2 The strength of this thesis, as well as the more general thesis of the underdetermination of empirical theories by observational data, depends on one's conception of what it is for a class of data statements to support a theory. For present purposes I will follow what appears to be Quine's latitudinarianism on this subject. Theories, together with auxiliary observational statements, make (entail) observational predictions. (Which statements count as observational for this purpose will not be an issue here.) A set of such observational predictions supports a theory to the extent that the theory, supplemented by true auxiliary observation statements, entails the members of the set. Two theories (appropriately supplemented with auxiliary observational statements) that entail the same members of the set, are equally well supported by the set.

3 I am assuming, in order to simplify the argument, that words are the minimal meaning bearing units, that languages contain finitely many such words, and that the translation of the infinitely many phrases and sentences of the two languages is the result of (i) the translation of the words that make them up plus (ii) combinatorial principles specifying the translations of syntactically complex expressions in terms of the translations of their parts.

4 See chapter 2 of Quine, Willard V.O. Word and Object, (Cambridge, MA: The MIT Press 1960)Google Scholar.

5 The importance that Quine attaches to stimulus meanings in testing theories of translation seems to reflect the importance one would intuitively place on information about the circumstances in which speakers would regard sentences of their language as expressing truths vs. the circumstances in which they would regard their sentences as expressing falsehoods. The circumstances in which speakers would assent to S seem intended by Quine to approximate the circumstances in which speakers would take S to express a truth; the case is similar with dissent and falsehood. It is worth remembering, however, that these are only approximations. Various nonsemantic factors — conversational implicatures, politeness, etc. — may bring it about in particular cases that one is unwilling to assent to S even though one believes it to be true, or that one is willing to assent to Seven though one does not believe it to be true. Once this is realized, principles like Quine's (i) may seem less plausible than corresponding principles in which reference to stimulus meanings is replaced by reference to the circumstances in which speakers believe sentences to express truths and those in which they believe them to express falsehoods.

6 See page 78 of Word and Object and pages 46-8 of Quine, Willard V.O. ‘Ontological Relativity,’ in Ontological Relativity and Other Essays (New York: Columbia University Press 1969)CrossRefGoogle Scholar.

7 See Quine, W.V.O. ‘On the Reasons for Indeterminacy of Translation,’ Journal of Philosophy 67 (1970) 178-83CrossRefGoogle Scholar; and ‘On Empirically Equivalent Systems of the World,’ Erkenntnis 9 (1975) 313-28, at 322. By logical incompatibility I mean (and I assume Quine also means) the relation standardly defined — either model-theoretically or substitutionally — in logic books, as opposed to looser notions of ‘conceptual incompatibility’ or ‘analytic incompatibility.’

8 See Quine, W.V.O. ‘Reply to Chomsky,’ in Words and Objections, Davidson, D. and Hintikka, J. eds. (Dordrecht: Reidel 1969), 303Google Scholar.

9 In speaking of ‘Quine's behaviorism,’ one should distinguish several different theses that Quine himself is not at pains to separate. One thesis is that all linguistic facts are determined by purely behavioral facts, nonintentionally described. Since claims about Quinean stimulus meanings themselves appear implicitly to invoke intentional notions, facts about stimulus meanings may not be among the meaning-determining facts recognized by this thesis. A second thesis — which is not really a version of behaviorism in the strict sense — expands the range of facts that are alleged to determine linguistic facts so as to include both purely behavioral facts and facts about Quinean stimulus meanings. A third thesis maintains that facts about stimulus meanings alone determine all linguistic facts. The first of these theses — which may not be able to accommodate even facts about stimulus meanings — is not of much use to Quine, whereas the second and third lack any principled motivation. In addition, in my view all three theses are false (on any reasonable interpretation of what it means for one set of facts, or truths, to determine another).

10 This is a preliminary statement of Quinean physicalism. In section IV I will take a closer look at several questions regarding how the doctrine should be formulated.

11 I use italics as a form of (single) quotation for sentences and phrases, especially when the quoted sentence or phrase itself includes single quotes.

