Quine

Citation preview

Quine Elliott Sober; Peter Hylton Proceedings of the Aristotelian Society, Supplementary Volumes, Vol. 74. (2000), pp. 237-299. Stable URL: http://links.jstor.org/sici?sici=0309-7013%282000%2974%3C237%3AQ%3E2.0.CO%3B2-A Proceedings of the Aristotelian Society, Supplementary Volumes is currently published by The Aristotelian Society.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/journals/aristotelian.html. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers, and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take advantage of advances in technology. For more information regarding JSTOR, please contact [email protected].

http://www.jstor.org Mon Dec 3 22:41:03 2007

QUINE by Elliott Sober and Peter Hylton I-Elliott

Sober

QUINE'S TWO DOGMAS ABSTRACT In 'Two Dogmas of Empiricism', Quine attacks the analytic/synthetic distinction and defends a doctrine that I call epistemological holism. Now, almost fifty years after the article's appearance, what are we to make of these ideas? I suggest that the philosophical naturalism that Quine did so much to promote should lead us to reject Quine's brief against the analytic/synthetic distinction; I also argue that Quine misunderstood Carnap's views on analyticity. As for epistemological holism, I claim that this thesis does not follow from the logical point that Duhem and Quine made about the role of auxiliary assumptions in hypothesis testing, and that the thesis should be rejected.

Q

uine's publication in 1951 of 'Two Dogmas of Empiricism' was a watershed event in 20th century philosophy. In that essay, Quine sought to demolish the concepts of analyticity and a priority; he also sketched a positive proposal of his own-epistemological holism. There can be little doubt that philosophy changed as a result of Quine's work. The question I want to address here is whether it should have. My goal is not to argue for a return to the halcyon days of the logical empiricists. Rather, I want to take stock. Now, almost fifty years after the publication of 'Two Dogmas,' what view should we take of analyticity, the a priori, and epistemological holism, and of what Quine said about these topics? Analyticity is an issue in the philosophy of language, a priority an issue in epistemology. As a first approximation, the two questions are as follows: Are there sentences that are true in virtue of the meanings of the terms they contain? And are there propositions that we can know to be true independently of sense experience?' A central critical objective of 'Two Dogmas' was to dismantle a view that sought to connect these two questions. The 1. The observant reader will notice that I describe a priority as a property of propositions and analyticity as a property of sentences. The reason for this will be discussed presently.

238

I-ELLIOTT

SOBER

positivists' linguistic theory of the a priori claimed that the special epistemological status of a priori propositions derives from the fact that they are expressed by analytic sentences and these sentences are analytic because of the meanings assigned to their constituent terms by the adoption of linguistic conventions. Notice that even a devastating criticism of the concept of analyticity would leave open the possibility that some other defence of the concept of a priority could be given. It is here that Quine's positive proposal is meant to do its work-epistemological holism excludes the possibility of a priori knowledge.

Quine on Analyticity. It would be a mistake to think that Quine's point was simply that nothing is a priori or analytic. He didn't regard these concepts as clear, but empty, like the concept of a round square. Rather, the point was supposed to be that these concepts are 'unclear'. And this didn't mean that the concepts are vague-that there are situations in which there is no fact of the matter as to whether the concepts apply. Being analytic wasn't supposed to be like being bald; it is perfectly clear what baldness is, even though there are grey areas in which there is no saying whether the concept applies.* Quine's main complaint in 'Two Dogmas' against the concept of analyticity is that it can be defined only in terms of other equally suspect concepts, such as meaning, necessity, synonymy, proposition, and definition. For example, one might say that the sentence 'bachelors are unmarried' is analytic because the truth of the sentence follows from the definitions of the terms that occur in it. However, if there are doubts about whether analyticity is a clear concept, there also will be doubts about what it takes for a sentence to express a definition. If this and similar proposals are unsatisfactory, what would count as a 'clear' definition of analyticity? Quine sometimes says 2. The main concept in Quine's epistemological holism is centrality-beliefs differ in their degree of centrality; there is no dividing line between beliefs that are a priori and analytic on the one hand and those that are a posteriori and synthetic on the other. If this were the whole message of 'Two Dogmas', one might reasonably summarize Quine's point by saying that he regarded the analytic/synthetic distinction as a vague one. However, Quine's criticism of analyticity goes beyond this, as we will see.

QUINE

239

that an acceptable characterization of analyticity would have to be specified in terms of observable behaviour (Quine 1953b, 1960, 1970); at other times, he says that extensional semantic concepts, such as reference and truth, are legitimate, and that analyticity would also be legitimate if it could be defined in terms of them (Quine 1953a). But analyticity cannot be defined in terms of behaviour or in terms of concepts drawn just from the theory of reference, so we must reject analyticity as 'unclear'. It is a testimonial to how influential Quine has been with respect to some of his other philosophical ideas that this criticism of analyticity now seems so thin. Quine taught us that the main question we should ask about a theory is whether it is indispensable as a device for explaining and predicting phenomena. The question, then, should not be whether analyticity can be defined, but whether the concept is useful in science. The concept of 'electron' cannot be defined in terms of observables, but that says .~ nothing about what we should think of the electron t h e ~ r yAs 3. Although Quine rejects the notion of analyticity, he occasionally endorses what he regards as acceptably sanitized readings of the concept. For example, Quine (1960) says that a sentence is 'stimulus analytic' if every speaker of the language is disposed to assent to it. And Quine (1974, p. 79) proposes that we say that '... a sentence is analytic if everybody learns that it is true by learning its words.' The sentence 'dogs exist' is probably stimulus analytic. Is it analytic according to the second proposal? Suppose that people learn that dogs exist in the process of learning to use the term. Do they thereby learn that 'dogs exist' is true by learning its words? Well, they learn that the sentence is true when they learn its words. But what about 'by'? The more-or-less traditional concept of analyticity says, roughly, that a sentence is analytic when one can know that it is true just by knowing what its terms mean. Learning that dogs exist may be a standard concomitant of learning to use the constituent terms, but the existence of dogs does not follow from statements that describe what the term means ( p a c e the ontological argument's treatment of a term that has the opposite spelling). Quine clearly does not intend his proposed definition of analyticity to coincide with the traditional concept, so it would be no objection to his proposal if it judged 'dogs exist' to be analytic. However, the proposal can be interpreted to rule this out: if learning the terms = learning what the terms mean, then speakers do not learn that 'dogs exist' is true just by virtue of learning the terms. Is it possible that Quine's proposal is an endorsement of the traditional concept rather than a replacement of it? 4. Grice and Strawson (1956, p. 203) point out that 'reductive analysis' is too much to hope for, and go on to suggest that concepts can be explained in 'less formal ways.' They then say that the analytic sentences are the ones whose denials 'have no sense (p. 205).' I, of course, agree with their complaint. However, I suspect that their suggestion will not accord with the notion of meaning required in linguistics and cognitive science. The denial of an analytic truth does have a meaning; this is part of the reason we are puzzled when people deny analytic truths. There is an important theoretical difference between 'bachelors are sometimes married' and strings of words that are not well-formed.

240

I-ELLIOTT

SOBER

Quine observes in 'Two Dogmas (p. 44)', 'objects at the atomic level are posited to make the laws of macroscopic objects ... simpler and more manageable; and we need not expect or demand full definition of atomic and subatomic entities in terms of macroscopic ones...' It also is worth recalling that Quine felt obliged to affirm that sets exist, since set theory is needed for mathematics and for mathematized natural science. The philosophical attractiveness of nominalism had to give way to the demands of science. Why didn't Quine take the same attitude towards analyticity?5 Quine's criticism of modal concepts has the same character. He claims that the concept of necessity is not 'clear', where the standard of clarity is in fact reductionistic. Necessity has to be definable in terms that are extensional, just as analyticity has to be definable in terms that are behavioural. This cannot be done, so the concepts should be rejected. Whether science would lose anything of value in rejecting these concepts is a question that never got broached. Quine assumed that an extensional language would be adequate for codifying the results and inferences found in mathematized science. Why did Quine bypass science and consider the concepts of analyticity and necessity only in terms of how they answer to 5. Occasionally, Quine has drawn back from his demand that analyticity be defined either behaviourally or extensionally. For example, in Word and Object, Quine (1960, p. 206) says '...we find it argued that the standard of clarity that I demand for synonymy and analyticity is unreasonably high; yet I ask no more, after all, than a rough characterization in terms of dispositions to verbal behaviour.' If 'rough characterization' means only that one be able to describe behaviours that would be evidence that a sentence is analytic in a speaker's idiolect, the requirement is easy enough to satisfy. As Carnap (1953) pointed out, if you ask a speaker whether it is conceivable to them that a sentence (e.g., 'either pigs exist or pigs do not exist') could turn out to be false, and they reply 'no, the sentence is true by definition,' that is evidence that the sentence is analytically true. The inference is not infallible, of course, but empirical evidence is rarely like that in any case. Quine's reply to Carnap's (1953) is his doctrine of the indeterminacy of translation. In Chapter 2 of Word and Object, Quine (1960) claims that the totality of a speaker's verbal behaviour (including not just what he actually says, but what he would say in various circumstances) can be consistent with two mutually incompatible accounts of what the speaker's terms mean. Quine claims, not just that these accounts will be behaviourally equivalent, but that there is no fact of the matter as to which is truehis claim is not just that behaviour underdetermines theory, but that semantic theories are prey to an indeterminacy. I reply, as others have done before, that underdetermination is a property of all theorizing, and that Quine has not shown that underdetermination should be coupled with a claim of indeterminacy in the case of theories about meaning.

QUINE

24 1

some antecedently assumed standard of philosophical clarity? With respect to analyticity, part of the reason may have been that cognitive science was not sufficiently developed when Quine wrote for him to have felt that the practices of science could trump the arguments of philosophers. He probably assumed that the concept of meaning was to a large extent a concept that philosophers owned. As for the concept of necessity, Quine believed that modal discourse is not part of the substance of science proper, but consists in a philosophical overlay that can be peeled away without damaging the properly scientific core. Scientists talk of 'laws' and make counterfactual inferences, but the real content of science, for Quine, is resolutely confined to the actual. This point of view has not stood the test of time. It now seems clear that the concept of probability, as it is used in scientific theories, is a modal concept6 Talk of the half-life of a radioactive material, for example, cannot be glossed in terms of the actual frequency of an event-type in a larger reference class. The point can be seen by asking what it means to say that a coin has a probability of 112 of landing heads when tossed. This doesn't mean that the coin will land heads half the time; a fair coin can be tossed an odd number of times and then destroyed. It once was thought that the statement means that if the coin were tossed again and again indefinitely, that the relative frequency of heads would necessarily approach a limit of 112. This would entail that modality is part of the content of scientific theories that use probabilities, but in fact the concept of objective probability is even less reducible to frequencies than this interpretation suggests. The most that can be said about the relationship of probabilities to hypothetical relative frequencies is given by the Law of Large Numbers (Skyrms 1980)-if a coin's probability of landing heads is 112, then the probability approaches unity that the frequency of heads will be within epsilon of 112, as the run of tosses is 6. I set to one side the way in which probability helps characterize the evidential credentials of scientific theories. Here I am interested in probabilistic concepts (e.g., fitness, half-life) that are used in theories. Also, in saying that probability is a modal concept, I am talking about the interpreted concept, not the uninterpreted formalism. The Kolmogorov axioms of probability can be interpreted in terms of actual relative frequency, and there is nothing modal about this interpretation. However, this interpretation does not do justice to the way that probability is used in scientific theories.

242

I-ELLIOTT

SOBER

made longer. A fair coin doesn't have to land heads 50% of the time, even in the limit. Probability doesn't reduce to hypothetical relative frequency; rather probability claims confer probabilities on hypothetical relative frequencies. Probability is i r r e d ~ c i b l e . ~ In any event, Quine's circularity argument against analyticity and necessity is a prime example of an appeal to 'first philosophy', a practice that Quine later criticized in his essay 'Epistemology Naturalized' (Quine 1969). First philosophy seeks a justification for the propositions we believe that is more secure than the evidence provided by natural science. Philosophers who think that Quine was right to reject first philosophy should be skeptical about the arguments in 'Two Dogmas of Empiricism' against analyticity and modality.8

Carnap on Analyticity. If Quine did not treat analyticity as a theoretical term in empirical science, neither did the the logical empiricist who was the main target of Quine's criticisms. Carnap thought that it is a matter of convention what sentences one regards as analytic. Claims about what is analytic are proposals; the question of truth and falsity does not arise (Creath 1991, 7. Stein (1992, pp. 288-289) makes a complementary point-'...the concept of the space of all states of a physical system is a central one in much of classical physics, and in all of quantum physics; and "all states" means, of course, all possible states.' 8. Although Quine often contrasted his naturalism with Carnap's foundationalism, in fact Quine's metaphilosophical position is quite Carnapian: He who wishes to investigate the questions of the logic of science must, therefore, renounce the proud claims of a philosophy that sits enthroned above the special sciences and must realize that he is working in exactly the same field as the scientific specialist, only with a somewhat different emphasis ... This may sound like the Quine of 'Epistemology Naturalized' (Quine 1969), but it isn't. It's the Carnap of Logical Syntax of Language (Carnap 1937, p. 332). Naturalism is sometimes understood as the view that a conflict between philosophical argument and scientific practice must always be resolved by rejecting the former. I do not understand it that way. If philosophy is in fact continuous with science, then we should be skeptical of any such formula. If philosophical critique is part of the practice of science, then the right resolution of such conflicts will sometimes be to change scientific practice. Fogelin (1997, pp. 554555, pp. 561-563) argues that Quine's naturalism is at odds with his arguments on behalf of the theses of indeterminacy of translation and of reference. For Fogelin, naturalism is characterized by the Peircean commitment to eschew Cartesian skeptical challenges. I see a different relationship between naturalism and skeptical problems, but that is a subject for another essay.

QUINE

243

p. 365; Stein 1992, pp. 279-282; Ricketts 1994). This is the point of Carnap's (1937) Principle of Tolerance-'in logic, there are no morals'. Carnap's views about meaning, and about analyticity in particular, need to be understood within the context of the threepart distinction that he and other empiricists drew among descriptive definitions, explications, and outright stipulations. A purely descriptive definition aims to record what terms actually mean in the language spoken within a community. An explication takes a pre-existing concept and clarifies it; a stipulation starts from scratch, conferring on a symbol a meaning that may be entirely unrelated to whatever meaning it may have had before. It is obvious why there is no philosophical interest in merely stipulating what a term means. But why did Carnap reject the project of recording the meanings of terms as they actually are used? Carnap's reason was that he thought that the terms in a natural language are too vague and confused. From Frege, he inherited the idea that the psychological states of language users vary from person to person and, being subjective, are ultimately uncommunicable; and under the influence of Tarski's theory of truth, he came to believe that natural languages are contradictory. Meaning in natural languages is not amenable to systematic and rigorous scientific study. Although the concept of analyticity has a piecemeal and fragmentary application to natural languages and their speakers (Carnap 1952b, 1953), its rigorous and systematic deployment can take place only within the confines of an artificial language. Since explications are proposals, they are to be judged by their usefulness, not by whether they are true. Still, one of the considerations that enters into the evaluation of an explication is its fidelity to pre-existing usage (Carnap 1950; Creath 1991, pp. 368369). For example, the separation of definitions from empirical claims in an axiomatization of Newtonian mechanics would be intended to clarify the question of what is a priori and what is empirical in that theory-i.e., the theory that scientists had developed before the philosophers came along. In such an axiomatization, the claim that 'F = m a ' is a definition of force would be intended to characterize a theory that was formulated and put to work by scientists; there would be no philosophical point in simply stipulating that 'F = ma' is analytic, or in stipulating that it

244

I-ELLIOTT

SOBER

is not. Philosophers can stipulate all they like; the character of Newtonian mechanics is not affected, and it is the latter that philosophers want to understand. Explication, then, can be thought of as a 'mixture' of description and stipulation, or as a 'compromise' between them. T o offer an explication is to make a normative remark. One is suggesting that terms be used in a way that retains some allegiance to pre-existing usage but also departs from that usage. To give an explication of what a term means is to say something that is neither true nor false. The same point holds with respect to claims that a certain sentence is analytic. For Carnap, to say that 'F = ma' is analytic, or that it is not, is to make a proposal. Thus, Quine and Carnap agree that claims about meaning and analyticity do not describe facts (Creath 1991, Stein 1992, Ricketts 1994), but they agree on this point for very different reasons. Quine takes the subject matter of such claims to be natural languages; there is no fact of the matter about what terms mean or about which sentences are analytic because the concepts of 'meaning' and 'analyticity' are not sufficiently clear. Carnap takes the appropriate subject matter of such claims to be artificial languages that are constructed with the goal of explicating usage in this or that natural language. The concepts of 'meaning' and 'analyticity' are scientifically respectable when deployed within an artificial language, but they are used to formulate proposals, not descriptions. They fail to describe facts for the simple reason that they are not intended to do so. What Carnap and Quine disagree about is whether the analytic/synthetic distinction, and the concept of meaning generally, is useful. Carnap (1963, p. 922) says that the concept of analyticity is 'indispensable for methodological and philosophical discussion'; the analytic/synthetic distinction is important because it serves to separate nonempirical from empirical elements in a system of beliefs. This epistemological distinction is normative; it concerns how we should think about the justifications that different beliefs possess. Quine thinks the distinction is useless, for two reasons. First, he thinks it isn't needed in the empirical study of natural languages. Second, it isn't needed in epistemology either; Quine's epistemological holism, which we will discuss later, entails that there is no distinction between a priori and a posteriori beliefs.

