page 2 of Perfect Speaker Theory

beliefs are transparent to himself. Another and consequent reason is that an OOS has the duty of restatement, i.e., he would fail to be doing his best if he did not stand ready to clarify or defend what he said if asked by linguistically competent hearers to do so. But his obligations here are a function of the defeasibility of his attempts to do his best as a speaker. By my definition a PS is one to whom this duty is inapplicable for the reasons given.

So a PS is better than optimal; and he therefore comes between an IS and an OOS, although he is somewhat closer to the latter than the former. A second stab at a definition adds these comments to conditions (1) to (3).

It should be immediately added, if it is not already clear, that the PS is conceived as one whose interlocutors are always ordinarily competent speakers of the same language. He is not a citizen of a kingdom of PSs.

Some comments are needed on the conditions. The third, that the PS is epistemically cautious in the way described, suggests that he is governed by an ethics of epistemic caution which at least imposes an obligation not just to be sensitive, but to articulate or to be ready to articulate sensitivity, to the possibility of epistemic defeat of any claim presupposed or made on its occasion. This condition is vital to resolving certain problems, one of which I consider. But it is not a requirement that a PS should be regressively hedging about his claims, making explicit such protases to his remarks as 'if there is a world at all...' and the like. A PS is not by stipulation absurd, only pedantic. But it is a corollary of this duty of (so to speak) maximal epistemic caution that the PS be as well-informed as is required to fulfil conditions (1) to (2). This is not a demand that the PS be an IS, i.e., omniscient: it is rather that his use of the expressions of the language should be conformable to what I shall later describe as a Best Dictionary for the Language–that is, one which makes use of the best current theories of what use of the language's expressions constitutes talk of.

A comment on the PS himself is prompted by the first two conditions and his correlative freedom from the duty of restatement, and this comment in turn underlines a substantial point often made about a feature of language. In line with (1) to (2), the PS never indulges in metaphor or irony if there is a risk of misunderstanding–which even with normally competent hearers there often enough is; and he never indulges in ambiguity. He is, in short, in danger of being a bore. But his potential boringness is interesting in this respect: that it reveals one of the constraints imposed on him by language (there are plenty of others). For whereas by (1) and (2) he shuns ambiguity, he cannot avoid vagueness, except by stipulative means, which he will not anyway normally wish to resort to. He is of course minimally vague: a major use of certain resources in the language, namely adjectives and adverbs, is the reduction of vagueness, allowing both ordinary speakers and a PS to be maximally specific; maximally but not completely, because vagueness is a built-in feature of language upon which a good deal of its utility turns. But this does not generate an inconsistency with condition (2), which is that the PS expresses exactly his intended meaning, for one can exactly mean to say what cannot be expressed otherwise than by use of a vague expression. For example: suppose the PS says, 'X is bald'. That can be exactly what he intends to convey, independently of questions about the degree to which X has less hair on his head than Y to whom the PS does not apply this predicate. It is of course possible for a non-arbitrary stipulation to be made with respect to some vague expression which precisifies it relative to a certain purpose. One can say that a person n % of whose scalp has a covering of fewer than n hairs per some measure, is bald. Suppose registered bald people by law have to wear a white hat on sunny days. Then the legal instrument which enacts this law would have to be precise: trichological police would need a definition to work to. Just such precisifications in fact obtain in registration of blind people to whom welfare benefits are due.

But as we see, a PS would normally neither need nor desire to go for precisification of vague expressions, however non-arbitrary relative to a purpose: for their vagueness is exactly what from time to time he needs. The constraint they impose is not a limitation.

These comments together give us a second stab at a definition of a PS which will suffice for the present. I shall sometimes speak of the three conditions as rules which bind the PS in his practice. Another important feature of PS-hood, a corollary of the third condition, emerges as the model is applied. I turn to that in a moment. First I will just mention a reason for using the adjective 'perfect' in my label. It is because it is informatively symmetrical with the use of the adjective made by Russell and others in their attempts to define a 'perfect language'. That was a programme aimed at specifying the underlying logical structure of natural language. In addition to the assumption that there is such a thing, there was a further, at least at the outset: that it admits of a uniquely correct representation. The ambition was to set out in algebraic description of logical form something which, like Leibniz's longed-for universal characteristic, would completely and unambiguously represent what is said by any natural languages. And this in turn was held to have exciting metaphysical potential, since the idea was that what there is can be read off from what the language says. Well: all I need say is that I propose we substitute idealisation of the speaker for idealisation of the language: instead of looking for the perfect language, let us try to describe a PS of ordinary language and see where it takes us. I think such a task justifies the assumption upon which it rests, namely, that it is not the language that says things, but its speakers.

I turn now to apply PST to cases. The suggestion is this: application of the theory reveals a certain pattern in what generates the problems, namely, a falling-short, either because of the usual vicissitudes of discourse in an imperfect world, or artificially by hypothesis, of a PS's conditions; and it therefore suggests a solution to them, which is to make explicit appeal to the point of utterance and/or to identify the epistemic deficit requiring remedy.

The first case I consider – and I consider all of them briefly – is that of the natural language analogues of the logical constants, classically interpreted. I assume familiarity with the standard examples, and just register them here. On a certain natural view, one central use of 'and' in English is to convey temporal succession, one central use of 'or' conveys the speaker's ignorance of the truth-value of the disjuncts, and one central use of conditional statements is to assert that accepting an antecedent is a ground for accepting its consequent, and normally that there is anyway uncertainty about whether the situation denoted or described by the antecedent obtains.

The difficulties felt about these natural uses of the natural language operators arise from their divergence from their formal analogues. Their content exceeds that of their formal analogues, and they are therefore more fruitful in implications. Suppose it is given that a particular disjunction is true. We thus know that at least one of the disjuncts is true. But we are not in a position to infer something of a different order, such as that an asserter of this disjunction is either ignorant of the disjunctsí truth-values, or might be dissembling, guessing, joking or some such. The additional content is provided by pragmatic considerations: those specified in terms of the speaker's intentions and certain contextual features.

Faced with questions about these divergences, our inclination is to look for mappings. Strawson, for example, held that the acceptability or truth of a conditional rests on whether acceptance of the antecedent is a ground for accepting the consequent, but that this is not sufficient for a conditional's truth or acceptability, for which the truth of the associated material conditional, entailed by the natural language conditional, is also required. From this it follows

....3/