12 The connection between the two Quinean routes to the indeterminacy of translation — one appealing to behaviorism and one appealing to physicalism — may be thought of as follows. Quine seems to assume that facts about stimulus meanings are determined by the physical truths. (This assumption could be questioned, due to the apparently intentional character of assent and dissent, but let it pass.) He also thinks that truths about stimulus meanings do not determine claims about meaning made by theories of translation. If his version of behaviorism is correct, then it follows that theories of translation aren't determined by any facts. Suppose, however, that one rejects his version of behaviorism while still accepting physicalism. One could then appeal to additional physical facts, over and above facts about stimulus meanings or the physical facts allegedly determining stimulus meanings, in an attempt to specify the totality of facts that determine the claims made by theories of translation. Quine's thesis of the underdetermination of translation by physics is aimed at defeating this strategy.

13 David Chalmers and Frank Jackson are two philosophers who seem to believe that (certain) physicalists may be on solid ground at least to the extent of claiming that semantic statements are necessary consequences of the physical truths only if they are a priori consequences of those truths. See Chalmers, The Conscious Mind (Oxford: Oxford University Press 1996)Google Scholar, and Jackson, ‘Armchair Metaphysics,’ in Philosophy of Mind, Michael, M. and O'Leary-Hawthorne, J. eds., (Dordrecht: Kluwer Academic Publishers 1994)Google Scholar; Jackson, ‘Postscript to “What Mary didn't Know,“’ in Materialism, Moser, P.K. and Trout, J.D. eds. (London: Routledge 1994)Google Scholar; Jackson, ‘Finding the Mind in the Natural World,’ reprinted in The Nature of Consciousness, Block, N. Flanagan, O. and Guzeldere, G. eds. (Cambridge, MA: The MIT Press 1994)Google Scholar; and Jackson, From Metaphysics to Ethics: A Defense of Conceptual Analysis, (Oxford: Oxford University Press 1998)Google Scholar. For a good critical discussion of Chalmers and Jackson, see Byrne, Alex ‘Cosmic Hermeneutics,’ Philosophical Perspectives 13 (1999).Google Scholar

14 It is important not to confuse the obvious fact that we know what our words, and those of our neighbors, mean with the utterly implausible claim that we arrive at this knowledge by deriving true claims about the meanings of our words, and those of our neighbors, as a priori consequences of purely physical truths. In the case of translation, it is clear that evidence involving speakers’ assent to and dissent from sentences in various situations does constitute part of the evidence that bears on our acceptance of one or another system of translation. Thus Quinean stimulus meanings do have a genuine epistemic bearing on translation. However, (i) as we have seen, it is doubtful that claims about stimulus meanings are themselves epistemic consequences of the set of purely physical truths, (ii) claims about stimulus meanings, and other current behavior of speakers, do not exhaust the evidence bearing on translation (see note 9), and (iii) the relationship between theories of translation we accept and our evidential base for such theories is not one of a priori consequence.

15 This result could be avoided if it could be shown that there are genuinely a priori semantic definitions of ‘gene’ and ‘DNA’ from which, together with the set of purely physical truths, the theoretical identification of genes with DNA could be derived. I am doubtful that such definitions exist. However, I will not attempt to prove this. By the same token, I offer no conclusive proof that there aren't genuinely a priori semantic definitions of meaning and reference from which, together with all physical truths, our ordinary claims about meaning and reference can be derived.

16 Here, when determination is defined in terms of logical consequence (plus auxiliary truths) it is natural to take ‘P’ and ‘Q’ to range over sets of sentences, rather than sets of propositions (statements). Although this, presumably, would be congenial to Quine, it contrasts with my favored way of understanding previous construals of the determination relation. When determination is construed as necessary consequence, or a priori consequence, it is natural take it to be a relation between sets of propositions. Fortunately, at this stage of the discussion the choice between sentences and propositions is not crucial, and we may consider candidates for the determination relation involving either one. Later, in sections III and IV, significant issues related to this choice will be discussed.

