23
ANALYTICITY AND APRIORITY: THE QUINE-PUTNAM DISPUTE PAUL YU Quine's attack on analyticity and apriority, first begun in 'Two Dogmas of Empiricism' and then repeated in a steady stream of books and papers, has become part of the lore of contemporary philosophy. What has been equally fascinating is Putnam's reaction, over a number of years, to the attack. Since a full-fledged study of Quine's writings and of Putnam's reaction to them would have to be book size, I shall be concerned principally with Quine's 'Two Dogmas' and two recent papers by Putnam, ' "Two Dogmas" Revisited' and 'Analyticity and Apriority: Beyond Wittgenstein and Quine'. 1 Putnam disagrees with Quine on two important questions, namely: (1) is the notion of analyticity - better, is any notion of analyti- city - tenable? (2) Are there a priori truths? They appear to be in agreement, however, on the following question: (3) Can analyticity ground apriority? That is, can the notion of analyticity explain adequately the existence of a priori truths? Given their disagreements about (1) and (2), clearly Quine and Putnam have different reasons for answering (3) in the negative. Since Quine thinks that no notion of analyticity is tenable, analyticity can explain nothing. In particular, it cannot explain apriority. (Actually Quine goes further and argues that there is no such thing as apriority to be explained.) Putnam, however, thinks that there are defensible notions of analyticity. It is just that such notions cannot ground apriority. (Unlike Quine, however, Putnam believes that there is such a thing as an apriori truth.) What I want to do in this paper is to explain in some detail the relationship between Quine's views and Putnam's views and then to adjudicate their dispute, since the resolution of dispute over such basic issues by philosophers of such stature cannot fail to have signi. ficant implications for much of philosophy. Before I do that, how- 41

Analyticity and apriority: The Quine-Putnam dispute

  • Upload
    paul-yu

  • View
    214

  • Download
    0

Embed Size (px)

Citation preview

ANALYTICITY AND APRIORITY: THE QUINE-PUTNAM DISPUTE

PAUL YU

Quine's attack on analyticity and apriority, first begun in 'Two Dogmas of Empiricism' and then repeated in a steady stream of books and papers, has become part of the lore of contemporary philosophy. What has been equally fascinating is Putnam's reaction, over a number of years, to the attack. Since a full-fledged study of Quine's writings and of Putnam's reaction to them would have to be book size, I shall be concerned principally with Quine's 'Two Dogmas' and two recent papers by Putnam, ' "Two Dogmas" Revisited' and 'Analyticity and Apriority: Beyond Wittgenstein and Quine'. 1 Putnam disagrees with Quine on two important questions, namely:

(1) is the notion of analyticity - better, is any notion of analyti- city - tenable?

(2) Are there a priori truths? They appear to be in agreement, however, on the following

question: (3) Can analyticity ground apriority? That is, can the notion of

analyticity explain adequately the existence of a priori truths? Given their disagreements about (1) and (2), clearly Quine and

Putnam have different reasons for answering (3) in the negative. Since Quine thinks that no notion of analyticity is tenable, analyticity can explain nothing. In particular, it cannot explain apriority. (Actually Quine goes further and argues that there is no such thing as apriority to be explained.) Putnam, however, thinks that there are defensible notions of analyticity. It is just that such notions cannot ground apriority. (Unlike Quine, however, Putnam believes that there is such a thing as an apriori truth.)

What I want to do in this paper is to explain in some detail the relationship between Quine's views and Putnam's views and then to adjudicate their dispute, since the resolution of dispute over such basic issues by philosophers of such stature cannot fail to have signi. ficant implications for much of philosophy. Before I do that, how-

41

PAUL YU

ever, I need to do a bit of stage-setting by making some background remarks in the first section about the entire tangled issue of analyticity and apriority.

The issue of a priori knowledge has certainly been a central one in philosophy. Before I discuss it, however, I propose to refine the notion a bit to avoid certain problems. To begin with, the traditional definition of 'apriori knowledge' is roughly this:

p is knowable a priori if and only if it is knowable independ- ently of all experience.

I propose to substitute 'unrevisable' in place of 'knowable a priori', where 'unrevisable' is defined as follows:

p is unrevisable if and only if it is rational to accept and it would never subsequently be rational to reject no matter how the world turns out (epistemically) to be.2.

The reasons for this substitution are twofold. First of all, what is most important to the traditional notion of the a priori is not the genetic consideration of whether some truth is discoverable come what may or independently of all experience (whatever that might mean), 3 but rather the justificatory consideration of whether there are beliefs which are not disconfirmable and which we may rationally hold come what may. Secondly, the traditional notion of the apriori, on this understanding, is an epistemological one. It is awkward, therefore, to talk about a priori knowledge, since the notion of truth is part of the notion of knowledge, and the notion of truth is generally viewed as metaphysical. To be sure, traditional theorists most likely believed that human cognitive ability is such that what is unrevisable is in fact true. But that is a detachable assumption of some significance, and detaching it as an explicit assumption can only produce clarity. The notion of unrevisability, then, is, as it were, the purely epistemological element of the traditional notion of the a priori. If my understanding is correct, unrevisability is the same thing as what is designated by the currently fashionable term, 'epistemic necessity'. 4

I turn now to the traditional rationalism-empiricism dispute con- cerning unrevisable beliefs. While it may be acknowledged that as general labels 'rationalism' and 'empiricism' are apt to be more mis- leading than enlightening, they are still not without point when the

42

ANALYTICITY AND APRIORITY

opposition is formulated clearly within the confimes of a single problem. As I am using these terms, both rationalists and empiricists are realists, where a realist (with respect to a given theory or discourse) holds that '(1) the sentences of that theory or discourse are true or false; and (2) that what makes them true or false is something ex- ternal - that is to say, it is not (in general) our sense data, actual or potential, or the structure of our minds, or our language, etc. 's Where rationalists and empiricists disagree is with respect to epistemic justification. The empiricist view may be characterized very broadly as the view that observation must be the last arbiter of what is accepted as true. A rationalist, on the other hand, holds that there is a special class of beliefs, logical and mathematical beliefs being the traditional paradigms, which have a different sort of validation, s Thus a rationalist would say that, whereas ordinary beliefs are indeed established by means of observation, logical and mathematical beliefs (and perhaps some others) are established by what might be called intuit ion, construed as an irreducible human cognitive facutly. Inuitition, he suggests, is no more mysterious than our ordinary cognitive abilities.* And beliefs which are certified by intuition are unrevisable and hence true. 7. Here it is worth noting for later reference that, although traditional rationalists and empiricists differ in their conceptions of human cognitive abilities and hence in their assessment of how revisability and unrevisability are connected to truth and falsity, they agree that the notion of unrevisability is im- portant only in connection with the notion of truth. That is, un- revisability is significant only because evidence is a mark (in some suitable sense) for truth, and it is of course with truth that we are ultimately concerned. I shall return to this point in section IV below.

