1 For more detailed discussion of the general conception of epistemic justification, see my book The Structure of Empirical Knowledge (Cambridge, MA: Harvard University Press 1985) [hereafter cited as SEK), ch. 1. Certain recent philosophers have questioned, or seemed to question, this requirement for knowledge, arguing instead that knowledge requires only that the process leading to the acceptance of the belief in question be reliable, i.e., that it in fact produce or tend to produce true beliefs, even though the person in question may have no reason of any sort for thinking that this is so. See, e.g., Goldman, AlvinEpistemology and Cognition (Cambridge, MA: Harvard University Press 1985)Google Scholar; and Nozick, RobertPhilosophical Explanations (Cambridge, MA: Harvard University Press 1981)Google Scholar, ch. 3. I have dealt with such views elsewhere (see SEK, ch. 3) and will mostly neglect them in this essay, where my aim is to consider one crucial element of a more traditional epistemological position.

2 If, as I believe, there are in fact no such basic or foundational beliefs, the problem becomes even more acute, with all knowledge depending in part at least on a priori justification. See SEK, Part I, especially ch. 4, for a critique of foundationalism; and SEK, Part II, for what I regard as the main alternative.

3 For a particular person to be justified in accepting the conclusion of such an argument on the basis of a prior acceptance of its premises, the reason in question must, I assume, be in some way available to him.

4 It is also anticipated in Locke’s concept of ‘trifling propositions’ and in Leibniz’s appeal to identical propositions to explain knowledge of necessity.

5 All references to the Critique of Pure Reason will be to the translation by Norman Kemp Smith (London: MacMillan 1929) and will use the standard pagination.

6 A possible response would be that P is meaningful only within the confines of experience, so that the specified limitation is in effect already built in and so does not alter the identity of the proposition. The argument for such a view from within Kant’s philosophical position is stronger for some propositions than for others, though it is not clear that he accepts it for any. I doubt, however, whether any such contention can be made compelling on general philosophical grounds. (And if it were, the result, for reasons discussed below, would be that P, as long as its a priori status was capable of being accounted for by Kant’s philosophy, was not synthetic after all.)

7 This is perhaps expressed most clearly in the B preface, but is also lurking just under the surface of the argument in many other places.

8 For a clear and relatively candid expression of this outlook, see Wesley, SalmonThe Foundations of Scientific Inference (Pittsburgh: University of Pittsburgh Press 1967), 39-40.Google Scholar

9 As this suggests, ‘analytic’ and ‘synthetic’ are normally construed as mutually exclusive and jointly exhaustive terms, and it is only on such a construal that the issue between rationalists and empiricists can be correctly couched in terms of whether there is synthetic a priori knowledge. Some empiricists have clouded the issue by defining ‘synthetic’ independently. For example, in Language, Truth and Logic (New York: Dover 1946), Ayer offers the following definitions:

… a proposition is analytic when its validity depends solely on the definitions of the symbols it contains and synthetic when its validity is determined by the facts of experience. (78] Since the meaning of ‘synthetic’ seems on this definition to be simply equated with that of ‘a posteriori’ (’validity’ presumably means the proposition’s being justified), there will obviously be no synthetic a priori knowledge. Equally obviously, however, this in no way rules out the possibility of non-analytic a priori knowledge, which would be quite enough to vindicate the rationalist claim.

10 For this reason, it would be clearer in some ways to follow Butchvarov and regard logical truths as themselves synthetic, reserving the label ‘analytic’ for those propositions which are reducible to logic and whose justification is thereby (partially) explained. (See Butchvarov, PanayotThe Concept of Knowledge (Evanston, IL: Northwestern University Press 1970), 106-8.)Google Scholar

11 I ignore here the fact that some of these logical ingredients would no doubt take the form of principles or rules of inference rather than theses or assertions. A principle of inference is, from an epistemological standpoint, just as much in need of epistemic justification as is an assertion, albeit in a somewhat modified sense: if one is to be epistemically justified in accepting conclusions on the basis of such a principle, one needs to have some reason for thinking that conclusions satisfying the principle will be true (or perhaps, for some kinds of principles, are likely to be true) if the corresponding premises satisfying the principle are true.

12 Thus Quinton, in a defense of the moderate empiricist thesis, characterizes the position he is defending as the thesis that a non-derivative a priori truth is one whose ‘acceptance as true is a condition of understanding the terms it contains.’ See Quinton, Anthony‘The A Priori and the Analytic,’ reprinted inSleigh, Robert ed., Necessary Truth (Englewood Cliffs, NJ: Prentice-Hall l972), 90.Google Scholar

13 Cf., e.g., Quinton, 101-6.

14 Butchvarov, 109-10

15 For discussion of some of these, see Appendix A of SEK.

16 It is also obvious that moderate empiricism, even if otherwise acceptable, could offer no adequate solution to the problem, discussed above in section I, of how to justify inferences that go beyond direct observation. The only way to construe such inferences as analytic would be by adopting reductive theses like phenomenalism and behaviorism. Such theses are widely regarded as untenable and would in any case only extend to part of the problem (in particular, there is no apparent reductive solution to the problem of induction). While this constitutes a further objection to moderate empiricism as an account of the a priori, the objections discussed in the text are obviously more fundamental.

