Tuesday, 1 November 2011

Tacit knowledge of theories of meaning

The conception of tacit knowledge I defend is that of context-dependent, conceptually-structured, practical knowledge. It stands opposed to knowledge which can be made explicit in situation-independent general terms. Thus, in general, tacit knowledge cannot be conveyed by situation-independent linguistic instruction. Such language and tacit knowledge stand opposed.

But there has, historically, been an appeal to tacit knowledge precisely in connection with language and in a way which is rather more straight forward than Harry Collins’ suggestion that tacit knowledge can reside in the patterns and frequencies of word use of linguistic communication. This is the idea that a speaker’s understanding of a language consists in tacit knowledge of a theory of meaning, or a grammar, for that language.

There are two main threads for this idea in the philosophy of language. One derives from Chomsky’s project to articulate an innate, universal grammar for human natural language.

If a person who cognized the grammar and its rules could miraculously become conscious of them we would not hesitate to say that he knows the grammar and its rules, and that this conscious knowledge is what constitutes his knowledge of language. [Chomsky 1980: 70].

The other is Donald Davidson’s aim to set out a theory of meaning for natural language based on an inversion of Tarski’s semantic conception of truth. Such a theory of meaning would contain a finite set of axioms, giving, for example, the reference of primitive terms and a recursive procedure to derive an instance of the T-schema for any declarative sentence of the language.

In Tarski’s work, T-sentences are taken to be true because the right branch of the biconditional is assumed to be a translation of the sentence for which truth conditions are being given. But we cannot assume in advance that correct translation can be recognised without pre-empting the point of radical interpretations; in empirical applications, we must abandon the assumption. What I propose is to reverse the direction of explanation: assuming translation, Tarski was able to define truth; the present idea is to take truth as basic and to extract an account of translation or interpretation. The advantages, from the point of view of radical interpretation, are obvious. Truth is a single property which attaches, or fails to attach, to utterances, while each utterance has its own interpretation; and truth is more apt to connect with fairly simple attitudes of speakers. [Davidson 1984: 134]

Under the general constraint that such a theory delivers true instances of the T-schema – true meaning theorems – across the board, individual instances such as ‘Snow is white’ is true iff snow is white will, Davidson argues, unpack the meaning of sentences of the object language.

In fact, the connection between Davidson’s project and tacit knowledge is not clear. He makes two comments on the subject. One is that knowledge of such a theory would suffice for understanding [Davidson 1984: 125]. The other is that it is a necessary condition for languages to be learnable that a constructive or compositional account of the language could be given [ibid: 3]. But he does not say explicitly, for example, that speakers have implicit knowledge of a theory of meaning which explains their ability. Others, however, have made this claim.

Michael Dummett, for example, argues that the best interpretation of Davidson’s approach is as a ‘full bloodied’ rather than ‘modest’. The requirement for full bloodedness for a theory of meaning is the requirement that it does not simply presuppose key facts about content by simply giving the meaning of the basic terms of one language by simply stating them in another language. Instead, it gives an account of the meaning of the primitive terms of a language, its basic predicates and referring terms by describing the practical abilities that an understanding of those terms gives to a speaker. This, in turn, presupposes a form of tacit or implicit knowledge.

A theory of meaning will, then, represent the practical ability possessed by a speaker as consisting in his grasp of a set of propositions; since the speaker derives his understanding of a sentence from the meanings of its component words, these propositions will most naturally form a deductively connected system. The knowledge of these propositions that is attributed to a speaker can only be an implicit knowledge. In general, it cannot be demanded of someone who has any given practical ability that he have more than an implicit knowledge of those propositions by means of which we give a theoretical representation of that ability. [Dummett 1976: 70]

The recursive structure of a theory of meaning of such a form reflects ‘our intuitive conviction that a speaker derives his understanding of a sentence from his understanding of the words composing it and the way they are put together’ [Dummett 1974: 109].