Now it is fairly clear that rationalism is not an intrinsically attractive view. Let me rehearse just briefly some difficulties which have been raised against the rationalist position. First of all, it is suggested that the rationalist account does not do justice to the difficulties involved in "a priori" discovery. Thus'proving a difficult theorem in mathematics is certainly unlike recognizing that 2 plus 2 equals 4. Secondly, conflicting intuitions are clearly always possible, and when there are conflicting intuitions it is not clear that there can be a non-arbitrary method for deciding the issue on the rationalist account. Thirdly, foundational studies in mathematics reveal para- doxes which challenge precisely our most deep-rooted intuitions. Finally, the rationalist view fails to provide us with a uni f ied account of knowledge since our senses are fundamentally divorced from our alleged intuition. And this account is further anomalous since, whereas

43

PAUL YU

we have complex and ramified theories about sense-mechanisms, intuition is primitive and unanalyzed, s To be sure rationalists can offer responses which mitigate to some extent the damage done by these objections - I shall not discuss them here. But it is fair to say that what attraction rationalism has resides principally in a lack in empiricism. That is, logical and mathematical beliefs (and perhaps some others) are decidedly odd beliefs which do not appear to be responsive to observation and which do appear to be unrevisable. The rationalist account, then, is best viewed as a challenge to em- piricism: how can a satisfactory account be found for logical, mathe- matical, etc. beliefs on the basis of empiricist principles? Empiricist reactions to this challenge can be divided, for analytic purposes, into two groups: (1) radical empireists, who reject the appearance of unrevisability of logical and mathematical beliefs and argue that there is no such thing as an unrevisable belief, and (2) moderate empiricists, who accept the existence of unrevisable beliefs but who then try to account for them in empiricist terms. I shall now con- sider these approaches in reverse order.

II

The principal moderate empiricist response to the rationalist challenge may be called the theory ofanalyticity. 9 According to this view, there are unrevisable beliefs all right, but all o f them are analytic. Actually there exist a number of different notions of analyticity. For reasons which will become obvious in section III, I shall confine my discussion in this section to what may be called the broad notion of analyticity, which is as follows:

a sentence is analytic if and only if it is true solely in virtue of the meanings of its constituent terms.

There is, curiously, no generally agreed upon account as to precisely how analyticity is to account for unrevisability. As I reconstruct it, the explanation is roughly this:

i. Matters o f meaning are entirely distinct from matters of fact.

ii. Matters of meaning are of course known by native speakers. That is, they have, qua speakers, semantic beliefs.

iii. Since matters of meaning are entirely distinct from matters of fact, semantic beliefs are unrevisable.

iv. Given the non-factual character of matters of meaning, what is unrevisable is in fact true.

44

ANALYTICITY AND APRIORITY

So meaning is invoked to explain truth, by way of semantic beliefs and unrevisability.

Two different sorts of questions have been raised against the theory of analyticity. First of all, it may be asked whether such notions of meaning and analyticity are sufficiently problem free to serve as explanatory principles. Secondarily, assuming the answer to the first question to be in the affirmative, it may be asked whether analyticity will indeed ground unrevisability satisfactorily, whether the explanation above is indeed adequate. I shall examine the first question, which is of considerable independent interest, in the next section. I shall examine the second question in section IV.

I mentioned earlier that there exist a number of different notions of analyticity. I have already discussed what I called the broad notion. Here is another, which may be referred to as the narrow notion of analyticity:

a sentence is analytic if and only if (i) it is a logical truth, or (ii) it may be turned into a logical truth by putting synonyms for synonyms.

Both the broad and narrow notions, which are undoubtedly the most widely known and discussed notions of analyticity, are meta- physical. There are also epistemological notions of analyticity. The following is one example:

a sentence is analytic if and only if it is knowable as true, if it is indeed true, on the sole basis of knowledge of its consti- tuent terms. 1~

All the notions of analyticity whith which I am familiar, including the three I have discussed, invoke the notion of meaning either directly or indirectly. As is well known, in 'Two Dol~nas' and in sub- sequent writings Quine has argued systematically against the notion of meaning and indeed against related semantic notions such as synonymy, semantical rule, and meaning postulate. One fundamental question that must first be answered, therefore, is whether the notion of meaning is tenable. That depends on, of course, what is taken as meaning. I turn now to a general discussion of meaning.

III

As a result of what might be called the psychological turn of theorizing about language, it is now widely agreed that the central question in the study of language is 'what is it for a speaker to have or know a language?' (rather than, say, 'what is a language?') There

45

PAUL YU

is, indeed, even a rough consensus on the correct answer to this question, viz.

A speaker S has or knows a language L iff S knows how to effect the following mappings:

(1) (i) acoustic sounds - phoneme sequences: (ii) phoneme sequences - syntactic structure; (iii) syntactic structure -- "meaning".

This rough consensus, however, is really rough, and in two quite different ways. Not only is the so-called consensus answer rough because so many details are missing, but the consensus is itself rough, since as it is stated (1) conceals at least two important contro- versies:

a. Some theorists hold that by "meaning" we should refer to semantic representations, whereas other theorists insist that by "meaning" we should refer to truth conditions.

b. Some theorists construe the ability to effect the mappingsspeci- fled by (1) as involving a body of knowledge, whereas others con- strue it as merely a skill (which involves no sort of knowledge). I propose to consider these two controversies by way of sketching a general account of meaning.

