1 Wittgenstein, Ludwig, Philosophical Investigations, 3rd ed. (New York: MacMillan, 1958)Google Scholar.

2 Davidson, Donald, Inquiries into Truth and Interpretation (Oxford: Clarendon Press, 1984)Google Scholar.

3 Hacking, Ian, “The Parody of Conversation,” in Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson, edited by LePore, E. (Oxford: Basil Blackwell, 1986), pp. 447–58Google Scholar.

4 Lewis, David, “Radical Interpretation,” Synthese, 23 (1974): 333–34Google Scholar. Davidson shies away from such a baldly metaphysical way of talking about meaning. I think the reason is that Lewis's talk of the totality of physical facts about a person can lead us to forget or ignore an essential fact about linguistic meaning. Languages are devices for communicating. Features of languages essential to their uses in communication must be features to which communicators have ready access. Hence whatever facts determine the facts about linguistic meaning must be readily accessible to communicators.

With this warning in mind, I nevertheless follow Lewis. Since our aim is a general understanding of meaning and the propositional attitudes, we should not constrain the theory with facts about how understanding actually goes on, if we can see how the same thing could go on for agents whose epistemic situation is less limited than ours.

5 See Davidson, Donald, “The Structure and Content of Truth,” The Journal of Philosophy, 87, 6 (06 1990): 315, 323CrossRefGoogle Scholar.

6 We do not assume that the interpreter knows the syntactic structure of the sentences; the problem is to work out the syntax and the semantics together. The reason for making the problem this much harder is the idea that truth is prior to reference: “There are two approaches to the theory of meaning, the building-block method, which starts with the simple, and builds up, and the holistic method, which starts with the complex (sentences, at any rate) and abstracts out the parts. The first method would be fine if we could give a non-linguistic characterization of reference, but of this there seems no chance. The second begins at the point (sentences) where we can hope to connect language with behaviour described in non-linguistic terms I suggest that words, meanings of words, reference, and satisfaction are posits we need to implement a theory of truth” (Davidson, Donald, “Realities without Reference,” in Davidson, Inquiries into Truth and Interpretation, pp. 221–22)Google Scholar.

7 Donald Davidson, “Belief and the Basis of Meaning,” in , Davidson, Inquiries into Truth and Interpretation, p. 153Google Scholar. See Fodor, J. and LePore, E., Holism: A Shopper's Guide (Cambridge, MA: Basil Blackwell, 1992) for a discussion of the status of the Principle of CharityGoogle Scholar.

8 If the base of evidence included only expressed sentential attitudes (speech and writing) the problem of radical interpretation would be impossible to solve, since without information bearing on why an agent produces her utterances it would be impossible to discover their meaning.

9 See McDowell's, John “On The Sense and Reference of a Proper Name,” in Reference, Truth and Reality, edited by Platts, Mark (London: Routledge & Kegan Paul, 1980), esp. p. 142, for a nice exposition of this point. There are psychological phenomena which are not propositional in nature, but even these should figure in confirming a theory of meaning, since they are connected with the attitudes in various familiar ways: rage, for instance, with the knowledge that one has been wrongedGoogle Scholar.

10 “Rational” here means, to begin with, rational according to formal theories of rationality, like deductive logic and decision theory. There may be more substantive theories of rationality as well.

11 Davidson, “The Structure and Content of Truth”; “Thought and Talk,” in Davidson, Inquiries into Truth and Interpretation; “Rational Animals,” in Actions and Events: Perspectives on the Philosophy of Donald Davidson, edited by LePore, E. and McLaughlin, B. (Oxford: Basil Blackwell, 1985)Google Scholar; and “Epistemology Externalized,” Dialectica, 45, 2–3 (1991)Google Scholar.

12 Lewis, , “Radical Interpretation,” p. 135Google Scholar.

13 I will leave it unspecified what counts as a community. The crucial point at issue between Davidson and Hacking is not the number of individuals. The crucial point (see below) is whether meaning depends on standards for correct usage which are established prior to occasions of their use.

14 Lewis, , “Radical Interpretation,” p. 125Google Scholar.

15 Davidson, Donald, “A Nice Derangement of Epitaphs,” in Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson, edited by LePore, E. (Oxford: Basil Blackwell, 1986), p. 436. I will not attempt to spell out what “conventional” means.Google ScholarLewis, David provides one theory in his Convention (Cambridge, MA: Harvard University Press, 1969). The suggestion is only that the meaning of an utterance of a word depends on a convention for its use that is established before the conversation begins. The claim that meaning is not conventional is compatible, for instance, with the claim that the norms of rationality are publicGoogle Scholar.

16 Donald Davidson, “Communication and Convention,” in Davidson, Inquiries into Truth and Interpretation, and “A Nice Derangement of Epitaphs.”

17 The ability to generate the new theory need not be based on grasp of conventions for modifying old theories. Davidson identifies the ability with our ability o t invent and confirm empirical theories. See Davidson, , “A Nice Derangement of Epitaphs,” p. 446Google Scholar.

18 Hacking does think that perhaps we need a recursive grammar but we do not need “one uniform recursive system for generating meanings” (Hacking, , “The Parody of Conversation,” p. 456)Google Scholar.