QUINE

245

Analyticity in Natural Languages. Carnap rejected the idea that the analytic/synthetic distinction marks a factual distinction among sentences in a natural language. Did he do so for good reasons? The fact that natural languages are vague and unclear does not show that the analytic/synthetic distinction fails to apply. For example, perhaps 'rich' is a vague term; perhaps there is no precise number of dollars that marks the difference between someone who is rich and someone who is not. Even so, the sentence 'rich people have more money than poor people' may still be analytic. It also is possible that the concept of being rich is useful in the science of economics. Similar points apply to the claim that natural languages are contradictory. What is true is that natural languages permit the construction of paradoxical sentences such as 'this sentence is false'. However, that doesn't show that the language is contradictory; after all, the language is not an assertion, but a device for constructing assertions. In any event, whatever imperfections there may be in natural languages, the task of semantics as an empirical discipline is to describe their semantic properties, warts and all. Philosophers may be more interested in reforming usage, but that does not show that the task of description is ill-conceived. Of course, it does not follow that the concepts of meaning and analyticity as currently understood are sufficiently clear. They are terms of art-in philosophy and in linguistics-and they may need to be elaborated in various ways, perhaps by the construction of formal systems. However, the point of doing so may be that one wishes to describe the semantic features of natural languages.9 Suppose a group of speakers uses a language that is as clear as Carnap might wish. Perhaps they have adopted the proposals set forth in one of Carnap's artificial languages, or perhaps their discourse is already so clear that an explication of what they are saying is really nothing more than a description. In either case, is and ought coincide. In this instance, claims about meaning and analyticity are not merely proposals; they are descriptions. But of 9. The same point holds if we interpret analyticity as Carnap did-not as a term intended to describe natural languages but as an epistemological category. Surely there are sentences for which it is a factual question whether they are a priori or a posteriori.

246

I-ELLIOTT

SOBER

what? What properties of utterances is one describing when one talks about what terms mean or about which sentences are analytic? This question needs to be answered even if one holds, as Carnap did, that the analytic/synthetic distinction makes clear sense only when the language under discussion is artificial. The standard answer to this question provided by logical empiricism is that analytic sentences in some way embody linguistic conventions. They are true, not because of the way the world is, but purely because of how speakers assign meanings to terms. Quine attacks this conventionalist theory of analyticity, so the theory is worth investigating, even if Carnap would have been reluctant to embrace it as an account of meaning in natural language.

Analyticity and Conventionalism. The logical empiricists were attracted to a conventionalist account of analyticity in part because of their conviction that there are important conventional elements in Einstein's special theory of relativity. When a light signal leaves a source, bounces off a mirror, and then returns to the source, what fraction of its travel time does it spend going out and what fraction does it spend coming back? The positivists thought that Einstein's insight was to recognize that any answer to this question is a matter of convention (Reichenbach 1949, Carnap 1966). Einstein chose 1/2, but that was a stipulation, not an empirical discovery. The positivists represented this claim by saying that Einstein had made a linguistic decision; he had adopted a 'linguistic framework' in which (S) A light ray that goes from a source, to a reflector, and back to the source, spends 1/2 its travel time going out from source to reflector and 1/2 coming back from reflector to source. is an analytic truth. Other choices of 'language' are possible, and within those alternative languages, sentence (S) would be analytically false. This was not a happy formulation. A real analytic truth is true, and what it says typically does not depend for its

QUINE

247

truth on our choice of language.'' Given standard English usage, the sentence 'bachelors are unmarried' is true; if English had involved different conventions about what words mean, the sentence might have meant something different, and so it might have been false. However, the proposition expressed by the English sentence-that bachelors are unmarried-does not depend for its truth on how English works." There are bachelors in France who would very much resent the English chauvinism involved in saying that the way English works affects whether (French) bachelors are unmarried. 'Quelle impertinence!' one can hear them exclaim. The fact that bachelors are unmarried is no more dependent on English than it is on French. Genuine (object-language) analytic truths express truths, and the propositions those sentences express are language-independent. Applied to sentence (S), this pattern of reasoning entails that if (Sj is true in Einstein's language, then what the sentence expresses is true, and moreover, it is true independently of how Einstein chose to set up his language. But surely this is not what the positivists wanted to say-their epistemological point is better put by saying that there is no fact of the matter about whether (S) is true. Adopting (Sj may be a matter of convention, but (S) is not made true by anything, not even by adopting a convention. Given what the sentence about travel time actually means in English, it would be a mistake to say that it is an analytic truth or an analytic falsehood. It is neither, and it isn't an empirical truth or an empirical falsehood, either. Conventionalists should say two things: First, (S) is neither true nor false and the same holds for all its sister propositions that assign fractions different from 1/2.12 Second, 10. I say 'typically' because the proposition expressed by a meta-linguistic statement does depend for its truth on facts about a language. Consider, for example, the claim that the term 'mother' means female parent in English. Whether this is so obviously depends on facts about English. However, object-language statements do not have this property. Whether mothers are female parents does not depend for its truth on facts about English. 11. This point was made forcefully by Frege (1884) and has been reiterated by many subsequent writers (e.g., Ewing 1940, Kneale 1947, Pap 1958, Harman 1967, Lewis 1969, Boghossian 1996). It does not depend on any particular 'theory of propositions'-i.e., on any particular view about how propositions are individuated. What is required is some distinction between a sentence and 'what the sentence says' (Boghossian 1996, p. 380). 12. Here the conventionalist disagrees with the skeptical realist, who claims that (S) has a truth value that it is impossible to ascertain. The conventionalist and the skeptical realist make the same epistemological point-that (S) cannot be known to be true-but they disagree about the reason why.

248

I-ELLIOTT

SOBER

an arbitrary choice must be made of one such statement for purposes of theory construction. Including (S) in one's theory is dispensable, but including some such principle is i n d i ~ ~ e n s a b l e . ' ~ It is important to recognize the epistemic status that the conventional elements in one's system of beliefs possess. If (S) lacks a truth value, then no observation can empirically disconfirm it; it is 'immune from empirical revision'. However, this is not because (S) is certain to be true (Ricketts 1994). (S) is a priori in the negative sense that its justification does not depend on experience, but what, if anything, can be said positively about the justification there is for believing it? Here it is useful to recall the distinction between pragmatic and epistemic reasons for believing a proposition, a distinction that is familiar in connection with Pascal's wager. Epistemic reasons have to do with the proposition believed; pragmatic reasons have to do with the act of believing. Pascal says that even if there is no epistemic reason to believe in God, there is a pragmatic one. Even if there is no evidence for the proposition that God exists, there still is a good reason for agents to get themselves to perform the act of believing. Conventionalists assert that there is no epistemic reason to believe (S), though there may be a pragmatic reason to adopt that convention. I hope these comments make it clear why conventionalism as a thesis about the content of various scientific theories provides a bad model for thinking about analyticity. Analyticity has to do with language, but conventionalism in science has nothing much to do with that. Furthermore, the example about bachelors shows that it is a mistake to think of linguistic practice as making analytic (object-language) sentences true. The linguistic conventions-the arbitrary pairing of words with meanings that is operative in a language-establish what proposition a given sentence in a language expresses. This is true for all sentences in 13. Thus, it should be clear how conventionalism as a thesis about the content of a specific scientific theory differs from 'trivial semantic conventionalism'-the claim that the pairing of words in a language with their meanings or referents is arbitrary. It is interesting that Carnap (1963, p. 916) makes this very point. In discussing the logical truth 'all black dogs are dogs,' he says that 'once the meanings of the individual words in a sentence of this form are given (which may be regarded as a matter of convention), then it is no longer a matter of convention or of arbitrary choice whether or not to regard the sentence as true; the truth of such a sentence is determined by the logical relations holding between the given meanings ...'

QUINE

249

a language, not just the analytic ones. Whether bachelors are unmarried depends on linguistic conventions no more than whether there are nine planets. The argument just described against the thesis of truth-by-convention makes use of a counterfactual; we are asked to consider, for example, whether bachelors would still be unmarried if English had developed differently. Those, like Quine, who think that counterfactuals are 'unclear' may be unimpressed with this argument. But once again, the rejection of first philosophy that Quine did so much to promote leads to a different assessment. It is true that philosophers have not been able to provide a fully adequate account of the rules that govern counterfactual reasoning. It also is true that extant accounts (e.g., those of Stalnaker 1968 and Lewis 1973) have had to make use of irreducibly modal concepts, and this will arguably be a feature of future theories. If so, philosophical theories of counterfactual reasoning will not measure up to Quine's standards of clarity. However, none of these points provides a compelling reason to 'reject' counterfactual reasoning. What needs to be remembered is that counterfactual reasoning is central to scientific inference. Eating lollipops is correlated with being young, but eating lollies doesn't make one youthful. Children would still be young if they stopped consuming lollipops. This is an entirely mundane type of remark, one that is often made in scientific contexts (see, for example, Fisher's smoking hypothesis). The use of counterfactuals in the argument against the thesis of truth-by-convention should be no more suspect.14 Quine's discussion of modality embodies an attempt to change science; the point, however, is to understand it.

The Argument of 'Truth by Convention'. Not only does the argument just presented against the idea of truth-by-convention involve counterfactual reasoning; in addition, it relies on a 14. It doesn't help the conventionalist to assert that the relation between our adopting conventions and the truth of various propositions is 'constitutive' rather than causal. The counterfactual test is still appropriate-if C makes P true (where 'makes' is understood constitutively), then it should turn out that P would not be true if C had not occurred.

250

I-ELLIOTT

SOBER

distinction-the distinction between a sentence and what the sentence expresses (a 'proposition'). This is a distinction that Quine might have qualms about drawing, since talk of propositions, for him, is not legitimate (but see footnote 11). Accordingly, it is no surprise that Quine advanced a quite different argument against conventionalism in his earlier essay 'Truth by Convention'. Quine (1936) considers whether logical truths like 'if pigs fly, then pigs fly' could be true by convention. He observes that there are infinitely many such sentences, which means that we would not be able to stipulate, one by one, that each is true. Rather, we might try to cover them all by making stipulations about general logical schemata; for example, one could propose the convention that all statements of the form 'if P, then P' are true. Quine then points out that applying this stipulation about the schema to each of its many instances requires the use of logic. We would have to reason as follows: The sentence 'if pigs fly, then pigs fly' has the form 'if P, then P'. All sentences of the form 'if P, then P' are true b y stipulation. 'If pigs fly, then pigs fly' is true b y stipulation.

As this derivation shows, logical truths aren't true just in virtue of the stipulations we make about various logical schemata. Rather, the first premiss reports a fact about logical form that is independent of the stipulation described in the second premiss; in addition, the argument uses a rule of inference that connects premisses to conclusion, and this rule is not made valid by the reported stipulation. In this argument, nonstipulated facts about logic are used to connect the general stipulation about the schema to a specific stipulation concerning the sentence at issue. Conventionalism thereby fails as a thesis about all logical truths. This is an important argument. However, we need to be clear on what it does and does not entail. It does not show that no sentence is true by convention. It is quite possible, for all this argument says, that there are finitely many statements that are true by convention. The problem with conventionalism that Quine identifies arises only when conventionalists are ambitious-when they claim that the number of sentences that are true by convention is infinite. Indeed, Quine says in both 'Truth by Convention' and in 'Two Dogmas' that he has no

QUINE

251

problem with the idea that explicit stipulation can render a sentence true. l 5 The argument discussed before that turns on the distinction between a sentence and the proposition it expresses suggests that Quine is too generous to the idea of explicit stipulation. To see why, let's consider a simple example of a stipulative definition. Suppose I announce that I'll use the term 'ap' to mean apple. If I subsequently say 'an ap is an apple', my statement will be true. I'll also be saying something true if I say 'some aps grow in Wisconsin'. How do these two utterances differ? In both cases, we figure out whether a sentence is true by deciding what proposition the sentence expresses and then determining whether that proposition is true. If we know a priori that an apple is an apple and we know that 'an ap is an apple' expresses this proposition, then we can say without further ado that the sentence is true. However, if we see that the sentence 'some aps grow in Wisconsin' expresses the proposition that some apples grow in Wisconsin, then we can't say, based just on that and the a priori knowledge at our disposal, whether the sentence is true, since the proposition it expresses is a posteriori. In short, even modest conventionalists, who propose only a finite number of linguistic stipulations, have a problem. They can't say that a sentence is made true solely by the act of adopting a linguistic convention. 15. Although Quine does not object to the idea that some sentences are made true by explicit linguistic convention, he doubts that it makes sense to talk of linguistic conventions that are merely implicit in language use; he is suspicious of talk of conventions that arise without an act of 'convening.' When I declare what my term 'ap' will mean, I do so within a pre-existing language. However, the meanings of terms in the first language a community comes to use cannot arise by people stipulating what their terms will mean. David Lewis (1969) answered Quine's question by describing how a community could evolve conventional practices without explicitly consenting to an agreement stated in a pre-existing language. Lewis' idea was to use game theory; terms can acquire meanings in the same way that the drivers in a society can come to drive on the same side of the road. It is arbitrary whether we drive on the left or the right, but it is not arbitrary that we all do the same thing. A community of agents who are guided by the wish to cooperate can evolve a communication system in which signs have significance. The question may be raised as to whether the same account applies when individuals are selfish rather than cooperative (Sober 1994b), but the general point remains-there is nothing unintelligible about the idea of conventions coming into existence without explicit stipulations. Still, as Lewis notes, this defense of the concept of convention does nothing to save the idea that analytic truths are made true by the conventions one adopts. It remains true that bachelors would still be unmarried if English assigned different meanings to the terms 'bachelor' and 'unmarried,' for example.

252

I-ELLIOTT

SOBER

Boghossian (1996) provides a nice formulation of the present point. In general, the truth conditions for a sentence in a natural language will have the following form: s is true in language L iff s expresses proposition p and p is true.

Facts about the meanings of terms in a sentence determine which proposition the sentence expresses. However, that isn't enough to settle whether the sentence is true; there is, in addition, the extra-linguistic question of whether the proposition is true. It is easy to forget this extra-linguistic contribution when the proposition in question is a priori. In the apple example, the fact that an apple is an apple is part of what makes the sentence 'an ap is an apple' true. This extra-linguistic contribution to truth value is present in any object-language sentence. We can illustrate this two-part contribution via the following derivation: 'An a p is an apple' is true in L iff an apple is an apple.

An apple is an apple.

'An ap is an apple' is true in L.

