end up behaving the same way; given that Amelia desires to fly, Principles of Aeronautical Engineering functions as a how-to book. The clause for input is satisfied because Leonard must cope with the same physical conditions Amelia must when testing her theories. Principles of Aeronautical Engineering explains these conditions by predicting them from a rich and inventive combination of physical law and mathematical approximation theory.

Thus, if we take Evans’ characterization of tacit knowledge to be correct, we are forced to say that Leonard has tacit knowledge of the propositions expressed in Principles of Aeronautical Engineering. However, it is very weird to say that Leonard has tacit knowledge of a physical theory rich enough to adequately describe and predict his flying behavior.

What I take to be most interesting about the insufficiency of Evans’ account is that it suggests that propositional knowledge may in fact be irrelevant to large swaths of cognitive science. Perhaps the cognitive revolution rests upon a simple mistake, and the relationship between a correct cognitive science and our own cognitive abilities is no more direct than that between Principles of Aeronautical Engineering and Leonard’s flying abilities.

Some of Michael Dummett’s earlier writings about the theory of meaning in fact tended to suggest this view, as admitted in this more recent passage (1991, p. 105).

If linguistic competence could be straightforwardly classified as a practical ability, we could say, as I once did say, that in framing a meaning-theory we are giving a theoretical representation of a practical ability-the ability to speak the language. We are representing this complex ability as consisting in the knowledge of a theory, that is of an articulated structure of propositions. On this account, we are analyzing a complex of practical abilities by feigning to attribute to who has these abilities a knowledge of the theory.

Dummett takes it to be clear that this earlier dogged insistence on seeing linguistic competence as only a matter of know-how is an implausible form of behaviorism, though he does not say why it is implausible. Rather, he seems to have been swept up in the anti-behaviorist Zeitgeist of Universal Grammar’s hey-day. Following Dummett’s suggestion, one should examine the problems with old-school, radical, behaviorism as a guide to characterizing the notion of tacit knowledge.