For ease of exposition, I shall consider the first controversy in the context of a theory, Davidson's, which sees the ability characterized by (1) (what we shall call, neutrally, linguistic competence) as in- volving a body of knowledge. Very roughly, according to Davidson, when a speaker has or knows a language L what he knows are sen- tences o f the form

(2) s is true L i f fp

where ' s ' is the standard name of some sentence o f L and where 'p ' is to be replaced by a suitable sentence o f the meta-language. 11 On the other hand, according to theorists who see "meaning" as semantic representations, when a speaker has or knows a language L what he knows are sentences of the form

(3) s in L translates as s* in L*

The difference between the truth-conditional view of Davidson and the translational view of the rival theoriests is obvious. If a speaker knows an instance of (2), then he can come to acquire a true belief about some object or substance in the world. On the other hand, S's knowledge of an instance of (3) may be quite useless in this respect. Briefly, a translational view of linguistic competence, one which sees speakers as only being able to relate structure o f one language

46

ANALYTICITY AND APRIORITY

with structures of another, cannot account for an important feature of our linguistic ability, viz. that we can come to acquire correct beliefs about the world by hearing sentences uttered by other speakers, t2 .

So far the controversy between the truth-conditional view and the translational view of linguistic competence has been couched in terms of theories which sees linguistic competence as involving a body of knowledge. Can the same point be made if linguistic com- petence is seen instead as a skill? The anser is clearly 'yes'. The im- portant point is that a speaker can come to acquire correct beliefs about the world on the basis of heard sentences. Whether this connection between the linguistic and the extra-linguistic is establish- ed via a body of knowledge or via a (mere) skill does not affect the point that linguistic competence must be truth-conditionally con- strued.

Some additional comments are called for before I turn to the second controversy. First a point about terminology. Call any account which attempts to specify the truth-conditions of the sentences of a language L a semantics of L, and call any account of what a speaker knows (or what he can do qua speaker) when he knows a language a theory of semantic competence for L. 13 Then I suggest that "meaning" is best construed as what is understood (or what can be understood) by a speaker by uttered sentences of his language.14 So ''meaning" is a central term in a theory 6f semantic competence, and is not (or is only for independent reasons) a central term in a semantics. Briefly, meaning is just what is under- stood by a speaker who is semantically competent. Now since the notion of meaning, or so I claim, has no firm footing in what may be called pre-theoretic or uncritical semantics, meaning is simply what a theory of meaning demands it to be. In a way, none of these terminological proposals is overly important. What is important is whether the overall theory which results gives the best possible account of what can plausibly be viewed as "meaning" phenomena. The second preliminary point concerns Tarski's truth def'mition. In an important paper, published in 1972, I-hrtry Field suggests that Traski's definition of truth may have been motivated by a desire to bring semantics in line with the demands of physicalism. Is Since in Tarski's def'mition truth is defined in terms of satisfaction, satis- faction is des in terms of denofation and application, and de- notation and application are inductively characterized, it looks as if Tarski's definition does not appeal to any semantic notions. This appearance, however, according to Field, is deceiving. The only

47

PAUL YU

reason Tarski apparently does not have to appeal to semantic notions is that he relies on enumeration when the notions of denotation and application are defined. Field concludes that a really adequate de- finition of truth would have to include not only the Tarski-style referential analysis of truth but also a theory of reference (i.e. a theory of denotation and application), t6

I am now ready to consider the second controversy. According to Davidson, who thinks that semantic competence involves a body of knowledge

(4) semantics of L =Tarskian truth definition

and

(5) S is semantically competent in L iff S knows the semantics of L.

For Davidson, then, the relationship between a semantics for L and a theory of semantic competence of L is quite intimate. Davidson's view, however, is objectionable in a number of ways. It may fairly be asked whether it is plausible to require all speakers (including many children) to have the semantic concepts and semantic knowledge required by Davidson's theory. Further, DavidsOn's theory requires speakers to know that 'Socrates' refers to Scorates. But if the causal theory of reference is even approximately correct, to know that 'Socrates' refers to Scorates one must know (minimally) that a causal chain exists between Socrates and the current use of 'Socrates'. Surely that is to require too much.

Michael Devitt has recently advanced a view of semantics and semantic competence according to which 17

(6) semantics of L = referential analysis of truth + theory of reference

and

(7) S is semantically competent iff S is able to perceive certain sounds as sentences-under-analysis and S is able to refer to objects and sets of objects by using names and general terms.

Devitt's view of semantic competence is a skill view, in the following way. On Davidson's account, it will be recalled, for S to be seman- tically competent with 'Socrates' if for S to know that 'Socrates' refers to Socrates. On Devitt's account, on the other hand, for S to be semantically competent with 'Socrates' is for S to be able to designate or refer to Socrates with 'Socrates'. And the latter, accord-

48

ANALYTICITY AND APRIORITY

ing to Devitt, is a matter of causal connection, not of knowledge. These are two respects in which Devitt's suggesting is unsatis-

factory. The first problem is that it leaves no room for anything like psychological representation. Without such a notion, however, an account of semantic competence would be quite incomplete. A number of considerations can be cited for this conclusion:

a. A speaker clearly conceptualizes objects. He thinks of Socrates, for example, as the Greek philosopher who was compelled to drink hemlock.

b. This notion of psychological representation or conceptualiza- tion is important when we try to give an account of linguistic com- munication. Thus A says to B

(8) Socrates is wise.

The cognitive content of A's claim is that Socrates is wise. This is what he has asserted whether he has the right concept of Scorates or not. On the other hand, the communicative content (in one sense of that term) is that the Greek philosopher who drank hemlock was wise. Without psychological representation such a distinction cannot be made. Is

c. The notion of psychological representation, as Putnam has em- phasized, is important in the context of language acquisition. Real connections between symbols and objects cannot account for language acquisition. 19

d. Finally, as linguists have emphasized, speakers have intuitions of ambiguity, synonymy, superordination, etc. which must also be accounted for by a theory of semantic competence. Some at least of these phenomena, so I claim, would need to be accounted for in terms of psychological representation. I suggest, then, the following modification of Devitt's account:

S is semantically competent iff S is able to perceive certain sounds as sentences-under-analysis

and

(9) S is able to refer to objects and sets of objects by using names and general terms and S assigns appropriate "con- ceptual roles" to terms

Without pretending to have produced a detailed account, I do con- tend that such an account as (9) is necessary if every aspect of one's linguistic ability is to be adequately explained.