The first premiss describes what the quoted sentence means by stating its truth conditions. The conclusion does not follow from this first premiss taken by itself; the second premiss is needed as well, and this premiss states a fact that is language-inde~endent.'~ 16. Although it sounds quite intuitive to say, for example, that bachelors would still be unmarried if we a d o ~ t e ddifferent linauistic conventions concerning the use of the term 'bachelor,' it is'not immediately obvious what is wrong with tlhe following argument: Because of our linguistic conventions, the English sentence 'a bachelor is an unmarried man' is true.

Because of our linguistic conventions, a bachelor is an unmarried man. The premiss seems true and the conclusion seems to follow from the premiss by a sort of disquotation principle. However, I claim that the conclusion is false. Where does the argument go wrong? I understand the premiss as follows. It says that (*) Because of our linguistic conventions, the English sentence 'a bachelor is an unmarried man' expresses some true proposition p. Notice that the fact brought into being, according to this premiss, involves a relation between the sentence and the class of true propositions. As such, it is similar to the following claim: Because of something Joe did, his herringbone suit now fits him. This sentence could be true because Joe went on a diet, or because Joe had his suit altered (or both). The relation got changed by changing one of the relata, but the sentence does not say which relatum got changed. The same point holds of (*). It is true; our adoption of certain linguistic conventions changed something. But what? There are two possibilities. Our behaviour could

QUINE

253

In 'Two Dogmas', Quine (1953c, pp. 36-37) says that it is tempting to think that 'the truth of a statement is somehow analyzable into a linguistic component and a factual component. Given this supposition, it next seems reasonable that in some statements the factual component should be null; and these are the analytic statements'. It should be clear that I do not share Quine's skepticism about the possibility of effecting the separation he describes. But, in addition, I disagree with the idea that analytic truths are 'nonfactua1'-that there is nothing besides the 'linguistic component' that makes them true.17 Fortunately, the legitimacy of analyticity as a category does not depend on this conventionalist position. I have suggested that Quine was right in 'Truth by Convention' to deny that explicitly adopted conventions can account for the truth of all the logical truths, since there are too many. I then suggested that Quine is too generous; on closer analysis, it is implausible to claim that even one logical truth is true just by virtue of the conventions we adopt. However, even if conventionalism is mistaken as an account of the 'ground' of logical truth, it still may be true that there is an irreducibly arbitrary element in our choice of logical principles. The epistemic side of conventionalism remains untouched. In addition, the semantic thesis that logical principles do not have truth values is not undermined by Quine's argument in 'Truth by Convention'; just as human conventions cannot make the law of excluded middle true, so the human incapacity to propose an infinite number of conventions cannot show that the law has a determinate and convention-independent truth value. The conventionalist claim that certain statements lack truth values but should be accepted on pragmatic grounds does not require that any sentence is literally made true just by language users adopting linguistic conventions. have changed which proposition the sentence expresses, or it could have changed the truth value of a proposition. Clearly, the former is correct. This is why the premiss in the quoted argument is true, but the conclusion does not follow. 17. Even though my claim that a definition never suffices to establish the truth of any object-language sentence conflicts with what Quine says in the passage just cited from 'Two Dogmas of Empiricism,' it accords with an idea he developed earlier in 'Truth by Convention'. There Quine (1936, p. 71) says that 'a definition is incapable of grounding the most trivial statement'. This is because he thinks that a definition is 'a convention of notational abbreviation;' 'a definition is not a premise to theory, but a licence for rewriting theory' (p. 71). Quine is saying that a definition is a metalinguistic statement that says that two object-language expressions are synonymous, or at least intersubstitutable salve veritate.

254

I-ELLIOTT

SOBER

Back to Frege. Quine's target in 'Two Dogmas' is Carnap and the logical empiricists, not Frege. However, in the light of the errors we have identified in the idea that analytic truths are true by convention, it makes sense to go back to Frege, whose conception of analyticity was not conventionalist. Frege (1884) said that analytic truths are deductive consequences of definitions. However, as noted above in connection with apples, the truth value of an object-language sentence is never settled solely by describing what the sentence means. The meaning of an object-language sentence is given by stating a biconditional that describes its truth conditions; from this biconditional, taken by itself, the truth of the left-hand side does not follow. However, Frege's idea easily accommodates this observation. The truth of an analytic (objectlanguage) sentence follows from logical truths when (metalinguistically formulated) definitions of the terms in the sentence are supplied. Definitions, as I am using the term, are meta-linguistic statements that describe the semantic properties of expressions or sentences in an object language. This may sound overly restrictive, since it entails, for example, that the object-language sentence 'a mother is a female parent' is not a definition. However, notice that the quoted sentence says nothing about English or any other natural language; what the sentence says would still be true even if English never existed. Definitions must be meta-linguistic statements if they are, literally, to describe the semantic properties of terms in a particular language. Without construing definitions in this way, analyticity fails to have anything to do with language. This Fregean conception of analyticity should be understood in such a way that it is consistent with the apparent fact that very few terms in a natural language can be defined noncircularly in that language. The term 'triangle' is an exception; even the term 'bachelor' turns out, on reflection, to be such that one can state necessary, but not necessary and sufficient, conditions for the term to apply-perhaps bachelors must be unmarried adult male human beings, but the converse seems not to be true.'' 18. It has often been noted that it sounds odd to say that the Pope is a bachelor; it also sounds odd to say of a man who lives out of wedlock with a woman for many years and has children with her, that he is a bachelor (Harman 1996).

QUINE

255

However, the case for analyticity doesn't depend on the existence of definitions in this sense. Carnap (1952a) used the term 'meaning postulate' to describe statements that provide analytically necessary or analytically sufficient conditions for the application of a term. Analyticity does not require that the terms in a natural language be definable noncircularly. Even in the absence of full-scale definitions, analytic truths can be made to fall into the pattern described before in connection with 'ap': 'Bachelors are unmarried' is true in English iff unmarried bach-

elors are unmarried.

Unmarried bachelors are unmarried.

.....................

'Bachelors are unmarried' is true in English.

In such two-premiss arguments, the first premiss describes what speakers of a natural language know if they know what the quoted sentence means.19 What of the second premiss in these displayed arguments? They express 'logical truths', but what are they? Quine restricts logic to first-order logic with identity. I see no reason to be so parsimonious. For example, the following sentence seems to me to be just as analytic as 'bachelors are unmarried', but second-order logic is needed to explain why: If Jack falls down and Jill falls down, then there is a characteristic that Jack and Jill both have.

In any event, I don't think it is necessary to settle here what logic is with any precision. For now, we can envision a plurality of notions of analyticity, each relativized to its own conception of what makes a truth a logical truth. Perhaps this tolerant pluralism can be whittled down by well-motivated arguments concerning the nature of logic.

VII Analyticity as Part of a Theory of Linguistic Competence. A cognitivist model of language use seeks to predict and explain 19. It is for this reason that "'bachelors are unmarried" is true in English iff grass is green' is a true biconditional, but not one that can serve to show that 'bachelors are unmarried' is analytic.

256

I-ELLIOTT

SOBER

linguistic behaviour by attributing certain sorts of knowledge to people. It will say that people who are able to speak English have this ability by virtue of their possessing an abstract representation of their language. This knowledge allows them to say when a string of words constitutes a well-formed English sentence and when it does not. In addition, speakers know various facts about what words mean and they know rules that allow them to assign meanings to sentences, based on the meanings assigned to their parts." A linguist will want to describe how the knowledge that English speakers possess allows them to understand what English sentences mean. It is not the mission of linguistics to say which of those English sentences is true. Linguists will want to provide an abstract and general theory that assigns truth conditions to 'there are nine planets' and 'there are ten planets'. It is for astronomy, not linguistics, to say which of these sentences is true. The same point holds for the sentences 'bachelors are unmarried' and 'bachelors are married'. The linguist wants a general theory of the semantics of the language that describes what each sentence means; which of them is true is not a problem for linguistics. The legitimacy of the concept of analyticity is thus not completely settled by seeing how theories in linguistics and cognitive science develop. Those empirical sciences will tell us what we should think of the concept of meaning (Putnam 1962), but meaning is only one of the two ingredients that go into the concept of analyticity. The other ingredient is logic. Still, it is not unreasonable to think that if speakers of a language know what sentences in their language mean, and also know some logic, then they will know (or be able to know), on that basis, that certain sentences of their language are true. Of 20. If theories of meaning are to be thought of in this way-as empirical theories that seek to explain and predict the behaviour of language users-then a standard criticism of verificationist theories of meaning needs to be reconsidered. Verificationism has often been criticized for being meaningless by its own standard-it isn't a correct definition of what the term 'meaning' means in ordinary usage, nor is it empirically testable. Carnap (1936-37) replied that verificationism is a proposal, not a statement that has a truth value; however, critics then were able to say that it is a proposal that they felt no particular obligation to accept. On the view I am suggesting, theories of meaning are no more 'proposals' than are theories of planetary motion. Claims about meaning are introduced to explain and predict the behaviour of language users, just as concepts like mass and acceleration are introduced to explain and predict the behaviour of planets.

QUINE

257

course, speakers of a language will not have at their disposal the full panoply of what logicians mean by 'logic'. But suppose that some elements of logic are psychologically real-speakers of a language have at least some understanding of the entailment relation. If so, their grasp of their language, along with whatever logical acumen they possess, will allow them to recognize that some of the sentences in their language are true. Knowledge of language allows speakers who know some logic to know that certain object-language sentences are true, where those sentences are not about language at all. This conception of analyticity leaves open the possibility that some languages may fail to contain analytic sentences. As an admittedly artificial example, consider a 'language' in which the only well-formed sentences are ones that report an object's temperature; each wff has the form 'object x has a temperature of n degrees Centigrade'. A theory of meaning can be provided for this language, one that describes what it takes to be a competent user of this representational system. However, no sentence in this language is analytic. This is enough to show that analytic truths are not an inevitable byproduct of the meaningfulness of sentences. Rather, they are to be found in languages that possess a certain type of expressive power-these languages contain sentences that competent speakers know have logical truths as their truth conditions.

VIII Knowledge of Language and Knowledge of Extra-Linguistic Propositions. The concept of analyticity builds a bridge between the truth of a sentence in a natural language and a proposition that is not linguistic in its subject matter. For example, if 'mothers are female parents' is an analytic truth in English, then speakers will know the following biconditional (or something like it): The English sentence 'mothers are female parents' is true iff mothers are mothers.

In principle, knowing this biconditional might allow speakers who already know the left-hand side of the biconditional to come to know the right, or it might permit individuals who already know the right-hand side to come to know the left. Do we know

258

I-ELLIOTT

SOBER

that the English sentence is true because we know that mothers are mothers, or do we know that mothers are mothers because we happen to know that the quoted English sentence is true? The linguistic theory of the apriori claims that knowledge of language is the source of the a priori knowledge we possess. However, it is entirely possible that the situation is precisely the reversethat our knowledge of language is predicated on our already having certain items of a priori knowledge at our disposal. Even if knowledge flows from logic to language in the kind of case just described, there may be other cases in which the direction is reversed. Consider, for example, Cantor's proof that there are nondenumerably many numbers between 0 and 1. A naive person who is told that there are more numbers between 0 and 1 than there are whole numbers will react with bafflement. However, if you write down a series of decimals, and then show how Cantor's diagonalization procedure can be used to construct a decimal not on the list, the claim becomes clear. Cantor's proof helps people see that the proposition about cardinality is true by exploiting certain obvious properties of decimal notation. Try to convince someone of Cantor's claim by using Roman numerals and you'll get nowhere. Cantor's proof allows one to learn something about numbers by examining properties of a notational system.21 Although knowing Cantor's theorem is facilitated by knowing facts about a language, it is clear that one can't learn that the theorem is true just by examining facts that are about language and nothing else. After all, Cantor's theorem is about numbers, and whatever numbers are, they are not language-bound entities. One must know that items of notation refer to numbers for the proof to tell you anything about numbers. Competent users of a representational system (whether it is a natural language or decimal notation) know that certain biconditionals are true. These biconditionals connect the truth of a representation with a proposition that is extra-linguistic in its subject matter. Knowing such biconditionals can permit knowledge to proceed from left to right, or from right to left. The 21. Clearly, examining a notational system isn't necessary for knowing that the theorem is true; after all, if you are told by a reliable person that Cantor's theorem is true, this allows you to know the proposition or to rationally believe it. And proofs of the theorem that do not refer to notational systems can be given as well.

QUINE

259

linguistic theory of the a priori envisioned a one-way street running from analyticity to a priority, but not in the opposite direction. A better picture is a two-way street, with perhaps more traffic flow in one direction than in the other. Earlier in this paper, I rejected the idea that analytic truths are literally 'made true' by the conventional assignment of meanings to words. I also have rejected the quite different, epistemic, idea that analytic truths can be known to be true just by knowing what they mean.22 A more satisfactory understanding of analyticity is Frege's-that the truth of an analytic sentence follows from logical truths plus definitions. One virtue of this proposal is its austerity. The proposal does not have packed into it an assumed link between analyticity and a priority, nor does it presuppose any special status for logic. These require separate arguments.

Is Analyticity 'Epistemologically Significant'? If meaning is a useful theoretical concept in cognitive science, then there can be no objection to the notion of analyticity, or so a naturalist should be willing to grant. However, philosophers sympathetic with Quine's critique of analyticity sometimes say that Quine's real contribution was to deprive analyticity of its 'epistemological significance'. Of course, even if analyticity were a legitimate scientific notion, it wouldn't follow that it has any epistemological or philosophical importance. How is this further issue to be addressed? I take it that the question of 'epistemological significance' concerns the issue of a priori knowledge. If a sentence is analytic, then what the sentence says will be a priori, if logic is a priori.23 As described in our various examples about bachelors, mothers, and apples, the biconditional that specifies the meaning of an 22. Boghossian (1996) also separates these metaphysical and epistemic conceptions of analyticity, but he endorses the latter. 23. If a sentence is analytic, then the fact that the sentence is true will not be decidable a priori if it is an empirical question what the sentence means in the particular (socially shared) natural language in question. Perhaps sentences that are analytic in a speaker's idiolect can be known by that speaker to be true a priori, but that is another matter. In any event, for socially shared languages, the connection between analyticity and a priority is effected via the proposition that the sentence expresses.

260

I-ELLIOTT

SOBER

object-language analytic truth will connect that sentence to a logically true proposition. Thus, the philosophical interest of 'analysis', if it has any, consists in the fact that it provides access to the a priori. This is what analytic philosophers care about that separates them from lexicographers. If there is no such thing as a priori knowledge, then analyticity loses its philosophical interest, however legitimate the concept of meaning might be in cognitive science. It is to the concept of the a priori that we now must turn.