17 This conception of determination is closely related to familiar conceptions of theoretical reduction, which are used by Friedman, Michael in ‘Physicalism and the Indeterminacy of Translation,’ Nous 9 (1975) 353-73Google Scholar to characterize Quine's theses of physicalism and the indeterminacy of translation. There Friedman recognizes two kinds of reduction, strong and weak. Strong reduction is reduction in the classical sense. A theory T₂ is classically reducible to a theory T₁ iff the theorems of T₂ are logical consequences of T₁ together with a set D containing a ‘definition’ for each primitive predicate ofT₂.

18 In this discussion I ignore certain practical complications such as the fact that some speakers speak more than one language, the fact that words of the language may be ambiguous, and the possibility that sometimes there may be no translation of a word in one language onto a word or phrase of the other language. Although these are real factors in translation, they are peripheral to Quine's philosophical claims about translation.

19 Here and in what follows bold italics are used to play the role of corner quotes.

20 In Ontological Relativity and Other Essays (New York: Columbia University Press 1969).

21 (R6) is true, on this account, because its consequent — for all speakers x and expressions α, α, as used by x, refers to all and only rabbits, only if it also refers to all and only temporal stages of rabbits, and so on — is vacuously true. (R7) is false because its antecedent, which is the same as the consequent of (R6), is vacuously true and its consequent is false. A slightly different rendering of the argument could be given by building into the consequent of (R6) (and antecedent of (R7)) the existential claim that some expression does refer to all and only rabbits. On this reading, (R6) turns out to be false, on Quine's view, while (R7) comes out true. A point in favor of this reading is that it fits Quine's comment in the quoted passage to the effect that the consequent of (R6) is ‘absurd.’

22 In section IV I examine the possibility of rejecting all claims, including disquotational ones, that purport to specify which objects words refer to, in the ordinary sense, while nevertheless accepting — as true — the thesis that one's words, and those of others, refer to some objects or other. For the moment, however, I will put aside this unusual possibility in order to examine the consequences of the more natural interpretation of Quine, according to which he is a complete eliminativist about ordinary reference. Once this position is shown to be self-undermining I will turn to the alternative possibility in order to determine whether or not it is viable.

23 Quine, ‘Ontological Relativity,’ 48-9Google Scholar

24 Quine, W.V.O. Pursuit of Truth (Cambridge, MA: Harvard University Press 1992)Google Scholar

25 The alternative, to take Quine to be using his two notions of reference to give an account of what we ordinarily mean by ‘refers,’ is to attribute to him what, from an ordinary point of view, is an obvious falsehood. When I say that someone else's expression refers to rabbits, I am not saying anything about myself or my present language. This is shown by the fact that (i) what I say could be true in a counterfactual situation in which I don't exist, or don't speak a language at all, and (ii) a person could believe the proposition I express by α, as used by Maria, refers to rabbits without believing anything about me or my language, as evidenced by the fact that my remark, John believes that α, as used by Maria, refers to rabbits, might be true even though John has no beliefs about me or my language. Even putting aside counterfactual situations, as well as considerations involving propositional attitude ascriptions, my remark that someone's expression α refers (in the ordinary sense) to rabbits might be true (in the actual context of utterance) in a case in which there is no expression in my language that means the same as α, so long as ‘rabbit’ in my language is coextensive with α.

26 Quine seems to intend to include such statements.

27 It should be noted that (iii) is not very plausible, as it stands. As formulated at present, it does not require the translation manual T I have adopted to satisfy any conditions of empirical adequacy. For example if I have adopted a translation manual T that translates your word ‘rabbit’ onto my phrase ‘prime number,’ then, according to (iii) my remark You use ‘rabbit’ to refer to prime numbers will be counted as true. But that seems unreasonable.

28 Because there are several Quinean options for interpreting claims like (i), and no clear specification of a determination relation that would allow us to settle precisely what is determined in Quine's sense and what is not, the conclusion that, on his view, claims like (i) are indeterminate should be regarded as plausible, but not absolutely conclusive. Should it turn out that there is of way of construing these claims as determinate in his sense the argument that follows would not be significantly affected.