Devitt is quite right in rejecting Davidson's theory, which requires us to have implausibly large number of semantic concepts and seman-

49

PAUL YU

tic beliefs/knowledge. Speakers need to have generative abilities and referential abilities, but that does not mean that they must know the semantics of their language. So far so good. But, and this brings me to my second criticism of Devitt, we still need to explain how speakers are able to do what they do (qua speakers). And it is not at all clear to me how we can produce such an explanation without attributing to the speakers all kinds of beliefs/knowledge. Such beliefs/knowledge are of course not semantic in the sense of containing semantic concepts (such as truth and reference) - Davidson's mistake is precisely to require semantic beliefs/knowledge in this sense for semantic competence. But they are semantic in the sense that they are what explain our semantic competence. The point is this: there need be no conflict between a knowledge view and a skill view of se- mantic competence. Indeed, it is clear that a theorist who insists that knowing a language is a skill which does not involve the possession of a body of knowledge is not putting forward a positive theory. Rather, he is simply denying one sort of explanation of that skill (viz. in terms of beliefs/knowledge) and leaving the explanation of that skill as an open issue. What I am suggesting is that, although some knowledge views (e.g. Davidson's) are problematic, I do not see how we are to give a satisfactory account of linguistic competence without attri- buting significant semantic beliefs/knowledge to speakers.

So much for my general account of meaning and semantic com- petence. Now it is clear that, in order for my conception of meaning and semantic competence to be defensible, I need to be able to draw a distinction between (1) what a speaker knows or understands just in virtue of mastering his language and (2) what a speaker knows or understands in virtue of mastering his language and knowing collateral information. That is, I need to be able to draw a distinction between semantic knowledge (or belief) and factual knowledge (or belie 0. for example, cannot even be formulated without such a distinction. This is precisely the distinction that Quine challenges. In a crucial, but little discussed, passage, Quine writes:

Even for such favored occasion sentences as 'Gavagai' and 'Rabbit ' , actually, sameness of stimulus meaning has its short- comings as a synonymy relation. The difficulty is that an in- formant's assent to our dissent from 'Gavagai?' can depend excessively on prior collateral information as a supplement to the present prompting stimulus . . . . . There may be a local rabbit-fly unknown to the linguist, and recognizable some way off by its long wings and erratic movements; and seeing such a fly in the neighborhood of an ill-glimpsed animal could help a

50

ANALYTICITY AND APRIORITY

native to recognize the latter as a rabbit . . . . And, to be less fanciful, there are all those stimulations that incorporate verbal hints from native kibitzers . . . . Intuitively the ideal would be to accord to the affirmative meaning of 'Gavagai' just those stimulations that would prompt assent to 'Gavagai?' on the strength purely of an understanding of 'Gavagai', un- aided by collateral information: unaided by recent observation of rabbits near the spot, unaided by knowledge of the nature and habits of the rabbit-fly, unaided by conversance with the kibitzer's language . . . . B u t . . . the trouble i s . . . that we hay# made no general experimental sense of a distinction between what goes into a native's learning to apply an expression and what goes into his learning supplementary matters about the objects concerned. True, the linguist can press such a distinction part way; he can filter out such idiosyncratic bits of collateral matter as the informant's recent observation of rabbits near the s p o t . . . But any socially shared information, such as that about the rabbit-fly or the ability to understand a bystander's remark, will continue to affect even that common denominator. There is no evident criterion whereby to strip such effects away and leave just the meaning of 'Gavagai' properly so-called - whatever meaning properly so-called may be.2~

To issue this challenge, of course, is not tantamount to showing that no difference exists between semantic beliefs and factual beliefs. But it does call for an answer. And it is to Putnam's credit that lie is one of the few philosophers who recognized the nature of the challenge and who actually proposed a general answer. Thus he wrote in 'Is Semantics Possibl.e?':

The hypothesis is that there are, in connection with almost any word . . . . certain core facts such that (1)one cannot con- vey the normal use of the word (to the satisfaction of native speakers) without conveying those core facts, and (2) in the case of many words and may speakers, conveying those core facts is sufficient to convey at least an approximation to the normal u s e . . . If this hypothesis is false, then I think Quine's pessimism is probably justified. But if this hypothesis is right, then I think it is clear what the problem of the theory of mean- ing is, regardless of whether or not one chooses to call it "theory of meaning"; the question is to explore and explain this empirical phenomenon. (pp. 59 -60) 21

If I understand Putnam correctly, he is suggesting that we can main- rain that there is an important difference between semantic beliefs and factual beliefs even i f we agree that semantic beliefs constitute a subset o f our factual beliefs. This is a basic insight, and one that has

51

PAUL YU

not been sufficiently appreciated. It is overlooked, for example, by Harman. A theorist who believes that there is a distinction to be drawn between semantic beliefs and factual beliefs, according to Harman, can be thought of as accepting the view that 'an important distinction must be made between the entries in a person's internal dictionary and the entries in his internal encyclopedia'. 22 But con- sider the sentence 'this is silent paint'. Is this semantically deviant - a claim that is at variance with the internal dictionary, or is it rather factually weird - a claim that is at variance with the internal en- cyclopedia? There is, Harman suggests, clearly no saying. 'The idea that there is an important epistemological distinction between dictionaries and encyclopedias does not survive a good look at a dictionary . . . . The main difference between a dictionary and an encyclopedia is that dictionaries have more and shorter entries. 23 But this is certainly not a good argument against the distinction, as it is drawn according to Putnam's suggestion, between semantic beliefs and factual beliefs. There is no need for a defender of that distinction to hold that the distinction in question is an epistemo- logical one. Harman's argument is effective only against a theorist who accepts the premise, now rendered quite dubious by the pre- ceding discussion, that matters of meaning are entirely distinct from matters of fact. Putnam's suggestion, however, is that there is an important difference between semantic beliefs and non-semantic facutai beliefs even though they are both factual beliefs (and hence have the same epistemologicat status.).

Now I should not want to be taken as saying that Putnam's actual characterization of the content of a theory of meaning is correct as it,stands. Indeed, if my suggestion that meaning is just what is under. stood (or can be understood) by semantically competent speakers is approximately correct, then Putnam's proposal is clearly inadequate. Thus I cannot convey the normal use of certain four letter words without noting, that they are not used in polite company. But that is hardly a semantic fact about such words. Again, I can convey the normal use of 'cordon bleu' by noting that it is a term of high praise for a culinary preparation, but that is not to give the meaning of 'cordon bleu'. A satisfactory way of drawing the distinction between semantic beliefs and(non-semantic) factual beliefs must not depend on the problematic notion of the normal use of a word. ~ The crucial point, however, is that Putnam has given us good reaons not to be persuaded by Quine's scepticism about meaning.

