(p.281) Appendix to Essay 10: Belief and the Basis of Meaning (1974)
(p.281) Appendix to Essay 10: Belief and the Basis of Meaning (1974)
Abstract and Keywords
Davidson responds to David Lewis’ and W.V. Quine's replies to Essay 10.
He discusses the various interpretative principles Lewis improved on (Charity, Rationalization, Truthfulness) and then raises the issue whether, as Lewis seemed to suggest, we can interpret utterances and beliefs independently of employing intentional idiom (that is, whether we can employ a reductive, physical framework to get the same interpretative results). Agreeing with Lewis on the supervenience of the mental on the physical (but disqualifying his ontic monism in a new footnote as not even ‘materialist’), Davidson stresses the analytical irreducibility of the former to the latter (he refers us to Essay 11 of his Essays on Actions and Events). In relation to both Lewis and Quine, he discusses the degree of determinacy ineliminable from his truth‐theoretic approach to interpretation.
David Lewis says that ontological parsimony is not the subject we're discussing, so we may as well assume any entities that can do us any good. That is exactly my attitude, and always has been. Abstract entities take up no space and cost us nothing. The question I raised was whether or not things like propositions, meanings as entities, and objects of beliefs do do us any good when we want to understand language and thought.
Now for a more important point. Lewis raises the question how a relevantly complete physical description of an agent (call him Karl) over time (the P corpus) is related to a correct description of Karl's propositional attitudes expressed in our language (Ao), Karl's attitudes expressed in Karl's language (Ak), and the meanings or truth conditions of Karl's sentences (M). P, he says, determines all the rest. There is a clear sense in which this seems to me to be right, and it's an assumption I accept along with Lewis. But there's that word ‘determine’. All the things that go in boxes Ao, Ak, and M are supervenient on what goes into P. Lewis gives the same characterization of supervenience as I would (the same one as Moore gave, although he was giving it for a special case): you can't have two things that are exactly alike in all of the things that you would list under P and differing in the things you would list under Ao, Ak, and M. This is a ‘determining’ I accept; it's a mild form of materialism.1 But there's (p.282) another sense of ‘determining’ that Lewis seems to think follows from this characterization of supervenience or is even identical to it which I would put in a different compartment. This is the idea that given all the truths in P you have all of the evidence you will ever need or could use in deciding what goes into the other boxes. This I think is false. I don't mean to question Lewis's remark that there is information in P that I've thrown away or wasted. That's a different point, and it may be right. But the idea that the whole evidential base could come from P seems to me wrong: some evidence will from the start describe in intentional terms things an agent does. But maybe this is too big a topic to debate here.2 Whatever is in Ak comes from P, according to Lewis; my question was how far we could get with what comes out of Ak. If Lewis can show that we can, or must, use more of what comes from P than is used in Ak, then he has scored a real point and I ought to go along with it.
I accept the constraining factors Lewis lists except for one to which I shall come, but perhaps he's done the work needed for me to respond by listing the things his method would take advantage of and my method would throw away.
Let me start with Lewis's improved Principle of Charity. The improved principle of charity says why assume that Karl is right about everything when you can see that he's behind a post and can't notice what's going on behind it, and so forth. Surely we can do better by allowing for explicable error. I agree. The improved principle of charity, insofar as it says there are cases where you can make exceptions right from the beginning, is what I espouse. I'm worried a little, though, about the very sophisticated way in which Lewis put it when he said, imagine we all start off with a common set of beliefs which change as they are exposed to the world. We imagine that Karl does what he did and I do what I do, and we construct his resulting beliefs, not by making his beliefs like mine, but rather by imagining what I would have believed if I'd done what he did and been where he was. As a roughly stated principle I feel there's a lot to it; as a sharp principle I'm not sure that it doesn't assume much too much about what must be going on, or should be going on, in Karl's head. (p.283) But it's an attractive idea. Whatever line we take on this, the central point that I had in mind is that up to the point where we get to work on the theory it's essential that attributions of all of these beliefs is something we do in order to get our method working and not something independently testable belief by belief. With this Lewis seems to agree.
The second point concerns the Rationalization Principle. I see two ways in which this could be understood. One would be to forge ahead with Ao before trying to get anywhere with Ak and M, on the grounds that this assumption about what Karl wants and believes will explain his observed behavior. Well, it's just a question of how far we can get with it. I've argued we can't hope to reconstruct detailed beliefs and desires until we also work out a theory of interpretation of what Karl says, and this still seems right to me. On the other hand, if the point is that once you've got the theory you should check it out and see whether or not it rationalizes not only speech behavior but the rest of behavior, there I do agree and presumably that does bring in some information from P. At that point of checking, if the way in which I described my method left that out, then that is indeed a fault. In other words there's something to test the theory against besides just speech behavior, and that's the rest of behavior. And I don't say that in passing as if it were only a little difference. I think it's a big and important one.