What Is Epistemological Holism? Quine's critique of analyticity aimed to undermine the linguistic theory of the a priori, but this critique, even if successful, would leave open the possibility that there are propositions that can be known or justified independently of experience. Quine's epistemological holism is intended to shut down that possibility. Not only can't the concept of analyticity be used to show that there is a priori knowledge; nothing else can either, because there is no such thing. The distinction between a priori and a posteriori is intended to separate two types of knowledge or justification. We can make rational judgments about the truth values of a posteriori propositions only by appealing to the testimony of the senses; however, for a priori propositions, the rational justification for assigning truth values does not depend on sense experience. What, then, does it depend on? The usual answer is that 'rational reflection' of some kind on the content of the proposition suffices to justify an assignment of truth value. This reflection might involve noticing that a sentence that expresses the proposition is analytic, as the logical empiricists thought; the Kantian alternative is that it might involve noticing that the proposition is synthetic, but is a presupposition of the possibility of empirical knowledge. The contrast between a priori and a posteriori is often thought to coincide with a contrast between beliefs that are certain and beliefs that are not. However, nothing I have said about apriority implies that this is so. Suppose I know that triangles have three sides by examining the content of that proposition-by analyzing its constituent concepts. I therefore bypass the use of sense experience with its attendant sources of error. However, it is

QUINE

26 1

entirely possible that the introspective examination of one's concepts is prey to unreliabilities of its own. Various philosophers have claimed that the exercise of 'reason' yields infallible results, but I do not. The a priorilaposteriori distinction simply distinguishes two sources of knowledge or justified belief. If the a priorila posteriori distinction makes sense, then we should be able to divide the different propositions we believe into two classes, each having its own distinctive sort of support. Quine's holism denies that this is possible. For Quine, the totality of what we believe is supported by the totality of the sense experiences we have. We adjust our total theory of the world by bringing it into accord with the sense experiences we have and also by making it conform to requirements such as logical consistency, simplicity, elegance, generality, and tractability. Logical reflection and sense experience jointly shape this total theory; it is a mistake to think that some of our beliefs are justified just by rational reflection while others depend for their justification on sense experience. An analogy may help convey the character of what Quine is claiming. Suppose a husband and wife deposit their salaries into a joint checking account from which they pay their bills. Does the wife's salary cover the car insurance, while the husband's takes care of the groceries? In some marriages this division may occur, but in the one I am imagining, it does not. The expenses as a whole are covered by the two paychecks together. When their income shrinks or grows, and when prices rise and fall, they revise their list of what they buy. But individual purchases never trace back to one paycheck rather than the other. Quine's holism separates his position from that of Mill, according to whom propositions that we might think are a priori are in fact empirical. Mill (1859) thought that our belief in simple arithmetic propositions like 2 + 3 = 5 is based on the experiences we have when we count. Arithmetic is a posteriori. However, for Mill, the justification that we have in believing that 2 + 3 = 5 can be separated from the justification we have for believing that grass is green; both are experiential, but different experiences justify different beliefs. This isn't holism, but a kind of epistemological atomism. Mill's position makes it clear how epistemological holism goes beyond the simple denial that there are a priori truths.

262

I-ELLIOTT

SOBER

Centrality and Minimum Mutilation. If our beliefs 'face the tribunal of sense experience not individually but as a corporate body', what determines how we should revise our beliefs when we encounter 'recalcitrant experience'? For example, what do we do if we expect it to be snowing when we look out the window, but don't have the experience of whiteness that we expected to have? We need to revise our beliefs, but which ones should we change? Why do we revise the belief that it is snowing, rather than the belief that our eyes are functioning properly, or the belief that snow is white, or still others? Quine says that our beliefs differ from each other in terms of their 'centrality', and that we follow a policy of 'minimum mutilation'-we revise less central beliefs rather than more central ones. Our logical and mathematical beliefs are the most central beliefs we have, which is why we rarely if ever are tempted to revise them in the light of experience. However, this is not because they enjoy some special sort of nonempirical justification. Logic and mathematics differ in degree, not in kind, from the beliefs we have that clearly seem to be empirical; our belief in the validity of modus tollens is just more central than our belief that it is snowing. What does 'centrality' mean? Quine's examples of logic and mathematics suggest that centrality has to do with the importance a belief has in organizing experience, which in some measure is influenced by the beliefs generality. We don't abandon logical principles of inference when we expect to see snow but don't; the reason is that logic applies to lots of other situations besides this one, and moreover plays an important role in organizing how we think about those situations. There are several problems with this suggestion. First, even though modus tollens is central in the sense just described, it isn't true that every theorem in pure mathematics is like this. A result in some recondite area of algebraic topology, for example, might play little or no general role in organizing how we think about the world. Why, then, are such theorems accorded so great a degree of immunity from revision in the light of recalcitrant sense experience? The same point applies to trivial analytic truths. Surely the belief that bachelors are unmarried isn't central to our understanding of the world; it

QUINE

263

is utterly unlike modus tollens in this respect (Putnam 1962, p. 39; Sidelle 1989). A second problem for Quine's proposal derives from the fact that general beliefs that organize our interpretation of wide swathes of experience are sometimes revised in the light of observations; this is what happens when scientists reject a general theory because they think that the evidence obliges them to do so. Why doesn't the principle of minimum mutilation dictate that we should always assume that the data are misleading when the evidence seems to disconfirm a theory that is central to our system of beliefs? The fact of the matter is that scientists often reject theories in the light of data, but they sometimes retain a theory in the face of contrary evidence, conjecturing that the data must be misleading in some way they can't quite figure out (this is roughly what Kuhn 1970 meant by an 'anomaly'). The rule of minimum mutilation isn't sufficiently elaborated for it to help illuminate this important phenomenon. Lastly, suppose the concept of centrality were sufficiently clear that we could see that logic and mathematics are the most central of our beliefs. The question then arises as to how the claim that logic and mathematics are maximally central differs from the claim that mathematics and logic are a priori. If rational revisability in the light of sense experience is governed by the principle of minimum mutilation, in what sense are our most central beliefs empirically re~isable?'~

The Holistic Concept of ConJirmation. Quine's holism has been extremely influential in philosophy generally, but perhaps most influential in philosophy of mathematics. Many philosophers hold that mathematical statements are empirically well supported because they play an indispensable role in natural science (Putnam 1971; Kitcher 1983; Maddy 1990; Hellman 1992, 24. To these criticisms, it might be replied that Quine never pretended that the principle of minimum mutilation is a complete epistemology. My response is that if the principle is just one of the considerations that governs belief revision, then it is left open whether Quine is right when he says that all beliefs are empirically revisable.

264

I-ELLIOTT

SOBER

forthcoming; Resnik 1992, 1 9 9 9 . ~ Calculus, ' for example, might appear to be an a priori subject, but the fact that it is used in predictively successful physical theories means that calculus enjoys the empirical confirmation that those theories have received. That calculus is used in relativity theory means that the empirical evidence that supports relativity theory also accrues to calculus; but the fact that calculus is used in Newtonian theory does not mean that the empirical disconfirmation of Newtonian theory also accrues to calculus. Ronald Reagan used to be called 'the Teflon President'-accusations of malfeasance never seemed to stick to him, whereas he was often given credit for the good things that happened while he was in office. Quinean philosophy of mathematics holds that mathematics has the same Teflon quality-it is empirically confirmed by the good theories in which it is used, but it is not empirically disconfirmed by the bad theories in which it also is used (Sober 1993, 1999). The confirmation relation that holism invokes is bizarre. Consider a perfectly mundane example. 1 draw a card at random from a standard deck of cards without looking at it. The probability that it is the seven of hearts is 1/52. You then inform me that the card is red. This information confirms the hypothesis that the card is the seven of hearts, not in the sense of insuring that the hypothesis is true, but in the sense of making the hypothesis more plausible than it was before; the probability that I have the seven of hearts has just increased to 1/26. However, this information does not confirm the hypothesis that the card I hold is a seven; the probability that I have a seven remains what it 25. Maddy (1992) criticizes the indispensability argument; Field (1980) accepts the Quinean assumption that mathematics would be justified if it were indispensable, but areues against the antecedent. Peressini (1997) identifies issues that need to be adcdresseiin the indispensability argument concerning the relationship between pure mathematics and mathematized natural science that are sevarable from the auestion of confirmational holism. It is worth noting that Quine has recently changed his mind about the holistic status of mathematics. In From Stimulus to Science, Quine writes: The accepted wisdom is that mathematics lacks empirical content. This is not contradicted by the participation of mathematics in implying [observational] categorical~for ... such participation does not confer empirical content. The content belongs to the implying set, and is unshared by its members. I do, then, accept the accepted wisdom. No mathematical sentence has empirical content, nor does any set of them (Quine 1995, p. 49; quoted by Gibson 1998, pp. 677-678). In his reply to Gibson, Quine (1998, p. 685) confirms that his present view is that 'mathematics lacks empirical content.'

QUINE

265

was, namely 1/13. The following pattern of inference (which Hempel 1965, p. 31 called 'the special consequence condition' of confirmation) is fallacious: 0 confirms H 0 = the card is red H entails S H = the card is the seven of hearts ..................... S = the card is a seven

0 confirms S Yet, holism maintains that when an observation confirms relativity theory, this observation also confirms the pure mathematics that is used in relativity theory. The example just described has a natural Bayesian representation. Bayesians say that 0 confirms H precisely when Pr(Hl0) > Pr(H) and that 0 disconfirms H precisely when Pr(Hl0) < Pr(H); confirmation means probability raising and disconfirmation means probability lowering. In the card example, 0 raises the probability of H but it does not raise the probability of S, even though H entails S. Although it is convenient to use a Bayesian model of confirmation to analyze this example, the force of the example does not depend on Bayesianism's being adequate in general. It is enough that Bayesianism makes sense in the case at hand. Another property of simple examples of confirmation is that there is a kind of symmetry between confirmation and disconfirmation. If observing that the card is red confirms the hypothesis that the card is the seven of hearts, then, if the card had not been red, that observation would have disconfirmed the hypothesis. This idea also can be given a Bayesian representation. It follows from Bayes's theorem that Pr(Hl0) > Pr(H) if and only if P r ( ~ 1 n o t - 0Pr(vT,lO). None of these three points answers the question that Hellman poses. Even if M is less than certain, it still may fail to be confirmed by 0. And the fact that M helps other pairs of theories plus auxiliary assumptions to make accurate predictions merely raises for them the same question that arises with respect to L and A. Having lots of the same thing provides lots of confirmation only if each test provides some confirmation, and this has not been shown. Finally, Hellman's point about the disjunction vT, does not show that P r ( M 0 ) > Pr(M). Hellman also says that mathematical propositions are confirmed by their role in empirically successful theories, but are not disconfirmed by their role in empirically unsuccessful theories. However, as noted above, Bayes's theorem entails that there is a symmetry between confirmation and disconfirmation. In addition, Hellman says that his position is consistent with there being propositions of pure mathematics that are immune from empirical revision. As he realizes, this is a substantial retreat from Quine's position. That some propositions of pure mathematics (like the ones that Plateau investigated) can be tested empirically is not in dispute.

QUINE

267

the version of the paper later published in From a Logical Point of View (Quine 1953c, p. 41). Quine (1991, p. 269) says that he 'didn't know about Duhem' when he wrote 'Two Dogmas', but inserted the footnote at the prompting of Hempel and Philipp Frank. In that footnote, he equates his epistemological holism with Duhem's thesis. Duhem (1914) made a claim about hypotheses in physicstaken on their own, they do not have observational consequences. If physical hypotheses are to make testable predictions, they must be conjoined with 'auxiliary assumptions'. Even though H by itself does not entail any observation statements, the conjunction (H&A) may do so. Quine generalized Duhem's point-what is true of physical hypotheses is true of all our beliefs; no hypothesis is 'testable in isolation'. I do not doubt the logical point with which Duhem and Quine begin. However, epistemological holism does not follow. The logical point has implications about what happens when a prediction fails to come true. If (H&A) entails 0 , and n o t - 0 turns out to be true, then there is a choice; one might abandon H, or abandon A, or abandon one or more logical principles. Quine thinks that the fact that we face a choice here shows that all our beliefs are potentially at risk when we make observations. We could abandon our logical and mathematical beliefs, rather than our physical theories, when predictions fail to come true. Only pragmatic considerations reflecting our allegiance to the policy of minimum mutilation lead us to do otherwise. It is often replied that this flexibility would disappear if we had independent reason to believe the background assumptions Aif the conjunction (H&A) entails a false prediction, and we know that A is true, then the culprit must be H, which we then reject. To this, Quine might reply-why should our prior conviction that A is true continue to hold sway? Why shouldn't the failure of our current prediction be a reason to re-evaluate A? Here is a way to think about testing that answers this question: Typically, scientists test a hypothesis by testing it against another hypothesis; testing is a comparative process in which one tries to find observations that favour one hypothesis over another (Sober 1994a, 1999). The logical point about the need for background assumptions is entirely correct. However, the typical situation in science is that the competing hypotheses make use of the same

268

I-ELLIOTT

SOBER

set of background assumptions. The conjunction (H1& A ) and the conjunction (H2& A) both make predictions. However, these conjunctions do not deductively entail what they predict. Rather, since observation is always subject to error, what happens is that these conjunctions confer probabilities on observations. The conjunction (H1& A) confers a probability on the observational outcome 0 , and the same is true of the conjunction (HZ& A). We now can define the idea that an observational outcome 0favours one hypothesis over another: Given background assumptions A, 0 favours H I over Hz if and only if Pr(O/H,& A ) > Pr(OJH2& A).

If 0 is very probable according to (H1& A) and very improbable according to (Hz &A), then 0 strongly favours H 1 over HZ. If the probabilities are closer together, the degree of favouring is less decisive. The point to notice is that 0's favouring H1 over H2 says nothing about whether 0 favours A over some alternative auxiliary assumption. The experiment just described does not test A at all because it does not test A against an alternative. The usual role that mathematics and logic play in empirical science is to furnish a common set of background assumptions, which all the competing hypotheses exploit. This is why logic and mathematics are not tested in typical scientific experiments (on this point, see Popper, 1963, p. 112).27 For logic and mathematics to be tested empirically, one logical or mathematical statement would have to be pitted against an alternative and a framework of shared background assumptions would have to be supplied that permits the two statements to make different predictions about observations. It is instructive to examine a case in the history of mathematics in which this actually happened. Plateau was a 19th century French mathematician 27. Friedman (1997, p. 12) claims that the 'fundamental problem' with Quinean holism is that it does not make sense to view a mathematized theory like general relativity as a conjunction in which the pure mathematics is isolated in one of the conjuncts; 'rather, the mathematical background of Einstein's theory functions as a necessary presupposition of that theory ...' Holism, I would say, does not require this partition; in effect, Hempel's special consequence condition can be thought of as expressing the content of holism. Friedman then goes on to note that Newtonian physics, special relativity, and general relativity are all now thought of as different theories that share the same mathematical machinery. It is because that machinery is common across these theories that the empirical evidence that favours one theory over another does not confirm (or disconfirm) the shared mathematical presuppositions.

QUINE

269

who wanted to figure out what the surface of least area is that fills various closed curves. A simple example of the type of problem that Plateau had in mind is a curve that has the shape of an ellipse. An elliptical curve can be filled by the ellipse it contains; any other surface that fills the curve will bulge into a third dimension and so must have more surface area. Although the answer to Plateau's question is obvious for this example, the answers aren't at all obvious for more complicated three-dimensional curves. What Plateau did was to dip wires bent in different shapes into soapy water and observe the resulting soap bubbles that adhered to the frames when they were removed from the water (Courant and Robbins 1969). Given the physical assumption that soap bubbles take on the surface of least area in this experiment, different mathematical hypotheses ('the surface of least area for curve C is s,' versus 'the surface of least area for curve C is s2', for example) make different predictions. This example is interesting in the history of mathematics precisely because it is so atypical. No such test of '2 + 3 = 5' against alternative arithmetic hypotheses has ever been carried out, nor is it remotely clear what this would be like. Quine suggests in 'Two Dogmas' that the alternative to epistemological holism is the idea that single sentences have their confirmation conditions separately-that one should be able to say, for each hypothesis, which observations would confirm it and which would infirm it. This is the second dogma of empiricism that Quine rejects. However, there is no need to choose between epistemological holism and this positivist idea. Confirmation and disconfirmation are three-place relations, not twoplaced-hypothesis H is confirmed or disconfirmed by observation 0 relative to background assumptions A. This logical point does not force one to say that anything that confirms/ disconfirms H also confirms/disconfirms A (see also Creath 1991, pp. 379-382).28 28. Although there are many passages that suggest that Duhem was indeed an epistemological holist, the following passage suggests a rather different interpretation: The only thing the experiment teaches us is that among the propositions used to predict the phenomenon and to establish whether it would be produced, there is at least one error; but where this error lies is just what it does not tell us. The physicist may declare that this error lies in exactly the hypothesis he wishes to refute, but is he sure it is not in another proposition? If he is, he accepts implicitly the accuracy of all the other propositions he has used, and the validity of his conclusion is as great as the validity of his confidence (Duhem 1914, p. 185).