Alexander Miller suggests that this project aims at answering three question:

(a) How is it possible, given the finitude of their capacities, for speakers of a natural language to understand a potential infinity of sentences?
(b) How is it possible to understand utterances of previously unencountered sentences?
(c) How is it possible for a natural language to be learnable? (i.e. how is it possible for explicit training with only a relatively small number of sentences to secure competence with a possibly very large set of sentences outwith that initial set?).

Dummett’s outline answer to this depends on the cogency of thinking that speakers have knowledge of the axioms that encode understanding of the language and which fix the derivation of suitable, interpretative, instances of the T-schema.

There is, however, a problem with thinking of knowledge of the axioms as knowledge in any ordinary sense. The problem was highlighted by Gareth Evans as a problem for thinking of grasp of the axioms of a theory of meaning as any kind of intentional state or propositional attitude. Evans argues that ‘It is of the essence of a belief state that it be at the service of many distinct projects, and that its influence on any project be mediated by other beliefs’ [Evans 1981: **]. He then considers the ascription of a belief that a particular substance is poisonous to a rat:

It is true that many philosophers would be prepared to regard the dispositional state of the rat as a belief. But such a view requires blindness to the fundamental differences which exist between the state of the rat and the belief of the man—differences which suggest that fundamentally different mechanism are at work. We might begin with this disanalogy: the rat manifests the 'belief' in only one way—by not eating—whereas there is no limit to the ways in which the ordinary belief that something is poisonous might be manifested. The [rational] subject might manifest it by, for example, preventing someone else from eating the food, or by giving it to a hated enemy, or by committing suicide with it. These variations stem from the different projects with which the belief may interact, but similar variations arise from combining the belief with other beliefs. It might, for example, lead to the subject's consuming a small amount of the food every day, when combined with the belief that the consumption of small doses of a poison renders one immune to its effects. [Evans 1981: **]

So although the rat and the rational subject may share some dispositions to behave (in general: avoiding eating the poisonous substance), for the rat, the dispositions are tied to a narrow range of behaviours whereas for the rational subject, there is no limit to the actions to which a belief that something is poisonous may contribute. Ascribing a belief to the rat adds nothing to a more minimal stimulus response account of its behaviour. This then presents a problem which Crispin Wright summarises as follows.

[S]omeone who is credited with implicit knowledge of a meaning-delivering theorem may express his knowledge in an indefinite variety of ways, including, in appropriate contexts, lying, assent and silence. But the (implicit) knowledge of a meaning theoretic axiom would seem to be harnessed to the single project of forming beliefs about the context of sentences which contain the expression... [Wright 1986: 227-8].

Thus whilst the output of a theory of meaning, codified as a potentially infinite set of instances of the T-schema or meaning theorems, can play a role in the broader life of a rational subject and thus can be the objects of intentional states or propostional attitudes, the axioms on which such a theory is based cannot.

Evans himself continues to speak of the speaker’s relation to axioms as a form of ‘tacit knowledge’ even though it is not, according to him, a propositional attitude. It is sub-doxastic. Further, because the form of generalisability is a condition on conceptual understanding. Evans takes this to be non-conceptual. So on his picture, a speaker’s grasp of the axioms of a theory of meaning is a non-conceptual, sub-doxastic tacit knowledge.

Given Evans’ argument that a speaker cannot have intentional attitudes towards the axioms of a theory of meaning which are merely dispositions, this presents a problem about the distinction between two possible approaches. Consider a language with a simple subject-predicate structure and ten names and ten predicates yielding 100 different possible sentences of the form Fa. Two distinct theories of meaning are now possible. One comprises a list of a meaning theorem for each of the 100 sentences. The other has one axiom for each of the names, one for each of the predicates and one setting out the meaning of a subject-predicate combination, as an instance of the T-schema (eg ‘a sentence coupling a name with a predicate is true iff the object denoted by the name satisfies the predicate’ [Evans 1981, p. 123]). Given that the speaker does not stand in any intentional attitude to the axioms and need not possess the concepts in which they are expressed, what is the rationale for preferring one theory over the other?