Finally, how about the Principle of Truthfulness? This is the constraint I question. How is it supposed to work? If it says that we can reasonably expect people to say what they believe is true, then I doubt that it's right. If it means that we are justified in being annoyed at them if they represent themselves as believing something when they don't, that's obviously true. But it's not clear what use we can make of that in constructing the theory. If the point is that once we have a fairly complete theory of interpretation and a set of beliefs we can then check among them to see whether or not some of the attitudes towards speech acts are geared to honest intentions, and so forth, then that's surely correct. Whether that's just more of the principle of charity I'm not quite clear. But whether we call it that or not it's certainly the kind of checking we're going to be doing throughout.
The last point had to do with indeterminacy. What Lewis said was that if there's indeterminacy of a serious sort, the kind I gather that Quine has insisted on, and which I certainly think in principle can (p.284) arise, then that proves there aren't enough constraints. But isn't that the very issue? The idea of the method I suggested is to show how we can have a workable theory of interpretation and a workable way of attributing beliefs without assuming that indeterminacy can be eliminated.
Now I turn to Quine's comments, which are generous, as usual. On most points we are in agreement, but here and there I would like to respond to question or criticisms. Quine wonders how my principle of charity across the board as a first step in constructing a theory of truth is going to work. He observes that the principle may work well enough for a sentence like ‘Es schneit’, but he thinks indeterminacy will enter with more theoretical (or less observational) sentences. I too think it probably will. But the degree of indeterminacy will probably be less if my suggestions are followed than if Quine's are, for two reasons. First, I want to apply the principle of charity to theoretical sentences as well as to others: I consider it a count against a theory of interpretation that it makes aliens wrong (by my account) about anything. Against it only prima facie, of course; when a final compromise is struck, error will emerge. Secondly, by insisting on a theory of truth, my approach introduces formal constraints that do not apply in Quine's radical translation. The formal constraints are that the theory be finitely stated, and that it entails a T‐sentence for each sentence of the object language. The restriction to finite axiomatizations eliminates certain sorts of triviality, and forces acceptable theories to reveal significant structure. Asking for T‐sentences as theorems keeps the theory of truth in touch with our best intuitions about truth, and makes the theory testable at a level where no semantics enter the statement of truth conditions. This is well illustrated by quantified sentences: a standard theory with a recursion on satisfaction will entail ‘ “Everything flows” is true if and only if everything flows’, while a substitutional theory with a recursion on truth will only entail ‘ “Everything flows” is true if and only if every sentence formed by substituting a singular term for “everything” in “Everything flows” is true’. From my point of view, a theory that yields only the latter result is unsatisfactory since it cannot be tested at the ground level, and the resulting theory consequently cannot be used for radical interpretation.
If we give up T‐sentences as a test of a good theory, then, we fail to connect our account of the concept of truth with paradigmatic cases of its use. But by retaining the test, as I would, we eliminate substitutional (p.285) quantification as a viable option; thus a source of indeterminacy that Quine mentions disappears. Quine worries that by insisting on provability we may be forcing a decision about what counts as necessary in the metalanguage. Of course there is no suggestion that the axioms are necessary truths; we are dealing with empirical theories. Some logic in the metalanguage must, indeed, be assumed. But this is the case whether there is a recursion on satisfaction, as I would urge, or on truth, as Quine is inclined to allow. In either case it takes logic to apply the account to the cases.
When I said that T‐sentences need only be true, I was not discussing their relation to the account of truth, but how they were to be tested. For Tarski logic alone guarantees the truth of T‐sentences; in an empirical theory, it is necessary to find empirical evidence.
Finally, a remark about the irreducibility of psychological terms and the indeterminacy of translation or interpretation. Of course I agree with Quine that indeterminacy infects not only the interpretation of sentences about psychological attitudes but also sentences whose subject matter is physical. This does not show, however, that the special character of the attitudes is not connected with indeterminacy generally because, even if we forget object language sentences about attitudes, the beliefs of speakers of that language enter directly into the construction of a theory of interpretation, and it is the tradeoffs between the beliefs the theorist attributes to speakers and the meanings the theorist assigns to their words that creates indeterminacy. I argued that even if such indeterminacy were, per impossible, eliminated, the irreducibility of the mental to the physical would remain. (p.286)
(1) I would now hesitate to call this materialism. All that this sort of supervenience enforces is a single ontology of objects and events, a form of monism. The objects and events are sometimes described in physical terms, sometimes in physiological terms, sometimes in psychological terms: the same things are described now as physical, again as mental or biological, etc. The only reason to call this physicalism would be if one accepts the additional claim that the ideal physics is more precise or complete than any other system of description and/or one thinks definitional or nomological reduction of the vocabularies of psychology or biology to the vocabulary of physics is possible. But why believe this? [Footnote added in 2000.]
(2) My reasons for this claim are given in ‘Mental Events’.