270

I-ELLIOTT

SOBER

XIV Moderate Holism. In various writings since 'Two Dogmas', Quine (1960, p. 13; 1975, p. 314; 1986, p. 427; 1991) has mentioned that this essay overstates his epistemological holism (Creath 1991; Gibson 1998). The claim should not be that the totality of one's system of beliefs faces the tribunal of sense experience as a corporate body, but that largish sets of beliefs do so. The most minimal sort of holism concerns 'clusters of sentences just inclusive enough ... to imply an observable effect of an observable experimental condition (Quine 1991, p. 268)'.29 It is interesting that all forms of holism, including even this most moderate formulation, are subject to the criticisms just adumbrated. Suppose two biologists disagree about the age of a fossil. Mr. One believes the hypothesis (HI) that the fossil is at most 50 million years old; Ms. Two believes H2 , which says that the fossil is at least 60 million years old. They agree that potassium-argon dating is a reliable technique in general and that the potassiumargon dating device they have in their lab is in good working order. This auxiliary assumption (A) permits the two hypotheses to issue in different predictions about what the device will say. The test is performed, and the device says that the fossil is at least 70 million years old. Mr. One has his belief in H, disconfirmed and Ms. Two has hers in H2 confirmed. What happens to their shared assumption about potassium-argon dating in general and the device in particular? According to Quinean holism, the same observation disconfirms A and also confirms A, for A is conjoined with H I in the mind of Mr. One while A is conjoined with HZ in the mind of Ms. Two. Notice that Duhem here does not prohibit the inference that a physical hypothesis is false or that it is disconfirmed by an observation. He simply points out that this conclusion depends on one's being willing to say that the auxiliary assumptions are correct. And the point about auxiliary assumptions is not that one can never have evidence that they are true, but that one cannot be absolutely certain that they are. All this is rather unholistic. 29. The shift from radical to moderate holism makes it important to consider the mathematical propositions that are not used in empirical natural science. What is their epistemological status? Quine (1990, p. 95) gives this answer: 'The main axioms of set theory are generalities operative already in the [empirically] applicable part of the domain. Further sentences... can still be submitted to considerations of simplicity, economy, and naturalness that contribute to the molding of scientific theories generally.' Quine goes on to remark that these considerations justify the axiom of choice and the continuum hypothesis.

QUINE

27 1

There is no difficulty in imagining that these two scientists might react in opposite ways to the device's performance. Perhaps Mr. One has his confidence in potassium-argon dating decline, while Ms. Two has hers increase. This is a familiar human reaction-people sometimes increase their confidence in messengers who bring good news, and sometimes they doubt the reliability of messengers (and even want to kill them) when the news is bad. Ego intrudes in science, just as it does in other walks of life. However, confirmation theory has always tried to separate the confirmation relation that obtains among propositions from the psychological quirks that may attach to people. Surely it is perfectly possible that the auxiliary assumptions in this example are independently confirmed, that both parties rationally agree that this is so, and that the status of these shared background assumptions is entirely unaffected by the experimental result. This cannot happen according to Quine's holism (whether it is extreme or moderate), but surely it can and does (Creath 1991, p. 380)~~' If Mr. One's confidence in A and his confidence in HI crumble simultaneously as a result of the test, what can he say was his reason for interpreting the observation as providing a reason for rejecting H I ? It can't be that he thought and continues to think that potassium-argon dating is a reliable method and that the device in his lab works properly, for this is just what he has abandoned. Rationally rejecting a hypothesis formerly believed typically requires that one not abandon the auxiliary assumptions that are used to test the hypothesis. Holism must be false for this to occur. Quine would interpret the behaviours of our two scientists by saying that they embrace various pragmatic principles, including the principle of minimum mutilation. I have already expressed my reservations about the ideas of centrality and minimum mutilation. I now want to focus on the word 'pragmatic'. What does this word mean, and what does it exclude? As I mentioned in 30. The fact that the auxiliary assumptions A are rationally retained by both parties in this example is not, by itself, a counterexample to the most modest sort of holism, which is the one that Quine endorses in the passage I quoted. Minimal holism says that it is H I and the conditional (if H I then 0 1 ) that stand or fall together and that the same is true of Hzand the conditional (if H2 then 0 2 ) The conditionals (if H I then 0 ) ) and (if Hz then 02)are each weaker than the auxiliary assumption A. However, everything said about A applies also to these two conditionals.

272

I-ELLIOTT

SOBER

the first part of this paper, I take my cue from Pascal's wager; 'pragmatic' means prudential, and it contrasts with evidential. With respect to Mr. One and Ms. Two, I claim that it isn't just 'useful' for them to think, before and after the experiment, that potassium-argon dating is reliable and that their laboratory device is in good working order. They have good evidence that this is so before the experiment is performed, and the result of the experiment does not provide any evidence that these auxiliary assumptions are false. The experiment tests one hypothesis about the fossil's age against another; the auxiliary assumptions are used, but they are not tested.

Nondistributive Holism. I so far have construed holism, whether it is moderate or radical, as obeying a distributive principle-if a conjunction is confirmed or disconfirmed, so are each of its conjuncts. What is true of the whole must be true of the parts. We have seen that this epistemological principle is unsatisfactory, regardless of how inclusive the conjunctions are supposed to be. However, there is a different way to formulate epistemological holism, one that denies distribution. The idea is that confirmation and disconfirmation accrue only to sufficiently inclusive conjunctions; what is true of the whole cannot be affirmed or denied of the parts. This nondistributive version of holism is inconsistent with the Quinean views one finds so often in philosophy of mathematics. One will not be able to say that propositions of pure mathematical are confirmed when they help empirical theories to make accurate predictions. Still, nondistributive holism leads to the conclusion that mathematics and empirical hypotheses are in the same boat, and that might be symmetry enough to vouchsafe the claim that mathematics is just as empirical as natural science.31 In any event, the story of Mr. One and Ms. Two shows that nondistributive holism is implausible. Before the experiment, Mr. 31. It is interesting that hypothetico-deductivism-the view that an observation sentence 0 confirms (disconfirms) X iff X deductively implies 0 and 0 is true (false)leads to nondistributive holism. If (H&A) entails 0 , but neither conjunct entails 0 on its own, then 0 would confirm (and not-0 would disconfirm) the conjunction, but neither conjunct, according to hypothetico-deductivism. Resnik (1997, p. 115) endorses hypothetico-deductivism in his defense of epistemological holism.

QUINE

273

One has a certain degree of confidence in the conjunction H I & A; afterwards, that degree of confidence has declined. Ms. Two likewise begins with a certain degree of confidence in H 2 & A , which increases as a result of the experiment. There is absolutely no reason to believe, either from what Quine has written or on any other basis, that distribution must be a mistake. As mentioned above, it can be perfectly reasonable for the two participants to think that the auxiliary assumption about potassium-argon dating is not tested in this experiment, and so their confidence in A remains unchanged; what happens is that the plausibility of H2 increases and that of H I declines. If the probability of a con~unctiongoes up or down and the conjuncts are probabilistically independent of each other, then at least one of the conjuncts must have its probability move in the same direction. However, it is false that each conjunct must move in the same direction as the conjunction. This is why Bayesianism is inconsistent with nondistributive and distributive holism alike.32Again, the lesson I draw from this does not depend on one's accepting Bayesianism as a completely adequate general account of confirmation. Bayesianism is uncontroversial in limited domains and holism of both sorts is therefore false.

XVI A Duhemian DeJinition of Empirical Unrevisability. It is perfectly consistent with the correct logical point with which Duhem and Quine began that there might be propositions that are empirically unrevisable. If we use the idea mentioned earlier that testing is a contrastive activity-that one tests a hypothesis H I by testing it against a competing hypothesis H2 -then we can formulate the following definition: The choice between competing hypotheses H, and Hz is empirically undecidable, within the context of background beliefs B, iff, for all observations 0, Pr(OlH, & B) = Pr(OlH2 & B), or else the probabilities are undefined." 32. This point about Bayesianism is not new; see, for example, Carnap (1950, pp. 392-393). 33. The last clause (that the probabilities may be undefined) is included because some questions may be empirically undecidable within the context of a set of background beliefs because those background beliefs don't allow the competing hypotheses to confer probabilities on observations.

274

I-ELLIOTT

SOBER

No observation can favour one hypothesis over the other in this circumstance. My reason for focusing on the difference between empirically decidable and empirically undecidable questions, rather than propositions, can be gleaned by considering the following three statements: There is a printed page in front of you. There is a salami in front of you. There is no printed page in front of you, but an undetectable evil demon is causing you to have just the experiences you would be having if there were a printed page in front of you.

Is the first statement empirical? Well, it is possible to cite experiences that discriminate between it and the second statement. However, there is no experience that discriminates between the first statement and the third. Whether a proposition is empirical depends on what you take its competitors to be (Sober 1994a, 1999). Notice that the concept of empirical undecidability just defined is not coextensive with the idea that a proposition can be justified by reason alone, independently of experience. For one thing, the choice between a proposition and its negation might be empirically undecidable and not decidable by rational reflection, either. For another, a proposition might be both empirically decidable and a priori, as the example of Plateau's soap bubble experiment illustrates (see also Rey 1998). The notion of empirical undecidability just defined might be generalized. Perhaps there are questions that are not only empirically undecidable with respect to a single set of background beliefs, but ones that are empirically undecidable relative to a range of possible sets of background beliefs. The strongest such concept would be that of a question that is empirically undecidable relative to all possible sets of background beliefs. This is an empty category,34and not of much interest. However, a more interesting concept can be obtained as follows. Let's begin with a censored version of our current set of beliefs; let Bo consist 34. Simply let B be the pair of conditionals [if HI then 0 and if Hzthen not-01, where 0 is some observation. Trivially, the choice between H I and Hz is empirically decidable, relative to that background assumption.

QUINE

275

of all the beliefs we have, minus the ones that depend for their justification on our believing H 1 rather than H2 (or vice versa). B, is the set of background beliefs that people might now use to see whether observations favour one hypothesis over the other. As we make further observations, we rationally modify B, and so construct a new set of beliefs B1;this, in turn, will be modified by new experience to yield B2, and so on. We now have a sequence of sets of background beliefs B,, B1, B2, etc. Each of these is justified by the observations available at the time, and none of them begs the question of whether HI or H2 is more plausible. Perhaps there are questions that are not only empirically undecidable now, relative to our present set of beliefs B,, but which will remain so, as our background beliefs are modified in the light of e~perience.~' The fact that hypotheses are able to predict observations only via the mediation of background beliefs does not show that this is i m p o s s i b ~ e . ~ ~

XVII Duhernian Circumspection. Although the logical point that hypotheses make observational predictions only when supplemented by background assumptions cannot be said to vindicate the Quinean contention that every belief is revisable in the light of experience, it does suggest that a certain humility is in order when one is asked whether observations could ever justify rejecting this or that hypothesis. It is difficult for us to imagine how there could be consistent alternatives to '2 + 3 = 5' that 35. It also is possible that the choice between H I and Hz is empirically decidable now, but turns out not to be so at a later date. Shifts in both directions are possible. 36. Skyrms (1984) suggests that a 'metaphysical' proposition H is one such that, for all observations 0 , Pr(Hl0) = Pr(H). Since Skyrms is a subjective Bayesian, he is content to have a proposition be metaphysical for one person, but not for another. As explained above, I prefer to define the notion of empirical undecidability as a property of questions, not of propositions. Furthermore, my construal does not require that probabilities be assigned to hypotheses. Grice and Strawson (1956) and Creath (1991) both describe a Duhemian notion of confirmational equivalence and suggest that it might provide a definition of synonymy that breaks out of the vicious circle that Quine criticized. I doubt that this is the concept of meaning that cognitive science requires; it is too verificationist. Confirmationally equivalent sentences can differ in meaning.

276

I-ELLIOTT

SOBER

make different probabilistic predictions about observation^.^^ However, as the historical example of nonEuclidean geometry suggests, this may simply be due to our lack of imagination, which future generations of investigators perhaps will be able to overcome. Putnam (1983, pp. 90-91) thinks that this is the deep objection that 'Two Dogmas' develops against the idea of a priori knowledge. However, this point does not establish that it could make sense to abandon the belief that 2 + 3 = 5 in response to an experimental outcome. The fact that we can't be certain that a proposition is immune from empirical revision does not show that the proposition is, in fact, empirically revisable. Likewise, it may seem obvious that two hypotheses are empirically equivalent, but we should take to heart the possibility that new auxiliary assumptions will be discovered that reveal that the two hypotheses in fact make different predictions. Even if (H1& B,) and (Hz & B,) make the same predictions, it is conceivable that a new set of auxiliary assumptions (B,) will be adopted with the result that (H, & B,) and (Hz & B,) make different predictions. But once again, that this is conceivable does not show H1 and Hz in fact make different predictions. All that follows is that our ability to determine whether the two hypotheses are empirically equivalent may be limited by our lack of omniscience concerning the future of science. These Duhemian admonitions also apply within the realm of the a priori. As Russell's paradox made vivid, the fact that a mathematical postulate seems to be self-consistent is no guarantee that it is. Creative work in mathematics can alter the epistemological status of a mathematical proposition, just as creative work in empirical science can change the status of a physical proposition. However, it would be wrong to conclude from this that no proposition is, in fact, a priori. Russell's paradox and its various remedies were developed without the mediation of empirical evidence. The fact that we can be mistaken in our judgments about what is a priori true doesn't show that nothing is apriori.

XVIII Conclusion. Although Quine attacked two 'dogmas' that he saw in the work of the logical empiricists, I think Quine's philosophy 37. Although contradictions entail everything in classical logic, Pr(O/H) is not defined if Pr(H) = 0.

QUINE

277

embodies two 'dogmas' of its own. The first is Quine's reductionism with respect to meaning and modality (Grice and Strawson 1956; Stein 1992, p. 277). Analyticity is a legitimate concept only if it can be defined in terms of concepts that Quine regards as 'clear'; he imposed the same requirement on the concept of necessity. The second Quinean dogma is epistemological holism (Creath 1991, p. 351, p. 384)-that, in the first instance, it is largish conjunctions of beliefs that are confirmed and disconfirmed; what holism says happens to the conjuncts in those conjunctions depends on whether the doctrine is formulated distributively or nondistributively. The objections I have presented to these two Quinean dogmas are different in character. Against the first, I mustered a characteristically Quinean consideration. The fundamental question about the concept of meaning is whether it is needed to explain and predict. If it is, and if meaning turns out not to be reducible in the way that Quine demands, so be it. I criticized the second dogma by adducing some elementary considerations about the confirmation relation. Quine's discussion of 'minimum mutilation' and of belief systems facing 'the tribunal of experience as a corporate body' needs to be made more precise. When even the first steps are taken in this enterprise, serious problems arise. The linguistic theory of the a priori was very ambitious; it sought to account for all instances of a priori knowledge by tracing them back to facts about language. Quine began to dismantle that imposing edifice in 'Truth by Convention'; 'Two Dogmas of Empiricism' is widely thought to have completed the demolition job, but, in my opinion, it dislodged not a single brick. Not that my criticisms of Quine's criticisms amount to a defense of the views that Quine was attacking. Analyticity should not be understood to mean truth-by-convention. Nor should we think that isolated statements are confirmed or disconfirmed by observations in the absence of background assumptions. If cognitive science produces predictive and explanatory theories that attribute to speakers a knowledge of what their terms mean, then the concept of meaning is on safe footing, and so is the concept of analyticity. And even though Duhem and Quine are right that auxiliary assumptions are needed to bring hypotheses into confirmational contact with

278

I-ELLIOTT

SOBER

observations, there still may be hypotheses that cannot be tested observationally, either now, or as science grows.38