19 Ibid., p. 457.

20 Ibid., p. 458.

21 Davidson, , “Structure and Content of Truth,” p. 316Google Scholar.

22 This problem has at least a family resemblance with other problems familiar in recent work in the philosophy of mind. “Naturalistic accounts” of meaning and thought content—accounts that put “mind in the world order,” as Fodor, Jerry has it in Psychosemantics (Cambridge, MA: The MIT Press, 1987), chap. 4—have related troubles with error. Consider Fodor's Crude Causal Theory: the content of a state of a creature is fixed by what causes it. But whenever an instance of the state exists then so does its cause. So, according to this theory, these contentful states never lack for an object to be about. Worse, they always represent their objects as what they are: as of one of the types of objects that cause instances of this state. So error is impossible. Fodor calls this the Disjunction Problem, and, as he notes, it is an extremely robust problem, arising in a variety of ways for a variety of theories. We may think of Hacking's problem for Davidson as a particularly virulent manifestation of the Disjunction ProblemGoogle Scholar.

23 See Davidson, “A Nice Derangement of Epitaphs,” for a discussion of the notion of literal meaning.

24 Davidson, , “Communication and Convention,” p. 272Google Scholar.

25 Compare David Lewis: “The best scheme [of interpretation] is the one that does the best job overall of conforming to the constraining principles [for Davidson, the Principle of Charity], taking one individual and time with another. (The individuals in question being not only Karl and others of his kind as they actually are, but also some of their might-have-been counterparts)” (“Postscripts to ‘Radical Interpretation’,” in Philosophical Papers [New York: Oxford University Press, 1983], Vol. 1, p. 120)Google Scholar.

26 See, e.g., Tarski, Alfred, “The Concept of Truth in Formalized Languages,” in Logic, Semantics, Metamathematics, edited by Corcoran, John and translated by. Woodger, J. H. (Indianapolis, IN: Hackett, 1983), pp. 152–278Google Scholar.

27 Davidson notes this requirement on a theory of truth that is to yield meanings in Inquiries into Truth and Interpretation (See “Introduction,” p. xiv, and “Reply o t Foster,” pp. 171–79). Specifying which counterfactuals a theory of truth should support is a very serious problem, since if things are different enough then a speaker's words will not mean what they mean in the actual world. Again, it is unlikely that this problem has a solution if the aim is to specify the counterfactuals without use of any notions connected with intentionality. But Davidson's aim is not reductive in this way; for him it is enough if we could specify which counterfactuals are needed without already possessing a theory of meaning for the subject being interpreted.

28 Davidson, , “The Structure and Content of Truth,” p. 313. Davidson does not write that T-sentences are natural laws, only that they “have the form and function of natural laws.” Presumably the conditions for being a natural law are stronger than the conditions for simply having the form and function of a natural law. (See note 27 above; T-sentences are not required to hold at all nomologically possible worlds.)Google Scholar

29 It may seem odd to think of a time-slice theory as lawlike, since one standard criterion for a sentence to be lawlike is that it should apply to new instances. I do not think the oddness is a problem. For time-slice languages, there just are no new instances. Once the language of the speaker changes, as she moves on to a new passing theory she is no longer a source of instances. (Someone else might somehow end up speaking exactly the language of the time-slice; this person could be another source of instances.)

30 See my “Sunburn: Independence Conditions on Causal Relevance,” Philosophy and Phenomenological Research, 52, 3 (09 1993), for a similar point about causal relevanceGoogle Scholar.

31 Another way to know what sentences people would hold true is to know what they would say in the relevant circumstances, and how they would say it. Of course the radical interpreter is supposed not to have knowledge of this sort.

32 This is the point at which following Lewis's more metaphysical approach to radical interpretation is essential, since it is completely implausible to suppose that any actual interpreter has access to the totality of physical facts about a person. A fully satisfactory account of how the Third Principle can be satisfied by the interpreter of a time-slice requires an answer to the problem raised in note 27 above. The problem arises for any interpreter, and hence it cannot be a problem unique to the “duettist.”

33 An actual sentence is a token of a sentence type. It is either an utterance or a mental event of coming to have a sentential attitude, individuated by reference t o the sentence toward which it is an attitude.

34 Meaning remains compositional: each of these indefinitely many term-meaning axioms specifies how tha t term with that meaning may be used in indefinitely many sentences other than the ones toward which the agent actually has attitudes. But compositionality is suffering with respect to the sentences toward which the agent actually has attitudes.

35 See Davidson's “Semantics for Natural Languages,” in Davidson, , Inquiries into Truth and Interpretation, pp. 56–57Google Scholar.

36 Davidson, , “The Structure and Content of Truth,” p. 279Google Scholar.

37 Ibid., p. 325.

38 Millikan, Ruth Garrett, “Truth-Rules, Hoverflies and the Kripke-Wittgenstein Paradox,” The Philosophical Review, 99, 3 (07 1990): 329. See also herCrossRefGoogle ScholarLanguage, Thought and other Biological Categories (Cambridge, MA: MIT Press, 1984)Google Scholar, and “Biosemantics,” The Journal of Philosophy, 86, 6 (06 1989): 281–97CrossRefGoogle Scholar.

39 Thanks to Victoria McGeer for showing me that an explanation is wanted for how the radically interpreted make mistakes, and for continuing to insist that my explanation is wanting. Thanks also to Louise Antony, Jay Atlas, Paul Hurley, Kirk Ludwig, Dugald Owen, Bruce Vermazen and two anonymous readers for Dialogue for valuable suggestions.