29 For more on the difference between ordinary reference and disquotational, or Tarski-reference, as well as the related difference between ordinary truth and Tarski-truth, see chapters 3 and 4 of Soames, Understanding Truth (New York: Oxford University Press 1999)CrossRefGoogle Scholar.

30 See 220-1 of Word and Object.

31 For an extended discussion of the difference between Tarski's truth predicate and our ordinary notion of truth see Chapter 4 of Understanding Truth.

32 Quine routinely uses the term ‘refers’ both for the relation between a singular term and its extension and for the relation between a predicate and individuals to which the predicate applies. (Sometimes in the semantic literature ‘refers’ is used only with singular terms, and ‘applies’ is reserved for the relation between predicates and individuals.) In this discussion I follow Quine's broad use of ‘refers’; similarly for ‘Tarski-refers’ and related notions.

33 See chapter 4 of Understanding Truth.

34 Note, the problem is not that the theorems of the physical theory adopted at a given time may leave out some physical truths, but that the physical language in which that theory is formulated is not capable of expressing all the physical facts.

35 In my seminar at Princeton in the fall of 1996, David Lewis raised the possibility that the thing that does the determining might not have to be a set of sentences, but rather might be construed as something non-linguistic — e.g. an arrangement of elementary particles. Perhaps. But when we try to say what an arrangement is, and what it is for various claims to be determined by the arrangement, it is difficult to avoid appealing to ordinary semantic and intentional notions that Quine's theses are meant to eliminate. For example, an arrangement might be identified with a set of (physically identical) possible worlds, and a proposition might be declared to be determined by such a set iff it is true at each member of the set. But then we have appealed to a notion of truth that goes beyond Quine's eliminativism. Depending on what possible worlds are taken to be — e.g. maximally complete sets of propositions, or maximal properties that the universe might have had — mere reference to such worlds may also be inconsistent with his eliminativism.

36 A nice discussion of this sort of self-undermining eliminativism can be found in chapter 4 of Byrne, Alex ‘The Emergent Mind,’ unpublished dissertation, Princeton University (1993)Google Scholar.

37 If the physical truths determined that it is not the case that ‘rabbit’ refers to rabbits, then it would not be indeterminate whether ‘rabbit’ refers to rabbits, as opposed to temporal stages of rabbits, and so on. Repeating the same results for temporal stages of rabbits, and all other potential referents, would eliminate all potential indeterminacy.

38 I take propositions to objects of propositional attitudes (like belief and assertion), bearers of truth value, and semantic contents of sentences. For an explanation of why propositions must be structured, and an elementary account of what that structure consists in, see Soames, ‘Direct Reference, Propositional Attitudes, and Semantic Content,’ Philosophical Topics 15 (1987) 47–87CrossRefGoogle Scholar; reprinted in Propositions and Attitudes, Salmon, N. and Soames, S. eds. (New York: Oxford University Press 1988)Google Scholar. See also Salmon, Nathan Frege's Puzzle (Cambridge, MA: The MIT Press 1986).Google Scholar

39 Although this inconsistency would not arise if (i) and (ii) were reformulated to state only that an atomic proposition is true iff it is determined by the physical truths, such a reformulation would not resolve the underlying difficulty — since one would still be faced with the problematic result that some disjunction, say, might itself be determined by physics, and so have what would appear to be legitimate truth-making credentials, even though it was characterized as untrue by virtue of the fact that none of the disjuncts were determined by physics.

40 What about the class of physical truths themselves? Might there be some atomic propositions — say about the position or velocity of certain particles — such that none of them can correctly be characterized as true even though certain disjunctions of them are true? Let us suppose that this is not ruled out. Then the way to formulate physicalism is to start with the physical truths, however characterized, and to formulate physicalism as a thesis about what it is for any nonphysical proposition to count as true. In the case of formulation 2, we simply understand the three conditions as stating conditions on the truth, or untruth, of nonphysical propositions. If it turns out that some physical truth is a disjunction of physical claims none of which can correctly be characterized as true, this will have no effect on our of understanding physicalism.