52

ANALYTICITY AND APRIORITY

IV

If my discussion in the preceding section is at least roughly correct, then the notion of meaning is, pace Quine, tenable. From that of course it does not follow either that there are also tenable notions of analyticity or that tenable notions of analyticity, if such there be, can ground unrevisability. What I shall argue is in fact this. The narrow notion of analyticity and the epistemological notion of analyticity are relatively problem free, but they cannot ground un- revisability. The broad notion of analyticity, at least on the usual assumptions, can ground unrevisability. But those assumptions are quite dubious. I realize that there may be other notions of analyticity, but I shall be quite content if I can establish the above claims.

(1) Assuming for the purposes of this discussion that the notion of logical truth is unproblematic, the tenability of the notion of meaning (and hence of synonymy) will indeed legitimize the narrow noaon of analyticity. However, since logical truths are on this account analytic by definition, there is not explanation as to why they are analytic or what point there is in calling them 'analytic'. A fort~ori there is also no explanation of why logical truths should be unrevisable. The legitimacy of the narrow notion of analyticity, therefore, does nothing for the moderate empiricist approach. 2s

(2) Since the epistemologlcal notion of analyticity refers ex- plicitly to faUible semantic belief, clearly it cannot ground un- revisability.

(3) As I argued in the last section, meaning is what a theory of meaning demands it to be. On the theory of meaning I favor and have argued for elsewhere, meaning is the "content" of a speaker's under- standing. For such a conception of meaning and theory to be defens- ible, however, I have argued, we must accept Putnam's suggestion that semantic beliefs are simply a subset of factual beliefs. But that is to give up the premise, crucial both to the theory of broad analyticity and to Quine's attach against semantic notions, that matters of meaning are epistemologtcally distinct from matters of fact. Hence broad ana- lyticity as I have construed it in section lI is not a defensible notion at all. It cannot, then, be used to ground analyticity. ~ This is a conclu- sion that is agreed upon by both Quine and Putnam, though for differ- ent reasons. Since Quine rejects all semantic noti<)ns, analyticity can- not be for him an explanatory device at all; so it cannot explain un- revisability. Though Putnam allows that there are defensible notions of analyticity, he does not think that they can ground unrevisabil/ty. 2~

53

PAUL YU

To be sure Putnam nowhere presents the argument I sketched above against the theory of broad analyticity. But I trust that it is plausible to think that that argument grows naturally out of Putnam's defense of the notion of meaning.

On the present account, knowledge of meanings would simply be a subset of our knowledge and would be, contrary to traditional accounts - viz. those attacked by Quine, thoroughly empirical. Semantic beliefs are acquired, can evolve, and in no sense ground truth. In the epistemological sense of 'analytic', for example, an English speaker would know that 'all vixens are female' is true, if it is indeed true, solely on the basis of his semantic knowledge or beliefs. In the metaphysical, narrow sense of 'analytic', to reject 'bachelors are unmarried' we would indeed have to change the meaning of bachelor - change an one-criterion belief, ff Putnam is right. But there will be nothing unempirical about this change.

V

The principal moderate empiricist attempt to answer the rationalist challenge (viz. the theory of analyticity) - to provide an account of unrevisable beliefs consistent with empiricist principles - is, as I have argued, a failure. Other empiricists, Quine most prominently amount them, have long wanted to argue for a far more radical position, namely, that there is no such thing as an unrevisable belief at all. Let me now turn to an examination of Quine's radical empiricism.

As Putnam notes, Quine has produced a ~tistorieal argument' against the existence of unrevisable beliefs. ~ Quine argues in effect that revisions o f what used to be'thought o f as an unrevisable truths have in fact already occurred. In a passage cited in both Putnam 1976 and Putnam 1979, Quine writes:

Any. statement can be held true come what may if we make drastic enough adjustments elsewhere in the system. Even a statement very close to the periphery can be held true in the face of recalcitrant experience by pleading hallucination or by amending certain statements of the kind called logical laws. Conversely, by the same token, no statement is immune from revision. Revision even of the logical law of the excluded middle has been proposed as a means of simplifying quantum mechanics; and what difference is there between such a shift and a shift whereby Kepler superseded Ptolemy, or Einstein Newton, or Darwin Aristotle? ('Two Dogmas', p. 43).

54

ANALYTICITY AND APRIORITY

It is not entirely clear what Quine is saying in this passage, but by the 'historical argument' Putnam is presumably referring to some such argument:

i. Beliefs of the form'p is unrevisable' have turned out to be false in the past.

ii. Hence such beliefs may turn out to be false in the future.

It is important to see the nature of Quine's argument here. As I read him, Quine holds in general that evidence is a matter of center of gravity, so that if sufficient evidence to the contrary turns up then any belief, bar none, may be revised. 29 The ~istorical argument' is intended to show that sufficient contrary evidence may turn up for any belief, even unrevisable truths so-called. Indeed, it is clear that Quine assumes the truth of both realism and empiricism: reality is independent of us and accessible to us through observation; hence any belief we hold must face the tribunal of sensory evidence and, if the 'historical argument' is correct, may need to be revised. It is not clear what assumptions are being made by Quine about the nature of human cognition, a clearly important concern in any dis- cussion of revisability, since presumably the nature of our cognition determines how we perceive reality and determines our concept of rational belief. But it is not implausible to think that Quine views us as fallible but intelligent cognizers: fallible in that we are biological organisms which have no apodictic intuition, but intelligent enough that we are able to recognize relevant evidence and to evolve success- ful theories. 3~

Curiously, Putnam accepted this argument completely as recently as 1976 but now rejects it in the 1979 paper. Thus he writes in Putnam 1976:

I believe that Quine's methodological argument against apriori. city is a correct one. (p. 206)

In Putnam 1979, after characterizing essentially the same argument, he writes:

It is this argument from the history of science that I challenge. (p. 434)

It should be noted immediately that Putnam still agrees with Quine in part. Thus Putnam points out that, not only are beliefs which are (in fact) false revishble, but beliefs which are (in fact) true - indeed beliefs which are viewed as unrevisable truths - may be revised. Thus suppose Kripke produces a derivation which purports to show that the Peano Axioms lead to a contradiction. The proof is checked by