Evans’ suggestion is that the second theory should be preferred if it turns out that a speaker has a disposition for each of the expressions given by the theory. For the list-like theory that is 100 distinct dispositions linking each sentence to its truth condition. For the theory that articulates an underlying structure, Evans argues that there are 20 dispositions: one for each of the names and the predicates. For each name, then for each predicate, the speaker will be disposed to judge the corresponding sentence composed of name and predicate is true if the object named satisfies the predicate. Likewise, for each predicate. Using Π as a universal substitutional quantifier a speaker U has tacit knowledge that a names or denotes John if and only if U has a disposition such that:

(Πφ)(Πψ) [if U tacitly knows that an object satisfies φ iff it is ψ ; and if U hears an utterance having the form φa; then U will judge that: the utterance is true iff John is ψ].

Tacit knowledge that F means bald corresponds to this disposition:

(Πx)(Πα) [if U tacitly knows that the denotation of α is x, and U hears an utterance having the form Fα, then U will judge that: the utterance is true iff x is bald].

As Wright puts it: ‘

‘Tacit knowledge’ ought to be a disposition which constitutes understanding; and what is it to understand a sub-sentential expression... except to be disposed to make the right judgements about the truth conditions of sentences containing it provided one understands the accompanying name or predicate? [Wright 1986: 230]

Hence the 21 axioms yield 20 inter-defined dispositions. If there is evidence that a speaker possess just these dispositions, that is evidence that the second, structural theory, mirrors their competence. (The fact that there are 20 rather than 21, including the axiom of compositionality, is the subject of a criticism by Wright and response by Davies that need not detain us [Davies 1987].) But as Wright points out, even the 20 dispositions are dispositions to make judgements about whole sentences. So, still, why prefer the structured theory over the unstructured list. Both will yield the same dispositions. Evans response is to stress the underlying causal structure that grounds the dispositions.

The semanticist aims to uncover a structure in the language that mirrors the competence speakers of the language have actually acquired. This does not mean , that he aims to uncover a theory that he supposes his subjects know, in any acceptable sense of that word. It means merely this: if (but only if) speakers of the language can understand certain sentences they have not previously encountered, as a result of acquaintance with their parts, the semanticist must state how the meaning of these sentences is a function of the meanings of those parts. He must assign semantical properties to the parts and state the general significance of the construction in such a way that a statement of what those sentences mean is deductively entailed. There may be more than one way of doing this. [Evans identity and predication: 25-6]

Martin Davies develops this thought further in what he calls the ‘mirror constraint’:

The salient structural facts about the competence of speakers are here presented as being of the following form: speakers who understand sentences s1, s2... sn are able, without further training, to understand sentence s. And the salient structural facts about a semantic theory are of this form: the resources used in derivations of meaning specifications for s1, s2... sn are jointly sufficient for the derivation of the meaning specification for s. The constraint on semantic theories... is just that these two structures should match. [Davies 1987: 446]

Tacit knowledge of one theory of meaning rather than another consists in the fact that the structure described by one, rather than the other, is the causal explanation of the speaker’s ability.

To conceive of semantic structure as psychological, rather than abstract, is to conceive of it as the causal-explanatory structure of the semantic ability of actual speakers. It is the kind of cognitive structure that permits speakers to recognize the meanings of previously unencountered sentences. [Davies 1986: 132]

Evidence for this should include, not just patterns of sentence use but also patterns of acquisition and loss of linguistic understanding as well as revision of meaning.

There are in the literature a number of other worries adjacent to this general approach of ascribing tacit knowledge of the axioms of a theory of meaning on the basis of sub-doxastic dispositions. One worry that Wright develops, for example, is whether any account can be given of the dispositions governing either names or predicates in the example above [Wright 1986: 232-3]. The account of the disposition that corresponds to understanding a name has to presuppose the disposition corresponding to understanding a predicate and vice versa. Hence the objection is: no non-question-begging account has been offered of what these dispositions are.