41 According to formulation 2 of physicalism the fact that the atomic proposition expressed by ‘gavagai’ refers to x (for any value of ‘x’) is not determined by the physical truths means that it is not true, and that its negation, the proposition expressed by ‘gavagai’ does not refer to x will be true (for any value of ‘x’).

42 Among these features are the needed determination relation and the required ‘gappy’ formulation of physicalism. On the subject of truth value gaps, it is worth noting that formulation 3 of physicalism requires a rather special kind of gap. For an extended discussion of gaps of this kind (though not in connection with Quinean theses) see chapters 6 and 7 of Understanding Truth. My view is that gaps of this special type do exist, and can be motivated quite naturally in certain cases. However I don't know of any natural explanation of how truth-value gaps of the required type could arise from facts about which propositions are, and which are not, determined by the physical truths. Thus I am skeptical about this aspect of the modified Quinean position.

A variation on formulation 3 of physicalism that posits a different kind of truth-value gap is as follows:

Physicalism (Formulation 3^*)

When this view is combined with the above assumptions about the determination relation, one gets the result that the claim that ‘gavagai’ refers to a given rabbit is not true, even though its negation, the claim that ‘gavagai’ doesn't refer to that rabbit, isn't true either.

Sometimes this type of three-valued system is combined with the method of supervaluations, which counts a logically compound sentence as true iff it comes out true on all classical extensions (in which all truth-value gaps are closed), false iff it comes out false on all classical extensions, and neither true nor false otherwise. Views of this sort typically allow some disjunctions to be true even though each of their disjuncts is not true, and some existential generalizations to be true even though each of their instances is untrue. In my view, this is untenable. It is part of what we mean by ‘or’ that Either A or B or C cannot be true if A is not true, B is not true, and C is not true; and it is part of what we mean by there is at least one F that There is at least one F that is G cannot be true if for each individual in the extension Of F, the formula x is G is not true relative to an assignment of that individual to ‘x.’

If I am right, this has consequences for the modified Quinean position. Since that position claims that certain disjunctions are determined by physics, and are thereby true, even though none of their disjuncts are determined by physics, it must not claim that these disjuncts are not true. Similarly for existential generalizations and their instances. Consequently, the modified Quinean position should not be taken to include formulation 3^* of physicalism. Rather, something like formulation 3 is required, with the special interpretation of truth-value gaps mentioned above.

Quine himself was, of course, no friend of truth-value gaps of any kind, and so he did not provide an interpretation of the kind of gaps needed by the modified Quinean position. As a result, this position cannot literally be attributed to him, but must be regarded as a reconstruction, or idealization, that may capture some of his actual intentions.

43 One may take an instance of a universal generalization to be the result of erasing the quantifier and substituting an occurrence of a single logically proper name for each occurrence of the newly free variable.

44 There is a complication here worth noting. When both B and A iff B express singular propositions, there are certain natural assumptions according to which it is possible for an agent to know both propositions while being in no position to infer by a priori reasoning the proposition expressed by A. For example let B be the sentence ‘Ruth Barcan published on quantified modal logic in 1946’ and let A iff B be the sentence ‘Rudolf Carnap published a system of quantified modal logic in 1946 iff Ruth Barcan published on quantified modal logic in 1946.’ Now suppose (i) that sentence B expresses the same singular proposition as sentence C, ‘Ruth Marcus published on quantified modal logic in 1946,’ (ii) that Ralph understands all the sentences, but does not know that sentence B expresses the same proposition as sentence C, and (iii) that Ralph believes the proposition expressed by A iff B in virtue of accepting the sentence A iff B, and Ralph believes the proposition expressed by B by virtue of accepting sentence C. On these assumptions Ralph believes the proposition expressed by B and also the proposition expressed by A iff B even though he is in no position to come to believe the proposition expressed by A by a priori reasoning alone.

45 Word and Object, 52-3