55

PAUL YU

both important logicians and by machine, and it holds up. Then it may be rational for me to believe that Peano Arithmetic is incon- sistent even though it (in fact) is not. 31 (Quine would find this dis- cussion agreeable, I think, though he never discusses the revision of beliefs which are (in fact) true). Still, Putnam now disagrees with Quine on a fundamental point: although he agreed with Qine com- pletely with regard to revisability in 1976, he now believes that there are in fact unrevisable beliefs. I am not quite certain what the precise explanation is for Putnam's change of heart. It cannot be because the kinds of examples discussed by Quine are regarded by Putnam as merely "physical" (and not mathematical or logical) since, as Putnam himself notes, there is the impressive case of geometry. Thus in Putnam 1976 he writes:

Unless one accepts the ridiculous claim that what seemed a priori was only the conditional statement that if Euclid's axioms, then Euclid's theorems, (I think that this is what is what Quine calls 'disinterpreting' geometry in 'Camap and Logical Truth'); then one must admit that the key propositions of Euclidean geometry [are?] interpreted propositions ('about form and void', as Quine says), and these interpreted propo- sitions were methodologically immune from revision (prior to the invention of rival theory) as Boolean logic was prior to the proposal of the quantum logical interpretation of quantum mechanics. The correct moral - the one Quine draws - is that some statements can only be overthrown by rival theory; but there is no such thing as an absolutely unrevisable statement. 32

Part of the reason for Putnam's aboutface, of course, must be that he thinks that he has a convincing reply to Quine's historical argu- ment'. Succinctly, Putnam claims that the conclusion of Quine's argu- ment does not follow since there is a simple counter-example. Accord- ing to Putnam, there is at least one unrevisable belief, namely:

(p) No~ every statement is both true and false.

How does Putnam argue for the unrevisability of p? He writes:

By our lights, to believe that all one's beliefs are both true and fa l se . . , is to give up both the notions of belief and t r u t h . . . In short, to believe that all statements are correct (which is what we are talking about)would be to have no notion of rationality. At least one statement is a priori, because to deny that state- ment would be to forfeit rationality itself. 3a

To say that p is revisable, then, according to Putnam, is to abandon the (current) concept of rationality. Since that is the only concept

56

ANALYTICITY AND APRIORITY

of rationality we have, we shall be left with no concept of rationality. But the concept of revisability can be explicated only with the help of the concept of rationality. Hence if we have no concept of rationality, we shall have no concept of revisability either. Hence the claim that p is revisable is incoherent. In brief, in Putnam's view, the existence of at least one unrevisable belief, viz. p, is assured given our (current) concept of rationality. Quine's llistorical argument' is no good, therefore, because not all unrevisable truths so-called are the same. However revisable other unrevisable truths so-called may be - Putnam agrees that there is force to Quine's 'historical argument - p is different since its unrevisability is 'certified by the theory of rationality itself') 4 Putnam notes that the fact that our (current) concept of rationality is provisional does not affect p's unrevisabflity. It is after all Quine's professed view that the best available theories should be the ultimate parameter for what is plausible and what is to be regarded as true. (In Quine's words, by who else's lights should we try to answer questions, if not by our own?)

I find Putnam's arguments unconvincing. To begin with, it is by no means clear what Putnam is referring to by 'the theory of rational- ity', and nowhere does he elaborate on his rather bald claim that to say that p is revisable is to abandon the theory of rationality (what- ever that is). More importantly, however, there is a simple way to amend Quine's 'historical argument' just a bit so that it is not met by Putnam's response, viz.:

i. Beliefs of the form 'p is unrevisable', in any reasonable conception of rationality have turned out to be false in the past.

ii. Hence such beliefs may turn out to be false in the future.

Vis-i-vis this argument, the provisional character of our concept of rationality is relevant, and Putnam has said nothing to show this argument to be untenable, finally, there is this, perhaps most im- portant point, a point which I have already broached earlier. Though rationalists and Quine have different ideas about the nature of human cognition and hence different ideas about.how revisability and un- revisability are connected to truth and falsity, for both the significance of unrevisabiLity derives ultimately from its connection to truth. As I put the matter earlier, for both Quine and his opponents unrevisability is significant only because evidence is a mark (in some suitable sense) for truth. Now Putnam claims that the unrevisabtltty of p is certi- fiable by our theory of rationality. Given Putnam's actual argument, it is clear that he has given us no reason to think that p is true (or

57

PAUL YU

even that p can sensibly be said to be either true or false). 3s But then unrevisability in Putnam's sense is of no interest whatever to Quine and Quine's traditional opponents.

It is worth noting that Quine is in fact quite conservative. Al- though in 'Two Dogmas' he cites quantum logic as a possible case of revision of logic, it is clear that on the whole he believes that we shall never have good enough reason to abandon "classical" logic and to adopt a deviant logic, a6 But there is still all the difference in the world between the following positions:

(1) logic will/n fact never get revised; and

(2) logic is in principle unrevisable.

The latter requires a guarantee, whereas the former clearly does not, about what the future will or will not hold. Quine's 'historical ar- gument', however, is a powerful argument against any such guarantee.

VI

Rationalism, I argued at the end of section I, is best viewed as a challenge to empiricism: how can a satisfactory empiricist account be found for such beliefs as logical and mathematical ones? The most straightforward radical empiricist view is of course the claim that unrevisable truths so-called are really the most general and the most well confirmed of our inductive generalizations. This view is generally attributed to Mill, though it is not clear how justified the attribution isl a7 Whatever its parentage, this view obviously faces grave diffi- culties. Unrevisable truths so-called are radically different from ordinary empirical generalizations both in point of confirmation and in point of disconfirmation. Thus they do not appear to be discon- firmable by contrary instances but are rather "protected". Indeed, it is not even clear what the instances, say of '2+2=4 ' are, numbers being such abstract entities. And, of course, we do not appear to collect any evidence in order to have a warranted belief that 2 plus 2 equals 4. How does Quine's account avoid these difficulties?