REFERENCES Boghossian, P., 1996, 'Analyticity Reconsidered', Nous 30: 360-391. Carnap, R., 1936-37, 'Testability and Meaning', Philosophy of Science, 3: 41971; 4: 1-40. Carnap, R., 1937, The Logical Syntax of Language (London: Routledge and Kegan Paul). Carnap, R., 1950, The Logical Foundations of Probability (Chicago: University of Chicago Press). Carnap. R., 1952a, 'Meaning Postulates', Philosophical Studies, 3: 65-73. Reprinted in Meaning and Necessity, 2nd edition, pp. 222-229 (Chicago: University of Chicago). Carnap, R., 1952b, 'Quine on Analyticity', in R. Creath (ed.), Dear Carnap, Dear Van-the Quine-Carnap Correspondence, 1990, pp. 427-432 (Berkeley: University of California Press). Carnap, R., 1953, 'Meaning and Synonymy in Natural Languages', Philosophical Studies, 6: 33-47. Carnap, R., 1963, 'W. V. Quine on Logical Truth', in P.A. Schilpp (ed.), The Philosophy of Rudolf Carnap (LaSalle, Illinois: Open Court). Carnap, R., 1966, An Introduction to the Philosophy of Science (New York: Basic Books). Chomsky, N., 1986, Knowledge of Language: Its Nature, Origin, and C're (New York: Praeger). Courant, R. and Robbins, H., 1941, What is Mathematics? (New York: Oxford University Press). Creath, R., 1991, 'Every Dogma Has Its Day', Erkenntnis, 35: 347-389. Duhem, P., 1914, The Aim and Structure of Physical Theory (Princeton: Princeton University Press, 1954). Ewing, A. C., 1940, 'The Linguistic Theory of A Priori Propositions', Proceedings ofthe Aristotelian Society, 15: 207-244. Field, H., 1980, Science Without Numbers (Princeton University Press). Fogelin, R., 1997, 'Quine's Limited Naturalism', Journal of Philosophy, 94: 543563. Frege, G., 1884, The Foundations of Arithmetic, J . L. Austin (trans.) (Oxford: Basil Blackwell, 1968). Friedman, M., 1977, 'Philosophical Naturalism', Proceedings and Addresses of the American Philosophical Association, 71: 7-22. Gibson, R., 1998, 'Quine's Philosophy-A Brief Sketch', in L. Hahn and L. Schilpp (eds.), The Philosophy of W . V. Quine, expanded edition, pp. 667-683 (La Salle, Illinois: Open Court Press). Grice, H. P. and Strawson, P. F.; 1956, 'In Defence of a Dogma', Philosophical Review 65: 141-158. Reprinted in P. Grice, Studies in the Way of Words (Cambridge Ma.: Harvard University Press, 1989). 38. My thanks to Paul Bloom, Tom Bontly, Michael Byrd, Richard Creath, Ellery Eells, Mehmet Elgin, Berent En$, Branden Fitelson, Jean Gayon, Roger Gibson, Geoffrey Hellman, Ernie Lepore, Richard Miller, Greg Mougin, Tony Peressini, Larry Shapiro, Alan Sidelle, Dennis Stampe, and Chris Stephens for comments on earlier drafts.

QUINE

279

Harman, G., 1967, 'Quine on Meaning and Existence I and II', Review of Metaphysics, 21:124-151, 343-367. Harman, G., 1996, 'Analyticity Regained?' Nous, 30: 392-400. Hellman, G., 1992, 'The Boxer and His Fists-the Constructivist in the Arena of Quantum Physics', Proceedings of the Aristotelian Society (Supplementary Volume), 66: 61-77. Hellman, G., 1999, 'Some Ins and Outs of Indispensability-A Modal-Structural Perspective', in A. Cantini, E. Casari, and S. Minari (eds.), Logic in Florence, pp. 2 5 4 0 (Kluwer). Hempel, C., 1965, 'Studies in the Logic of Confirmation', in Aspects ofScientijic Explanation and Other Studies in the Philosophy of Science, pp. 3-46 (New York: Free Press). Kitcher, P., 1983, The Nature of Mathematical Knowledge (Oxford: Oxford University Press). Kneale, W., 1947, 'Are Necessary Truths True by Convention?' Proceedings of the Aristotelian Society, 21 : 118- 133. Kuhn, T., 1970, The Structure of Scienti$c Revolutions, 2nd edition (Chicago: University of Chicago Press). Lewis, D., 1969, Convention (Cambridge: Harvard University Press). Lewis, D., 1973, Counterfactuals, revised edition, 1986 (Cambridge: Harvard University Press). Maddy, P., 1990, Realism in Mathematics (Oxford: Oxford University Press). Maddy, P., 1992, 'Indispensability and Practice', Journal ofPhilosophy, 89: 275289

Mi?(-J., 1859, A System of Logic-Ratiocinative and Inductive (New York: Harper). Musgrave, A,, 1986, 'Arithmetical Platonism-Is Wright Wrong or Must Field Yield?' in M. Fricke (ed.), Essays in Honor of Bob Durant (Dunedin, New Zealand: Otago University Philosophy Department). Pap, A,, 1958, Semantics and Necessary Truth; An Inquiry into the Foundations of Analytic Philosophy (New Haven: Yale University Press). Peressini, T., 1997, 'Troubles with Indispensability-Applying Pure Mathematics in Physical Theory.' Philosophia Mathematica, 5: 210-227. Popper, K., 1963, Conjectures and Refutations (London: Routledge & Kegan Paul). Putnam, H., 1962, 'The Analytic and the Synthetic', in Mind, Language, and Reality-Philosophical Papers, vol. 2, pp. 33-69, 1975 (Cambridge: Cambridge University Press) Putnam, H., 1971, 'Philosophy of Logic', New York: Harper and Row. Reprinted in Mathematics, Matter, and Method-Philosophical Papers (Cambridge: Cambridge University Press, 1975). Putnam, H., 1975a, 'The Meaning of "Meaning" ', in Mind, Language, and Reality-Philosophical Papers, vol. 2, pp. 215-271 (Cambridge: Cambridge University Press). Putnam, H., 1975b, 'The Refutation of Conventionalism', in Mind, Language, and Reality-Philosophical Papers, vol. 2, pp. 153-191 (Cambridge: Cambridge University Press). Putnam, H., 1983, "'Two Dogmas" Revisited', in Realism and Reason-Philosophical Papers, vol. 3 (Cambridge: Cambridge University Press). Quine, W., 1936, 'Truth by Convention', in The Ways of Paradox, pp. 77-106, 1966 (Cambridge: Harvard University Press). Quine, W., 1953a, 'Notes on the Theory of Reference', in From a Logical Point of View, pp. 130-138 (Cambridge, MA: Harvard University Press).

280

I-ELLIOTT

SOBER

Quine, W., 1953b, 'The Problem of Meaning in Linguistics', in From a Logical Point of View, pp. 47-64. (Cambridge, MA: Harvard University Press). Quine, W., 1953c, 'Two Dogmas of Empiricism', in From a Logical Point of View, pp. 20-46 (Cambridge, MA: Harvard University Press). Quine, W., 1960, Word and Object (Cambridge: MIT Press). Quine, W., 1963, 'Carnap and Logical Truth', in P. A. Schilpp (ed.), The Philosophy of Rudolf Carnap (LaSalle, Illinois: Open Court). Quine, W., 1969, 'Epistemology Naturalized', in Ontological Relativity and Other Essays (New York: Columbia University Press). Quine, W., 1970, Philosophy of Logic (Englewood Cliffs, NJ: Prentice-Hall). Quine, W., 1974, The Roots of Reference (La Salle, Illinois: Open Court). Quine, W., 1975, 'On Empirically Equivalent Systems of the World', Erkenntnis, 9: 313-328. Quine, W., 1981, 'Success and the Limits of Mathematization', in Theories and Things (Cambridge: Harvard University Press). Quine, W., 1986, 'Reply to Hilary Putnam', in L. Hahn and L. Schilpp (eds.), The Philosophj) of W . V. Quine (La Salle, Illinois: Open Court Press). Quine, W., 1990, The Pursuit of Truth (Cambridge: Harvard University Press). Quine, W., 1991, '"Two Dogmas" in Retrospect', Canadian Journal of Philosophy, 21: 265-274. Quine, W., 1998, 'Reply to Roger F. Gibson, Jr.', in L. Hahn and L. Schilpp (eds.), The Philosophy of W . V. Quine, expanded edition, pp. 684-685 (La Salle, Illinois: Open Court Press). Reichenbach, H., 1949, 'The Philosophical Significance of the Theory of Relativity', in P. A. Schilpp (ed.), Albert Einstein-Philosopher-Scientist (New York: Harper). Resnik, M., 1992, 'Applying Mathematics and the Indispensability Argument' in J. Echeverria, et a1 (eds.), The Space of Mathematics, pp. 115-31 (New York: Walter de Gruyter). Resnik, M., 1995, 'Scientific vs. Mathematical Realism-The Indispensability Argument', Philosophia Mathematica, 3: 166- 174. Resnik, M., 1997, Mathematics as a Science of Patterns (Oxford: Oxford University Press). Rey, G., 1998, 'A Naturalistic A Priori', Philosophical Studies, 92: 25-43. Ricketts, T., 1994, 'Carnap's Principle of Tolerance, Empiricism, and Conventionalism', in P. Clark (ed.), Reading Putnam (Oxford: Oxford University Press). Sidelle, A., 1989, Necessity, Essence, and Individuation (Ithaca, NY: Cornell University Press). Skyrms, B., 1984, Pragmatics and Empiricism (New Haven: Yale University Press). Sober, E., 1993, 'Indispensability and Mathematics', Philosophical Review, 102: 35-57. Sober, E., 1994a, 'Contrastive Empiricism', in From a Biological Point of View (Cambridge: Cambridge University Press) Sober, E., 1994b, 'The Primacy of Truth-Telling and the Evolution of Lying', in From a Biological Point of View (Cambridge: Cambridge University Press). Sober, E., 1999, 'Testability.' Proceedings and Addresses of the American Philosophical Association, 73: 47-76. Also available at the following URL http:// philosophy.wisc.edu/sober.

Stalnaker, R., 1968, 'A Theory of Conditionals', in N. Rescher (ed.), Studies in Logical Theory (Oxford: Blackwell). Stein, H., 1992, 'Was Carnap Entirely Wrong, After All?' Synthese, 93: 275295.

QUINE by Elliott Sober and Peter Hylton II- Peter Hylton REFERENCE, ONTOLOGICAL RELATIVITY,

AND REALISM

ABSTRACT Section I of this essay discusses Quine's views about reference, contrasting them with those of Russell. For the latter, our language and thought succeed in being about the world because of our acquaintance with objects; the relation of reference-roughly, the relation between a name and its bearer-is thus fundamental. For Quine, by contrast, the fundamental relation by which our language comes to be about the world, and to have empirical content, is that between a sentence and stimulations of our sensory surfaces; reference, while important, is a derivative notion. Section I1 shows how this view of reference as derivative makes possible the notorious Quinean doctrine of ontological relativity. Section 111 raises the issue of realism. It argues that somewhat different notions of realism are in play for Quine and for Russell-for Russell, objects, and our knowledge of objects, play the fundamental role, while for quine objectivity and truth are fundamental, with ontology being derivative.

TQ

here are three things I want to do. One is to talk about ulne's views on reference, contrasting them with those of Russell. The second is to indicate how those views make possible the disputed and seemingly paradoxical Quinean doctrine of ontological relativity. The third, which will be very brief, is to say a little about realism-and here again I shall revert to the contrast between Russell and Quine.

Russell's views on reference form a sharp and useful contrast with those of ~ u i n e . Russell ' postulated a direct and immediate relation between the mind and entities outside the mind, a 1. Russell was an important influence on Quine, second only to Carnap. In a tribute to the latter, Quine says: 'I see him [Carnap] as the dominant figure from the 1930s onwards, as Russell had been in the decades before.' 'Homage to Rudolf Carnap', reprinted in Ways of Paradox (Cambridge, MA: Harvard University Press, 1966, second ed. 1976), p. 40. I shall speak here of Russell's views in the first two decades of this century, and not be concerned with subsequent shifts of doctrine.

282

11-PETER

HYLTON

relation he called acquaintance; this relation he held to lie at the base of all knowledge. Thus, in Problems of Philosophy he says: 'The faculty of being acquainted with things other than itself is the main characteristic of a mind. Acquaintance with objects essentially consists in a relation between the mind and something other than the mind; it is this that constitutes the mind's power of knowing thing^.'^ His insistence upon the directness and immediacy of the relation is to be explained, at least in part, in terms of his opposition to Idealism. The Idealists held that our knowledge is always mediated by a complex structure; we can thus have knowledge of the world, at least as far as it is knowable, which is not of a straightforwardly empirical sort. It was in reaction to this that Russell, along with G. E. Moore, postulated the notion of acquaintance. It was to be a cognitive relation, holding between the mind and objects, which relied upon no kind of structure or theory: a pre-conceptual relation, and an immediate relation, rather than a mediated one. The mind is, as it were, in direct contact with objects outside it. There are reasons, arising especially from the existence of illusion and error, for thinking that we cannot have this kind of immediate cognitive relation to ordinary physical objects. Russell came to think that we are acquainted both with abstract objects and with sense-data (as well as the contents of our own minds, and past objects of acquaintance). Russellian sense-data, it is worth adding, are the appropriate relata for the relation of acquaintance, and are outside the mind.3 The relation of acquaintance, for Russell, provides an answer to the question: how do our thoughts and our language succeed in being about the world outside our minds? Acquaintance is the point at which this contact is made. While we appear to know about many things which are not objects of acquaintance, all such knowledge must ultimately be reducible to knowledge of objects which are. Thus in Problems of Philosophy, again, Russell says: 'Every proposition which we can understand must be composed wholly of constituents with which we are acquainted.' (p. 58; 2. Bertrand Russell, The Problems of Philosophy (Oxford: Oxford University Press, 1912, reset edition 1946), p. 42. 3. For an elaboration of this very hasty summary, see the present author's Russell, Idealism, and the Emergence of Analytic Philosophy (Oxford: Oxford University Press, 1990).

QUINE

283

emphasis in the original). This demand gives rise to an extremely far-reaching programme of philosophical analysis: to show that sentences which appear to express propositions which violate this demand in fact express propositions which are in accord with it. In a fully analysed sentence-one which reveals in fully explicit fashion the structure of the proposition which it expressesevery word will refer to an object of acquaintance. (This does not imply that all of our sentence are about objects of acquaintance; Russell employs the theory of descriptions to show how propositions which contain only objects of acquaintance may be about objects with which we are not acquainted.) Two features of this view are salient from our point of view. First, acquaintance is entirely presuppositionless, and wholly independent of beliefs and concepts. It is a pre-conceptual and pre-theoretical relation; this is why it can serve as the foundation for all theory. Second, it is a relation between a person and an object, not between a person and a fact or a proposition or a sentence. For Russell, then, reference, a version of the relation between a name and the named object, is a presuppositionless relation which is at the foundation of all knowledge. Quine, as we shall see, rejects these Russellian views entirely, in every aspect. It is, however, worth noting that he does not reject the sort of question to which those views are an answer. We might put the question like this: how do we come to have knowledge at all? or: how can our thoughts be about the world? Quine in fact has an answer to questions of this sort, as he construes them. And he construes them, as he does all serious questions, along naturalistic lines-roughly, as scientific questions. He answers them in terms of the effects that the world has on us, and of our responses to those effects. Various forms of energy, most obviously light, sound, heat, and pressure, impinge on the surfaces of our bodies; in some cases, a person responds differentially to such impingements. Thus we may respond in one way when a certain pattern of light is impinging on our eyes, and in another way when it is not. The body contains sensory surfaces, which are stimulated by the impingement of the relevant forms of energy; these stimulations affect behaviour. It is a long way from forms of energy impinging on the body to our discourse and our thought being about the world. We need to see, in broad outline, how Quine thinks we get from one

284

11--PETER HYLTON

to the other, from the impingement of energy on our sensory surfaces to the empirical content of our theories. The key here is Quine's notion of an observation sentence. (The notion of an observation sentence is a bit more complicated than I'll be suggesting here, or than Quine himself often suggests; but I don't think the complications will matter for present purposes, so I will not go into them.) An observation sentence satisfies three conditions. First, it is a complete utterance whose truthvalue varies with occasion of utterance ('There's milk in the glass', rather than 'Milk is good for young children'). Second, for each individual, it is directly tied to stimulation of that individual's sensory surfaces: the individual's willingness to utter the sentence, or to agree when someone else utters it, depends only on which of that individual's sensory surfaces are being stimulated at that time. Third, from the point of view of the community of language-speakers, observation sentences are utterances about which there will be general agreement, in any given circumstances. If I am in a situation which stimulates my sensory surfaces in such a way as to lead me to agree or disagree with a particular observation sentence, then any other speakers of the same language, who are (as Quine says) 'witnesses to the o c ~ a s i o nwould '~ have their sensory surfaces stimulated in a way that results in the same verdict. Observation sentences will thus be uncontroversial, to the point of triviality. 'It's raining', 'It's red' (or just 'Red'), and 'There's a horse', might count as examples. Not by chance, these are the sorts of sentences that an infant learning the language acquires first: they have no presuppositions, their mastery does not at all depend upon a mastery of other aspects of the language. Within the small community of language-speakers where the infant lives, other sentences will count as observational: 'Mama', for example, or 'Rover', said of the family dog.5 In spite 4. From Stimulus to Science (Cambridge MA: Harvard University Press, 1995), p. 22. 5. Implicit here is a point which would require elaboration in a more detailed treatment: what counts as an observation sentence depends on the size of the community of language-speakers. For some purposes it is relevant to think of all normally functioning speakers of the language, but for the purposes of explaining a child's acquisition of language a smaller community-perhaps a very small one-may be appropriate. In other ways too the notion of an observation sentence is less cut-and-dried than I am making it sound here. Observationality is a matter of degree, although Quine has not always been clear on this point, and may vary along a number of dimensions.