Miller, however, argues that this worry can be assuaged by a comparison with the role of beliefs and desires in rationalising action. In that case, an account of the role of either belief or desire has to presuppose the other. But we, rightly, do not take that to threaten the elucidation of either (or both) but rather demonstrate the holism of the mental. Tacit knowledge of the axioms governing a language has to be ascribed as a whole relative to the mirror constraint [Miller ** 158-9].

But this response to the objection that the dispositions corresponding to states of ‘tacit knowledge’ of a theory of meaning cannot be given non-vacuous clarification does not undermine the earlier point. ‘Tacit knowledge’ of the axioms of the theory is no form of intentional state for the subject.

Given thattalk of tacit knowledge is not very well established in natural language and thus its use is a matter, in part at least, for stipulation, there is no very firm objection to calling the relation between a subject and his or her dispositional grasp of the axioms of a theory of meaning a matter of tacit knowledge. But it is worth stressing the difference between this use of that phrase and the use I have articulated.

First, ‘tacit knowledge’ of the axioms of a theory of meaning is not a matter of belief or any other intentional attitude. It is not conceptually articulated for the subject. It is, obviously, conceptually articulated in theories of meaning and thus conceptually articulated for the theorist. But, aside from their outputs, the contents of such theories are not objects of awareness for subjects.

Second, since they are constituted by dispositions which are not at the service of a variety of distinct projects, what is tacitly known does not play a rationalising role in the life of the subject merely a causal role [Byne 2004: 79].

Third, it is open to question whether such states carry content or meaning at all. All that the mirror constraint clearly does is establish that a theory of meaning tracks causal-enabling states of a subjects physiology. Evans even suggests that neurophysiological data would be decisive in matching a theory of meaning to a subject: behavioural evidence is merely second best. But that surely supports only a causal-structural approach rather than a content-laden one.

In the terms of the dilemma expressed in previous chapters, such a conception may merit the description ‘tacit’ but only at the cost of failing to count as knowledge. By contrast, the account we have given is pitched at the level of the rational subject, is conceptually structured and, whatever causally enables the motor skills may be involved in some aspects of practical knowledge, this does not rule out the fact that the knowledge has a content albeit one which can only be expressed in context-specific ways.

There is a further and interesting difference. The project of articulating a theory of meaning for a natural language is one of formalising the knowledge that a competent speaker has. It aims to codify the ability which is exercised in particular situations in situation-independent and universal terms. Thus what we have taken to be at the heart of tacit knowledge – its situation-specificity – is what this alternative conception aims to trump. Situation-specific expertise is explained through the provision of a general theory. But for the fact that the theory itself is merely ‘tacitly’ known, the project aims to turn what is tacit knowledge for most speakers into explicit knowledge for theorists. (Of course, knowledge of an axiomatised structural theory of meaning for a natural language cannot be a matter of explicit knowledge for most speakers. If it were the project would not be as difficult as it has proved.)

If one does not follow Evans in calling sub-personal causal dispositional states ‘tacit knowledge’ is there any sense in which linguistic understanding involves tacit knowledge? Yes. The right response to Wittgenstein’s regress argument balances what is explicit in explanations of meaning with what is tacit in the sense of situation-specific understanding. To grasp the meaning of a word is to have a potentially unlimited competence in its use even if it is explicable in finite and particular explanations. But such grasp the meaning involves the recognition of any particular use that that! use is correct, accords with its meaning. Such recognition is a context-dependent demonstrative thought which accord with what we take to be the most promising understanding of what is tacit.

Polanyi’s says that we know more than we can tell. One way to approach that claim is to sketch a concept of knowledge in which the substance that we know is somehow hidden from us, not part of the space of our reasons. Tacit knowledge of a theory of meaning is an example. Such knowledge would be tacit but it is far from clear how it would count as knowledge. It is better therefore to keep a firm grip on the nature of knowledge but grant that our ability to know outstrips our ability to summarise what we know.