17 Reprinted in Quine, From a Logical Point of View, 2nd ed. (Harvard: Harvard University Press 1961). Page references to this paper will use the abbreviation TD.

18 And of course, even more obviously, the thesis in question would not mean in any case that the alleged unintelligibility of one of these concepts thereby extends to the other. For this to follow, Quine needs to have assumed in effect that the two terms in question are synonymous, a dubious assumption indeed, as we have seen - even if it were not made by one who, as we shall see, also rejects the intelligibility of the very concept of synonymy.

19 See Quine, Philosophy of Logic (Englewood Cliffs, NJ: Prentice-Hall1970), 2.Google Scholar

20 The issue throughout obviously concerns primarily a priori justification which is direct or intuitive, not that which relies on demonstration.

21 ‘In Defense of a Dogma,’ in Sleigh, Robert ed., Necessary Truth (Englewood Cliffs, NJ: Prentice-Hall1972), 86-7Google Scholar

22 It is interesting to note that Quine himself seems to recognize in one place the possibility of change of meaning, though he avoids using that specific phrase:

By a less extraordinary coincidence, … an eternal sentence that was true could become false because of some semantic change occurring in the continuing evolution of our own language. Here again we must view the discrepancy as a difference between two languages: English as of one date and English as of another. The string of sounds or characters in question is, and remains, an eternal sentence of earlier English, and a true one; it just happens to do duty as a falsehood in another language, later English. [Philosophy of Logic, 14]

But then the shift to later English does not involve ‘giving up’ the original sentence in any epistemologically interesting sense, and it is unclear why precisely the same thing cannot be said in the more specific case of an allegedly a priori claim. See also the discussion of deviant logics (ibid., ch. 6, esp. 74).

23 Orenstein, AlexWillard Van Orman Quine (Boston: Twayne 1977), 85-6Google Scholar

24 The indeterminacy thesis was first developed in Quine, Word and Object (Cambridge, MA: MIT Press 1960), ch. 2. (This book will hereafter be cited as WO.) It has since been elaborated and refined in many other places. Though a full account of the thesis would involve many further details and ramifications, these are inessential for present purposes.

25 As already noted, Quine himself would not use the term ‘meaning’; but I can see no reason to deprive ourselves of this useful formulation.

26 For an elaboration of this response, see my ‘Analytic Philosophy and the Nature of Thought,’ forthcoming.

27 I ignore here the problems posed by the fact that one particular belief can seemingly belong to many, indeed indefinitely many, different systems of belief.

28 For my attempt to sketch such an argument for a notion of coherence which includes the idea of explanatory adequacy as a major component, see SEK, ch. 8.

29 It is important to be clear that the issue here is not merely a verbal issue concerning the proper understanding of the term ‘justification’ as it occurs in epistemic contexts. What has been shown is that a Quinean epistemological view can offer no (non-question-begging) reason for thinking that the beliefs it sanctions are even likely to be true. And that is enough to make such a position a very strong version of skepticism, one which is unable to vindicate the ordinary claims of science and common sense- no matter what use is made of the term ’justification.’

30 It is less than clear just how such epistemic standards are to be represented in a Quinean framework. It is more natural to take them as principles or rules rather than as beliefs, i.e., in linguistic terms, roughly as imperative sentences rather than declarative ones. Quine has little to say about this issue, however, and I shall not worry about it here. It is clear at least that such standards, however they may best be represented, cannot for a Quinean be construed as immune to the possibility of revision.

31 As Harman points out, ‘Quine’s theory of evidence may also be thought of as a coherence theory of evidence: a person attempts to make his total conceptual scheme as coherent as possible …, ’ where the various standards mentioned above constitute the components of the idea of coherence (’Quine on Meaning and Existence II,’ Review of Metaphysics 21 (1967-68), 351). In these terms, the point being made in the text is the familiar one that the epistemic authority of coherence cannot itself be established by appeal to coherence. For more discussion, see SEK, 108-10.

32 Ironically enough, the point here is similar to an argument which Quine, and Harman following him, offers against the moderate empiricist appeal to convention: just as the logical force of the conventionalist’s conventions cannot derive from the conventions themselves, so also the logical force of the sentences of the web cannot be given merely by other sentences in the web. Hence, if those sentences are all there is, they lose all logical force and the need for revision collapses (except, possibly, for sentences which are strictly observational, if there are any of these).

33 See SEK, ch. 3, and the references offered there.

34 Versions of this paper were presented at Western Washington University and at the University of Nebraska. I am grateful to the audiences there for many useful comments and suggestions. Ann Baker was a constant source of advice and encouragement.