QUINE

285

of their triviality, observation sentences are, in Quine's view, at the foundation of all knowledge. Everything we know about the world is due to the impingement of energy on our sensory surfaces; this for Quine is a scientifically established fact. And observation sentences encapsulate those impingements, insofar as they are relevant to our knowledge. All of our knowledge, however theoretical, is thus ultimately answerable to observation sentences. Our theories are empirically testable because they imply what Quine calls observation categorical~.These are, roughly, sentences saying that whenever one observation sentences holds, another will also hold ('where there's smoke there's fire', for example). An empirical test, in extreme cases, will consist in finding some observation categorical which, according to the theory, should be true, and then seeing whether falsifying instances can be found.6 It is just the uncontroversial nature of observation sentences which fits them to play this role. By the same token, Quine defines the empirical content of a set of sentences as the observation categoricals which it implies. Russell asked how our words and thoughts came to be about the world. Quine's answer to that question, as he construes it, is: via the impingement of energy on the sensory surfaces of our bodies, as this is encapsulated in observation sentences, and in the observation categoricals formed from them. Now this answer, unlike Russell's, has nothing directly to do with reference, in the sense of the relation between a name and its bearer. The willingness to utter, or to assent to, an observation sentence, is brought about by the occurrence of the relevant stimulation of It may be that no sentence is fully observational; but this does not suggest that there are no sentences that are sufficiently observational to play the sort of role for which he requires them. 6. Because observation sentences are, in given circumstances, quite uncontroversial, it will also be uncontroversial whether given circumstances falsify an observation categorical. But since an observation categorical is a generalization, its truth will not vary from occasion to occasion, and may be quite controversial. Note also that in practice a test of a theory will seldom need to go all the way down to the level of observation sentences and the corresponding observational categorical~.Almost always there will be higher-level sentences which are still agreed upon by the various parties to a dispute; perhaps we should think of these sentences as observational for the linguistic community concerned (advanced organic chemists, or whatever the relevant group may be). But still in such cases there will be more fully observational sentences to which appeal could be made if there were no agreement at higher levels, and Quine's idealization of the process seizes upon this fact.

286

11-PETER

HYLTON

your sensory surfaces. But the observation sentence does not refer to such stimulation, or to the energy impinging on your surfaces. Obviously the child happily uttering 'Mama' in the presence of its mother is not in any sense talking about the light reaching its retinas, or the stimulations of the retinal nerves. To talk about those things takes a good deal of linguistic and theoretical sophistication, whereas a crucial fact about observation sentences is that one can master them while completely devoid of all such sophistication. It is for this reason that they are the child's 'entering wedge' into language, as Quine says more than once;7 it is for this reason also that they can play the role of ultimate evidence, for while may we disagree about all the sophistication we do not disagree about observation sentences. Observation sentences, then, do not refer to stimulations of our sensory surfaces, or anything of that kind. D o they refer to other objects? This question is trickier. It may look as if the utterance 'Mama', by a particular child, refers (let us say) to that child's mother. In that context, indeed it may seem simply to be a name of that person. Equally, it may look as if the sentence 'There's a horse' contains the term 'horse', which (of course) refers to horses. If we think of an observation sentence as it might occur in the discourse of an adult, who has mastered the language and possesses a range of knowledge about various matters, then these appearances are correct. Thought of in that way, an observation sentence will typically contain, or simply be, a referring term. Quine claims, however, that this is true only because the adult has mastered not only the use of the observation sentence, but also the use of more sophisticated parts of the language. To refer to a horse, he argues, it is not sufficient to say 'Horse' or 'There's a horse' when one is in the presence of a horse. One needs also to have some idea of the answer to such questions as: what counts as one horse, and what counts as two? under what circumstances do we have the same horse over again, and when do we have a new horse? The beginner, who simply makes the noise 'Horse' when receiving (say) the sort of visual stimulations which one typically receives when looking at a horse (or, more realistically, at the picture of a horse) cannot even formulate such questions, much less begin to answer them. Yet is 7. For example Pursuit of Truth (Cambridge MA: Harvard University Press, 1990), p. 5.

QUINE

287

it, Quine insists, only in the mouth of one who has these capacities that such words can be said to refer at all: there is more to reference than merely making a sound in response to patterns of stimulation. Another way to make this point is to say that observation sentences are indeed sentences. Even a one-word observation sentence, such as 'Mama', or 'Red', or 'Horse', must be counted in context-uttered all by itself, in response to present stimulation-as a sentence, and not as a referring term. Clearly the criterion of sentencehood here is not narrowly syntactical. What does it mean to say that observation sentences are indeed sentences, and not referring terms? In part it means that an observation sentence is complete: all by itself, as it stands, it is to be assessed as correct or incorrect, true or false. A referring term, by contrast, picks out an object or a kind of object in order to say something about it; only in the wider context in which something is said about the object does the question of correctness arise. In part also insisting that observation sentences are sentences and not referring terms means that no question of reference arises until we analyse observation sentences, break them down into significant parts. This analysis enables us to see them as made up of parts which may refer. The analysis, however, draws on more sophisticated parts of the language, which are not required simply for an initial mastery of the observation sentence. (All of this applies equally to one-word observation sentences, though in such cases the only significant part is orthographically identical to the whole.) The upshot of this discussion is that the means by which language comes into contact with the world, in Quine's view, is not referential. Instead, the crucial relation is, roughly speaking, that between an observation sentence and the circumstances which make the utterance of that sentence correct or incorrect in the eyes of the community of language-speakers as a whole. (To speak less roughly: these are circumstances which for any speaker of the language will give rise to stimulations of that person's nerve endings which will lead him or her accept the sentence. The long version here makes a difference, because one might think of circumstances as made up of objects in various relations. For Quine, however, it is the stimulations, not the objects, which are crucial.) This is not to say that Quine denies that our language

288

11-PETER

HYLTON

contains expressions which refer to the world; on the contrary, he insists upon the referential aspect of language. In the Preface to Word and Object, he describes that book, beyond its first two chapters, as 'reflect[ing]... on the development and structure of our own referential apparatus'.' And clarifying the referential structure of our language is, in his view, an important part of the philosopher's task. What he does deny is that reference is fundamental, in the sense in which it is fundamental for Russell. Reference, for Quine, is a derivative notion; the fundamental notion is that of the relation between a complete utterance and stimulations of our sensory surfaces-those stimulations which would occur in circumstances which lead the sentence to accepted by speakers of the given language. If reference is not the foundation of language-use, how can language be referential at all? This is a question which Quine takes seriously-construing it, of course, as a scientific question-and to which he has suggested an a n ~ w e rWhat . ~ is important, for his philosophical view as a whole, is not so much the details of the answer but that there be some answer, of a purely naturalistic sort, compatible with the basic picture; for this reason the answer he puts forward is avowedly tentative and speculative.'0 On the account that he suggests, the first steps of reification take place very early, and at the observational level. There are situations in which the observation sentences 'Blue' and 'Pebble' are both appropriately asserted; some but not all of those situations make it appropriate to utter 'Blue pebble'. Those that do can already, in retrospect, be seen as postulating an object-a pebble-and saying something about it. It is only with hindsight, however, that this distinction at the observational level should be seen as genuine reification, as introducing objects in the familiar sense. Objects such as pebbles have 8. Word and Object (Cambridge Ma: MIT Press, 1960), p. ix. 9. Most obviously in Roots of Reference (La Salle: Open Court, 1974) and in parts of Word and Object. 10. Thus Quine says: 'In Roots of Reference I was posing a Kantian sort of question: how is reification possible?'; 'Comment on Parsons', in eds. R. Barrett and R. Gibson, Perspectives on Quine (Oxford: Basil Blackwell, 1990), p. 291. Again, in Pursuit of Truth, he says that one of the advantages of beginning an account of language-acquisition with observation sentences is that we 'are freed to speculate on the nature of reification and its utility for scientific theory' (p. 8, my emphasis).

QUINE

289

identity conditions which provide for their re-identification from time to time. This is a matter which cannot be dealt with at the purely observational level, for it involves rudimentary physical theory. Thus Quine says: 'reification of bodies across time is beyond the reach of observation sentences and categoricals. Substantial reification is theoretical.'" Reification also requires the ability to use words such as 'same', 'another', 'an', means of forming plurals, relative pronouns, and so on. These expressions and transformations, Quine suggests, are acquired contextually, together, and little by little.I2 We might again put the point by saying that merely standing in front of something making a noise is not yet naming an object, not even if one consistently makes the same noise when in front of (or when wanting) the same object. As another author put it, 'a great deal of stage-setting in the language is presupposed if the mere act of naming is to make sense'.I3 For the adult language-user the stage is set, hence we are likely to see the child's first recognizable noises as already full-fledged names. But the capacity to name things, in our sense, is acquired little by little. There is, of course, far more to be said about Quine's account of the acquisition of fully referential language-and far more than Quine says about the process itself. But it is not my purpose here to explore these matters in any detail. What is important for us is that reference, for Quine. is not the fundamental relation between language and the world; it is not the means by which language acquires its empirical content and comes to be about the world. That relation is, rather, the relation between an observation sentence and the situations which typically lead to the sorts of stimulations under which the sentence is uttered, or commands assent when uttered by someone else. Reference, on Quine's account, is a relation between the language, or some 11. Pursuit of Truth, p. 25

12. See Word and Object, p. 93. In a striking image. Quine there compares such acquisition to a climbers ascent of a 'chimney', o r narrow space between two rock faces: 'The contextual learning of these various particles goes on simultaneously, we may suppose, so that they are gradually adjusted to one another and a coherent pattern of usage is evolved matching that of society. The child scrambles up an intellectual chimney, supporting himself against each side by pressure against the others.' 13. Ludwig Wittgenstein, Philosophical Investigations (New York: Macmillan, 1953), section 257.

290

11-PETER

HYLTON

linguistic expressions, and the world, just not one which is fundamental in the sense we have indicated. The capacity to refer is a capacity to use language which is more sophisticated than the most elementary, purely observational, sort. A child acquires this capacity, little by little, as it grows up.

I turn now to the famous, even notorious, Quinean doctrine known as 'ontological relativity' or the 'inscrutability of reference'.14This form of relativism about ontology is not derivative upon relativism about truth: on the contrary, it occurs even if we suppose all problems about truth to be settled-even if we suppose that we possess a unique theory of the world whose complete truth is not in doubt. Such a completely true theory would, after all, be a body of true sentences. The referential burden of language, however, is not simply a matter of which sentences are true, but also a matter of how we see those sentences as making ontological claims. It is thus a matter of how the whole sentences are analysed into parts and how significance is attributed to those parts. Here, in the gap between the significance of the sentence and the significance of its parts, Quine sees empirical slack, which is manifest in ontological relativity. We have already seen a clear example of the gap between the significance of the whole sentence and the significance of its constituent parts. In the case of observation sentences, we saw that, taken as wholes, they got their meaning simply from their relation to sensory stimulation. Thought of simply in that way, as responses to stimulation, observation sentences do not make any ontological claims. It become possible to see them as making such claims only when they are seen as part of a more sophisticated theory, which includes the notion of identity, means of forming plurals, and so on. To repeat an earlier point: the relation between language and the world by which our language comes to have empirical meaning and empirical content is not a relation between names and objects but rather a relation between sentences and sensory stimulations. 14. Quine makes no distinction here; see eds Edwin Hahn and Paul A. Schilpp, The Philosophy of W . V. Quine (La Salle: Open Court, 1986; second edition 1998), p. 459.

QUINE

29 1

This idea is fundamental for all of Quine's views about language, and underlies ontological relativity. For the idea applies not only to observation sentences but to all sentences, to language as a whole. It is most clearly seen in the case of an observation sentence, for this is a sentence whose empirical meaning is wholly exhausted by its relation to sensory stimulations. Other sentences are responsive not only to current stimulations but also to the other beliefs which one has; if two scientists, say, agree about what they are currently seeing but disagree about whether it is evidence for some other claim, this will surely be because of further disagreements between them. The claim is thus answerable not only to what they are currently observing, but also to the other things about which they disagree. The empirical meaning of such a sentence is not exhausted by its links to current stimulation; it also consists of its links to other sentences. These sentence-to-sentence links, and the dependencies of meaning which they set up, are far too complex and many-sided for us be able to reconstruct them. Let me quote a passage of Quine on this point. He is considering the inference which a chemist may make from seeing a greenish tint in a test-tube to saying 'There was copper in it'. Such an inference-from the observation sentence 'It's greenish', let's say, to 'There was copper in it'-obviously draws on the general background knowledge which our chemist possesses. Quine comments: The intervening theory is composed of sentences associated with one another in multifarious ways not easily reconstructed even in conjecture. There are so-called logical connections, and there are so-called causal ones; but any such interconnections of sentences must be due finally to the conditioning of sentences as responses to sentences as stimuli. If some of the connections count more particularly as logical or causal, they d o so only by reference to so-called logical or causal laws, which in turn are sentences within the theory. The theory as a whole ... is a fabric of sentences variously associated to one another and to non-verbal stimulation by the mechanism of conditioned response. (Word and Object, p. 11)

The meaning of 'There was copper in it' is thus by no means exhausted by its direct links to stimulation-by an account of which stimulations would prompt one to agree with it, and which

292

11-PETER

HYLTON

to disagree with it. The sentence also plays a theoretical role, and is thus linked to countless other sentences of our theory. For our immediate purposes the thing to emphasize about this account of the significance of language is that it all takes place at the level of sentences. It is observation sentences which are linked to stimulations; it is links between one sentence and others which plays a role in the meaning of non-observational sentences. We might put the point like this. An observation sentence is directly linked to stimulations, and those links determine its correct use and thus its meaning. For a non-observation sentence links to stimulation are equally important, but in this case those links are partly or wholly indirect: the sentence is linked to other sentences which are in turn linked to other sentences and so on, terminating in observation sentences, or in the observation categorical~formed from them. This terminus provides the empirical meaning for any sentence, however indirect and complex the connections may be. (Quine's holism further complicates this already complicated picture, for it has the consequence that in many cases a sentence has a link to another only when it is taken as part of a more inclusive class of sentences. Such links, and hence the more inclusive class, would have to be taken into account in considering the meaning of the individual sentence. But this does not affect the point here, which is that the empirical meaning of sentences consist in their relation to stimulation, whether direct, as in the case of observation sentences, or via other sentences or classes of sentences.) What we said about observation sentences may thus be generalized to all sentences, though with a crucial qualification. For any sentence, there is a sense in which what one needs to know to be a competent user of that sentence does not depend upon its structure, upon its analysis into constituent parts. If one uses the sentence as a whole correctly, both in relation to stimulations and in relation to other sentences, then one is a competent user of it; that is all that can be required of one, for that is all there is to language-use. But now the qualification is that it will in fact be impossible for us to use theoretical sentences correctly without an analysis of those sentences into significant constituent parts. Take the most obvious case, that of the 'so-called logical connections' between sentences. A logical connection may be thought of as summing up infinitely many facts of the form: whenever it

QUINE

293

is appropriate to assert that sentence it is also appropriate to assert this one. But we could not learn all of these facts one-byone; there are infinitely many of them. What we learn-implicitly at first, explicitly from logic books-is that whenever it is appropriate to assert a sentence of this form, it is appropriate also to assert a corresponding sentence of that forin; this sums up infinitely many facts in a perspicuous, learnable, fashion. But this requires that we attribute forms to sentences, i.e. see them as made up of constituent parts and patterns which are significant and which recur in other sentences. Let us see where we are in the argument. Reference is a relation between words-significant constituent parts of sentences-and objects. The fundamental relation of language to the world, the relation which guarantees empirical significance to our language, however, does not take place at that level, but rather at the level of the relation between sentences and stimulations. We are bound to attribute significance to constituent parts of sentences because it is only in that way that we can see the patterns and analogies in language which make it possible for us to use it. This situation leaves it open that there might in fact be more than one way to attribute structures to our sentences, or at least more than one way in which to attribute sense and reference to the parts of sentences. Note that if there were more than one account of these matters, then there would be no way to bring evidence to bear on the question, which of them is correct. The evidence here consists of language-use, which, according to Quine, is the use of sentences. The imagined dispute is between two ways of breaking sentences down into constituent parts, or attributing significance to those parts. The differences here will, we are supposing, make no difference to the ways in which the whole sentences are used. (Note also that, for the same reason, the idea of a causal explanation of how language is learned and how it functions will not differentiate between the two accounts. For Quine it is our acquisition of the ability to use sentences which is to be explained; an explanation of the ability to refer arises out of this account. Reference is not an independent phenomenon, demanding an independent explanation. The point, again, is that reference is derived, not fundamental).