Quine proposes construing logical and mathematical beliefs (and perhaps some others) as very theoretical hypotheses. By so doing, Quine succeeds in explaining why such beliefs are apparently immune to (simple) refutation by contrary instances, since he points out that high level principles of physics, say, are equally not refutable in any simple way. (There is also some difficulty in saying what their instances are, though the difficulty is perhaps not as great as that of

58

ANALYTICITY AND APRIORITY

'2+2=4'.) The confirmation and disconfirmation of all such high level principles are tortuously indirect. Such beliefs are overthrown, if they are overthrown at all, only by a rival theory. This is indeed a point about Quine's account that Putnam explicitly recognizes and applauds. 3s Kripke, Putnam mentions, has argued in an unpublished paper that Quine's conception is incoherent. According to Kripke, the notion of testing a statement makes no sense unless something is fixed. Why should we accept the view that quantum mechanics requires us to change our logic, Kripke asks. If nothing is unrevisable, why do we not conclude instead that we should revise the statement that quantum mechanics requires us to change our logic? 39 Quine has suggested at least an outline of an answer to this objection. The answer is that, as a matter o f fact, certain beliefs are more central, more "protected" than others - w h a t Putnam has dubbed 'con- textually a priori', 4~ so that sense can be made of the notions of testing and revision. Perhaps the priorities ought to be rearranged; that is of course always in the offing. Now there is no denying that Quine has not provided anything like a detailed account for con- ceptual changes of a fundamental nature, but that is hardly a telling objection against such a generally attractive and unified empiricist view.

VII

Let me summarize briefly the major points I have made. (1) I have argued with Putnam against Quine that there are un-

problematic notions of meaning and analyticity, and I have tried to offer at least a general framework for a theory of meaning.

(2) I have tried to show that Quine and Putnam are in agree- ment, though for different reasons, that there is no notion ofanalyti- city that can ground unrevisability.

(3) Finally, I have defended Quine's thesis that there is no such thing as an unrevisable belief by reiterating his historical argument and by showing that Putnam's putative example of an unrevisable belief cannot escape the scope of Quine's argument.*

CENTRAL MICHIGAN UNIVERSITY MOUNT PLEASANT, MICHIGAN 48859

USA

* I would like to thank Stephen Schwartz and an anonymous referee of this journal for helpful comments on earlier versions of this paper. Research

59

PAUL YU

for this paper was partially supported by a grant from the Faculty Research and Creative Endeavors Committee (grant number 4-22594) at Central Michigan University and partially by a grant from the Fulbright program. I hereby grate- fully acknowledge their financial assistance.

NOTES

t The former is in Gilbert Ryie (ed.), Contemporary Aspects of Philosophy, Stocksfield, England: Oriel Press, 1976 and the latter is in Peter French, Theodore Uehllng Jr. and Howard Wettstein (eds.), Midwest Studies in Philosophy: Studies in Metaphysics, Minneapolis: University of Minnesota Press, 1979. For brevity of reference henceforth the former will be referred to as 'Putnam 1976' and the latter as 'Putnam, 1979'. In between these two Putnam also wrote 'There is At Least One A Priori Truth'. Erkenntnis, V. 13, 1978. But since that is a transitional paper which adds little to Putnam's discussion, I shall largely ignore it. A fuller study of the issues discussed in the present paper would have to include at least the following further works by Quine and Putnam. For Quine: 'Carnap and Logical Truth', Synthese, V. 12, 1960; World and Object, Cambridge: M.I.T. Press, 1960; 'Necessary Truth', Voice of America Forum Lectures, Philosophy of Science Series, No. 7, 1964; 'On a Suggestion by Kata', Journal o f Philo- sophy, V. 64, 1967; 'Epistemology Naturalized' in Ontological Relativity and Other Essays, New York and London: Columbia University Press 1969; Philosophy of Logic, Englewood: Prentice-Hall, 1969; 'RepLies' in Donald Davidson and Jaako Hintikka (eds.), Words and Objections, Dordrecht-Holland: D. Reidel Publishing Company, 1969; and The Roots of Reference, LaSalle Illinois: Open Court, 1973. For Putnam: "the Ana- lytic and Synthetic' in Herbert Feigl and Grover Maxwell (eds.), Minnesota Studies in the Philosophy of Science, V. 3, Minneapolis: University of Minnesota Press, 1962; 'It Ain't Necessarily So', Journal of Philosophy, V. 54, 1962; 'Is Logic Empirical?in Robert Cohen and Marx Wartofsky (eds.), Boston Studies in the Philosophy o f Science, V. 5, Dordrecht-HoUand: D. Reidel, 1968; 'Is Semantics Possible? in Howard Kiefer and Milton Munitz (eds.), Language, Belief and Metaphysics, New York: State Uni- versity of New York Press, 1970 *The Meaning of "Meaning"' in Keith Gunderson (ed.), Language, Mind and Knowledge, Minneapolis: University of Minnesota Press, 1969: ~'lae Refutation of Conventionalism' in Milton Munitz (ed.), Semantics and Meaning, New York: New York University Press, 1975; "Truth and Necessity in Mathematics' in Mathematics, Matter and Method: Philosophical Papers, V.1, London: Cambridge University Press, 1975; and 'What is Mathematical Truth?' ibid.

s This is essentially Putnam's definition of a 'a priori' in Putnam 1979,p.433. 3 I do not decry the genetic notion, but I do want to point out that there are

two (at least) distinguishable notions of the a prioro. For a development of

60

ANALYTICITY AND APRIORITY

the genetic-discovery notion of the a priori see Philip Kitcher, 'A Priori Knowledge', Philosophical Review, V. 84 No. 1, Jan. 1980.

4 See Putnam 1979, p. 430. 5 Putnam, 'What is Mathematical Truth?', pp. 69 -70 . Putnam attributes the

formulation to Michael Dummett. There are, I should hasten to add, other important and fruitful construals of 'rationalism' and 'empiricism'. Thus rationalism is sometimes understood as the doctrines that innate ideas are components of our knowledge, and empiricism as the doctrine that observa- tion is the only source of any concept or belief about the truth. I am not concerned with these other construals of 'rationalism' and 'empiricism' in this paper. For explicit statements of this sort of view, see Roderick Chisholm, Theory o/ Knowledge, second edition, p. 49, Englewood Cliffs, New Jersey: Prentice-Hall, 1977, and A. Ewing, "The Linguistic Theory of A Priori Pro- positions' in I-L D. Lewis (ed.), Clarity is Not Enough, London: George Allen & Unwin Ltd., 1963. Here indeed it is not clear what traditional rationalists are saying, whether (1) logical beliefs etc. are responsive to observation, though we possess a special sort of congnition which can encompass all observations, or whether (2) logical beliefs etc. are not responsive to observations at all. I suspect that both positions have been held.

' For further discussion see Carl Hempel, 'On the Nature of Mathematical Truth' in Herbert Feigl and Wlifrid Sellars (eds.), Readings in Philosophical Analysis, New York: Appleton-Century-Crofts, 1949.

0 The other major moderate proposal is, of course, conventionalism. For a survey of recent literature see chapter 5 of my doctoral dissertation,Knowl- edge and the A Priori (unpublished), Ann Arbor Mierof'tlms, 1973.

to See Richard Grandy, 'Classification: Kinds and Conventions' (unpublished), pp. 1-3.

tl Due to the limitation of space my exposition of Davidson will be unfair to him in several respects. For example, Davidson is quite careful to say that he is only purporting to state knowledge which would suffice for com- petence, not knowledge which is in fact implicit in competence. See Donald Davidson, "Truth and Meaning', Synthese, V. 17, 1967 and 'Reply to Foster' in Truth andMeantng, edited by Gareth Evans and John McDowall, Oxford University Press, 1976.

1= For an excellent recent discussion of the controversy between truth- conditional theorists and translational theorists see Ernest Lepore and Barry Loewer, ~Franslational Semantics', Synthese, V. 48, 198.

a= 'Semantic competence' is meant to be, abstractly speaking, a component of our total linguistic competence.

14 See my 'Grammar and Understanding', Canadian Journal of Philosophy V. 9 No. 2, 1979,p. 268.

Is See I-Iartry Field, ~l'arski's Theory of Truth ' , Journal of Philosophy, V. 69, 1972.

61

PAUL YU

16 Thus a referential analysis o f truth is simply an account of how truth con- ditions depend on referential conditio.ns. It is up to the theory o f reference to say whether those referential conditions are fulf'dled or not.

1~ See Michael Devitt, Designation, Columbia University Press, 1981, particu- laxly chapters 3 and 4.

la For a very illuminating discussion of such matters see Nathan Salmon, Review of Linsky's Names and Descriptions, Journal of Philosophy, V. 76, 1979.

i. See Putnam, 'Is Semantics Possible?' 2o World and Ob/eet, pp. 37-8. emphasis mine. sl For some other major attempts on Putnam's part to work out a general

theory of meaning see 'The Analytic and the Synthetic' where he develops an account of one-criterion words and `The Meaning of "Meaning"' where he develops an account of natural kind words.

22 Thought, p. 97.

23 ibid., pp, 98-9, emphasis mine. 24 Two proposals which immediately suggest themselves axe the following: (1)

semantic beliefs axe just those factual beliefs which axe shaxed community- wide; and (2) semantic beliefs axe those factual beliefs which axe first acquired whcn one learns to use expressions of one's language. The former agrees with Quine that there is no distinction between semantic beliefs and community-wide beliefs; the latter may not. It is beyond my present am- bitions, howcver, to consider such proposals in detail.

~s Cf. Putnam 1976, pp. 2 0 3 - 4 and Putnam 1979, p. 209. 2~ The standard attack on the broad notion of analyticity consists simply of

the query: what could the 'in virtue of ' in ' truth in virtue of meaning' come to? That is, how could the notion of meaning possibly explain truth? See, for example, Itarman, Tl~ought, pp. 1 2 - 5 . 1 hope my criticism of the theory of broad analyticity is more convincing than the standard attack.

27 See references cited in note 25 above. 28 Putnam's discussion show some curious lapses:

(1) In one place he suggests that analyticity is not a significant notion vis- a-vis unrcvisability, and this is clearly meant as a historical remark. As a historical remark, however, it is plainly false. Sec Putnam 1976, p. 209f. (2) Putnam also suggests that Quine confused analyticity and apriority in ~rwo Dogmas'. (ibid., p. 207). That this is false is shown by the organiza- tion of ~fwo Dogmas'. In the first hall" Quint tries to show that analyticity cannot ground apriority; in the second half he goes further and tries to show that there is no such thing as apriority at all. But these axe historical and exegetical issues with which I am not concerned.

a, Word and Object, p. 18 ~0 See 'Propositional Objects ' in Ontological Relativity, especially pp. 1 5 7 -

60 and ~rhe Nature of Natural Knowledge' in Samuel Guttenplan (ed.) Mind and Language, Oxford: Oxford University Press, 1975.

31 Putnam 1979, pp. 4 3 1 - 2 . I take i t that this is just reaffirmation of a point

62

ANALYTICITY AND APRIORITY

which has been made by both Putnam and Kripke, namely that a belief may be metaphysically necessary but epistemically contingent. See, for example, Putnam, ~rhe Meaning of "Meaning" ', p. 151. pp. 208-9 . See also Putnam, Mathematics, Matter and Method: Philo- sophical Papers, V. 1, Introduction, pp. viii-x.

ss Putnam 1979,pp. 4 3 4 - 5 . Putnam 1979, p. 436.

u Putnam is aware of this difficulty. Thus in Putnam 1979 he asks without answering the following question: ' . . . . . is the fact (if it is a fact) that a rational being could not believe the denial of "2+2=4" barring epistemo- logical catastrophe) an explanation of the truth of "2+2=4"? Or is it rather a fact about our rationality? (p. 431). At the end of the aper he offers the following, rather lame, observation: 'My own guessis that the truthsoflogic we are speaking of are so basic that the notion of explanation collapses when we try to "explain" why they are true. I do not mean that there is something "unexplainable" here; there is simply no room for an explana- tion of what is presupposed by every explanatory activity - and that goes for philosophical as well as scientific explanations . . . . ' (p. 441). Putnam does not want to accept the tempting but, he thinks, erroneous Wittgen- steinian view that some truths arise in us. But then he has obvious difficulty in explaining how p differs from other truths.

s6 This is particularly evident in his chapter on deviant logics in Philosophy o f Logic. I am inclined to believe, though this is certainly not the place to argue for it, that Quine's attitude is too conservative. To take just one instance, it may well be that an adequate solution of the so-called sorites paradox requires us to abandon 2-valued logic. For discussion see Richard Grandy, 'Boolean Valuations and Truth' (unpublished) and David Sanford, 'Competing Semantics of Vagueness: Many Values Versus Super-Truth', Synthese, V. 33, 1976.

s~ For Mill's view see John Stuart Mill, A System of Logic, Book II, Chapter VI. Pap has suggested that Mill might not be discussing arithmetical claims so much as counting procedures. See Arthur Pap. An Introduction to the Philosophy of Science, pp. 90-3 , New York and London: The Free Press and Collier-Macmillan, Ltd., 1962. See the passage cited in connection with note 32. See Putnam 1979, p. 440. See, for example, 'There is At Least One A Priori Truth' , p. 154.

M

! )

40

63