294

11-PETER

HYLTON

So, on Quinean assumptions, there might be two ways of attributing significance to the parts of our sentences, and no way, even in principle, of deciding between them. Here we have, in outline, Quine's doctrine of ontological relativity, or at least the necessary background of that doctrine. Let us see how Quine encapsulates some of the points we have been making: Reference and ontology recede thus to the status of mere auxiliaries. True sentences, observational and theoretical, are the alpha and the omega of the scientific enterprise. They are related by structure, and objects figure as mere nodes of the structure. What particular objects there may be is indifferent to the truth of observation sentences, indifferent to the support they lend to theoretical sentences, indifferent to the success of the theory in its predictions. (Pursuitof Truth, p. 31)

Objects, and reference to objects, are central to our theory of the world, for we cannot get by without analyzing our sentences so as to reveal structure: otherwise we could never master the sentence-to-sentence links which are essential to language beyond the observational level. For all that, however, the role of objects is secondary; the truth of sentences, and their links to one another, are primary. Hence the idea that there could be more than one way of analyzing sentences, while leaving untouched both their truth-values and all the infinitely complex links among them. So far we have spoken of this idea as no more than an abstract possibility. Quine, however, is in no doubt at all that there actually are such alternative methods of analysis. He says, indeed, that the matter 'admits of trivial proof .I5 But I'm not going to go into this, because I want to leave time to say a little about realism. It is, however, relevant to say a very little about the allegedly paradoxical consequences of ontological relativity. The threat of paradox arises from a seeming regress: you can translate my utterances of the word 'rabbit', say, as not really being about rabbits but as about something else; but then your words too can be translated in any one of various ways by someone else, and so on. Where is this regress to end? Quine says: 'in practice we end the regress of background languages by acquiescing in our 15. 'Reply to John Woods' Hahn and Schilpp, second edition only, p. 728.

QUINE

295

mother tongue and taking its words at face value.' ('Ontological Relativity', p. 49). (These are words which some have found more helpful than others.) It is crucial here that what give our words their meaning is our use of them, not our translations of them.16 For it is with translation, but only with translation, that the issue of inscrutability of reference arises; while we are simply using our language there is no such issue. Let us see two longer comments of Quine's on this subject: To say what objects someone is talking about is to say no more than how we propose to translate his terms into ours.... The point is not that we ourselves are casting about in vain for a mooring. Staying aboard our own language and not rocking the boat, we are borne smoothly along on it and all is well; 'rabbit' denotes rabbits, and there is no sense in asking, 'Rabbits in what sense of "rabbit"?' Reference goes inscrutable if, rocking the boat, we contemplate a permuational mapping of our language, or if we undertake translation. (Theories and Things , p. 20.) And, again:17 Within the home language, reference is best seen (I now hold) as unproblematic but trivial, on a par with Tarski's truth paradigm. Thus 'London' denotes London (whatever that is) and 'rabbit' denotes rabbits (whatever they are). Inscrutability of reference emerges only in translation. The point here, I think, is that while we are speaking our familiar language the inscrutability of reference simply gets n o grip. It does not in any way interfere with language-use, or force us to modify our account of that use. And this is true also of that special case of language-use in which we use some words to say what other words refer to. There is, of course, n o saying what objects a word refers to except by using, and thereby taking for granted, some language. But this should not seem threatening or paradoxical: there is no saying anything except by using, and taking for granted, some language. If this continues to seem 16. I have attempted a more discursive treatment of the issues discussed in the next two paragraphs in, 'Translation, Meaning, and Self-Knowledge'; Proceedings of the Aristotelian Society, vol. XCI, Part 3 (1990/91), pp. 269-290. 17. 'Reply to Paul A. Roth', eds. Hahn and Schilpp, p. 460. The emphases here are Quine's.

296

11-PETER

HYLTON

paradoxical it is perhaps because the point we made at the start, when contrasting Quine with Russell, is hard to absorb. We tend to treat the notion of an object as fundamental, rather than seeing objects merely as 'neutral nodes [in] the structure of [our] theory' (Pursuit of Truth, p. 33). We tend, therefore, to think that there should be something that an object really and truly is, is in itself, which outruns the role it plays in our theory of the world, or in some as yet only imaginary wholly correct theory. The lesson of Quine's approach to reference, by contrast, is that there is no more to an object than its role in theory; ontological relativity drives that lesson home.18

Finally I should like, very briefly, to say something about realism. Russell, of course, was a realist-not, at some points, about physical objects, but always about the objects with which we are acquainted. You can't be acquainted with something unless it's really there-somewhere-for your mind to latch on to. How are we to understand the notion of an object, an of an object's being there in a sense which abstracts from the idea of its being in space and time? Can we make sense of the idea of an object's existing or not existing as a primitive idea, not explicable in other terms? I think that this idea has a sense for Russell because it draws on the idea of acquaintance-of our being in immediate epistemic contact with an object. Of course there is no definition of existence in terms of acquaintance; acquaintance, indeed, presupposes existence. But the idea that the notion of the existence of an object makes sense-that the notion of an object itself makes sense-without any further explanation is I think bolstered by the role that acquaintance plays in Russell's thought. -What I'm talking about here is not the truth or the plausibility of Russell's views so much as their sense: what gives 18. In 1968 Quine said: 'What makes sense is to say not what the objects of a theory are, absolutely speaking, but how one theory of objects is interpretable or reinterpretable in another.' (Ontological Relativity New York, Columbia University Press, p. 50). I think now he might be inclined to say that the idea of speaking absolutely in this sense is a mere chimera from the start, and not an apparently sensible idea to be ruled out-not, so to speak, something that we might be able to do but cannot, but rather an incoherent idea which dissolves under examination. See again the essay referred to in footnote, 16, above.

QUINE

297

his ideas the meaning which they have in the system of his thought? And my suggestion is that his notion of acquaintance helps to give sense to his ideas of an object and of existence. I sometimes think of these Russellian ideas, and of the conception of realism that goes along with them, as pre-Kantian. They make the idea of an object, and of an object's existing, foundational; the notion of judgment plays a secondary role. It is no accident, of course, that Russell's thought throughout the relevant period is self-consciously anti-Idealist, and in particular anti-Kantian. Quine, however, is also a realist-or at least he repeatedly says he is: 'a robust realist', he sometimes calls himself. But clearly there is a rather different conception of realism involved here. We might call it a Kantian conception, because it makes judgment, and the truth or objectivity of judgment, fundamental notions; ontology becomes derivative. On this account objectivity is fundamental, and must of course be explained in a way which does not presuppose objects; on the pre-Kantian view objects are fundamental, and objectivity will presumably be explained in terms of them. One strand of thought in analytic philosophy has clung to the pre-Kantian conception of realism, though seldom quite as explicitly as Russell did; but that has, I think, always been only one strand. It is worth emphasizing this point, because in our thinking about the tradition we are, I think, often inclined to interpret it all in the pre-Kantian vein. Much of the recent reinterpretation of Frege which has taken place over the last fifteen years or so has essentially involved coming to terms with the fact that what realism amounts to in Frege's hands is quite unlike Russell's realism.I9 And one thing-though not I suspect the only thing-going on in Putnam's adoption in the 1970s of what he calls 'internal realism' is his coming to accept a Kantian conception of realism, or perhaps just to appreciate the implications of such a conception. So that's one thing I wanted to do, just to point out the existence of these two conceptions of realism within the analytic tradition. 19. See for example T. G. Ricketts, 'Objectivity and Objecthood: Frege's Metaphysics of Judgment', Frege Synthesized, ed. L. Haaparantant (Dordrecht: Reidel, 1986), pp. 65-95.

298

11-PETER

HYLTON

Another thing I want to do-and this is the last-is to raise the question of the compatibility of Quine's realism with ontological relativity. What is left of our insistence that the objects we refer to are real, if we acknowledge that it is indeterminate just which objects those are? 'Rover' must be taken as referring to a real object, we insist; we are then bound to add, however, that this object may be the family dog, or its space-time complement, or a set of canine time-slices, or a set of real numbers corresponding to the space-time points occupied by the dog, or who knows what else? What is the force of the original insistence, in the face of these additions? It is hard to dismiss these questions entirely. For Quine, however, I think their force is not so much to undermine realism about objects as to show us what it can coherently come to. The clue here is provided by an idea already emphasized: that there is no more to an object than its role in the theory-its role in the totality of our beliefs, or the totality of truth about the world. Quine, taking the logician's ingenuity to the point of perversity, adds to this idea the claim that we can systematically switch objects from role to role. The result is ontological relativity. By re-construing predicates, we can, for example, switch the roles of Rover and his space-time complement, and so on for all other physical objects. The possibility of this sort of switching, if granted, may seem to undermine realism by indicating that we don't really know what we are being realistic about; we feel as if we are in the odd position of insisting that something must exist, but having to acknowledge that we cannot say what. But if all that there is to an object is the role that it plays in theory, then what is it that we do not know? According to Quine, as we saw, the correct way to think of an object is simply as marking a certain role-a 'neutral node'-in the structure of our theory. So it would seem that all that being a realist about an object can come to is, so to speak, taking that role, that node, seriously. But this is simply to take seriously the theory of which it is an aspect. So there is no issue concerning realism about objects which is separate from the issue of realism about the theory which mentions them: to repeat, ontology is derivative upon truth; hence, if we are realists about truth we are more or less automatically realists about objects too-ontological relativity notwithstanding.

QUINE

299

This conclusion may seem to be too anodyne to be the moral of the seemingly bizarre idea of ontological relativity, for really it just repeats the ideas of the start of the paper, where we contrasted Russell with Quine, which seemed moderate enough. I am inclined to think, however, that those ideas are less moderate-less in conformity with unreconstructed common sensethan may have appeared; and that the apparent oddness of ontological relativity was lurking from the start. We are, as Quine says, ' b ~ d ~ - m i n d e dit' ;comes ~ ~ entirely natural to us to think in terms of bodies, and of objects more generally, and to take them as fundamental to our knowledge. The pre-Kantian view has considerable appeal to common sense. Russell's way of articulating this nexus of ideas, when dissected and exposed to the light of day, may easily come to seem offensive to common sense, but it is an attempt to articulate ideas which are, I think, very natural. The apparently paradoxical character of ontological relativity arises in part because Quine's ingenuity enables him to draw conclusions that most of would never have dreamed of. But it also arises because the notion of an object, and the nexus of ideas that goes along with it, goes very deep in our conception of the world. Ontological relativity dramatizes the fact that Quine's work repudiates those ideas complete^^.^'

20. Roots of Reference, p. 54 21. For comments and discussion I should like to thank W. D. Hart and Gary Kemp, as well as audiences at Northwestern University, the University of California at Davis, Edinburgh University, Glasgow University, and Ohio State University.

http://www.jstor.org

LINKED CITATIONS - Page 1 of 2 -

You have printed the following article: Quine Elliott Sober; Peter Hylton Proceedings of the Aristotelian Society, Supplementary Volumes, Vol. 74. (2000), pp. 237-299. Stable URL: http://links.jstor.org/sici?sici=0309-7013%282000%2974%3C237%3AQ%3E2.0.CO%3B2-A

This article references the following linked citations. If you are trying to access articles from an off-campus location, you may be required to first logon via your library web site to access JSTOR. Please visit your library's website or contact a librarian to learn about options for remote access to JSTOR.

References Analyticity Reconsidered Paul Artin Boghossian Noûs, Vol. 30, No. 3. (Sep., 1996), pp. 360-391. Stable URL: http://links.jstor.org/sici?sici=0029-4624%28199609%2930%3A3%3C360%3AAR%3E2.0.CO%3B2-P

Testability and Meaning Rudolf Carnap Philosophy of Science, Vol. 3, No. 4. (Oct., 1936), pp. 419-471. Stable URL: http://links.jstor.org/sici?sici=0031-8248%28193610%293%3A4%3C419%3ATAM%3E2.0.CO%3B2-U

Quine's Limited Naturalism Robert J. Fogelin The Journal of Philosophy, Vol. 94, No. 11. (Nov., 1997), pp. 543-563. Stable URL: http://links.jstor.org/sici?sici=0022-362X%28199711%2994%3A11%3C543%3AQLN%3E2.0.CO%3B2-3

Philosophical Naturalism Michael Friedman Proceedings and Addresses of the American Philosophical Association, Vol. 71, No. 2. (Nov., 1997), pp. 5+7-21. Stable URL: http://links.jstor.org/sici?sici=0065-972X%28199711%2971%3A2%3C5%3APN%3E2.0.CO%3B2-9

http://www.jstor.org

LINKED CITATIONS - Page 2 of 2 -

In Defense of a Dogma H. P. Grice; P. F. Strawson The Philosophical Review, Vol. 65, No. 2. (Apr., 1956), pp. 141-158. Stable URL: http://links.jstor.org/sici?sici=0031-8108%28195604%2965%3A2%3C141%3AIDOAD%3E2.0.CO%3B2-E

Analyticity Regained? Gilbert Harman Noûs, Vol. 30, No. 3. (Sep., 1996), pp. 392-400. Stable URL: http://links.jstor.org/sici?sici=0029-4624%28199609%2930%3A3%3C392%3AAR%3E2.0.CO%3B2-E

Indispensability and Practice Penelope Maddy The Journal of Philosophy, Vol. 89, No. 6. (Jun., 1992), pp. 275-289. Stable URL: http://links.jstor.org/sici?sici=0022-362X%28199206%2989%3A6%3C275%3AIAP%3E2.0.CO%3B2-Z

Mathematics and Indispensability Elliott Sober The Philosophical Review, Vol. 102, No. 1. (Jan., 1993), pp. 35-57. Stable URL: http://links.jstor.org/sici?sici=0031-8108%28199301%29102%3A1%3C35%3AMAI%3E2.0.CO%3B2-B

Testability Elliott Sober Proceedings and Addresses of the American Philosophical Association, Vol. 73, No. 2. (Nov., 1999), pp. 47-76. Stable URL: http://links.jstor.org/sici?sici=0065-972X%28199911%2973%3A2%3C47%3AT%3E2.0.CO%3B2-N